The Interplay between Financial and Labor Markets By Alexey Levkov B.A., The Hebrew University of Jerusalem, 2002 M.A., The Hebrew University of Jerusalem, 2004 A.M., Brown University, 2006 A Dissertation Submitted in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in the Department of Economics at Brown University Providence, Rhode Island May 2010 © Copyright 2010 by Alexey Levkov The dissertation by Alexey Levkov is accepted in its present form by the Department of Economics as satisfying the dissertation requirements for the degree of Doctor of Philosophy. Date Ross Levine, Advisor Recommended to the Graduate Council Date Kenneth Chay, Reader Date Andrew Foster, Reader Approved by the Graduate Council Date Sheila Bonde, Dean of the Graduate School iii VITA Alexey Levkov was born in Ukraine in 1978 and is an Israeli citizen since 1990. After serving for three years in the Israeli Armed Forces he was accepted at the Hebrew University of Jerusalem. Alexey graduated from the Hebrew University in 2004 with a Bachelor degree in Statistics and Economics and a Master Degree in Economics. In 2005 Alexey started at Brown University, pursuing a PhD in Economics. During his years at Brown Alexey was a teaching assistant in “Data, Statistics, Finance” and “Economics of the Middle East” as well as a teaching fellow in “Introduction to Econometrics”. He was consultant for the World Bank in 2007 and worked as a research assistant for Ross Levine, Joshua Angrist, and Rachel Friedberg. Alexey received the Umezawa–Stoltz Prize for third–year research paper, was nominated for the third meeting with Nobel Laureates in Economics, and received merit dissertation fellowship during the years 2009 and 2010. One of the papers that Alexey co-authored was discussed in the printed edition of The Economist on November 15, 2008 and another co-authored paper is forthcoming in The Journal of Finance. iv PREFACE AND ACKNOWLEDGMENTS The three chapters of the dissertation illustrate the interplay between financial and labor markets. The chapters examine how removal of geographical restrictions on bank branching in the United States between 1960 and 1999 affected union membership, distribution of income, and black-white wage differential in the non-banking sectors. In the first chapter of the dissertation I use bank branch deregulation as an instrumental variable to identify an exogenous increase in the competitiveness of the non-financial sector and evaluate its impact on union membership. The results indicate that by making the economy more competitive, branch deregulation has a first-order negative impact on union membership in the nonbanking sectors. The second chapter of the dissertation is a joint work with Thorsten Beck from Tilburg University and Ross Levine from Brown University. In this chapter we find that bank branch deregulation materially tightened the distribution of income in the United States by boosting incomes in the lower part of the income distribution while having little impact on incomes above the median. Our findings are explained by rising relative wage rates and working hours of unskilled workers following bank deregulation. The third chapter of the dissertation is a joint work with Ross Levine and Yona Rubinstein, both from Brown University. Here we use bank branch deregulation as an instrumental variable to identify an exogenous increase in the competitiveness of the non-financial sector and evaluate its impact on black-white v wage differential in the entire U.S. economy. We find that bank deregulation reduced the racial wage gap by spurring the entry of nonfinancial firms. Consistent with taste-based theories, competition reduced both the racial wage gap and racial segregation in the workplace, particularly in states with a comparatively high degree of racial prejudice. The dissertation relates to an emerging literature that examines the channels underlying the “finance-growth nexus” and advertises the role of labor markets in driving this relationship. I am grateful to Ross Levine and Yona Rubinstein for close mentoring and for inspiring me to explore the links between financial innovations and labor markets, which are at the core of this dissertation. While working on the dissertation I also received excellent comments from Joshua Angrist, Dror Brenner, Kenneth Chay, Kfir Eliaz, Oded Galor, Boris Gershman, Martin Goetz, Juan Carlos Gozzi, Andrew Foster, Victor Lavy, Ivo Welch, and seminar participants at Bar Ilan University, Ben Gurion University, Brown University, Haifa University, NBER Workshop on “Income Distribution and Macroeconomics”, Royal Holloway College, Tel Aviv University, The Federal Reserve Bank of Boston, The Hebrew University of Jerusalem, and the World Bank. I benefited from a generous financial support from the William R. Rhodes Center, the Watson Institute, and the Department of Economics at Brown University. vi Contents 1 The E¤ect of Competition on Unions 1 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 History of Bank Deregulation in the United States . . . . . . . . . . . 10 1.3 Statistical Model and Identi…cation Strategy . . . . . . . . . . . . . . 16 1.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 1.4.1 Union Membership and Coverage . . . . . . . . . . . . . . . . 23 1.4.2 Personal Characteristics of Workers . . . . . . . . . . . . . . . 24 1.4.3 New Incorporations Per Capita . . . . . . . . . . . . . . . . . 25 1.4.4 Timing of Bank Deregulation . . . . . . . . . . . . . . . . . . 25 1.4.5 External Financial Dependence . . . . . . . . . . . . . . . . . 26 1.4.6 Wrongful-Discharge Protections and Collective Bargaining Laws 26 1.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 1.5.1 The Impact of Firm Entry on Union Membership . . . . . . . 28 1.5.2 The Impact of Deregulation on Union Wage Premium . . . . . 32 1.5.3 Estimates by External Financial Dependence . . . . . . . . . . 34 1.5.4 Evidence from the Panel Study of Income Dynamics . . . . . . 37 vii 1.5.5 Integrating the Competitive Explanation with Other Explanations of Union Decline . . . . . . . . . . . . . . . . . . . . . . 41 1.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 1.7 Appendix A: Hazard Model . . . . . . . . . . . . . . . . . . . . . . . 44 1.8 Appendix B: Formation of Microdata Samples . . . . . . . . . . . . . 48 2 Big Bad Banks? The Winners and Losers from Bank Deregulation in the United States 74 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 2.2 Data and Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . 82 2.2.1 Branch Deregulation . . . . . . . . . . . . . . . . . . . . . . . 82 2.2.2 Income Distribution Data . . . . . . . . . . . . . . . . . . . . 84 2.2.3 Control Variables . . . . . . . . . . . . . . . . . . . . . . . . . 86 2.2.4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 2.3 Branch Deregulation and Income Distribution . . . . . . . . . . . . . 89 2.3.1 Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . 89 2.3.2 Deregulation and the Distribution of Income . . . . . . . . . . 90 2.3.3 Deregulation and Income for Di¤erent Income Groups . . . . . 93 2.3.4 Dynamics of Deregulation and the Distribution of Income . . . 94 2.3.5 Mechanisms: Impact of Deregulation as a Function of Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 2.4 Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 viii 2.4.1 Theories of How Financial Markets A¤ect the Distribution of Income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 2.4.2 Evidence on the Entrepreneurship Channel . . . . . . . . . . . 100 2.4.3 Evidence on the Education Channel . . . . . . . . . . . . . . . 101 2.4.4 Evidence on the Labor Demand Channel . . . . . . . . . . . . 104 2.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 3 Race Discrimination and Competition 121 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 3.2 Bank Deregulation and Competition . . . . . . . . . . . . . . . . . . 131 3.2.1 A Brief History of Bank Branch Regulation . . . . . . . . . . 131 3.2.2 Bank Deregulation and Competition in the Non-Financial Sector133 3.3 Conceptual and Statistical Framework . . . . . . . . . . . . . . . . . 134 3.3.1 Conceptual Framework . . . . . . . . . . . . . . . . . . . . . . 134 3.3.2 Statistical Framework . . . . . . . . . . . . . . . . . . . . . . 137 3.4 Data and Econometric Design . . . . . . . . . . . . . . . . . . . . . . 139 3.4.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 3.4.2 Generating Relative Wages and the Racial Bias Indexes . . . . 141 3.4.3 Econometric Design . . . . . . . . . . . . . . . . . . . . . . . . 149 3.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 3.5.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 3.5.2 Bank Deregulation and Blacks’Relative Wages . . . . . . . . 156 3.5.3 Competition and Blacks’Relative Wages . . . . . . . . . . . . 160 ix 3.6 The E¤ect of Competition on Segregation . . . . . . . . . . . . . . . 164 3.6.1 Racial Prejudices, Competition, and Segregation . . . . . . . . 164 3.6.2 The E¤ect of Competition on Segregation: Results . . . . . . 165 3.6.3 Competition and Blacks’Relative Wages Within Industries . . 167 3.7 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 3.7.1 Relative Hours Worked . . . . . . . . . . . . . . . . . . . . . . 169 3.7.2 Selection, Migration, and Self-Employment . . . . . . . . . . . 171 3.7.3 Swimming Upstream 3.7.4 Racial Discrimination or the Poor . . . . . . . . . . . . . . . . 174 . . . . . . . . . . . . . . . . . . . . . . 173 3.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 Bibliography 195 x List of Tables 1.1 Timing of Bank Branch Deregulation and Pre-Existing Union Membership and Coverage: The Duration Model . . . . . . . . . . . . . . 61 1.2 Wald Estimate of the Impact of Log New Incorporations Per Capita on Union Membership . . . . . . . . . . . . . . . . . . . . . . . . . . 62 1.3 The Impact of Log New Incorporations Per Capita on Union Membership: OLS, IV, and TSLS Estimates . . . . . . . . . . . . . . . . . . . 63 1.4 The Impact of Log New Incorporations Per Capita on Union Membership: OLS, IV, and TSLS Estimates with Region - Speci…c Time Fixed E¤ects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 1.5 External Financial Dependence for Manufacturing Sectors . . . . . . 65 1.6 Characteristics of Workers by External Financial Dependence . . . . 66 1.7 The Impact of Bank Deregulation on Union Membership in Manufacturing Sectors By External Financial Dependence . . . . . . . . . . . 67 1.8 The Impact of Bank Deregulation on Union Coverage: Evidence from the Panel Study of Income Dynamics . . . . . . . . . . . . . . . . . . xi 68 1.9 Changes in Workers’Outcomes as a Result of Changes in Union Coverage: Estimates from a Panel of Workers . . . . . . . . . . . . . . . 69 1.10 Timing of Branch and Interstate Bank Deregulation . . . . . . . . . . 70 1.11 Variables Used for Analyses of Panel Study of Income Dynamics . . . 71 1.12 Variables Used for Analyses of Panel Study of Income Dynamics (cont.) 72 1.13 Sample Restrictions Imposed on Microdata Samples . . . . . . . . . . 73 2.1 Timing of Bank Deregulation and Pre-Existing Income Inequality: The Duration Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 2.2 The Impact of Deregulation on Income Inequality . . . . . . . . . . . 117 2.3 The Impact of Deregulation on Income Inequality as a Function of Initial State Characteristics . . . . . . . . . . . . . . . . . . . . . . . 118 2.4 Decomposing the Impact of Deregulation on Income Inequality to Betweenand Within-Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 2.5 The Impact of Deregulation on Earnings Inequality . . . . . . . . . . 120 3.1 The Racial Bias Index, Survey Measures of Racial Prejudice, and Relative Wages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 3.2 Bank Deregulation and Log New Incorporations Per Capita . . . . . . 184 3.3 Bank Deregulation and Relative Wage Rates . . . . . . . . . . . . . . 185 3.4 The Impact of Log New Incorporations Per Capita on Relative Wage Rates: OLS and 2SLS Estimates . . . . . . . . . . . . . . . . . . . . . 186 3.5 The Impact of Log New Incorporations on the Relative Wages of Blacks: OLS and 2SLS Estimates . . . . . . . . . . . . . . . . . . . . . . . . . 187 xii 3.6 The Impact of Log New Incorporations Per Capita on Employment of Blacks in “White”Industries: OLS and 2SLS Estimates . . . . . . . 188 3.7 The Impact of Log New Incorporations Per Capita on Relative Wage Rates: OLS and 2SLS Estimates . . . . . . . . . . . . . . . . . . . . . 189 3.8 Relative Log Hourly Wages and Annual Working Hours in High Racial Bias States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190 3.9 Bank Deregulation and Selection on Observable Characteristics . . . 191 3.10 Dates of Intrastate and Interstate Deregulations, by States . . . . . . 192 3.11 Summary Statistics: Number of observations . . . . . . . . . . . . . . 193 3.12 Racial Bias Index by States, 1970 . . . . . . . . . . . . . . . . . . . . 194 xiii List of Figures 1-1 Trends in Union Membership and Number of New Incorporations Per Capita in the United States . . . . . . . . . . . . . . . . . . . . . . . 51 1-2 Location of Bank Branches in Texas Before Branch Deregulation . . . 52 1-3 Location of Bank Branches in Texas After Branch Deregulation . . . 53 1-4 Pre-Existing Union Membership and Coverage and the Timing of Interstate Bank Deregulation . . . . . . . . . . . . . . . . . . . . . . . . 54 1-5 The Impact of Bank Branch Deregulation on Log New Incorporations Per Capita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 1-6 The Impact of Bank Branch Deregulation on Union Membership . . . 56 1-7 Changes in New Incorporations Per Capita and Union Membership by States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 1-8 The Location of Unionized Workers in the Non-Unionized Wage Distribution Before and After Bank Branch Deregulation . . . . . . . . . 58 1-9 Correlation Between Measures of External Financial Dependence Among Manufacturing Sectors . . . . . . . . . . . . . . . . . . . . . . . . . . 59 1-10 Explaining the Decline in Unions . . . . . . . . . . . . . . . . . . . . 60 xiv 2-1 Timing of Bank Deregulation and Pre-Existing Income Inequality: Graph- xv ical Analysis (A) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 2-2 Timing of Bank Deregulation and Pre-Existing Income Inequality: Graphical Analysis (B) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 2-3 The Impact of Deregulation on Di¤erent Percentiles of Income Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 2-4 The Dynamic Impact of Deregulation on Gini Coe¢ cient of Income Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 2-5 The Impact of Deregulation on the Relative Wages of Unskilled Workers113 2-6 The Impact of Deregulation on the Relative Working Hours of Unskilled Workers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 2-7 The Impact of Deregulation on Unemployment Rate . . . . . . . . . . 115 3-1 Trends and Innovations in the Relative Wage Rates of Blacks Prior to Bank Deregulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178 3-2 The Impact of Deregulation on Entry of Firms . . . . . . . . . . . . . 179 3-3 The Impact of Deregulation on the Relative Wage Rates of Blacks . . 180 3-4 The Impact of Log New Incorporations Per Capita on the Relative Wage Rates of Blacks: Di¤erent OLS and 2SLS Speci…cations . . . . 181 3-5 The Location of Blacks in the White Wage Distribution Before and After Deregulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182 Chapter 1 The E¤ect of Competition on Unions 1.1 Introduction The decline of American unions has been unrelenting since its peak in 1945 with an accelerated reduction in union membership since the late 1970s. At the same time, the U.S. economy has become increasingly competitive, with the number of new incorporations per capita increasing steadily over the last four decades. These trends are depicted in Figure 1-1. What role did increasingly competitive markets play in the decline of union membership? One view holds that less competitive product markets allow unions to negotiate larger wage gains (Abowd and Lemieux (1993); Christo…des and Oswald (1992); Hirsch (2008); Segal (1964); Stewart (1990)). According to Hicks-Marshallian laws of factor demand, the demand for labor is less elastic in less competitive product 1 2 markets. This, in turn, enables unions to demand higher wages for their members in less competitive markets without risking large losses of employment. A countervailing view holds that unions could succeed in competitive industries by organizing workers across the entire industry. When an entire industry is unionized, …rms may survive with higher union costs as long as their competitors face similar costs. Unions might also succeed comparatively well in competitive industries if monopolistic industries more e¤ectively coordinate to …ght unions and strikes (Levinson (1967)). A central obstacle in estimating the causal e¤ect of competition on unions is endogeneity: the presence of unions may a¤ect the degree of competition in an industry. Such reverse relation may arise if …rms are reluctant to enter markets with high union membership due to potentially lower pro…ts (Abowd (1989); Clark (1984); Hirsch (1991); Lee and Mas (2009); Ruback and Zimmerman (1984)). This reverse causality may bias the causal e¤ect of competition on unions and overestimate the potentially negative impact of competition on unions. In this paper I present evidence on the causal e¤ect of competition on union membership and coverage by using innovations in …nancial markets as instrumental variables for the entry of new non-…nancial …rms. The …nancial innovations used in this paper are the state-by-state removal of restrictions on bank branching within state borders. These policy changes occurred in di¤erent states in di¤erent years. Furthermore, past research and evidence presented below shows that the timing of these liberalizations was exogenous to pre-existing conditions in the labor markets (Black and Strahan (2001); Kroszner and Strahan (1999); Levine, Levkov and Rubinstein (2008)), and that bank branch deregulation signi…cantly increased the rate 3 of new …rm entry among non-…nancial …rms (Black and Strahan (2002); Kerr and Nanda (2009)). I report several …ndings from analyzing data that span the 1978–2006 period, and combine information on union membership and coverage of prime–age men from May and Outgoing Rotation Groups Current Population Surveys as well as Panel Study of Income Dynamics, on new incorporations per capita from Dun and Bradstreet, on timing of bank branch deregulation from Kroszner and Strahan (1999), on external …nancial dependence for manufacturing sectors from Cetorelli and Strahan (2006), on wrongful–discharge protections from Autor, Donohue and Schwab (2006), and on collective bargaining laws from Valletta and Freeman (1988). The validity of the identi…cation strategy rests on the assumption that bank deregulation is a legitimate instrument for entry of new …rms. Consistent estimation of the causal impact of competition on union membership requires that (1) union membership does not predict the timing of bank deregulation; (2) that bank deregulation intensi…es competition; and (3) that bank deregulation in‡uences union membership only through its impact on competition. I …rst examine whether pre-existing unionism a¤ects the timing of bank deregulation. If labor unions supported bank regulation because rents were shared with workers, then deregulation should occur later in states where labor unions have greater in‡uence. Using a hazard model and incorporating the political-economy factors from Kroszner and Strahan (1999), I show that neither the levels nor the rates of change in union membership or coverage explain the timing of bank deregulation. Moreover, results from Granger (1969) causality test indicate that changes in union membership 4 did not precede bank deregulation. Both of these results help to rule out potential impact of unions on the timing of bank deregulation. Next, I document that bank branch deregulation has a substantial, …rst-order e¤ect on competition in the economy, as measured by new incorporations per capita. This e¤ect is not only statistically signi…cant, but also quite large relative to the e¤ect of the state business cycle on new business incorporations. Finally, I provide a few arguments that back up the exclusion restriction, which is the critical assumption that drives the bank deregulation IV story. First, some of the two-stage least squares (TSLS) estimates of the impact of competition on union membership are overidenti…ed because several estimates could be constructed from subsets of the instruments. In all speci…cations the overidentifying restrictions are not rejected, thus providing some con…dence in the validity of the instrument. More convincingly, I have estimated the impact of bank branch deregulation on union membership in a sample of manufacturing sectors with low dependence on external …nance, a sample in which bank branch deregulation had no impact on competition (Cetorelli and Strahan (2006)). If bank deregulation a¤ects union membership for reasons other than competition, I would expect deregulation to be related to union membership for this sample. On the other hand, if deregulation a¤ects union membership only through its e¤ect on competition, I would not expect any relationship in this sample. The estimates suggest that deregulation has no e¤ect on union membership in manufacturing sectors with low dependence on external …nance, a …nding which supports the estimation framework employed throughout this paper. The intensi…cation of product market competition, as measured by the accel- 5 eration of new incorporations, reduced union membership. Two-stage least squares estimates of the impact of log new incorporations per capita on union membership indicate that a ten percent increase in new incorporations per capita results in a 1:2 1:9 percentage point decrease in union membership. This result is robust to inclusion of personal characteristics of workers, industry …xed e¤ects, wrongful-discharge protections and collective bargaining laws, and region-speci…c time …xed e¤ects. I also …nd that the decline in union membership is associated with the decline in the union wage premium. I supplement the TSLS analysis by exploring the “reduced-form”impact of bank deregulation on union membership among manufacturing sectors with di¤erent dependence on external …nance. I build on the work of Cetorelli and Strahan (2006) who …nd that removal of restrictions on interstate banking, and not branching within states, increased the total number of establishments in manufacturing sectors with above median dependence on external …nance. Consistently with the …ndings in Cetorelli and Strahan (2006), I …nd that only interstate bank deregulation reduced union membership and that the reduction in union membership occurs only in manufacturing sectors with external …nancial dependence above the median. These …ndings are consistent with the view that …rm entry reduces unionism. At a more descriptive level, I document a monotonic relationship between changes in competition and union membership by estimating these changes separately for each state. Speci…cally, I show that states with larger changes in entry of new incorporations per capita following bank deregulation experienced larger declines in union membership. 6 Still, there are several obstacles with inferring from the evidence above that …rm entry reduces unionism by making states’ economies more competitive. First, the average …rm size fell following deregulation (Cetorelli and Strahan (2006)). If unions primarily target large …rms, then a reduction in …rm size may explain the fall in unionism independently from the competition channel. Second, the composition of …rms and workers has changed after bank deregulation (Black and Strahan (2002); Kerr and Nanda (2009)). If workers do not organize immediately to bargain collectively with employers, then this might explain why unionism declines after bank deregulation. To address these concerns, I use data from the Panel Study of Income Dynamics (PSID), so that I can trace individual workers throughout their employment histories. In particular, the PSID provides information of exact tenure with current employer, which enables to estimate the impact of …rm entry on union coverage among “mature” employees that have been with their current employers for long periods of time. This excludes new workers in existing …rms and workers in new …rms that may not be union members for reasons that have nothing to do with the competitiveness of the economy. The evidence from the PSID shows a reduction of 3 6 percentage points in union coverage following deregulation. The reduction in union coverage is larger for workers with longer tenures with their current employer, which is inconsistent with the argument that changes in unionism after deregulation are driven by changes in composition of workers. Based on the evidence, I conclude that intensi…ed competitiveness of the U.S. 7 economy has a …rst-order e¤ect on declining unionism. Between 1978 and 2006 union membership among wage and salary male workers fell by about 20%. At the same time, entry of new incorporations per capita increased by about 100%. Using the most conservative estimates, a ten percent increase in new incorporations per capita reduces union membership by 1:24 percentage points. Thus a 100% increase in new incorporations per capita reduces union membership by 12:4 percentage points, or about sixty percent of the overall reduction in union membership. The economically large impact of competition on unionism is net of other potential explanations of union decline. By using industry …xed e¤ect, I am estimating the change in union membership as a result of …rm–entry within industries, thus accounting for the movement of workers from manufacturing towards the service sector. By having state …xed e¤ects, I account for state–speci…c (and thus region–speci…c) observable and unobservable characteristics that do not vary over time. State …xed e¤ects also account for the movement of workers to the South which has a long tradition of right–to–work laws. I also allow for the time trend in union membership to vary by regions by including region–speci…c time …xed e¤ects. Inclusion of wrongful–discharge protections and collective bargaining laws accounts for di¤erences in workers’demand for unionism. Finally, time …xed e¤ects control for national changes in union membership and thus account for oil and energy crises that were especially harmful for the highly unionized automobile industry. The large negative e¤ect of competition on unionism that I estimate is consistent with the empirical …ndings in Ashenfelter and Johnson (1972), Hirsch and Berger (1984), and Lee (1978) who …nd a positive association between industry concentration 8 and the likelihood of unionism and with the results in Slaughter (2007) who …nds larger union representation in industries with lower global engagement. The results are also consistent the empirical literature that …nds a larger union wage premium in less competitive industries (Abowd and Lemieux (1993); Christo…des and Oswald (1992); Segal (1964); Stewart (1990)) and with the literature that …nds a reduction in union bargaining power following deregulation in the airline (Cappelli (1985); Card (1986); Card (1996a); Cremieux (1996)) and the trucking (Rose (1987); Belzer (1995); Peoples (1996); Belman and Monaco (2001)) industries. This paper’s results are also consistent with Farber (1987), Farber (1989), and Farber and Krueger (1993) who …nd that one of the main reasons for declining unionism in the United States is lower demand for union representation. Indeed, if unions can no longer provide higher wages for their members due to higher competitive pressures, then workers might not vote for further representation by the union. Finally, the results are consistent with Freeman (1988) who argues that most of the union decline can be explained by increasing management opposition to unions due to competitive pressures. This paper’s results, however, are at odds with the empirical evidence in Weiss (1966), who …nds that for a given degree of union strength, a greater concentration in an industry yields a smaller rate of increase in wages. Weiss explains his …ndings by hypothesizing that in concentrated industries wages are already high, so that unions are not able to add much. Similarly, the results are inconsistent with (a) Nickell, Vainiomaki and Wadhwani (1994) who …nd very similar union wage premium in …rms operating in markets with di¤erent degrees of competition as measured by market share and with (b) Kahn (1979) who doesn’t …nd any signi…cant relation 9 between industry concentration and union membership. Finally, this paper’s …ndings regarding union wage premium are consistent with the large literature that documents a substantial wage premium for union workers, but inconsistent with the careful study of DiNardo and Lee (2004), who …nd that workers in establishments that barely rejected union representation have very similar outcomes to workers in establishments that barely approved representation.1 This paper relates to a large literature that shows a …rst–order relationship between …nancial development and economic growth (Demirgüç-Kunt and Maksimovic (1998); Jayaratne and Strahan (1996); King and Levine (1993a); King and Levine (1993b); Levine and Renelt (1992); Levine and Zervos (1998); Rajan and Zingales (1998) are notable recent contributions to this literature). If unions impose ine¢ ciency in production, then this paper’s results show a speci…c channel through which …nancial innovations increase productivity and growth in the economy.2 This paper thus relates to an emerging literature that examines the channels underlying the …nance–growth nexus and advertises the role of labor markets in driving this relationship. The rest of the paper is organized as follows. The next section discusses the history of bank deregulation in the United States. Section 1.3 presents the statistical model 1 See Lewis (1986) and Jarrell and Stanley (1990) for a review of the union wage premium literature. 2 The impact of unions on productivity is not clear–cut. In a landmark study, Freeman and Medo¤ (1984) conclude that “... most studies ... …nd that unionized establishments are more productive than otherwise comparable nonunion establishments” (p. 169). Their conclusion, however, is challenged by Hirsch (2007) who concludes that “The most important point to bring away from the productivity evidence may be the absence of a large positive e¤ect due to unions”(p. 211, italics in the origin). A case study by Krueger and Mas (2004) further shows that unions are associated with lower product quality. 10 and identi…cation strategy. Section 1.4 describes the data. Section 1.5 presents the main …ndings and Section 1.6 concludes. 1.2 History of Bank Deregulation in the United States Geographic restrictions on banks have their origins in the U.S. Constitution, which limited states from taxing interstate commerce and issuing …at money. In turn, states raised revenues by chartering banks and taxing their pro…ts. Since states received no charter fees from banks incorporated in other states, state legislatures prohibited the entry of out-of-state banks through interstate bank regulations. To maximize revenues from selling charters, states also e¤ectively granted local monopolies to banks by restricting banks from branching within state borders. These intrastate branching restrictions frequently limited banks to operating in one city (Flannery (1984)). By protecting ine¢ cient banks from competition, geographic restrictions created a powerful constituency for maintaining these regulations even after the original …scal motivations receded. Indeed, banks protected by these regulations successfully lobbied both the federal government and state governments to prohibit interstate banking (White (1982); Economides, Hubbard and Palia (1996)). In the last quarter of the twentieth century, however, technological, legal, and …nancial innovations diminished the economic and political power of banks bene…ting from geographic restrictions. In particular, a series of innovations lowered the 11 costs of using distant banks. This reduced the monopoly power of local banks and weakened their ability and desire to lobby for geographic restrictions. For example, the invention of automatic teller machines (ATMs), in conjunction with court rulings that ATMs are not bank branches, weakened the geographical link between banks and their clientele. Furthermore, the creation of checkable money market mutual funds made banking by mail and telephone easier, thus further weakening the power of local bank monopolies. Finally, the increasing sophistication of credit scoring techniques, improvements in information processing, and the revolution in telecommunications reduced the informational advantages of local bankers, especially with regards to small …rms. These national developments interacted with preexisting state characteristics to shape the timing of bank deregulation across the states. As shown by Kroszner and Strahan (1999), branch deregulation occurred later in states where potential losers from deregulation (small, monopolistic banks) were …nancially stronger and had a lot of political power. On the other hand, deregulation occurred earlier in states where potential winners of deregulation (small …rms) were relatively numerous. Most states deregulated geographic restrictions on banking between the mid-1970s and 1994, when the Riegle-Neal Act e¤ectively eliminated these restrictions. Figures 1-2 and 1-3 illustrate the substantial changes in the geographical location of bank branches. The …gures indicate the location of bank branches in Texas in the period before (…gure 1-2) and after bank branch deregulation (…gure 1-3). Although the …gures depict changes in location of branches in a single state, similar results are found for virtually all states. The circles in the …gures represent the total number 12 of bank branches within a zip code. The size of each circle is proportional to the number of bank branches. As shown, bank branch deregulation increased the spread of branches within state borders and increased their concentration in already established locations, thus intensifying competition among bank branches. The forces driving bank deregulation were exogenous to competition in the non…nancial sector and pre-existing presence of unions. In particular, the timing of deregulation was neither shaped by new …rm formation (Black and Strahan (2002); Kerr and Nanda (2009)), nor by the degree of income inequality (Beck, Levine and Levkov (forthcoming)), or black-white wage di¤erential (Levine et al. (2008)) in each state. Moreover, as shown in Table 1.1 neither the preexisting levels nor rates of change in union membership or coverage explain the timing of bank deregulation. Table 1.1 shows estimates from a hazard model where the dependent variable is log expected time to bank branch deregulation. Each coe¢ cient measures the percentage change in the hazard of bank branch deregulation as a result of a marginal change in: (a) proportion of wage and salary workers who are union members, (b) changes in proportion of wage and salary workers who are union members, (c) proportion of wage and salary workers who are covered by a collective bargaining agreement, but not necessarily union members, and (d) changes in proportion of wage and salary workers who are covered by a collective bargaining agreement, but not necessarily union members. Standard errors are adjusted for state level clustering and appear in parentheses. Columns (1)–(4) use wage and salary workers in all industries, while columns (5)–(8) use only workers in manufacturing. All speci…cations control for political economy factors that a¤ect the timing of bank branch deregulation (Kroszner 13 and Strahan (1999)). These factors are listed in the notes to Table 1.1. In all speci…cations the coe¢ cients of interest are statistically insigni…cant, indicating that neither the preexisting levels nor rates of change in union membership or coverage explain the timing of bank branch deregulation.3 Interstate bank deregulation can not be examined in the context of a hazard model due to dependence of timing of deregulation between states. Nevertheless, Figure 1-4 provides some evidence that the timing of interstate deregulation was not a¤ected by preexisting presence of unions. The upper panel of Figure 1-4 plots the level of union membership and the rate of change in union membership before interstate deregulation and the year of deregulation for each state, where union membership is the proportion of wage and salary workers who are members of a union. The lower panel uses union coverage instead of membership, where union coverage is the proportion of wage and salary workers who are covered by a collective bargaining agreement, but not necessarily union members. The dashed lines are the regression lines that depict the degree of correlation between the timing of interstate deregulation and preexisting presence of unions. The absolute values of t-statistics for the correlations in the plots are: 0:53 in plot (a), 0:46 in plot (b), 0:35 in plot (c), and 0:47 in plot (d). The results in Figure 1-4 indicate that neither the preexisting levels nor rates of change in union membership or coverage explain the timing of bank deregulation. An extensive literature examines the impact of both interstate and branch bank deregulation. Along many dimensions, branch deregulation exerted a dominant e¤ect 3 Estimation of the hazard model is explained in details in the Appendix. 14 on banking system performance. For example, Jayaratne and Strahan (1998) …nd that removing branching restrictions improved banking e¢ ciency by reducing interest rates on loans, raising them on deposits, lowering overhead costs, and shrinking loan losses. While interstate and branch deregulation both improved bank e¢ ciency, branch deregulation exerted a more robust impact on e¢ ciency when simultaneously controlling for interstate deregulation. Within the banking industry, Ashenfelter and Hannan (1986) …nd a positive impact of branch deregulation and the share of female employees across several banking markets in Pennsylvania and New Jersey. At a more aggregate level, branch deregulation also accelerated the growth rate of per capita GDP and personal income (Jayaratne and Strahan (1996), Huang (2008)), lowered economic volatility (Morgan, Rime and Strahan (2004), Demyanyk, Ostergaard and Sorensen (2007)), reduced the gender gap in earnings and employment among bank employees (Black and Strahan (2001)), and reduced income inequality in the economy (Beck et al. (forthcoming)). More speci…cally for the purposes of this paper, interstate and branch bank deregulation intensi…ed competition among …rms in the non-…nancial sector by reducing barriers to entry. Black and Strahan (2002) …nd that deregulation helped entrepreneurs start new businesses, with the rate of new incorporations per capita in a state increasing by six percentage points following deregulation. Kerr and Nanda (2009) …nd that deregulation increased the number of new start-ups by six percentage points and expanded the number of facilities of existing …rms by four percentage points across all sectors in the economy. Furthermore, they …nd a dramatic increase in both the entry and exit of …rms, suggesting that deregulation increased contestability 15 throughout the economy. Figure 1-5 shows the dynamic impact of branch deregulation on log new incorporations per capita. Speci…cally, the …gure plots estimates of 1 25 and the corre- sponding 95% con…dence intervals from the following speci…cation: ln (entry)st = s + t + 10 1 Dst + 9 2 Dst + ::: + +15 25 Dst + st where entryst is the number of new incorporations per capita in state s at time t. Dstj equals one for states in the j th year before branch deregulation and equals zero +k otherwise. Dst equals one for states in the k th year after branch deregulation and equals zero otherwise. s and t are state and year …xed e¤ects, respectively. I exclude the year of deregulation, thus estimating the dynamic e¤ect of branch deregulation on log new incorporations per capita relative to the year of deregulation. Vertical lines in the plot mark 95% con…dence intervals, adjusted for state level clustering. The results in Figure 1-5 indicate that trends in log new incorporations per capita did not precede deregulation, which helps to rule out reverse causality. As shown, the impact of deregulation is insigni…cantly di¤erent from zero for all years before deregulation. After deregulation, on the other hand, new incorporations per capita increase signi…cantly. The impact of deregulation becomes signi…cantly di¤erent from zero in the second year after deregulation, grows for the next four years, then becomes steady for the next six years, and …nally increases a little more to reach its highest level in the fourteenth year after deregulation. 16 1.3 Statistical Model and Identi…cation Strategy This section describes the procedure for estimating the impact of competition on union membership. It then outlines the main problems with such estimation when using non-experimental data and o¤ers possible solutions to overcome these obstacles. The structural model of interest is: unionst = 0 + 1 compst + "st (1.1) where unionst is the proportion of union members in state s and time t, compst is a measure of competition among …rms in state s and time t, and "st is the error term. For simplicity of exposition I assume that all the relevant state characteristics as well as state and year …xed e¤ects have been accounted for in previous steps. The coe¢ cient of interest, 1, measures the percentage change in union membership as a result of a marginal change in competition among …rms. A negative and signi…cant estimate of 1 implies that …rm competition reduces union membership. There are several obstacles to estimating the structural coe¢ cient of interest in (1:1). First, competition among …rms is not directly observed. Instead, it is usually proxied by concentration ratios, barriers to entry, or various measures of market power. In this paper I will use entry of new incorporations per capita (entryst ) as a proxy for product market competition in each state and year. Thus, in practice 17 equation (1:1) becomes: where ust = unionst = 0 + 1 entryst +[ = 0 + 1 entryst + ust 1 (compst 1 (compst (1.2) entryst ) + "st ] entryst ) + "st . The error term ust contains the di¤erence between compst and entryst . If this di¤erence is correlated with entryst then the resulting Ordinary Least Squares (OLS) estimate of 1 will be biased towards zero due to attenuation bias. Another obstacle to estimate the causal impact of competition on union is potential reverse causality. Firms may be reluctant to enter markets with high union membership due to potentially lower pro…ts (Abowd (1989); Clark (1984); Hirsch (1991); Lee and Mas (2009); Ruback and Zimmerman (1984)). This relation can be summarized by the following equation: entryst = 0 + 1 unionst (1.3) + vst The reverse relationship in (1:3) poses a problem for the OLS estimate of 1 in (1:2) because it results in potentially non-zero covariance between entry and the error term u: Cov (entry; u) = If 1 < 0 and 1 < 0 such that 1 = (1 1 1 1 1) V ar (u) 1 1 < 0, the correlation between entry and u is negative, suggesting that in the presence of reverse causation the OLS estimate of 1 18 in equation (1:2) overestimates the negative impact of competition on union membership.4 The biases associated with a “naive” OLS estimation of equation (1:1) may be solved by using an instrumental variable which is strongly correlated with entry of new incorporations and a¤ects union membership only through its impact on new incorporations. Speci…cally, let Dst be an indicator which equals one in the years after bank branch deregulation and zero otherwise. Equations (1:6a) and (1:6b) provide the basis for an instrumental variable estimate of the e¤ect of entry of …rms on union membership: The parameter 12 unionst = 11 + 12 Dst + "st (1.6a) entryst = 21 + 22 Dst + (1.6b) captures the “reduced-form”e¤ect of bank deregulation on union membership. The parameter 4 st 22 captures the “…rst-stage”e¤ect of bank deregulation The covariance between entryst and ust (for simplicity I will omit the indexing st): Cov (entry; u) = Cov ( = 0 + 1 union + v; u) (1.4) 1 Cov (union; u) + Cov (v; u) Assuming Cov (v; u) = 0, equation (1:4) becomes: Cov (entry; u) = 1 Cov (union; u) = 1 Cov ( 0 = 1 1 Cov (entry; u) + (1.5) 1 entry + u; u) + Solving (1:5) for covariance between entry and u yields: Cov (entry; u) = 1 1 1 1 V ar (u) 1V ar (u) 19 on entry of new incorporations. The Instrumental Variable (IV) estimator of entry of …rms on union membership is the ratio between the two parameters, namely 12 = 22 . The validity of the identi…cation strategy rests on the assumption that bank deregulation is a legitimate instrument for entry of new incorporations. Consistent estimation of the causal impact of competition on union membership requires, among other things, that union membership does not predict the timing of bank deregulation. If labor unions supported bank regulation because rents were shared with workers, then deregulation should occur later in states where labor unions have greater in‡uence.5 The results from the duration model in Table 1.1, however, indicate that neither the preexisting levels nor rates of change in union membership or coverage explain the timing of bank branch deregulation. Moreover, Figure 1-6 shows that changes in union membership did not precede deregulation. Union membership falls after bank deregulation. Both of these results help to rule out potential impact of union membership on the timing of bank deregulation. For being a legitimate instrument, bank deregulation must be correlated with entry of new …rms. Previous work has shown that by reducing barriers to entry, bank deregulation increased the rate of new incorporations per capita and start-up creation (Black and Strahan (2002), Kerr and Nanda (2009)). Figure 1-5 is a graphical depiction of this relation, which is the …rst stage of the IV estimation. The …gure clearly shows a signi…cant increase in the number of new incorporations per capita following bank branch deregulation. This e¤ect is statistically signi…cant at the 5 5 The potential reverse causation in (1:6a) will create a correlation between Dst and "st and therefore bias 12 . 20 percent and is quite large relative to the e¤ect of the state business cycle on new business incorporations. For example, an increase in personal income growth of one percent would generate an increase in incorporations of a little more than one percent initially, and this increase would eventually peter out after about six years. The critical assumption that drives the bank deregulation IV story is that bank deregulation in‡uences union membership only through its impact on competition. If bank deregulation in‡uences union membership for other reasons, my approach is called into question. It is therefore useful to consider other potential impacts of bank deregulation. First, as documented by Cetorelli and Strahan (2006), the average …rm size falls following deregulation. If unions primarily target large …rms, then a reduction in …rm size may explain the fall in union membership independently from the competition channel. Second, the number of new …rms in the economy has risen after bank deregulation (Black and Strahan (2002); Kerr and Nanda (2009)) changing the composition of workers and …rms. If new workers do not organize immediately to bargain collectively with employers, then this might explain why union membership falls after bank deregulation, again, independently from the competitive channel. There are few arguments that back up the exclusion restriction assumption. First, Figure 1-6 indicates that union membership falls for many years following bank deregulation and that bank deregulation has a trend e¤ect on union membership. If changes in union membership are simply driven by addition of new workers who do not organize immediately to bargain collectively with employers, then we should see a level e¤ect of deregulation on union membership, not a trend e¤ect. Second, some of the TSLS estimates in tables 1.3 and 1.4 are overidenti…ed be- 21 cause several estimates of the impact of log new incorporations per capita could be constructed from subsets of the instruments. In all speci…cations the overidentifying restrictions are not rejected, thus further providing some con…dence in the validity of the instruments. Finally, and perhaps most convincingly, I have estimated the impact of bank branch deregulation on union membership in a sample of manufacturing sectors with low dependence on external …nance, a sample in which bank branch deregulation had no impact on entry of new establishments (Cetorelli and Strahan (2006)). If bank deregulation a¤ects union membership for reasons other than competition, I would expect deregulation to be related to union membership for this sample. On the other hand, if deregulation a¤ects union membership only through its e¤ect on competition, I would not expect any relationship in this sample. The estimates suggest that deregulation has no e¤ect on union membership in manufacturing sectors with low dependence on external …nance, a …nding which supports the estimation framework employed throughout this paper. The regression of interest throughout the IV analysis is: unionst = s + t + entryst + X0st +"st where unionst is the proportion of union members in state s and time t, (1.7) s and t are state and year …xed e¤ects, respectively, entryst denotes the predicted value of the natural logarithm of new incorporations per capita, Xst is a vector of time–varying observable characteristics of states, and "st is the error term. Union membership is 22 the average state–year value of residuals from an OLS regression of union membership indicator on a series of dummy variables that indicate years of completed education (0–8, 9–11, 12, 13–15, and 16+), potential experience (age – years of completed education –6) and its square, and industry …xed e¤ect. The natural logarithm of new corporations per capita is predicted in the …rst stage by bank branch deregulation. I consider several ways to use bank deregulation as an instrument. First, I use a dummy variable which equals one in all year after deregulation and zero otherwise. I also use years since deregulation as well as a quadratic function of years since deregulation. Finally, I use a non–parametric function of years relative to deregulation to instrument log new incorporations per capita, by including a series of dummy variables for each year before and after bank branch deregulation. When estimating equation (1:7) I allow for possible serial correlation of errors over time by clustering the standard errors of the estimated coe¢ cients at the state level. 1.4 Data To consistently estimate the e¤ect of competition among …rms on union membership I collect data on union membership, new incorporations per capita, timing of bank deregulation, as well as labor bargaining laws and wrongful-discharge protections for all states in the period 1978–2006. I supplement these data with estimates of external …nancial dependence for manufacturing sectors from Cetorelli and Strahan (2006). In accord with previous literature on bank deregulation, I drop Delaware and South Dakota because of large concentration of credit card banks in these states. 23 1.4.1 Union Membership and Coverage Union membership and coverage are obtained from the May Supplement to the Current Population Survey (CPS) for the years 1978–1981 and from CPS Outgoing Rotation Groups (ORG) for the years 1983–2006. The 1982 Current Population Survey did not include any union status questions and thus is excluded from the analyses. The sample period is 1978–2006 because of limitation of some state characteristics discussed below. I restrict the sample of CPS respondent to prime-age (25–54) white men who work for wage and salary, excluding those who work in agriculture and have non-missing union status. I further restrict the sample to those who work or with a job but currently not working. Respondents are counted as union members if they respond “yes” to the following question, asked to employed wage and salary workers: “On this job, is ___ a member of a labor union or of an employee association similar to a union?” Respondents who answer “no” to this question are then asked: “On this job, is ___ covered by a union or employee association contract?”Respondents are counted as covered if they are union members or if they are not members but say they are covered by a union contract. I supplement the CPS data with union coverage information from the Panel Study of Income Dynamics (PSID) for the years 1977-1993. I use union coverage and not union membership because it is asked more frequently and available for all years. I do not use post 1993 data because information on the state of residence is not consistently available after 1993. Similarly to the CPS, I restrict the PSID sample to prime-age 24 (25-54) white male heads of households who are not self-employed and belong to the “core”sample of 1968 families originally interviewed by the PSID and have nonmissing union status. Respondents are counted as covered by a union contract if they respond “yes”to the following question: “Is your current job covered by a union contract?”. Data appendix provides more details on the sample restrictions imposed on the CPS and the PSID. Tables 1.11 and 1.12 list all the variables used for the construction of the PSID sample, while Table 1.13 provides further details about the construction of the CPS and the PSID microdata samples. 1.4.2 Personal Characteristics of Workers The CPS …les provide information on years of completed education as well as age at the time of the survey. Questions regarding years of completed education were changed starting from the 1992 CPS (see, for example, Polivka (1996)). After the redesign years of completed education are no longer available in a continuous form, but only in categories. I use a time consistent measure of years of completed education by constructing …ve categories for years of completed education: 0-8, 9-11, 12, 13-15, and 16+. To calculate potential experience in data years coded with the revised education question, I use …gures from Park (1994) to assign years of completed education to each worker based upon highest degree held. Years of potential experience are then calculated as age minus assigned years of education minus 6, rounded down to the nearest integer value. 25 Years of completed education are also available in the PSID. Similarly to the analyses of CPS …les, I construct …ve categories for years of completed education: 0-8, 9-11, 12, 13-15, and 16+. One of the advantages of PSID over CPS is availability of exact tenure with the current employer. This information will be crucial when examining whether or not changes in union coverage are driven by “mature”or “new” workers. 1.4.3 New Incorporations Per Capita Information on new incorporation is obtained from Dun and Bradstreet. This series comes from the individual states and is available from 1964 until 2006. However, only post 1978 data are used due to limitations of wrongful discharge protections (discussed below) prior to 1978. New incorporations are adjusted to per capita terms by dividing new incorporations by population estimates, obtained from the U.S. Census Bureau. Following Black and Strahan (2002), I use the natural logarithm of new incorporations per capita as a proxy for competition among …rms. 1.4.4 Timing of Bank Deregulation Timing of bank deregulation is obtained from Kroszner and Strahan (1999). These data indicate the year in which di¤erent states permitted: (1) branching via mergers and acquisitions (M&As) through the holding company structure, which was the …rst step in the deregulation process, followed by de novo branching, and (2) out–of–state banking companies to buy banks headquartered in a state. Table 1.11 lists the years 26 of bank deregulation for each state. 1.4.5 External Financial Dependence Estimates of external …nancial dependence for manufacturing sectors are from Cetorelli and Strahan (2006), Table II. As in Cetorelli and Strahan (2006), I use two measures of …nancial dependence of manufacturing sectors. The …rst measure is the proportion of capital expenditures …nanced with external funds. A negative value indicates that …rms in the indicated sector have free cash ‡ow, whereas a positive value indicates that …rms must issue debt or equity to …nance their investment. The …gures represent the median value for COMPUSTAT …rms (that have been on COMPUSTAT for at least 10 years) in each sector over the 1980 to 1997 period. A second measure is the median ratio of loans to assets for small …rms from the Federal Reserve 1998 Survey of Small Business Finance. Larger values of the loans/assets ratio indicate greater dependence on external …nance. 1.4.6 Wrongful-Discharge Protections and Collective Bargaining Laws Wrongful-discharge protections for each state are from Autor et al. (2006). These protections were created by U.S. state courts in the 1970s and the 1980s and limited the ability of employers to …re workers. Presence of these protections in a state might reduce workers’ demand for unionism. In robustness checks, therefore, I include three dummy variables that indicate presence of “public policy”, “good-faith”, and 27 “implied contract” protections. The “public policy” protection provides employees with protections against discharges that would prevent an important public policy, such as performing jury duty; “good-faith”prevents employers from …ring workers for bad cause such as just before a substantial commission is due; and “implied contract” protection comes into force when an employer implicitly promises not to terminate a worker without good cause. Data on wrongful-discharge protections are available for the period 1978-1998 and exclude the District of Columbia. Between 1999 and 2006, I impute the presence of wrongful-discharge protections for each state by assigning the 1998 values. The wrongful-discharge data are supplemented with data on labor bargaining laws from Valletta and Freeman (1988). These data describe the status of public sector collective bargaining policies for each state. Speci…cally, these data include indicators for whether or not states permit collective bargaining and striking. An updated version of these data is obtained from the National Bureau of Economic Analysis data collection section and includes public sector collective bargaining policies by state from 1955 until 1996. Between 1997 and 2006, I impute collective bargaining policies by assigning the 1996 values. 28 1.5 1.5.1 Results The Impact of Firm Entry on Union Membership In Table 1.2 I use bank branch deregulation to estimate the impact of log new incorporations per capita on union membership using Wald (1940) estimator. This estimator computes the impact of log new incorporations per capita on union membership as the ratio of the change in union membership to the change in log new incorporations as a result of bank deregulation. The change in union membership as a result of bank branch deregulation is :016, while the corresponding change in log new incorporations per capita is :082. The ratio of these two changes, :194, is a Wald estimate of the impact of log new incorporations per capita on union membership. The Wald estimator is likely to provide a consistent estimate in this case if bank deregulation is uncorrelated with determinants of union membership other than log new incorporations per capita. The last row in Table 1.2 provides the OLS estimate of the impact of log new incorporations per capita on union membership. The OLS estimate is the coe¢ cient on log new incorporations per capita from a regression of union membership on log new incorporations per capita and state and year …xed e¤ects. The Wald estimate ( :194) is much more negative than the OLS estimate ( :007), suggesting that error in measuring competition in a state biases the OLS estimate towards zero much more than the potential reverse causation biases the OLS estimate away from zero. The Wald estimate suggests that a 10 percent increase in new incorporations per capita results in 1:94 percent reduction in union membership, which is about one–…fth of 29 the standard deviation of union membership in the sample. In Table 1.3 I use alternative speci…cations of bank deregulation to estimate the impact of log new incorporations per capita on union membership. Columns (1) and (2) replicate the OLS and Wald estimates from Table 1.2. In column (3) I use years since bank deregulation as an instrument for log new incorporations per capita, while in column (4) I use years since bank deregulation and its square. Finally, in column (5) I use a nonparametric function of years relative to bank deregulation by including a series of dummy variables for each year before and after deregulation. In all speci…cations I control for state and year …xed e¤ects. The speci…cations in panel B additionally control for collective bargaining laws and labor protection laws. In the linear speci…cation of instruments, the instrumented impact of log new incorporations per capita in statistically insigni…cant from zero. The linear speci…cation also has relatively low …rst stage F –statistic of 2:28 in panel A and 1:37 in panel B. Both the quadratic and the non-parametric speci…cations of instruments, on the other hand, yield a statistically signi…cant impact of log new incorporations per capita on union membership. These estimates range from panel B to :124 in column (5) of :164 in (4) of panel A. The TSLS estimates in columns (4) and (5) of Table 1.3 are overidenti…ed because several estimates of the impact of log new incorporations per capita could be constructed from subsets of the instruments. The p-values of overidentifying restrictions tests presented at the bottom of each panel test the hypothesis that the di¤erent combinations of instruments yield the same estimated impact of log new incorporations per capita on union membership, i.e., that all instruments are exogenous. 30 The p-values are calculated from a J statistic, which under the null hypothesis is distributed 2 m 1, where m is the number of instruments. In all speci…cations the overidentifying restrictions are not rejected, providing some con…dence in the validity of the instruments. Table 1.4 is similar to Table 1.3 except that all speci…cations in panels A and B also control for region–speci…c time …xed e¤ects. This ‡exible speci…cation allows for a di¤erent time trend in the di¤erent regions (North East, Mid West, South, and West). Most of the estimates lose their statistical signi…cance due to larger standard errors. The estimates in the last column, however, remain statistically signi…cant at the 5% (panel A) and at the 10% (panel B) levels. To clarify the timing of the impact of deregulation on union membership, I examine the dynamics of the relationship between branch deregulation and union membership. In Figure 1-6 I trace the “reduced–form”impact of deregulation on union membership for every year before and after bank branch deregulation. I exclude the year of deregulation, thus estimating the dynamic impact of deregulation on union membership relative to the year of deregulation. Speci…cally, I plot estimates of 1 25 from the following regression: unionst = s + t + 10 1 Dst + 9 2 Dst + ::: + +15 25 Dst + est (1.8) where unionst is the proportion of union members in state s at time t. Dstj equals one for states in the j th year before branch deregulation and equals zero otherwise. +k Dst equals one for states in the k th year after branch deregulation and equals zero otherwise. s and t 31 are state and year …xed e¤ects, respectively. Vertical lines in the plot mark 95% con…dence intervals, which are adjusted for state level clustering. Equation (1:8) is a Granger (1969) causality test. The test is a check on whether, conditional on state and year …xed e¤ects, past deregulation predicts union membership, while future deregulation does not. As shown, there are no upward or downward trends in union membership before deregulation which helps to rule out reverse causality. Rather, union membership falls signi…cantly following bank deregulation. The pattern of coe¢ cients depicted in Figure 1-6 provides evidence that bank branch deregulation led to signi…cant union decline rather than vice versa. To provide further evidence about the impact of …rm entry on union membership following bank branch deregulation, I calculate percentage change in new incorporations per capita and union membership for each state as a result of bank deregulation. Speci…cally, for each state I contrast the average value of union membership in all years before bank deregulation from the average after deregulation and divide it by the average value of union membership in all years prior to deregulation. I adjust the changes for the fact that the post deregulation period is di¤erent for the di¤erent state by dividing percentage changes in union membership by years since deregulation. The interpretation of the resulting …gures is the percentage change in union membership per year of deregulation. I follow a similar procedure for new incorporations per capita. I then sort states according to percentage change in new incorporations per capita. The results are shown in Figure 1-7. The states of IA, AR, MS, KY, MN, WI, and MO have the largest percentage gain in new incorporations per capita following bank deregulation. These states also 32 have the largest percentage decline in union membership. The states of NH, CT, OH, VA, OR, HI, and WA, on the other hand, have the smallest percentage gain in new incorporations per capita and the smallest decline in union membership. The results in Figure 1-7 depict the monotonicity of the reduction in union membership with respect to percentage changes in new incorporations per capita. 1.5.2 The Impact of Deregulation on Union Wage Premium If unions successfully bargain a wage premium for their members and if bank deregulation reduces union membership by decertifying unions, then bank deregulation should lead to a reduction in union wage premium. Figure 1-8 provides a reduced form e¤ect of bank deregulation on union wage premium by comparing the changes in union wage premium across the full distribution of wages of union and non-union workers. The …gure plots the location of union members in the non–union log real conditional wage distribution, before and after bank branch deregulation.6 The plots was constructed using the following procedure: First, I regress log real hourly earnings of non–union members on …ve indicators of years of completed education (0–8, 9–11, 12, 13–15, and 16+), potential experience (age –years of completed education –6) and its square, industry …xed e¤ects, and state …xed e¤ects. I run the 6 Hourly wages are converted to constant 1982 dollars using the Consumer Price Index. I restrict the sample to prime age (25-54) white male wage and salary workers, whose real wages are above one–half of the minimum wage in 1982 dollars and who work at least 40 hours per week. I further drop workers with real wages above the 99th percentile of year–speci…c distribution of real wages. 33 regression separately for every year using the following speci…cation: ln wagenon ist union = X0ist t + uist (1.9) This forces the resulting residuals of non–union members to sum up to zero in every year. Next, I calculate residuals for union members, rist , based on their own personal characteristics (X) and the estimated return to these characteristics from equation (1:9): rist = wageunion ist X0ist b t (1.10) This procedure creates log real hourly earnings of union members relative to non– union members (they are the benchmark because their residuals are zero for every year by construction) who have the same observable characteristics and the same time–varying return to these characteristics. There are two main advantages of the two–step procedure described in equations (1:9) and (1:10). First, given the changes in the structure of wages since the mid 1970s (see Katz and Autor (1999) for a review), the procedure allows for the return to observable characteristics to vary over time. Second, the procedure not only allows to compare wages of union and non-union workers with the same observable characteristics but also forces the time–varying return to these characteristics to be the same for union and non–union workers. Next, I keep 100 union workers, each corresponding to a di¤erent percentile (1 100) of union workers’ log real conditional hourly earnings distribution and I calculate their position in non–union workers’ relative log real hourly earnings dis- 34 tribution. I repeat this procedure before (solid line) and after (dashed line) bank branch deregulation. The results in Figure 1-8 demonstrate a signi…cant reduction in union wage premium after branch deregulation across the entire distribution of wages. The median union worker, for example, corresponds to 78th percentile in the nonunion workers’ wage distribution before deregulation. After deregulation, however, the median union worker falls to 63rd percentile. The reduction in union wage premium following bank deregulation, seems to be consistent with the large literature that documents a substantial 10% 15% union wage premium but is inconsistent with the careful study of DiNardo and Lee (2004) who do not …nd any evidence for union wage premium in establishments that barely rejected union representation versus establishments that barely accepted it.7 1.5.3 Estimates by External Financial Dependence To provide more evidence on the potential link between …rm entry and union membership, I use data from Cetorelli and Strahan (2006) on external …nancial dependence for manufacturing sectors. Estimates of external …nancial dependence for each manufacturing sector are provided in Table 1.5. As in Cetorelli and Strahan (2006), I use two measures of …nancial dependence of manufacturing sectors. The …rst measure is the proportion of capital expenditures …nanced with external funds. A negative value indicates that …rms in the indicated sector have free cash ‡ow, whereas a positive value indicates that …rms must issue debt or equity to …nance their investment. A 7 See Lewis (1986) and Jarrell and Stanley (1990) for a review of the union wage premium literature. 35 second measure is the median ratio of loans to assets for small …rms from the Federal Reserve 1998 Survey of Small Business Finance. Larger values of the loans/assets ratio indicate greater dependence on external …nance. As depicted in Figure 1-9, the two measures of dependence on external …nance are positively correlated, with a correlation coe¢ cient of :51. Cetorelli and Strahan (2006) analyze the impact of interstate and branch bank deregulation on the number of establishments for sectors with di¤erent degrees of dependence on external …nance. They …nd an increase in the total number of establishments in sectors with more …nancial dependence. Importantly, their results hold only for interstate bank deregulation and not bank branch deregulation. Building on the work of Cetorelli and Strahan (2006), I estimate the impact of bank deregulation on union membership in manufacturing sectors, separately by the median external …nancial dependence.8 If unions decline because of …rm entry, then one would expect union membership to be negatively associated with interstate bank deregulation. Branch deregulation, on the other hand, should not have a signi…cant impact on union membership within the manufacturing sector. In this respect branch deregulation serves as a “placebo treatment”. Before analyzing the potential impact of both types of bank deregulation on union decline in sectors with di¤erent degrees of dependence on …nance, it is useful to check wether or not the workers in these sectors are similar in their characteristics. Speci…cally, my strategy would be called into question if, for example, sectors with 8 The median proportion of capital expenditures …nanced with external funds is 0. The median loans/assets ratio is :3. 36 relatively low dependence on external …nance have no union members. Table 1.6 reports mean values of workers’characteristics, separately by the median dependence on external …nance. In columns (1) and (2) I divide workers by the median proportion of capital expenditures …nanced with external funds, while in columns (4) and (5) workers are divided by the median loans–to–assets ratio. Columns (3) and (6) report the di¤erence between columns (2) and (1) and columns (5) and (4), respectively, and the accompanying standard errors. The results show that workers in sectors with above median proportion of capital expenditures …nanced by external funds are more educated than their counterparts below the median. There are no other signi…cant di¤erences with respect to union membership or coverage, potential experience in the workforce, weekly working hours, or full–time participation. Nor are there any di¤erences between workers in sectors below or above the median loans– to–assets ratio (column 6). The results in Table 1.7 show a signi…cant decline in union membership following bank deregulation in sectors with above median values of loans/assets ratio (panel A, column 4). Consistent with the …ndings of Cetorelli and Strahan (2006), union membership declines only following interstate deregulation and not branch deregulation. A point estimate of :018 means that union membership falls by 1:8 percentage points, which is about one–seventh of the standard deviation of union membership in manufacturing. The results also hold when controlling for collective bargaining and labor protection laws in panel B. In unreported regressions, I also con…rm all the results in Table 1.7 when conditioning on region–speci…c time …xed e¤ects. These results are available upon request. Somewhat surprisingly, there is no signi…cant change 37 in union membership in sectors above the median proportion of capital expenditures …nanced with external funds (column 2). Overall, the results in Table 1.7 provide additional evidence for the negative relationship between …rm entry and union membership. 1.5.4 Evidence from the Panel Study of Income Dynamics The critical assumption that drives the bank deregulation IV story is that bank deregulation in‡uences union membership only through its impact on competition. If bank deregulation in‡uences union membership for other reasons, my approach is called into question. It is therefore useful to consider other potential impacts of bank deregulation. First, as documented by Cetorelli and Strahan (2006), the average …rm size falls following deregulation. If unions primarily target large …rms, then a reduction in …rm size may explain the fall in union membership independently from the competition channel. Second, the number of new …rms in the economy has risen after bank deregulation (Black and Strahan (2002); Kerr and Nanda (2009)) changing the composition of workers and …rms. If new workers do not organize immediately to bargain collectively with employers, this might explain why union membership falls after bank deregulation, again, independently from the competitive channel. To provide more evidence on whether or not …rm entry reduces union membership by making the economy more competitive, I collect data from the Panel Study of Income Dynamics (PSID). The advantage of the PSID, among other things, is availability of exact tenure with current employer, which enables me to estimate the 38 impact of …rm entry on union membership among “mature”employees that have been with their current employer for long periods of time. This excludes new workers in existing …rms and workers in new …rms that may not be union members for reasons that have nothing to do with competitiveness of the economy. Table 1.8 presents estimates of the impact of bank branch deregulation on union coverage using sample of prime age (25 –54) white male heads of household from the “core”PSID sample who work for wage and salary. All estimates are Ordinary Least Squares and are weighted by sampling weights provided by the PSID. Speci…cally, Table 1.8 reports the estimate of unionist = from the following speci…cation: s + t + Dst + X0ist +"ist where unionist is union coverage indicator (0 time t, s and t (1.11) 1) of person i who resides in state s in are state and year …xed e¤ects, Dst is a dummy variable taking the value of unity in the post–branching period ( > t), and Xist is a vector of personal characteristics that includes years of completed education, tenure with the current employer, and tenure squared.9 Equation (1:11) is estimated at the individual level and not at the state–year level due to the relatively small sample. Equation (1:11) is a generalization of the di¤erence–in–di¤erences (DID) approach where the impact of deregulation is estimated as the di¤erence between the change in union coverage before and after deregulation with the di¤erence in union coverage 9 I use union coverage and not union membership because it is available more frequently in the PSID. 39 for a control group. In this speci…cation the control group is constructed from the average of all workers in the sample, rather than from a di¤erent set of workers not experiencing any change in the bank branching laws. To see this, note that E (unionis +k unionis j jDs +k = 1) E (unionis +k unionis j) = (1 where p is the fraction of the total sample that deregulated in a given year (which is small). The estimation of p) +k is subject to possibly severe serial correlation problem, which results in inconsistent standard errors (Moulton (1990)). Several factors make serial correlation an especially important issue in the context of DID estimation. First, equation (1:11) relies on a relatively long time period from 1977 to 1996. Second, union coverage is serially correlated and third, the bank branch deregulation indicator changes little within state over time. When estimating equation (1:11) I therefore cluster the standard errors at the state level. I also follow the non–parametric procedure of block–bootstrapping the standard errors, as suggested in Bertrand, Du‡o and Mullainathan (2004) and Angrist and Pischke (2009). I construct a bootstrap sample by drawing with replacement 49 matrices Vs , where Vs is the entire time series of observations for state s. I then run a regression of union coverage on bank deregulation dummy, state and year …xed e¤ects and workers’personal characteristics and obtain the estimate of . I draw a large number (200) of bootstrap samples and calculate the standard deviation of the resulting 200 estimates of . The results in Table 1.8 show a signi…cant reduction in union coverage following 40 bank branch deregulation. The reduction is only marginally signi…cant when using workers with all levels of tenure. The reduction becomes larger, however, with workers’tenure. This result is inconsistent with the argument that changes in union coverage are driven by changes in the composition of workers. The largest reduction in union coverage occurs among workers with at least 6 years of tenure with their current employer. The results hold with or without controlling for workers’personal characteristics and using alternative approaches to calculate the standard errors. Overall, the results in Table 1.8 reduce concerns that deregulation a¤ects union membership and coverage by changing the composition of the workforce. Table 1.9 provides some economic insight about the economic behavior of unions. By using the panel nature of the PSID, I am able to track workers over time and analyze changes in their outcomes as a result of changes in union coverage following bank branch deregulation. Speci…cally, I am following the same individuals for …ve years before and …ve years after bank branch deregulation. In addition to sample restrictions described at the beginning of the section, I restrict the sample to workers who reside in the same state during the 10 year period before and after deregulation. I then divide workers into three groups: workers who lost union coverage after bank branch deregulation in a state (panel A), workers who gained union coverage (panel B), and workers whose union coverage status hasn’t changed following bank deregulation (column C). The di¤erent columns of Table 1.9 represent di¤erent outcomes. These outcomes are: log real hourly wages in column (1), log weekly working hours in column (2), log annual working weeks in column (3), weekly working hours in column (4), annual 41 working weeks in column (5), and full-time full-year indicator which equals unity for those who report working at least 35 hours per week and at least 40 weeks per year. Workers who lost union coverage following bank deregulation report working 4:2% more hours per week and 2:6% more weeks per year, but do not experience any signi…cant changes in their wages or full-time, full-year participation. Workers who gained union coverage following deregulation, on the other hand, have an increase in wages of 18:9% without any changes in their employment patterns (working hours and weeks or full-time, full-year participation). As expected, there are no changes in any of the outcomes for workers who did not experience changes in union coverage. The results in Table 1.9 are consistent with a simple model of union behavior presented in Farber (1986). Union members seem to enjoy higher wages and work less hours per week and less weeks per year. The results are inconsistent, however, with the …ndings in DiNardo and Lee (2004) who employ regression discontinuity design to analyze wages and employment outcomes in establishment that barely rejected union representation and establishments that barely approved it. 1.5.5 Integrating the Competitive Explanation with Other Explanations of Union Decline Competition is not the only potential explanation for declining unions in the United States. Other, potentially more important explanations include the movement of workers from manufacturing to the service industry, large gains in employment in the South of the United States which has a long tradition of right–to–work laws, oil and 42 energy crises that signi…cantly hurt the automobile industry, and government provisions of better working conditions and laws against discrimination which lowered workers demand for unionism. A careful empirical analysis of the impact of competition on unions must take into the account the entire menu of potential causes of union decline. Figure 1-10 evaluates the relative importance of more competitive economy in the menu of other potential explanations for union decline. The solid line represents the actual trend in union membership. The solid line with connected full circles is based on residuals calculated from the regression of a union membership indicator of each worker on industry …xed e¤ect. This line, therefore, describes the hypothetical decline in union membership if there were no changes in industrial/occupational composition of workers over time. Similarly, the solid line with connected hollow circles is based on residuals calculated from the regression of union membership indicator of each worker on industry and state …xed e¤ect. This line describes the hypothetical decline in union membership if there were no changes in industrial/occupational and regional composition of workers over time. Finally, the dashed line is based on residuals calculated from the regression of union membership indicator of each worker on industry and state …xed e¤ect as well as a series of dummy variables for each year before and after bank branch deregulation. This line represents the hypothetical decline in union membership if there was no bank branch deregulation in the United States (and no changes in industrial/occupational and regional composition of workers). If bank deregulation a¤ects union membership only through its impact on competition, then the dashed line rep- 43 resents the hypothetical decline in union membership if the U.S. economy did not become more competitive over time. As can be seen, about two-thirds of the decline in union membership can be attributed to more competitive economic environment in the United States. 1.6 Conclusion Removal of geographical restrictions on bank branching in the United States provides a natural experiment for studying the e¤ect of competition among …rms in the non-…nancial sector on union membership. Bank branch deregulation helps entrepreneurship in the non-…nancial sector by fostering competition and consolidation in the banking sector. Using bank branch deregulation as an instrument for …rm entry, I …nd the resultant intensi…cation of competition materially reduces union membership among non–agricultural wage and salary male workers. After accounting for other potential reasons for union decline, competition between …rms seems to explain about three quarters of the overall decline in union membership in the U.S. since the late 1970s. The results in this paper provide support for the view that unions primarily target less competitive industries. Moreover, the results indicate that the decline in unionism following bank branch deregulation is associated with a reduction in union wage premium and with an increase in working hours among workers who lost union representation. This is consistent with the view that shocks to competition in the product market increase the elasticity of demand for labor and therefore increase the 44 trade-o¤ between employment and wages. There are several potential directions for future research. First, it would be interesting to study the impact of competition on actual voting for union representation. This will provide a deeper understanding of the mechanisms that underlie the competition–unionism relation. Second, this paper relates to a large literature that studies the causes of the recent rise in earnings inequality in the United States. One of the reasons emphasized in this literature is union decline (Card (1996b), Card (2001)). The results in my paper, however, indicate that competition plays a major role in explaining union decline. Future research should incorporate the rising competitiveness of the U.S. economy in the menu of potential reasons for changing earnings inequality. 1.7 Appendix A: Hazard Model Let T be a continuous random variable denoting years until bank deregulation, and let (t) = lim dt!o Pr (t T < t + dtjT > t) dt be the hazard function. Intuitively, (t) dt is the probability of deregulation in a short interval of length dt after t, conditional that the state is still regulated by time t. By 45 the law of conditional probability, lim Pr (t dt!o T < t + dtjT > t) Pr (t T < t + dt; T > t) = lim = dt!o dt dt Pr (T > t) Pr (t T < t + dt) fT (t) = lim = = dt!o dt Pr (T > t) 1 FT (t) fT (t) = F T (t) where fT (t) and F T (t) are the probability density function and the survivor function of T , respectively. Thus, the hazard function can be written as (t) = fT (t) = F T (t) dF T (t) =dt F T (t) This is a di¤erential equation in t whose solution (subject to the initial condition F T (0) = 1) is Z F T (t) = exp t (s) ds (1.12) 0 Since fT (t) = dF T (t) =dt, from equation (1:12) we get fT (t) = (t) exp Z t (s) ds (1.13) 0 Consider state i which did not deregulate by 1980. The probability of still being regulated h years after 1980 is Pr (T 1980 + h; T > 1980) = Pr (T > 1980) Pr (T 1980 + h) F T (1980 + h) = = Pr (T > 1980) F T (1980) 1980 + hjT > 1980) = Pr (T 46 The probability that state i deregulated between 1980 and 1980 + si (0 < si < h), is Pr (T = 1980 + si ; T > 1980) = Pr (T > 1980) Pr (T = 1980 + si ) fT (1980 + si ) = = Pr (T > 1980) F T (1980) Pr (T = 1980 + si jT > 1980) = Let NR be the number of states still regulated in 1980 + h, and let ND be the number of states that deregulated between 1980 and 1980 + si . Assuming independence of timing of deregulation between states, the likelihood function is given by L= NR Y i=1 F T (1980 + h) F T (1980) ND Y j=1 fT (1980 + sj ) F T (1980) (1.14) Substituting equations (1:12) and (1:13) into equation (1:14) gives the following expression for the likelihood function L = 8 o9 (s) ds = 0 n R o 1980 ; exp (s) ds 0 o9 n R 1980+sj = (s) ds (1980 + sj ) exp 0 n R o 1980 ; exp (s) ds 0 NR < exp Y i=1 : 8 ND < Y j=1 : Typically, the hazard function the way in which n R 1980+h (1.15) has two components. The …rst component describes shifts at any given point in time between states who have di¤erent characteristics and the second component is the time pro…le of deregulation which is common to all states. Following Lancaster (1979), I assume the following functional 47 form for the hazard function: (t) = 1 (union) 2 (t) where union is union membership in a given state and year. It is common to assume that: 1 (union) = exp f 0 + 1 uniong Assuming that the hazard function increases over time, the functional form for (1.16) 2 (t) can be expressed as: 2 1 (t) = t ; >1 (1.17) Equations (1:16) and (1:17) completely specify the hazard function , and thus the likelihood function L in equation (1:15). Notice that from equation (1:16) it follows that: log The coe¢ cient of interest is 1 1 = 0 + 1 union which stands for the percentage change in the hazard of bank deregulation as a result of a marginal change in union membership. Estimates of 1.1. 1 that maximize the likelihood function in equation (1:15) are reported in Table 48 1.8 Appendix B: Formation of Microdata Samples May and Outgoing Rotation Groups Current Population Survey (CPS) samples are obtained from the NBER data collection section and Unicon Research, respectively. These data can be obtained from and . I limit the CPS samples to prime age (25-54) white males, who work for wage and salary in the non-agricultural sector. I further limit the sample to individuals who either work, or have a job but currently not working. I exclude individuals with missing information on union membership, union coverage, and industry of employment. Finally, I drop individuals with missing or zero sampling weights. Following the literature on bank deregulation in the United States, I exclude Delaware and South Dakota due to large concentration of credit card banks in these states. The resulting CPS sample includes 1; 317; 387 observations. The …rst two columns in Table 1.13 provide more details on the sample restrictions imposed on the original CPS samples. When analyzing changes in the union wage premium following bank branch deregulation, I further limit the sample to individuals who report working at least forty hours per week (full-time workers), have hourly wages above $1:675 (1=2 of the minimum wage in 1982) and below the 99th percentile of year-speci…c distribution of real hourly wages of full-time workers. Hourly wages are adjusted to constant 1982 dollars using the Consumer Price Index. CPS measures of education are not standardized over time. Beginning with the 1992 CPS, education is no longer classi…ed according to years of schooling, but ac- 49 cording to the highest degree received classi…ed into categories (see Polivka (1996) for a review). I therefore classify the workers in the sample into …ve educational categories, according to highest degree received: 0 8, 9 11, 12, 13 15, and 16+. I calculate years of potential experience as the maximum between zero and age (in the survey year) minus years of schooling minus 6, where years of schooling are imputed for the post 1992 period using …gures in Park (1994). The Panel Study of Income Dynamics (PSID) sample is collected by the Institute for Social Research at the University of Michigan and is available for free download at . The full set of variable used in the construction of the PSID sample is listed in Tables 1.11 and 1.12. I restrict the PSID sample to prime age (25-54) white male heads of household from the “core” PSID sample of 1968 families, who are not self-employed and either work or have a job but currently not working. I further limit the sample to individuals with non-missing information on union coverage and non-missing state of residence. State of residence is not consistently reported in the PSID after 1993 and thus I do not use post 1993 data. Finally, I exclude individuals with missing tenure with the current employer and individuals who reside in Delaware and South Dakota. The resulting sample includes 18; 329 observations. The last column in Table 1.13 provides more details on the sample restrictions imposed on the original PSID sample. In the PSID, hourly earnings are reported separately for hourly and for salaried workers. In 1993, wages are not reported for salaried workers. In this year only, I approximate wages for salaried workers by dividing their total annual earnings (in the survey year) by total annual working hours in the year prior to the survey. When 50 analyzing wages, I limit the sample to individuals with hourly wages above $1:675 (1=2 of the minimum wage in 1982) and below the 99th percentile of year-speci…c distribution of real hourly wages. Hourly wages are adjusted to constant 1982 dollars using the Consumer Price Index. 51 Figure 1-1: Trends in Union Membership and Number of New Incorporations Per Capita in the United States 3 Union membership .25 2 .2 1 .15 0 .1 1965 1970 1975 1980 Union membership 1985 Year 1990 1995 2000 New incorporations per capita 2005 Log new incorporations (per 10,000) .3 52 Figure 1-2: Location of Bank Branches in Texas Before Branch Deregulation 53 Figure 1-3: Location of Bank Branches in Texas After Branch Deregulation 54 Figure 1-4: Pre-Existing Union Membership and Coverage and the Timing of Interstate Bank Deregulation Union Membership (b) Changes Year of interstate deregulation Year of interstate deregulation (a) Levels 2000 HI 1995 KS ND NE ARNM MS CO VT TX LA OK NH WY AL SC AZ NCDC FLGA VA ID TN MD UT 1990 1985 MT IA WV CA WIWA MOORMN PA IN NJ IL NV OH KY RI CT MA AK MI NY 1980 ME 0 .1 .2 .3 2000 HI 1995 MT KS ND IA NE AR NM CO MS WV VT AL OK TX LA WY WAWI CA NH PA OR IN AZ MIMN SC MOIL NJ TN NC FLID VA DC GAOH MD NV UT KY RI CT MA NY 1990 1985 AK 1980 ME .4 -.03 -.02 Union membership -.01 0 Changes in union membership Union Coverage (d) Changes Year of interstate deregulation Year of interstate deregulation (c) Levels 2000 HI 1995 1990 SC NC 1985 MT KS ND IA NE AR NM MS COVT WV TX LA OK NH WY AL CA WI WA AZ MO OR MN IN PA IL NJ MI DC VA FL GA IDTN MD NV OH UT KY RI CTMA AK NY 1980 ME 0 .1 .2 Union coverage .3 .4 2000 HI 1995 1990 1985 UT MT KS ND IA NE AR NM CO WV MS VT AL TX OK LA WA WY WI CANH PA ORIN MIMN AZSC MO IL NJ TN NC IDFL VA GDC A OH MD NV KY RI CT MA NY AK 1980 ME -.03 -.02 -.01 Changes in union coverage 0 55 Figure 1-5: The Impact of Bank Branch Deregulation on Log New Incorporations Per Capita Percentage change in new incorporations per capita .4 .2 0 -.2 -.4 -10 -5 0 5 Years relative to deregulation 10 15 56 Figure 1-6: The Impact of Bank Branch Deregulation on Union Membership Percentage change in union membership .06 .04 .02 0 -.02 -.04 -.06 -10 -5 0 5 Years relative to deregulation 10 15 57 Figure 1-7: Changes in New Incorporations Per Capita and Union Membership by States IA AR MS KY MN WI MO CO TN AL TX IN MT LA MI GA FL IL KS WY NE NM ND WV OK PA UT MA NH CT OH VA OR HI WA -.1 0 .1 Percentage change per year after bank deregulation Union membership (x4) Number of new incorporations per capita .2 58 Figure 1-8: The Location of Unionized Workers in the Non-Unionized Wage Distribution Before and After Bank Branch Deregulation 100 Percentile of log real hourly wages among non-unionized workers 90 80 70 60 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100 Percentile of log real hourly wages among unionized workers Before After 59 Figure 1-9: Correlation Between Measures of External Financial Dependence Among Manufacturing Sectors Median Loans/Assets for 1998 SSBF Firms .6 Petroleum and coal products .5 Lumber and wood products Textile mill products .4 Paper and allied products Furniture and fixtures Printing and publishingChemicals and allied products Miscellaneous Primary manufacturing metal industries Rubber and plastic products Instruments and related products Stone, clay, glass, and concrete products Food and kindred products .3 Industrial machinery and equipment .2 Apparel and other textile Fabricated metal products Electrical and electronic equipment .1 Transportation equipment Leather and leather products 0 -1 -.75 -.5 -.25 0 .25 .5 External Financial Dependence for Mature COMPUSTAT Firms .75 1 60 Figure 1-10: Explaining the Decline in Unions Union membership (1978=100) 100 95 90 85 80 1975 1980 1985 1990 1995 2000 2005 Year Actual Union Membership With Industry Fixed Effect With Industry and State Fixed Effect With Industry and State Fixed Effects and Bank Deregulation 2010 61 Table 1.1: Timing of Bank Branch Deregulation and Pre-Existing Union Membership and Coverage: The Duration Model All Industries (1) Union membership (2) (3) Manufacturing Only (4) -.710 .032 (.395) (.174) -.606 -.220 (.716) (.383) Change in union coverage 270 (7) (8) (.409) -.227 Union coverage Number of observations (6) -.165 (.747) Change in union membership (5) 270 270 -.086 -.057 (.339) (.206) 270 270 269 270 269 Note - The model is a Weibul hazard model where the dependent variable is the log expected time to bank branch deregulation. The hazard of deregulation is a likelihood that a state deregulates at time t, given that the state has not yet deregulated. Each coefficient measures the percentage change in the hazard of deregulation as a result of a marginal change in either the level of union membership and coverage or changes in union membership and coverage. Standard errors are adjusted for state-level clustering and appear in parentheses. Union membership is the percentage of nonagricultural wage and salary employees who are union members. Union coverage is the percentage of nonagricultural wage and salary employees who are covered by a collective bargaining. In columns (1) – (4), union membership and coverage are averaged to the state-year level using workers in all industries. In columns (5) – (8), union membership and coverage are averaged to the state-year level using only workers in manufacturing. All specifications control for political economy variables that affect the timing of bank branch deregulation (Kroszner and Strahan, 1999). These variables are: (1) small bank share of all banking assets, (2) capital ratio of small banks relative to large, (3) relative size of insurance in states where banks may sell insurance, (4) an indicator which takes upon a value of one if banks may sell insurance, (5) relative size of insurance in states where banks may not sell insurance, (6) small firm share, (7) share of state government controlled by Democrats, (8) an indicator which takes upon a value of one if a state is controlled by one party, (9) average yield on bank loans minus Fed funds rate, (10) an indicator which takes upon a value of one if state has unit banking law, and (11) an indicator which takes upon a value of one if state changes bank insurance powers. Sample period is 1978 to 1994, excluding 1982, and the sample comprises 36 states that deregulated after 1978. States drop from the sample once they deregulate. 62 Table 1.2: Wald Estimate of the Impact of Log New Incorporations Per Capita on Union Membership (1) (2) (3) Before Bank Deregulation After Bank Deregulation Difference (2) - (1) Union membership .195 .179 -.016** (.007) Log new incorporations per capita 1.023 1.105 .082*** (.030) Wald estimate of the impact of log new incorporations per capita on union membership OLS estimate of the impact of log new incorporations per capita on union membership -.194* (.113) -.007 (.009) Note – The sample size is 1,372 state – year observations and consists of 49 states between the years 1978 and 2006, excluding 1982. Delaware and South Dakota are excluded because of large concentration of credit card bank in these states. The year 1982 is excluded because union status questions were not asked in this year. Average union membership for each state and year is calculated using May Current Population Survey files for the years 1978-1981 and Outgoing Rotation Groups files for the years 1983-2006. The underlying sample includes prime age (25-54) white men who work for wage and salary, excluding those who work in agriculture. Specifically, union membership in each state and year is the average residual from an OLS regression of union membership indicator on five dummies of years of completed education (0-8, 9-11, 12, 13-15, and 16+), potential experience and its square, and industry fixed effects, pooling all years together. I use CPS sampling weights when calculating the residuals and averaging them to the state – year level. The number of new incorporations is from Dun and Bradstreet . New incorporations are divided by population estimates from the U.S. Census Bureau. Timing of bank branch deregulation for each state is from Kroszner and Strahan (1999). All specifications in column (3) control for state and year fixed effects. Wald estimate of the impact of log new incorporations per capita on union membership uses bank branch deregulation indicator as an instrument for log new incorporations. Branch deregulation indicator equals one during all years in which a state permits in – state branching. Standard errors are clustered at the state level and appear in parentheses. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent, respectively. 63 Table 1.3: The Impact of Log New Incorporations Per Capita on Union Membership: OLS, IV, and TSLS Estimates Specification of Instrument(s) OLS (1) Dummy (2) Linear (3) Quadratic (4) Nonparam. (5) Panel A: Without Controlling for Labor Laws Log new incorporations per capita -.007 -.194* -.191 -.164* -.150* (.009) (.113) (.207) (.092) (.084) 7.61 [.008] 2.28 [.138] 7.67 [.008] 9.66 [.003] 1,372 <.746> 1,372 <.957> 1,372 F-statistic of excluded instruments [p-value] Number of observations 1,372 1,372 Log new incorporations per capita -.013 -.199** -.259 -.141** -.124** (.011) (.098) (.245) (.065) (.056) 8.08 [.007] 1.37 [.248] 8.16 [.006] 10.89 [.002] <.271> <.845> 1,284 1,284 1,284 1,284 Panel B: Controlling for Labor Laws F-statistic of excluded instruments [p-value] Number of observations 1,284 Note – The sample size in panel A is 1,372 state–year observations and consists of 49 states between the years 1978 and 2006, excluding 1982. Delaware and South Dakota are excluded because of large concentration of credit card bank in these states. The year 1982 is excluded because union status questions were not asked in this year. Average union membership for each state and year is calculated using May Current Population Survey files for the years 1978-1981 and Outgoing Rotation Groups files for the years 1983-2006. The underlying sample includes prime age (25-54) white men who work for wage and salary, excluding those who work in agriculture. Specifically, union membership in each state and year is the average residual from an OLS regression of union membership indicator on five dummies of years of completed education (0-8, 9-11, 12, 13-15, and 16+), potential experience and its square, and industry fixed effects, pooling all years together. I use CPS sampling weights when calculating the residuals and averaging them to the state– year level. Column (1) reports OLS estimate of the impact of log new incorporations per capita on average union membership in a state. In column (2), I use bank branch deregulation indicator as an instrumental variable for log new incorporations per capita. Branch deregulation indicator equals one during all years in which a state permits in–state branching. In column (3), I use years since branch deregulation as an instrumental variable. In column (4), I use years since branch deregulation and its square as instruments. Finally, in column (5) I use a series of dummy variables for each year before and after deregulation. All specifications include state and year fixed effects. In panel B, I also control for collective bargaining laws obtained from Valetta and Freeman (1988) and later updated by NBER until 1996. Specifically, I include (a) an indicator which equals one if a state permits collective bargaining and equals zero otherwise, and (b) an indicator which equals one if a state permits striking and equals zero otherwise. All specifications in panel B also include three indicators for presence of laws that limit the ability of employers to fire workers. Presence of the laws by states was obtained from Autor, Donohue III, and Schwab (2006). These laws are: (a) “public policy” provides employees with protections against discharges that would prevent an important public policy, such as performing jury duty; (b) “good-faith” prevents employers from firing workers for bad cause such as just before a substantial commission is due; and (c) “implied contract” protection comes into force when an employer implicitly promises not to terminate a worker without good cause. Standard errors are clustered at the state level and appear in parentheses. * and ** indicate statistical significance at the 10 and 5 percent, respectively. 64 Table 1.4: The Impact of Log New Incorporations Per Capita on Union Membership: OLS, IV, and TSLS Estimates with Region - Speci…c Time Fixed E¤ects Specification of Instrument(s) OLS (1) Dummy (2) Linear (3) Quadratic (4) Nonparam. (5) Panel A: Without Controlling for Labor Laws Log new incorporations per capita -.020** -.276 -.133 -.214* -.187** (.009) (.180) (.255) (.117) (.094) 4.80 [.033] 1.78 [.188] 5.14 [.028] 8.07 [.007] 1,372 1,372 <.513> 1,372 <.678> 1,372 F-statistic of excluded instruments [p-value] Number of observations 1,372 Log new incorporations per capita -.017 -.229 -.049 -.194 -.138* (.011) (.139) (.204) (.120) (.072) 4.68 [.036] 1.54 [.221] 3.49 [.068] 8.27 [.006] <.175> <.469> 1,284 1,284 1,284 1,284 Panel B: Controlling for Labor Laws F-statistic of excluded instruments [p-value] Number of observations 1,284 Note – The sample size in panel A is 1,372 state–year observations and consists of 49 states between the years 1978 and 2006, excluding 1982. Delaware and South Dakota are excluded because of large concentration of credit card bank in these states. The year 1982 is excluded because union status questions were not asked in this year. Average union membership for each state and year is calculated using May Current Population Survey files for the years 1978-1981 and Outgoing Rotation Groups files for the years 1983-2006. The underlying sample includes prime age (25-54) white men who work for wage and salary, excluding those who work in agriculture. Specifically, union membership in each state and year is the average residual from an OLS regression of union membership indicator on five dummies of years of completed education (0-8, 9-11, 12, 13-15, and 16+), potential experience and its square, and industry fixed effects, pooling all years together. I use CPS sampling weights when calculating the residuals and averaging them to the state– year level. Column (1) reports OLS estimate of the impact of log new incorporations per capita on average union membership in a state. In column (2), I use bank branch deregulation indicator as an instrumental variable for log new incorporations per capita. Branch deregulation indicator equals one during all years in which a state permits in–state branching. In column (3), I use years since branch deregulation as an instrumental variable. In column (4), I use years since branch deregulation and its square as instruments. Finally, in column (5) I use a series of dummy variables for each year before and after deregulation. All specifications include state and year fixed effects as well as region – specific year fixed effects. In panel B, I also control for collective bargaining laws obtained from Valetta and Freeman (1988) and later updated by NBER until 1996. Specifically, I include (a) an indicator which equals one if a state permits collective bargaining and equals zero otherwise, and (b) an indicator which equals one if a state permits striking and equals zero otherwise. All specifications in panel B also include three indicators for presence of laws that limit the ability of employers to fire workers. Presence of the laws by states was obtained from Autor, Donohue III, and Schwab (2006). These laws are: (a) “public policy” provides employees with protections against discharges that would prevent an important public policy, such as performing jury duty; (b) “good-faith” prevents employers from firing workers for bad cause such as just before a substantial commission is due; and (c) “implied contract” protection comes into force when an employer implicitly promises not to terminate a worker without good cause. Standard errors are clustered at the state level and appear in parentheses. * and ** indicate statistical significance at the 10 and 5 percent, respectively. 65 Table 1.5: External Financial Dependence for Manufacturing Sectors Median Loans/Assets For 1998 Survey of Small Business Finance Firms External Financial Dependence for Mature COMPUSTAT Firms Leather and leather products 0.04 -0.96 Tobacco manufactures Apparel and other textile N/A 0.13 -0.92 -0.61 Fabricated metal products Food and kindred products 0.12 0.27 -0.24 -0.24 Furniture and fixtures Miscellaneous manufacturing 0.36 0.31 -0.23 -0.20 Stone, clay, glass, and concrete products Printing and publishing 0.28 0.33 -0.20 -0.07 Instruments and related products Transportation equipment 0.29 0.06 -0.04 0.01 Industrial machinery and equipment Primary metal industries 0.21 0.31 0.01 0.03 Lumber and wood products Rubber and plastic products 0.49 0.30 0.04 0.04 Paper and allied products Petroleum and coal products 0.37 0.60 0.06 0.09 Textile mill products Electrical and electronic equipment 0.47 0.14 0.10 0.22 Chemicals and allied products 0.33 0.28 Sector Note – External financial dependence equals the proportion of capital expenditures financed with external funds. A negative value indicates that firms have free cash flow, whereas a positive value indicates that firms must issue debt or equity to finance their investment. The figures represent the median value for COMPUSTAT firms (that have been on COMPUSTAT for at least 10 years) in each sector over the 1980 to 1997 period. The loans/assets ratio is the median ratio of loans to assets for small firms from the Federal Reserve 1998 Survey of Small Business Finance. Source – Cetorelli and Strahan (2006). 66 Table 1.6: Characteristics of Workers by External Financial Dependence Union member Covered by a union Proportion of Capital Median Loans/Assets Expenditures Financed With External Funds For 1998 Survey of Small Business Finance Firms (1) Below (2) Above (3) Difference (4) Below (5) Above (6) Difference Median Median (2) - (1) Median Median (5) - (4) .333 .358 .025 .318 .346 .028 .387 (.032) .039 .346 .362 (.032) .016 .224 .257 (.030) .033 .348 Education < 12 years .301 .221 (.033) -.080*** Education = 12 years .383 .402 (.029) .019 .403 .400 (.026) -.003 Education > 12 years .316 .377 (.035) .061** .373 .343 (.029) -.030 Potential experience 19.9 19.7 (.028) -.186 19.6 19.7 (.032) .151 Weekly working hours 38.7 40.1 (.622) 1.48 40.3 39.5 (.604) -.788 Working full-time .897 .910 (1.40) .014 .912 .902 (1.389) -.010 (.016) (.014) Note – The table reports mean characteristics of workers in manufacturing sectors with different degrees of external financial dependence. All data are for 1978, before bank deregulation in most states. Columns (1) and (2) divide workers by the median proportion of capital expenditures financed with external funds using data on mature COMPUSTAT firms. Column (3) reports the differences between columns (2) and (1) and the accompanying standard errors. Column (4) and (5) divide workers by the median loans-to-assets ratio for 1998 Small of Small Business Finance firms. Column (6) reports the differences between columns (5) and (4) and the accompanying standard errors. Standard errors are clustered at the state level and appear in parentheses. ** and *** indicate statistical significance at the 5 and 1 percent, respectively. (2) (1) -.011 1,283 .72 -.011 (.011) -.015 (.015) 1,281 .69 -.003 (.014) -.013 (.014) Panel B: Controlling for Labor Laws .67 1,364 (.015) (.015) -.004 1,279 .70 -.019** (.009) .000 (.014) .68 1,366 (.009) (.015) -.018* .006 (4) Above Median Note – The sample is at the state – year level and consists of 49 states between the years 1978 and 2006, excluding 1982. Delaware and South Dakota are excluded because of large concentration of credit card bank in these states. The year 1982 is excluded because union status questions were not asked in this year. Average union membership for each state and year is calculated using May Current Population Survey files for the years 1978-1981 and Outgoing Rotation Groups files for the years 1983-2006. The underlying sample includes prime age (25-54) white men who work for wage and salary in manufacturing. Union membership in each state and year is the average residual from an OLS regression of union membership indicator on five dummies of years of completed education (0-8, 9-11, 12, 13-15, and 16+) and potential experience and its square, pooling all years together. I use CPS sampling weights when calculating the residuals and averaging them to the state–year level. I calculate the residuals separately by the median external financial dependence. In columns (1) and (2), workers are divided by the median external financial dependence for mature COMPUSTAT firms. In columns (3) and (4), workers are divided by the median loans-to-assets ratio for 1998 SSBF firms. All specifications include state and year fixed effects. In panel B, I also control for collective bargaining laws obtained from Valetta and Freeman (1988) and later updated by NBER until 1996. Specifically, I include (a) an indicator which equals one if a state permits collective bargaining and equals zero otherwise, and (b) an indicator which equals one if a state permits striking and equals zero otherwise. All specifications in panel B also include three indicators for presence of laws that limit the ability of employers to fire workers. Presence of the laws by states was obtained from Autor, Donohue III, and Schwab (2006). These laws are: (a) “public policy” provides employees with protections against discharges that would prevent an important public policy, such as performing jury duty; (b) “goodfaith” prevents employers from firing workers for bad cause such as just before a substantial commission is due; and (c) “implied contract” protection comes into force when an employer implicitly promises not to terminate a worker without good cause. Branch deregulation indicator equals one during all years in which a state permits in–state branching. Interstate banking deregulation indicator equals one during all years in which a state permits out-of-state banking companies to buy banks headquartered in the state. Standard errors are clustered at the state level and appear in parentheses. * and ** indicate statistical significance at the 10 and 5 percent, respectively. .59 1,284 Number of observations .002 (.015) Interstate deregulation R2 -.002 (.016) Branch deregulation Number of observations .70 1,365 (.012) (.014) Interstate deregulation .60 1,371 -.005 (.016) -.014 -.002 (.016) .002 Branch deregulation R2 (3) Below Median Panel A: Without Controlling for Labor Laws Above Median Below Median Median Loans/Assets for 1998 Survey of Small Business Finance Firms Proportion of Capital Expenditures Financed With External Funds Table 1.7: The Impact of Bank Deregulation on Union Membership in Manufacturing Sectors By External Financial Dependence 67 68 Table 1.8: The Impact of Bank Deregulation on Union Coverage: Evidence from the Panel Study of Income Dynamics No Covariates With Covariates Tenure Tenure All 3+ 6+ All 3+ 6+ (1) (2) (3) (4) (5) (6) Branch deregulation (clustered s.e.s) -.028 (.018) -.039 (.019)** -.061 (.022)*** -.029 (.018) -.042 (.019)** -.065 (.021)*** [block – bootstrapped s.e.s] [.015]* [.015]** [.020]*** [.016]* [.016]** [.020]*** R2 Number of observations .07 .08 .09 .12 .11 .12 18,329 12,803 9,297 18,329 12,803 9,297 Note – The dependent variable is union coverage indicator. The sample is at the worker level and consists of respondents to Panel Study of Income Dynamics (PSID) surveys in the years 1977 – 1993. The sample is restricted to prime age (25 – 54) white male heads of household from the “core” PSID sample who work for wage and salary. All estimates are Ordinary Least Squares and are weighted by sampling weights provided by the PSID. All specifications include state and year fixed effects. Specifications in columns (4) – (6) also control for years of completed education, tenure, and tenure squared. Branch deregulation indicator equals one during all years in which a state permits in–state branching. In parentheses I report standard errors which are clustered at the state level. In brackets, I report block – bootstrapped standard errors. I construct a bootstrap sample by drawing with replacement 49 matrices Vs, where Vs is the entire time series of observations for state s. I then run a regression of union coverage on bank deregulation dummy, state and year fixed effects and workers’ personal characteristics (columns 4-6) and obtain the estimated impact of bank deregulation on union coverage indicator. I draw a large number (200) of bootstrap samples and calculate the standard deviation of the resulting 200 estimates of the impact of bank deregulation on union coverage indicator. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent, respectively. 69 Table 1.9: Changes in Workers’Outcomes as a Result of Changes in Union Coverage: Estimates from a Panel of Workers Log Log Log Wages Weekly Hours Annual Weeks Weekly Hours Annual Weeks Full-Time Full-Year Working (1) (2) (3) (4) (5) (6) Panel A: Workers Who Lost Union Coverage After Bank Deregulation Branch deregulation (clustered s.e.s) .017 (.036) .042 (.022)* .026 (.018) 1.642 (.914)* 1.468 (.695)** .034 (.042) [block-bootstrapped s.e.s] [.032] [.017]** [.015]* [.706]** [.569]** [.037] R2 Number of observations .48 813 .15 813 .18 813 .14 873 .19 873 .21 873 Branch deregulation .189 -.020 .031 -1.171 1.488 .053 (clustered s.e.s) [block-bootstrapped s.e.s] (.067)*** [.059]*** (.029) [.027] (.047) [.043] (1.078) [.897] (1.254) [1.222] (.058) [.055] R2 Number of observations .51 .26 .09 .26 .14 .18 562 562 562 624 624 624 Branch deregulation .038 .006 .006 .116 .213 .013 (.025) [.023] (.008) [.007] (.006) [.005] (.358) [.302] (.276) [.216] (.010) [.009] Panel B: Workers Who Gained Union Coverage After Bank Deregulation Panel C: Workers Whose Union Coverage Did Not Change (clustered s.e.s) [block-bootstrapped s.e.s] R2 Number of observations .26 .05 .05 .06 .08 .04 6,772 6,772 6,772 7,691 7,691 7,691 Note – The dependent variables are: log real hourly wages in column (1), log weekly working hours in column (2), log annual working weeks in column (3), weekly working hours in column (4), annual working weeks in column (5), and full-time full-year indicator in column (6). Full-time full-year workers are those who report working at least 35 hours per week and at least 40 weeks per year. The sample is at the worker level and consists of respondents Panel Study of Income Dynamics (PSID) respondents in years 1977 – 1993. The sample is restricted to prime age (25 – 54) white male heads of household from the “core” PSID sample who work for wage and salary. Additionally, I limit the sample to workers whom I can trace for five years before bank branch deregulation and five years after deregulation in their state. In panel A, the sample is restricted to workers who lost union coverage after bank deregulation. In panel B, the sample is restricted to workers who gained union coverage after bank deregulation. In panel C, the sample is restricted to workers whose union coverage status did not change after bank deregulation. All estimates are Ordinary Least Squares and are weighted by sampling weights provided by the PSID. All specifications include completed years of education, tenure, tenure squared, and state and year fixed effects. Branch deregulation indicator equals one during all years in which a state permits in– state branching. In parentheses I report standard errors which are clustered at the state level. In brackets, I report block – bootstrapped standard errors. I construct a bootstrap sample by drawing with replacement 49 matrices Vs, where Vs is the entire time series of observations for state s. I then run a regression of union membership on bank deregulation dummy, state and year fixed effects and workers’ personal characteristics (columns 4-6) and obtain the estimated impact of bank deregulation on union membership indicator. I draw a large number (200) of bootstrap samples and calculate the standard deviation of the resulting 200 estimates of bank deregulation on union membership. *, **, and *** indicate statistical significance at the 10, 5, and 1 percent, respectively. AR CA CO CT DC FL GA HI ID IL IN IA KS KY LA ME MD MA MI MN MS MO Arkansas California Colorado Connecticut District of Columbia Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri 1990 1986 1987 1993 1960 1984 1988 1975 1987 1990 1989 1999 1960 1988 1983 1986 1960 1988 1991 1980 1994 1960 1960 1960 1981 1986 1988 1986 1986 1985 1983 1987 1978 1992 1984 1986 1991 1985 1986 1985 1997 1985 1985 1988 1983 1989 1987 1982 1986 1987 Inter-state Type of deregulation: Branch Wyoming West Virginia Wisconsin Virginia Washington Utah Vermont Tennessee Texas Rhode Island South Carolina Oregon Pennsylvania Ohio Oklahoma North Carolina North Dakota New Mexico New York New Hampshire New Jersey Nebraska Nevada Montana State Note - Dates of branch and interstate deregulation are taken from Kroszner and Strahan (1999). AL AK AZ Alaska Arizona State code Alabama State WY WV WI VA WA UT VT TN TX RI SC OR PA OH OK NC ND NM NY NH NJ NE NV MT State code 1988 1987 1990 1978 1985 1981 1970 1985 1988 1960 1960 1985 1982 1979 1988 1960 1987 1991 1976 1987 1977 1985 1960 1990 Branch 1987 1988 1987 1985 1987 1984 1988 1985 1987 1984 1986 1986 1986 1985 1987 1985 1991 1989 1982 1987 1986 1990 1985 1993 Inter-state Type of deregulation: Table 1.10: Timing of Branch and Interstate Bank Deregulation 70 Household ER30219 ER30248 ER30285 ER30315 ER30345 ER30375 ER30401 ER30431 ER30465 ER30500 ER30537 ER30572 ER30608 ER30644 ER30691 ER30735 ER30808 Year 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 Head of V22406 V20651 V19349 V18049 V16631 V15130 V14114 V13011 V11606 V10419 V8961 V8352 V7658 V7067 V6462 V5850 V5350 Age V22407 V20652 V19350 V18050 V16632 V15131 V14115 V13012 V11607 V10420 V8962 V8353 V7659 V7068 V6463 V5851 V5351 Gender V23276 V21420 V20114 V18814 V17483 V16086 V14612 V13565 V11938 V11055 V9408 V8723 V8099 V7447 V6802 V6209 V5662 Ethnicity V22489 V20720 V19420 V18120 V16682 V15181 V14166 V13068 V11668 V10519 V9010 V8379 V7711 V7102 V6499 V5941 V5384 Tenure V23333 V21504 V20198 V18898 V17545 V16161 V14687 V13640 V12400 V11042 V9395 V8709 V8085 V7433 V6787 V6194 V5647 Education V22448 V20693 V19393 V18093 V16655 V15154 V14146 V13046 V11637 V10453 V9005 V8374 V7706 V7095 V6492 V5872 V5373 Status Empl. V22451 V20696 V19396 V18096 V16658 V15157 V14149 V13049 V11640 V10456 V9006 V8375 V7707 V7096 V6493 V5875 V5376 Employed Self Table 1.11: Variables Used for Analyses of Panel Study of Income Dynamics Union V22454 V20699 V19399 V18099 V16661 V15160 V14152 V13052 V11649 V10458 V9008 V8377 V7709 V7098 V6495 V5877 V5382 Coverage 71 V12503 V13703 V14803 V16303 V17703 V19003 V20303 V21603 1986 1987 1988 1989 1990 1991 1992 1993 V8203 1982 V11103 V7503 1981 1985 V6903 1980 V8803 V6303 1979 V10003 V5703 1978 1984 V5203 1977 1983 Residence Year State of ER30864 ER30803 ER30730 ER30686 ER30641 ER30605 ER30569 ER30534 ER30497 ER30462 ER30428 ER30398 ER30372 ER30342 ER30312 ER30282 ER30245 Weight Sampling V22577 V20797 V19497 V18197 V16759 V15258 V14202 V13106 V11706 V10562 V9035 V8404 V7742 V7119 V6516 V5905 V5418 (last year) Week Hours per V22575 V20796 V19496 V18196 V16758 V15257 V14203 V13105 V11705 V10561 V9034 V8403 V7741 V7118 V6515 V5904 V5417 (last year) Year Weeks per V21634 V20344 V19044 V17744 V16335 V14835 V13745 V12545 V11146 V10037 V8830 V8228 V7530 V6934 V6336 V5731 V5232 (last year) Hours Annual V22463 V20703 V19403 V18103 V16665 V15164 V14156 V13056 V11653 V10462 V9013 V8382 V7714 V7121 V6518 V5907 V5420 Hour or Paid by Salaried V22470 V20707 V19407 V18107 V16669 V15168 V14160 V13060 V11657 V10466 V9017 V8386 V7718 V7125 V6522 V5911 V5424 Hourly If Paid … V20704 V19404 V18104 V16666 V15165 V14157 V13057 V11654 V10463 V9014 V8383 V7715 V7122 V6519 V5908 V5421 Salaried If Hourly Earnings V22464 … … … … … … … … … … … … … … … … (current) Earnings Annual Table 1.12: Variables Used for Analyses of Panel Study of Income Dynamics (cont.) 72 73 Table 1.13: Sample Restrictions Imposed on Microdata Samples Current Population Survey May Supplement Outgoing Rotation Groups Panel Study of Income Dynamics (1978-1981) (1983-2006) (1977-1993) 486,668 11,001,341 671,390 Sample restrictions (observations deleted): Head of household … … (543,207) Prime-age (25-54) in the year of the survey (253,284) (6,763,618) (45,573) Male (121,456) (2,195,285) (20,866) White (12,098) (280,646) (20,764) Works for wages and salary, not in agriculture* (18,518) (332,373) (9,594) Either working, or with job but not at work (3,170) (61,861) (147) Non-missing union membership and coverage** (19,326) (76,041) (417) Non-missing industry (10) (0) (0) Non-missing state of residence (0) (0) (197) Non-missing tenure … … (152) (1,194) (31,739) (114) (0) (3) (12,030) 1,259,775 18,329 Total number of observations in the raw data Not residing in Delaware or South Dakota Non-missing and positive sampling weight*** Total number of observations that satisfy sample restrictions above 57,612 1,317,387 Note – In the PSID the restrictions are different in that: * restricts to not self-employed; ** only restricts to non-missing union coverage; and *** restricts to the “core” sample of 1968 families. Chapter 2 Big Bad Banks? The Winners and Losers from Bank Deregulation in the United States 2.1 Introduction Although income distributional considerations have played a central –if not the central – role in shaping the policies that govern …nancial markets, researchers have devoted few resources toward evaluating the actual impact of …nancial regulations on the distribution of income. For instance, Thomas Je¤erson’s fears that concentrated banking power would help wealthy industrialists at the expense of others spurred him to …ght against the Bank of the United States, and similar anxieties fueled Andrew Jackson’s veto of the re-chartering of the Second Bank of the United States (Hammond (1957); Bodenhorn (2003)). During the 20th century, politicians in many U.S. 74 75 states implemented and maintained restrictions on bank branching, arguing that the formation of large banks would disproportionately curtail the economic opportunities of the poor (Southworth (1928); White (1982); Kroszner and Strahan (1999)). Today, in the midst of a …nancial crisis, unease about centralized economic power and growing income inequality have fueled debates about Gramm-Leach-Bliley and the desirability of sti¤er regulations on …nancial conglomerates.1 While beliefs about the in‡uence of …nancial regulations on the distribution of income a¤ect policies, we econometrically evaluate the causal impact of bank regulations on income distribution. Theoretical debates and welfare concerns further motivate our analyses of the distributional e¤ects of bank regulation. If banking is a natural monopoly, then unregulated, monopolistic banks may earn rents through high …xed fees that disproportionately hurt the poor as developed in models by Greenwood and Jovanovic (1990), Banerjee and Newman (1993), and Galor and Zeira (1993). Countervailing arguments, however, challenge the view that restricting the consolidation and expansion of banks will help the poor. Regulatory restrictions on bank mergers, acquisitions, and branching could create and protect local banking monopolies, curtailing competition and raising fees that primarily hurt the poor. In light of this debate, we evaluate whether regulatory restrictions on bank expansion increased, decreased, or had no e¤ect on income inequality. Furthermore, bank regulations could directly a¤ect social welfare by altering the distribution of income. As summarized by Kah1 On compensation in the …nancial sector and income inequality in the overall economy, see Philippon and Reshef (2009) and Kaplan and Rauh (2009). Many have recently suggested that …nancial deregulation, including bank branch deregulation, the Gramm-Leach-Bliley Act, and the accommodating supervisory approach at the Federal Reserve, contributed both to the …nancial crisis and to growing income inequality, e.g., Krugman (2009) and Moss (2009). 76 neman and Krueger (2006) and Clark, Frijters and Shields (2008), people care about relative income, as well as absolute income and risk. Thus, understanding the impact of …nancial regulations on the distribution of income provides additional information on the welfare implications of …nance. More speci…cally, this paper assesses how branch deregulation altered the distribution of income in the United States. From the 1970s through the 1990s, most states removed restrictions on intrastate branching, which intensi…ed bank competition and improved bank e¢ ciency and performance (Flannery (1984); Jayaratne and Strahan (1998)). Researchers have examined the impact of these reforms on economic growth (Jayaratne and Strahan (1996); Huang (2008)), entrepreneurship (Black and Strahan (2002); Kerr and Nanda (2009)), economic volatility (Morgan et al. (2004); Demyanyk et al. (2007)), and the wage gap between men and women bank executives (Black and Strahan (2001)). In this paper, we provide the …rst evaluation of the impact of branch deregulation on the distribution of income in the overall economy and help resolve a debate about bank deregulation that extends across two centuries. The removal of branching restrictions by states provides a natural setting for identifying and assessing who won and lost from bank deregulation. As shown by Kroszner and Strahan (1999), national technological innovations triggered branch deregulation at the state level. Speci…cally, (i) the invention of automatic teller machines (ATMs), in conjunction with court rulings that ATMs are not bank branches, weakened the geographical bond between customers and banks; (ii) checkable money market mutual funds facilitated banking by mail and telephone, which weakened local bank monopolies; and, (iii) improvements in communications technology lowered the 77 costs of using distant banks. These innovations reduced the monopoly power of local banks, weakening their ability and desire to …ght against deregulation. Kroszner and Strahan (1999) further show that cross-state variation in the timing of deregulation re‡ects the interactions of these national technological innovations with preexisting state-speci…c conditions. For example, deregulation occurred later in states where politically powerful groups viewed large, multiple-branch banks as potentially serious competitors. Moreover, as we demonstrate below, neither the level nor rate of change in the distribution of income before deregulation helps predict when a state removed restrictions on bank branching, suggesting that the timing of branch deregulation at the state level is exogenous to the state’s distribution of income. Consequently, we employ a di¤erence-in-di¤erences estimation methodology that exploits the exogenous cross-state, cross-year variation in the timing of branch deregulation to assess the causal impact of bank deregulation on the distribution of income. The paper’s major …nding is that deregulation of branching restrictions substantively tightened the distribution of income by disproportionately raising incomes in the lower half of the income distribution. While income inequality widened in the overall U.S. economy during the sample period, branch deregulation lowered inequality relative to this national trend. This …nding is robust to using di¤erent measures of income inequality, controlling for time-varying state characteristics, and controlling for both state and year …xed e¤ects. We …nd no evidence that reverse causality or prior trends in the distribution of income account for these …ndings. Furthermore, the economic magnitude is consequential. Eight years after deregulation, the Gini coe¢ cient of income inequality is about 4% lower than before deregulation after con- 78 trolling for state and year …xed e¤ects. Put di¤erently, deregulation explains about 60% of the variation of inequality after controlling for state and year …xed e¤ects. Removing restrictions on intrastate bank branching reduced inequality by boosting incomes in the lower part of the income distribution, not by shrinking incomes above the median. Deregulation increased the average incomes of the bottom quarter of the income distribution by more than 5%, but deregulation did not signi…cantly a¤ect the upper half of the distribution of income. These results are consistent with the view that the removal of intrastate branching restrictions triggered changes in banking behavior that had disproportionately positive repercussions on lower income individuals. To provide additional evidence that bank deregulation tightened the distribution of income by a¤ecting bank performance and not through some other mechanism, we show that the impact of deregulation on the distribution of income varied across states in a theoretically predictable manner. In particular, if branch deregulation tightened the distribution of income by improving the operation of banks, then deregulation should have had a more pronounced e¤ect on the distribution of income in those states where branching restrictions were particularly harmful to bank operations before deregulation. Based on Kroszner and Strahan (1999), we use four indicators of the degree to which intrastate branching restrictions hurt bank performance prior to deregulation. For example, in states with a more geographically di¤use population, branching restrictions were particularly e¤ective at creating local banking monopolies that hindered bank performance. After deregulating, therefore, we should observe a bigger e¤ect on bank performance in states with more di¤use populations. This 79 is what we …nd. Across the four indicators of the cross-state severity of branching restrictions and their impact, we …nd that deregulation reduced income inequality more in those states where these branching restrictions had been particularly harmful to bank operations. These …ndings increase con…dence in the interpretation that deregulation reduced income inequality by enhancing bank performance. We …nish by conducting a preliminary exploration of three possible explanations of the channels underlying these …ndings. We view this component of the analysis as a preliminary exploration because each of these explanations warrants independent investigation with individual-level, longitudinal datasets. Nevertheless, we provide this extension to further motivate and guide future research on the channels linking bank performance and the distribution of income. The …rst two explanations stress the ability of the poor to access banking services directly. In Galor and Zeira (1993), for example, credit market imperfections prevent the poor from borrowing to invest more in education, which hinders their access to higher paying jobs. Deregulation that eases these credit constraints, therefore, allows lower income individuals to invest more in education, reducing inequality. A second explanation focuses on the ability of the poor to become entrepreneurs. In Banerjee and Newman (1993), …nancial imperfections are particularly binding on the poor because they lack collateral and because their incomes are relatively low compared to the …xed costs of obtaining bank loans. Thus, deregulation that improves bank performance by lowering collateral requirements and borrowing costs will disproportionately bene…t the poor by expanding their access to bank credit. A third explanation highlights the response of …rms to the lower interest rates 80 triggered by deregulation, rather than stressing increased access to credit by lower income individuals. While the drop in the cost of capital encourages …rms to substitute capital for labor, the cost reduction also increases production, boosting the demand for capital and labor. On net, if the drop in the cost of capital increases the demand for labor and this increase falls disproportionately on lower income workers, then deregulation could reduce inequality by a¤ecting …rms’demand for labor. Although branch deregulation stimulated entrepreneurship and increased education, our results suggest that deregulation reduced income inequality primarily by boosting the relative demand for low-skilled workers. More speci…cally, deregulation dramatically increased the rate of new incorporations (Black and Strahan (2002)) and the rates of entry and exit of non-incorporated …rms (Kerr and Nanda (2009)). However, the reduction in total income inequality is fully accounted for by a reduction in earnings inequality among salaried employees, not by a movement of lower income workers into higher paying self-employed activities or by a change in income di¤erentials among the self-employed. Furthermore, the self-employed account for only about 9% of the working age population, and this percentage did not change signi…cantly after deregulation. On education, Levine and Rubinstein (2009) …nd that the fall in interest rates caused by bank deregulation reduced high school dropout rates in lower income households. Yet, changes in educational attainment do not account for the reduction in income inequality triggered by branch deregulation during our sample period. Rather, consistent with the view that bank deregulation increased the relative demand for low-income workers, we …nd that deregulation increased the relative wages and relative working hours of unskilled workers, thus accounting for a tight- 81 ening of income distribution. Beyond the increase in the relative wages and working hours of low-income workers, bank deregulation also lowered the unemployment rate, further emphasizing the labor demand channel. This paper relates to policy debates concerning the current …nancial crisis and the role of …nancial markets in promoting economic development. First, the international policy community increasingly emphasizes the bene…ts of providing the poor with greater access to …nancial services as a vehicle for …ghting poverty and reducing inequality. In a broad cross-section of countries, Beck, Demirg-Kunt and Levine (2007) …nd that an overall index of banking sector development is associated with a reduction in income inequality across countries, but they do not analyze the impact of a speci…c, exogenous policy change. Burgess and Pande (2005) …nd that when India reformed its banking laws to provide the poor with greater access to …nancial services, this policy change reduced poverty by boosting wages in rural areas. Our …ndings also suggest that …nancial development might help the poor primarily by intensifying competition and boosting wage earnings, not by increasing the income of the poor from self-employment. Second, given the severity of the global …nancial crisis, many governments are reevaluating their approaches to bank regulation. Many economists and policymakers stress the potential dangers of …nancial deregulation. Though our work does not examine the current crisis, the results do indicate that regulations that impeded competition among banks during the 20th century were disproportionately harmful to lower income individuals, which should not be ignored as countries rethink and redesign their regulatory systems. 82 The remainder of the paper proceeds as follows. Section 2.2 describes the data and econometric methodology. Section 2.3 provides the core results, while Section 2.4 provides further evidence on how deregulation in‡uences labor market conditions. Section 2.5 concludes. 2.2 Data and Methodology To assess the e¤ect of branch deregulation on income distribution, we gather data on the timing of deregulation, income distribution, and other banking sector and statelevel characteristics. This section presents the data and describes the econometric methods. 2.2.1 Branch Deregulation Historically, most U.S. states had restrictions on branching within and across state borders. With regards to intrastate branching restrictions, most states allowed bank holding companies to own separately capitalized and licensed banks throughout a state. Other states were “unit banking states,” in which each bank was typically permitted to operate only one o¢ ce. Beginning in the early 1970s, states started relaxing these restrictions, allowing bank holding companies to consolidate subsidiaries into branches and permitting de novo branching throughout the state. This deregulation led to signi…cant entry into local banking markets (Amel and Liang (1992)), consolidation of smaller banks into large bank holding companies (Savage (1993); Calem (1994)), and conversion of exist- 83 ing bank subsidiaries into branches (McLaughlin (1995)). This relaxation, however, came gradually, with the last states lifting restrictions following the 1994 passage of the Riegle-Neal Interstate Banking and Branching E¢ ciency Act. Consistent with Jayaratne and Strahan (1996), and others, we choose the date of deregulation as the date on which a state permitted branching via mergers and acquisitions (M&As) through the holding company structure, which was the …rst step in the deregulation process, followed by de novo branching. Appendix Table I (available on the Internet Appendix, at http://www.afajof.org/supplements.asp) presents the deregulation dates. Twelve states deregulated before the start of our sample period in 1976. Arkansas, Iowa and Minnesota were the last states to deregulate, only after the passage of the Riegle-Neal Act in 1994. We have data for 50 states and the District of Columbia. Consistent with the literature on branch deregulation, we drop Delaware and South Dakota because the structure of their banking systems were heavily a¤ected by laws that made them centers for the credit card industry. Over this period, states also deregulated restrictions on interstate banking by allowing bank holding companies to expand across state borders. We con…rm this paper’s results using the date of interstate deregulation instead of the date of intrastate deregulation. However, when we simultaneously control for inter- and intrastate branch deregulation, we …nd that only intrastate deregulation enters significantly. Thus, we focus on intrastate rather than interstate deregulation throughout the remainder of this paper. 84 2.2.2 Income Distribution Data Information on the distribution of income is from the March Supplement of the Current Population Survey (CPS), which is an annual survey of about 60,000 households across the United States. The CPS is a repeated, representative sampling of the population, but it does not trace individuals through time. The CPS provides information on total personal income, wage and salary income (earnings), proprietor income, income from other sources, and a wide array of demographic characteristics in the year prior to the survey. Most importantly for our study, we start with the 1977 survey because the exact state of residence is unavailable prior to this survey. Each individual in the CPS is assigned a probability sampling weight corresponding to his or her representativeness in the population. We use sampling weights in all our analyses. We measure the distribution of income for each state and year over the period 1976 to 2006 in four ways. First, the Gini coe¢ cient of income distribution is derived from the Lorenz curve. Larger values of the Gini coe¢ cient imply greater income inequality. The Gini coe¢ cient equals zero if everyone receives the same income, and equals one if a single individual receives all of the economy’s income. We present results with both the natural logarithm of the Gini coe¢ cient as well as the logistic transformation of the Gini coe¢ cient (logistic Gini) in the regression analyses. While using the logistic Gini does bound the minimum value at zero, using the log of Gini allows one to interpret the regression coe¢ cient as a percentage change. Our second measure of income distribution is the Theil index, which is also increasing in the 85 degree of income inequality. If all individuals receive the same income, the Theil index equals zero, while the Theil index equals Ln(N) if one individual receives all of the economy’s income, where N equals the number of individuals. An advantage of the Theil index is that it is computationally easy to decompose the index into that part of inequality accounted for by di¤erences in income between groups in the sample and that part of inequality accounted for by di¤erences within each group. Third, we examine the di¤erence between the natural logarithm of incomes of those at the 90th percentile and those at the 10th percentile (Log (90/10)). Finally, we use the di¤erence between the natural logarithm of incomes at the 75th percentile and that at the 25th percentile ((Log (75/25)). Internet Appendix Table II provides more detailed information on the construction of these income distribution indicators. Consistent with studies of the U.S. labor market, our main sample includes primeage (25-54) civilians that have non-negative personal income, and excludes (i) individuals with missing observations on key variables (education, demographics, etc), (ii) individuals with total personal income below the 1st or above the 99th percentile of the distribution of income, (iii) people living in group quarters, (iv) individuals who receive zero income and live in households with zero or negative income from all sources of income, and (v) a few individuals for which the CPS assigns a zero (or missing) sampling weight. As discussed below, the results are robust to relaxing these standard de…nitions of the relevant labor market. Internet Appendix Table III provides details on the construction of the sample. There are 1,859,411 individuals in our sample. The average age in the sample is 38 years, 49% are female, and 75% are white, non-Hispanic individuals. In the sample, 86 49% have a high school degree or less, while 27% graduated from college. Only 9% of the individuals report being self-employed (entrepreneurs). In Internet Appendix Table IV, we present basic descriptive statistics on the …ve measures of income inequality, which are measured at the state-year level. In particular, we have data for the 31 years between 1976 and 2006 and for 48 states plus the District of Columbia. Thus, there are 1,519 state-year observations. Besides providing information on the means of the inequality indicators and their minimum and maximum values, we also present three types of standard deviations of the natural logarithms of the inequality indexes: cross-state, within-state, and within state-year. The cross-state standard deviation of Y is the standard deviation of Yst Yes , where Yes is the average value of Y in state s over the sample period. The within state standard deviation of Y is the standard deviation of Yst Yet , where Yet is the average value of Y in year t. The within state-year standard deviation of Y is the standard deviation of Yst Yes Yet , where Yes is the average value of Y in state s and Yet is the average value of Y in year t. These standard deviations help in assessing the economic magnitude of the impact of bank branch deregulation on the distribution of income. 2.2.3 Control Variables To control for time-varying changes in a state’s economy, we use the U.S. Department of Commerce data to calculate the growth rate of per capita Gross State Product (GSP). We also control for the unemployment rate, obtained from the Bureau 87 of Labor Statistics, and a number of state-speci…c, time-varying socio-demographic characteristics, including the percentage of high-school dropouts, the proportion of blacks, and the proportion of female-headed households. We also test whether the impact of deregulation on income inequality varies in a predictable way with di¤erent state characteristics at the time of deregulation. As we discuss below, we control for the interaction of branch deregulation with a unit banking indicator, the small bank share, the small …rm share and population dispersion. The unit banking indicator equals one if the state had unit banking restrictions prior to deregulation and zero otherwise. The following states had unit banking before deregulation: Arizona, Colorado, Florida, Illinois, Iowa, Kansas, Minnesota, Missouri, Montana, Nebraska, North Dakota, Oklahoma, Texas, Wisconsin, West Virginia, and Wyoming. The small bank share equals the fraction of banking assets in the state that are held by banks with assets below the median size bank of each state, while the small …rm share equals the proportion of all establishments operating in a state with fewer than twenty employees. Data on the small …rm share and small bank share are from Kroszner and Strahan (1999). Population dispersion equals one divided by population per square mile, which is obtained from the U.S. Census Bureau. 2.2.4 Methodology We use a di¤erence-in-di¤erences speci…cation to assess the relation between branch deregulation and income distribution, based on the following regression set-up: Yst = + Dst + Xst + As + Bt + "st (2.1) 88 In equation (2.1) Yst is a measure of income distribution in state s in year t, As and Bt are vectors of state and year dummy variables that account for state and year …xede¤ects, Xst is a set of time-varying, state-level variables and "st is the error term. The variable of interest is Dst , a dummy variable that equals one in the years after state s deregulates and equals zero otherwise. The coe¢ cient, , therefore indicates the impact of branch deregulation on income distribution. A positive and signi…cant suggests that deregulation exerts a positive e¤ect on the degree of income inequality, while a negative and signi…cant indicates that deregulation pushed income inequal- ity lower. In total, we have data for 48 states plus the District of Columbia, over 31 years, so the 1,519 state-year observations serve as the basis for much of our analysis. The di¤erence-in-di¤erences estimation technique allows us to control for omitted variables. We include year-speci…c dummy variables to control for nation-wide shocks and trends that shape income distribution over time, such as business cycles, national changes in regulations and laws, long-term trends in income distribution, and changes in female labor force participation. We include state-speci…c dummy variables to control for time-invariant, unobserved state characteristics that shape income distribution across states. We estimate equation (2.1) allowing for state-level clustering of the errors, i.e. allowing for correlation in the error terms over time within states.2 2 In robustness tests, reported in the Internet Appendix, we con…rm the results using both bootstrapped standard errors and seemingly unrelated regression (SUR) standard errors. Bootstrapped standard errors are calculated using the following procedure: First, we take a random sample of 1,519 state-year observations from our data and calculate the impact of deregulation on income inequality while accounting for state and year …xed e¤ects. The sample size is done with replacement such that a certain state in a certain year may appear several times. We take 500 such samples and estimate the impact of deregulation on income inequality 500 times. The standard deviation of the resulting estimates is the bootstrapped standard error. Second, following Bekaert, Harvey 89 2.3 2.3.1 Branch Deregulation and Income Distribution Preliminary Results Our empirical analysis rests on the assumption that the cross-state timing of bank branch deregulation was una¤ected by the distribution of income. Figures 2-1 and 2-2 show that neither the level of the Gini coe¢ cient before deregulation nor its rate of change prior to deregulation explains the timing of branch deregulation. In a regression of the year of deregulation on the average Gini coe¢ cient before deregulation or the rate of change of the Gini coe¢ cient in the years before deregulation, the t-statistic on the inequality indicators are 0.20 and -1.16 respectively. Additional evidence that income inequality did not a¤ect the timing of branch deregulation emerges from a hazard model study of deregulation. Following Kroszner and Strahan (1999), Table 2.1 reports tests of whether the Gini coe¢ cient of income inequality in‡uences the likelihood that a state deregulates in a speci…c year given that it has not deregulated yet. While the Kroszner and Strahan (1999) sample period starts in 1970, we do not have Gini data available before 1976. Also, since we use the original Kroszner and Strahan (1999) dataset, our sample period ends in 1994, when there were three states that had not yet deregulated –Arkansas, Iowa and Minnesota. Table 2.1 indicates that the timing of bank branch deregulation does not vary with the degree of pre-existing income inequality. Column 1 reports the results of a regression with only the Gini coe¢ cient of income inequality, while columns 2 –5 proand Lundblad (2005) we estimate SUR standard errors, restricting the o¤-diagonal elements of the weighting matrix to be identical. 90 vide regression results controlling for numerous state-level control variables, including those state characteristics employed by Kroszner and Strahan (1999). As shown, the Gini coe¢ cient does not enter signi…cantly in any of the Table 2.1 regressions. 2.3.2 Deregulation and the Distribution of Income In Table 2.2, we assess the impact of branch deregulation on income inequality using …ve indicators of income inequality and two regression speci…cations. In Panel A, the regressions simply condition on state and year …xed e¤ects, which are not reported. Panel B also includes numerous time-varying, state-speci…c characteristics: the growth rate of per capita gross state product, the proportion of blacks in the population, the proportion of high-school dropouts in the population, the proportion of female-headed households in the population, and the unemployment rate. The Table 2.2 results indicate that bank branch deregulation substantially reduced income inequality. The branch deregulation dummy enters negatively and signi…cantly at the 5% level in all 10 regressions. For example, consider the logistic Gini. The column 1 results suggest that deregulation induced a 3.9% reduction in the logistic Gini, which is economically large. To gauge the economic e¤ect of this result, we compare the coe¢ cient estimate to the standard deviation of the logistic Gini coe¢ cient after accounting for state and year …xed e¤ects. This standard deviation is 6.5% as shown in Internet Appendix Table IV, suggesting that branching deregulation explains about 60% of the variation of income inequality after controlling for ‡uctuations in inequality accounted for by state and year e¤ects. That said, state and 91 year …xed e¤ects explain much more of the total variation in inequality than branch deregulation. The R-square in the logistic Gini regression (column 1) of Table 2.2 is 0.36, but branch deregulation explains, on average, only two percentage points of this R-square. The Table 2.2 results indicate that deregulation tightened the distribution of income even when controlling for several time-varying state-level factors. Higher unemployment is associated with higher income inequality, though the other state characteristics do not enter independently signi…cantly across the …ve inequality measures. Given that unemployment is highly correlated over time within a state, we also run regressions including up to …ve lags of unemployment. This does not change the statistical or economic signi…cance of the coe¢ cient on bank deregulation as shown in the Internet Appendix, Tables XA and XB). Most importantly for the purposes of this paper, the results on deregulation are robust to controlling for unemployment, per capita economic growth, an economy’s socio-demographic traits, and educational attainment. Numerous robustness tests, which are reported in the Internet Appendix, con…rm these …ndings. First, we are concerned that some other time-varying, state-speci…c characteristic could be both highly correlated with the timing of each state’s branch deregulation and powerfully linked to changes in income inequality. Consequently, we also control for the state-speci…c timing of di¤erent labor protection laws, which were constructed by Autor et al. (2006). We …nd that the timing of these labor reforms is not correlated with branch deregulation and the labor market laws do not explain changes in the distribution of income. Thus, bank deregulation is not simply proxying 92 for labor market reforms that underlie the resultant tightening of the distribution of income. Furthermore, we control for an array of time-varying, state-speci…c traits, including the size of each state’s aggregate economy, the level of real per capita income in each state, or lagged values of each state’s Gini coe¢ cient. Adding these regressors does not alter the results. Second, we are also concerned that the migration of labor across state lines could a¤ect the results. If deregulation induces interstate labor reallocations that tighten the distribution of income, we want to identify and understand these dynamics. Thus, we regress the share of immigrants per state-year on the branch deregulation dummy, while controlling for year and state-…xed e¤ects. We do not …nd any signi…cant e¤ects of branch deregulation on migration ‡ows. We also control for migration ‡ows directly in the Table 2.2 regressions and obtain the same conclusions. Third, although we use the standard sample of prime age workers (25-54), we conduct a number of robustness tests regarding sample selection. In particular, the results hold when using di¤erent age groups, such as 18-64, 1854, 25-64, 25-25, 36-45, and 46-54. Furthermore, since the inclusion or exclusion of outliers, could a¤ect the results, we redo the analyses and con…rm the …ndings when (i) including all observations and (ii) excluding individuals with incomes below the 1st and above the 99th percentiles of the year-speci…c income distribution as shown in the Internet Appendix, Table VII. Fourth, Figures 2-1 and 2-2 seem to suggest that Hawaii, Utah and Virginia might be outliers and we therefore re-do all of the analyses without these states. All of the results hold, as shown in the Internet Appendix, Table VIII. Fifth, since Iowa was the last state to deregulate in 1999, we re-do the regressions for the period 1976 to 1999, thus dropping the last seven years of 93 our sample period. All the …ndings are con…rmed, as shown in the Internet Appendix, Table VIII. Finally, the results hold when examining household income, rather than individual income. 2.3.3 Deregulation and Income for Di¤erent Income Groups Although the results in Table 2.2 demonstrate that income inequality fell after intrastate branch deregulation, the analyses do not yet provide information on whether the distribution of income tightens because the rich get poorer, or because deregulation disproportionately help the poor. We now address this issue by examining the impact of branch deregulation on the incomes of individuals across the full distribution of incomes. More speci…cally, we compute the logarithm of income for the ith percentile of the distribution of income in each state s and year t, Y (i)st . We do this for i equal to 5, 10, 15, . . . , 90, 95. We then run 19 regressions of the form: Y (i)st = + Dst + As + Bt + "st (2.2) where the regressions are run for each ith percentile of the income distribution. Figure 2-3 depicts the estimated coe¢ cient, , from each of these 19 regressions and also indicates whether the estimates are signi…cant at the 5% level. Figure 2-3 shows that intrastate branch deregulation tightened the distribution of income by disproportionately raising incomes in the lower part of the income distribution, not by lowering the incomes of the rich. Speci…cally, deregulation boosted 94 incomes below the 40th percentile of the distribution of income. Deregulation did not have a signi…cant impact on other parts of the income distribution. Rather than reducing incomes above the median income level, deregulation reduced income inequality by increasing incomes at the lower end of the income distribution. 2.3.4 Dynamics of Deregulation and the Distribution of Income We next examine the dynamics of the relation between deregulation and inequality. We do this by including a series of dummy variables in the standard regression to trace out the year-by-year e¤ects of intrastate deregulation on the logarithm of the Gini coe¢ cient: log (gini)st = + 10 1 Dst + 9 2 Dst + ::: + +15 25 Dst + As + Bt + "st (2.3) where the deregulation dummy variables, the “D’s,” equal zero, except as follows: D j equals one for states in the jth year before deregulation, while D+j equals one for states in the jth year after deregulation. We exclude the year of deregulation, thus estimating the dynamic e¤ect of deregulation on income distribution relative to the year of deregulation. As and Bt are vectors of state and year dummy variables, respectively. At the end points, Dst10 equals one for all year that are ten or more years +15 before deregulation, while Dst equals one for all years that are …fteen or more years after deregulation. Thus, there is much greater variance for these end points and the estimates may be measured with less precision. After de-trending and centering the 95 estimates on the year of deregulation (year 0), Figure 2-4 plots the results and the 95% con…dence intervals, which are adjusted for state level clustering. Figure 2-4 illustrates two key points: innovations in the distribution of income did not precede deregulation and the impact of deregulation on inequality materializes very quickly. As shown, the coe¢ cients on the deregulation dummy variables are insigni…cantly di¤erent from zero for all years before deregulation, with no trends in inequality prior to branch deregulation. Next, note that inequality falls immediately after deregulation, such that D+1 is negative and signi…cant at the 5% level. Thus, the particular mechanisms and channels connecting bank deregulation with the distribution of income must be fast acting. The impact of deregulation on inequality grows for about eight years after deregulation and then the e¤ect levels o¤, indicating a steady-state drop in the Gini coe¢ cient of inequality of about 4%. In sum, changes in inequality do not precede deregulation and deregulation has a level e¤ect on inequality, but does not have a trend e¤ect. 2.3.5 Mechanisms: Impact of Deregulation as a Function of Initial Conditions We next assess whether the impact of deregulation on the distribution of income varies in predictable ways across states with di¤erent initial conditions. If the impact of deregulation on income inequality varies in a theoretically predictable manner, this provides greater con…dence in the conclusions, sheds empirical light on the mechanisms through which deregulation in‡uences the distribution of income, and also 96 reduces concerns about reverse causality. Speci…cally, if bank deregulation reduced income inequality by boosting bank performance, then the impact of bank deregulation should be stronger in states where branch regulation had a more harmful e¤ect on bank performance prior to deregulation. Following Kroszner and Strahan (1999), we consider four initial conditions that re‡ect the harmful e¤ects of branch regulation before deregulation. To proxy for the initial conditions, we use data from 1976, though the results are robust to using values measured in the year before each state deregulated. First, unit banking –where states typically restricted banks to having one o¢ ce – was the most extreme form of branching restriction and exerted the biggest e¤ect on bank performance before deregulation. Thus, we expect that deregulation exerted a particularly large impact on income inequality in states that had unit banks before they deregulated. Second, states with a high share of small banks will tend to bene…t disproportionately from eliminating branching restrictions that protect small banks from competition. Thus, we expect that deregulation had an especially large impact on inequality in states with a comparatively high ratio of small banks at the time of deregulation. Third, small …rms tend to face greater barriers to obtaining credit from distant banks than larger …rms, suggesting that local branching restrictions that protect local banking monopolies were particularly harmful in states dominated by small …rms. Thus, we expect that deregulation had a bigger impact in states with a large proportion of small …rms prior to deregulation. Finally, we examine population dispersion. Local banking monopolies will be particularly well protected if the population is di¤use, so that other banks tend to be far away. This suggests that deregulation would have 97 a bigger e¤ect on inequality in states with high initial population dispersion. These four initial conditions are not independent. States that had adopted unit banking before deregulation tended to have a higher share of small banks and …rms and more dispersed populations. The correlations between the four characteristics are far from perfect, however. The highest pair-wise correlation coe¢ cient is 0.53. Since we do not have strong reasons to favor one indicator over another, we provide the results on each in our assessment of whether intrastate branch deregulation has a particularly large e¤ect on the distribution of income in those economies where theory suggests the impact will be largest. The results in Table 2.3 indicate that the impact of branch deregulation on income inequality was stronger in states where branching restrictions had been especially harmful to bank activities before deregulation. As shown in Table 2.3, branch deregulation reduced income inequality more in states that had (i) unit banking (column 1), (ii) a more dispersed population (column 2), (iii) a higher share of small banks (column 3), and (iv) a larger share of small …rms (column 4). More speci…cally, deregulation exerted a strong, negative e¤ect on inequality in unit banking states, while this e¤ect was weaker, both economically and statistically, in non-unit banking states. In terms of population dispersion, the e¤ect of deregulation on the logistic Gini holds across the 25th, 50th and 75th percentile of the distribution of population dispersion, but is stronger for states with initially more dispersed population. In terms of the share of small banks and the share of small …rms, the results indicate that branch deregulation exerted an economically large and statistically signi…cant impact on income inequality in those states with above the median values of these pre- 98 deregulation characteristics. Branch deregulation reduced inequality more in states where branching restrictions had been particularly harmful to the operation of the banking system before liberalization, suggesting that branch deregulation tightened the distribution of income by enhancing bank performance. 2.4 2.4.1 Channels Theories of How Financial Markets A¤ect the Distribution of Income Having found that branch deregulation decreased income inequality by a¤ecting bank performance, we now explore three potential channels underlying these …ndings. The …rst two explanations rely on (i) branch deregulation improving the ability of the poor to access banking services directly and (ii) the poor using this improved access to either purchase more education or become entrepreneurs. The third explanation focuses on …rms’demand for labor, not on the poor directly using …nancial services. These explanations are not mutually exclusive. In terms of entrepreneurship, …nancial imperfections represent particularly severe impediments to poor individuals opening their own businesses for two key reasons: (i) the poor have comparatively little collateral and (ii) the …xed costs of borrowing are relatively high for the poor. From this perspective, branch deregulation that improves credit markets will lower the barriers to entrepreneurship disproportionately for poor individuals (Banerjee and Newman (1993)). 99 In terms of human capital accumulation, …nancial imperfections in conjunction with the high cost of schooling represent particularly pronounced barriers to the poor purchasing education, perpetuating income inequality (Galor and Zeira (1993)). In this context, …nancial reforms that ease …nancial market imperfections will reduce income inequality by allowing talented, but poor, individuals to borrow and purchase education. Textbook price theory provides a third channel through which bank deregulation a¤ects income inequality that does not involve the poor directly increasing their use of …nancial services. Jayaratne and Strahan (1998) show that branch deregulation reduced the cost of capital. Reductions in the cost of capital induce …rms to (i) substitute capital for labor and (ii) expand output, which increases demand for capital and labor. On net, if the output e¤ect dominates, the reduction in the cost of capital will increase the demand for labor. Even under these conditions, however, the impact of deregulation on inequality is ambiguous because we do not know if the increased demand for labor falls primarily on higher- or lower-income workers. If deregulation disproportionately increases the demand for lower-income workers, then branch deregulation could tighten the distribution of income by a¤ecting …rms’demand for labor, not by directly increasing the use of …nancial services by relatively low-income individuals. 100 2.4.2 Evidence on the Entrepreneurship Channel To provide an initial assessment of the entrepreneurship channel, we decompose the impact of bank branch deregulation on income inequality into that part accounted for by a reduction in the income gap between the self-employed and wage earners and that part accounted for by a reduction in income inequality among the self-employed and among wage earners. We conduct this decomposition in two-steps. First, using the Theil index, we decompose income inequality into the “between” component, which measures income inequality between the self-employed and wage earners, and the “within” component, which is composed of inequality among the self-employed and inequality among wage earners. As detailed in the Internet Appendix, the Theil index is easily decomposable into between and within group components. Thus, we now examine the Theil index (rather its log) in decomposing income inequality for each state and year. We then estimate the impact of deregulation on each of these components controlling for state and year …xed e¤ects. This yields that part of the estimated change in income inequality from deregulation that is accounted by a reduction in inequality between the self-employed and wage earners and that part accounted for by a reduction in inequality within the two groups. Enhanced entrepreneurship does not directly account for the impact of deregulation on the distribution of income. As shown in Panel A of 2.4, none of the change in income inequality is accounted for by a reduction in between group inequality. All of the reduction in income inequality from deregulation is accounted for by a reduction in income inequality among salaried workers. The change in between group 101 inequality is actually positive, but insigni…cant. These results are unsurprising in light of the following observations: (i) the self-employed account for only 9% of the sample, (ii) the proportion of self-employed individuals did not increase following branch deregulation, and (iii) the self-employed do not, on average, have higher incomes than salaried employees after accounting for educational di¤erences (Hamilton (2000)). These results do not suggest that the relation between branch deregulation and entrepreneurship is unimportant. Bank deregulation boosted the rate of entry and exit of …rms (Black and Strahan (2002); Kerr and Nanda (2009)). Nonetheless, the decomposition …ndings indicate that direct changes in entrepreneurial income and the movement of lower-income salaried workers into higher-income entrepreneurial activities do not account for the tightening of the distribution of income following deregulation. 2.4.3 Evidence on the Education Channel In Panel B of Table 2.4, we conduct a similar decomposition but focus on education groups. We divide the sample into those with some education beyond a high school degree (about 51% of the sample) and those with educational attainment of a high school degree or less (about 49% of the sample). Since Panel A shows that all of the reduction in income inequality is accounted for by a reduction in inequality among wage earners, we focus only on wage earners in conducting the decomposition by educational attainment. The reduction in income inequality triggered by branch deregulation is accounted 102 for by both a closing of the gap between low- and high-educated workers and by a fall in inequality among low-educated workers. From Panel B of Table 2.4, 73% (0.0074/0.0102) of overall income inequality is accounted for by a reduction in inequality within the two education categories, and the bulk of this reduction arises because of a tightening of the distribution of income among the less educated group. Furthermore, 27% (0.0028/0.0102) of the reduction in income inequality explained by bank deregulation is accounted for by a reduction in the income gap between education groups. The between group results are consistent with at least two possible explanations: (i) bank deregulation eased credit constraints and induced lower-income individuals to increase their investment in education, thereby reducing income inequality and (ii) bank deregulation increased the demand for workers in the lower-education group, reducing between group inequality.3 To evaluate whether an increase in relative educational attainment by low-skilled workers following bank deregulation accounts for the reduction in income inequality, Table 2.5 presents two additional analyses. First, we test whether bank deregulation lowers earnings inequality among workers of di¤erent ages. Speci…cally, we assess whether there is a di¤erential e¤ect of branch deregulation on income inequality for the 25-35, 36-45, and 46-54 age groups. Since Figure 2-4 shows that the impact of 3 We also examine whether bank deregulation reduced income inequality by a¤ecting the income gap between black and white individuals or the gap between women and men. First, when splitting the sample between black and white workers, we …nd that only 20% of the reduction in income inequality is accounted for by a tightening of the income gap between blacks and whites, while 80% of the reduction in total income inequality is accounted for by a tightening of income inequality within the group of whites. Second, when splitting the sample between women and men, we …nd that the reduction in income inequality is accounted for by a tightening of income inequality among women and among men, but not a reduction in income inequality between women and men. Also, see Demyanyk (2008), who examines the impact of bank deregulation on proprietors di¤erentiated by race and gender. 103 deregulation on income inequality is almost immediate and Levine and Rubinstein (2009) …nd that the main impact of deregulation on education involves a reduction in high school dropout rates, then if deregulation is reducing earnings inequality by increasing education, we should obverse this primarily among relatively young workers, not those who are older than 35. If we …nd the same relation between deregulation and earnings inequality across the di¤erent age groups, this suggests that increased educational attainment is not the primary channel through which bank deregulation reduce income inequality during our estimation period. Second, we more directly control for education by eliminating the educational attainment component of wage earnings. Speci…cally, in the analyses thus far, we have computed measures of earnings inequality based on the unconditional wage earnings of individuals. We now condition each individual’s earnings on educational attainment. That is, we compute that part of an individual’s earnings that are unexplained by years of education. Then, we assess the impact of branch deregulation on measures of earnings inequality that are computed based on conditional earnings. If branch deregulation also reduces these conditional earnings inequality measures, this suggests that deregulation is not reducing earnings inequality only by its e¤ect on educational attainment. In particular, we …rst regress log earnings on …ve dummy variables corresponding to the number of years of educational attainment (0-8, 9-11, 12, 13-15, and 16+) and year …xed e¤ects. We then collect the residuals to calculate the conditional earnings inequality measures. In robustness tests, reported in the Internet Appendix, we also control for gender and ethnicity, and obtain the same results. 104 As shown in Table 2.5, education does not account for the impact of bank deregulation on earnings inequality, suggesting that branch deregulation reduced earnings inequality primarily by boosting …rms’relative demand for low-income workers. First, across the …ve earnings inequality indicators, we do not …nd any di¤erential e¤ect of branch deregulation on income inequality among the 25-35, 36-45, and 46-54 age groups. The easing of credit constraints in response to bank deregulation is most likely to a¤ect the educational choices of individuals in school, or just out of school. It seems unlikely that branch deregulation will cause a su¢ ciently large and rapid increase in the educational attainment of workers above the age of 35, such that the resulting increase in earnings would tighten economy-wide measures of earnings inequality in the year after deregulation. Second, bank deregulation reduces conditional earnings inequality, where the conditioning is done based on educational attainment. As shown in Panel B, the estimated impact of deregulation on earnings inequality holds for conditional earnings and there is no di¤erential impact on the 25-35, 36-45, and 46-54 age groups. These …ndings imply that deregulation is not reducing earnings inequality only through its e¤ect on educational attainment. 2.4.4 Evidence on the Labor Demand Channel We now conduct a more focused examination on whether branch deregulation reduced income inequality by increasing the relative demand for unskilled workers. Speci…cally, we assess the impact of branch deregulation on the relative wages and relative working hours of unskilled vis-à-vis skilled workers, where unskilled workers are those 105 with 12 or fewer years of completed education and skilled workers are those with 13 or more years of education. Our goal is to abstract from (i) di¤erences in experience, race, and gender between unskilled and skilled workers and (ii) potentially time-varying returns to experience, race, and gender and focus only on di¤erences in wage rates and hours worked between unskilled and skilled workers. We follow a two-step procedure for computing the relative wage rates and relative working hours of unskilled workers while controlling for di¤erences in experience, race, and gender between skilled unskilled and skilled workers and accounting for time-varying returns to these characteristics. In this examination of relative wages and working hours, we exclude unemployed individuals and instead directly focus on the impact on bank deregulation on unemployment below. For simplicity, we describe the procedure for wage rates and simply note that we follow the same twostep procedure for relative working hours. We …rst estimate the following log hourly wage equation using the sample of skilled workers: s wist = Xist s t (2.4) + "ist w where wist is the log real hourly wage of skilled worker i in state s during time t and Xist is a vector of person-speci…c observable characteristics that includes the level, square, cubic and quartic in potential experience, gender and race indicators, and interaction terms between potential experience and gender and race. Equation (2.4) is estimated separately for every year between 1976 and 2006. This yields timevarying returns to observable characteristics, i.e., s t. This is important given the 106 changes in the structure of wages in the United States since the mid 1970s (Katz and Autor (1999)). Critically, equation (2.4) also contains a constant term in Xist , so that estimating equation (2.4) separately in each year provides an estimate of the conditional mean skilled wage rate in each year as part of s t. In the second step, we generate the estimated relative wage rate of each unskilled u worker i in state s during time t as the worker’s actual log real wage rate (wist ) minus the estimated wage rate that a skilled worker with the same characteristics would earn: u rist = wist where Xuist s t Xuist s t (2.5) is computed based on the condition that each unskilled worker’s observ- able characteristics (Xuist ) are rewarded at the same time-varying estimated prices ( st ) as his skilled counterpart. In this way, we abstract from potential time-varying di¤erences in the valuation of race, gender, and experience across unskilled and skilled workers in the labor market and focus on relative wage rates and working hours. Furthermore, in computing the relative wage rates of unskilled workers in equation (2.5), we subtract the estimated time-varying constant term from equation (2.4), i.e., we subtract the conditional mean skilled wage rate in each year in calculating the relative wage rate of unskilled workers. We calculate the relative working hours in exactly the same manner as above, but use weekly working hours instead of wages. We then run regressions similar to those underlying Figure 2-4. Speci…cally, r (w)ist = + 10 1 Dst + 9 2 Dst + ::: + +15 25 Dst + As + Bt + "st (2.6) where r (w)ist 107 is the log real relative wage of unskilled worker i, who resides in state s in year t, and the “D’s,”equal zero, except as follows: D j equals one for states in the jth year before deregulation, while D+j equals one for states in the jth year after deregulation. We use an analogous procedure for relative weekly working hours. Together, Figures 2-5 and 2-6 indicate that bank deregulation boosted both the relative wage rates and relative working hours of unskilled workers in comparison to skilled workers. Figure 2-5 shows that the relative wages of unskilled workers show a signi…cant increase three years after branch deregulation, a trend that continues thereafter, with an overall increase of almost 9% 15 years after branch deregulation. Figure 2-6 shows an immediate impact of branch deregulation on the relative hours worked of unskilled vis-à-vis skilled workers, a trend that continues for the following 15 years, with an overall e¤ect of 1.5 hours per week. Figure 2-7 provides additional evidence for how branch deregulation a¤ects labor demand. Recall, when examining relative wages and relative hours worked, we examine only those in the labor force and excluded the unemployed. We now focus only on the relation between bank deregulation and the unemployment rate. Speci…cally, we examine the dynamic e¤ect of branch deregulation on unemployment by running the following regression: log (unemployment)ist = + 10 1 Dst + 9 2 Dst + ::: + +15 25 Dst + As + Bt + "st (2.7) Figure 2-7 shows that bank deregulation was associated with a signi…cant drop in the unemployment rate starting two years after deregulation, with a cumulative 108 e¤ect of more than two percentage points after 15 years. Beyond bank deregulation’s positive e¤ect on both the relative wage rates and working hours of unskilled workers, branch deregulation also reduced the unemployment rate.4 2.5 Conclusions Policymakers and economists disagree sharply about who wins and who loses from bank regulations. While some argue that the unregulated expansion of large banks will increase banking fees and reduce the economic opportunities of the poor, others hold that regulations restrict competition, protect monopolistic banks, and disproportionately help the rich. More generally, an in‡uential political economy literature stresses that income distributional considerations, rather than e¢ ciency considerations, frequently exert the dominant in‡uence on bank regulations as discussed in Claessens and Perotti (2007) and Haber and Perotti (2008). We …nd that removing restrictions on intrastate branching tightened the distribution of income by increasing incomes in the lower part of the income distribution while having little impact on incomes above the median. This …nding is robust to 4 We extend the analysis of bank deregulation and unemployment along two dimensions. First, the paper’s core results in Table 2.2 hold when (1) excluding the unemployed from the sample or (2) when controlling for contemporaneous and numerous lagged values of the unemployment rate. These results suggest that the relation between branch deregulation and income inequality is not completely accounted for by a reduction in unemployment following deregulation. Second, when assessing the impact of branch deregulation on income inequality for di¤erent levels of initial unemployment rates, we …nd that states with initially higher levels of unemployment also experience a signi…cantly greater reduction in income inequality after branch deregulation, while states with an initial unemployment rate below the median level across states experience a weaker or even insigni…cant reduction in income inequality after branch deregulation. As noted, however, bank deregulation is associated with a tightening of the distribution of income even when excluding unemployed individuals from the sample. These results are reported in the Internet Appendix. 109 an array of sensitivity analyses. We …nd no evidence that reverse causality drives the results. Moreover, the impact of deregulation on income distribution varies in a theoretically predictable manner across states with distinct economic, …nancial, and demographic characteristics at the time of deregulation. These …ndings support the view that branch regulation in the United States restricted competition, protected local banking monopolies, and impeded the economic opportunities of the relatively poor. We also present evidence that the impact of branch deregulation on income inequality is an indirect one. There is no evidence that branch deregulation reduces inequality by boosting incomes of the self-employed or by increasing educational attainment. Rather, the e¤ect of branch deregulation on income inequality is driven by a reduction in inequality between skilled and unskilled workers and a reduction in income inequality among unskilled worker. In addition, we show that the relative wages and the relative working hours of unskilled vis-à-vis skilled workers increased signi…cantly after branch deregulation. This is consistent with branch deregulation leading to a greater demand for labor that falls disproportionally on lower-skilled workers who therefore see both their working hours and their wage rates increase. 110 Figure 2-1: Timing of Bank Deregulation and Pre-Existing Income Inequality: Graphical Analysis (A) (A) 2000 IA 1995 Year of bank deregulation AR MN 1990 CO MO WI IN IL KS NH NM KY MT FL OKWY TX LA MI WV ND MS WA NE OR TN HI 1985 MA GA PA UT 1980 AL CT OH VA NJ 1975 -.1 -.05 0 .05 .1 Average Gini coefficient prior to bank deregulation Figure 2-2: Timing of Bank Deregulation and Pre-Existing Income Inequality: Graphical Analysis (B) (B) 2000 IA 1995 Year of bank deregulation AR MN CONM KY WI MT MO IN FL WY TX IL LA OK ND NHKS MI MS TN ORWANE MA 1990 1985 WV HI GA PA AL 1980 UT CT OH VA 1975 -.02 0 .02 Average change in the Gini coefficient prior to bank deregulation .04 111 Figure 2-3: The Impact of Deregulation on Di¤erent Percentiles of Income Distribution .15 Percentage change .1 .05 0 -.05 5 10 15 20 25 30 35 40 45 50 55 60 65 Percentile of income distribution Significant at 5% 70 75 80 Not significant 85 90 95 112 Figure 2-4: The Dynamic Impact of Deregulation on Gini Coe¢ cient of Income Inequality .04 Percentage change .02 0 -.02 -.04 -.06 -10 -5 0 5 Years relative to branch deregulation 10 15 113 Figure 2-5: The Impact of Deregulation on the Relative Wages of Unskilled Workers Percentage change in relative real wages .12 .09 .06 .03 0 -.03 -.06 -10 -5 0 5 Years relative to branch deregulation 10 15 114 Figure 2-6: The Impact of Deregulation on the Relative Working Hours of Unskilled Workers 2 Change in relative weekly working hours 1.5 1 .5 0 -.5 -1 -1.5 -10 -5 0 5 Years relative to branch deregulation 10 15 115 Figure 2-7: The Impact of Deregulation on Unemployment Rate .01 Percentage change 0 -.01 -.02 -.03 -10 -5 0 5 Years relative to branch deregulation 10 15 116 Table 2.1: Timing of Bank Deregulation and Pre-Existing Income Inequality: The Duration Model The model is a Weibul hazard model where the dependent variable is the log expected time to bank branch deregulation. All the right-hand side variables are included in levels. The sample period is 1976 to 1994 and the sample comprises 37 states that deregulated after 1977. States drop from the sample once they deregulate. The Gini coefficient of income inequality is calculated from total personal income from the March Current Population Surveys (CPS). Control variables include real per capita GDP, proportion blacks, proportion highschool dropouts, proportion female-headed households, and unemployment rate in a state. Data on real per capita GDP are from the Bureau of Economic Analysis. Proportion blacks, high-school dropouts, and femaleheaded households are calculated from the CPS. Data on unemployment rate are obtained from the Bureau of Labor Statistics. The political-economy factors are from Kroszner and Strahan (1999). These factors include: (1) small bank share of all banking assets, (2) capital ratio of small banks relative to large, (3) relative size of insurance in states where banks may sell insurance, (4) an indicator which takes upon a value of one if banks may sell insurance, (5) relative size of insurance in states where banks may not sell insurance, (6) small firm share, (7) share of state government controlled by Democrats, (8) an indicator which takes upon a value of one if a state is controlled by one party, (9) average yield on bank loans minus Fed funds rate, (10) an indicator which takes upon a value of one if state has unit banking law, and (11) an indicator which takes upon a value of one if state changes bank insurance powers.. Standard errors are adjusted for state level clustering and appear in parentheses. (1) (2) (3) (4) (5) 0.02 0.02 0.03 0.03 0.01 (0.03) (0.05) (0.02) (0.03) (0.03) Controls Political-economy factors No No Yes No No Yes Yes Yes Yes Yes Regional indicators No No No No Yes Observations 408 408 408 408 408 Gini coefficient of income inequality 117 Table 2.2: The Impact of Deregulation on Income Inequality The table shows estimates of the impact of bank branch deregulation on the different measures of income inequality. Bank deregulation indicator equals one during all years in which a state permits in-state branching and equals zero otherwise. The measures of income inequality are: (1) logistic transformation of the Gini coefficient, (2) natural logarithm of the Gini coefficient, (3) natural logarithm of Theil index, (4) natural logarithm of the ratio of 90th and 10th percentiles, and (5) natural logarithm of the ratio of 75th and 25th percentiles. The number of observations in each regression corresponds to 49 states (we exclude Delaware and South Dakota) times 31 years between 1976 and 2006. All regressions control for state and year fixed effects. There are no other control variables in panel A. In panel B, we control for the growth rate of real per capita GDP, proportion of blacks, proportion of high-school dropouts, proportion of female-headed households, and the unemployment rate. Standard errors are clustered at the state level and appear in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. Logistic Log Log Log Log Gini (1) Gini (2) Theil (3) 90/10 (4) 75/25 (5) -0.039 (0.013)*** -0.022 (0.008)*** -0.041 (0.016)** -0.135 (0.058)** -0.077 (0.020)*** 0.36 1,519 0.35 1,519 0.43 1,519 0.74 1,519 0.60 1,519 -0.031 (0.011)*** -0.018 (0.006)*** -0.032 (0.014)** -0.101 (0.050)** -0.066 (0.017)*** Growth rate of per capita GDP (2000 dollars) -0.053 (0.072) -0.028 (0.041) -0.050 (0.081) -0.140 (0.229) -0.114 (0.119) Proportion blacks -0.390 (0.265) -0.218 (0.154) -0.462 (0.320) -0.826 (1.451) -0.231 (0.473) 0.256 (0.124)** 0.140 (0.071)* 0.219 (0.147) 0.432 (0.635) -0.072 (0.155) 0.030 (0.100) 0.017 (0.058) 0.028 (0.125) 0.226 (0.501) 0.102 (0.153) 0.011 (0.002)*** 0.006 (0.001)*** 0.013 (0.003)*** 0.069 (0.014)*** 0.023 (0.003)*** 0.40 1,519 0.39 1,519 0.46 1,519 0.75 1,519 0.63 1,519 Panel A: No Controls Bank deregulation R2 Observations Panel B: With Controls Bank deregulation Proportion high-school dropouts Proportion female-headed households Unemployment rate R2 Observations 118 Table 2.3: The Impact of Deregulation on Income Inequality as a Function of Initial State Characteristics The table presents estimates of the impact of bank deregulation on the logistic transformation of the Gini coefficient of income inequality as a function of initial state characteristics. Bank deregulation equals one during all years in which a state permits in-state branching and equals zero otherwise. All models control for state and year fixed effects. Since we control for state fixed effects the initial state characteristics are dropped from the regressions. Unit banking states are: CO, AR, FL, IL, IA, KS, MN, MO, MT, NE, ND, OK, TX, WI, WV, and WY. Population dispersion equals one divided by population per square mile, which we obtain from the 1960 estimates of U.S. Census Bureau. We use the 1976 values of the share of small banks and small firms in a state, which we obtain from Kroszner and Strahan (1999). These data exclude 12 states that deregulated before 1976. Standard errors are adjusted for state level clustering and appear in parentheses. *, **, and *** indicate statistical significance levels at the 10%, 5%, and 1% levels, respectively. (1) (2) (3) (4) Bank deregulation -0.022 (0.014) -0.026 (0.013)* 0.026 (0.025) 1.797 (0.524)*** Deregulation x (unit banking) -0.033 (0.017)* Deregulation x (initial population dispersion) -0.313 (0.138)** Deregulation x (initial share of small banks) -0.503 (0.189)** Deregulation x (initial share of small firms) Linear combination -2.062 (0.590)*** -0.055 (0.017)*** Evaluated at the 25th percentile -0.029 (0.013)** -0.011 (0.016) -0.011 (0.016) Evaluated at the 50th percentile -0.030 (0.013)** -0.029 (0.015)* -0.028 (0.015)* Evaluated at the 75th percentile -0.037 (0.013)*** -0.043 (0.016)*** -0.046 (0.015)*** 1,519 1,147 1,147 Observations 1,519 119 Table 2.4: Decomposing the Impact of Deregulation on Income Inequality to Between- and Within-Groups The table reports the impact of bank branch deregulation on the Theil index of income inequality. Bank deregulation indicator equals one during all years in which a state permits in-state branching and equals zero otherwise. The number of observations in each decomposition is 1,519, corresponding to 49 states (we exclude Delaware and South Dakota) times 31 years between 1976 and 2006. All decompositions control for state and year fixed effects. In panel A, we divide the sample into two mutually exclusive groups: (i) those who are self-employed and (ii) those who work for wages. In panel B we divide the sample of wage workers into two mutually exclusive groups: (i) those with twelve or less years of completed education and (ii) those with thirteen or more years of completed education. In the first column, in both panels, we estimate the overall impact of branch deregulation on the Theil index of inequality using all groups. In the next column, we estimate the impact of deregulation on inequality between the different groups. In the third column, we estimate the impact of deregulation on inequality within the different groups combined. The second and the third columns add up to the first column. In the next columns we estimate the impact of deregulation on income inequality separately within each of the groups. Standard errors are adjusted for state level clustering and appear in parentheses. * and ** indicate statistical significance levels at the 10% and 5% levels, respectively. Employment Groups: Panel A: All Workers Bank deregulation Panel B: Salaried Workers Bank deregulation Total Between Groups Within Groups Self Employed Salaried -0.0103 0.0002 -0.0105 -0.0077 -0.0102 (0.0043)** (0.0003) (0.0042)** (0.0074) (0.0042)** Total Between Groups Within Groups Education Groups: High Some School or College or Less More -0.0102 -0.0028 -0.0074 -0.0086 -0.0039 (0.0042)** (0.0011)** (0.0035)** (0.0043)* (0.0038) 120 Table 2.5: The Impact of Deregulation on Earnings Inequality The table shows estimates of the impact of bank branch deregulation on the different measures of earnings inequality. Bank deregulation indicator equals one during all years in which a state permits in-state branching and equals zero otherwise. The measures of earnings inequality are: (1) logistic transformation of the Gini coefficient, (2) natural logarithm of the Gini coefficient, (3) natural logarithm of Theil index, (4) natural logarithm of the ratio of 90th and 10th percentiles, and (5) natural logarithm of the ratio of 75th and 25th percentiles. The number of observations in each regression corresponds to 49 states (we exclude Delaware and South Dakota) times 31 years between 1976 and 2006 times 3 age groups (25-35, 36-45, and 46-54). All regressions control for state and year fixed effects, age fixed effects, age-specific state fixed effects, and age-specific year fixed effects. In panel A, we use total annual earnings of wage and salary workers. In panel B, we use total annual earnings of wage and salary workers, which are conditional on years of completed education. Specifically, we first regress log real annual earnings on six dummies of years of completed education (0-8, 9-11, 12, 13-15, and 16+) and year fixed effects and then calculate measures of inequality based on the residuals. When calculating the residuals we use sampling weights provided by the CPS. Standard errors are clustered at the state level and appear in parentheses. ** and *** indicate statistical significance at the 5% and 1% levels, respectively. Logistic Log Log Log Log Gini (1) Gini (2) Theil (3) 90/10 (4) 75/25 (5) Panel A: Unconditional Earnings Bank deregulation -0.041 (0.018)** -0.025 (0.011)** -0.053 (0.023)** -0.108 (0.038)*** -0.054 (0.020)*** (Bank deregulation) x (Ages 36-45) 0.006 (0.015) 0.004 (0.010) 0.013 (0.020) 0.034 (0.036) -0.005 (0.015) (Bank deregulation) x (Ages 46-54) 0.011 (0.018) 0.007 (0.012) 0.020 (0.024) 0.034 (0.042) -0.010 (0.018) R2 0.26 0.26 0.26 0.40 0.39 Observations 4,557 4,557 4,557 4,557 4,557 Bank deregulation -0.040 -0.039 -0.083 -0.005 -0.002 (Bank deregulation) x (Ages 36-45) (0.015)** 0.004 (0.015)** 0.004 (0.031)** 0.019 (0.002)*** 0.001 (0.001)** -0.000 (Bank deregulation) x (Ages 46-54) (0.014) 0.005 (0.014) 0.005 (0.033) 0.016 (0.002) 0.002 (0.001) -0.001 (0.015) (0.015) (0.034) (0.002) (0.001) 0.50 4,557 0.50 4,557 0.40 4,557 0.48 4,557 0.46 4,557 Panel B: Earnings Conditional on Education R2 Observations Chapter 3 Race Discrimination and Competition 3.1 Introduction More than four decades after the Civil Rights Act, large disparities in wage rates persist between black and white Americans (Smith and Welch (1989), Donohue and Heckman (1991), Altonji and Blank (1999), and Neal (2006)). Yet, researchers have not fully determined the degree to which blacks earn less than whites with the same observable skills because of di¤erences in unobserved skills or because of racial wage discrimination, whereby blacks are paid less than identically productive whites. Becker (1957) argues that taste-based discrimination, the disutility that white employers attach to hiring black workers, can produce racial wage discrimination and that intensi…ed product market competition can reduce this wage gap between identically productive workers. With lower entry barriers, …rms with less of a taste for 121 122 discrimination can initiate pro…table operations by hiring equally productive black workers at lower wage rates than their white counterparts, boosting the relative demand for black workers. Thus, like any ine¢ ciency, competition erodes racial wage discrimination. Rather than emphasizing racial biases and competition, Arrow (1972) and Phelps (1972) stress that a gap in skills and imperfect information can explain racial wage di¤erentials. If black workers are on average less productive than white workers due to characteristics that are unobserved by employers, those employers will use the observable characteristic –race –as a signal of productivity. In turn, employers will pay blacks a lower wage rate than whites with identical observable skills. In this paper, we evaluate the causal impact of intensi…ed competition on blacks’ relative wages and assess whether the mechanisms linking competition and the racial wage gap are consistent with a taste-based explanation of the racial wage gap. In particular, Becker (1957) theory suggests that competition will increase the relative demand for black workers only in economies where employers have a taste for discrimination. We examine this. Indeed, we provide the …rst evaluation of whether the impact of competition on blacks’ labor market opportunities varies positively with the economy’s taste for discrimination. This provides information on whether competition diminishes the manifestation of racial prejudices in labor markets. Speci…cally, we use bank deregulation to identify an exogenous intensi…cation of competition among non…nancial …rms, and evaluate its impact on the racial wage gap in the overall economy. From the mid-1970s to 1994, individual states of the United States relaxed restrictions on both the entry of banks from other states and 123 the branching of banks within states. The resultant intensi…cation of competition among banks reduced …nancial market imperfections and lowered entry barriers facing non…nancial …rms. Black and Strahan (2002), Kerr and Nanda (2009), and our own estimates demonstrate that bank deregulation substantively spurred the entry of new …rms. We assess whether this greater competition among …rms boosted blacks’ relative wages. That is, we exploit the cross-state, cross-time variation in bank deregulation to identify exogenous changes in entry barriers facing non…nancial …rms and then estimate whether the resultant intensi…cation of competition increases blacks’ relative wages. Critically, we do not focus on black’s relative wages within the banking industry. Rather, we evaluate whether bank deregulation that spurred competition in the entire economy reduced the racial wage gap in the state’s overall economy. Our estimation strategy requires that bank deregulation is exogenous to competition and blacks’labor market outcomes. Geographic restrictions on banking protected local banks from competition for much of the 20th century (White (1982)). By the mid-1970s, however, technological innovations reduced the economic advantages of these restrictions, weakening the ability and desire of banks to …ght deregulation and triggering the dismantling of these statutes over the next two decades. Kroszner and Strahan (1999) show that (1) the invention of automatic teller machines weakened the geographical bond between customers and banks; (2) checkable money market mutual funds facilitated banking by mail and telephone; and, (3) improvements in data processing, telecommunications, and credit scoring techniques weakened the informational advantages of local bankers. These national innovations interacted with preexisting state characteristics to produce considerable variation in the timing of 124 bank deregulation across states. In states where probable losers from deregulation, such as small banks, were politically powerful, deregulation occurred later, while in states where probable winners from deregulation, such as small …rms, were relatively powerful, deregulation occurred sooner. The timing of deregulation was neither associated with competition in the non…nancial sector (Black and Strahan (2002)), nor, as we demonstrate below, with the racial wage gap. Hence the history of bank deregulation and its impact on entry barriers facing non…nancial …rms provides a natural laboratory for evaluating the causal impact of competition on blacks’relative wages. Our approach provides information on the mechanisms linking competition and relative wages by evaluating whether the impact of competition on the racial wage gap varies by the degree of racial bias. According to Becker (1957), competition will boost blacks’relative wages only when the marginal employer receives disutility from hiring black workers. Thus, we use interracial marriage rates prior to deregulation to proxy for cross-state di¤erences in the taste for discrimination. Using the 1970 census, we compute the predicted rate of intermarriage based on individual and state characteristics. We interpret the di¤erence between the predicted rate of intermarriage and the actual rate as positively related to the taste for discrimination. Although imperfect, this racial bias index is (1) measured prior to bank deregulation, (2) uncorrelated with the timing of bank deregulation, and (3) strongly associated with recent survey measures of racial prejudices employed by Charles and Guryan (2008). We use the racial bias index to evaluate whether the impact of competition on blacks’ relative wages varies positively with the economy’s taste for discrimination, as suggested by the taste-based view of racial wage discrimination. Econometrically, di¤erentiating 125 by year, state, and taste for discrimination yields a quasi-triple di¤erence estimation framework that, as we explain below, allows us to relax standard identi…cation assumptions. Employed with this framework, we turn to the data. Using individual-level data from the Current Population Survey for survey years 1977 to 2007, state-level data on new incorporations per capita to proxy for competition in the non…nancial sector from Black and Strahan (2002), data on married couples from the 1970s Census, and the dates of bank deregulation from Kroszner and Strahan (1999), we evaluate the impact of deregulation and competition on blacks’relative wages. The …ndings suggest that intensi…ed competition substantially reduced racial wage discrimination by ameliorating the manifestation of racial prejudices in labor markets. We …rst …nd that bank deregulation increased the rate of new incorporations across states with di¤erent values of the racial bias index. Dynamically, the impact of deregulation on the rate of new incorporations grows over time. Second, bank deregulation increased blacks’relative wage rates, but only in “high racial bias” states. In states with above the median level of the racial bias index, deregulation eliminated about one-third of the initial racial wage gap after …ve years. Furthermore, the dynamic impact of deregulation on blacks’relative wages mirrors that of deregulation on new incorporations, with blacks’relative wages rising for many years following bank deregulation. Third, blacks’relative wages are positively associated with the rate of new incorporations in high racial bias states. Thus, while bank deregulation boosted the rate of new incorporations in high and low racial bias, there is a positive association between blacks’relative wages and both bank deregulation and new incorporations 126 only in high racial bias states. Moreover, the two-stage least squares results indicate that an exogenous intensi…cation of competition only boosted blacks’relative wages in states with a su¢ ciently high taste for discrimination. Using inter- and intrastate bank deregulation as instrumental variables to identify exogenous shocks to the rate of new incorporations, we …nd that increases in the rate of new incorporations only reduced the racial wage gap in high racial bias states, such that a ten percent increase in the rate of new incorporations reduced the black-white wage di¤erential by 2.5 percent. Furthermore, exogenous increases in the rate of new incorporations also increased the relative working hours of black workers in high racial bias states, consist with the interpretation that intensi…ed competition boosted the relative demand for black workers. The …ndings do not simply imply that states with a high degree of racial bias converge toward low racial bias states, nor that blacks’relative wages increase over time. Rather, among high racial bias states, exogenous increases in competition reduce the wage gap while accounting for state and year …xed e¤ects. The results, therefore, cannot be attributed to convergence of blacks’relative wages in high racial bias states toward those in other states nor to any common time e¤ects (businesscycles) among high racial bias states. Indeed, by conditioning on state and year e¤ects, we control for all national in‡uences, such as federal statutes, as well as statespeci…c factors. The results imply that competition boosts blacks’ relative wages by eroding the adverse e¤ects of racial prejudices on the relative demand for black workers. To provide additional information on how competition shapes the impact of racial 127 prejudices on labor markets, we also examine segregation. Becker (1957) theory predicts that when employers are heterogeneous in both quality and taste for discrimination, black workers will be hired by employers with the weakest racial prejudices, creating segregation in the workforce. Moreover, the taste-based theory predicts that an intensi…cation of product market competition will reduce segregation. A lowering of entry barriers that allows new employers with less of a taste for discrimination to enter the market increases the employment opportunities of black workers, reducing segregation. Thus, if our …ndings on blacks’relative wages re‡ect the causal impact of competition on how racial prejudices a¤ect labor markets, then we should also observe greater integration following an intensi…cation of competition. We examine this additional testable implication and …nd that an exogenous increase in the rate of new incorporations reduced racial segregation across industries. To proxy for the degree of segregation, we construct several measures of the degree to which each industry is disproportionately composed of white workers or run by white managers. In high racial bias states, we …nd that an increase in the rate of new incorporations sparked by bank deregulation induced more blacks to work in industries that historically were composed disproportionately of white workers or had a high proportion of white managers. These …ndings are fully consistent with the view that intensi…ed competition reduced the e¤ects of racial prejudices on blacks’relative wages and racial segregation. This paper’s …ndings are robust to several potentially confounding in‡uences. First, the intensi…cation of competition from bank deregulation could have increased blacks’relative wages by disproportionately helping occupations and industries with 128 a comparatively high proportion of blacks, not by reducing the manifestation of racial prejudices on labor markets (Black and Spitz-Oener (2007)). However, we …nd that blacks’wages rise relative to comparable white workers within the same occupation and industry. Second, bank deregulation could trigger changes in the skill composition of the labor force through the selection of workers into the labor force, interstate migration, and changes in self-employment (Butler and Heckman (1977); Mulligan and Rubinstein (2008)). We …nd no evidence that deregulation substantively a¤ected the relative skill composition of black workers. Third, bank deregulation could have changed the prices of unobserved skills in which average black and white workers are di¤erentially endowed. Following Juhn, Murphy and Pierce (1991), however, we …nd that bank deregulation improved black workers’location throughout white workers’ residual wage distribution, indicating that competition boosted blacks’relative wages in particular, not the relative wages of comparatively low income workers in general. We are not the …rst to examine competition and discrimination. Becker (1957), Shepherd and Levin (1973), and Oster (1975) compare market concentration and relative wages across industries, obtaining mixed results. Ashenfelter and Hannan (1986), however, stress the importance of examining labor market integration, not relative wages, when comparing industries or when examining an industry over time because relative wages are primarily established in the overall economy, not in separate industries. Consistent with Becker’s theory, they …nd a negative association between market concentration and the share of female employees across several banking markets in Pennsylvania and New Jersey. In contrast to the Ashenfelter and Hannan (1986) approach, several studies trace the impact of competition on relative 129 wages within a single industry. Heywood and Peoples (1994) and Peoples and Talley (2001) …nd that the deregulation of trucking increased the relative wages of black workers. Black and Strahan (2001) …nd that bank deregulation increased competition between banks, disproportionately reducing the rents paid to male workers relative to female bank employees. Within manufacturing, Black and Brainerd (2004) …nd that globalization intensi…ed competition and thereby reduced the gender wage gap. Our major contribution is that we provide the …rst evaluation of whether the impact of an exogenous intensi…cation of product market competition on blacks’relative wages and racial integration varies positively with the economy’s taste for discrimination. That is, we not only assess whether competition boosts the relative demand for black workers in general, we examine whether competition boosts blacks’relative wages and integration only in those environments in which the taste-based theory suggests that competition will enhance blacks’labor market opportunities. Toward this end, we use the rate of new incorporations per capita as a measure of product market competition, rather than market concentration, because Becker (1957) identi…es the entry of new …rms as the mechanism through which lower entry barriers changes the relative demand for black workers. Our results are fully consistent with the central implication of the taste-based theory: An intensi…cation of product market competition diminishes the manifestation of racial prejudices in labor markets. Our work complements Charles and Guryan (2008) study of the relation between racial prejudices and blacks’relative wages. Using state-level survey measures of racial prejudices to gauge relative demand for black workers and the share of black workers in the labor force, they provide the …rst empirical support for Becker (1957) hypothesis 130 that a stronger taste for discrimination by the marginal …rm reduces blacks’relative wage rates. Rather than evaluating the relation between racial prejudices at the margin and relative wages, we examine the impact of changes in competition on changes in relative wage rates, while distinguishing states by the taste for discrimination. This paper also relates to research on …nance and income inequality (Levine (2005); Beck et al. (forthcoming); Demirgüç-Kunt and Levine (2009)). We show that exogenous improvements in the functioning of banks substantively enhanced the economic opportunities of an historically disadvantaged group by diminishing the impact of racial prejudices on labor market opportunities. Our …ndings do not reject, and might even complement, statistical discrimination explanations of racial wage di¤erentials. Di¤erences in productive skills might play an additional role in explaining the racial wage gap. Indeed, Neal and Johnson (1996) explain a large proportion of the racial wage gap using a measure of cognitive achievement. Heckman, Stixrud and Urzua (2006), however, warn that gaps in cognitive achievement scores could re‡ect the historical rami…cations of racial prejudices, highlighting potential interactions between statistical and taste-based theories. For example, competition that boosts blacks’relative wages might enhance incentives for blacks to acquire more skills. This would increase the average skill level of blacks, potentially reducing statistical discrimination and triggering self-reinforcing dynamics that reduce disparities in wages, education, and health (Coate and Loury (1993); Benabou (1996); Card and Krueger (1992); Durlauf (1996); Durlauf (2005); Durlauf and Fafchamps (2005); Jencks and Phillips (1998); Altonji and Pierret (2001); Fryer and Levitt (2004); Almond, Chay and Greenstone (2008)). 131 In what follows, Section 3.2 discusses the use of bank deregulation as an exogenous source of variation in competition. Section 3.3 outlines the conceptual and statistical framework. Section 3.4 describes the data and econometric design. Section 3.5 presents the core results on relative wages, while Section 3.6 provides the results on racial integration. After providing robustness tests in Section 3.7, Section 3.8 concludes. 3.2 3.2.1 Bank Deregulation and Competition A Brief History of Bank Branch Regulation Geographic restrictions on banks have their origins in the U.S. Constitution, which limited states from taxing interstate commerce and issuing …at money. In turn, states raised revenues by chartering banks and taxing their pro…ts. Since states received no charter fees from banks incorporated in other states, state legislatures prohibited the entry of out-of-state banks through interstate bank regulations. To maximize revenues from selling charters, states also e¤ectively granted local monopolies to banks by restricting banks from branching within state borders. These intrastate branching restrictions frequently limited banks to operating in one city. By protecting ine¢ cient banks from competition, geographic restrictions created a powerful constituency for maintaining these regulations even after the original …scal motivations receded. Indeed, banks protected by these regulations successfully lobbied the government to prohibit interstate banking and intrastate branching (South- 132 worth (1928); White (1982); Economides et al. (1996)). In the last quarter of the 20th century, however, technological, legal, and …nancial innovations diminished the economic and political power of banks bene…ting from geographic restrictions. In particular, a series of innovations lowered the costs of using distant banks. This reduced the monopoly power of local banks and weakened their ability and desire to lobby for geographic restrictions. For example, the invention of automatic teller machines (ATMs), in conjunction with court rulings that ATMs are not bank branches, weakened the geographical link between banks and their clientele. Furthermore, the creation of checkable money market mutual funds made banking by mail and telephone easier, thus further weakening the power of local bank monopolies. Finally, the increasing sophistication of credit scoring techniques, improvements in information processing, and the revolution in telecommunications reduced the informational advantages of local bankers, especially with regards to small and new …rms. These national developments interacted with preexisting state characteristics to shape the timing of bank deregulation across the states. As shown by Kroszner and Strahan (1999), deregulation occurred later in states where potential losers from deregulation (small, monopolistic banks) were …nancially stronger and had a lot of political power. On the other hand, deregulation occurred earlier in states where potential winners of deregulation (small …rms) were relatively numerous. Most states deregulated geographic restrictions on banking between the mid-1970s and 1994, when the Riegle-Neal Act e¤ectively eliminated these restrictions. The forces driving bank deregulation were exogenous to competition in the non- 133 …nancial sector and the racial wage gap. The timing of deregulation was not shaped by new …rm formation (Black and Strahan (2002), Kerr and Nanda (2009)), nor by the strength of labor unions (Black and Strahan (2001)), nor by the degree of earnings inequality (Beck et al. (forthcoming)) in each state. Moreover, we show below that the racial wage gap does not explain the timing of bank deregulation. 3.2.2 Bank Deregulation and Competition in the Non-Financial Sector An extensive literature examines the rami…cations of bank deregulation. For example, Jayaratne and Strahan (1998) …nd that removing geographic restrictions improved banking e¢ ciency by reducing interest rates on loans, raising them on deposits, lowering overhead costs, and shrinking loan losses. Beyond banking, deregulation accelerated a state’s rate of economic growth (Jayaratne and Strahan (1996); and Huang (2008)), lowered economic volatility (Demyanyk et al. (2007)), improved the selfemployment opportunities of disadvantaged groups (Demyanyk (2008)), and reduced income inequality (Beck et al. (forthcoming)). More speci…cally for the purposes of this paper, inter- and intrastate bank deregulation intensi…ed competition among …rms in the non-…nancial sector by reducing barriers to entry. Black and Strahan (2002) …nd that deregulation helped entrepreneurs start new businesses, with the rate of new incorporations per capita in a state increasing by six percentage points following deregulation. Kerr and Nanda (2009) …nd that interstate deregulation increased the number of new start-ups by six percent- 134 age points and expanded the number of facilities of existing …rms by four percentage points across all sectors in the economy. Furthermore, they …nd a dramatic increase in both the entry and exit of …rms, suggesting that deregulation increased contestability throughout the economy.1 Below, we con…rm that inter- and intrastate bank deregulation boosted the rate of new incorporations per capita and we use this to identify an exogenous, positive shock to competition in our analysis of racial discrimination. 3.3 3.3.1 Conceptual and Statistical Framework Conceptual Framework Becker (1957) seminal analysis of racial discrimination and competition motivates our empirical analysis. In Becker’s model, employers are heterogeneous in both quality and ‘taste for discrimination,’which is de…ned as the degree to which they su¤er “disutility”from employing minority workers.2 In equilibrium, minority workers must ‘compensate’employers either by being more productive at a given wage or by accepting a lower wage for identical productivity. In turn, market pressures cause blacks to be hired by the least racially biased employers. Thus, the joint distribution of 1 Several interrelated factors explain the impact of deregulation on competition in the overall economy. First, deregulation fueled competition among banks and reduced lending rates. This facilitated the expansion of existing …rms and the entry of new ones. Furthermore, the country’s more innovative banks were developing better techniques for evaluating …rms. Sophisticated credit-scoring techniques in conjunction with dramatic advances in information processing enhanced the ability of banks to evaluate and …nance new and small businesses. By easing the acquisition of banks across and within state boundaries, deregulation helped spread these superior techniques for evaluating …rms Hubbard and Palia (1995). Deregulation also permitted the formation of larger, more geographically diversi…ed banks. Diamond (1984) theory of intermediation suggests that greater diversi…cation reduces the monitoring costs of lending to riskier, more opaque …rms. Indeed, Berger, Saunders, Scalise and Udell (1998) shows that small business lending increases after small banks are acquired. 2 Bertrand and Mullainathan (2004) …nd that resumes with traditionally white names receive 50 percent more calls for interviews than identical resumes with distinctively black names. 135 employer taste for discrimination and employer quality combine with the proportion of black workers in the economy to in‡uence the racial wage gap. For example, the racial wage di¤erential will be larger when all employers have a greater taste for discrimination, holding other things constant. Similarly, the racial wage di¤erential vanishes if employers receive no disutility from hiring blacks. As a …nal example, an increase in the relative supply of black workers, holding the distribution of existing …rms constant, will tend to widen the racial wage di¤erential as …rms with a stronger taste for discrimination are induced to hire black workers through lower relative wage rates. De…ne racial discrimination as the percentage di¤erence in the wage rates of identical black and white workers, so that, the log hourly wage rates of black workers WstB in economy s during period t di¤er from those of identical white workers WstW by a racial discrimination premium dst : WstB WstW dst : (3.1) In anticipation of examining the states of the United States, we use the subscript s to designate an economy. Racial discrimination is not de…ned as an economy’s attitude toward minorities, though these tastes –these racial prejudices –in‡uence the wage gap between identical black and white workers. In this framework, an increase in competition – a reduction in entry barriers – can reduce the black-white wage rate di¤erential. With lower entry barriers, new employers with less of a taste for discrimination than incumbents can earn greater 136 pro…ts than existing …rms by hiring equally productive black workers at lower wage rates than their white counterparts. Thus, the lowering of entry barriers boosts the relative demand for black workers, eroding the manifestation of racial prejudices on relative wage rates.3 Competition does not change any individual’s preferences toward hiring minority workers. Rather, competition makes racial wage discrimination more costly by facilitating the entry of employers with less of a taste for discrimination. The model further suggests that the marginal impact of competition on blacks’ relative wage rates varies positively with the degree of racial prejudice in the economy. At the most basic level, if employers have no taste for discrimination, then the racial wage gap equals zero in the model. Under these conditions, competition does not a¤ect black’s relative wage rates. Competition only increases the relative wage rates of black workers in the model when employers have a taste for discrimination.4 In turning toward an empirical assessment of the relationship between competition and racial discrimination, we use the entry of new …rms as a proxy measure of competition. One key advantage of new …rm entry as a proxy for competition, rather than more traditional measures based on market share, is that Becker (1957) theory focuses on the actual entry of new …rms: The entry of new …rms with di¤erent tastes toward hiring minorities from those of existing …rms reduces racial discrimination. Furthermore, we (1) explicitly account for cross-state di¤erences in the taste for 3 Note, perfect competition does not necessarily eliminate the black-white wage di¤erential as argued by Becker (1957). Depending on the joint distribution of …rm quality and taste for discrimination, an equilibrium racial wage gap might still obtain with zero entry costs. 4 One can conceive of distributions of existing employer quality and racial prejudices inconjunction with particular distributions of the quality, racial prejudices, and entry barriers of shadow …rms, such that an increase competition will not boost blacks’relative wages. We evaluate this below. 137 discrimination and (2) control for state and year …xed e¤ects to hold other factors constant that theory suggests a¤ect the racial wage gap, such as cross-state di¤erences in the racial composition of the workforce. Thus, we evaluate whether the impact of competition on the racial wage gap varies positively with the degree of racial bias in the economy. 3.3.2 Statistical Framework We now outline a statistical framework for assessing the impact of competition on the racial wage gap. Below, we de…ne the construction of the actual data series and provide details on the econometric methods. B Let Wist equal the log hourly wage rate of black worker i in state s in time t. Further, de…ne Rist as the relative wage rate of black worker i in state s in time t, which equals the di¤erence between the log hourly wage rate of black worker i B B (Wist ) with observable characteristics (Xist ) and the wage rate of a white worker with identical observable traits B W st (Xist ). We call B W st (Xist ) the conditional wage rate of black worker i, where the conditioning is done on black worker i receiving the same wage rate as the average white worker with identical observed characteristics in state B s in time t (Xist ). For simplicity of illustration, yet without loss of generality, assume that states can be divided into those with a high taste for discrimination and those where people do not receive as much disutility from working and interacting with minorities. We allow the impact of deregulation and competition on blacks’ relative wages to vary 138 by a state’s degree of taste for discrimination. Let Ts be a binary variable which is equal to one if the taste for discrimination in state s is high and zero if it is low. We de…ne the construction of this racial bias index below. Thus, we begin with a standard relative log hourly wage equation given by a linear-in-the-parameters speci…cation: Rist B W st (Xist ) B Wist = 0 Nst + 1 Nst Ts + s + 0t + 1t (3.2) + "ist ; where Nst denotes the entry of new …rms, which serves as a proxy for competition, in state s in time t, Ts is zero-one racial bias index, 0t and 1t s is a vector of state …xed e¤ects, are vectors of time e¤ects in low and high racial bias states respectively, and "ist is an error term composed of a person speci…c idiosyncratic shock and any unobserved state-year …xed e¤ects. Note, this speci…cation allows time e¤ects to vary across high and low racial bias states, so that racial bias states and 1t 0t measures the time e¤ect in low measures the time e¤ect in high racial bias states. In terms of the parameter estimates on competition, 0 is the causal impact of competition on the relative wage rates of black workers in low racial bias states, while 0 + 1 is the e¤ect of competition on the black-white wage gap in high racial bias states. Becker’s theory predicts that (1) competition boosts the relative wages of black workers in states with a su¢ ciently high degree of racial bias, i.e., 0 + 1 > 0; and (2) the impact of competition on blacks’ relative wages is larger in states with a higher degree of racial bias, i.e., 1 > 0. Since the marginal employer might have a nonzero taste for discrimination even in low racial bias states, our speci…cation does 139 not necessarily imply that 3.4 3.4.1 0 = 0, only that 0 + 1 > 0 and 1 > 0. Data and Econometric Design Data In this study we use micro-level and state aggregate data sources. For the micro-level data on labor market characteristics, we use the Integrated Public Use Microdata Series (IPUMS) from the U.S. Current Population Survey (CPS), March Supplements for the survey years 1977 to 2007 and the Census of Population for 1970, Form 1 State, and Form 2 State one-percent samples. These are combined with aggregate state level data on bank deregulation, taken from Kroszner and Strahan (1999), and new incorporations as a proxy for competition in the non…nancial sector, which we obtained from Black and Strahan (2002). CPS Samples for the Years 1977 to 2007 The CPS March Annual Demographic Supplements provide information on earnings, along with weeks and hours worked in the calendar year preceding the March survey so that the 1991 survey provides information on earnings in 1990. We start in Survey year 1977 because that is when the CPS reports information on each person’s state of residence. To enhance comparability and connect our analyses to the literature, we restrict our sample to non-Hispanic white and black adult civilian males between the ages of 18 and 65 during the working year, and exclude persons living in group quarters or with missing data on relevant demographics. Our main wage sample 140 further excludes the self-employed, persons in the military, agricultural, or private household sectors, persons with inconsistent reports on earnings, and individuals with allocated earnings. We classify the adult population into six educational categories: (i) persons with 0–8 years of schooling completed; (ii) high school dropouts; (iii) high school graduates, 12 years of schooling; (iv) some college; (v) college graduate; and (vi) advanced degree. Potential work experience is constructed as the maximum between zero and age (in year of survey) minus years of schooling completed minus seven. Wage rates are de…ned as real annual earnings divided by the product of weekly working hours and annual working weeks. We use the Consumer Price Index to de‡ate earnings to 2000 dollars and set hourly earnings to missing if any of these components is missing or zero. Following Autor, Katz and Kearney (2008), workers with top coded earnings have their annual earnings set to 1.5 times the annual topcode amount. We trim outliers with hourly wages below the 1st percentile and above the 97th percentile of the year-speci…c distribution of hourly earnings of full-time, fullyear workers. This trimming virtually eliminates individuals with top-coded annual earnings. The results are robust to altering the de…nition of outliers. Finally, in accord with previous research on bank deregulation, we drop Delaware and South Dakota from our analyses due to large concentration of credit card banks in these states. Table 3.11 provides more details on the construction of our sample. 141 The 1970 Census We use the 1970 Census to construct information on the rate of racial intermarriage in each state. The Census samples are the largest microdata set containing detailed marriage and demographics information. Our primary sample includes married whites and blacks between that ages of 18 to 65, and excludes couples in which at least one person is living in group quarters or has missing data on race, gender, state of residence, marital status and educational attainment. State Level Data on Bank Deregulation and New Incorporations We obtain the dates of interstate and intrastate bank deregulation from Kroszner and Strahan (1999) and Amel (1993). Most states removed these geographic restrictions on banking between the mid-1970s and 1994, when they were eliminated by federal legislation. Table 3.10 provides the deregulation dates for each state. To measure new …rm entry, we use the rate of new incorporations per capita from Black and Strahan (2002), who obtain these data from Dun and Bradstreet. Speci…cally, we use the log of new business incorporations per capita for each state over the period 1977-1994. 3.4.2 Generating Relative Wages and the Racial Bias Indexes Relative Wages We …rst compute the estimated relative wage rate for each black worker i in the ^ ist ), which equals the worker’s actual wage rate minus the estimated wage sample (R 142 rate that the average white worker with identical characteristics would earn. We follow a two-step procedure for computing the log hourly wage rate that a white worker with identical characteristics as his black counterpart would earn. We …rst estimate the following Mincerian log hourly wage equation using the sample of white workers: 0 W Wist = Xist W t (3.3) + eist ; W is the log hourly wage of white worker i in state s during time t, Xist is where Wist a vector of person-speci…c observable determinants of log hourly wages (e.g., quartic in potential experience, and six education categories), Xist also includes state …xed e¤ects, eist captures the component of wages idiosyncratic to white worker i. Equation (3:3) is estimated separately for every year between 1976 and 2006. This yields time-varying returns, or “prices” to observable characteristics, i.e., W t . Further, the average value of unobservable traits among white workers in state s during time t are incorporated into the estimation of (3.3) by the inclusion of state …xed e¤ects in each of the 31 separate regressions. Below, we analyze the potential biases induced by unobservable traits. This …rst step has two noteworthy and crucial properties. First, given the changes in the structure of wages in the United States since the mid 1970s (Katz and Autor (1999)), we allow the Mincerian returns to observable skills W t to vary by year. This is crucial for our analyses due to the well-documented skill gap between black and white workers. Failure to account for time-varying returns to observables will lead to erroneous estimates of the dynamic pattern of relative wages, potentially biasing 143 our assessment of the impact of competition on the black-white wage gap. Second, by allowing state …xed e¤ects to vary by year, we control for all timevarying, state-speci…c characteristics that might a¤ect the wage rates of white workers including the e¤ect of bank deregulation. Speci…cally, we include a vector of state dummy variables in (3.3), which is estimated separately for each year. Thus, we control for the state’s unemployment rate, its gross state product, changes in the industrial composition of production, the racial composition of each state, state-level productivity di¤erences, and regulatory reforms on the wage rates of white workers. By controlling for these wage rate determinants in general, we can more precisely focus on the impact of bank deregulation on blacks’relative wages in particular. In some extensions, we also control for time-varying occupation- and industryspeci…c e¤ects. To do this, we add a vector of occupation and/or industry dummy variables to (3.3), which is estimated separately for each year. We use the three digit occupation and industry codes recorded in the CPS, which are based on the 1950 Census Bureau classi…cation system to provide a consistent set of industry and occupation codes throughout the sample. This means including 408 additional dummy variables each year (144 industries and 262 occupations). While a worker’s occupation and industry could re‡ect racial wage discrimination, we condition on occupation and industry in robustness tests to assess whether blacks’wages change relative to white workers with the same observable skills, who are working in the same occupation and industry. In the second step, we generate the estimated relative wage rate of each black B worker i in state s during time t as the worker’s actual wage rate (Wist ) minus the 144 estimated wage rate that a white worker with the same characteristics would earn W Xist ^t B0 , using the estimated parameters from (3.3):5 B ^ ist = Wist R B0 W Xist ^t ; (3.4) W where (Xist ^t ) is computed based on the following conditions: (1) each black worker’s B0 B ) are rewarded at the same estimated prices observable Mincerian characteristics (Xist W (^t ) as his white counterpart and (2) each black worker in state s during year t receives as part of his wage rate the value of the unobservable traits of the average white worker in that state and year. Racial Bias Indexes Becker (1957) theory implies that the impact of an intensi…cation of competition on the relative demand, and hence the relative wage rates, of black workers depends positively on the taste for discrimination, holding other factors constant. We do not directly observe the taste for discrimination. Consequently, we compute and use several estimates of the degree of racial bias in each state. We develop two types of racial bias indices based on the accumulated stock of racial intermarriage in 1970. The “simple” racial bias index equals the di¤erence between the rate of intermarriage that would exist if married people were randomly matched and the actual intermarriage rate. The random rate equals 2P (1 P ), where P is the proportion of blacks among the married population. Larger values of 5 To connect this to equation (3.2) of the statistical model, note that the estimated conditional B0 W wage rate of black worker i is ^ W (X B ) = X ^ . st ist ist t 145 the simple racial bias index indicate that intermarriage occurs less in practice than if marriage pairings were random. We interpret larger values as (partially) re‡ecting racial bias. In the second type of racial bias index, we account for other factors that might induce the actual rate of intermarriage to deviate from the random rate. Intermarriage depends on the opportunities for interracial social contacts, so that the relative sizes of the black-white populations might independently a¤ect intermarriage (Blau (1977)). Furthermore, since the odds of interethnic unions increase with couples’ educational attainment (Massey and Denton (1987); Qian (1997); Rubinstein and Brenner (2010)), we also control for education and age. Speci…cally, based on the 1970s census, we estimate the following equation for all married couples (excluding couples in which either the husband or wife is neither white nor black) in the United States: Iis = bHis + cWis + dSs + is ; (3.5) where Iis equals one if couple i in state s is racially mixed and zero otherwise, His and Wis are vectors of age and education characteristics for the two spouses respectively, Ss are state characteristics, is is the unexplained component of intermarriage, while b, c, and d are coe¢ cients. Our benchmark speci…cation conditions on nine categories of education, along with age entered as a quartic. For state characteristics, we include the random intermarriage rate de…ned above along with the percentage of blacks among married couples. We experimented with numerous speci…cations, including 146 and excluding the random intermarriage rate and the percentage of blacks, changing the speci…cation of education and age controls, and conditioning on metropolitan and urban locations. These combinations produce the same conclusions. From equation (3.5), we compute the intermarriage racial bias index for each state. Let s equal the average value of is across couples in state s. Recognizing that minf s g < 0,we compute the racial bias index as Tes = Tes equals zero for the state with the largest s. s + maxf s g, so that We interpret large values as signaling a stronger taste for discrimination. Table 3.12 provides the value of the racial bias index, Tes ; for each state and the District of Columbia. Furthermore, Ts = 1 if Tes medianfTes g, and Ts = 0 if Tes < medianfTes g. The intermarriage racial bias index is positively correlated with survey-based mea- sures of racial prejudice. Table 3.1 (Panel A) shows that the intermarriage racial bias index is positively related to three survey-based measures of racial prejudice used by Charles and Guryan (2008) in their study of relative wages and racial prejudices: (1) the fraction of whites supporting a law against interracial marriage, (2) the fraction of whites would not vote for a black president, and (3) the fraction of whites supporting the right to segregate neighborhoods by race. Thus, the racial bias index based on intermarriage in 1970 is closely associated with subjective measures of racial attitudes measured over the period 1972 to 2004. The intermarriage racial bias index is negatively correlated with the relative wage rates of black workers. Panel B of Table 3.1 shows that the intermarriage racial bias index is strongly, negatively associated with black’s relative wage rates in the years prior to both inter- and intrastate bank deregulation, suggesting that the racial bias 147 index captures cross-state di¤erences in the relative demand for black workers. As emphasized by Becker (1957), the relative supply of blacks in the workforce should also a¤ect the relative wage rate of black workers. We assess this prediction by including the proportion of blacks in the workforce in 1970 in the relative wage rate regression. As shown in Panel B of Table 3.1, states in which black workers compose ten percent or more of the labor force tend to have low relative wage rates for black workers. The intermarriage racial bias index, however, remains negatively and signi…cantly associated with the relative wage rate of black workers. We also use the Charles and Guryan (2008) survey-based estimates of the degree of racial prejudice of the marginal …rm in each state to categorize high- and lowracial bias states. As shown in Panel B of Table 3.1, states with above the median levels of this marginal racial prejudice indicator have signi…cantly lower relative wages of black workers. Nonetheless, the racial bias index based on racial intermarriage remains negatively and signi…cantly associated with blacks’relative wages even when controlling for the marginal racial prejudice indicator and when controlling for both the marginal racial prejudice indicator and the proportion of blacks in the workforce. For the purposes of this paper, there are advantages to using the intermarriage racial bias index rather than survey-based measures of racial attitudes. The intermarriage racial bias index is based on actual choices made prior to deregulation not survey responses made during the period of deregulation. Moreover, our empirical strategy requires that the measure of racial bias is invariant to bank deregulation and the resultant change in competition. If we di¤erentiate states based on a measure of racial bias that itself re‡ects the e¤ects of deregulation on the relative demand 148 and supply of black workers, this will confound our strategy of identifying the causal impact of product market competition on the relative demand for black workers. The racial attitude surveys, however, are conducted during the period of bank deregulation. Thus, the Charles and Guryan (2008) estimate of the degree of racial prejudice of the marginal …rm, which is based on the racial attitude surveys and the relative supply of black workers, includes the e¤ects of bank deregulation on product market competition. Although this is not a problem for their study of the connection between the taste for discrimination of the marginal …rm and relative wage rates, these survey-based measures are inappropriate for our purposes. Furthermore, when assessing the impact of competition on blacks’relative wages, Becker (1957) framework advertises the advantages of our intermarriage racial bias index rather than an estimate of the taste for discrimination of the marginal …rm, which incorporates both the relative demand for and supply of black workers. Theory does not necessarily predict that an increase in competition will increase blacks’ relative wages more in states with initially low relative black wage rates because low relative wage rates re‡ect both demand and supply. Rather, theory implies that an intensi…cation of competition will increase the relative demand for black workers only in states with a su¢ cient taste for discrimination. Holding the relative supply of black workers constant, therefore, theory predicts that an intensi…cation of competition will increase blacks’ relative wages in states with a su¢ ciently high taste for discrimination. This is what we evaluate. We distinguish states by their overall taste for discrimination, and assess whether competition boosts blacks’relative wage rates while conditioning on other state and year characteristics, including the relative 149 supply of black workers. In sum, we include state and year …xed e¤ects and evaluate whether an exogenous increase in competition boosts the relative demand for black workers more in states with larger values of the intermarriage racial bias index. Measuring racial bias with error will bias the results against …nding a statistically signi…cant connection between racial bias, competition, and the black-white wage gap. We do not require that the racial bias index is a perfect measure of racial attitudes. We simply require that it provides some information on racial prejudices across states. 3.4.3 Econometric Design The Impact of Competition on Blacks’Relative Wages To obtain a consistent estimate of the impact of competition, as measured by the rate of incorporations, on relative wages, we need an instrumental variable that is correlated with the rate of new incorporations but not independently correlated with blacks’relative wages. It is important to use instrumental variables because blacks’ relative wages could a¤ect the actual entry of …rms. For example, …rms could enter to exploit the opportunity to hire less expensive labor in states with a large racial wage gap. If this occurs, OLS will underestimate the causal impact of competition on blacks’relative wages. Thus, to assess the causal impact of competition on racial wage discrimination while di¤erentiating economies by their tastes for discrimination, our identifying strategy assumes that (i) banking deregulation is exogenous to blacks’ relative wages, (ii) new incorporations per capita is a fair proxy for competition, (iii) 150 the racial bias index re‡ects tastes for discrimination, and (iv) bank deregulation does not a¤ect the racial wage gap beyond its impact on the rate of new incorporations. From equations (3.2) and (3.4), the following second stage regression captures the causal relationship of interest, ^ ist = R 0 Nst + 1 Nst Ts + s + 0t + 1t + (3.6) ist ; where the predicted value of the log of new incorporations per capita (Nst ) is obtained from the …rst stage regression using bank deregulation as an instrument:6 0 Nst = Dst 0 0 + Dst 1 Ts + s + 0t + 1t + st ; (3.7) where Dst is a vector indicating years since bank deregulation, Ts equals one in high racial bias states and zero in low racial bias states, coe¢ cients, s is a vector of state-speci…c e¤ects, in low and high racial bias states respectively, ist 0t 0 and and 1t 1 are corresponding represent time e¤ects is an error term composed of a person speci…c idiosyncratic shock to relative wages and any unobserved state-year …xed e¤ects, and st is an error term. The standard errors are clustered at the state- year level throughout the analyses. As emphasized, we assess whether the impact of competition on the racial wage gap depends on the degree to which states have a stronger or weaker taste for discrimination. Our estimation strategy allows us to relax the standard exclusion restriction that 6 The …rst stage regression is conducted at the individual level, so it is weighted by the proportion of black workers in each state. 151 bank deregulation only a¤ects blacks’ relative wages through its a¤ect on the rate of new incorporations. We conduct a quasi triple di¤erence estimation by including state and year …xed e¤ects and by separately analyzing states with above and below the median value of the racial bias index. This yields an estimate of the di¤erential impact of an increase in the rate of new incorporations on blacks’ relative wages in high and low racial bias states, i.e., we assess whether > 0. To obtain a 1 consistent estimate of this di¤erential impact using 2SLS, we do not require the standard exclusion restriction to hold. Rather, we simply require that any bias arising from bank deregulation a¤ecting blacks’relative wages beyond its impact through the rate of new incorporations is the same in high and low racial bias states, which we assess empirically below.7 7 More formally with regard to the exclusion restriction, consider a modi…ed version of equation (3.6) that allows bank deregulation to a¤ect blacks’relative wages beyond new …rm entry (Nst ): ^ ist = R 0 Nst + 1 Nst Ts + s + 0t + 1t 0 + Dst + ist where re‡ects the direct impact of bank deregulation on blacks’ relative wages. Using equation (3.7), it is straightforward to show that: p lim ( 2SLS 0) = 0 + ; p lim ( 0 + 1) 2SLS = 0 + 1 + 0 0 + 1 and p lim 2SLS 1 = p lim ( 0 + 2SLS 1) p lim ( 0) 2SLS = 1 + 0 + = 1 0 1 1 0( 0 + 1) Under the standard exclusion restriction, bank deregulation has no direct impact on blacks’relative wages, = 0, so that 2SLS provides consistent estimates for 0 and 1 . However, even if bank deregulation in‡uences blacks’ relative wages directly, 2SLS provides a consistent estimate of 1 as long as the impact of bank deregulation on the entry of new …rms is the same in high and low racial bias states, i.e., if 1 = 0: Moreover, if both 1 > 0 and > 0. 2SLS will underestimate the di¤erential impact of competition on blacks’relative wages in high racial bias states, so that p lim 2SLS < 1 . We evaluate both of 1 these conditions below. 152 Reduced Form Estimator We also assess the reduced form impact of banking deregulation on black workers’ relative wages by estimating the following wage equation using OLS: 0 ^ ist = Dst R where 0 and 1 0 0 + Dst are coe¢ cients, s 1 Ts + s + 0t + 1t + is a vector of state …xed e¤ects, vectors of time e¤ects in low and high racial states respectively, and term. For simplicity, we use s; 0t ; and 1t (3.8) ist ; 0t ist and 1t are is an error as generic representations of state and year …xed e¤ects, while recognizing that the actual values will di¤er across equations. In extensions of equation (3.8), we assess the dynamic e¤ects of deregulation on black workers’relative wages by allowing the relationship between relative wages and deregulation to vary by each year before and after bank deregulation. Although OLS produces unbiased estimates of the impact of bank deregulation on the racial wag gap under standard identifying assumptions, including the assumption that deregulation is uncorrelated with ist , OLS does not necessarily identify a channel running from competition to blacks’relative wages. Identifying this channel motivates our use of the two-stage least squares (2SLS). Examining both the 2SLS and reduced form speci…cations provides a more comprehensive assessment of the determinants of racial wage discrimination than using only one method. If the rate of new incorporations is a sound proxy for competition and bank deregulation is a valid instrument, then the 2SLS estimator provides information on the causal impact of competition on blacks’ relative wage rates, putting 153 aside for now the complexities associated with accurately measuring the relative wage rates of equivalent black and white workers. Yet, the reduced form analysis is independently valuable. It provides information on whether bank deregulation disproportionately bene…ted an historically disadvantaged group in the economy, expanding our understanding of the impact of …nancial sector policies on the economy. 3.5 3.5.1 Results Preliminaries Our empirical analysis rests on the assumption that the cross-state timing of bank deregulation was not a¤ected by the racial wage gap. Figure 3-1 shows that neither the level of the estimated wage gap before deregulation (Panel A) nor its rate of change prior to deregulation (Panel C) explains cross-state di¤erences in the timing of interstate bank deregulation. Panels B and D of Figure 3-1 con…rm these …ndings for the case of intrastate deregulation. The size of the “bubbles”in the …gures represent the size of the black workforce in each state, which corresponds to the weighting in the relative wage regressions below. Our strategy also requires that bank deregulation increases the rate of new incorporations in the overall economy. In Table 3.2, we show that both interstate bank deregulation and intrastate branch deregulation exert a strong, positive impact on the log of new incorporations per capita over time. In columns (1) (3), we use simple dummy variables that equal zero before a state deregulates and one afterwards. Inter- 154 state deregulation enters signi…cantly and positively, but intrastate does not, which is consistent with the …ndings in Black and Strahan (2002). The results in Table 3.2 emphasize that the positive impact of deregulation on the rate of new incorporations grows over time. In columns (4) (6), we include the number of years since deregulation and its quadratic. Interstate and Intrastate equal the number of years since interstate and intrastate bank deregulation respectively, and equal zero before deregulation. Both linear terms enter positively and signi…cantly, while the quadratic terms are negative, but the coe¢ cients are an order of magnitude smaller.8 Economically, the coe¢ cients in columns (4) and (5) indicate that …ve years after either inter- or intrastate deregulation the rate of new incorporations is about 10 percent greater than before deregulation. Furthermore, simultaneously deregulating inter- and intrastate restrictions boosts the rate of new incorporations by 18 percent after …ve years as shown in column (6). Figure 3-2 more fully illustrates the positive, dynamic impact of both interstate and intrastate deregulation on the rate of new incorporations. In Figure 3-2, we trace out the year-by-year relationship between both interstate and intrastate deregulation and the logarithm of new incorporations. We do this for two samples of states, those with above the median level of the racial bias index and those with below median levels. Speci…cally,we report estimated coe¢ cients from the following regression: Nst = + 8 1 Inter 9 +:::+ 18 Inter+9 + 1 Intra 9 +:::+ 18 Intra+9 + s + t +"st ; (3.9) The impact of each form of deregulation on competition grows over time, reaching a maximum about a decade after interstate deregulation, and over two decades after intrastate deregulation. where Inter j equals one for the j th 155 year before interstate deregulation, and Inter+k equals one for the k th year after interstate deregulation, while Intra j equals one for the j th year before intrastate deregulation, and Intra+k equals one for the k th year after intrastate deregulation. These dummy variables equal zero in other years. We present results starting 9 years before each form of bank deregulation and trace out the year-by-year dynamics of the relationship between deregulation and the wage gap until 9 years after each type of bank deregulation. The year of deregulation is omitted and the regressions include state ( s ) and year ( t ) …xed e¤ects. After detrending the series, Figure 3-2 illustrates the level and trend of the logarithm of new incorporations following each type of bank deregulation relative to the level and trend before deregulation.9 There are three critical observations from Figure 3-2. First, interstate and intrastate bank deregulation boost the rate of new incorporations. This is crucial since we use bank deregulation to identify an exogenous intensi…cation of competition. Second, the impact of bank deregulation on the rate of new incorporations is not immediate. The e¤ect of bank deregulation on the rate of new incorporations is still growing after …ve years. If bank deregulation a¤ects blacks’ relative wages by increasing the rate of new incorporations, therefore, we should also …nd that the dynamic impact of deregulation on black’s relative wages materializes over time. Third, the positive impact of inter- and intrastate bank deregulation on the rate 9 Speci…cally, we compute the trend in the coe¢ cients on the dummy variables on bank deregulation prior to deregulation. We then detrend the entire series of estimated coe¢ cients based on the pre-deregulation trend. The resulting …gure illustrates the level and trend of the logarithm of new incorporations after bank deregulation relative to the patterns before deregulation. 156 of new incorporations occurs in both states with above the median level of the racial bias index and in states with below the median level of the racial bias index. The the marginal impact of intrastate deregulation on the rate of new incorporations in low racial bias states is less pronounced than in high racial bias states.10 Though the impact of bank deregulation on new incorporations does not have to be identical in high and low racial bias states, our empirical strategy requires that deregulation boosts the rate of new incorporations in both high and low racial bias states because we propose to evaluate whether the marginal impact of an exogenous increase in competition is greater in high racial bias states. 3.5.2 Bank Deregulation and Blacks’Relative Wages Reduced Form Analyses of Bank Deregulation We next assess the reduced form impact of Interstate and Intrastate on the relative ^ ist . For each form of deregulation, we present three wage rates of black workers R speci…cations. First, blacks’relative wages are regressed on bank deregulation using the full sample. Second, we add an interaction term of deregulation and the racial bias dummy for each state, which equals one if the value of the racial bias index is greater than or equal to the sample median and zero otherwise. As suggested by theory, the impact of competition-enhancing bank deregulation on blacks’ relative wages should be greater in more racially biased states. Third, rather than include an 10 As developed in footnote (7), when bank deregulation has a larger e¤ect on the rate of new incorporations in high racial bias states, the 2SLS estimator will tend to underestimate the di¤erential impact of the rate of new incorporations on blacks’relative wages in high versus low racial bias states. 157 interaction term, we split the sample by the median value of the racial bias index, which allows the coe¢ cients on state and year …xed e¤ects to di¤er across the two subsamples. Throughout the analyses, we include state and year …xed e¤ects. Table 3.3 shows that bank deregulation has a large, signi…cant impact on the relative wage rates of black workers in states with su¢ ciently high values of the racial bias index. In the regressions including the interaction of deregulation with the racial bias dummy, the impact of deregulation on blacks’ relative wages is increasing in the state’s racial bias index. The results hold for both inter- and intrastate bank deregulation. When splitting the sample between high and low racial bias states, the results indicate that a drop in entry barriers triggers a bigger increase in the relative demand for black workers in more racially biased economies. Furthermore, by splitting the sample between high and low racial bias states, we employ a quasi-triple di¤erence speci…cation. In particular, there might be concerns that even though bank deregulation di¤ers in its timing across states, there might be a confounding factor that reduces racial discrimination and is coincident with the state-speci…c timing of bank deregulation. By showing that bank deregulation only increases blacks’relative wages in high racial bias states as predicted by theory, this reduces the possibility that an unobserved state-year e¤ect is driving the results, and it is fully consistent with the view that intensi…ed competition reduces the manifestation of racial prejudices in labor market outcomes. The estimated reduction in the racial wage gap from bank deregulation is economically meaningful. Consider column (4) of Table 3.3, which provides the regression results for states with above the median value of the racial bias index. Among these 158 states, deregulation boosts the wage rates of black workers by 6 percentage points more than their white counterparts after …ve years (6 = 0:012 5 100). Since the average racial wage gap in these high-bias states was 20 percent in 1976, the results suggest that interstate deregulation eliminates almost 30 percent of the initial racial wage gap. The results are virtually identical when using Intrastate, as shown in column (8). Dynamic Analysis of the E¤ect of Bank Deregulation We next illustrate the dynamic relation between bank deregulation and the relative wages of blacks. In Figure 3-3, we trace out the year-by-year relationship between deregulation and the wage gap by including a series of dummy variables in equation (3.8) for inter- and intrastate deregulation respectively. Speci…cally, D j equals one for the j th year before deregulation, and D+k equals one for the k th year after deregulation. These dummy variables equal zero in other years. The year of deregulation is omitted and the regressions include state and year …xed e¤ects. In examining the dynamic impact of deregulation on the racial wage gap, we use two samples of states. In Panel A of Figure 3-3, the subsample includes states with above the median values of the racial bias index. Panel B reports the dynamic relation between the relative wage rates of black workers and bank deregulation for the subsample of states with below the median values. The dashed line reports the estimated coe¢ cients on the interstate deregulation dummy variables, while the solid line provides the estimated coe¢ cients on the intrastate deregulation dummy variables. 159 Three crucial messages emerge from Figure 3-3. First, the impact of both interstate and intrastate bank deregulation on blacks’ relative wages is much greater in states where the racial bias index is above the median than in states with lower values of the racial bias index. For example, the impact of interstate bank deregulation on blacks’relative wages rises over time in states with high values of the racial bias index, while interstate bank deregulation has virtually no e¤ect on relative wage rates in states with low values of the racial bias index. Second, there is no evidence that trends or innovations in the wage gap precede either interstate or intrastate bank deregulation. Rather, blacks’ relative wages rise after bank deregulation for an extensive period in states with high values of the racial bias index. Third, the impact of deregulation on black’s relative wages grows over time. This is consistent with the dynamics of the relationship between deregulation and the rate of new incorporation documented in Figure 3-2 and Table 3.2. While demonstrating the powerful impact of bank deregulation on the racial wage gap, these results do not provide direct evidence on the underlying causal mechanisms. We now examine the relationship between the rate of new incorporations and blacks’ relative wages to assess whether, and under which conditions, an exogenous increase in the rate of new incorporations reduces the black-white wage gap. 160 3.5.3 Competition and Blacks’Relative Wages Reduced Form Analyses of Competition In examining the relationship between competition and the racial wage gap, we begin with reduced form OLS regressions. In Table 3.4, the dependent variable is blacks’ ^ ist . The key regressor is the log of new incorporations per capita, relative wages R which we use as a proxy for competition. The estimation is conducted on the full sample, and we also split the sample into states with below and above the median level of the racial bias index. In Panel A, we use the benchmark measure of blacks’relative wages, which is computed while conditioning on the standard Mincerian characteristics, education and potential work experience. In Panel B, we use an alternative measure of blacks’ relative wages that also conditions on occupation, as discussed above. There is a strong, positive association between the rate of new incorporations and the relative wages of black workers in states with above the median values of the racial bias index (column 3). The OLS estimates indicate that ten percent increase in the rate of new incorporations is associated with a 1.4 percent increase in blacks’relative wages in high racial bias states. In contrast, there is no relationship between the wage gap and our proxy for competition in states with low values of the racial bias index (column 2). These results hold both when using the benchmark, Mincerian measure of blacks’relative wages (Panel A) and also when conditioning on occupation (Panel B). 161 2SLS Analyses of Competition The …nal six columns of Panel A and Panel B of Table 3.4 report 2SLS estimates, where two di¤erent sets of instrumental variables are used to identify changes in the rate of new incorporations. First, the “Linear”instruments simply include Interstate and Intrastate. Second, the “Non-Parametric” instruments included dummy variables for each year before and after both interstate and intrastate deregulation. These instruments are drawn from the analyses reported above in Table 3.2 and Figure 33. Furthermore, in reported robustness tests, we …nd that using Interstate and Intrastate plus their quadratic terms as instruments produces similar results. As shown, the instrumental variables pass the validity tests. They signi…cantly explain new incorporations as shown by the F-test of the excluded instruments. Furthermore, the instruments pass the test of the over-identifying restrictions (OIR test), meaning that the hypothesis that the instruments only a¤ect blacks’relative wages through their e¤ect on new incorporations is not rejected.11 The exogenous increase in the rate of new incorporations dramatically boosted the wage rates of black workers relative to their white counterparts in states with above the median values of the racial bias index. As reported in columns (6) and (9) of both Panels A and B, an acceleration of the rate of new incorporations increased blacks’relative wages in high racial bias states. In contrast, the results in columns 11 In unreported robustness tests, we also show that the results are not driven by states in which deregulation did not induce an increase in competition, which would run counter to theory and our identi…cation strategy. Thus, we run the …rst-stage regression while omitting each state one-at-atime. We then …nd which states are “‡attening” the estimated relationship between competition and deregulation in the …rst stage. When we eliminate these states, the results strengthen. This robustness test suggests that the e¤ects of deregulation on racial discrimination are driven by states in which the "treatment" is a¤ecting product market competition, not by some spurious channel. 162 (5) and (8) indicate that a faster rate of new incorporations did not increase blacks’ relative wages in states with below the median values of the racial bias index. The economic impact the rate of new incorporations on blacks’relative wages is large in states with above the median level of the racial bias index. With either set of instrumental variables, the estimates indicate that a ten percent acceleration in the rate of new incorporations increases blacks’relative wages by about 2.5 percent in high racial bias states.12 Combining these results with those in Figure 3-2, the results suggest that bank deregulation boosted the rate of new incorporations by over 20% after …ve years in high racial bias states, which in turn increased blacks’ relative wages by about …ve percent in these same states. These estimates indicate that by increasing competition, bank deregulation boosted blacks’relative wages by one-quarter of the initial racial wage gap in these states, which equaled, on average, 20 percent in the years before bank deregulation. Competition and Blacks’Relative Wages: Sensitivity Analyses The results are robust to using either the Charles and Guryan (2008) measure of racial prejudices (CG) or the intermarriage racial bias index (LLR) to categorize states as high- or low-racial bias states. Table 3.5 presents the OLS and 2SLS analyses of the relation between the racial wage gap and the rate of new incorporations. We use the linear instrument set and compute blacks’relative wages conditional on standard 12 The 2SLS parameter estimate is larger than the OLS estimate. This is consistent with the reverse causality argument made above. Speci…cally, if …rms are attracted to states where blacks’ relative wages are particularly low, OLS will underestimate the impact of a lowering of entry barriers on blacks’relative wages. 163 Mincerian traits and occupation. We use a common sample of states that is slightly smaller than in Table 3.4 because the CG measure is unavailable for Hawaii, Idaho, Maine, Nebraska, Nevada, and New Mexico. The strong positive impact of the rate of new incorporations on blacks’relative wages is robust to using the CG racial prejudice indicator to classify states. In states with above the median values of the two racial bias indicators, the log of new incorporations per capita is positively associated with blacks’relative wages. Figure 3-4 shows that the results are robust to considering the full range of possible combinations of (1) estimation strategy (OLS and 2SLS), (2) method for computing blacks’relative wages (either conditioning on standard Mincerian controls (R) or also conditioning on occupation (Ro)), (3) method for categorizing states by taste for discrimination (LLR or CG), and (4) using linear or non-parametric instrumental variables (Linear or N on param:). Figure 3-4 plots each point estimate along with its 95% con…dence interval. As shown, the results are robust. In terms of the instrumental variable results, there is only one speci…cation in which the rate of new incorporation does not enter positively and signi…cantly at the …ve percent level, and instead enters with a p-value of (0:10). This exception involves using the CG indicator to de…ne racial attitudes, and we have already discussed the advantages, in the context of our particular study, of using the intermarriage racial bias measure (LLR). 164 3.6 The E¤ect of Competition on Segregation 3.6.1 Racial Prejudices, Competition, and Segregation Besides making predictions regarding relative wages, Becker (1957) taste-based theory of discrimination also predicts that when employers are heterogeneous in both productive quality and the “disutility” they receive from employing black workers, there will be racial segregation as black workers are hired by the least racially biased employers. Indeed, if …rms are similar except for the racial prejudices of employers, segregation will reduce racial wage di¤erentials as workers sort according to the racial preferences of employers. This led Welch (1975) to emphasize the segregation prediction of the taste-based theory of discrimination. While racial wage di¤erentials are a fundamental measure of labor market discrimination and the focus of our examination, segregation o¤ers an additional margin along which to assess whether the relations between competition and the racial characteristics of labor markets are consistent with the taste-based theory. More speci…cally, the taste-based theory suggests that an intensi…cation of competition will reduce segregation. Lowering entry barriers allows new employers with less of a taste for discrimination than existing employers to enter. This increases the number of employers willing to hire black workers at prevailing wage rates. By expanding the employment opportunities of blacks, competition will reduce segregation. If our earlier results on blacks’relative wages re‡ect the causal impact of intensi…ed competition on how racial prejudices a¤ect labor markets, then we should also observe a reduction in segregation following an intensi…cation of competition. 165 3.6.2 The E¤ect of Competition on Segregation: Results In light of this testable implication, we turn back to the data and evaluate the impact of an intensi…cation of competition on the racial allocation of workers while di¤erentiating states by the degree of racial prejudice. To analyze the same time period and workers used in our examination of blacks’relative wages, we study the racial composition of workers at the industry level, using data on the 144, 3-digit industry categories in the CPS. If there are cross-industry di¤erences in the racial prejudices of employers, then the taste-based theory predicts that competition will expand the cross-industry labor market opportunities available to black workers and induce blacks to move to industries that were previously dominated by whites. While one may question whether the racial prejudices of employers di¤er by industry, Becker (1957) provides an economic rational for examining segregation at the industry level: competition may di¤er by industry. If there are cross-industry di¤erences in entry barriers, and hence competition, employers with stronger racial prejudices will have comparatively greater success in less competitive industries holding other factors constant. Thus, while the keystone of our analysis is blacks’relative wages, racial integration across industries provides additional evidence on the mechanisms linking competition and racial discrimination in labor markets. Consequently, we construct and use several measures of the extent to which an industry is particularly “white.” First, we calculate the share of white workers by industry. Second, since the racial composition of workers in an industry might simply re‡ect the human capital needs of the industry in conjunction with the di¤erential 166 racial composition of human capital skills, we also estimate the degree to which the proportion of white workers in an industry is greater than the proportion explained by the underlying characteristics of workers. To do this, we regress (for each year) the proportion of white workers in each of the 144 industries on the characteristics of the white workers in that industry, including education, a quartic in potential experience, as well as occupation and state …xed e¤ects, i.e., the same set of regressors that we employ to generate wage residuals. We collect the average residuals in each industry. These provide crude and residual (“unexplained”) measures of the “whiteness” of each industry. Third, motivated by Ashenfelter and Hannan (1986), we calculate the proportion of white managers in each industry and use this proportion as a measure of the degree to which an industry is dominated by whites. Fourth, we also construct the unexplained proportion of white managers, using the same conditioning regressors. We next estimate the impact of competition on the racial composition of the industry in which each black worker is employed. We use the same speci…cation employed in our relative wage regressions, except the dependent variable is one of the measures of the “whiteness” of the industry in which each black works. Thus, we regress industry whiteness on the log of new incorporations per capita, controlling for state and year …xed e¤ects. We do this using OLS and 2SLS. We divide states by the degree of racial bias, using both the LLR and CG measures of racial bias to categorize states. Thus, we evaluate whether an exogenous increase in competition induces black workers to move to “white” industries, while di¤erentiating states by racial bias. In the analyses, we obtain the same results whether we use the crude or residual measures of the degree to which an industry is composed of white workers 167 or managers. For simplicity, we present the results for the unexplained proportion of white workers and the crude measure of the fraction of white managers. Consistent with the taste-based theory, Table 3.6 indicates that an acceleration of the rate of new incorporations in high racial bias states induced blacks to work in “whiter” industries. These results hold when examining (1) the unexplained proportion of white workers and (2) the proportion of white managers. The results hold when using OLS or 2SLS, and whether we divide states by the LLR or the CG indicator of racial prejudices. The sizes of the estimated coe¢ cients suggest that the impact of new incorporations on the reallocation of black workers across industries is economically small, but not inconsequential. For instance, consider the sample of high racial bias states based on the racial bias index (LLR) and the results using the proportion of white managers. The estimated coe¢ cient indicates that a one standard deviation increase in the log of new incorporations per capita (0.4) boosts the proportion of white managers in which the average black worker is employed by 0.0032 (0.4*0.08), which is almost one-tenth of the cross-industry standard deviation of white managers (0.04). In sum, the …ndings are consistent with the view that intensi…ed competition reduced racial segregation in the workforce. 3.6.3 Competition and Blacks’ Relative Wages Within Industries Given this …nding on the movement of black workers to historically white industries, we were concerned that the earlier results on blacks’ relative wages could re‡ect a 168 shift of black workers to better paying industries, rather than an increase in blacks’ relative wages within industries. To assess whether the shift of black workers to white industries accounts for the increase in blacks’relative wages, we evaluate the impact of an increase in the rate of new incorporations on blacks’ relative wages, where we not only compute blacks’relative wages by conditioning on education, potential experience, and occupation, but also by conditioning on industry. As noted above, we recognize the problems with this conditioning since an individual’s industry could be endogenously explained by the rate of new incorporations. Nonetheless, as a robustness check, we compare the wages of black workers with the same observable traits as their white counterparts who are working in the same industry and the same occupation to see whether the relation between the rate of new incorporations and blacks’relative wages is accounted for by the movement of black workers to higher paying industries. The results in Table 3.7 suggest that the intensi…cation of competition boosted blacks’ wages relative to comparable white workers within the same industry and occupation. Increased racial integration does not fully account for the increase in blacks’ relative wages following the boost in the rate of new incorporations. Both results –the increase in blacks’relative wages and the increase in racial integration in the workplace –are consistent with the taste-based view of racial discrimination. 169 3.7 Robustness Checks In this section, we address concerns about several factors that could confound our ability to draw accurate inferences about the impact of competition on racial wage discrimination. Some of these factors work against the reported …ndings, leading us to underestimate the bene…cial e¤ects of bank deregulation and the rate of new incorporations on blacks’relative wages. In these cases, we simply discuss our robustness tests without presenting tables. Other factors either play a central role in Becker (1957) theory or potentially lead us to overestimate the impact of competition on racial discrimination. In these cases, we present correspondingly more information. 3.7.1 Relative Hours Worked We were concerned that blacks’relative wages could also rise if deregulation induced the labor supply curve of black males to shift leftward. If this occurs, the working hours of blacks could actually fall after deregulation relative to those of whites. Table 3.8 reports the e¤ects of bank deregulation and the log of new incorporations per capita on the relative working hours of blacks in high racial bias states using two approaches. We examine high racial bias states because this is where the rate of new incorporations increased blacks’ relative wages. In the …rst approach, we trace the impact of bank deregulation, through the rate of new incorporations, to blacks’relative wages. We then examine the impact of these projected relative wages on blacks’ relative annual hours worked. If an outward shift in the demand curve is causing the increase in blacks’ relative wages, then we expect to …nd a positive 170 coe¢ cient on blacks’relative wages in the relative working hours regression. In the second approach, we examine the impact of the log of new incorporations per capita on the relative working hours of blacks without tracing the e¤ect through relative wages. Speci…cally, we reproduce the 2SLS analyses in Table 3.4 except that the dependent variable is the di¤erence between the actual number of hours worked of each black worker and the projected annual hours worked of a white worker with identical traits. The di¤erence between the actual and projected hours worked re‡ects the racial gap in hours. We use bank deregulation to identify an exogenous increase in new incorporations and assess the impact on this gap in working hours. To compute relative working hours, we …rst estimate a labor supply equation every year on a sample of white males, while conditioning on state …xed e¤ects and the same Mincerian characteristics used in the wage equation. Then, we use the resulting coe¢ cient estimates to calculate the predicted number of hours worked of a white worker with each black worker’s characteristics. Finally, we compute the relative working hours of each black worker as the di¤erence between his actual and predicted working hours. Since there is a meaningful kink in the labor supply curve between working and not working, we use both OLS and Tobit speci…cations and also examine the subsample of blacks with positive working hours. We use a standard bootstrapping procedure to correct the standard errors since the regressors are estimated. We …nd that bank deregulation that increased the rate of new incorporations and boosted blacks’relative wages also increased the relative working hours of blacks. The evidence suggests that bank deregulation increased the relative demand for black 171 workers. As shown, the impact is particularly pronounced among workers. This suggests that while deregulation increased the relative demand for black workers, bank deregulation did not signi…cantly attract new black workers into the workforce. Most important given the focus of this paper, the Table 3.8 results clearly demonstrate that bank deregulation and competition did not shift black’s labor supply curve to the left. 3.7.2 Selection, Migration, and Self-Employment We were concerned that changes in the skill composition of black males in the economy could a¤ect our evaluation of blacks’relative wages. Consequently, we calculate the projected wage rates for all working age (non-institutionalized) blacks in each state, whether they are working or not. We do this using the estimated returns to observable traits from equation (3.3) and using the actual traits of each black male. In this way, we compute the value of observable traits of all black males. Then, we evaluate the impact of bank deregulation on the composition of skills in the workforce. Table 3.9 provides regression results of the projected wage rates of all relevant black males on a dummy variable if the person works, Interstate, and the interaction between Interstate and the dummy variable for working or not, as well as state and year …xed e¤ects. There are similar regressions for Intrastate. The summation of the coe¢ cients on Interstate and the interaction term provide information on whether the average value of the traits of workers changes after deregulation. The coe¢ cient on Interstate provides information on the change in the average value of the traits 172 of individuals who are not working following deregulation. Deregulation did not have a signi…cant e¤ect on the average value of the traits of black workers. There is no evidence that bank deregulation substantively a¤ected the skill composition of black workers. To the extent that observable traits are correlated with unobservable characteristics, these results further imply that the composition of unobservable traits did not change much following bank deregulation. Deregulation could also a¤ect migration across states. To assess this, we estimate the e¤ect of deregulation on the fraction of black males within states. We …nd that the share of black males within states increased slightly after deregulation. This is consistent with a situation in which deregulation boosted the rate of new incorporations, reduced the racial wage gap, and attracted blacks from other states. Yet, as shown in Table 3.9, the net compositional changes of blacks in the economy due to deregulation did not have much of an e¤ect on the skill composition of working blacks. There is no indication that migration leads us to overstate the bene…cial e¤ects of deregulation. Similarly, the boost in blacks’relative wages could attract black males with comparatively low unobserved skills into the labor force, leading us to underestimate the degree to which the rate of new incorporations reduces racial wage di¤erentials. A quantile regression at the median helps in assessing the importance of this potential bias by putting less weight on entrants of black workers with low unobserved skills. We …nd no evidence that selection based on unobservables is causing us to underestimate the true e¤ect of the rate of new incorporations. While the log of new incorporations per capita increases the relative demand for black workers, the number 173 of new black males pulled into the labor force is relatively small, such that the median regression yields virtually identical results to the OLS coe¢ cient estimates. 3.7.3 Swimming Upstream Biases could arise from changes in the “prices”of unobserved skills. Although national trends in returns to unobserved skills will not a¤ect our results because we control for year …xed e¤ects, the intensi…cation of competition when a state deregulates could increase returns to unobservable traits. If the average white worker has more of these unobserved traits than the average black worker, the average wage rate of whites will rise relative to that of blacks. This e¤ect will cause the estimated value of blacks’ relative wages to fall, even though racial discrimination is not rising. Under these conditions, we will underestimate the true, positive e¤ect of deregulation on the relative wages of blacks. This is sometimes called “swimming upstream” (Juhn et al. (1991); Blau and Kahn (1997); Blau and Kahn (2000); and Mulligan and Rubinstein (2008)). To assess the importance of swimming upstream, we follow the literature and use quantile regressions. The goal is to compare black and white workers that are more similar in unobserved skills than when using OLS, which compares averages from both groups. In unreported regressions, we con…rm the existence of swimming upstream, suggesting that we are underestimating the bene…cial e¤ects of bank deregulation on blacks’relative wages when using OLS. The median regressions produce similar coe¢ cient estimates to those from OLS. Moreover, in moving from lower quantiles to 174 higher quantiles, we …nd that deregulation reduced a larger proportion of the racial wage gap. Under the assumption that the average white has more unobserved skills than the average black, these …ndings are consistent with the view that the racial wage gap closed more among white and black workers with comparable unobserved skills. 3.7.4 Racial Discrimination or the Poor Since bank deregulation exerts a disproportionately positive impact on the poor and blacks are on average comparatively poor (Beck et al. (forthcoming)), the current paper’s analyses could re‡ect this income distributional e¤ect, rather than the impact of bank deregulation and competition on blacks in particular. Three observations, however, suggest that this is not the case. First, bank deregulation and the rate of new incorporations boosted blacks’ relative wages in states with a high degree of racial bias. This is di¢ cult to reconcile with the view that our results simply re‡ect a tightening of the distribution of income. Second, the results hold when computing relative wages conditional on occupation and industry. Thus, our …ndings indicate that even within low-paying (and high-paying) occupations and industries, blacks’relative wages rose with competition. Third, and most directly, we perform a rank analysis and compare the change in blacks’relative wages with those of comparable whites across the full distribution of relative wage rates. If deregulation is simply helping the poor, we should not see that blacks converge toward whites at each point in the wage distribution. 175 The results show that bank deregulation, and the accompanying boost in the log of new incorporations per capita, disproportionately helped black workers across the full distribution of wages. Figure 3-5 shows the rank plot for the high racial bias states, and for the sample of states with below the median level of the racial bias index. The solid and dashed lines represent the location of blacks within the conditional log hourly wage distribution of whites before and after deregulation respectively. The median black among the high racial bias states, for example, corresponds to the 28th percentile white worker prior to deregulation and the 32nd percentile white work after deregulation. The median black, therefore, gained four ranks in the white wage distribution as a result of deregulation, but only in high racial bias states. Consistent with the earlier results, there is little change in relative wage rates in the low racial bias states. These results suggest that deregulation exerted a particularly pronounced e¤ect on black workers. 3.8 Conclusions In this paper, we examine whether an increase in product market competition reduced the manifestation of racial prejudices in labor markets. As Becker (1957) argued, taste-based discrimination by employers can (1) produce an equilibrium gap between the wages of identical black and white workers and (2) produce racial segregation in the workforce. He further stressed that greater competition could erode the racial wage gap by reducing the impact of racial prejudices on the relative demand for black workers and reduce racial segregation by increasing the number of employers willing 176 to hire black workers. A central implication of the taste-based discrimination theory is that greater competition will reduce the black-white wage di¤erential and increase racial segregation only in economies where employers have a su¢ ciently strong “taste for discrimination.” Our results indicate that an exogenous intensi…cation of competition substantively boosted blacks’relative wages and reduced racial segregation in states with a su¢ ciently high degree of racial bias. In reduced form speci…cations, bank deregulation that lowered entry barriers facing non…nancial …rms reduced the racial wage gap. In 2SLS, we use bank deregulation to identify an exogenous intensi…cation of competition. We …nd that the resultant increase in competition eliminated more than one-…fth of the preexisting black-white wage di¤erential in high racial bias states over a …ve year period. Furthermore and critically, we …nd that intensi…ed competition reduced racial segregation, especially in high racial bias states. These …ndings suggest that competition reduced the impact of racial prejudices on blacks’relative wages and enhanced the opportunities of black workers. Looking forward, much work remains. The paper emphasizes the powerful role of competition in expanding the economic opportunities of minorities. By reducing racial wage di¤erentials, competition could also increase the incentives for blacks to acquire skills. Thus, future research might merge and extend taste-based and statisticalbased explanations of racial discrimination. This paper also advertises the need for additional research on …nance and economic opportunity. In this paper, we show that improvements in the functioning of banks substantively enhanced the economic opportunities of a disadvantaged group. These improvements materialize not because 177 banks make more loans to black entrepreneurs, but because improvements in banking disproportionately enhanced the labor market opportunities of blacks. 178 Figure 3-1: Trends and Innovations in the Relative Wage Rates of Blacks Prior to Bank Deregulation B. Year of intrastate deregulation Year of interstate deregulation A. 2000 1995 1990 1985 1980 -.5 -.4 -.3 -.2 -.1 0 Relative wage rates of blacks prior to interstate deregulation 2000 1995 1990 1985 1980 1975 .1 -.5 1995 1990 1985 1980 -.1 -.05 0 .05 Change in relative wage rates of blacks prior to interstate deregulation 0 .1 D. 2000 Year of intrastate deregulation Year of interstate deregulation C. -.4 -.3 -.2 -.1 Relative wage rates of blacks prior to intrastate deregulation .1 2000 1995 1990 1985 1980 -.1 -.05 0 .05 Change in relative wage rates of blacks prior to intrastate deregulation .1 179 Figure 3-2: The Impact of Deregulation on Entry of Firms A. Racial Bias Index > Median Percentage change in new corporations per capita .6 .4 .2 0 -.2 -10 -5 0 Years before/after deregulation Intrastate Deregulation 5 10 Interstate Deregulation Percentage change in new corporations per capita B. Racial Bias Index < Median .3 .2 .1 0 -.1 -10 -5 0 Years before/after deregulation Intrastate Deregulation 5 Interstate Deregulation 10 180 Figure 3-3: The Impact of Deregulation on the Relative Wage Rates of Blacks Percentage change in the relative wage rates of blacks A. Racial Bias Index > Median .3 .2 .1 0 -.1 -10 -5 0 5 Years before/after deregulation Intrastate Deregulation 10 15 Interstate Deregulation Percentage change in the relative wage rates of blacks B. Racial Bias Index < Median .3 .2 .1 0 -.1 -10 -5 0 5 Years before/after deregulation Intrastate Deregulation 10 Interstate Deregulation 15 181 Figure 3-4: The Impact of Log New Incorporations Per Capita on the Relative Wage Rates of Blacks: Di¤erent OLS and 2SLS Speci…cations Percentage point change in relative wages of blacks .45 .35 .25 .15 R G LL G ,C R o, ,C Ro L L m, R , ,R R m, a , r a m r R a a a r LL R C G r a m on-p G n-p -pa LL o, o, pa ,C on No R, , RNon- LS N r, R .05 arS r, R LS SN ar, ea S e a L S e n 2 i e n S i L ) L )2 Lin R )2 Li n S L 2S (10 SG LS (12 LL R (11 LS SL (9) G SL, C 2S o, LL 2S )2 , C (5), 2Ro 6) 8 ) ( ,R R, ( R 7 , S ( , S S S OL OL OL OL (2) (4) (1) (3) -.05 OLS 2SLS Specification 182 100 90 80 70 60 50 40 30 20 10 0 Percentile of white wage distribution Before and after interstate dereg. States with racial bias > median Before and after intrastate dereg. States with racial bias > median 100 90 80 70 60 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90100 Percentile of black wage distribution 0 10 20 30 40 50 60 70 80 90100 Percentile of black wage distribution Before and after interstate dereg. States with racial bias < median Before and after intrastate dereg. States with racial bias < median 100 90 80 70 60 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90100 Percentile of black wage distribution Percentile of white wage distribution Percentile of white wage distribution Percentile of white wage distribution Figure 3-5: The Location of Blacks in the White Wage Distribution Before and After Deregulation 100 90 80 70 60 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90100 Percentile of black wage distribution 183 Table 3.1: The Racial Bias Index, Survey Measures of Racial Prejudice, and Relative Wages Panel A: Correlation Coefficients Between the Different Measures of Taste for Discrimination Racial bias index Observations Panel B: Taste for Discrimination and Relative Wages of Blacks Racial bias index > median (1) Fraction whites who would not vote for black president (2) Fraction whites who support right to segregate neighborhoods (3) 0.36 {0.02} 0.35 {0.02} 0.31 {0.04} 43 43 43 Fraction whites who support law against interracial marriage Dependent Variable: Relative Wages of Blacks (1) (2) (3) (4) -.072*** -.065*** -.058*** (.014) -.042*** (.012) -.002 (.015) (.015) (.017) -.082*** 10,076 10,076 10,076 -.079*** (.013) Marginal racial prejudice > median Share of ≥ blacks 10% in 1970 (.013) Observations 10,076 NOTE –Panel A reports correlation coefficients between (1) The racial bias index, which is based on interracial marriages in 1970, and (2) three recent survey-based indicators of racial prejudice from Charles and Guryan (2008). Panel B reports estimated coefficients from four regressions, where the dependent variable is blacks’relative wage rates. Relative wages are conditional on five indicators of years of completed education (0-8, 9-11, 12, 13-15, and 16+) and a quartic in potential experience. Estimates are weighted by sampling weights provided by the Current Population Survey. In column (1), the regressor is an indicator which equals one if the racial bias index above the median and zero otherwise. In column (2) the regressor is an indicator which equals one if the marginal racial prejudice above the median and zero otherwise. The marginal racial prejudice index is the pth percentile of the distribution of an aggregate index of racial prejudice, where p is the percentile of workforce that is black. The marginal racial prejudice index is taken from Charles and Guryan (2008). Column (3) includes simultaneously the regressors from columns (1) and (2). In column (4) we also control for an indicator which equals one if the proportion of blacks in the workforce in 1970 is above 10%. The regressions include black workers prior to interstate and intrastate bank deregulation, so that the reported number of observations equals 10,076. All regressions include year fixed effects. We do not include state fixed effects because the regressors are fixed for each state and do not change over time. Standard errors are clustered at the state-year level and appear in parentheses; pvalues are in brackets. *, **, and *** indicate significance at the 10%, 5%, and 1% respectively. 184 Table 3.2: Bank Deregulation and Log New Incorporations Per Capita (1) Interstate dummy (2) (3) .084*** .082** (.031) .040 (.031) .038 (.041) (.041) Intrastate dummy (4) (5) (6) Interstate .032** .029** Interstate squared (.015) -.002 (.014) -.002 Intrastate (.001) .021*** (.001) .019** Intrastate squared (.008) -.0004* (.008) -.0004* (.0002) (.0002) 882 882 Observations 882 882 882 882 NOTE –The table shows the impact of various measures of bank deregulation on log new incorporations per capita. Robust standard errors are adjusted for state-level clustering and appear in parentheses. Intrastate dummy equals one in the years after a state permits branching via mergers and acquisitions and zero otherwise. Interstate dummy equals one in the years after a state permits interstate banking and zero otherwise. Interstate is equal to years since interstate deregulation and is equal to zero before interstate deregulation. Intrastate is equal to years since intrastate deregulation and is equal to zero before intrastate deregulation. New incorporations are from Dun and Bradstreet. Dates of intrastate and interstate bank deregulations are from Kroszner and Strahan (1999) and Amel (2008). The sample is for the years 1977-1994 and excludes Delaware and South Dakota. All regressions include state and year fixed effects. There are no other covariates. *, **, and *** indicate significance at the 10%, 5%, and 1%, respectively. 73,801 3% .006 (.015) .001 (.003) 73,801 21% .037** (.016) .004 (.003) .003*** (.001) (2) 48,367 9% .013 (.016) .003 (.003) (3) Below Median 25,434 29% .061* (.034) .012* (.007) (4) Above Median 73,801 13% .023*** (.006) .005*** (.001) (5) (6) 73,801 25% .044*** (.007) .005*** (.001) .004*** (.001) All States 48,367 15% .022*** (.007) .004*** (.002) (7) Below Median 25,434 27% .057*** (.012) .011*** (.002) (8) Above Median Racial Bias Index: Intrastate Deregulation NOTE - The dependent variable is the relative wage rates of blacks. Relative wages are conditional on five indicators of years of completed education (0-8, 9-11, 12, 13-15, and 16+) and a quartic in potential experience. Estimates are weighted by sampling weights provided by the Current Population Survey. Standard errors are adjusted for state-year clustering and appear in parentheses. All regressions include state and year fixed effects. The reported number of observations is for blacks only. Details about sample construction are in Appendix Table 2. In columns (1)-(4), “years since deregulation” stands for years since interstate deregulation; in columns (5)-(8), “years since deregulation” stands for years since intrastate deregulation. In columns (2) and (6), years since deregulation is interacted with an indicator which equals one if the racial bias index is above the median and zero otherwise. In columns (1), (2), (5), and (6) we include the entire sample. In columns (3) and (7) we include only states with racial bias index below the median. In columns (4) and (8) we include only states with racial bias index above the median. The racial bias index is based on rate of interracial marriages using the 1970 Census of Population. Appendix Table 3 lists the racial bias index for each state. The average initial racial wage gap is 17% for all states, 15% for states with a racial bias index below the median, and 21% for states with a racial bias index above the median. *, **, and *** indicate significance at the 10%, 5%, and 1% respectively. Observations as a share of sample's initial wage gap Impact after five years Impact after five years (Years since deregulation) x (Racial bias index > median) Years since deregulation (1) All States Racial Bias Index: Interstate Deregulation Table 3.3: Bank Deregulation and Relative Wage Rates 185 (2) (1) (3) Above Median (4) All States (5) Below Median 37,876 24,754 (.023) 13,122 (.038) .003 37,876 (.064) 21.8 .174 24,754 (.071) 8.2 .134 13,122 (.071) 26.8 .267*** 37,876 .016 (.020) 24,754 -.029 (.023) 13,122 .122*** (.037) 37,876 .023 (.064) 21.8 .046 24,754 -.026 (.071) 8.2 .358 13,122 .214*** (.068) 26.8 .123 37,876 .020 (.046) 3.0 .845 24,754 -.079 (.053) 2.3 .683 .518 24,754 (.054) 2.3 -.122** (8) Below Median 13,122 .198*** (.058) 113.5 .082 .086 13,122 (.062) 113.5 .235*** (9) Above Median Racial Bias Index: NOTE - The dependent variable is the relative wage rates of blacks. In panel A, relative wages are conditional on five indicators of years of completed education (0-8, 9-11, 12, 13-15, and 16+) and a quartic in potential experience. In panel B, relative wages are conditional on five indicators of years of completed education (0-8, 9-11, 12, 13-15, and 16+), a quartic in potential experience, and occupation fixed effects. Estimates are weighted by sampling weights provided by the Current Population Survey. Standard errors are adjusted for state-year clustering and appear in parentheses. All regressions include state and year fixed effects. The reported number of observations is for blacks only. Details about sample construction are in Appendix Table 2. The sample is limited to the years 1977-1994 due to availability of log new incorporations per capita data. In columns (1), (4), and (7) we include the entire sample. In columns (2), (5), and (8) we include only states with racial bias index below the median. In columns (3), (6), and (9) we include only states with racial bias index above the median. The racial bias index is based on rate of interracial marriages using the 1970 Census of Population. Appendix Table 3 lists the racial bias index for each state. Columns (1)-(3) report Ordinary Least Squares estimates of the impact of log new incorporations per capita on the relative wage rates of blacks. Columns (4)-(9) report Two Stage Least Squares estimates of the impact of log new incorporations per capita on the relative wage rates of blacks. In columns (4)-(6), log new incorporations per capita are instrumented by years since interstate deregulation and years since intrastate deregulation. In columns (7)-(9), log new incorporations per capita are instrumented by dummy variables for each year before and after interstate deregulation and dummy variables for each year before and after intrastate deregulation. The F-test of excluded instruments reports the F-statistic from the first-stage. The OIR test reports the p-value of a J-statistic that test over-identifying restrictions. *, **, and *** indicate significance at the 10%, 5%, and 1% respectively. Observations F-test of excluded instruments OIR test (p-value) Log new incorporation per capita .608 37,876 (.048) 3.0 .019 (7) (6) Panel B: Relative Wage Rates are Conditional on Education, Potential Experience, and Occupation OIR test (p-value) Observations F-test of excluded instruments (.022) All States 2SLS: Non-Parametric Above Median Racial Bias Index: 2SLS: Linear Panel A: Relative Wage Rates are Conditional on Education and Potential Experience Log new incorporation per capita .018 -.038 .137*** .042 -.080 Below Median All States Racial Bias Index: OLS Table 3.4: The Impact of Log New Incorporations Per Capita on Relative Wage Rates: OLS and 2SLS Estimates 186 29,121 8,093 .264 32.8 -.051 (.057) 2SLS 12,942 .259 26.8 .224*** (.068) 29,121 .571 16.2 .165** (.076) Racial Bias Above Median LLR CG (7) (8) NOTE - The dependent variable is the relative wage rates of blacks. Relative wages are conditional on five indicators of years of completed education (0-8, 9-11, 12, 13-15, and 16+), a quartic in potential experience, and occupation fixed effects. Estimates are weighted by sampling weights provided by the Current Population Survey. Standard errors are adjusted for state-year clustering and appear in parentheses. All regressions include state and year fixed effects. The reported number of observations is for blacks only. Details about sample construction are in Appendix Table 2. The sample is limited to the years 1977-1994 due to availability of log new incorporations per capita data. “LLR”stands for the racial bias index and is based on rate of interracial marriages using the 1970 Census of Population. Appendix Table 3 lists the racial bias index for each state. “CG” stands for the marginal racial prejudice which is obtained from Charles and Guryan (2008). In columns (1) and (5) we include only states with racial bias index below the median. In columns (3) and (7) we include only states with racial bias index above the median. In columns (2) and (6) we include only states with marginal racial prejudice below the median. In columns (4) and (8) we include only states with marginal racial prejudice above the median. Columns (1)-(4) report Ordinary Least Squares estimates of the impact of log new incorporations per capita on the relative wage rates of blacks. Columns (5)-(8) report Two Stage Least Squares estimates of the impact of log new incorporations per capita on the relative wage rates of blacks. Log new incorporations per capita are instrumented by dummy variables for each year before and after interstate deregulation and dummy variables for each year before and after intrastate deregulation. The F-test of excluded instruments reports the F-statistic from the first-stage. The OIR test reports the p-value of a Jstatistic that test over-identifying restrictions. *, **, and *** indicate significance at the 10%, 5%, and 1% respectively. 12,942 .375 8,093 -.024 (.071) 24,272 24,272 .039 (.028) Observations .124*** (.037) OIR test (p-value) -.029 (.030) Racial Bias Below Median LLR CG (5) (6) 8.1 -.029 (.023) Racial Bias Above Median LLR CG (3) (4) F-test of excluded instruments Log new incorporation per capita Racial Bias Below Median LLR CG (1) (2) OLS Table 3.5: The Impact of Log New Incorporations on the Relative Wages of Blacks: OLS and 2SLS Estimates 187 8,093 .000 (.002) .001 (.001) 24,754 29,121 .004** (.001) .004** (.002) 13,122 Above Median (2) 8,093 -.001 (.002) -.001 (.003) 24,754 Below Median (3) 29,121 .007** (.003) .012*** (.004) 13,122 Above Median (4) OLS 8,064 .000 (.003) .002 (.002) 24,687 Below Median (5) 29,034 .009*** (.002) .007*** (.002) 13,064 Above Median (6) Racial Prejudice: 2SLS 8,064 .000 (.003) -.000 (.004) 24,687 Below Median (7) 29,034 .015*** (.004) .008* (.004) 13,064 Above Median (8) Racial Prejudice: NOTE –The dependent variable in columns (1)-(4) is proportion of “excess”whites in an industry, where the proportion of “excess”whites is the proportion of whites that is unexplained by years of completed education (0-8, 9-11, 12, 13-15, and 16+), a quartic in potential experience, and occupation fixed effects. The dependent variable in columns (5)-(8) is proportion of white managers in an industry. Estimates are weighted by sampling weights provided by the Current Population Survey. Standard errors are adjusted for state-year clustering and appear in parentheses. All regressions include state and year fixed effects. The reported number of observations is for blacks only. Details about sample construction are in Appendix Table 2. The sample is limited to the years 1977-1994 due to availability of log new incorporations per capita data. In columns (1), (3), (5), and (7) we include only states with racial prejudice below the median. In columns (2), (4), (6), and (8) we include only states with racial prejudice above the median. In panel A, racial prejudice is the racial bias index which is based on rate of interracial marriages using the 1970 Census of Population. Appendix Table 3 lists the racial bias index for each state. In panel B, racial prejudice is the marginal racial prejudice which is obtained from Charles and Guryan (2008). Columns (1), (2), (5), and (6) report Ordinary Least Squares estimates, while columns (3), (4), (7), and (8) report Two Stage Least Squares. Log new incorporations per capita are instrumented by dummy variables for each year before and after interstate deregulation and dummy variables for each year before and after intrastate deregulation. *, **, and *** indicate significance at the 10%, 5%, and 1% respectively. Observations Panel B: Marginal Racial Prejudice Log new incorporation per capita Observations Log new incorporation per capita Panel A: Racial Bias Index Below Median (1) 2SLS Racial Prejudice: OLS Racial Prejudice: Dependent variable: working for a white manager Dependent variable: working in a white industry Table 3.6: The Impact of Log New Incorporations Per Capita on Employment of Blacks in “White”Industries: OLS and 2SLS Estimates 188 (2) 37,876 24,754 -.024 (.022) (1) .018 (.020) 13,122 .123*** (.035) (3) Above Median 37,876 -.012 (.061) 21.8 .316 (4) All States 24,754 .010 (.070) 8.2 .813 (5) Below Median 13,122 .190*** (.067) 26.8 .051 (6) Above Median Racial Bias Index: 2SLS: Linear 37,876 .005 (.042) 3.0 .814 (7) All States 24,754 -.043 (.048) 2.3 .638 (8) Below Median 13,122 .172*** (.056) 113.5 .122 (9) Above Median Racial Bias Index: 2SLS: Non-Parametric NOTE - The dependent variable is the relative wage rates of blacks. Relative wages are conditional on five indicators of years of completed education (08, 9-11, 12, 13-15, and 16+), a quartic in potential experience, occupation fixed effects, and industry fixed effects. Estimates are weighted by sampling weights provided by the Current Population Survey. Standard errors are adjusted for state-year clustering and appear in parentheses. All regressions include state and year fixed effects. The reported number of observations is for blacks only. Details about sample construction are in Appendix Table 2. The sample is limited to the years 1977-1994 due to availability of log new incorporations per capita data. In columns (1), (4), and (7) we include the entire sample. In columns (2), (5), and (8) we include only states with racial bias index below the median. In columns (3), (6), and (9) we include only states with racial bias index above the median. The racial bias index is based on rate of interracial marriages using the 1970 Census of Population. Appendix Table 3 lists the racial bias index for each state. Columns (1)-(3) report Ordinary Least Squares estimates of the impact of log new incorporations per capita on the relative wage rates of blacks. Columns (4)-(9) report Two Stage Least Squares estimates of the impact of log new incorporations per capita on the relative wage rates of blacks. In columns (4)-(6), log new incorporations per capita are instrumented by years since interstate deregulation and years since intrastate deregulation. In columns (7)-(9), log new incorporations per capita are instrumented by dummy variables for each year before and after interstate deregulation and dummy variables for each year before and after intrastate deregulation. The F-test of excluded instruments reports the F-statistic from the first-stage. The OIR test reports the p-value of a J-statistic that test over-identifying restrictions. *, **, and *** indicate significance at the 10%, 5%, and 1% respectively. Observations F-test of excluded instruments OIR test (p-value) Log new incorporation per capita Below Median All States Racial Bias Index: OLS Table 3.7: The Impact of Log New Incorporations Per Capita on Relative Wage Rates: OLS and 2SLS Estimates 189 21.5 .19 F-test of excluded instruments OIR test (p-value) 20,556 21.5 .19 Yes Yes 16,951 21.5 .19 Yes Yes (223) 424* (3) Hours>0 OLS 20,556 21.7 .39 Yes Yes 59 (105) (4) All 2SLS 16,951 21.5 .19 Yes Yes (0.285) .658** (5) All OLS 16,951 21.3 .49 Yes Yes .271*** (.106) (6) All 2SLS Log(Annual Hours) NOTE - The dependent variable is either hours worked or the log of hours worked. Thus, some specifications include all working-age black males, while others include only working black males. Estimates are weighted by sampling weights provided by the Current Population Survey. Standard errors are adjusted for state-year clustering and appear in parentheses. In regressions (1) –(3) and (4), the standard errors are bootstrapped. All regressions include state and year fixed effects. The analysis excludes states with below the median values of the racial bias index. Data on new incorporations per capita are available for the period 1977-1994. “Years since deregulation” includes both years since interstate deregulation and years since intrastate deregulation. The F-test of excluded instruments reports the F-statistic from the first-stage. The OIR test reports the p-value of a J-statistic that test over-identifying restrictions. *, **, and *** indicate significance at the 10%, 5%, and 1%, respectively. 20,556 Yes Years since deregulation squared Observations Yes 376* (223) 377* (223) (2) (1) Years since deregulation Instruments: Log new incorporations per capita Projected relative log hourly wage All Tobit All OLS Annual Hours Table 3.8: Relative Log Hourly Wages and Annual Working Hours in High Racial Bias States 190 116,593 H0: (Years since deregulation) x (1 if person reports wages) = 0 Observations (2) 116,593 (.0005) .0001 -.001 (.000) .001*** (.000) .075*** (.003) 77,301 (.001) -.003* -.005*** (.001) .002*** (.000) .064*** (.003) (3) 77,301 (.0006) .0003 -.001 (.001) .001*** (.000) .070*** (.004) (4) Racial Bias Index Below Median 39,292 (.002) .001 -.001 (.003) .002*** (.000) .069*** (.005) (5) 39,292 (.0008) -.0002 -.000 (.001) .000 (.000) .082*** (.005) (6) Racial Bias Index Above Median NOTE - The dependent variable is the predicted relative wages of blacks. Relative wages are conditional on five indicators of years of completed education (0-8, 9-11, 12, 13-15, and 16+) and a quartic in potential experience. Estimates are weighted by sampling weights provided by the Current Population Survey. Standard errors are adjusted for state-year clustering and appear in parentheses. All regressions control for state and year fixed effects. The reported number of observations is for blacks only. Details about sample construction are in Appendix Table 2. The sample is limited to the years 1977-1994 due to availability of log new incorporations per capita data. In columns (1) and (2) we include the entire sample. In columns (3) and (4) we include only states with racial bias index below the median. In columns (5) and (6) we include only states with racial bias index above the median. The racial bias index is based on rate of interracial marriages using the 1970 Census of Population. Appendix Table 3 lists the racial bias index for each state. *, **, and *** indicate significance at the 10%, 5%, and 1% respectively. -.002 (.001) Impact of deregulation on observable skills of black workers Intrastate x (1 if person reports wages) Intrastate Interstate x (1 if person reports wages) -.004*** (.001) .002*** (.000) Interstate (1) .066*** (.003) 1 if person reports wages All States Table 3.9: Bank Deregulation and Selection on Observable Characteristics 191 192 Table 3.10: Dates of Intrastate and Interstate Deregulations, by States State State code Type of deregulation: IntraInterstate state 1981 1987 1960 1982 Alabama Alaska AL AK Arizona Arkansas AZ AR 1960 1994 California Colorado CA CO Connecticut District of Columbia State State code Type of deregulation: IntraInterstate state 1990 1993 1985 1990 Montana Nebraska MT NE 1986 1989 Nevada New Hampshire NV NH 1960 1987 1985 1987 1960 1991 1987 1988 New Jersey New Mexico NJ NM 1977 1991 1986 1989 CT DC 1980 1960 1983 1985 New York North Carolina NY NC 1976 1960 1982 1985 Florida Georgia FL GA 1988 1983 1985 1985 North Dakota Ohio ND OH 1987 1979 1991 1985 Hawaii Idaho HI ID 1986 1960 1997 1985 Oklahoma Oregon OK OR 1988 1985 1987 1986 Illinois Indiana IL IN 1988 1989 1986 1986 Pennsylvania Rhode Island PA RI 1982 1960 1986 1984 Iowa Kansas IA KS 1999 1987 1991 1992 South Carolina Tennessee SC TN 1960 1985 1986 1985 Kentucky Louisiana KY LA 1990 1988 1984 1987 Texas Utah TX UT 1988 1981 1987 1984 Maine Maryland ME MD 1975 1960 1978 1985 Vermont Virginia VT VA 1970 1978 1988 1985 Massachusetts Michigan MA MI 1984 1987 1983 1986 Washington West Virginia WA WV 1985 1987 1987 1988 Minnesota Mississippi MN MS 1993 1986 1986 1988 Wisconsin Wyoming WI WY 1990 1988 1987 1987 Missouri MO 1990 1986 NOTE - Dates of intrastate and interstate deregulations are taken from Kroszner and Strahan (1999). 193 Table 3.11: Summary Statistics: Number of observations Restriction / Selection Rule Observations All observations in sample years 1977 to 2007 5,085,135 Civilian adults, not in group quarters, with positive sampling weight and non-missing demographics such as: age, gender, state and region of residence, 3,805,475 marital status, and education Excluding: Observations in Delaware and South Dakota Women 3,712,856 1,749,618 Younger than 18 or older than 65 More than 50 years of potential experience 1,392,503 1,337,897 Hispanics or other race groups but Whites or Blacks 1,149,855 Main sample: Whites 1,033,262 Blacks 116,593 Wage sample: All Whites 756,996 683,195 Blacks 73,801 NOTE - March Current Population Survey data were obtained from . We start in Survey year 1977 because that is when the CPS reports information on each person's exact state of residence. The 2007 Survey is the latest Survey available. We exclude Delaware and South Dakota due to large concentration of credit card banks in these two states. The ‘wage sample’differs from the ‘main sample’in that we drop self-employed and agricultural workers, workers in private household sector, those with wages below the 1st and above the 97th percentile of yearspecific wage distribution of full-time, full-year workers (i.e., those who work at least 50 weeks per year and at least 35 hours per week). Finally, we include in the ‘wage sample’only wage and salary workers. 194 Table 3.12: Racial Bias Index by States, 1970 States with racial bias index < median State States with racial bias index > median Racial Bias Index State Racial Bias Index Alaska Hawaii 0.00 0.07 Arkansas Virginia 0.30 0.30 Washington New York 0.10 0.11 South Dakota Colorado 0.30 0.30 Nevada California 0.12 0.15 North Carolina Texas 0.32 0.32 District of Columbia Delaware 0.18 0.24 Nebraska Minnesota 0.32 0.32 South Carolina New Jersey 0.24 0.25 Mississippi Oregon 0.33 0.33 Pennsylvania Michigan 0.25 0.26 Louisiana Georgia 0.33 0.34 Kentucky Illinois 0.26 0.26 Oklahoma Indiana 0.35 0.35 Maryland Connecticut 0.27 0.27 Alabama Wisconsin 0.35 0.36 Rhode island New Mexico 0.27 0.27 Vermont Utah 0.36 0.37 Kansas Massachusetts 0.28 0.28 Idaho Tennessee 0.37 0.39 Ohio Missouri 0.28 0.28 Iowa Montana 0.39 0.40 Arizona Florida 0.29 0.29 North Dakota West Virginia 0.43 0.45 Maine Wyoming 0.45 0.46 New Hampshire 0.46 NOTE - The racial bias index is based on inter-racial marriage data obtained from the 1970 Census of Population. The sample includes married whites and blacks between that ages of 18 to 65, and excludes couples in which at least one person is living in group quarters or has missing data on race, gender, state of residence, marital status, or educational attainment. The racial bias index is based on the difference between the estimated rate of inter-racial marriage in 1970, where the estimation is based on each state’s racial composition along with each individual’s education and age characteristics, and the actual rate of inter-racial marriage. Larger values of the racial bias index signify that the actual rate of inter-racial marriage is correspondingly smaller than the estimated rate. Bibliography Abowd, John M., “The E¤ect of Wage Bargains on the Stock Market Value of the Firm,”The American Economic Review, 1989, 79 (4), 774–800. and Thomas Lemieux, “The E¤ects of Product Market Competition on Collective Bargaining Agreements: The Case of Foreign Competition in Canada,” The Quarterly Journal of Economics, 1993, 108 (4), 983–1014. Almond, Douglas, Kenneth Y. Chay, and Michael Greenstone, “Civil Rights, the War on Poverty, and Black-White Convergence in Infant Mortality in the Rural South and Mississippi,”American Economic Review, 2008. Altonji, Joseph G. and Charles R. Pierret, “Employer Learning and Statistical Discrimination,”The Quarterly Journal of Economics, 2001, 116 (1), 313–350. and Rebecca M. Blank, “Race and Gender in the Labor Market,”in Ashenfelter, Orley and Card, David, ed., Handbook of Labor Economics, Vol. 3, Amsterdam: North-Holland, 1999, pp. 3143–3259. 195 196 Amel, Dean, “State Laws A¤ecting the Geographic Expansion of Commercial Banks,” Unpublished manuscript, Board of Governors of the Federal Reserve System, 1993. Amel, Dean F. and Nellie J. Liang, “The Relationship between Entry into Banking Markets and Changes in Legal Restrictions on Entry,” Antitrust Bulletin, 1992, 37, 631. Angrist, Joshua D. and Jörn-Ste¤en Pischke, Mostly Harmless Econometrics: An Empiricist’s Companion, Princeton University Press, 2009. Arrow, Kenneth J., “Some Mathematical Models of Race Discrimination in the Labor Market,”in Pascal, Anthony H., ed., Racial Discrimination in Economic Life, Lexington, MA: Health Press, 1972, pp. 187–204. Ashenfelter, Orley and George E. Johnson, “Unionism, Relative Wages, and Labor Quality in U.S. Manufacturing Industries,” International Economic Review, 1972, 13 (3), 488–508. and Timothy Hannan, “Sex Discrimination and Product Market Competition: The Case of the Banking Industry,”The Quarterly Journal of Economics, 1986, 101 (1), 149–174. Autor, David H., John J. Donohue, and Stewart J. Schwab, “The Costs of Wrongful-Discharge Laws,” The Review of Economics and Statistics, 2006, 88 (2), 211–231. 197 , Lawrence F. Katz, and Melissa S. Kearney, “Trends in U.S. Wage Inequality: Revising the Revisionists,” The Review of Economics and Statistics, 2008, 90 (2), 300–323. Banerjee, Abhijit V. and Andrew F. Newman, “Occupational Choice and the Process of Development,”The Journal of Political Economy, 1993, 101 (2), 274– 298. Beck, Thorsten, Asli Demirg-Kunt, and Ross Levine, “Finance, Inequality, and the Poor: Cross-Country Evidence,”Journal of Economic Growth, 2007, 12, 27–49. , Ross Levine, and Alexey Levkov, “Big Bad Banks? The Winners and Losers from Bank Deregulation in the United States,” The Journal of Finance, forthcoming. Becker, Gary S., The Economics of Discrimination, University of Chicago Press, 1957. Bekaert, Geert, Campbell R. Harvey, and Christian Lundblad, “Does Financial Liberalization Spur Growth?,” Journal of Financial Economics, 2005, 77 (1), 3–55. Belman, Dale L. and Kristen A. Monaco, “The E¤ects of Deregulation, DeUnionization, Technology, and Human Capital on the Work and Work Lives of Truck Drivers,”Industrial and Labor Relations Review, 2001, 54 (2A), 502–524. 198 Belzer, Michael H., “Collective Bargaining After Deregulation: Do the Teamsters Still Count?,”Industrial and Labor Relations Review, 1995, pp. 636–655. Benabou, Roland, “Equity and E¢ ciency in Human Capital Investment: The Local Connection,”Review of Economic Studies, 1996, 63, 237–264. Berger, Alan N., Anthony Saunders, Joseph M. Scalise, and Gregory F. Udell, “The E¤ects of Bank Mergers and Acquisitions on Small Business Lending,”Journal of Financial Economics, 1998, 50 (2), 187–229. Bertrand, Marianne and Sendhil Mullainathan, “Are Emily and Greg More Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination,”The American Economic Review, 2004, 94 (4), 991–1013. , Esther Du‡o, and Sendhil Mullainathan, “How Much Should We Trust Di¤erences-in-Di¤erences Estimates?,” Quarterly Journal of Economics, 2004, 119 (1), 249–275. Black, Sandra E. and Alexandra Spitz-Oener, “Explaining Women’s Success: Technological Change and the Skill Content of Women’s Work,” Technical Report 2007. and Elizabeth Brainerd, “Importing Equality? The Impact of Globalization on Gender Discrimination,” Industrial and Labor Relations Review, 2004, pp. 540–559. 199 and Philip E. Strahan, “The Division of Spoils: Rent-Sharing and Discrimination in a Regulated Industry,”The American Economic Review, 2001, 91 (4), 814–831. and , “Entrepreneurship and Bank Credit Availability,” The Journal of Finance, 2002, LVII (6), 2807–2833. Blau, Francine D. and Lawrence M. Kahn, “Swimming Upstream: Trends in the Gender Wage Di¤erential in the 1980s,” Journal of Labor Economics, 1997, 15 (1), 1–42. and , “Gender Di¤erences in Pay,”The Journal of Economic Perspectives, 2000, 14 (4), 75–99. Blau, Peter M., Inequality and Heterogeneity, New York: Free Press, 1977. Bodenhorn, Howard, State Banking in Early America: A New Economic History, New York: Oxford University Press, 2003. Burgess, Robin and Rohini Pande, “Do Rural Banks Matter? Evidence from the Indian Social Banking Experiment,” American Economic Review, 2005, 95 (3), 780–795. Butler, Richard and James J. Heckman, “The Government’s Impact on the Labor Market Status of Black Americans: A Critical Review,” in Hausman, Leonard J. and Ashenfelter, Orley and Rustin, Bayard and Schubert, Richard F. and Slaiman, Donald, ed., Equal Rights and Industrial Relations, Madison, WI: Industrial Relations Research Association, 1977, pp. 235–281. 200 Calem, Paul S., “The Impact of Geographic Deregulation on Small Banks,” Business Review, 1994, pp. 17–31. Cappelli, Peter, “Competitive Pressures and Labor Relations in the Airline Industry,”Industrial Relations, 1985, 24 (3), 316–338. Card, David, “The Impact of Deregulation on the Employment and Wages of Airline Mechanics,”Industrial and Labor Relations Review, 1986, 39 (4), 527–538. , “Deregulation and Labor Earnings in the Airline Industry,” NBER Working Paper No. 5687 1996. , “The E¤ect of Unions on the Structure of Wages: A Longitudinal Analysis,” Econometrica, 1996, 64 (4), 957–979. , “The E¤ect of Unions on Wage Inequality in the US Labor Market,”Industrial and Labor Relations Review, 2001, 54 (2), 296–315. and Alan B. Krueger, “School Quality and Black-White Relative Earnings: A Direct Assessment,”The Quarterly Journal of Economics, 1992, 107 (1), 151– 200. Cetorelli, Nicola and Philip E. Strahan, “Finance as a Barrier to Entry: Bank Competition and Industry Structure in Local U.S. Markets,” The Journal of Finance, 2006, LXI (1), 437–461. 201 Charles, Kerwin K. and Jonathan Guryan, “Prejudice and Wages: An Empirical Assessment of Becker’s The Economics of Discrimination,” Journal of Political Economy, 2008, 116 (5), 773–809. Christo…des, Louis N. and Andrew J. Oswald, “Real Wage Determination and Rent-Sharing in Collective Bargaining Agreements,” The Quarterly Journal of Economics, 1992, 107 (3), 985–1002. Claessens, Stijn and Enrico Perotti, “Finance and Inequality: Channels and Evidence,”Journal of Comparative Economics, 2007, 35 (4), 748–773. Clark, Andrew E., Paul Frijters, and Michael A. Shields, “Relative Income, Happiness, and Utility: An Explanation for the Easterlin Paradox and Other Puzzles,”Journal of Economic Literature, 2008, 46 (1), 95–144. Clark, Kim B., “Unionization and Firm Performance: The Impact on Pro…ts, Growth, and Productivity,”The American Economic Review, 1984, 74 (5), 893– 919. Coate, Stephen and Glenn C. Loury, “Will A¢ rmative-Action Policies Eliminate Negative Stereotypes?,” The American Economic Review, 1993, 83 (5), 1220– 1240. Cremieux, Pierre-Yves, “The E¤ect of Deregulation on Employee Earnings: Pilots, Flight Attendants, and Mechanics, 1959-1992,” Industrial and Labor Relations Review, 1996, pp. 223–242. 202 Demirgüç-Kunt, Asli and Ross Levine, “Finance and Inequality: Theory and Evidence,”Annual Review of Financial Economics, 2009, 1, 287–318. and Vojislav Maksimovic, “Law, Finance, and Firm Growth,” The Journal of Finance, 1998, LIII (6), 2107–2137. Demyanyk, Yuliya, “U.S. Banking Deregulation and Self-Employment: A Di¤erential Impact on Those in Need,”Journal of Economics and Business, 2008, 60 (1-2), 165–178. , Charlotte Ostergaard, and Bent E. Sorensen, “U.S. Banking Deregulation, Small Businesses, and Interstate Insurance of Personal Income,” The Journal of Finance, 2007, LXII (6), 2763–2801. Diamond, Douglas W., “Financial Intermediation and Delegated Monitoring,”The Review of Economic Studies, 1984, 51 (3), 393–414. DiNardo, John and David S. Lee, “Economic Impacts of New Unionization on Private Sector Employers: 1984-2001,” The Quarterly Journal of Economics, 2004, 119 (4), 1383–1441. Donohue, John and James J. Heckman, “Continuous Versus Episodic Change: The Impact of Civil Rights Policy on the Economic Status of Blacks,” Journal of Economic Literature, 1991, 29 (4), 1603–1643. Durlauf, Steven N., “A Ttheory of Persistent Income Inequality,”Journal of Economic Growth, 1996, 1 (1), 75–93. 203 , “Racial Pro…ling as a Public Policy Question: E¢ ciency, Equity, and Ambiguity,”American Economic Review, 2005, 95 (2), 132–136. and Marcel Fafchamps, “Social Capital,” in Aghion, Philippe and Durlauf, Steven, ed., Handbook of Economic Growth, Vol. 1, Amsterdam: North-Holland, 2005, pp. 1639–1699. Economides, Nicholas, Glenn R. Hubbard, and Darius Palia, “The Political Economy of Branching Restrictions and Deposit Insurance: A model of Monopolistic Competition among Small and Large Banks,” Journal of Law and Economics, 1996, XXXIX (2), 667–704. Farber, Henry S., “The Analysis of Union Behavior,” in Ashenfelter, Orley and Layard, Richard, ed., Handbook of Labor Economics, Vol. 2, Amsterdam: NorthHolland, 1986, pp. 1039–1089. , “The Recent Decline of Unionization in the United States,”Science, 1987, 238 (4829), 915. , “Trends in Worker Demand for Union Representation,” The American Economic Review, 1989, 79 (2), 166–171. and Alan B. Krueger, “Union Membership in the United States: The Decline Continues,”in Kaufman, Bruce and Kleiner, Morris, ed., Employee Representation: Alternatives and Future Directions, Industrial Relations Research Association, 1993, pp. 105–134. 204 Flannery, Mark J., “The Social Costs of Unit Banking Restrictions,” Journal of Monetary Economics, 1984, 13 (2), 237–249. Freeman, Richard B., “Contraction and Expansion: The Divergence of Private Sector and Public Sector Unionism in the United States,” The Journal of Economic Perspectives, 1988, 2 (2), 63–88. and James Medo¤, What Do Unions Do?, New York: Basic Books, 1984. Fryer, Roland G. and Steven D. Levitt, “Understanding the Black-White Test Score Gap in the First Two Years of School,”Review of Economics and Statistics, 2004, 86 (2), 447–464. Galor, Oded and Joseph Zeira, “Income Distribution and Macroeconomics,”The Review of Economic Studies, 1993, pp. 35–52. Granger, Clive W. J., “Investigating Causal Relations by Econometric Models and Cross-Spectral Methods,”Econometrica, 1969, 37 (3), 424–438. Greenwood, Jeremy and Boyan Jovanovic, “Financial Development, Growth, and the Distribution of Income,” The Journal of Political Economy, 1990, 98 (5), 1076–1107. Haber, Stephen and Enrico Perotti, “The Political Economy of Finance,”Journal of Economic Literature, 2008. 205 Hamilton, Barton H., “Does Entrepreneurship Pay? An Empirical Analysis of the Returns to Self-Employment,” Journal of Political Economy, 2000, 108 (3), 604–631. Hammond, Bray, Banks and Politics in America: From the Revolution to the Civil War, Princeton University Press, 1957. Heckman, James J., Jora Stixrud, and Sergio Urzua, “The E¤ects of Cognitive and Noncognitive Abilities on Labor Market Outcomes and Social Behavior,” Journal of Labor Economics, 2006, 24 (3), 411–482. Heywood, John S. and James H. Peoples, “Deregulation and the Prevalence of Black Truck Drivers,”Journal of Law and Economics, 1994, 37, 133–155. Hirsch, Barry T., “Union Coverage and Pro…tability Among U.S. Firms,” The Review of Economics and Statistics, 1991, 73 (1), 69–77. , “What Do Unions Do for Economic Performance?,”in Bennett, James T. and Kaufman, Bruce E., ed., What Do Unions Do? A Twenty–Year Perspective, New Jersey: Transaction Publishers, 2007, pp. 193–327. , “Sluggish Institutions in a Dynamic World: Can Unions and Industrial Competition Coexist?,”Journal of Economic Perspectives, 2008, 22 (1), 153–176. and Mark C. Berger, “Union Membership Determination and Industry Characteristics,”Southern Economic Journal, 1984, 50 (3), 665–679. 206 Huang, Rocco R., “Evaluating the Real E¤ect of Bank Branching Deregulation: Comparing Contiguous Counties Across US State Borders,”Journal of Financial Economics, 2008, 87 (3), 678–705. Hubbard, Glenn R. and Darius Palia, “Executive Pay and Performance: Evidence from the US Banking Industry,” Journal of Financial Economics, 1995, 39 (1), 105–130. Jarrell, Stephen B. and T. D. Stanley, “A Meta-Analysis of the Union-Nonunion Wage Gap,”Industrial and Labor Relations Review, 1990, 44 (1), 54–67. Jayaratne, Jith and Philip E. Strahan, “The Finance-Growth Nexus: Evidence from Bank Branch Deregulation,” The Quarterly Journal of Economics, 1996, 111 (3), 639–670. and , “Entry Restrictions, Industry Evolution, and Dynamic E¢ ciency: Evidence from Commercial Banking,”Journal of Law and Economics, 1998, XLI, 239–273. Jencks, Christopher and Meredith Phillips, The Black-White Test Score Gap, Washington, DC: Brookings Institution Press, 1998. Juhn, Chinhui, Kevin M. Murphy, and Brooks Pierce, “Accounting for the Slowdown in Black-White Wage Convergence,”in Kosters, Marvin, ed., Workers and Their Wages: Changing Patterns in the United States, Washington, DC: American Enterprise Press, 1991, pp. 107–143. 207 Kahn, Lawrence M., “Unionism and Relative Wages: Direct and Indirect E¤ects,” Industrial and Labor Relations Review, 1979, 32 (4), 520–532. Kahneman, Daniel and Alan B. Krueger, “Developments in the Measurement of Subjective Well-Being,”Journal of Economic Perspectives, 2006, 20 (1), 3–24. Katz, Lawrence F. and David H. Autor, “Changes in the Wage Structure and Earnings Inequality,” in Ashenfelter, Orley and Card, David, ed., Handbook of Labor Economics, Vol. 3, Amsterdam: North-Holland, 1999, pp. 1463–1555. Kerr, William R. and Ramana Nanda, “Democratizing Entry: Banking Deregulations, Financing Constraints, and Entrepreneurship,” Journal of Financial Economics, 2009, 94 (1), 124–149. King, Robert G. and Ross Levine, “Finance and Growth: Schumpeter Might be Right,”The Quarterly Journal of Economics, 1993, 108 (3), 717–737. and , “Finance Entrepreneurship and Growth: Theory and Evidence,” Journal of Monetary Economics, 1993, 32 (3), 513–542. Kroszner, Randall S. and Philip E. Strahan, “What Drives Deregulation? Economics and Politics of the Relaxation of Bank Branching Restrictions,” The Quarterly Journal of Economics, 1999, 114 (4), 1437–1467. Krueger, Alan B. and Alexandre Mas, “Strikes, Scabs, and Tread Separations: Labor Strife and the Production of Defective Bridgestone/Firestone Tires,”Journal of Political Economy, 2004, 112 (2), 253–289. 208 Lancaster, Tony, “Econometric Methods for the Duration of Unemployment,” Econometrica, 1979, 47 (4), 939–956. Lee, David S. and Alexandre Mas, “Long-Run Impacts of Unions on Firms: New Evidence from Financial Markets, 1961-1999,”NBER Working Paper No. 14709 2009. Lee, Lung-Fei, “Unionism and Wage Rates: A Simultaneous Equations Model with Qualitative and Limited,”International Economic Review, 1978, 19 (2), 415–433. Levine, Ross, “Finance and Growth: Theory and Evidence,” in Aghion, Philippe and Durlauf, Steven, ed., Handbook of Economic Growth, Amsterdam: NorthHolland, 2005, pp. 865–934. , Alexey Levkov, and Yona Rubinstein, “Racial Discrimination and Competition,”NBER Working Paper No. 14273 2008. and David Renelt, “A Sensitivity Analysis of Cross-Country Growth Regressions,”The American Economic Review, 1992, 82 (4), 942–963. and Sarah Zervos, “Stock Markets, Banks, and Economic Growth,” The American Economic Review, 1998, 88 (3), 537–558. and Yona Rubinstein, “Credit Constraints, Teen Pregnancy, and High School Diplomas,”"Brown University, mimeo" 2009. Levinson, Harold M., “Unionism, Concentration, and Wage Changes: Towards a Uni…ed Theory,”Industrial and Labor Relations Review, 1967, 20 (2), 198–205. 209 Lewis, Gregg H., “Union Relative Wage E¤ect,”in Ashenfelter, Orley and Layard, Richard, ed., Handbook of Labor Economics, Vol. 2, Amsterdam: North-Holland, 1986, pp. 1139–1181. Massey, Douglas S. and Nancy A. Denton, “Trends in the Residential Segregation of Blacks, Hispanics, and Asians: 1970-1980,” American Sociological Review, 1987, pp. 802–825. McLaughlin, Susan, “The Impact of Interstate Banking and Branching Reform: Evidence from the States,” Current Issues in Economics and Finance, 1995, 1 (2), 1–5. Morgan, Donald P., Bertrand Rime, and Philip E. Strahan, “Bank Integration and State Business Cycles,”The Quarterly Journal of Economics, 2004, 119 (4), 1555–1584. Moulton, Brent R., “An Illustration of a Pitfall in Estimating the E¤ects of Aggregate Variables on Micro Units,” The Review of Economics and Statistics, 1990, 72 (2), 334–338. Mulligan, Casey B. and Yona Rubinstein, “Selection, Investment, and Women’s Relative Wages Over Time,” Quarterly Journal of Economics, 2008, 123 (3), 1061–1110. Neal, Derek A., “Why Has Black-White Skill Convergence Stopped,”in Hanushek, Erik and Welch, Finis, ed., Handbook of the Economics of Education, Amsterdam: North-Holland, 2006. 210 and William R. Johnson, “The Role of Premarket Factors in Black-White Wage Di¤erences,”The Journal of Political Economy, 1996, 104 (5), 869–895. Nickell, Stephen J., Jari Vainiomaki, and Sushil Wadhwani, “Wages and Product Market Power,”Economica, 1994, 61 (244), 457–473. Oster, Sharon M., “Industry Di¤erences in the Level of Discrimination Against Women,”The Quarterly Journal of Economics, 1975, 89 (2), 215–229. Park, Jin Heum, “Estimation of Sheepskin E¤ects and Returns to Schooling Using the Old and the New CPS Measures of Educational Attainment,” Technical Report, Princeton University 1994. Peoples, James, “Trucking Deregulation and Labour Earnings in the USA: A ReExamination,”Applied Economics, 1996, 28, 865–874. and Wayne K. Talley, “Black-White Earnings Di¤erentials: Privatization Versus Deregulation,”The American Economic Review, 2001, 91 (2), 164–168. Phelps, Edmund S., “The Statistical Theory of Racism and Sexism,”The American Economic Review, 1972, pp. 659–661. Polivka, Anne E., “Data Watch: The Redesigned Current Population Survey,”The Journal of Economic Perspectives, 1996, 10 (3), 169–180. Qian, Zhenchao, “Breaking the Racial Barriers: Variations in Interracial Marriage Between 1980 and 1990,”Demography, 1997, pp. 263–276. 211 Rajan, Raghuram G. and Luigi Zingales, “Financial Dependence and Growth,” The American Economic Review, 1998, 88 (3), 559–586. Rose, Nancy L., “Labor Rent Sharing and Regulation: Evidence from the Trucking Industry,”The Journal of Political Economy, 1987, 95 (6), 1146–1178. Ruback, Richard S. and Martin B. Zimmerman, “Unionization and Pro…tability: Evidence from the Capital Market,”The Journal of Political Economy, 1984, 92 (6), 1134–1157. Rubinstein, Yona and Dror Brenner, “Pride and Prejudice: Evidence from the ‘Promised Land’,”Technical Report 2010. Savage, Donald T., “Interstate Banking: A Status Report,” Federal Reserve Bulletin, 1993, 79, 1075. Segal, Martin, “The Relation Between Union Wage Impact and Market Structure,” The Quarterly Journal of Economics, 1964, 78 (1), 96–114. Shepherd, William G. and Sharon G. Levin, “Managerial Discrimination in Large Firms,”The Review of Economics and Statistics, 1973, 55 (4), 412–422. Slaughter, Matthew J., “Globalization and Declining Unionization in the United States,”Industrial Relations, 2007, 46 (2), 329–346. Smith, James P. and Finis R. Welch, “Black Economic Progress After Myrdal,” Journal of Economic Literature, 1989, 27 (2), 519–564. 212 Southworth, Shirley D., Branch Banking in the United States, New York, NY: McGraw-Hill, 1928. Stewart, Mark B., “Union Wage Di¤erentials, Product Market In‡uences and the Division of Rents,”The Economic Journal, 1990, 100 (403), 1122–1137. Valletta, Robert G. and Richard B. Freeman, “The NBER Public Sector Collective Bargaining Law Data Set, Appendix B,” in Freeman, Richard B. and Ichniowski, Casey, ed., When Public Employees Unionize, Chicago: NBER and University of Chicago Press, 1988, pp. 399–419. Wald, Abraham, “The Fitting of Straight Lines if Both Variables are Subject to Error,”The Annals of Mathematical Statistics, 1940, pp. 284–300. Weiss, Leonard W., “Concentration and Labor Earnings,”The American Economic Review, 1966, 56 (1/2), 96–117. Welch, Finis R., “Human Capital Theory: Education, Discrimination, and Life Cycles,”The American Economic Review, 1975, 65 (2), 63–73. White, Eugene N., “The Political Economy of Banking Regulation, 1864-1933,” The Journal of Economic History, 1982, 42 (1), 33–40.