Three Essays in Corporate Management and Firm Policy

THREE ESSAYS IN CORPORATE MANAGEMENT AND FIRM POLICY

By YANKUO QIAO

A dissertation submitted to the Graduate School–Newark Rutgers, The State University of New Jersey In partial fulﬁllment of the requirements For the degree of Doctor of Philosophy Graduate Program in Management Written under the direction of Ivan Brick And approved by

Dissertation Director: Professor Ivan E. Brick

The dissertation entitled “Three Essays in Corporate Management and Firm Pol- icy” examines the relationship between corporate decision making and firm performance, with a particular focus on CEO transition and power distribution in the top management team. It includes three essays. These essays investigate, both theoretically and empirically, the nexus of top executives, firm value and various firm policies on corporate social responsibility, investment and voluntary disclosure. It strives to make a significant contribution to the literature on corporate finance, corporate governance, management structure and organizational behavior. The first essay, “CEO’s Investment Cycle in Social Capital and Its Impact on Firm Value”, demonstrates that there is a CEO-centric influence upon the level of firm investment in social capital, as measured by Corporate Social Responsibility (CSR). The CEO specific effect is widely recognized in the extant literature as a persistent and significant impact on various firm policies such as financing and investment. This paper hypothesizes that if managerial style is an important influence upon the firm’s investment in CSR, we would expect the greatest change in the level of CSR to occur when the new CEO takes over the company. The paper presents empirical evidence in support of this hypothesis. Moreover, it is found that the sudden change of CSR during the transitory period of CEO turnover is likely to be permanent in that the new CEO will on average not reverse the policy later in her tenure and the social responsibility policy becomes stable over time. One exception to this general pattern of CSR policy change is when the new CEO is an outsider. The latter finding is consistent with attunement theory which

ii states that an outsider successor may be reluctant to aggressively change CSR due to the initial lack of viable relationship and trustful dialogue with different stakeholder groups of the firm. Thus, the capability and incentive of the outsider successor for changing CSR activities will be undermined or be self-refrained, and the outsider CEO tends to continue her predecessor’s policy in the early stage of tenure. Furthermore, based on the CEO-centric effect and the investment cycle in social capital, the paper investigates the value relevance of CSR during CEO transition period, when the investment in CSR changes dramatically. The empirical evidence suggests that CSR is value enhancing when “trust” becomes an important intangible asset during crisis periods. The second essay, “Impact of Internal Governance on a CEO’s Investment Cycle”, examines the impact of internal governance on the cyclical pattern of corporate investment policy. The extant literature defines internal governance as the mechanism by which senior executives help discipline the CEO to maximize shareholder value. Weisbach (1995) finds that a year or two before the CEO retires, the firm experiences a decrease in total investment, suggesting the CEO cuts her investment to conserve current resources. Pan, Wang and Weisbach (2016) find evidence of a CEO’s investment cycle, in which investment increases over a CEO’s tenure, whereas disinvestment decreases. These papers suggest that older CEOs incur agency costs as they try to extract rents as their investment horizon declines. The empirical evidence of this paper confirms their results, and additionally shows that good internal governance helps reduce older CEOs underinvesting before their exit. No significant relationship in either economic or statistical sense is found between internal governance and investment for younger CEOs. This suggests that older CEOs who face good internal governance underinvest less. It is also found that new incoming CEOs divest profitably these projects acquired by their predecessors under good internal governance. These results are robust to: normal CEO retirements (exclude performance-related turnovers), sudden CEO deaths, and controlling for measures of board size, proportion of outsiders on the board, CEO pay-performance sensitivity, CEO pay slice, CEO power, firm complexity

iii and if the CEO was an outsider or not. The third essay, “To Delegate or Not to Delegate? On the Quality of Voluntary Corporate Financial Disclosure and Its Market Impacts”, investigates the impact of delegation structure of the top management team upon the quality of corporate voluntary disclosure on financial outcomes. The paper develops two competing hypotheses pertaining to the functional relationship between the degree of delegation and the management forecast accuracy. On the one hand, as indicated by the literature on internal governance, the efficacy of the top management team is optimized when neither the CEO nor the subordinate managers are dominant. On the other hand, an extensive literature has documented the importance and centrality of the CEO as well as the relevance of the subordinate managers to voluntary disclosure activities. The empirical findings are in support of an inverted hump-shaped relationship between the degree of delegation and the quality of voluntary information provision, suggesting that an internal optimality of responsibility sharing between the CEO and her immediate subordinates does not exist for information production, transmission and dissemination. Partial delegation and mixed executive duties lead to deteriorating quality of voluntary disclosure. In particular, the paper analyzes several aspects of managerial earnings forecasts (MFs), the most influential type of voluntary financial disclosure. The documented curvilinear forms are generally persistent across multiple quality metrics of MFs. Consistent with the literature on executive horizon and risk propensity, the curvilinear relation is more significant when the top management team is led by an older CEO. The paper utilizes an identification strategy of structural equations, which controls for selection bias and reverse causality. To theoretically underpin the empirical findings, a model of internal information production is developed in the framework of Bayesian Nash Equilibrium. The paper further documents that when the delegation structure is clear, namely either the CEO or subordinates are in charge, the liquidity of the company’s stock improves. The empirical evidence suggests that the variation of liquidity driven by delegation structures is not actively incorporated in stock prices.

iv ACKNOWLEDGEMENT

This dissertation lays a new cornerstone for my research in finance and corporation. It could also in part represent my work done at Rutgers Business School–Newark and New Brunswick in the past five years. It would not have been possible for me to complete this dissertation in absence of the support, mentoring and guidance from many people at Rutgers and beyond, including faculties, fellow students, friends, and family, to all of whom I wish to extend my sincerest gratitude and appreciation. First and foremost, I would like to express the deepest gratefulness to my advisor, Dr. Ivan Brick, Dean’s Professor in Business and Chair of the Department of Finance & Economics at Rutgers Business School, for being able to work under his wing when I needed help and guidance the most. He has offered unreserved help and invaluable advice to my dissertation and career development. I feel very fortunate to have the opportunity to do research jointly with Dr. Dar- ius Palia. It was an enjoyable and educational experience working under his super- vision. The knowledge and skills I attained from collaborating with him helped me immensely in writing the dissertation. I hope that I will have chances to continue working together with him in the future. I would like to give special thanks to Dr. Tavy Ronen and Dr. N.K. Chi- dambaran for their insightful comments and constructive suggestions about my dissertation. They helped me develop necessary critical thinking skills to carry out research for the dissertation and beyond. Dr. Tavy Ronen, in particular, helped me expand my horizon and developed expertise in corporate bond and market microstructure. I also owe a debt of gratitude to many other faculty members at Rutgers Busi- ness School, including Dr. Cheng-Few Lee, Dr. Yangru Wu, Dr. Zhaodong (Ken) Zhong, Dr. Oded Palmon, Dr. Michael Barnett, Dr. Nancy DiTomaso, Dr. Simi Kedia, Dr. Feng Gao, Dr. Zhengzi (Sophia) Li, Dr. Ben Sopranzetti, Dr. Robert Patrick, Dr. Jin-Mo Kim, Dr. Daniel Weaver, Dr. Daniela Osterrieder, Dr. Yuzhao

v Zhang, Dr. Harvey Poniachek and Dr. Parul Jane for providing general guidance on my research, teaching and doctoral study. I also wish to thank some fellow graduate students and friends, including Ge Wu, Alex Abakah, Yaqing Xiao, Peixuan Yuan, Yushan Tang, Hengguo Da, Rouzhi Wang, Yuanyuan Xiao, An Qin, Yifei Chen, Shuo Liang, Jiayin Li and Jun Li. Additional thanks to Taja-Nia Hender- son, Goncalo Filipe, Jane Foss, Cheryl Daniels, Tiﬀany Nelson-Mccullough and Michele Scott for the great support in administration. Finally, immeasurable appreciation belongs to my family, especially my par- ents, Qian Qiao and Lengzhuo Xu, and my wife, Xue Wang. Without their tireless support and encouragement, this work would have been impossible.

vi DEDICATION To my wife, Xue, and our children, Junyang and Junhuan

vii TABLE OF CONTENTS

ABSTRACT OF THE DISSERTATION ii

ACKNOWLEDGEMENT v

DEDICATION vii

TABLE OF CONTENTS viii

LIST OF TABLES xi

LIST OF ILLUSTRATIONS xv

Chapter 1: CEO’s Investment Cycle in Social Capital and Its Impact on Firm Value 1 1.1 Introduction ...... 2 1.2 Theoretical Background and Literature Review ...... 11 1.3 Hypotheses ...... 17 1.4 Methodology ...... 25 1.5 Measures, Data and Summary Statistics ...... 33 1.5.1 Data and Sample Construction ...... 33 1.5.2 Summary Statistics ...... 37 1.6 Results and Analyses ...... 39 1.6.1 Tests of Hypothesis 1 - Is CSR CEO Centric? ...... 39 1.6.2 Tests of Hypothesis 2 – Disentangling CEO fixed Effects From Firm Fixed Effects ...... 42 1.6.3 Tests of Hypothesis 3 – Is CSR Policy Change Permanent? . 44 1.6.4 Tests of Hypothesis 4 – Test of Attunement Theory . . . . . 45 1.6.5 Tests of Hypothesis 5 – Does CSR Affect Firm Value? . . . . 46 1.7 Robustness Check ...... 51 1.8 Concluding Remarks and Future Research ...... 59 References ...... 62

viii Chapter 2: Impact of Internal Governance on a CEO’s Investment Cycle 89 2.1 Introduction ...... 90 2.2 Literature Review ...... 95 2.3 Data, Variable Construction and Sample ...... 100 2.4 Empirical Results ...... 107 2.4.1 Proxy for Internal Governance ...... 107 2.4.2 Impact of Internal Governance on a CEO’s Investment Cycle ...... 109 2.4.3 Impact of Internal Governance on the Proﬁtability of Asset Divestitures ...... 111 2.4.4 Robustness Tests ...... 113 2.5 Conclusions ...... 121 References ...... 124 Appendix: Construction of the Internal Governance Measure ...... 129

Chapter 3: To Delegate or Not to Delegate? On the Quality of Vol- untary Corporate Financial Disclosure and Its Market Impacts 151 3.1 Introduction ...... 152 3.2 Theoretical Background and Literature Review ...... 163 3.2.1 Why Firms Disclose? ...... 163 3.2.2 Internal Governance ...... 165 3.2.3 Importance of CEO ...... 167 3.2.4 Inﬂuential Top Managers and Disclosure Activities . . . . . 169 3.2.5 Delegation and Internal Information Production ...... 171 3.2.6 Aging CEO ...... 173 3.2.6.1 Heterogeneity in Executive Horizons ...... 173 3.2.6.2 Heterogeneity in Risk Propensities ...... 174 3.2.7 Voluntary Disclosure, Liquidity and Stock Returns . . . . . 175 3.3 Hypothesis Development ...... 177

ix 3.4 Methodology and Sample ...... 186 3.4.1 Measures and Methodology ...... 186 3.4.1.1 Voluntary Disclosure Quality ...... 186 3.4.1.2 Stock Liquidity ...... 195 3.4.2 Data, Sample Construction and Summary Statistics . . . . . 199 3.5 Results and Analyses ...... 204 3.5.1 Quality of Voluntary Disclosure ...... 204 3.5.1.1 Graphical Analysis ...... 205 3.5.1.2 Simple Quadratic Regression Analysis ...... 208 3.5.1.3 Multivariate Regression Analysis ...... 209 3.5.1.4 System of Structural Equations ...... 210 3.5.2 Stock Liquidity ...... 212 3.5.3 Value relevance of the Liquidity Eﬀect ...... 214 3.5.3.1 Portfolio Analysis ...... 215 3.5.3.2 Meditation Analysis ...... 216 3.5.4 Robustness Check ...... 219 3.5.4.1 Alternative Instrument ...... 219 3.5.4.2 Alternative Delegation Measure ...... 221 3.5.4.3 Regressions in δ Quantiles ...... 222 3.5 Conclusion and Discussion ...... 224 References ...... 229 Appendix A: A Theory of Corporate Internal Information Production . . 243 A.1 The First-Best Case ...... 246 A.2 Bayesian Nash Equilibrium Cases ...... 247 A.3 Comparative Statics Analysis ...... 259 Appendix B: An Econometric Reasoning ...... 264 Appendix C: A Diagram of Hypotheses ...... 268

x LIST OF TABLES

Table 1.1 Variable Definitions ...... 72 Table 1.2 Summary Statistics of CEO Turnover ...... 74 Table 1.3 Descriptive Statistics ...... 75 Table 1.4 Regressions of CEO Fixed Effects on CSR ...... 76 Table 1.5 Regressions of CEO Turnover on Absolute CSR Change . . . . 77 Table 1.6 Regressions of CEO Turnover on CSR Change ...... 78 Table 1.7 Regressions of CEO Turnover and Tenure on CSR Change . . . 79 Table 1.8 Regressions of CEO Turnover and Outsider on CSR Change . . 80 Table 1.9 Trust, CSR and Firm Performance ...... 81 Table 1.10 Industry Norm, CSR and Firm Performance ...... 82 Table 1.11 Trust, CSR and Firm Value ...... 83 Table 1.12 Regressions of CEO-FEs on CSR & Additional Controls . . . 84 Table 1.13 Regressions of Turnover on ∆CSR & Additional Controls . . 85 Table 1.14 Regressions of Predicted CEO Turnover on CSR ...... 86 Table 2.1 Variable Definitions ...... 134 Table 2.2 Descriptive Statistics ...... 137 Table 2.3 Regressions of Firm Performance on Internal Governance for Younger and Older CEOs ...... 138 Table 2.4 Changes in Investment Rates Around CEO Turnover ...... 140 Table 2.5 Changes in Investment Rates Around CEO Turnover For Older/Younger CEOs and Firms with Good/Bad Internal Governance ...... 141 Table 2.6 Regressions of Changes in Investment Rates Before and in the Year of CEO Turnover (-2,0) on Internal Governance for Older/ Younger CEOs ...... 142 Table 2.7 Regressions of Changes in Investment Rates After CEO Turnover (1,+2) on Internal Governance for Older/Younger CEOs 143

xi Table 2.8 Regressions of Property Sales in the Year of CEO Turnover on Internal Governance for Older/Younger CEOs ...... 144 Table 2.9 Regressions of Gains or Losses on Property Sales in the Year of CEO Turnover on Internal Governance for Older/Younger CEOs 145 Table 2.10 (Robustness Test 1): Regressions of Changes in Investment Rates Around Sudden CEO Deaths (-2, 0) For Older CEOs . . . . 146 Table 2.11 (Robustness Test 2): Regressions of Investment Policy Vari- ables Before and in the Year of CEO Turnover Controlling for Performance-Related Turnover, Hiring of Outsider CEO, and CEO Pay-Performance Sensitivities ...... 147 Table 2.12 (Robustness Test 3): Regressions of Firm Performance on Internal Governance and CEO Pay Slice for Younger and Older CEOs ...... 148 Table 2.13 (Robustness Test 4): Regressions of Investment Policy Vari- ables Before and in the Year of CEO Turnover on Internal Gov- ernance for Older/Younger CEOs Controlling for CEO Power and Geographical Complexity ...... 149 Table 2.14 (Robustness Test 5): Regressions of Investment Policy Vari- ables After CEO Turnover (1,+2) on Internal Governance for Older/Younger CEOs Controlling for CEO Power and Geographical complexity ...... 150 Table 3.1.A Variable Deﬁnitions ...... 270 Table 3.1.B Coding Rules for MFs Accuracy ...... 272 Table 3.1.C Construction of CEO Power Index of Prestige Power, Ex- pertise Power and Ownership Power ...... 273 Table 3.1.D Data Merging and Sample Selection Procedures ...... 274 Table 3.2.A Summary Statistics of Total Sample ...... 275 Table 3.2.B Test of Diﬀerence Between Firm-Year Observations With and Without Voluntary Disclosure ...... 276

xii Table 3.3 Simple Quadratic Regressions of Disclosure Quality Metrics on δ ...... 281 Table 3.4.A Multivariate Regressions of Disclosure Quality Metrics on δ for the Whole Sample ...... 282 Table 3.4.B Multivariate Regressions of Disclosure Quality Metrics on δ for the Sample of Older CEOs ...... 283 Table 3.4.C Multivariate Regressions of Disclosure Quality Metrics on δ for the Sample of Younger CEOs ...... 284 Table 3.5.A Regressions of Disclosure Quality Metrics on δ for the Sample of Older CEOs with Heckman Correction ...... 285 Table 3.5.B Regressions of disclosure quality metrics on δ for the sample of older CEOs With endogeneity control ...... 286 Table 3.5.C Regressions of Disclosure Quality Metrics on δ for Older CEOs With Heckman Correction and Endogeneity Control . . . . . 287 Table 3.6.A Regressions of Info on Stock Liquidity Metrics for the Whole Sample ...... 288 Table 3.6.B Regressions of Info on Stock Liquidity Metrics for Older CEOs ...... 289 Table 3.6.C Regressions of Info on Metrics of Stock Liquidity for Younger CEOs ...... 290 Table 3.7.A Equal-Weighted and Value-Weighted δ Portfolio Alpha by Fama-French 3 Factor Model ...... 291 Table 3.7.B Double Sorted δ Portfolio Alpha by Fama-French 3 Factor Model ...... 292 Table 3.8 Meditation Analysis of δ’s Impact on Valuation Through In- formation Eﬃciency ...... 293 Table 3.9.A Meditation Analysis of the Predictability of δ for Future Raw Return Through Information Eﬃciency ...... 294

xiii Table 3.9.B Meditation Analysis of the Predictability of δ for Future Raw Return Through Information Eﬃciency ...... 295 Table 3.10.A Regressions of Disclosure Quality Metrics on δ With Alter- native Endogeneity Control ...... 296 Table 3.10.B Regressions of Disclosure Quality Metrics on δ With Alter- native Endogeneity Control and Heckman Correction ...... 297 Table 3.11.A Regressions of Disclosure Quality Metrics on Alternative Delegation Responsibilities Measure ...... 298 Table 3.11.B Regressions of Disclosure Quality Metrics on Alternative Delegation Measure With Endogeneity Control ...... 299 Table 3.11.C Regressions of Disclosure Quality on Alternative Delegation Measure With Endogeneity Control and Heckman Correction . . . . 300 Table 3.12.A Regressions of F AbsBias on δ Quantiles ...... 301 Table 3.12.B Regressions of F AbsError on δ Quantiles ...... 302

xiv LIST OF ILLUSTRATIONS

Figure 1.1 An Integrated Theoretical Framework ...... 87 Figure 1.2 CEO’s Investment Cycle in Social Capital ...... 88 Figure 2.1 CEO’s Investment Cycle Surrounding CEO Turnover . . . . . 140 Figure 2.2 CEO’s Investment Cycles Surrounding CEO Turnover by Age and Good/Bad Internal Governance ...... 141 Figure 3.1 Graphical Analysis of the Curvilinear Relation Between F AbsBias and δ ...... 277 Figure 3.2 Graphical Analysis of the Curvilinear Relation Between F AbsError and δ ...... 278 Figure 3.3 Graphical Analysis of the Curvilinear Relation Between Accuracy and δ ...... 279 Figure 3.4 Graphical Analysis of the Curvilinear Relation Between Optimism and δ ...... 280

xv CHAPTER 1: CEO, CSR AND FIRM VALUE 1

Chapter 1: CEO’s Investment Cycle in Social Capital and Its Impact on Firm Value

Ivan E. Brick† Yankuo Qiao†

In the extant literature, the CEO specific effect is recognized as a persistent and significant impact on various firm policies such as financing and investment. In this paper, we demonstrate that there is a CEO-centric influence upon the level of

ﬁrm investment in social capital, as measured by Corporate Social Responsibility

(CSR). We hypothesize that if managerial style is an important inﬂuence upon the ﬁrm’s investment in CSR we would expect the greatest change in the level of

CSR to occur when the new CEO takes over the company. We present empirical evidence in support of this hypothesis. Moreover, we ﬁnd that the sudden change in CSR during the transitory period of CEO turnover is likely to be permanent in that the new CEO will on average not reverse the policy later in her tenure and

†Rutgers Business School–Newark and New Brunswick CHAPTER 1: CEO, CSR AND FIRM VALUE 2 the social policy becomes stable over time. One exception to this general pattern of CSR policy change is when the new CEO is an outsider. The latter finding is consistent with attunement theory which states that an outsider successor may be reluctant to aggressively change CSR due to the initial lack of viable relationship and trustful dialogue with different stakeholder groups of the firm. Thus, the capability and incentive of the outsider successor for changing CSR activities will be undermined or be self-refrained, and the outsider CEO tends to continue her predecessor’s policy in the early stage of tenure. Furthermore, based on the CEO- centric effect and the cyclical pattern of social capital investment, we investigate the value relevance of CSR during CEO transition period, in which the investment in CSR changes dramatically. We find that CSR is value enhancing when “trust” becomes an important intangible asset during crisis period.

1 Introduction

The literature has widely documented that CEO has signiﬁcant impact on corporate policies and corporate ﬁnancial performance. For example, Malmendier and

Tate(2005) show that over-conﬁdent managers overinvest when they have abundant internal funds, but curtail investment when they require external ﬁnancing.

In addition, Malmendier, Tate, and Yan(2011) and Ben-David, Graham, and

Harvey(2013) demonstrate over-confident managers undertake less hedging than less confident managers. Similarly, we expect that personality traits of the man- CHAPTER 1: CEO, CSR AND FIRM VALUE 3 ager will influence the investment in social capital of the firm. Some preliminary linkages between the human factors and the firm investment in Corporate Social

Responsibility (CSR) have been proposed in the early literature on the theory of corporate social issues (see, for example, Holmstrom 1982; Mintzberg 1978).

Petrenko, Aime, Ridge, and Hill(2016) find that narcissistic CEOs tend to over invest in CSR activities that increase attention to the CEO. Tang, Qian, Chen, and Shen(2015) and McCarthy, Oliver, and Song(2017) find a negative relationship between CSR and CEO hubris. Therefore, CEO is likely to be the key element among the managerial factors that affect the policies and programs adopt and developed by the firm. For instance, UN Global Compact Annual Review,

2007, revealed that 71% of the surveyed ﬁrms whose CSR policies and practices were developed and managed by the CEO. In this paper, we provide empirical evidence that the CEO is the key element aﬀecting CSR policies and uncover the general pattern that delineates CEO’s investment cycle in social capital throughout tenure. Moreover, we demonstrate that the CEO, on average, chooses CSR policies that are inconsistent with managerial consumption of perquisites.

If CEO personal attributes influence corporate policies then one would expect that when a CEO is replaced, if the new CEO personality traits and experiences differ from the old CEO then one should expect a noticeable change in firm policy.

For example, Weisbach(1995) and Pan, Wang, and Weisbach(2016) show that during the turnover transition period, the investment rate of the firm decreases CHAPTER 1: CEO, CSR AND FIRM VALUE 4 and the rate of asset disposition increases.1 It makes sense to expect the largest annual changes in firm CSR policy occur if the firm experiences a CEO turnover since the new CEO brings to the firm a different managerial style, personality and biases. We present empirical evidence in support of this hypothesis. In addition, as the tenure of the CEO increases, corporate policies stabilize. Pan et al.(2016) empirically demonstrate that this is the case for the firm’s investment and asset disposition rate. Consequently, we expect a similar stabilization of CSR policy as the new CEO’s tenure increases. This paper documents that on average, the biggest change in the CSR activities occurs during the first two years of the new

CEO ascendancy. Moreover, the paper demonstrates that over time the CEO will not reverse the established tone of CSR policy at the beginning of her tenure.

The results may be aﬀected if we do not distinguish between forced and voluntary turnover. We therefore replicate our results after eliminating from the sample all forced CEO turnovers.2

It is possible that the changes in CSR policy are diﬀerent if the successor is from outside the organization. One might argue that the change in CSR is greatest if the board of directors install an outsider since the Board wants a directional

1 These empirical results are also consistent with the replacement of a myopic CEO who nears retirement by a CEO with a longer investment horizon, as modeled by Acharya, Myers and Rajan (2011). 2 It is possible that CEO turnover may be the result that the board of directors wishes to change the CSR policy of the firm. But if this were the case, then our results would have been affected when we removed forced turnovers from our sample. It is also possible, however, that the board of directors do not want to remove the CEO for this reason and wants to wait until the CEO to voluntarily retire to affect firm’s CSR policy. If that were the case, we should expect the biggest change in CSR policy to occur when an outside CEO is appointed. But our results support the attunement theory discussed below which predicts the opposite. CHAPTER 1: CEO, CSR AND FIRM VALUE 5 change for the firm. Additionally, the managerial style of the outsider should be greatly different from the outgoing CEO. On the other hand, according to the attunement theory (see for example, Orlitzky and Swanson 2002), an outsider successor may be reluctant to aggressively change CSR due to the initial lack of viable relationship and trustful dialogue with different stakeholder groups of the

firm. Thus, the capability and incentive of the outsider successor for changing CSR activities will be undermined or be self-restrained, and the outsider CEO tends to continue his or her predecessor’s policy in the early stage of tenure. Alternatively, the appointment of an outsider may imply that the board of directors would like directional change for the firm with respect to corporate policies such as investment, financing and organizational structure. Due to limited attention of the management team, (see, for example, Barkema and Schijven 2008; Bettman,

Johnson, and Payne 1986; Haleblian and Finkelstein 1993; Payne, Bettman, and

Johnson 1988) it is likely that the new CEO will give priority to these matters before considering a change in the CSR policy of the firm. In particular, according to Tang, Mack, and Chen(2018), the social responsible activities that benefit the welfare of stakeholders are essentially different from other economic- and strategic- driven decisions such as investment in tangible assets, research and development, and merger and acquisitions, which are closely associated with shareholder value and are likely the priority of the board. We present empirical evidence in support of the attunement theory and the limited attention hypothesis. The sudden change in

CSR policy during CEO transition is smaller if the new entering CEO is considered CHAPTER 1: CEO, CSR AND FIRM VALUE 6 as an outsider. As far as we know, no one has previously provided any direct test of the attunement theory.

There has been little research demonstrating the level of importance of the

CEO-centric influence upon CSR activities. While research has shown that such personality traits as CEO confidence and narcissism affects the level of CSR,3 these papers show that such traits are contributing factors. The empirical problem is that unless the researcher uses appropriate proxies for all possible CEO personality traits, skill sets, leadership styles and life experiences, one cannot ascertain as to whether these idiosyncratic traits are the major determinants of the level of CSR investment. We demonstrate that the CEO influence is the major determinant of the CSR activity.

The only exception to our knowledge is Kang(2017) who shows that CEO’s inﬂuence is a major determinant of the level of CSR. Our paper diﬀers from Kang

(2017) in two ways. Firstly, from a standpoint of empirical design, we primarily examine the influence of the successor CEO upon changes in CSR policy so we can better disentangle the CEO fixed effects from the firm fixed effects. In particular, it could be argued that the CEO fixed effects dummies are just a correlated proxy for unobservable firm fixed effects especially since the number of CEOs in our sample that has worked in two firms is very limited. It is also possible that economic and statistical significance of observable firm characteristics used in determinants

3 See, for example, McCarthy et al.(2017); Petrenko et al.(2016); Tang et al.(2015) CHAPTER 1: CEO, CSR AND FIRM VALUE 7 of CSR studies are greatly diminished if firm specific variables do not vary over time. Consequently, we study the changes in CSR policy surrounding the CEO turnover event to disentangle CEO fixed effects from firm fixed effects. Secondly and more importantly, the empirical setting proposed in this paper allows us to identify the timing of a CEO’s investment in CSR and the general pattern of

CSR policy change throughout the CEO’s tenure. Our paper finds that CEOs on average tend to change the investment in CSR at the beginning of their tenure and are less likely to reverse or significantly change CSR policy later in their tenure, leading to a stable CSR policy over time and making the original change permanent. Our paper also complements Bernard, Godard, and Zouaoui(2018) who study the impact of CEO turnover for French firms. That study uses the ratings provided by Viego of French firms which measure output or effectiveness of CSR investment. Bernard et al.(2018) find that there is lagged affect upon CSR outcomes following the CEO turnover. This is consistent with our results that the

CSR input change at the beginning of the CEO transition which ultimately leads to profound impact on sustainable performance.

The remaining question is whether the resulting change in CSR is improving

ﬁrm value. One might interpret that since CSR investment is CEO centric, CSR acts like a perquisite for the CEO and does not add value to the ﬁrm. For example,

Masulis and Reza(2015) ﬁnd that corporate giving is positively associated with charity preferences of CEOs and members of the board of directors. Moreover,

CEO affiliated charity contributions decline with the level of CEO ownership of CHAPTER 1: CEO, CSR AND FIRM VALUE 8 the firm. However, one might argue that the CEO’s CSR investment strategy reflects her skill set to manage the company. Consequently, if the skill set of the new CEO is not able to employ CSR assets profitably as did the former CEO, this new CEO would choose not to build up social capital and invest less in CSR activities than the previous CEO. It also might be true that the current CEO’s skill set does not match well with CSR investments. However, the new CEO has that skill set and is capable of transforming social capital into the capital stock of the

firm, in which case you might find an increase in CSR investment. All the above possibilities pertaining to the variation of CSR investment and its value-relevance are centered on the skill set of CEOs. If so, then we should not expect the changes in CSR policies around the transition of the new CEO negatively impacting firm value. The advantage of using the CEO transition period to understand how

CSR impacts firm value is that our paper demonstrates that the CSR policy changes occur during the first two years of the new CEO’s tenure, allowing us to compare the impact of sudden changes in CSR policy upon firm value. To examine this question, we adopt a matching method as our inference strategy given the strong simultaneity and reverse causality between firm value and social provisions. Specifically, for each year and industry sector, we form two groups of turnover firms with similar size of total assets. The treatment group is of firms whose CEOs adjust CSR significantly while the control group consists of firms whose CEOs adjust CSR moderately. We find that the change of T obin0s Q is positively related to the change of CSR promulgated by the incoming new CEO. CHAPTER 1: CEO, CSR AND FIRM VALUE 9

We also ﬁnd that changes the new entering CEOs made during 2003 and 2004 are positively associated with the holding period returns during the ﬁnancial crisis the

US endured towards the end of that decade. This latter result is consistent with the

ﬁndings in Lins, Servaes, and Tamayo(2017) that CSR activities engender trust among corporate stakeholders.4 Thus we can infer that on average, CSR activities do not serve as CEO perquisites but rather as value enhancing investments.

Our paper makes the following contributions to the literature. First, the paper complements the ﬁndings of studies on CSR determinants (see, for example,

Brick, Venezia, and Palmon 2018; Kang 2017; Manner 2010; McCarthy et al. 2017;

McWilliams and Siegel 2001; Petrenko et al. 2016; Stanwick and Stanwick 1998;

Tang et al. 2018, 2015; Thomas and Simerly 1994). In particular, we empirically demonstrate that the CEO-centric influence is the dominating determinant of CSR. Moreover, our finding is further strengthened by the empirical evidence that the change of CSR is more significant in the period surrounding CEO turnover than subsequent periods. Second, our empirical analyses contribute to the literature on the relationship between CEO characteristics and managerial styles (see for example Bertrand and Schoar 2003) by providing evidence that in addition to

firm financial, investment and strategic organization policies, CEO has persistent impact on CSR as well. The findings are in support of the upper-echelons theory of Hambrick and Mason(1984) and stakeholder theory (Donaldson and Preston

4 Also see Buchanan, Cao, and Chen(2018) who find that firms with high institutional ownership most enjoy the positive impact of engendering trust upon firm value during the financial crisis. CHAPTER 1: CEO, CSR AND FIRM VALUE 10

1995; Freeman 1984; Jones 1995). Third, we illustrate that, among CEO characteristics, CEO tenure and outsider attribute are two dominating factors inﬂuencing

CSR. The empirical findings are also consistent with the attunement theory (see, for example, Orlitzky, Schmidt, and Rynes 2003) in that outsider successors are self-restrained to change CSR aggressively due to lack of mutual trust with various stakeholder groups. Finally, our paper contributes to the literature on the value relevance of CSR. We provide additional evidence that CSR activities enhance firm value during crisis period when trust of investors becomes desperately important to a firm’s capital stock. The empirical evidence indicates that sudden change in CSR policy during the transitory period of CEO turnover is not driven by the myopic motives of CEOs. Moreover, complementary to the most extant studies of CSR and trust, we find that trusted firms are more resilient in response to the devastating effect of financial crisis.

The structure of the paper is organized as follows: Section 2 presents related literature that provides the motivation for our empirical analysis. Section 3 develops the hypotheses that we propose for the empirical tests. Section 4 formulates the regression models for the tests corresponding to hypotheses. Section 5 describes the data and provides summary statistics. Section 6 reports the results and interpretations of the empirical tests in accordance with the hypotheses and theory. Section 7 performs robustness check for the generality of the empirical results. Section 8 gives concluding remarks and a brief outlook on the further research. CHAPTER 1: CEO, CSR AND FIRM VALUE 11

2 Theoretical Background and Literature Review

McWilliams and Siegel(2001) provide a broad deﬁnition that CSR encompasses actions to further some social good to one that is narrowly focused on maximizing shareholder wealth (Friedman 1970). CSR is in essence how an organization responds to the expectations and demands of all the stakeholders not only in the business relationships, but also in the external society. For the purpose of this research, we follow the comprehensive viewpoint of Carroll(1999), stating that

“the social responsibility of business encompasses the economic, legal, ethical, and discretionary expectations that society has of organizations at a given point in time” (Carroll 1979, p.500). This inclusive perception of CSR is aligned with the framework of modern stakeholder theory (Freeman 1984), in which the corporation is expected to perceive and attune interests of external stakeholders such as employee, customers, local community and government, in addition to acting in maximizing shareholders’ wealth.

The signiﬁcant inﬂuence of CEO characteristics in determination of corporate activities is widely documented outside the academic literature such as in business press and public media, and is perceived as a prevailing view among managers.

Hambrick and Mason(1984) synthesize the past studies across different fields in literature and propose the theory of upper echelons perspective, serving as a standard paradigm for the perceptual process of decision making centered at CHAPTER 1: CEO, CSR AND FIRM VALUE 12 the CEO characteristics. According to Hambrick and Mason(1984), the decision making and strategic choice are complex processes heavily influenced by behavioral factors instead of pure economic optimization mechanism. Hambrick(2007) stated

“The central premise of upper echelons theory is that executives’ experiences, values, and personalities greatly influence their interpretations of the situations they face and, in turn, affect their choices”. The behavioral components influence the decision maker, namely, the CEO. There is evidence that CEO characteristics affect CSR policies. In particular, Petrenko et al.(2016) find that narcissistic

CEOs tend to over invest in CSR activities that increase attention to the CEO.

Tang et al.(2015) and McCarthy et al.(2017) ﬁnd a negative relationship between

CSR and CEO hubris. Therefore, CEO is likely to be the key element among the managerial factors that affect the social policies and initiatives adopt and developed by the firm. Therefore, both implicit and explicit characteristics of top managers will play a central role of filtration in the process where strategic choices are formed for a better firm performance. The upper-echelons model provides a theoretical foundation for a preliminary linkage between top managers and CSR.

In support of this “upper echelons perspective”, Bertrand and Mullainathan

(2001) document empirical evidence that managerial styles of individual managers affect the outcomes of corporate policies by specifying the manager fixed effects consistently across different companies. The cross-sectional analysis on a manager-

ﬁrm matched panel data set captures large extent of heterogeneity in investment,

ﬁnancial and organization activities left unexplained by previous empirical mod- CHAPTER 1: CEO, CSR AND FIRM VALUE 13 els such as Titman and Wessels(1988), Smith Jr and Watts(1992), and Bradley,

Jarrell, and Kim(1984) that accounted for firm-, industry- and market-level characteristics. As such, Hambrick and Mason(1984) and Bertrand and Schoar(2003) provide theoretical and empirical underpinning that the involvement of individual managers, especially CEO, do have significant influence upon the firm polices given other factual conditions in the overall business environment.

The managerial attributes of CEO that inﬂuence ﬁrm policies may be di- chotomized into two comprehensive categories: observable characteristics and unobservable characteristics. The observable characteristics of CEO are mainly described in demographic variables such as age and tenure, which are unambiguous, convenient to measure and found favor in diverse areas of research, such as sociology, political science and marketing (Hambrick and Mason 1984; Thomas and

Simerly 1994). Even in the early literature of organizations, a signiﬁcant amount of studies has recognized the importance of manager demographic attributes in the process of shaping organizational outcomes (see, for example, Chaganti and Samb- harya 1987; Gupta and Govindarajan 1984; Stinchcombe, McDill, and Walker

1968; Thomas, Litschert, and Ramaswamy 1991). For instance, age and tenure can properly gauge the managerial experience as well as capture the risk preference. Bertrand and Schoar(2003) document that some CEO-specific demographic variables such as education, cohort and tenure have significant effect on CEO’s managerial style. Especially, a CEO in earlier birth cohort tends to be conservative and run the company with lower level of investment rate and lower leverage, CHAPTER 1: CEO, CSR AND FIRM VALUE 14 while a CEO with MBA degree tends to invests aggressively. Traditionally, the literature has used various proxies to capture the personality traits and life experiences of the CEO to demonstrate the influence of the CEO upon corporate policies. But clearly no set of proxies can fully capture all of the personality traits and experiences of the CEO and as a result, many studies may suffer from omitted variable bias. In this study we use CEO fixed effects and the corporate event of

CEO turnover to remove such bias.

Our paper empirically demonstrates the CEO-centric impact upon corporate social responsibility (CSR) activities, and explore various characteristics, both observable and unobservable, of CEO to identify the underlying CEO-related factors aﬀecting CSR policies. Furthermore, we document that CEO attributes are the main drivers of CSR policy. Papers such as Petrenko et al.(2016), Tang et al.

(2015) and McCarthy et al.(2017) find that specific CEO characteristics affect

CSR, but they do not explore the strength of these determinants. They cannot explore the strength of these determinants because they are not controlling for all CEO characteristics simultaneously. Our approach enables us to examine the strength of CEO characteristics because the methodology controls for all CEO centric omitted variables.

The remaining question is whether the resulting change in CSR during the new

CEO transition period is improving ﬁrm value. One might interpret that since

CSR investment is CEO centric, CSR acts like a perquisite for the CEO and does CHAPTER 1: CEO, CSR AND FIRM VALUE 15 not add value to the firm. For example, Masulis and Reza(2015) find that corporate giving is positively associated with the charity preferences of the CEO and the board of directors. Moreover, CEO affiliated charity contributions decline with the level of CEO ownership of the firm. This result is consistent with the arguments proposed by Friedman(1970) and Jensen and Meckling(1976) who viewed expenditures on CSR as a part of CEO non-pecuniary consumption incurred by the owner-manager behavior. Similarly, Brown, Helland, and Smith(2006) find that corporate philanthropy is associated with firms with less efficient monitoring of the management such as firms with large number of members of board of directors.

In contrast, the instrumental stakeholder theory (see, for example, Donaldson and

Preston 1995; Freeman 1984; Jo and Harjoto 2012; Jones 1995; Lins, Servaes, and

Tamayo 2019) argues that CSR can improve the value of the ﬁrm. Participating in

CSR allows the firm to retain or attract better employees, reduce the possibility of regulatory action against the company, and increase brand awareness of the products. Ferrell, Liang, and Renneboog(2016) find support that firms strongly engage in CSR are associated with better corporate governance and with fewer agency conflicts. Consistent with instrumental stakeholder theory, there is a large body of empirical literature suggesting a positive linkage between CSR and firm performance. For example, a meta-analysis of 52 studies conducted by Orlitzky et al.(2003) confirms the existence of a bidirectional and simultaneous relationship between CSR and firm performance as opposed to a trade-off. More specifically, there are studies that show that a higher CSR score is on average associated with CHAPTER 1: CEO, CSR AND FIRM VALUE 16 a lower cost of capital (Dhaliwal, Li, Tsang, and Yang 2011; El Ghoul, Guedhami,

Kwok, and Mishra 2011; Goss and Roberts 2011). Other studies ﬁnd a positive association between ﬁrm value and how employees are treated (Edmans 2011, 2012;

Fauver, McDonald, and Taboada 2018).

However, the studies in general are divided and some papers demonstrate that

ﬁrms invest in CSR as a means to mitigate negative publicity regarding corporate activities (see, for example, Prior, Surroca, and Trib´o 2008; Yip, Van Staden, and

Cahan 2011; Zhang, Kandampully, and Choi 2014), or as a result of CEO’s desire to build socially elevated ﬁgure and public favorable image (see, for example,

Hayward, Rindova, and Pollock 2004). One reason for the mixed empirical results is that there is an endogenous relationship between CSR and ﬁrm performance that has been well documented in the literature (see, for example, Edmans 2011,

2012).

In recognition of the CEO centric eﬀect upon CSR, one might argue that the

CEO’s CSR investment strategy reﬂects her skill set to manage the company.

Consequently, if the skill set of the new CEO is not able to employ CSR assets profitably as did the former CEO, this new CEO would invest less in CSR activities than the previous CEO. It also might be true that the current CEO’s skill set does not match well with CSR investments. However, the new CEO has that skill set, in which case you might find an increase in CSR investment. All the above possibilities pertaining to the variation of CSR investment and its value-relevance CHAPTER 1: CEO, CSR AND FIRM VALUE 17 are centered on the skill set of CEOs. If so, then we should not expect that changes in CSR policies around the transition of the new CEO should impact negatively on firm value. The advantage of using the CEO transition period to understand how CSR impacts firm value is that our paper demonstrates that the CSR policy changes occur during the first two years of the new CEO’s tenure, allowing us to examine the impact of sudden changes in CSR policy upon firm value.

3 Hypotheses

One purpose of the paper is to empirically examine the strength of the CEO- centric influence upon CSR activities. Kang(2017) includes CEO fixed effects in his CSR empirical specification and finds that the CEO fixed effects dominate firm characteristics in predicting firm CSR activity. We further posit that the CEO

fixed effects capture the effect of the unique skill set embedded in the functional background of each individual CEO, in addition to other unobservable traits. This leads to our first hypothesis:

Hypothesis 1: The observable and unobservable CEO characteristics play a central role in determining CSR policy.

It could be argued that the CEO fixed effects dummies are just a correlated proxy for unobservable firm specific effects. According to the methodology introduced by Bertrand and Schoar(2003), the CEO fixed effects can be disentangled CHAPTER 1: CEO, CSR AND FIRM VALUE 18 from firm fixed effects only if we keep track of CEOs across firms and include in the sample individuals who had been appointed to CEO at least in two firms.

However, due to the limited number of ﬁrms with CSR data available, such sample composition is not applicable. Even if it was, it could be argued that since

CEOs are not appointed to firms randomly, firms that want to initiate more CSR provisions may hire CEOs who worked in high CSR firms, thereby overestimating the CEO fixed effects in determining CSR. From an empirical standpoint, it is also possible that economic and statistical significance of observable firm characteristics used in determinants of CSR studies are greatly diminished if firm specific variables do not vary much over time. In sum, notwithstanding the methodology introduced in Bertrand and Schoar(2003), the inclusion of high dimensional fixed effects dummies of both CEO and firm makes little contribution to disentangling one effect from the other in the context of our CSR study. Consequently, we study the changes in CSR policy surrounding the CEO turnover event to disentangle CEO fixed effects from firm fixed effects while controlling for the variations in firm-, industry- and market-level characteristics. It makes sense to expect the largest annual change in firm CSR policy occurs if the firm experiences a CEO turnover since the new CEO brings to the firm a different managerial style, personality traits and biases. That is, if each CEO does have unique effect upon CSR, we would expect to observe the biggest change of CSR in the period of ongoing managerial succession compared to those annual changes in subsequent periods.

One might argue that, similar to other long-term investment, the development of CHAPTER 1: CEO, CSR AND FIRM VALUE 19

CSR initiatives may take years to implement and function, resulting in signiﬁcant changes in CSR provisions in later years of the CEO’s tenure. Such concern is alleviated by the nature of our proxy for CSR level and will be addressed in section

5. Moreover, we expect that the change in CSR is associated with the event of

CEO turnover as opposed to the changes in other variables that have been used in the literature as determinants of CSR. From this novel perspective, we would be able to quantify the inﬂuence of CEO upon CSR activities. Therefore, we predict that:

Hypothesis 2: The change in CSR is associated with the event of CEO turnover.

To better understand the inﬂuence of the CEO upon CSR activities, we empirically examine whether changes of CSR policy diminishes as the tenure of the

CEO increases. We posit that, the CSR activities shall be altered by the new CEO more aggressively in his or her early tenure of the post-turnover period to reﬂect her skill set, biases, and personality traits. Once implemented and if the CEO’s personality traits and skill sets do not vary over time, the CSR policy under the new CEO will stabilize. Pan et al.(2016) empirically demonstrate that this is the case for the ﬁrm’s investment and asset disposition rate. Consequently, we expect a similar stabilization of CSR policy as the new CEO’s tenure increases.

Therefore, our next hypothesis is:

Hypothesis 3: In the post-turnover period, the annual change of CSR activities will stabilize such that the CEO will not reverse the policy change made by the CHAPTER 1: CEO, CSR AND FIRM VALUE 20 incoming CEO at the beginning of her tenure.

It is possible that the willingness of the new CEO to change CSR policy will depend on whether or not the new CEO appointed comes from outside the organization. An insider is a manager who has been promoted from within the company to become the CEO. An outsider is a manager from outside the company who has been appointed by the board to become the CEO. Brockman, Lee, and Salas(2016) suggests that insider-outsider attribute can roughly proxy the degree of CEO’s skill sets. In addition to insider-outsider attribute, the specifications of models in those studies also include other CEO-specific variables, such as age, board positions, and ownership. One might argue that the change in CSR is greatest if the board of directors install an outsider since the board wants a directional change for the firm. Additionally, the managerial style of the outsider should be greatly different from the outgoing CEO. On the other hand, according to the attunement theory (see for example, Orlitzky and Swanson 2002), an outsider successor may be reluctant to aggressively change CSR due to the initial lack of viable relationship and trustful dialogue with different stakeholder groups of the firm. Thus, the capability and incentive of the outsider successor for changing CSR activities will be undermined or be self-restrained, and the outsider CEO tends to continue his or her predecessor’s policy in the early stage of tenure.

The attunement model first proposed by Swanson(1999) and later by Orlitzky et al.(2003) provides a normative-descriptive framework of the firm and society, CHAPTER 1: CEO, CSR AND FIRM VALUE 21 solves the dilemma between what a corporation should or should not do and what a corporation can or cannot do on behalf of the social welfare, and identifies channels through which executive managers influence the CSR. Centered by CEO- led executives, the original portrayal of attunement (Swanson 1999) consists of four bi-directionally interconnected organizational constituents: normative receptivity of executives, hierarchical expansion, value-discovery culture and value expanded detection. According to Orlitzky and Swanson(2002), the CEO-led executives are receptive to social values in ways determined by their implicit and explicit characteristics. The values can thereby be conceivably conveyed to employees of the company through the formal channel of hierarchical structure of executive ladders or via informal channels of corporate culture. Simultaneously, different interest groups formed by internal and external stakeholders may put pressure on the CEO and circumscribe the process of value discovery and expansion. Building on the previous studies (see, for example, Cohen and Prusak 2001; Hosmer 1995), attunement theory also recognizes the important role of viable relationship or trustful dialogue in organizational processes, which are necessary to collectively hold the bidirectional networks for attunement.

According to the attunement theory, the influence of a new CEO upon CSR depends on how quickly the new CEO becomes attuned to the importance of the various stakeholder communities in effectively running the company. Thus, especially in the early stage after succession, the outsider may be reluctant to change the CSR due to his or her difficulties of recognizing the roles and value impact CHAPTER 1: CEO, CSR AND FIRM VALUE 22 of the various interest groups inside and outside the company, and is therefore incapable of alternating CSR activities or reluctant to do so. Alternatively, the appointment of an outsider may imply that the board of directors would like directional change for the firm with respect to corporate policies such as corporate strategy, investment and financing. Due to limited attention of the management team, (see, for example, Barkema and Schijven 2008; Bettman et al. 1986; Hale- blian and Finkelstein 1993; Payne et al. 1988), it is likely that the new CEO will give priority to these stockholder matters before considering a change in the CSR policy of the firm that affects outside stakeholders. Accordingly, we posit:

Hypothesis 4A: An outsider CEO is less likely to change CSR activities compared to an insider CEO as predicted by the attunement theory.

However, on the other hand, an outsider successor may aggressively alternate the policies related to CSR activities, based upon the personal experience of her previous managerial positions outside the company and the desire of the board of directors to initiate a change. That is, when a firm decides or has to hire an outsider CEO, the usual practice is to form a particular committee in search for appropriate candidates. The hiring process involves extensive searching activities to pinpoint candidates with the right experience, functional track and skill set that match the profile of the firm. If in fact we observe that outsider successors on average change CSR more aggressively, it would mean that the change in CSR is not necessarily driven by the CEO but by the board of directors and therefore CHAPTER 1: CEO, CSR AND FIRM VALUE 23 our current empirical design is overestimating the effect of CEO. In contrast, if the empirical finding is not in support of such rationale, our previous argument about the CEO-centric effect upon CSR is further corroborated and our empirical setting is, to a large degree, establishing a causal relation that the sudden change in CSR policy during CEO transitions is as a result of the shift in value, personal traits and skill set between outgoing and incoming CEOs. Therefore, to formally investigate the effect and implication of the outsider attribute, we propose an alternative hypothesis.

Hypothesis 4.B: An outsider CEO is more likely to aggressively change CSR activities compared to an insider CEO due to his or her background of past work experience.

One might interpret that since CSR investment is CEO centric, CSR acts like a perquisite for the CEO and does not add value to the firm. However, one might argue that the CEO’s CSR investment strategy reflects her skill set to manage the company. Consequently, if the skill set of the new CEO is not able to employ CSR assets profitably as did the former CEO, this new CEO would invest less in CSR activities than the previous CEO. It also might be true that the current CEO’s skill set does not match well with CSR investments. However, the new CEO has that skill set, in which case you might find an increase in CSR investment.

All the above possibilities pertaining to the variation of CSR investment and its value-relevance are centered on the skill set of CEOs. If so, then we should not CHAPTER 1: CEO, CSR AND FIRM VALUE 24 expect that changes in CSR policies around the transition of the new CEO should negatively impact ﬁrm value.

Hypothesis 5: If the new CEO signiﬁcantly changes the investment in CSR, the value of the ﬁrm should not decrease.

We argue that all of our 5 hypotheses are interdependently related. Hypoth- esis 1 validates the general premise of CEO-centric effect of CSR policy. The empirical methodology used to test Hypothesis 2 disentangles managerial effect from firm effect on CSR policy change. Hypothesis 3 and Hypothesis 4 help identify that the CEO is the main determinant of the CSR policy. The remaining question is that if the CEO is main driver of the CSR policy, is CSR serving as a perquisite for the CEO? We investigate the value relevance of CSR from novel perspectives. Figure 1 portrays an integrated conceptual diagram of the theoretical model that frames our hypotheses development and empirical tests, where the key variables of concern and the corresponding theories in the literature are in the shapes of oval and rectangular, respectively, and the single and two-way arrows delineate the primary relationships suggested by the postulated hypotheses for different CEO tenure stages separated by the dashed lines. CHAPTER 1: CEO, CSR AND FIRM VALUE 25

4 Methodology

We hypothesize that each individual CEO’s unique skill set is the key element of determining CSR policy. Thus, we should include dummies capturing for the

CEO’s personal inﬂuence in the model speciﬁcation of the determinants of CSR.

The regression model is constructed in the form below to test whether the explanatory power of CEO fixed effects outweigh other firm-specific variables that are usually specified as determinants of CSR. In particular:

CSRit = β0 + β1Φk + γXit + δj + θt + εit (1)

where the dependent variable is CSR. We measure the level of investment in CSR, using KLD net scores from KLD Research & Analytics, Inc. SOCRATES database.

A more detailed description of the variable is provided in the data section of the paper. Φk denotes CEO fixed effects, Xit is the set of standard control variables for firm characteristics, δj and θt are industry, and year fixed effects, respectively.

According to Hypothesis 1, we expect that the explanatory power of CEO fixed effects to outweigh that of firm characteristics.

More speciﬁcally, Xit proxies for ﬁrm characteristics that are documented in the literature as determinants of CSR (see, for example, Cordeiro and Tewari 2015;

McWilliams and Siegel 2001; Udayasankar 2008). We control for the size of the CHAPTER 1: CEO, CSR AND FIRM VALUE 26

firm by including the logarithm of total assets (AT ). The literature has shown that CSR is negatively related to the financial risk of the firm so we include a control variable of the market leverage ratio (LR) which is defined as long-term debt divided by the sum of the total debt and market value of equity. Since more profitable firms are better able to afford investment in CSR and consequently, we include as a control variable ROA. Firms that have high growth potential and require greater future discretionary investment are more likely to reduce investment in CSR. Accordingly, we include the ratio of the market value of equity to the book value of equity (M/B) as proxies for expected future discretionary investment, as well as the level of research and development expenditure (R&D) and the level of advertisement expense (ADV ), both scaled by total assets. We also include control variable selling, general and administrative expenditure scaled by total assets (SG&A) and the ratio of net plant, property and equipment to the

ﬁrm’s total assets (PP &E) as proxy for the level of ﬁrm tangibility. We follow

Brick et al.(2018) who include corporate marginal tax rate ( MTR) and they ﬁnd that the level of CSR is positively related to the marginal tax rate of the ﬁrm.5

The higher the tax rate, the less expensive in terms of after-tax dollars is the investment in CSR. Similarly, we posit, as does Brick et al.(2018), that the ﬁrm’s willingness to engage in earnings management can be construed as either its willingness to test the borders of ethics or as disinterest in transparency or fairness.

In particular, the proxy we use to measure the ethical judgment of the CEO is

5 The corporate marginal tax rates provided by John Graham via his website: https://faculty.fuqua.duke.edu/ jgraham/taxform.html. CHAPTER 1: CEO, CSR AND FIRM VALUE 27 the diﬀerence between the level of accruals scaled by the total assets and the average level of accruals of the industry classiﬁed by the Fama-French 49 approach

(exACCR). The amount of accruals is calculated as the difference between net income and cash from operations (i.e., the difference between items IB and OANCF in COMPUSTAT) and it is a proxy for the effort of the CEO to manage earnings of the firm. We identify industry fixed effects using two alternative approaches, the Fama-French 49 industry classification and the 4-digit SIC codes.

According to Hypothesis 2, the change in CSR is associated with the event of

CEO turnover as opposed to the changes in other variables currently well-used as determinants of CSR. To test the hypothesis, we estimated the following regression labeled as model 2:

∆CSRit = β0 + β1Φit + γ∆Xit + η∆Zit + δj + θt + εit (2)

where the dependent variable, denoted as ∆CSR, is the change in CSR. We can also use the absolute value change in CSR since we are trying to estimate how the new CEO deviates from the policies implemented by the old CEO and not whether or not CEOs increased or decreased the level of CSR investment. In this case, for the sake of consistency, we would use the absolute value of the change for each of our control variables. The results are strictly analogous and therefore we generally do not report the results except for the simple version. CHAPTER 1: CEO, CSR AND FIRM VALUE 28

Φit denotes a dummy tenure variable that takes the value of unity during the specific CEO turnover transition period and zero otherwise. Consistent with past studies (see, for example, Huson, Malatesta, and Parrino 2004; Murphy and Zim- merman 1993; Parrino 1997; Weisbach 1988), we define the period surrounding succession process, in which the turnover effect takes place, is the first full year of the new CEO. In accordance with our hypothesis, we shall expect to see positively significant coefficient of Φit, indicating that the event of CEO turnover contributing to the change in CSR.

∆Xit represents the vector of the changes of standard control variables for the firm characteristics used in previous models. In alternative specifications, we also use absolute change of independent variables. We obtain strictly analogous results demonstrating that CEO fixed effects are dominant. Hence, to save space we only present the results of changes. ∆Zit is the vector of the levels or changes of CEO-specific variables, such as age, gender, board positions and ownership that have been used in prior studies (see, for example, Bertrand and Schoar 2003;

6 Brockman et al. 2016; Kang 2017). δj and θt are industry, and year fixed effects, respectively. Although the specification adopts a first difference form (FD), one can hardly assert that the unit level effect is truly fixed over time. According to

Grieser and Hadlock(2019), signiﬁcant proportion (over ﬁfty percent) of studies in the extant empirical literature ignores the potential serial correlation within

6 For some variables in the control group which identifies the static status, such as the gender or board positions, we use categorical level variables instead of change of values. CHAPTER 1: CEO, CSR AND FIRM VALUE 29 unit levels while implementing fixed-effect (FE) and first-difference (FD) models.

Therefore, to mitigate the potential time trend within industries, we choose to include industry fixed effects dummies to control relatively macro-level intertem- poral correlation, which is deemed to be more severe. As such, our econometric specification demonstrates combined features of fixed effects model and first difference model (FE-FD). Apparently, the dependent variable of the absolute value of CSR change follows asymmetric non-Gaussian distribution. As such, in addition to linear models, we propose to utilize maximum likelihood techniques to estimate a Generalized Linear Model (GLM), in which we conjecture that the dependent variable follows a gamma distribution and apply log link function for the estimation. Constrained by the algorithm of the nonlinear maximum likelihood estimation, we couldn’t include relatively high dimensional fixed effects dummies identified by 4-digit SIC codes, which significantly increase the complexity of the model specification. Thus, we only use Fama French industry classification when estimating the GLM. Alternatively, considering the absolute value dependent variable as right censored at zero, we use Tobit models in addition to GLM and linear regression models to further validate the statistical inference of the empirical tests.

According to Hypothesis 3, we expect that the CEO shall aggressively change the corporate social responsibility (CSR) activities in the early stage of his post- turnover tenure and the dramatic change will gradually wind down as he or she remains longer in the office. In the new specification of model 2, Φit contains the explanatory variables, T urnoverit and T enurek. T enure2 6 equals one if the − CHAPTER 1: CEO, CSR AND FIRM VALUE 30 tenure is greater than two years but less or equal to 6 years. We chose 6 years as the breakoff because the sample median tenure is 6 years. Our third tenure variable is T enure6+ which equals one if the tenure of the CEO is greater than 6 years.

In order to test the relationship between the change of CSR and the CEO’s outsider attribute, we adjust model 2 in which contains the explanatory variable,

Outsiderit. We define the dummy variable Outsiderit taking value 1 during CEO transition if the individual is appointed to CEO within 2 years since joining the company. If the coefficient of Outsiderit is positively significant, the empirical evidence is in favor of Hypothesis 4.A, meaning that the outsider attribute of CEO tends to drive the fluctuation of CSR in general. If the coefficient of

Outsiderit is negatively signiﬁcant, the empirical evidence is in favor of Hypoth- esis 4.B, meaning that the CEO with outsider attribute tends to avoid changing

CSR much in general. We hypothesize that the sudden change in CSR policy during the transitory period of CEO turnover reflects the heterogeneity in managerial styles, values and skill sets between outgoing and incoming CEOs and therefore shouldn’t cause detrimental effect on firm value. For example, if the new entering

CEO couldn’t manage social assets as proﬁtable as her predecessor, the optimal policy for the new CEO is to disinvest in CSR activities. In order to test the value relevance of the sudden change in CSR policy as a result of CEO turnover, we propose to estimate the following regression model labeled as model 3: CHAPTER 1: CEO, CSR AND FIRM VALUE 31

∆T obinqit = β0 + β1CSRit + γ∆Xit + δj + θt + εit (3)

where the dependent variable is the change in T obin0 Q during CEO transition and ∆CSRit is the corresponding change in CSR policy. ∆Xit adopts the set of standard control variables for firm characteristics from the previous specification with slight adjustment for the purpose of testing CSR value relevance. First, to mitigate simultaneity and reverse causality, we drop M/B and replace market leverage ratio (LR) with book leverage ratio since both of which are highly correlated with the dependent variable. Consistent with previous model specifications, we include industry fixed effects dummies, δj, and year fixed effects dummies, θt.

Obviously, one empirical challenge is the serious endogeneity caused by the simultaneity and reverse causality between firm performance and social provisions, which might significantly distort the causal inference. To mitigate the confounding effects of endogeneity on the true relationship between firm value and CSR, we utilize the methodology introduced by Ho, Imai, King, and Stuart(2007) as our primary inference strategy. As suggested in Ho et al.(2007), preprocessing data using nonparametric method such as matching could ameliorate the spurious statistical inference, especially for causal inference, of parametric regression models. As such, our empirical strategy will include two stages wherein we first preprocess the sample into matched pairs and then use the matched observations to estimate Model 3. Specifically, for each year and industry sector, we form two CHAPTER 1: CEO, CSR AND FIRM VALUE 32 groups of turnover firms with similar size of total assets. The treatment group consists of firms whose CEOs adjust CSR significantly while the control group consists of firms whose CEOs adjust CSR moderately. To retain more data for the later parametric method of regression models, we regard the first quantile and third quantile of the CSR absolute change, respectively, per year and industry as the cutoff numbers. For each year and industry, CEO turnovers with CSR absolute change above the third quantile belong to the treatment group whereas those lower than first quantile are included in control group. For the same reason, we use FF 49 to identify industry for matching. To match pairs with similar size of total assets, for each year and industry, we pick each individual firm from the treatment group and pair it with the other firm from the control group that minimizes the difference of total assets between the two. We continue this pairing process without replacement until one group is exhausted. One caveat is that the above algorithm only matches pairs with closest size in total assets while the resultant pairs might be of considerable discrepancy in assets value. In an alternative sample, we further require that the CSR change of a firm in treatment group is in the opposite direction of the CSR change of the paired firm in the control group. CHAPTER 1: CEO, CSR AND FIRM VALUE 33

5 Measures, Data and Summary Statistics

5.1 Data and Sample Construction

One methodological concern is how to construct a inclusive and reliable measure of CSR. To capture the overall performance of CSR in a broad scope, we measure the level of social-friendly activities conducted by the ﬁrm using data from KLD

Research & Analytics, Inc. SOCRATES database. KLD net scores, as the most general and popular measures for engagement and investment in CSR activities, assign ratings for up to thirteen dimensions. To focus on skills or managerial styles for stakeholder management, we select seven major dimensions to construct the proxy: product quality and safety, environment, governance, employee relations, diversity, community relations and human rights investment, each specifying a linkage between the ﬁrm and internal or external stakeholders (see, for example,

Tang et al. 2018). Following the practice in Lins et al.(2017), we do not consider the penalization of participating in the controversial industries since there is noth- ing the CEO can do, except to change the primary line of business and exit those industries. Moreover, we control industry fixed effects in all the model specifications. The scores are assigned by sector analyst under the proprietary framework based on a wide variety of data sources across company filings, government and nongovernment data, public media sources and direct communication or survey through connections with executives of the company. Such a scoring approach CHAPTER 1: CEO, CSR AND FIRM VALUE 34 of incorporating a variety of data sources ranging from mandatory to voluntary disclosures particularly benefit the empirical design and the context of our study.

With regard to our arguments about the timing and dynamics of CEO inﬂuences upon CSR activities throughout the tenure, one might argue that the signiﬁcant change may happen close to the end of the tenure, since the resultant impact of certain CSR initiatives may take longer to sink in. However, KLD scores are not pure measures for CSR performance (output) but for CSR investment (input), in the sense that if the CEO decides to make huge investment in some long-term

CSR project which takes years to implement or cut down socially-responsible expenditures over time, the information will for sure be released or leaked through media, government ﬁlings, company self-disclosure and private conversations with

KLD experts, changing KLD scores accordingly. Therefore, as a measure for the

CSR decisions, KLD scores will always adjust ahead of the full implementation and impact of CSR initiatives take place, thereby consistent with the purpose of our empirical analyses.

The sum of KLD net scores summarizes quite well the social activities of the

ﬁrm and is used by literature (see for example Agle, Mitchell, and Sonnenfeld

1999; Deckop, Merriman, and Gupta 2006) and matches our research purpose. It is noteworthy that in SOCRATES database, the KLD scores are not completely assigned to each dimension in each year. Therefore, in order to make KLD scores comparable throughout different years and increase the efficiency of parameter estimates in the regression analyses, we follow Deng, Kang, and Low(2013) and CHAPTER 1: CEO, CSR AND FIRM VALUE 35 scale the sum of KLD scores (strengths and conccerns) by the number of items scored within each dimension in the year, and then take the difference between total scaled strength scores and total scaled concern scores. The measures based on KLD scores to gauge the CSR activities of the firm are widely used or modified by the literature, such as Agle et al.(1999), Deckop et al.(2006), and Bebchuk,

Cohen, and Ferrell(2009).

We collect CEO age, gender, shareholder ownership and construct the tenure of the CEO using annual observations between 1995 and 2013 from Execucomp.

By identifying the number of days between the date that the individual joined the company and the date that she was appointed to CEO, we are able to categorize

CEOs into two groups, outsider and insider. In our definition, the CEO is recognized as an outsider if the individual was appointment to CEO in less than two years since she entered the company. In order to maximize the power of empirical tests, we improve the completeness of the sample by manually filling the gaps in the dataset, which are caused by the missing dates when the CEO joined the company. We search through 10-K, definitive proxy statement, and other fillings available on the website of U.S. Security and Exchange Commission, looking for the relevant information mentioned in the biography of the CEO. Moreover, we augment to the information by using online resources such as Bloomberg Exec- utive Profile & Biography and Wikipedia to develop a full picture of the CEO’s career path in the case that the biography on the fillings is somewhat ambiguous.7

7 It is worth noting that in some observations the individual became CEO before the recorded CHAPTER 1: CEO, CSR AND FIRM VALUE 36

We use Compustat data to obtain the level of total assets (AT ), market leverage ratio, market to book ratio, research and development expenditure (R&D), tangibility (PP &E), return on assets (ROA) and selling, general and administrative expenditure (SG&A) to depict the firm characteristics in the aspects of firm scale and complexity, growth opportunity, solvency, profitability and governance.

We also obtain data of other ﬁrm characteristics, such as advertisement expenses

(ADV ), and level of excess accruals (exAccr). We also include a proxy for the marginal tax rate (MTR), obtained from John Graham’s website. In order to remove the firm’s size effect on variables such as R&D and SG&A, we further scale those variables by total assets. Analogously, we collect data of CEO’s age, gender, board positions and ownership from Execucomp, along with the hand-collected insider-outsider attribute, to identify CEO characteristics. We also use Execu- comp to identify a change in the identity of the CEO. The detailed description, calculation and data sources are summarized in Table 1. After combining the data obtain from the above sources, we have a final sample of 14,952 observations over the period between 1995 and 2013.

In the paper, we concentrate on the effect of CEO turnover and CEO tenure upon the variability of CSR. Therefore, for the time series of annual observations of each firm, we first take the difference between the values of observed variables date that he joined the company. After exploring the CEO profile, we found that in one scenario the problem is caused by the fact that the individual worked as CEO in the predecessor of the current firm. We hand checked all the abnormal observations and made adjustments accordingly based on the information available from aforementioned sources. For instance, in the above case we change the date that the CEO joined the company to the date that the individual joined the predecessor of the current company. CHAPTER 1: CEO, CSR AND FIRM VALUE 37 of the current fiscal year and those of the previous fiscal year, as shown below.

It should be noted that the annual observations for a certain company in our data collected from KLD SOCRATES database over the period between 1995 and 2013 are not continuous time series since it is possible that for some ﬁrm- year observations there are missing entries for one or more variables. This leads to additional attrition of data for the regressions that use the change of the dependent and independent variables. As a result, our new sample used in the regressions of change is only 10,964 observations, over the period between 1995 and 2013.

Moreover, since it’s hard to determine whether the level of CSR is due to the successor or the predecessor if a new CEO is hired in the middle of the fiscal year, we define the change for the first full year of CEO tenure as the change between the fiscal year end of the last full year of the departing CEO and the first fiscal year end of the new entering CEO, identifying the first full year operation of the successor surrounding turnover.

Table 2 provides the number of CEO turnovers in each ﬁscal year for our sample as well as the turnover rate. Accounting for the missing values in control variables, this table reports the turnover statistics based upon the merged sample used in the multivariate regression analyses. This reduces our sample from 14,952 to 12,991.

Our sample contains 1,013 CEO turnover events, representing a turnover rate of 7.8%. This turnover rate is lower than those documented in CEO transition literature since the KLD data availability varies every year. The time series of the data is not continuous, resulting in a signiﬁcant portion of CEOs whose transition CHAPTER 1: CEO, CSR AND FIRM VALUE 38 periods are not captured by the sample. The details are summarized in Table 2.

5.2 Summary Statistics

We report the summary statistics for diﬀerent samples in Table 3. Panel A and panel B report the descriptive statistics for the levels and changes of variables of ﬁrm characteristics, respectively. Panel C reports the descriptive statistics for the variables of CEO characteristics. Panel D provides the summary statistics for both the level and the change of CSR in absolute value. The statistics calculated for level variables use the original sample of 14,947 observations; the statistics calculated for change variables and CEO characteristics use the sample of 13,716 observations (except for ∆MTR where we have only 11,521 observations), consisting of both usual annual change and change of the period surrounding turnover event. Note that the sample distributions of market to book ratio (M/B) and the change of advertising expense (∆ADV ) are extremely skewed and are characterized by outliers. As a result, we winsorize the variables at the level of 1 percent and 99 percent.

As shown in panel D, the distribution of the KLD measure of CSR is strongly centralized with high kurtosis and average around zero. It is noteworthy that the metric of CSR absolute change follows asymmetric distribution. Therefore, to gain more eﬃciency in regression analysis, we may use nonlinear estimation techniques. The age of CEOs (Age) for this sample ranges from 32 to 96, with CHAPTER 1: CEO, CSR AND FIRM VALUE 39 average age around 56 years old, and over the half of the individuals in the sample are in their 50s. The majority of CEOs in the sample holds the chair position of the board but one can infer that approximately a third of the CEOs in our sample sit on the board while another individual serves as chair of the board of directors.

With reference to other control variables of ﬁrm and CEO characteristics, we have roughly equivalent results of basic statistics such as mean, median and standard deviation to those in studies in the corporate ﬁnance literature (see, for example,

Bae, Kang, and Wang 2011; Brockman et al. 2016).

6 Results and Analyses

6.1 Tests of Hypothesis 1 – Is CSR CEO Centric?

Table 4 summarizes the results of regression analysis for testing Hypothesis

1. The table contains four columns. Column (1) does not include CEO fixed effects. Column (2) reports the regression assuming only CEO fixed effects as control variables. Columns (3) and (4) include both CEO fixed effects and industry

fixed effects. Columns (3) identify industries based upon the four-digit SIC code classification, the finest industry classification method and column (4) identify industries based upon the Fama and French 49 industry classification. Because of missing data for some of our control variable, our sample reduces to 12,992. CHAPTER 1: CEO, CSR AND FIRM VALUE 40

In column (1) of Table 4, we report the parameter estimates of regression model only with variables of firm characteristics that have been used in past studies. Consistent with the literature, the coefficients of the majority of the variables are statistically significant (see, for example, Davidson, Dey, and Smith

2019; Kang 2017). Column (2) of Table 4 reports the results with only the CEO

fixed effects and the joint test of the group of fixed effects dummies illustrates that the total power of the CEO fixed effects are statistically significant at the level of 0.01%. Interestingly, as shown in columns (3) – (4) of Table 4, which reports the estimation results of the regression containing the CEO fixed effects dummies and the firm characteristics, the coefficients of firm characteristics largely become statistically insignificant, except for leverage ratio (LR), MTR and scaled excess accruals (exACCR). It is noteworthy that albeit calculated by firm level variables, the scaled excess accrual is a proxy for the managerial willingness of engaging in earnings management, gauging the CEO’s borders of ethics. Thus, the statistical significance of scaled excess accruals should be credited to the explanatory power of CEO characteristics. Moreover, the joint test of CEO fixed effects (β = 0) have consistently strong statistics across all the regressions, whereas the statistical significance of the joint test of firm variables (γ = 0) drops dramatically after including the CEO fixed effects into the model specification, indicating that in line with our Hypothesis 1, CEO fixed effects explain the most part of the CSR variation and hold a valid position in the model specification.8 More importantly,

8 The model’s other speciﬁcations corresponding to columns (1) to (3) in Table 3 using alternative industry classiﬁcation approaches generate similar empirical results and are thereby not CHAPTER 1: CEO, CSR AND FIRM VALUE 41

the adjusted R-squared (Adj. R2) becomes twice as large after including the CEO

fixed effects dummies. The dramatic increment of Adj. R2 evidently demonstrates the CEO-centric effect upon CSR.

It could be argued that the CEO fixed effects dummies are just a correlated proxy for unobservable firm fixed effects. To mitigate this concern, we re-estimate the regressions of Table 4 by including the firm-CEO joint fixed effects dummies

(i.e., the product of the ﬁrm and CEO dummies) in the model speciﬁcation. We obtain analogous results and therefore we do not formally report the results in a table.

It should be noted that although in face of multi-layer high dimensional fixed effects, the usual practice is to construct linked fixed effects dummies, such a method still could not fully disentangle, in this case, CEO effects from firm fixed effects. The joint statistical significance is the aggregate effect of the unobservable characteristics of the firm and the CEO. To our knowledge, the only methodology of trying to disentangle CEO fixed effects from firm fixed effects is introduced by

Bertrand and Schoar(2003), in which their sample includes CEOs who worked in at least two different firms. As addressed in the hypotheses section, such setting is not applicable to our CSR data. Therefore, to identify and illustrate the power of pure CEO effect, we further our analysis in a first difference (FD) setting, in which the variation of CSR is regressed against the CEO turnover, a proxy for the reported in the table for conciseness purpose. This is also the case for all tables. CHAPTER 1: CEO, CSR AND FIRM VALUE 42

“change of CEO” eﬀect.

6.2 Tests of Hypothesis 2 – Disentangling CEO Fixed Ef-

fects From Firm Fixed Eﬀects

Table 5 presents the results of the regression model where the dependent variable is the absolute value of CSR change because our primary focus is on the magnitude of change in CSR as opposed to its direction under the new leadership. Due to missing data for our change variables, the sample is further reduced to 10,964 in the regressions. We include the absolute values of the changes of standard control variables for the ﬁrm characteristics. We now include CEO characteristics as control variables such as if the CEO is a director (Director), if the CEO is the chair of the board (Chair), age of the CEO (Age), gender of the CEO (Gender), and change in CEO share ownership (CEOshare). In addition, we add the variable

T urnover in order to directly test Hypothesis 2. Column (1) and (2) report results of OLS estimator while column (3) and (4) report results of heteroscedastic consistent standard error (HCSE).

Consistent with the empirical prediction of Hypothesis 2, T urnover has signiﬁcantly positive impact on the absolute variation of CSR. For example, the regression model presented in column (1), the coeﬃcient of T urnover is positive

(0.132) and significant at the level of 0.1%, meaning that if the CEO is classified as a new successor, the absolute change in the KLD scores during the transition CHAPTER 1: CEO, CSR AND FIRM VALUE 43 period is 0.132. The effect of T urnover is also economically significant. Note that the average absolute change of the CSR metric is 0.264. The coefficient of

T urnover (0.132) in column (1) represents a change of 50% from the average level of the sample. Note that for regressions from column (1) to column (4), the coefficients of Chair, a dummy variable equal to 1 if the CEO is the chair of the board, is also positive and significant at the same level of 0.1%. However, the effect of chair is quite less significant compared to that of T urnover in terms of economic significance.

Columns (1) and (2) include industry fixed effects for our two different industry classifications. Across different specifications, the coefficients on T urnover remains roughly unchanged both in quantity and statistical significance. In column

(3) and (4), we further enhance the statistical inference by using heteroscedasticity consistent estimator. Since ∆CSR is always greater than zero and follows | | a trimmed, asymmetric distribution, we conjecture that ∆CSR follows gamma | | distribution and perform a nonlinear maximum likelihood regression in column

(5) for Fama and French industry classification.9 As shown in Table 5, the parameter estimate of T urnover is even more significant than that in column (3), indicating that our conjecture is correct and the efficiency of the regression is thereby increased. Hence, based on the information reported in Table 5, we can reach a preliminary conclusion that T urnover, as a proxy for CEO effect during

9 Recall that there are insufficient number of firms in each four digit SIC classification to perform the maximum likelihood estimation. CHAPTER 1: CEO, CSR AND FIRM VALUE 44 transition period, plays an important role in driving the fluctuation of CSR. The significant change in CSR in terms of absolute magnitude typically incurs amid

CEO transition.

It is also possible that using absolute value of change in CSR is masking the true eﬀect of CEO turnover on CSR since we are assuming that turnover has exactly the same eﬀect on both positive and negative changes in CSR. Hence, we reproduce Table 5 by using the change in values for both the dependent and control variables. The results, as reported in Table 6, and are strictly analogous to those reported in Table 5.

6.3 Tests of Hypothesis 3 – Is CSR Policy Change Perma-

nent?

To test whether or not the new CEO aggressively reverses the CSR decision over subsequent years in her tenure, we augment the turnover variable with our three tenure dummy variables. The first dummy tenure variable is T urnover which equals to one for the first two years of the new CEO. T enure2 6 equals one if the − tenure is greater than two years but less or equal to 6 years. We chose 6 years as the breakoff because the sample median tenure is 6 years. Our third tenure variable is T enure6+ which equals one if the tenure of the CEO is greater than 6 years. For the firm to be included in the sample, the firm has to have a turnover in the sample period and CSR data are available during the transition time. This CHAPTER 1: CEO, CSR AND FIRM VALUE 45 reduces our sample to 3,520 firm-year observations. Moreover, because we wanted to observe if the new CEO reverses the initial CSR policy change, we dichotomize the sample into two groups. The first group contains firms whose CEOs make a positive change in CSR during the transition period (columns (1) and (2)) and the second group of CEOs make a negative change in CSR during the transition period

(columns (3) and (4)). We report only the HCSE estimates in Table 7. Note that only the turnover variable is significantly positive for the positive change CSR group and significantly negative for the negative change CSR group. However, the other tenure variables are insignificant, implying that there is no subsequent changes in policy and the initial change that the new CEO makes is permanent.

This can also be seen in Figure 2. This ﬁgure depicts the change in CSR that occurs during and after the CEO transition. Panel A depicts the change if the new CEO initiates a positive change while Panel B depicts the change if the new

CEO initiates a negative change.

6.4 Tests of Hypothesis 4 – Test of Attunement Theory

Table 8 reports the parameter estimates of the regression model where we include a dummy variable outsider, which equals to one during the transition period if the manager was appointed as the CEO of the ﬁrm within two years joining the company. To test Hypothesis 4, we use a subset of the original dataset, in which only CEOs with turnover observations are included. Analogous to the results CHAPTER 1: CEO, CSR AND FIRM VALUE 46 in the past sections, ∆CSR is more aﬀected by CEO characteristics instead of

firm characteristics. In particular, the majority of the firm characteristics are insignificant or only significant marginally with limited statistical power, but the

CEO characteristics of outsider and board positions are significant. Among CEO characteristics, the coefficient of the outsider attribute is significantly negative consistent with the attunement theory and the limited attention hypothesis. The negative significance of outsider also indicates that when hiring a new CEO, the company is on average not actively in search of individuals with certain social skills to carry out reforms in social policy. Therefore, our methodology of CEO turnover is not overestimating the effect of CEO on social policy.

6.5 Tests of Hypothesis 5 — Does CSR Aﬀect Firm Value?

Our results indicate that CSR policy is CEO centric. The remaining question is whether the resulting change in CSR is improving ﬁrm value. One might interpret that since CSR investment is CEO centric, CSR acts like a perquisite for the CEO and does not add value to the ﬁrm. On the other hand, CEOs who decrease the level of investment in CSR are reducing wasteful expenditures and improving

firm value. However, one might argue that the CEO’s CSR investment strategy reflects her skill set to manage the company. Consequently, if the skill set of the new CEO is not able to employ CSR assets profitably as did the former CEO, this new CEO would invest less in CSR activities than the previous CEO. It also CHAPTER 1: CEO, CSR AND FIRM VALUE 47 might be true that the current CEO’s skill set does not match well with CSR investments. However, the new CEO has that skill set, in which case you might

find an increase in CSR investment. All the above possibilities pertaining to the variation of CSR investment and its value-relevance are centered on the skill set of CEOs. If so, then we should not expect the changes in CSR policies around the transition of the new CEO negatively impacting firm value. The advantage of using the CEO transition period to understand how CSR impacts firm value is that our paper demonstrates that the CSR policy changes occur during the first two years of the new CEO’s tenure, allowing us to compare the impact of sudden changes in CSR policy upon firm value.

As suggested in Ho et al.(2007), preprocessing data using nonparametric method such as matching could ward off the spurious statistical inference, especially for causal inference, of parametric regression models. We adopt the matching method as our inference strategy given the strong simultaneity and reverse causality between firm value and social provisions. Following the procedure specified in the methodology section, we pick each individual firm from treatment group and pair it with the other firm from control group that minimizes the difference of total assets between the two. We delete from this sample pairs where we do not have entries for all of the control variables. This results in 159 pairs.

Table 9 provides summary regression results when we regress change in T obin0s Q from the last ﬁscal year of the old CEO to the ﬁrst full year of the new CEO. CHAPTER 1: CEO, CSR AND FIRM VALUE 48

Column (1) includes all matched firm pairs where the treatment group had a significant change in CSR and the control group does not (but the change could be in the same direction as the treatment group. Column (2) includes all matched firm pairs where the control group has a change in CSR in the opposite direction of the treatment group. Note that there is no significant relationship between change in T obin0s Q and change in CSR. Accordingly, one can interpret the results indicating that changes in CSR do not materially impact the firm’s bottom line and therefore does not impact upon value.

It is possible, however, that one might ﬁnd a stronger eﬀect if we added a variable that captures whether or not the incoming CEO who is greatly changing the

CSR policy is moving positively or negatively away from the industry norm CSR policy, deﬁned as the median CSR metric of the industry (FF 49). Accordingly, we introduce a dummy variable NORM to capture this policy change. Our results are summarized in Table 10. The table has four columns. In columns (1) and

(2), we set NORM equal to one if the new CEO’s CSR policy of treatment firms is moving towards the industry norm. In columns (3) and (4), NORM is equal to one if the new CEO’s CSR policy is moving upwards away from the industry norm. Notice in columns (1) and (2), the coefficients of the dummy variable are negative implying that reducing CSR decreases firm value. On the other hand, in columns (3) and (4), the dummy variable coefficient is positive implying changing the CSR policy upwards away from the industry norm increases firm value. Thus, we have some evidence that there is a positive relation between CSR and firm CHAPTER 1: CEO, CSR AND FIRM VALUE 49 value.

Lins et al.(2017) show that CSR activities engender trust among corporate stakeholders. It is possible that one only would observe the impact of CSR upon

firm value is during times when trust becomes an important commodity. One would expect that trust is more important when the economy is in a deep recession as what happened between 2007 and 2009. Hence, columns (3) and (4) only include firms that undergone a turnover event during this period. Our sample is further reduced to 37 pairs of firms when the control group does not change CSR significantly (but the change could be of the same direction as the treated sample), and to 23 pairs when the change in CSR in the treatment group is in the opposite direction of the control group. We calculate the stock return using the year end closing price for each stages. Notice that notwithstanding the limited sample size, in these columns there is a positive relationship between change in firm value and change in CSR. We find that the change of T obin0s Q is positively related to the change of CSR promulgated by the incoming CEO. The latter result of change in

T obin0s Q is consistent with the stock return ﬁndings of Lins et al.(2017) who ﬁnd that CSR activities engender trust among corporate stakeholders. Thus we can infer that on average, CSR activities do not serve as CEO perquisites but rather as value enhancing investments.

We also provide an alternative methodology for testing the importance of CSR on firm value. According to Lins, Volpin, and Wagner(2013), the 2008-2009 finan- CHAPTER 1: CEO, CSR AND FIRM VALUE 50 cial crisis starts from August 2008, shortly before Lehman Brothers bankruptcy, to March 2009, when S&P 500 hit its lowest point. We further classify the crisis period into two stages 2007-2008 and 2008-2009, to demonstrate the resilience of trustworthy firms, which invest sufficiently in social capital, in face of the shock of the financial crisis. A firm is identified as trustworthy firm if the firm ushers in a new CEO who boost socially responsible investments in 2003, right after the period in which a list of formerly reputable firms such as Enron, Xerox, Worldcom, and Nortel among many others are charged with corporate fraud, devastating the capital stock of trust. In contrast, Lins et al.(2017) define trustworthy firms based upon the relative level of CSR in 2007 among all firms regardless of whether there is a CEO turnover. We calculate the stock return using the year end close price for each stages. As shown in Table 11, the trustworthy firms perform much better

(based upon annual returns) than the other control group in the second stage of

financial crisis, demonstrating strong power of resilience. Specifically, we perform simple regression analysis in Table 11, in which we regress the annual stock return against the variable, T rusted, which is an indicator variable that takes the value of unity if the firm is classified as a trustworthy firm. To ameliorate the effect of influential observations in the small sample regression, we use robust estimators.

In particular, we intend not to include other simultaneous firm characteristics as controls since the dummy variable of trusted firm is classified 5 years ago and thus is not endogenous to omitted firm fundamentals. Keeping only one variable of interest could mitigate the confounding effects of multicollinearity. The results CHAPTER 1: CEO, CSR AND FIRM VALUE 51 are in alignment with the findings in Lins et al.(2017). Furthermore, we conduct univariate tests in Panel B of Table 11, which generates consistent results with the regressions.

In sum, our results support the notion of the importance of CEO in determining CSR policy. Our empirical results delineate the general pattern of CEO investment cycle in social capital throughout tenure. Moreover, our results are consistent with the literature, in that outsider CEOs are less likely to change CSR activities because the new CEO has not developed the necessary relationships with stakeholders other than shareholders (attunement theory) or because the board of directors would like the new CEO to first concentrate on policy changes that affect shareholders. Finally, we find that CSR can increase firm value especially when the trust of investors becomes a valuable part of capital stock during economic recession.

7 Robustness Check

Although the empirical results produced by the above statistical tests have provided material evidence for the recognition of CEO-centric influence upon Corpo- rate Social Responsibility (CSR) activities and with the predictions in corresponding hypotheses, the results may be affected if we do not distinguish between forced and voluntary turnover. The concern that forced CEO departure and succession CHAPTER 1: CEO, CSR AND FIRM VALUE 52 may materially distort the statistical results since in these cases you would expect greater policy changes in order to improve firm performance. In other words, perhaps the significant results we obtain are not due to CEO centric policies, but simply due to policy changes of CEOs involved in a forced turnover (see for example, Huson et al. 2004; Kim 1996). Thus, the strategic leadership and managerial effort brought by the high quality manager in a forced turnover is more likely to result than in voluntary turnover events in reshaping firm policies such as investment policy, financing policy and societal oriented policies. On the other hand, the amount of managerial effort and good faith put by managers who voluntarily and involuntarily depart are identical while the realized level of performance varies due to uncertain component of manager luck. In other words, the replacement of managers may be solely due to bad luck. Although, in essence, the board punishes bad luck instead of lack of ability, the successful replacement candidate is likely to have a managerial style that is different from incumbent CEO. Therefore, both of the two hypotheses concerned with forced managerial turnover indicate that forced CEO transition is more likely to incur significant change in firm policies, including social policies assessed by this study’s CSR variable. In order to conclu- sively investigate the key research question of the CEO-centric effect particularly upon CSR policy and the relevant hypotheses, we remove the proportion of forced turnover observations from all the transition data from the sample to examine the generality of our previous results to voluntary turnovers.

In order to distinguish between the events of voluntarily turnover and forced CHAPTER 1: CEO, CSR AND FIRM VALUE 53 turnover, we, in principle, follow the criteria proposed in Parrino(1997) to determine if the turnover was forced. According to Parrino(1997), the forced departure of CEO can be identified through a process of three procedures. Firstly, the forced departure is identified if the public business news release explicitly announces that the individual leaves office due to forced termination of contract, policy differences or any other reasons related to firm’s activities. Secondly, if the direct messages of termination are not disclosed publicly, we believe, with one exception as noted below, that the individuals, who are above age 60, leave office voluntarily due to normal retirement. The classification of individuals, who are under age 60, is subject to the following two criteria. Those turnovers are considered as forced if the public information either do not discloses reasons such as death, poor health or the acceptance of another position (inside the firm or elsewhere, including governmental agency), or do not release the message of retirement at least six months before the succession. Finally, the cases of forced departure classified in the second procedure are reinvestigated using extensive information surrounding the transition period and records in the individual’s biography. The turnover is classified as voluntary if it turns out that the individual takes a comparable position elsewhere or departs for the ex-ante undisclosed reasons unrelated firm to policy and performance, such as personal interests confirmed by the departing CEO biography or subsequent press release. The CEO’s age and the date of departure can be obtained from Execucomp, which also provides fragmentary information about reasons of departure for reference. The business news and reports from various CHAPTER 1: CEO, CSR AND FIRM VALUE 54 sources such as newspaper, journal articles, company financial reports, analyst reviews and etc., were searched using the database of Factiva. The extensive information such as CEO biography, achievements and company history is obtained via internet resources such as Bloomberg Executive Profile & Biography,

Wikipedia, and SEC ﬁlings.

We also amended Parrino’s criteria in the following ways. Specifically, those individuals aged 60 to 65 who relinquish the duties as CEO during a significant downturn of firm performance or under legal investigation, indicated clearly by the press release around the transition period, are considered as leaving office non-voluntarily. Similarly, individuals from 55 to 60 who hand off duties to retire, pursue other opportunities or interest are considered as voluntary turnover if the reasons are reflected by their biography or consistent with company’s tradition of early retirement demonstrated by the historical turnovers, such as companies in the high-tech industry. Due to the limited information resources during the transition period, we experienced difficulty in determining the voluntariness of the

CEO departure in some cases. To utilize a conservative strategy in the robustness check, we conduct the main analyses using the sample in which both of the forced turnover observations and those observations we could not classify are eliminated.

The size of the forced or not sure turnovers excluded from the sample is around

17% of the total turnovers, which is quite comparable to the statistics in Parrino

(1997). We ﬁnd that the results are strictly analogous to those reported Tables 6

- 8. We also re-estimate the regression equations of Tables 6 – 8 where we include CHAPTER 1: CEO, CSR AND FIRM VALUE 55 a dummy variable indicating if the turnover was forced. The dummy variable is not signiﬁcant and the results for our other variables are strictly analogous as the tabulated results. Hence, we do not formally report the results of the additional tests.

An implication of our study is that corporate social responsibility is CEO centric. This would mean that if a manager leaves one company for another, then the corporate social responsibility policy of the two ﬁrms should be similar. We found 41 such examples in our data. We then took the diﬀerence between the

KLD score for the last year the CEO was employed in the ﬁrst company and the

first year the CEO was employed in the second company. The average change is 0.028 which is not significantly different from zero, compared to the average absolute change of CSR around 0.264 in the whole sample as shown in Table 3.

Ideally we should be running a regression examining the KLD scores similar to our previous regressions which would require having at least 5-years continuous data in the second firm since the CEO took office but we only find 9 CEOs with sufficient data.

It should be noted that the argument that CEO is the “dominating” effect upon CSR is too disingenuous. It does not mean that the influence of CEO could completely overturn the interaction among organizational structure, external market conditions and firm fundamentals. Otherwise, one might interpret the results here to imply that the identical CEO of two different firms results in the CHAPTER 1: CEO, CSR AND FIRM VALUE 56 same level of CSR at different points in time. This would not be correct since over time, CEOs acquire skills and experiences and the two firms may face different organization structures and market conditions. In our sample it should be noted that there is a 5–10 years gap between CEO positions of the same individual.

One concern regarding our results is that the literature has shown that investments in CSR activities impact ﬁrm’s proﬁtability (Ferrell et al. 2016; Lee and

Faff 2009), the cost of capital (El Ghoul et al. 2011; Goss and Roberts 2011), and access to capital markets (Cheng, Ioannou, and Serafeim 2014). In addition, Fer- rell et al.(2016) find a negative relationship between agency conflicts within the

ﬁrm and CSR. Clearly, the board of directors is supposed to manage these con-

flicts. In fact, Jo and Harjoto(2012) find a causal relationship between corporate governance (board of director characteristics) and firm engagement in CSR activities. Consequently, perhaps our results are due to omitted variables that capture both managerial ability and characteristics of the board of directors. That is, it is possible that these two proxies may affect both CSR and profitability because the CEO’s ability with the guidance of the board of director to manage conflicts creates transparency that allows firms to access capital markets more easily and reduce costs.

Consequently, we include in our regressions proxies for the characteristics of board of directors (namely, number of directors and percentage of outside directors) and the proxy for CEO managerial ability. We collect the data for the total CHAPTER 1: CEO, CSR AND FIRM VALUE 57 number of directors and the number of independent directors from ISS. We obtain the ability proxy using the methodology of Demerjian, Lev, and McVay(2012). 10

We report in Table 12 the results of incorporating these additional variables when regressing the level of CSR against our control variables. Table 12 is the analogue of Table 4 and the structure of the two tables is identical. Notice, however, that our sample compared to that reported in Table 4 is almost cut in half. Notice that as found by Jo and Harjoto(2012), the lagged corporate governance variables are significant. In particular, board size is positively correlated with the level of CSR investment when we do not include CEO fixed effects. The percentage of independent board of directors becomes positively significant when we include CEO fixed effects. Interestingly, ability is not significant. More importantly, the CEO fixed effects are significant and greatly increase the explanatory power of our model.

Table 13 reports the regression results when we regress the change of CSR against our control variables, the dummy variables for T urnover, as well as the board of director characteristics and ability proxy. Column (1) and (2) report results of OLS estimator while column (3) and (4) report results of heteroscedastic consistent standard error (HCSE). Again, the reported results are strictly analogous to that reported in our previous tables.

Our ﬁnal robustness test checks remove the possibility that endogeneity can aﬀect our results. It is possible that CEO turnover may be the result that the

10 We gratefully thank to Dr. Peter Demerjian for providing the managerial ability data through his website. http://faculty.washington.edu/pdemerj/data.html CHAPTER 1: CEO, CSR AND FIRM VALUE 58 board of directors wishes to change the CSR policy of the firm. But if this were the case, then our results would have been affected when we remove forced turnovers from our sample. It is possible, however, that the board of directors do not want to remove the CEO for this reason and wants to wait until the CEO to voluntarily retire to affect the CSR policy. If that were the case, we should expect the biggest change in CSR policy to occur when an outside CEO is appointed. But our results support the attunement theory which predicts the opposite. Never- theless, endogeneity may still exist if there are omitted variables such as lagged

firm characteristics that affect both CSR policy and turnover. Accordingly, we follow the procedures of Petrenko et al.(2016) and Tang et al.(2018) to remove this potential bias. Particularly, in the first stage, we regress the turnover indicator against a set of antecedents such as lagged ROA, market to book, R&D and

CEO characteristics such as CEO age, ownership and whether the CEO serves as a director or chair. We then include in the second stage the predicted value of T urnover variable. We replicate the regressions of Table 6 using this control variable and the results are reported in Table 14. Note that the inclusion of the predicted T urnover variable does not aﬀect our results.11

11 Another way to solve the potential endogeneity issues once for all is to replace the shock of CEO turnover, which is probably an outcome of miscellaneous reasons and thus considered as bad control, with clean cut experimental shocks. To identify a natural interference upon the business operation and decision making process of the ﬁrm and management, we could utilize sudden CEO deaths. We collect data of news report via Factiva to identify sudden death event based on the news contend and the criterion in the extant literature (see for example, Worrell, Davidson and Garrison, 1986). Unfortunately, we could only identify 17 such CEO events and clearly the power of any statistical test would be minimal. CHAPTER 1: CEO, CSR AND FIRM VALUE 59

8 Concluding Remarks and Future Research

In the study we demonstrate there is a heavy CEO-centric influence upon corporate social responsibility (CSR) activities. We show that the CEO fixed effects become the empirically dominating factor in determining the level of CSR, indicating that a set of CEO-related factors, captured by fixed effects dummies, are omitted in the specification of model for CSR that includes only firm characteristics. We further examine whether the largest changes in CSR center around CEO turnover events using a first difference model. If the CEO is the main driver of CSR policy, we would expect that the biggest changes in CSR will occur when the new CEO takes over. Our empirical results are consistent with that prediction for new CEOs. We also find that changes in CSR for CEO turnover gradually fade away as the tenure of the new CEO grows longer. However, we find changes in CSR policy are reduced around the CEO turnover event for those CEOs who are appointed from outside the company. This result is consistent with the implications of attunement theory which predicts that outsider successor tends not to drastically change CSR at the beginning of his or her tenure.

Although empirical results are consistent with the CEO-centric inﬂuence upon

CSR activities and for the predictions in relevant hypotheses, the results may be affected if we do not distinguish between forced and voluntary turnover. We therefore replicate our results after eliminating from the sample all the forced CEO CHAPTER 1: CEO, CSR AND FIRM VALUE 60 turnovers. Our final robustness test examines the possibility that endogeneity might affect our results. It is possible that CEO turnover may be the result that the board of directors wishes to change the CSR policy of the firm. But if this were the case, then our results would have been affected when we removed forced turnovers from our sample. It is possible, however, that the board of directors do not want to remove the CEO for this reason and wants to wait until the

CEO to voluntarily retire to affect firm’s CSR policy. If that were the case, we should expect the biggest change in CSR policy to occur when an outsider CEO is appointed. But our results support the attunement theory which predicts the opposite. Nevertheless, endogeneity may still exist if there are omitted variables that affect CSR policy and turnover. Accordingly, we follow the procedures of

Petrenko et al.(2016) and Tang et al.(2018) to ameliorate this potential bias.

Our results remain the same.

The remaining question is whether CSR is improving ﬁrm value. We believe that the idea and the results in this study could contribute to the reconciliation of the literature on the value relevance of CSR. One might interpret that since

CSR investment is CEO centric, CSR acts like a perquisite for the CEO and does not add value to the firm. However, one might argue that the CEO’s investment strategy reflects her skill set to manage the company. Consequently, if the skill set of the new CEO is not able to employ CSR assets profitably, for example, you might expect that CEO who is still acting in the interest of the shareholders, would invest less than the previous CEO. It also might be true that the current CEO’s CHAPTER 1: CEO, CSR AND FIRM VALUE 61 skill set does not match well with CSR investments. However, the new CEO has that skill set, in which case you might find an increase in CSR investment. All the above possibilities pertaining to the variation of CSR investment and its value- relevance are centered on the skill set of CEOs. As such, if the view of CEO-centric effect upon CSR advocated in this study is established, the conflicting results of the linkage between CSR and firm performance in the extant literature is explainable.

Identifying the skill set is an interesting question for further research. CHAPTER 1: CEO, CSR AND FIRM VALUE 62

References

Agle, B. R., Mitchell, R. K., & Sonnenfeld, J. A. (1999). Who matters to CEOs?

An investigation of stakeholder attributes and salience, corpate performance,

and CEO values. Academy of management journal, 42 (5), 507–525.

Bae, K.-H., Kang, J.-K., & Wang, J. (2011). Employee treatment and ﬁrm

leverage: A test of the stakeholder theory of capital structure. Journal of

ﬁnancial economics, 100 (1), 130–153.

Barkema, H. G., & Schijven, M. (2008). Toward unlocking the full potential of

acquisitions: The role of organizational restructuring. Academy of manage-

ment journal, 51 (4), 696–722.

Bebchuk, L., Cohen, A., & Ferrell, A. (2009). What matters in corporate gover-

nance? The Review of Financial Studies, 783 , 823.

Ben-David, I., Graham, J. R., & Harvey, C. R. (2013). Managerial miscalibration.

The Quarterly Journal of Economics, 128 (4), 1547–1584.

Bernard, Y., Godard, L., & Zouaoui, M. (2018). The eﬀect of CEOs’ turnover on

the corporate sustainability performance of French ﬁrms. Journal of business

Ethics, 150 (4), 1049–1069.

Bertrand, M., & Mullainathan, S. (2001). Are CEOs rewarded for luck? The

ones without principals are. The Quarterly Journal of Economics, 116 (3),

901–932.

Bertrand, M., & Schoar, A. (2003). Managing with style: The eﬀect of managers CHAPTER 1: CEO, CSR AND FIRM VALUE 63

on ﬁrm policies. The Quarterly journal of economics, 118 (4), 1169–1208.

Bettman, J. R., Johnson, E. J., & Payne, J. W. (1986). Cognitive eﬀort and

decision making strategies: A componential analysis of choice. (Tech. Rep.).

DUKE UNIV DURHAM NC.

Bradley, M., Jarrell, G. A., & Kim, E. H. (1984). On the existence of an optimal

capital structure: Theory and evidence. The Journal of Finance, 39 (3),

857–878.

Brick, I., Venezia, I., & Palmon, O. (2018). On the relation between executive

compensation and corporate social responsibility. Recent Trends in Behav-

ioral Finance.

Brockman, P., Lee, H. S. G., & Salas, J. M. (2016). Determinants of CEO com-

pensation: Generalist–specialist versus insider–outsider attributes. Journal

of Corporate Finance, 39 , 53–77.

Brown, W. O., Helland, E., & Smith, J. K. (2006). Corporate philanthropic

practices. Journal of corporate Finance, 12 (5), 855–877.

Buchanan, B., Cao, C. X., & Chen, C. (2018). Corporate social responsibility, ﬁrm

value, and inﬂuential institutional ownership. Journal of Corporate Finance,

52 , 73–95.

Carroll, A. B. (1979). A three-dimensional conceptual model of corporate perfor-

mance. Academy of Management Review, 4 (4), 497–505.

Carroll, A. B. (1999). Corporate social responsibility: Evolution of a deﬁnitional

construct. Business & Society, 38 (3), 268–295. CHAPTER 1: CEO, CSR AND FIRM VALUE 64

Chaganti, R., & Sambharya, R. (1987). Strategic orientation and characteristics

of upper management. Strategic Management Journal, 8 (4), 393–401.

Cheng, B., Ioannou, I., & Serafeim, G. (2014). Corporate social responsibility and

access to ﬁnance. Strategic Management Journal, 35 (1), 1–23.

Cohen, D., & Prusak, L. (2001). In good company. Boston: Harvard Business

School Press.

Cordeiro, J. J., & Tewari, M. (2015). Firm characteristics, industry context, and

investor reactions to environmental CSR: A stakeholder theory approach.

Journal of Business Ethics, 130 (4), 833–849.

Davidson, R. H., Dey, A., & Smith, A. J. (2019). CEO materialism and corporate

social responsibility. The Accounting Review, 94 (1), 101–126.

Deckop, J. R., Merriman, K. K., & Gupta, S. (2006). The eﬀects of CEO pay

structure on corporate social performance. Journal of Management, 32 (3),

329–342.

Demerjian, P., Lev, B., & McVay, S. (2012). Quantifying managerial ability: A

new measure and validity tests. Management Science, 58 (7), 1229–1248.

Deng, X., Kang, J.-k., & Low, B. S. (2013). Corporate social responsibility

and stakeholder value maximization: Evidence from mergers. Journal of

Financial Economics, 110 (1), 87–109.

Dhaliwal, D. S., Li, O. Z., Tsang, A., & Yang, Y. G. (2011). Voluntary nonﬁnancial

disclosure and the cost of equity capital: The initiation of corporate social

responsibility reporting. The Accounting Review, 86 (1), 59–100. CHAPTER 1: CEO, CSR AND FIRM VALUE 65

Donaldson, T., & Preston, L. E. (1995). The stakeholder theory of the corporation:

Concepts, evidence, and implications. Academy of Management Review,

20 (1), 65–91.

Edmans, A. (2011). Does the stock market fully value intangibles? Employee

satisfaction and equity prices. Journal of Financial Economics, 101 (3),

621–640.

Edmans, A. (2012). The link between job satisfaction and ﬁrm value, with im-

plications for corporate social responsibility. Academy of Management Per-

spectives, 26 (4), 1–19.

El Ghoul, S., Guedhami, O., Kwok, C. C., & Mishra, D. R. (2011). Does corpo-

rate social responsibility aﬀect the cost of capital? Journal of Banking &

Finance, 35 (9), 2388–2406.

Fauver, L., McDonald, M. B., & Taboada, A. G. (2018). Does it pay to treat

employees well? International evidence on the value of employee-friendly

culture. Journal of Corporate Finance, 50 , 84–108.

Ferrell, A., Liang, H., & Renneboog, L. (2016). Socially responsible ﬁrms. Journal

of Financial Economics, 122 (3), 585–606.

Freeman, R. E. (1984). Strategic management: A stakeholder approach. Boston,

MA: Pitman Publishing Inc.

Friedman, M. (1970). The social responsibility of business is to increase its proﬁts.

In New York Times Magazine.

Goss, A., & Roberts, G. S. (2011). The impact of corporate social responsibility CHAPTER 1: CEO, CSR AND FIRM VALUE 66

on the cost of bank loans. Journal of Banking & Finance, 35 (7), 1794–1810.

Grieser, W. D., & Hadlock, C. J. (2019). Panel-data estimation in ﬁnance:

Testable assumptions and parameter (in) consistency. Journal of Financial

and Quantitative Analysis, 54 (1), 1–29.

Gupta, A. K., & Govindarajan, V. (1984). Business unit strategy, managerial

characteristics, and business unit eﬀectiveness at strategy implementation.

Academy of Management Journal, 27 (1), 25–41.

Haleblian, J., & Finkelstein, S. (1993). Top management team size, ceo dominance,

and ﬁrm performance: The moderating roles of environmental turbulence

and discretion. Academy of Management Journal, 36 (4), 844–863.

Hambrick, D. C. (2007). Upper Echelons Theory: An Update. Academy of Man-

agement Review.

Hambrick, D. C., & Mason, P. A. (1984). Upper Echelons: The Organization as

a Reﬂection of Its Top Managers. Academy of Management Review, 9 (2),

193–206.

Hayward, M. L., Rindova, V. P., & Pollock, T. G. (2004). Believing one’s own

press: The causes and consequences of CEO celebrity. Strategic Management

Journal, 25 (7), 637–653.

Ho, D. E., Imai, K., King, G., & Stuart, E. A. (2007). Matching as nonparametric

preprocessing for reducing model dependence in parametric causal inference.

Political Analysis, 15 (3), 199–236.

Holmstrom, B. (1982). Moral hazard in teams. The Bell Journal of Economics, CHAPTER 1: CEO, CSR AND FIRM VALUE 67

324–340.

Hosmer, L. T. (1995). Trust: The connecting link between organizational theory

and philosophical ethics. Academy of Management Review, 20 (2), 379–403.

Huson, M. R., Malatesta, P. H., & Parrino, R. (2004). Managerial succession and

ﬁrm performance. Journal of Financial Economics, 74 (2), 237–275.

Jensen, M. C., & Meckling, W. (1976). Theory of the ﬁrm: managerial behavior,

agency costs and ownership structure. Journal of Financial Economics, 3 .

Jo, H., & Harjoto, M. A. (2012). The causal eﬀect of corporate governance on

corporate social responsibility. Journal of Business Ethics, 106 (1), 53–72.

Jones, T. M. (1995). Instrumental stakeholder theory: A synthesis of ethics and

economics. Academy of Management Review, 20 (2), 404–437.

Kang, J. (2017). Unobservable CEO characteristics and CEO compensation as

correlated determinants of CSP. Business & Society, 56 (3), 419–453.

Kim, Y. (1996). Long-term ﬁrm performance and chief executive turnover: An

empirical study of the dynamics. The Journal of Law, Economics, and

Organization, 12 (2), 480–496.

Lee, D. D., & Faﬀ, R. W. (2009). Corporate sustainability performance and

idiosyncratic risk: A global perspective. Financial Review, 44 (2), 213–237.

Lins, K. V., Servaes, H., & Tamayo, A. (2017). Social capital, trust, and ﬁrm

performance: The value of corporate social responsibility during the ﬁnancial

crisis. The Journal of Finance, 72 (4), 1785–1824.

Lins, K. V., Servaes, H., & Tamayo, A. (2019). Social capital, trust, and corporate CHAPTER 1: CEO, CSR AND FIRM VALUE 68

performance: how csr helped companies during the ﬁnancial crisis (and why

it can keep helping them). Journal of Applied Corporate Finance, 31 (2),

59–71.

Lins, K. V., Volpin, P., & Wagner, H. F. (2013). Does family control matter?

International evidence from the 2008–2009 ﬁnancial crisis. The Review of

Financial Studies, 26 (10), 2583–2619.

Malmendier, U., & Tate, G. (2005). CEO overconﬁdence and corporate invest-

ment. The Journal of Finance, 60 (6), 2661–2700.

Malmendier, U., Tate, G., & Yan, J. (2011). Overconﬁdence and early-life expe-

riences: The eﬀect of managerial traits on corporate ﬁnancial policies. The

Journal of Finance, 66 (5), 1687–1733.

Manner, M. H. (2010). The impact of CEO characteristics on corporate social

performance. Journal of Business Ethics, 93 (1), 53–72.

Masulis, R. W., & Reza, S. W. (2015). Agency problems of corporate philanthropy.

The Review of Financial Studies, 28 (2), 592–636.

McCarthy, S., Oliver, B., & Song, S. (2017). Corporate social responsibility and

CEO conﬁdence. Journal of Banking & Finance, 75 , 280–291.

McWilliams, A., & Siegel, D. (2001). Corporate social responsibility: A theory of

the ﬁrm perspective. Academy of Management Review, 26 (1), 117–127.

Mintzberg, H. (1978). Patterns in strategy formation. Management Science,

24 (9), 934–948.

Murphy, K. J., & Zimmerman, J. L. (1993). Financial performance surrounding CHAPTER 1: CEO, CSR AND FIRM VALUE 69

CEO turnover. Journal of Accounting and Economics, 16 (1-3), 273–315.

Orlitzky, M., Schmidt, F. L., & Rynes, S. L. (2003). Corporate social and ﬁnancial

performance: A meta-analysis. Organization studies, 24 (3), 403–441.

Orlitzky, M., & Swanson, D. L. (2002). Value attunement: Toward a theory of

socially responsible executive decision-making. Australian Journal of Man-

agement, 27 (1 suppl), 119–128.

Pan, Y., Wang, T. Y., & Weisbach, M. S. (2016). CEO Investment Cycles. The

Review of Financial Studies, 29 (11), 2955–2999.

Parrino, R. (1997). CEO Turnover and Outside Succession a Cross-Sectional

Analysis. Journal of Financial Economics, 46 (2), 165–197.

Payne, J. W., Bettman, J. R., & Johnson, E. J. (1988). Adaptive strategy selection

in decision making. Journal of Experimental Psychology: Learning, Memory,

and Cognition, 14 (3), 534.

Petrenko, O. V., Aime, F., Ridge, J., & Hill, A. (2016). Corporate social responsi-

bility or CEO narcissism? CSR motivations and organizational performance.

Strategic Management Journal, 37 (2), 262–279.

Prior, D., Surroca, J., & Trib´o,J. A. (2008). Are socially responsible managers

really ethical? Exploring the relationship between earnings management

and corporate social responsibility. Corporate governance: An international

review, 16 (3), 160–177.

Smith Jr, C. W., & Watts, R. L. (1992). The investment opportunity set and cor-

porate ﬁnancing, dividend, and compensation policies. Journal of Financial CHAPTER 1: CEO, CSR AND FIRM VALUE 70

Economics, 32 (3), 263–292.

Stanwick, P. A., & Stanwick, S. D. (1998). The determinants of corporate social

performance: An empirical examination. American Business Review, 16 (1),

86.

Stinchcombe, A. L., McDill, M. S., & Walker, D. R. (1968). Demography of

Organizations. American Journal of Sociology, 74 (3), 221–229.

Swanson, D. L. (1999). Toward an integrative theory of business and society: A

research strategy for corporate social performance. Academy of Management

Review, 24 (3), 506–521.

Tang, Y., Mack, D. Z., & Chen, G. (2018). The diﬀerential eﬀects of CEO nar-

cissism and hubris on corporate social responsibility. Strategic Management

Journal, 39 (5), 1370–1387.

Tang, Y., Qian, C., Chen, G., & Shen, R. (2015). How CEO hubris aﬀects

corporate social (ir) responsibility. Strategic Management Journal, 36 (9),

1338–1357.

Thomas, A. S., Litschert, R. J., & Ramaswamy, K. (1991). The performance im-

pact of strategy-manager coalignment: An empirical examination. Strategic

Management Journal, 12 (7), 509–522.

Thomas, A. S., & Simerly, R. L. (1994). The Chief Executive Oﬃcer and Corporate

Social Performance: An Interdisciplinary Examination. Journal of Business

Ethics, 13 (12), 959–968.

Titman, S., & Wessels, R. (1988). The determinants of capital structure choice. CHAPTER 1: CEO, CSR AND FIRM VALUE 71

The Journal of Finance, 43 (1), 1–19.

Udayasankar, K. (2008). Corporate social responsibility and ﬁrm size. Journal of

Business Ethics, 83 (2), 167–175.

Weisbach, M. S. (1988). Outside directors and CEO turnover. Journal of Financial

Economics, 20 , 431–460.

Weisbach, M. S. (1995). CEO turnover and the ﬁrm’s investment decisions.

Journal of Financial Economics, 37 (2), 159–188.

Yip, E., Van Staden, C., & Cahan, S. (2011). Corporate social responsibility

reporting and earnings management: The role of political costs. Australasian

Accounting, Business and Finance Journal, 5 (3), 17–34.

Zhang, T. C., Kandampully, J., & Choi, H. Y. (2014). The role of employee well-

ness programme in the hospitality industry: a review of concepts, research,

and practice. Research in Hospitality Management, 4 (1 & 2), 45–54. CHAPTER 1: CEO, CSR AND FIRM VALUE 72

Table 1: Variable Deﬁnitions This table reports the variables used in our empirical analysis and their deﬁnitions.

Variable Description

As a general measure of the corporate social performance in the paper, we form it by adding up scaled KLD net scores across seven di- CSR mensions for stakeholder management: product quality and safety, environment, employee relations, corporate governance, diversity and community relations.

As an indicator for the effect of CEO turnover, it is defined as a T urnover dummy variable that takes the value of unity if the observation is under the influence of managerial succession

AT Logarithmic total assets

M/B Market to book ratio

Tangible assets (plant, property and equipment) scaled by total PP &E assets

LR Market leverage ratio

ROA Return (EBITDA) on assets ratio

R&D Research and development expenses scaled by total assets

ADV Advertising expense scaled by total assets

Proxy for the firm’s willingness to engage in earnings management. The proxy is calculated by subtracting the average level of accruals exACCR of the industry from the level of accruals scaled by the firm’s total assets. The amount of accruals are calculated as the difference between net income and cash from operations

Marginal tax rate is the amount of tax paid on one more dollar of MTR income, which is the proxy for the cost of investment in CSR in terms of after-tax dollars

SG&A Non-production expense scaled by the total assets CHAPTER 1: CEO, CSR AND FIRM VALUE 73

A individual is deﬁned as outsider if the one was appointed to Outsider CEO in less than 2 years since he or she joined the company

T enure Number of years in oﬃce

The dummy variable takes the value of unity if the CEO is also a Director board director, and zero, otherwise

The dummy variable takes the value of unity if the CEO holds the Chair chair position of the board, and zero, otherwise

Age The present age of the CEO for the current ﬁscal year

Gender The gender of the CEO

Number of shares owned by CEO scaled by the total shares out- CEOshare standing

Dirsum Total number of directors serving on the board

Outpercn Percentage of outsider directors

MA Managerial ability constructed in Demerjian et al.(2012) CHAPTER 1: CEO, CSR AND FIRM VALUE 74

Table 2: Summary Statistics of CEO Turnover This table reports the summary statistics of our sample for the period 1996 to 2017. See Table 1 for variable deﬁnitions.

Turnover statistics by year

Number of Number of Fiscal Year Unique CEO Turnover Rate Average Tenure CEOs Turnovers

1994 34 3 0.088 12.651 1995 268 20 0.075 8.431 1996 287 8 0.028 9.042 1997 303 26 0.086 8.803 1998 322 26 0.081 8.574 1999 318 31 0.097 7.983 2000 308 27 0.088 8.317 2001 505 41 0.081 8.080 2002 418 26 0.062 7.687 2003 721 38 0.053 8.754 2004 722 36 0.050 8.669 2005 661 48 0.073 8.507 2006 702 46 0.066 8.367 2007 948 90 0.095 8.177 2008 970 86 0.089 8.211 2009 1151 93 0.081 8.325 2010 1149 92 0.080 8.484 2011 1106 84 0.076 8.980 2012 1112 93 0.084 8.874 2013 986 99 0.100 8.819

Total 12991 1013 0.078 8.680 CHAPTER 1: CEO, CSR AND FIRM VALUE 75

Table 3: Descriptive Statistics This table reports the descriptive statistics of CSR measure, CEO characteristics and ﬁrm fundamentals (level & change). See Table 1 for variable deﬁnitions.

N Mean Median p25 p75 Min Max Std. Dev. Panel A: Firm Characteristics (level) AT 14947 7.992 7.832 6.736 9.061 2.963 14.698 1.697 M/B 14946 3.039 2.195 1.458 3.511 -0.527 18.039 2.832 LR 14952 0.209 0.212 0.041 0.295 0.000 0.554 0.153 PP &E 14947 0.628 0.620 0.440 0.838 0.014 1.000 0.250 R&D 14947 0.026 0.000 0.000 0.024 0.000 0.887 0.052 ROA 14947 0.048 0.048 0.015 0.087 -2.075 0.783 0.095 ADV 14947 0.012 0.000 0.000 0.006 0.000 0.963 0.038 exACCR 14761 0.045 -0.004 -0.038 0.060 -2.085 7.232 0.259 MTR 13164 0.206 0.318 0.025 0.350 0.000 0.391 0.156 SG&A 14947 0.197 0.182 0.027 0.294 0.000 2.957 0.208 Panel B: Firm Characteristics (change) ∆AT 13716 0.084 0.062 -0.004 0.146 -1.382 2.535 0.196 ∆M/B 13716 -0.016 0.000 0.000 0.000 -4.524 3.861 0.872 ∆LR 13716 0.001 0.000 -0.020 0.016 -0.467 0.746 0.060 ∆PP &E 13716 0.003 0.000 -0.019 0.017 -0.605 0.692 0.064 ∆R&D 13716 0.000 0.000 0.000 0.000 -0.579 0.606 0.022 ∆ROA 13716 -0.002 0.000 -0.017 0.016 -2.133 2.134 0.102 ∆ADV 13716 0.000 0.000 0.000 0.000 -0.022 0.023 0.005 ∆exACCR 13716 0.007 -0.002 -0.024 0.031 -1.129 1.158 0.237 ∆MTR 11521 -0.007 0.000 -0.004 0.004 -0.367 0.373 0.108 ∆SG&A 13716 -0.001 0.000 -0.009 0.008 -1.058 1.004 0.054 Panel C: CEO Characteristics T urnover 13716 0.083 0.000 0.000 0.000 0.000 1.000 0.276 Director 13716 0.962 1.000 1.000 1.000 0.000 1.000 0.191 Chair 13716 0.646 1.000 1.000 1.000 0.000 1.000 0.478 Age 13164 56.840 57.000 52.000 61.000 32.000 96.000 7.468 Gender 13716 0.978 1.000 1.000 1.000 0.000 1.000 0.146 CEOshare 13716 -0.098 0.000 -0.020 0.037 -53.200 87.600 2.769 Outsider 13716 0.267 0.000 0.000 0.000 0.000 1.000 0.442 Panel D: CSR Measures (level) CSR 14952 -0.145 -0.167 -0.499 0.125 -2.976 3.750 0.592 ∆CSR 13535 0.264 0.175 0.000 0.333 0.000 4.432 0.338 | | CHAPTER 1: CEO, CSR AND FIRM VALUE 76

Table 4: Regressions of CEO Fixed Effects on CSR This table reports the results of the regressions that demonstrate the CEO fixed effects upon CSR activities. Our dependent variable is the level of CSR. t-statistics are in the parentheses denoted by ***, ** and * for the significance levels at the 0.1%, 1% and 5% levels, respectively.

(1) (2) (3) (4) CEO FE yes yes yes Firm Characteristics: AT 0.09*** -0.04* -0.045** (20.10) (-2.26) (-2.60) M/B 2.76e-04 -0.006** -0.006** (0.13) (-2.79) (-2.66) LR -0.211*** -0.203*** -0.21*** (-6.39) (-6.06) (-6.26) PP &E 0.042 -0.143* -0.139* (1.09) (-2.23) (-2.16) R&D 0.155 -0.584* -0.621** (1.04) (-2.55) (-2.71) ROA 0.204*** -0.077 -0.075 (3.40) (-1.18) (-1.14) ADV 0.193 -0.211 -0.171 (1.13) (-0.60) (-0.54) exACCR 0.174*** 0.183*** 0.186*** (7.32) (4.22) (4.28) MTR 0.123*** 0.104* 0.107* (3.50) (2.47) (2.53) SG&A 0.121** 0.096 0.076 (2.99) (1.13) (0.90) Fixed Eﬀects: Year yes yes yes yes Four-Digit SIC Industry yes yes yes Fama French 49 Industry yes

Joint test: βk = 0, k = 1, 2, ... F-statistic 6.40*** 5.96*** 5.91*** P-value 0.0000 0.0000 0.0000 Joint test: γ = 0 F-statistic 56.54*** 9.67*** 10.26*** P-value 0.0000 0.0000 0.0000 Adj.R2 0.247 0.528 0.590 0.589 N 12,992 14,952 12,992 12,992 CHAPTER 1: CEO, CSR AND FIRM VALUE 77

Table 5: Regressions of CEO Turnover on Absolute CSR Change This table reports the results of the regressions in which the absolute value of the change in CSR is the dependent variable. The main control variable of interest is T urnover. The ﬁrm characteristics variables and CEOshare are also absolute change. Pesudo R2 are reported for GLM and Tobit regressions. t-statistics are in the parentheses denoted by ***, ** and * for the signiﬁcance levels at the 1%, 5% and 10% levels, respectively.

OLS HCSE GLM Tobit (1) (2) (3) (4) (5) (6) T urnover 0.126*** 0.122*** 0.126*** 0.122*** 0.496*** 0.310*** (10.35) (9.97) (7.79) (7.58) (11.11) (4.93) Firm Characteristics: ∆AT -0.019 -0.013 -0.019 -0.013 -0.076 -0.113 | | (-0.85) (-0.57) (-0.96) (-0.64) (-0.82) (-0.99) ∆M/B 0.015*** 0.014*** 0.015*** 0.014** 0.068*** 0.052** | | (3.56) (3.38) (3.10) (3.05) (4.03) (2.21) ∆LR -0.029 0.007 -0.029 0.007 0.049 -0.154 | | (-0.45) (0.11) (-0.51) (0.12) (0.17) (-0.45) ∆PP &E -0.15* -0.149* -0.15*** -0.149** -0.257 -0.706*** | | (-2.33) (-2.29) (-3.02) (-2.93) (-0.91) (-2.79) ∆R&D 0.125 0.097 0.125 0.097 0.093 0.444 | | (0.85) (0.65) (1.32) (1.00) (0.16) (0.75) ∆ROA 0.024 0.042 0.024 0.042 0.143 -0.005 | | (0.62) (1.04) (0.66) (1.15) (0.59) (-0.03) ∆ADV 0.27 0.102 0.27 0.102 4.842 -0.392 | | (0.37) (0.13) (0.38) (0.13) (1.46) (-0.10) ∆exACCR -0.029 -0.035 -0.029* -0.035* -0.169* -0.198** | | (-1.51) (-1.78) (-1.67) (-1.97) (-2.31) (-2.33) ∆MTR -0.01 -0.002 -0.01 -0.002 -0.159 -0.029 | | (-0.33) (-0.08) (-0.37) (-0.09) (-1.21) (-0.22) ∆SG&A -0.207** -0.22** -0.207*** -0.22*** -0.857* -1.392*** | | (-2.76) (-2.88) (-3.62) (-3.54) (-2.56) (-2.86) CEO Characteristics: Director 0.015 0.009 0.015 0.009 0.081 0.019 (0.85) (0.48) (0.79) (0.47) (1.24) (0.27) Chair 0.022** 0.018* 0.022*** 0.018* 0.073* 0.070** (3.25) (2.55) (3.16) (2.46) (2.45) (2.32) Age -0.001* -0.001* -0.001*** -0.001* -0.003 -0.004** (-2.50) (-2.06) (-2.68) (-2.21) (-1.84) (-2.32) Gender 0.032 0.043* 0.032 0.043 -0.073 0.169** (1.55) (2.00) (1.43) (1.77) (-0.59) (2.12) ∆CEOshare 0.001 0.001 0.001 0.001 0.002 0.004 | | (0.76) (0.69) (0.69) (0.64) (0.35) (0.45) Fixed Eﬀects: Year yes yes yes yes yes yes FF 49 yes yes yes yes 4-Digit SIC yes yes

Adj. R2 0.247 0.256 0.253 0.256 0.238 0.382 N 10,964 10,964 10,964 10,964 10,964 10,964 CHAPTER 1: CEO, CSR AND FIRM VALUE 78

Table 6: Regressions of CEO Turnover on CSR Change This table reports the results of the regressions where the change in CSR is the dependent variable. The main control variable of interest is CEO turnover. t- statistics are in the parentheses denoted by ***, ** and * for the signiﬁcance levels at the 1%, 5% and 10% levels, respectively.

(1) (2) (3) (4) T urnover 0.076 0.073 0.076 0.073 (4.76)*** (4.52)*** (3.62)*** (3.47)*** Firm characteristics: ∆AT -0.027 -0.024 -0.027 -0.024 (-1.06) (-0.94) (-1.25) (-1.10) ∆M/B 0.010** 0.010** 0.010* 0.010* (2.12) (2.04) (1.95) (1.91) ∆LR -0.033 -0.036 -0.033 -0.036 (-0.45) (-0.48) (-0.53) (-0.55) ∆PP &E -0.120* -0.121* -0.120** -0.121** (-1.80) (-1.78) (-2.32) (-2.28) ∆R&D 0.035 0.029 0.035 0.029 (0.18) (0.15) (0.27) (0.23) ∆ROA -0.019 -0.021 -0.019 -0.021 (-0.42) (-0.44) (-0.47) (-0.49) ∆ADV 0.841 0.916 0.841 0.916 (0.90) (0.97) (0.91) (0.96) ∆exACCR 0.027 0.021 0.027 0.021 (1.28) (1.00) (1.50) (1.16) ∆MTR 0.066* 0.062 0.066* 0.062* (1.72) (1.59) (1.85) (1.70) ∆SG&A -0.047 -0.049 -0.047 -0.049 (-0.49) (-0.51) (-0.63) (-0.65) CEO characteristics : Director 0.046* 0.059** 0.046* 0.059** (1.95) (2.36) (1.77) (2.19) Chair 0.011 0.004 0.011 0.004 (1.17) (0.42) (1.13) (0.41) Age 0.000 0.001 0.000 0.001 (0.54) (0.89) (0.58) (0.95) Gender 0.042 0.048 0.042 0.048 (1.49) (1.62) (1.37) (1.48) ∆CEOshare -0.000 -0.000 -0.000 -0.000 (-0.09) (-0.17) (-0.08) (-0.15) Fixed Eﬀects: Year yes yes yes yes FF 49 yes yes 4-Digit SIC yes yes

Adj. R2 0.139 0.153 0.139 0.153 N 10,964 10,964 10,964 10,964 CHAPTER 1: CEO, CSR AND FIRM VALUE 79

Table 7: Regressions of CEO Turnover and Tenure on CSR Change This table reports the results of the regressions where change in CSR is the dependent variable using HCSE. The main control variable of interest is T enurek. t-statistics are in the parentheses denoted by ***, ** and * for the signiﬁcance levels at the 1%, 5% and 10% levels, respectively.

Positive change in CSR Negative Change in CSR (1) (2) (3) (4) T urnover 0.467*** 0.466*** -0.318*** -0.330*** (12.88) (11.64) (-11.91) (-11.31) T enure2 6 -0.108 -0.088 -0.052 -0.052 − (-0.97) (-0.61) (-0.59) (-0.59) T enure6+ -0.138 -0.085 -0.050 -0.030 (-1.41) (-0.65) (-0.62) (-0.38) Firm characteristics: ∆AT -0.048 -0.071 -0.014 0.030 (-0.66) (-0.80) (-0.21) (0.36) ∆M/B 0.006 0.003 0.009 0.007 (0.47) (0.18) (0.75) (0.61) ∆LR 0.355 0.336 -0.118 -0.185 (1.41) (1.20) (-0.67) (-0.93) ∆PP &E -0.040 -0.051 0.061 0.157 (-0.16) (-0.18) (0.32) (0.69) ∆R&D 0.266 0.140 1.327 1.370 (0.26) (0.10) (1.62) (1.46) ∆ROA 0.235 0.213 -0.074 -0.052 (1.31) (1.00) (-0.70) (-0.42) ∆ADV -2.038 -2.648 2.655 0.908 (-0.63) (-0.72) (0.77) (0.22) ∆exACCR -0.021 -0.090 0.056 0.049 (-0.25) (-1.01) (1.03) (0.87) ∆MTR 0.124 0.161 -0.143 -0.146 (0.82) (0.97) (-1.50) (-1.39) ∆SG&A 0.577 0.556 -0.685** -0.630* (1.49) (1.04) (-2.55) (-1.89) CEO characteristics: Director 0.153 0.190 0.090 0.067 (1.24) (1.07) (1.46) (0.93) Chair 0.056 0.029 0.030 0.022 (1.54) (0.67) (1.08) (0.56) Age 0.001 0.001 0.003 0.003 (0.32) (0.44) (1.37) (1.03) Gender 0.133 0.172 0.115* 0.140 (1.35) (1.46) (1.68) (1.60) ∆CEOshare -0.011* -0.014** 0.001 0.003 (-1.85) (-2.16) (0.19) (0.40) Fixed Eﬀects: Year yes yes yes yes FF 49 yes yes 4-Digit SIC yes yes

Adj. R2 0.224 0.312 0.213 0.276 N 1,223 1,223 1,448 1,448 CHAPTER 1: CEO, CSR AND FIRM VALUE 80

Table 8: Regression of CEO Turnover and Outsider on CSR Change This table reports the results of the regressions where the change in CSR is the dependent variable. The main control variable of interest, Outsider, is a dummy variable indicating whether the new CEO is from outside the organization. t- statistics are in the parentheses denoted by ***, ** and * for the signiﬁcance levels at the 1%, 5% and 10% levels, respectively.

(1) (2) (3) (4) T urnover 0.095*** 0.094*** 0.095*** 0.094*** (3.87) (3.66) (3.38) (3.23) Outsider -0.098** -0.101** -0.098** -0.101** (-2.43) (-2.35) (-2.29) (-2.30) Firm characteristics: ∆AT -0.000 0.004 -0.000 0.004 (-0.01) (0.07) (-0.01) (0.09) ∆M/B 0.014* 0.014 0.014 0.014 (1.70) (1.63) (1.58) (1.48) ∆LR -0.029 -0.033 -0.029 -0.033 (-0.18) (-0.20) (-0.23) (-0.26) ∆PP &E -0.146 -0.135 -0.146 -0.135 (-0.89) (-0.79) (-1.06) (-0.94) ∆R&D 0.365 0.320 0.365 0.320 (0.59) (0.48) (0.72) (0.62) ∆ROA -0.072 -0.110 -0.072 -0.110 (-0.70) (-1.03) (-0.70) (-1.05) ∆ADV 2.688 2.024 2.688 2.024 (1.35) (0.95) (1.43) (1.01) ∆exACCR 0.058 0.040 0.058 0.040 (1.01) (0.66) (1.17) (0.75) ∆MTR 0.064 0.085 0.064 0.085 (0.79) (1.00) (0.80) (1.02) ∆SG&A -0.064 -0.088 -0.064 -0.088 (-0.27) (-0.35) (-0.35) (-0.43) CEO characteristics : Director 0.040 0.070 0.040 0.070 (0.75) (1.17) (0.60) (0.94) Chair 0.028 0.009 0.028 0.009 (1.49) (0.42) (1.42) (0.41) Age 0.001 0.001 0.001 0.001 (0.41) (0.46) (0.45) (0.48) Gender 0.081 0.070 0.081 0.070 (1.43) (1.07) (1.39) (1.04) ∆CEOshare -0.000 -0.002 -0.000 -0.002 (-0.07) (-0.31) (-0.08) (-0.40) Fixed Eﬀects: Year yes yes yes yes FF 49 yes yes 4-Digit SIC yes yes

Adjusted R2 0.112 0.149 0.112 0.149 N 3,520 3,520 3,520 3,520 CHAPTER 1: CEO, CSR AND FIRM VALUE 81

Table 9: Trust, CSR and Firm Performance

This table reports the results of the regressions, where the change in T obin0s Q is the dependent variable. The main control variable of interest is the change of CSR. The sample consists of matched pairs by industry and year. The treatment group consists of firms with a major change in CSR during the CEO transition period. The control sample consists of firms where the change in CSR is below the median change of all firms experiencing a turnover during the sample period. t-statistics are in the parentheses denoted by ***, ** and * for the significance levels at the 1%, 5% and 10% levels, respectively.

Matched sample

Entire sample period Crisis period (2007 – 2009) (1) (2) (3) (4) Opposite sign Opposite sign ∆CSR 0.028 -0.010 0.482*** 0.535** (0.35) (-0.09) (3.59) (2.65) ∆AT -0.385 -1.227** -0.249 -0.281 (-1.11) (-2.49) (-1.32) (-0.86) ∆LR -0.140 -0.021 1.227*** 1.541** (-0.60) (-0.06) (2.93) (2.21) ∆ROA 1.574** 3.235*** 0.241 1.127 (2.18) (3.35) (0.68) (1.68) ∆PP &E 1.662 0.385 0.257 0.981 (1.27) (0.22) (0.40) (0.73) ∆R&D 4.675 8.500 -5.860 -8.890 (0.94) (1.36) (-1.65) (-1.40) ∆ADV 3.075 -21.634 0.345 -2.875 (0.27) (-1.34) (0.05) (-0.25) ∆SG&A 1.693 -1.739 0.002 0.388 (1.40) (-0.75) (0.00) (0.23)

Year Fixed Eﬀects yes yes yes yes Industry Fixed Eﬀects Yes Yes No No Adj. R2 0.576 0.627 0.213 0.201 N 318 182 74 46 CHAPTER 1: CEO, CSR AND FIRM VALUE 82

Table 10: Industry Norm, CSR and Firm Performance

This table reports the results of the regressions where the change in T obin0s Q is the dependent variable. The main control variable of interest is the change of CSR. The sample consists of matched pairs by industry and year. The treatment group consists of firms with a major change in CSR during the CEO transition period. The control sample consists of firms where the change in CSR is below the median change of all firms experiencing a turnover during the sample period. The relevant variable is NORM which equals one if the new CEO of the treated sample is moving towards the industry norm (columns 1 and 2) or upwards and away from the industry norm (columns 3 and 4). t-statistics are in the parentheses denoted by ***, ** and * for the significance levels at the 1%, 5% and 10% levels, respectively.

Matched sample (1) (2) (3) (4) Move down Move up Opposite sign towards away from the Norm the Norm ∆CSR 0.028 -0.010 -0.032 0.047 (0.35) (-0.09) (-0.40) (0.57) NORM -0.334*** 0.355** (-2.67) (2.35) ∆AT -0.385 -1.227** -0.348 -0.270* (-1.11) (-2.49) (-1.00) (-1.76) ∆LR -0.140 -0.021 -0.165 -0.327 (-0.60) (-0.06) (-0.68) (-0.89) ∆ROA 1.574** 3.235*** 1.664** -0.159 (2.18) (3.35) (2.22) (-0.65) ∆PP &E 1.662 0.385 1.538 1.492* (1.27) (0.22) (1.29) (1.87) ∆R&D 4.675 8.500 5.059 1.543 (0.94) (1.36) (1.00) (1.27) ∆ADV 3.075 -21.634 3.384 5.293 (0.27) (-1.34) (0.32) (0.99) ∆SG&A 1.693 -1.739 1.925 2.366 (1.40) (-0.75) (1.52) (0.21)

Year Fixed Eﬀects yes yes yes yes Industry Fixed Eﬀects yes yes yes yes

Adj. R2 0.576 0.627 0.593 0.583 N 318 182 318 318 CHAPTER 1: CEO, CSR AND FIRM VALUE 83

Table 11: Trust, CSR and Firm Value This table reports the stock return difference between firms that introduced a new CEO in 2003 that increase the CSR (denoted as trustworthy firms) and those that did not increase the CSR (denoted as Not Trusted. Panel A reports the regression results where the dependent variable is the annual return and independent variable is an indicator equal to one if the firm is trustworthy. Panel B provides a summary of the univariate tests. t-statistics are in the parentheses denoted by ***, ** and * for the significance levels at the 1%, 5% and 10% levels, respectively.

Panel A

Stock Return 2007-2008 Stock Return 2008-2009

(1) (2) (1) (2)

Robust HCSE Robust HCSE

Trusted -0.053 -0.053 1.047*** 1.047** (-0.53) (-0.53) (2.76) (2.46) constant -0.362*** -0.362*** 0.144 0.144 (-5.43) (-5.51) (0.59) (0.95)

Adjusted R2 -0.025 -0.025 0.167 0.167 N 31 31 31 31

Panel B

Trusted Not Trusted Test of Diﬀerence

2007-2008

mean -0.415 -0.362 -0.053 t-stat -5.532*** -5.526*** -0.528 N 14 17

2008-2009

mean 1.19 0.144 1.047 t-stat 2.973*** 0.959 2.446** N 14 17 CHAPTER 1: CEO, CSR AND FIRM VALUE 84

Table 12: Regressions of CEO-FEs on CSR & Additional Controls This table reports the results of the regressions that demonstrate the CEO fixed effects upon CSR activities. Our dependent variable is the level of CSR. This table includes the lagged values of total number of directors (laggednumDir) as well as the percentage of outside directors (Lagged Outpercn). In addition, we include proxy for managerial ability (MA). t-statistics are in the parentheses denoted by ***, ** and * for the significance levels at the 0.1%, 1% and 5% levels, respectively.

(1) (2) (3) (4) AT 0.111 0.001 -0.003 (14.49)*** (0.02) (-0.12) M/B -0.005 -0.009 -0.009 (-2.04)** (-2.84)*** (-2.81)*** LR -0.207 -0.175 -0.176 (-4.74)*** (-4.00)*** (-4.04)*** PP &E 0.015 -0.163 -0.162 (0.28) (-1.89)* (-1.88)* R&D 0.175 -0.64 -0.64 (0.89) (-2.14)** (-2.14)** ROA 0.255 -0.155 -0.142 (3.14)*** (-1.59) (-1.46) ADV 0.326 0.059 -0.136 (1.32) (0.13) (-0.36) exACCR 0.214 0.248 0.241 (5.53)*** (3.02)*** (2.94)*** MTR 0.067 0.16 0.158 (1.44) (2.70)*** (2.66)*** SG&A 0.356 0.253 0.248 (5.96)*** (1.95)* (1.92)* LaggedDirsum 0.021 0.005 0.004 (5.29)*** (0.96) (0.77) LaggedOutpercn 0.042 0.15 0.144 (0.82) (2.91)*** (2.80)*** MA 0.006 0.014 0.015 (0.22) (0.54) (0.58) Fixed Eﬀects: Year yes yes yes yes Four-Digit SIC Industry yes yes yes Fama French 49 Industry yes

Joint test: βk = 0, k = 1, 2, ... F-statistic 6.33*** 5.78*** 5.78*** P-value 0.0000 0.0000 0.0000 Joint test: γ = 0 F-statistic 41.59*** 4.88*** 4.81*** P-value 0.0000 0.0000 0.0000 Adj.R2 0.297 0.626 0.703 0.702 N 7,852 14,952 7,852 7,852 CHAPTER 1: CEO, CSR AND FIRM VALUE 85

Table 13: Regressions of Turnover on ∆CSR & Additional Controls This table reports the results of the regressions where the change in CSR is the dependent variable including board of director characteristics and ability proxy. t-statistics are in the parentheses denoted by ***, ** and * for the signiﬁcance levels at the 1%, 5% and 10% levels, respectively.

(1) (2) (3) (4) T urnover 0.084 0.083 0.084 0.083 (3.56)*** (3.41)*** (2.78)*** (2.66)*** Firm characteristics: ∆AT -0.012 -0.017 -0.012 -0.017 (-0.30) (-0.43) (-0.35) (-0.49) ∆M/B 0.015 0.015 0.015 0.015 (2.38)** (2.29)** (2.16)** (2.12)** ∆LR 0.018 0.031 0.018 0.031 (0.18) (0.29) (0.21) (0.34) ∆PP &E -0.134 -0.139 -0.134 -0.139 (-1.43) (-1.45) (-1.82)* (-1.82)* ∆R&D 0.038 0.044 0.038 0.044 (0.16) (0.18) (0.25) (0.28) ∆ROA -0.012 -0.014 -0.012 -0.014 (-0.18) (-0.21) (-0.20) (-0.23) ∆ADV 1.511 1.762 1.511 1.762 (1.22) (1.38) (1.18) (1.33) ∆exACCR 0.026 0.037 0.026 0.037 (0.54) (0.76) (0.53) (0.74) ∆MTR 0.052 0.050 0.052 0.050 (0.91) (0.86) (1.01) (0.95) ∆SG&A 0.021 0.015 0.021 0.015 (0.14) (0.10) (0.17) (0.12) Board characteristics: Lagged Dirsum -0.001 -0.001 -0.001 -0.001 (-0.23) (-0.26) (-0.22) (-0.25) Lagged Outpercn -0.055 -0.067 -0.055 -0.067 (-1.10) (-1.31) (-1.18) (-1.39) CEO characteristics: Director 0.070 0.073 0.070 0.073 (1.97)** (1.93)* (1.98)** (1.86)* Chair 0.008 -0.005 0.008 -0.005 (0.63) (-0.35) (0.62) (-0.35) Age -0.000 0.000 -0.000 0.000 (-0.01) (0.28) (-0.01) (0.30) Gender 0.051 0.060 0.051 0.060 (1.31) (1.46) (1.10) (1.28) ∆CEOshare 0.002 0.001 0.002 0.001 (0.79) (0.55) (0.98) (0.67) MA -0.030 -0.028 -0.030 -0.028 (-1.36) (-1.27) (-1.16) (-1.08) Fixed Eﬀects: Year yes yes yes yes FF 49 yes yes 4-Digit SIC yes yes

Adj. R2 0.130 0.153 0.130 0.153 N 6,234 6,234 6,234 6,234 CHAPTER 1: CEO, CSR AND FIRM VALUE 86

Table 14: Regressions of Predicted CEO Turnover on CSR This table reports the results of the regressions where the change in CSR is the dependent variable. The main control variable of interest is CEO turnover. In this table, we include the predicted turnover variable obtained from the ﬁrst stage regression. t-statistics are in the parentheses denoted by ***, ** and * for the signiﬁcance levels at the 0.1%, 1% and 5% levels, respectively.

Entire Sample Positive change in CSR Negative Change in CSR (1) (2) (3) (4) (5) (6) P redicted T urnover 0.220** 0.242** 1.356*** 1.560*** -1.037*** -1.356*** (2.23) (2.24) (2.90) (2.61) (-3.65) (-3.12) Firm characteristics: ∆AT -0.033 -0.033 0.053 0.118 -0.146 -0.109 (-1.32) (-1.26) (0.51) (0.80) (-1.69)* (-0.95) ∆M/B 0.009 0.010 0.007 0.010 0.015 0.022 (1.47) (1.54) (0.44) (0.46) (0.97) (1.27) ∆LR -0.029 -0.016 0.364 0.219 -0.181 -0.251 (-0.38) (-0.21) (1.01) (0.49) (-0.66) (-0.79) ∆PP &E -0.129 -0.131 -0.263 -0.254 0.280 0.421 (-2.17)** (-2.15)** (-0.80) (-0.64) (1.04) (1.20) ∆R&D 0.024 0.014 -0.368 -0.352 2.733 3.066 (0.18) (0.10) (-0.34) (-0.17) (2.11)** (1.79)* ∆ROA 0.032 0.020 0.850*** 0.917** -0.193 -0.107 (0.61) (0.37) (2.76) (2.24) (-1.18) (-0.53) ∆ADV 0.796 0.911 2.559 -1.279 2.499 1.504 (0.75) (0.82) (0.51) (-0.21) (0.63) (0.30) ∆exACCR 0.043** 0.036* 0.050 -0.035 0.010 0.017 (2.14) (1.74) (0.46) (-0.22) (0.16) (0.24) ∆MTR 0.037 0.030 -0.023 0.027 -0.066 -0.071 (0.89) (0.72) (-0.11) (0.10) (-0.54) (-0.49) ∆SG&A -0.121 -0.113 0.431 0.692 -1.356*** -1.349*** (-1.40) (-1.28) (0.76) (0.73) (-3.61) (-2.82) CEO characteristics : Director 0.053* 0.071** 0.030 0.078 0.090 0.103 (1.72) (2.24) (0.19) (0.27) (0.91) (0.87) Chair 0.005 -0.001 0.020 -0.009 0.046 0.066 (0.40) (-0.11) (0.38) (-0.12) (1.29) (1.31) Age -0.000 -0.000 -0.008** -0.010** 0.009*** 0.010** (-0.47) (-0.54) (-2.51) (-2.08) (3.09) (2.53) Gender 0.026 0.040 0.004 -0.009 0.122 0.118 (0.67) (0.91) (0.02) (-0.05) (0.95) (0.64) ∆CEOshare 0.000 0.000 -0.022*** -0.020** 0.014*** 0.014* (0.32) (0.26) (-2.62) (-2.09) (2.68) (1.96) Fixed eﬀects dummies: Year yes yes yes yes yes yes FF 49 yes yes yes 4-Digit SIC yes yes yes Adj. R2 0.143 0.163 0.149 0.254 0.180 0.270 N 8,039 8,039 792 792 855 855 CHAPTER 1: CEO, CSR AND FIRM VALUE 87 Figure 1: AnAn Integrated integrated Theoretical theoretical Framework modelcorresponding frames theories our in hypothesesdelineate the development the and literature primary empirical are relationshipslines. tests, in suggested where the by the the shapes key postulated of variables hypotheses of oval for concern and diﬀerent and rectangular, CEO the respectively, tenure and stages the separated single by the and dashed two-way arrows CHAPTER 1: CEO, CSR AND FIRM VALUE 88 Figure 2.B Figure 2.A Figure 2: CEO’sThe Investment below Cycle graphs in depict Socialnew the Capital CEO changes in initiates CSR a that positive occur change while during and Figure after B the depicts CEO the transition. CSR change Figure if A the depicts new the CEO CSR initiates change if a the negative change. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 89

Chapter 2: Impact of Internal Governance on a CEO’s Investment Cycle

Ivan E. Brick† Darius Palia†‡ Yankuo Qiao†

This paper examines the impact of internal governance on a CEO’s investment cycle. Extant literature defines internal governance as the mechanism by which senior executives help discipline the CEO to maximize shareholder value. Weis- bach(1995) finds that a year or two before the CEO retires, the firm experiences a decrease in total investment. Pan, Wang, and Weisbach(2016) find evidence of a CEO’s investment cycle, in which investment increases over a CEO’s tenure, whereas disinvestment decreases. These papers suggest that older CEOs incur agency costs as they try to extract rents as their investment horizon declines. We confirm their results, and additionally find that good internal governance helps reduce older CEOs underinvesting before their exit. For younger CEOs we do

†Rutgers Business School–Newark and New Brunswick ‡Columbia Law School CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 90 not find any relationship between internal governance and investment. This suggests that older CEOs who face good internal governance underinvest less. We also find that new incoming CEOs divest these projects profitably. These results are robust to: normal CEO retirements (exclude performance-related turnovers), sudden CEO deaths, and controlling for measures of board size, proportion of outsiders on the board, CEO pay-performance sensitivity, CEO pay slice, CEO power, firm complexity and if the CEO was an outsider or not.

1 Introduction

Many papers (see, for example, Fama and Jensen 1983a, 1983b; Jensen 2000;

Jensen and Meckling 1976) have suggested that value-maximization for the residual claimant, namely shareholders, is the preferred efficient goal of public corporations. Managerial agency theory has posited that CEOs who own less than 100% in their firm deviate from shareholder wealth maximization because any suboptimal decision that personally benefits the CEO costs the CEO less than the value loss to shareholders.

Moreover, the CEO-shareholder problem can become more acute if the investment horizon of the CEO is shorter when compared to the investment horizon of their stockholders. This is more likely to be case when the CEO approaches retirement. Murphy and Zimmerman(1993) find CEO turnover-related changes CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 91 in R&D, advertising, capital expenditures, and accounting accruals are due to poor firm performance. Weisbach(1995) finds that a year or two before the CEO retires, the firm experiences a decrease in total investment. He suggests that this decrease reflects the agency costs of the replaced CEO – who is sacrificing future cash flows in order to realize immediate increased cash flows. In a more recent and relevant paper, Pan et al.(2016) find that investment increases over a CEO’s tenure, whereas disinvestment decreases. They call this pattern a CEO’s investment cycle, which results in significant asset growth and employment over the

CEO’s tenure. The authors suggest that the CEO agency problems are reduced with the arrival of a new CEO, who does not enjoy the same private beneﬁts as her predecessor. In this paper, we conﬁrm the results of Pan et al.(2016) using a similar methodology employed by Aggarwal, Fu, and Pan(2017).

This paper examines the impact of internal governance on a CEO’s investment cycle. Acharya, Myers, and Rajan(2011) theorize that internal governance may mitigate the CEO horizon problem. The potential CEO successor of the company, they argue, should have a longer horizon than the current CEO. If the current

CEO puts in place policies that destroy the capital and reputational stock of the

ﬁrm, then the successor will ﬁnd herself when named CEO running a diminished

ﬁrm. Accordingly, lower level managers who hope to succeed the current CEO will oppose CEO policies that do not maximize the value of the ﬁrm. The current CEO is “forced” to abide by the wishes of the lower management, since the current CEO needs the assistance of the lower level manager to produce current earnings that CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 92 support the current stock price, which is an important factor in the level of the

CEO’s current compensation. On the other hand, if most administrative duties regarding the firm’s business operation are conducted by the subordinates, and the success of the company is so dependent on the effort exerted by the subordinates, then no one is in charge in coordinating the activities of the subordinate to ensure value maximization. Consequently, the optimal internal governance should be such that the responsibilities of the running the firm should be shared as opposed to be run solely by the CEO or only by key top management subordinates.

We deﬁne internal governance by the degree to which the other members in the top management team assist in the decisions with the CEO. According to

Acharya et al.(2011) internal governance works best when neither the CEO nor the subordinate managers are dominant1. Accordingly, they deﬁne a variable δ which is “the fraction of tasks assigned to the CEO” (p.700). A fully decentralized team would have δ = 0, and one where the CEO makes all the contribution is where δ = 1. To operationalize this metric, we use a similar metric suggested by

Aggarwal et al.(2017) by calculating the number of executive titles of the CEO

(f) scaled by the total number of executive titles carried by the top management team of five executives (f + g). In practice, we do three steps. First, we use the regular expression (regex) procedure in R to calculate the number of titles for each executive. Second, if regex does not classify titles for some firm-years, we manually check the proxy statements for filling in the titles. Third, we manually checked

1 For more detailed explanations of internal governance see Section II. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 93 for a random sample of firm-years, that the regex procedure correctly captured the titles. The exact procedure is explained in detail in the Appendix. Finally, we empirically estimate the optimal δ using a non-linear regression specification of firm performance metrics as a function of δ and other control variables.

A number of empirical papers have found that internal governance is beneﬁ- cial in other (non-investment related) contexts. For example, Landier, Sauvagnat,

Sraer, and Thesmar(2012) measure good internal governance by the number of executives appointed before the current CEO and finds that firm’s profitability increases with that number. Jain, Jiang, and Mekhaimer(2016) measure internal governance as the difference in horizons between a CEO and his immediate subordinates and finds that firms with better internal governance have lower information asymmetry and higher liquidity. Finally, Cheng, Lee, and Shevlin(2016) uses the number of years to retirement to capture key subordinate executives’ horizon incentives and their compensation relative to CEO compensation to capture their influence within the firm. They find that the extent of real earnings management decreases with key subordinate executives’ horizon and influence.

We contribute to the internal governance literature by demonstrating that an optimal use of non-CEO senior management can reduce the agency cost of the

firm. In particular, we first confirm Pan et al.(2016) results of CEOs underinvesting before turnover suggesting that the decrease in the investment rate prior to the CEO departure reflects the agency costs of the replaced CEO, who is sacrific- CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 94 ing future cash flows in order to realize immediate increased cash flows. However, we find that firm performance is increasing and then decreasing in δ for older

CEO’s (whose age is greater than or equal to the median CEO age of 56 years).

No statistically significant relationship is found for younger CEOs. Our results are robust to: normal CEO retirements (exclude performance-related turnover), sudden CEO deaths, and controlling for board size, proportion of outsiders on the board, CEO pay-performance sensitivity, CEO pay slice2, CEO power, firm complexity and if the CEO was an outsider or not. Such findings are consistent with the theoretical development of Acharya et al.(2011) that the internal governance mechanism should be effective only if the CEO is myopic and has a short executive horizon. Our results compliment the findings of Cheng et al.(2016) who find that the extent of real earnings management decreases with key subordinate executives’ horizon and influence and that of Jain et al.(2016) who find that firms with better internal governance have lower information asymmetry and higher liquid.

Additionally, we examine the disinvestment policy on CEO turnover. We ﬁnd that good governance has a positive impact on the value and proﬁtability of asset disposals for new CEOs who follow old exiting CEOs, and no impact on new

CEOs who follow exiting young CEOs. These ﬁndings suggest that with good internal governance, the older outgoing CEO is less likely to overpay for previously

2 CEO pay slice is the ratio of the CEO’s pay to the sum of top ﬁve (inclulding CEO) senior executives’ pay. It has been used to capture tournament incentives (Kale, Reis, and Venkateswaran (2009)), or CEO entrenchment (Bebchuk, Cremers, and Peyer 2011; Feng, Ge, Luo, and Shevlin 2011). CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 95 acquired assets and the reason for the disposition of assets is due to the diﬀerential skill-asset match between the incoming CEO and the older outgoing CEO. Taken together, the empirical evidence suggests that good internal governance improves not only the deteriorating investment policy preceding myopic CEO departure in terms of both dollar amount and quality of assets acquired, but ensures the asset disposals incurred at the beginning of a CEO’s tenure are likely not to be sold at a loss.

This paper proceeds as follows. Section II reviews the related literature and

Section III describes our data, variable construction, and sample characteristics.

The empirical results are reported in Section IV, and Section V presents our conclusions.

2 Literature Review

Importance of CEO: A number of papers have shown that CEOs have a large and significant impact on investment and financial policies of the firms as well as

firm performance. Bertrand and Mullainathan(2001) document that managerial styles represented by biographical characteristics of individual executives have significant effect upon the financial outputs of the firm. They find that CEO fixed- effects can substantially explain the heterogeneity in firm investment, financing and organization strategies, and firm performance. Palia(2000) finds that CEOs CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 96 of lower quality (that graduated from lower-ranked universities) are more likely to be CEOs of regulated companies than manufacturing firms who attract CEO with higher quality (that graduated from higher-ranked universities). Additionally, he

ﬁnds that the regulated industries oﬀer their CEOs a lower pay-performance sensitivity than manufacturing CEOs. This suggests that labor markets sort lower

(higher) quality CEOs into more regulated (non-regulated) industries. Heaton

(2002) argues that overoptimistic managers believe that capital markets under- value their firm’s risky securities, and may decline to invest in positive net present value projects that are externally financed. Malmendier and Tate(2005) find that managerial overconfidence can account for corporate investment distortions. Over- confident managers overestimate the returns to their investment projects and view external funds as unduly costly. Thus, they overinvest when they have abundant internal funds, but curtail investment when they require external financing. Baker and Wurgler(2013), show that managerial biases and nonstandard preferences can have a significant impact on the firm’s financing and investment decisions.

CEO Turnover: A number of studies find significant changes in firm policies and performance surrounding CEO turnover. Coughlan and Schmidt(1985),

Warner, Watts, and Wruck(1988), Weisbach(1988), and Huson, Malatesta, and

Parrino(2004) find strong evidence that accounting earnings and market value of the firm decline before CEO turnover. Weisbach(1988) shows that the positive relationship between prior performance and the probability of CEO turnover is more pronounced for firms with a lower proportion of independent directors on CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 97 the board. Huson et al.(2004) find that the extent to which the firm performance improves following the CEO dismissal is positively related to the appointment of an outsider CEO and the presence of a monitoring board. Strong and Meyer

(1987), Elliott and Shaw(1988), Dechow and Sloan(1991), Murphy and Zimmer- man(1993), and Weisbach(1995), demonstrate that during the transition period, there are signiﬁcant asset divestures and write-oﬀs, as well as reduction in capital expenditures resulting in overall investment downsizing. Strong and Meyer

(1987), Elliott and Shaw(1988), Weisbach(1995), Dechow and Sloan(1991) find that the outgoing CEO tends to constrain discretionary expenditures such as R&D development and advertising to boost earnings-based compensations, resulting in declining R&D in the final years of the CEO’s tenure3. Pan et al.(2016) find that the investment rate increases over a CEO’s tenure, whereas disinvestment decreases.

Internal Governance: It is well known that public corporations suﬀer from agency problems because of conﬂict of interest between CEOs and shareholders.

Jensen and Meckling(1976) argues that CEOs who own less than 100% of the ﬁrm deviate from maximizing shareholder wealth by consuming a suboptimal level of perquisites since the cost to the CEO is less than the loss in value to shareholders.

The managerial agency problem can become more acute if the investment horizon of the CEO is short compared to the stockholders of the ﬁrm. This is likely to be the case as CEOs approach retirement since the CEO obtains immediate beneﬁts

3 Butler and Newman (1989) ﬁnd contrary evidence. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 98 from increasing current earnings/cash ﬂows at the expense of future earnings. Age has been widely used as a proxy for the executives’ employment horizon (see, for example, Brickley, Linck, and Coles 1999; Dechow and Sloan 1991; Gibbons and

Murphy 1992; Jain et al. 2016; Matˇejka, Merchant, and Van der Stede 2009).

Acharya et al.(2011) suggest that the internal governance may mitigate the

CEO horizon problem. The potential CEO successor of the company should have a longer horizon than the current CEO. If the current CEO puts in place policies that destroy the capital and reputational stock of the ﬁrm, then the successor will

find herself when named CEO running a diminished firm. Accordingly, lower level managers who hope to succeed the current CEO will oppose CEO policies that do not maximize the value of the firm. The current CEO is “forced” to abide by the wishes of the lower management, since the current CEO needs the assistance of the lower level manager to produce current earnings that support stock price which is an important factor in the level of CEO current compensation.

To achieve optimal results, as shown by Acharya et al.(2011) and Landier,

Sraer, and Thesmar(2009) the CEO must allow other members of the top management team to take part in making decisions.4 The level of sharing of executive responsibilities among top management is referred to as the internal governance of

4 Landier et al.(2009) has a slightly different hierarchical setting than Acharya et al.(2011). In Landier et al.(2009), the vertical organizational structure consists of an informed Decision Maker (DM), in charge of selecting projects and a likely uninformed Implementer (I) in charge of execution. In the face of a dissenting and unmotivated I, DM chooses to use objective information in the selection of projects to ensure successful outcomes. Preference heterogeneity between DM and I (or dissent) leads to more informed decision making, less self-serving activities by DM which results in higher profitability. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 99 the firm. Therefore, if CEO’s power is optimally limited by effective internal governance mechanisms by the executive subordinates, the finalized decisions regarding corporate activities are collective agreements and products of compromise between

CEO and other top subordinates. In contrast, if the internal corporate governance is not effective and if the CEO is sufficiently powerful, then decision making is centralized and ex post performance should be much more variable (Adams, Almeida, and Ferreira(2005)). This is especially the case when the influential CEO shares the human resources role of the board and has strong voice on determining her successor and on career future of her subordinates, given competitive conditions of managerial labor market. On the other hand, if most administrative duties regarding the firm’s business operation are conducted by the subordinates, they are no longer willing to follow the leadership of the CEO who is capable of coordinating the various business activities that can maximize firm value. That is, the success of the company is so dependent on the effort exerted by the subordinates, they may not have the incentive or ability to expend the effort to maximize the value of the firm and optimally invest on behalf of the shareholders since no one is in charge in leading the firm. Consequently, the optimal internal governance should be such that the responsibilities of the running the firm should be shared as opposed to be run solely by the CEO or only by key top management subordinates.

A number of empirical papers have found that internal governance is beneficial in other (non-investment related) contexts. Landier et al.(2012) measures good internal governance by the number of executives appointed before the current CEO CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 100 and finds that firm’s profitability increases with that number. Jain et al.(2016), measure internal governance as the difference in horizons between a CEO and his immediate subordinates and find that firms with better internal governance have lower information asymmetry and higher liquidity. Finally, Cheng et al.(2016) uses the number of years to retirement to capture key subordinate executives’ horizon incentives and their compensation relative to CEO compensation to capture their influence within the firm. They find that the extent of real earnings management decreases with key subordinate executives’ horizon and influence. In all of these papers, there appears to be a linear relationship between the internal governance metric and the output performance metric (profitability, information asymmetry and liquidity, earnings management).

Aggarwal et al.(2017) use titles to capture internal governance. They also

ﬁnd a non-linear relationship between internal governance and ﬁrm performance.

However, they do not examine the change in investments before and after CEO turnover which is the focus of this paper.

3 Data, Variable Construction and Sample

We begin by obtaining the job titles and employment history of the CEO and the other top four subordinate managers of S&P 1500 ﬁrms from Execucomp for the years 1996 to 2017. This substantially extends the data employed by Pan et al. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 101

(2016) and Aggarwal et al.(2017) who end their sample in 2009. We use Excecu-

Comp’s variables TITLEANN and CEOANN to help identify the executive’s job title and CEO annual ﬂag, respectively. We also obtain the CEO’s tenure in each

ﬁscal year and clean up the data to ensure that only one CEO is identiﬁed for any

ﬁscal year.

According to Acharya et al.(2011) internal governance works best when neither the CEO nor their subordinate managers are dominant. The authors define a variable δ = f/ (f + g), which is the fraction of tasks assigned to the CEO. A fully decentralized team would have δ = 0, and one where the CEO makes all the contribution is where δ = 1. To operationalize this metric, we follow the procedure used by Aggarwal et al.(2017). In particular, we first calculate the number of executive titles of the CEO (f) scaled by the total number of executive titles carried by the entire top management team of five executives (f + g). We use the regular expression (regex) procedure in R to calculate the number of titles for each executive. (See the Appendix for more details.)

*** Appendix***

To determine the optimal level of internal governance, we need suitable performance metrics. Since the main channel through which internal governance mitigates agency problem is to effectively constrain the CEO’s myopic motives of under-investing in the firm’s capital stock, our performance proxy should be able to efficiently recognize the growth potential rather than only focus on current cash CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 102 earnings. Following the convention of the extant literature, we use proxies for both accounting performance (ROA) and market performance (M/B)5. Specifically, we define the firm’s return-on-assets ROA as earnings before interest and deprecia- tion (EBITDA) scaled by the total assets at the beginning of the fiscal year. We define M/B as market value of equity divided by the book value of equity. Both

ROA and M/B is winsorized at the 1% level for all Compustat firms. According to Chakravarthy(1986), M/B is an ideal measure for the success of strategic management, which ensures the firm’s long-term adaption to its business environment, given that it faces potential distortions from management. Additionally, M/B is a more forward looking measure than ROA, as it incorporates the market’s perception of the firm’s growth opportunities. Given that both performance measures,

ROA and M/B, are also strongly associated with the condition of the industry in which it operates, we use industry-adjusted performance measures (ROA, M/B) at the two-digit SIC level.

The theoretical model of internal governance is developed in Acharya et al.

(2011). The steady state capital stock for a myopic CEO case is given below.

b θ 1 SS b 1 1 γb k = [γ(1 δ)δ − b 1 ] − − (1 + r) − where kss is the steady state capital stock which is positively related to ﬁrm value;

5 We get similar results using Tobin’s Q deﬁned as the (book value of total assets – deferred taxes - book value of stockholders’ equity + market value of equity)/book value of total assets. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 103

θ is business environment; δ is the fraction of tasks assigned to the CEO, the main variable of interest; and value ranges of other model parameters (b > 1, 1 γb > − 0) ensure convergence in steady state. To examine the functional relationship between ﬁrm value and δ, we diﬀerentiate kss with respect to δ. This results in

ss b b ∂k b 1 θ ( 1 1) b 1 1 θ 1 γb − = [γ(1 δ)δ − b 1 ] − [δ − ((b 1)(1 δ)δ− ) 1][γ b 1 ] ∂δ − (1 + r) − − − − (1 + r) −

The sign of the above comparative statics is determined by that of the expres-

b 1 1 sion in the middle brackets, δ − ((b 1)(1 δ)δ− ) 1 , which turns from positive − − − to negative as δ goes from zero to 1. This indicates a hump shape relationship between ﬁrm performance and the dominance of CEO, as measured by the fraction of tasks δ.

Accordingly, we empirically estimate the following regression speciﬁcation.

2 0 OutcomeV ariableit = β0 + β1δit + β2δit + β xit 1 + γi + λt + εit (1) −

The dependent variables are the industry-adjusted ﬁrm performance variables

(ROA, M/B). We include a linear and squared term for δ given that the optimal internal governance as measured by the fraction of titles held by the CEO as posited by the theory should be non-linear. More speciﬁcally, according to Acharya CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 104

et al.(2011), β1 should be positive and β2 should be negative, and the optimal

δ∗ = β /(2β ). We include x a vector of firm level covariates, and γ and λ − 1 2 it i t are firm and year fixed-effects, respectively. The standard errors of all the fixed- effects models in the paper are two-way clustered by firm and year. Accordingly, companies of relatively effective internal governance can be defined with a dummy variable (IG) that takes the value of unity if the firm’s δ is within the confidence interval of

1 1 δ∗ σ , δ∗ + σ (2) − 2 δ 2 δ

in which σδ denotes the sample standard deviation of δ ; and companies with relatively ineffective governance are defined as zero when δ falls outside the confidence intervals.

To determine if optimal internal governance mitigates agency cost related to the shortening of the investment horizon of the CEO, we will examine the relationship between the investment rate and our internal governance dummy variable as described above. Consistent with Pan et al.(2016), we use Investment rate as our proxy for the level of corporate investment, which is deﬁned as the sum of the capital expenditures and acquisitions at the end of the period divided by total assets at the beginning of the period. We calculate the investment rate variable from Compustat. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 105

Our multivariate regressions will include other control variables as well. All

financial statement variables are obtained from Compustat. The first control variable is firm size, and to mitigate any skewness issues, we take the natural logarithm of the total assets (Log(assets)). In order to take care of any non-linear relationships with firm size, we also include the squared term (Log(assets))2. One might expect that the agency costs of the firm increases with leverage (Green and Tal- mor 1986; Jensen and Meckling 1976), and consequently, we include the variable

Leverage deﬁned as the sum of long-term debt plus short-term debt in current liabilities divided by beginning period total assets. We also control for external governance as characterized by board characteristics, which is hypothesized to play an important role of constraining the discretionary power of CEO and is a potential substitute for internal governance. In order to do so, we collect data from

ISS on board characteristics such as the number of directors and the percentage of outside directors on the board. We hence merge Execucomp, Compustat and

ISS to construct our sample. Since the data in ISS legacy database starts in 1996, our sample starts from 1996. Table 1 summarizes the deﬁnitions of each of our variables used.

The final sample consists of 29,323 firm-year observations. For the purpose of this empirical study, we omit any observations from the sample if we cannot construct an internal governance measure (δ) for the firm. The sample spans

fiscal years 1996 to 2017, covers 3,529 CEO turnovers, and 3,343 distinct firms, for a total of 6,612 unique CEO-firm combinations. Note that in some cases, CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 106 a company can have multiple turnover events in one fiscal year, which usually involves at least one interim CEO who occupies the office for several months.

We ignore the interim CEOs in the transition and only count one turnover event with the initial predecessor and the final official successor who took the helm for more than one year, since we only have fiscal annual data for internal governance measure (δ)6. Detailed summary statistics of our sample are shown in Table 2.

Speciﬁcally, the average fraction of corporate titles of the CEO is 0.261, which is

7% greater compared to that in Aggarwal et al.(2017). The increase in sample means may either reflect the structural changes since the 2008 financial crisis or the result of our improvement on the specification of internal governance measure (δ) using the regular expression regex technique (see the Appendix for more details).

The sample distribution of δ is quite symmetric with extreme values ranging from smallest 0.055 to largest 0.643, 1% values of 0.111, 99% values of 0.428, median values of 0.25 which is very close to average values of 0.26. With reference to both internal governance and other corporate ﬁnancial variables, we have roughly similar means, medians and standard deviations to those in Pan et al.(2016), and

Aggarwal et al.(2017). Additionally, we ﬁnd that the variation of δ is more due to the variation of responsibility of non-CEO executives, indicating our internal governance variable is not simply the inverse of the deﬁnition of CEO power used by Adams et al.(2005).

6 Our turnover sample size is somewhat smaller than Pan, Wang and Weisbach (2016). One reason is that our paper does not include interim CEOs. We do this because the tenure of interim CEOs is usually less than a year. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 107

4 Empirical Results

4.1 Proxy for Internal Governance

We begin by estimating the relationship between internal governance and firm performance for each firm-year observation, utilizing the quadratic model specification introduced by Aggarwal et al.(2017). The panel regressions employ firm and year fixed-effects as described in Equation (1). The regressions use our entire sample, including non-transition years. The theory of Acharya et al.(2011) suggests that internal governance works best to motivate the older myopic manager’s under-investment problem.

We split the sample into ﬁrms wherein the CEO’s age is below the median CEO age of 56 years, and where the CEO’s age is greater than the median CEO age of

56 years. Panel A of Table 3, presents the results for younger CEOs, and Panel

B presents the results for the older myopic CEOs. According to the theory, we would expect to find that the coefficient on the linear δ to be significantly positive and the coefficient on the square term should be significantly negative only when

CEOs are myopic. Following the empirical speciﬁcations of Aggarwal et al.(2017), age is considered as an important variable to measure the executive horizons of

CEOs. As an originally far-sighted CEO become older, her executive horizons will naturally become shorter and the executive may turn myopic. Assuming CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 108 the original population of CEOs are a combination of far-sighted and myopic executives due to diﬀerent preferences and career plans, splitting the sample by

CEO age might give us a better sample of myopic executives.

We run regressions wherein the dependent variables are either industry-adjusted

M/B or industry-adjusted ROA. We also include as independent variables δ, δ2 and the lagged control variables firm size ((Log(assets)), (Log(assets))2), board characteristics (Dirsum and Outpercn), leverage (Leverage) and R&D (R&D), respectively. In Panel A (the sub-sample of younger CEOs), we find no statistically significant relationship between firm performance and the coefficients of

δ and δ2 . In Panel B (the sub-sample of older CEOs), there is a statistically signiﬁcant positive relationship between ﬁrm performance and δ , followed by a statistically negative relationship with δ2 . These results are consistent with the theory of Acharya et al.(2011) and with the empirical results of Aggarwal et al.

(2017).

For the next section, we need to define an optimal region wherein there is effective internal governance. Using the estimates reported in column (4), we find that the optimal level of internal governance measure is 0.33. Next, we set the range of effective internal governance as 0.33 plus and minus half of the standard deviation of δ in sample. In all our subsequent tests, firms of the outgoing CEO that have a δ within this range will be deemed as having good internal governance.

The empirical results indicate that internal governance is most important when CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 109 the CEO ages. We also perform the sensitivity of these results for age above 56, and our results are generally analogous. (These results not formally reported but are available from the authors.)

4.2 Impact of Internal Governance on a CEO’s Investment

Cycle

We perform univariate analysis to examine the trend of Investment rate during the transition period of CEO in Table 4. We examine the change in the investment rate for the entire sample to check whether we have similar results to those in

Pan et al.(2016). The change in the investment rate is the diﬀerence between the investment rate two years prior to the turnover and year t, where t = -1, 0, 1 and

2. Table 4 and its accompanying ﬁgure summarizes the univariate results. For the entire sample, we observe a decreasing investment rate from t = -2 to t = 0 and

1. This result is consistent with the results reported by Pan et al.(2016).

Table 5 and its accompanying figure summarizes the change in the investment rate if the age of the outgoing CEO two years prior to leaving the firm is 56 or above and the firm has good internal governance. The results indicate that we no longer see a decreasing investment rate from t = -2 to t = 0 and 1 if the CEO optimally shares executive duties with her management team. In the case of bad internal governance, the under-investment problem still occurs. These empirical findings confirm that internal governance ameliorates the under-investment concerns in CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 110 myopic CEOs as raised by Landier et al.(2009) and Acharya et al.(2011).

While both Tables 4 and 5 shows that internal governance reduces the under- investment problem in the transition period for myopic CEOs only, we further examine if these results are robust to the inclusion of other ﬁrm-speciﬁc variables.

In order to do so, we used lagged controls for industry-adjusted M/B7, logarithm of the firm’s total assets, firm leverage, R&D, number of directors and the percentage of outside directors. Note that in these regressions a significantly negative regression coefficient on IG implies that good internal governance mitigates the under-investment problem. We also include firm and year fixed-effects in the following specification, and the standard errors are two-way clustered, by firm and year.

0 ∆InvestmentRate = β0 + β1IG + β2xit 1 + γi + λt + εit (3) −

In order to examine the variation of investment policy during the transition period, we construct the econometric model in which the dependent variable is the change of investment rate between two years prior to turnover year (Table 6) and from the year after turnover to year 2 (Table 7). For Table 6, the sample for columns (1) and (2) are those ﬁrms with CEOs whose age was less than 56 two years prior to leaving, whereas the sample for columns (3) and (4) are those ﬁrms with CEOs whose age was at least 56 two years prior to leaving. For Table 7, the

7 We include the lagged industry adjusted market to book variable as a control variable to account for any persistence in the performance measure. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 111 sample for columns (1) and (2) are those ﬁrms with incoming CEOs whose age was less than 56 one year after turnover, whereas the sample for columns (3) and

(4) are those firms with incoming CEOs whose age was at least 56 one year after turnover. The internal governance variable, IG, is significantly negative, indicating the general trend of decreasing investment rate for the two years prior to the departure of the older CEO is greatly mitigated. In contrast, internal governance seems to be much less important for younger CEOs and for new CEOs. Pan et al.(2016) suggest that the CEO agency problems are reduced with the arrival of a new CEO, who does not enjoy the same private benefits as her predecessor.

Such ﬁndings are also consistent with the theoretical development of Acharya et al.(2011) that the internal governance mechanism should be eﬀective only if the

CEO is myopic and has a short executive horizon. Additionally, for all departing executives, firms tend to decrease the investment rate prior to departure as leverage of the firm increases, perhaps to conserve resources due to the increased fixed obligation responsibility. Moreover, younger incoming CEOs tend not to increase investment rate during their first year of tenure as leverage increases.

4.3 Impact of Internal Governance on the Proﬁtability of

Asset Divestitures

The flip side to understanding the under-investment problem is that CEOs might dispose of assets. We use as our dependent variable property sales (Compustat CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 112 item SPPE) that is scaled by beginning period assets and internal governance defined for the outgoing CEO. All other control variables are same as before and the regression results are given in Table 8. In column (4) of Table 8, the internal governance variable, IG, is significantly positive, indicating that the incoming

CEO tends to incur more divestitures at the beginning of her tenure given a myopic predecessor under good internal governance.

There are two possible explanations for this seemingly controversial result. One is that the new CEO recognizes that the old CEO made poor investment decisions and the new CEO is correcting the course of the firm. If this were the case, we would expect that the firm would recognize losses upon the disposal of these assets. Second, it is possible that the outgoing CEO made appropriate acquisitions during her tenure, but the asset mix does not match well with the skill set of the incoming CEO. If this were the case, then the disposed assets should not incur any loss and perhaps even a gain. Clearly, if the empirical evidence is in support of the second explanation, the beneficial effect of good internal governance on investment policy is further verified in that it not only mitigates the underinvestment problem but also improves the quality of acquired assets. To examine this possibility, we regress the gains/losses of property sales (Compustat item SPPIV) that is scaled by the beginning period assets. Note that when SPPIV is a positive number, it represents the losses incurred from an asset disposal. The regressions results are presented in Table 9. The coefficient on IG is only statistically negative for the older CEOs who depart the firm and not for younger predecessor CEOs whose CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 113 executive horizon is considered long. This indicates that good internal governance reduces the probability that the new CEO is disposing assets at a loss. Taken together, the empirical evidence suggests that good internal governance improves the deteriorating investment policy preceding myopic CEO departure in terms of both dollar amount and quality of assets acquired, and the asset disposals incurred at the beginning of a CEO’s tenure are likely due to mismatch between old asset mix and the skill set of the new CEO.

4.4 Robustness Tests

It is reasonable that our results are influenced by CEO turnover being endoge- nously determined. For instance, without controlling the specific causes of CEO dismissals, it is possible that rather than serving as the antecedent of investment policy changes, CEO turnover is a consequence of unsatisfying investment decisions preceding the eventually realized transition to the new CEO. Therefore, when we focus our empirical analysis on the CEO transition period, the statistical inference about the effect of internal governance upon firm investment policy could be distorted given the correlation between the internal governance measure and CEO turnover, and the reverse casualty between investment policy and CEO turnover.

One way to solve the potential endogeneity issue is to utilize sudden CEO deaths. As a sudden loss and a natural experimental shock to the ﬁrm and man- CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 114 agement team, the announcement of executive death should in itself elicit distur- bances on the strategic implementation of the investment policy and is exogenous to our explanatory variables of interest. We collect data of news report via Fac- tiva ranging from 1988 to 2018 and identify sudden death event based on the news content (see for example, Worrell, Davidson III, Chandy, and Garrison(1986)).

Speciﬁcally, we regard CEO death due to accident, sudden health catastrophe and death owed to unlikely fatal illness as surprise events. We use R API of SEC

Edgar to process 10-K and DEF-14 ﬁlings for titles of CEO and her subordinates who were employed early than 1992 and are not covered by Execucomp. We also only examined CEOs who two-years before death were older than or equal to 56 years. We found 17 such cases. Due to the limited sample, we restrict our control variables to those related to investment, such as logarithmic total assets, leverage ratio and R&D expenditures.

The regression results of the relationship between investment policy changes and internal governance in face of top executive sudden death is in Table 10. The main dependent variable is the change of investment rate from 2 years before the event to the end of the ﬁscal year in which the death is announced. Addi- tionally, utilizing the exogenous nature of the natural experiment sample, we run regressions on the two components of investment, namely capital expenditures

(CAP X) and acquisitions, to further reveal the channels through which good internal governance mitigate the cyclical turbulences of investment policy. We use the robust estimator to run the econometric model, which accounts for the CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 115 heterogeneity of outliners and is by design, the most efficient estimator for small sample analysis. In column 1, the coefficient of good internal governance dummy is negative and significant at five-percent based on small sample t-statistics. In column (2), the statistical significance of the coefficient of good internal governance dummy is at one-percent, indicative of a more salient effect of good internal governance on acquisitions in particular. Taken together in column (3), we estimate the robust regressions on the investment policy as a whole and the revealed negative relationship between internal governance and change in investment rates for older myopic CEOs is extremely significant at one-percent with a t-statistic around 10. Notwithstanding the limited sample size, these results indicate that internal governance helps reduce the investment problem for older myopic CEOs when turnover is assumed to be exogenous (columns (3) and (4) of Table 6), and where turnover is exogenous (Table 10).

The second robustness test controls for performance-related turnover, hiring of outsider CEO and CEO pay-performance sensitivities. One might expect reverse causality between CEO turnover and internal governance due to poor ﬁrm performance. Therefore we restrict our sample to voluntary turnover by excluding any performance-related turnover. In order to distinguish between the events of voluntarily turnover and forced turnover, we use the procedure of Parrino(1997).

According to Parrino(1997), forced departure of CEO can be identified through three steps. First, forced departure is identified if the public business news release explicitly announces that the individual leaves office due to forced termination CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 116 of contract, policy differences or any other reasons related to firm’s activities.

Secondly, if the direct messages of termination are not disclosed publicly, we believe, with one exception as noted below, that the individuals, who are above age

60, leave oﬃce voluntarily due to normal retirement. The classiﬁcation of individuals, who are under age 60, is subject to the following two criterions. Those turnovers are considered as forced if the public information either do not disclose reasons such as death, poor health or the acceptance of another position

(inside the firm or elsewhere, including governmental agency), or do not release the message of retirement at least six months before the succession. Finally, the cases of forced departure classified in the second procedure are reinvestigated using extensive information surrounding the transition period and records in the individual’s biography. The turnover is classified as voluntary if it turns out that the individual takes a comparable position elsewhere or departs for the ex-ante undisclosed reasons unrelated to firm’s policy and performance, such as personal interest confirmed by the departing CEO biography or subsequent press release.

The CEO’s age and the date of departure can be obtained from Execucomp, which also provides fragmentary information about reasons of departure for references.

The business news and reports from various sources such as newspaper, journal articles, company ﬁnancial reports, analyst reviews and etc., were searched using the database of Factiva. The extensive information such as CEO biography, awards and company history are obtained via internet resources from Bloomberg

Executive Proﬁle & Biography, Wikipedia, and SEC ﬁlings. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 117

Additionally, there might be a positive correlation between the need to hire an outside successor and the extent of shared responsibility among the executive team. If the executive team is dominated by the CEO, the board of directors may very well want to hire an outsider to correct the imbalance. Since the outsider is more likely to change ﬁrm policies (see, for example, Cragun, Nyberg, and

Wright(2016)), we will find that the change of investment policy is greater for outside CEO replacements and (poorly internally governed) firms. Alternatively, if the firm is beginning to formulate a CEO succession plan, the board of directors would want greater coordination between the retiring CEO and her executive team so that the board can evaluate potential successor from within. This would result in more shared responsibility between the CEO and her executive team and a less drastic change in investment policy during the transition period. To reduce these concerns, in Table 11 we re-estimated our basic regression equation by including a dummy variable if CEO successor is an outsider. We define a new

CEO as an outsider if the new CEO had less than two years of employment as an executive in the ﬁrm before being appointed (as opposed to serving on the board of directors). We do not count service on the board because as a board member, she may have wanted to change the direction of the ﬁrm. We also include the pay for performance (PPS) metric as developed by Core and Guay(1999) as an additional control governance variable. Note that it cuts the sample size in half.

The results are summarized in Table 11. Note that in column (1), IG remains negatively related to the change in investments before turnover. This is consistent CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 118 with the results in Table 6. We also ﬁnd that IG remains positively related to the amount and proﬁtability of assets divested, which is consistent with the results in Tables 8 and 9. Note that the CEO’s pay-performance sensitivity has no impact on investment and divestures. These results suggest that our main

ﬁnding on investment and divestures are not impacted by performance-related turnover, hiring of outsider CEO and CEO pay-performance sensitivities.

The third robustness test examine if the CEO pay slice captures the internal governance task-oriented argument of Acharya et al.(2011). Although, CEO pay slice has been used to capture tournament incentives (Kale et al. 2009, or CEO entrenchment (Bebchuk et al. 2011; Feng et al. 2011), we examine if we find a non-linear relationship with firm-performance. In Table 12, we find CEO pay slice is not related to firm performance in quadratic form for older CEOs (who theory would say are myopic and need subordinate discipline). We also find that the non-linear relationship for δ remains in older CEOs when we include CEO pay slice. We also find that CEO pay slice has a low correlation of 0.062 with δ.

It is possible that our internal governance variable is proxying for CEO power.

In particular, one might expect that sharing responsibility is inversely related to

CEO power. If the CEO is suﬃciently powerful, then decision making is centralized and ex post performance should be much more variable (Adams et al.

(2005)). Consequently, we will include an additional control variable that is a proxy for CEO power to see if our results are aﬀected by such inclusion. Accord- CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 119 ing to Finkelstein(1992), there are four dimensions of CEO power that are inter- related: structural power, ownership power, expert power and prestige power. The degree of responsibility sharing mainly measures the structural power and is positively correlated with other aspects of executive powers. Consequently, we deﬁne

CEO P ower as the sum of a sequence of dummies that identiﬁes the other three aspects of CEO power: prestige power, expertise power and ownership power.

The P restige P ower dummy takes a value of one if the CEO is also the founder of the ﬁrm. Expertise P ower is a dummy variable that takes the value of 1 if the number of the segments of business of the ﬁrm is higher than industry median by 2-digit SIC. Ownership P ower is the sum of two dummy additional variables:

Share takes the value of 1 if the percentage of shares owned by the CEO is higher than industry median; and P ay takes the value of 1 if the pay slice of the CEO is higher than industry median.

Additionally, one might expect that the greater is the complexity of organization, the more likely the CEO needs help in running the firm and more likely she is willing to share responsibility with her subordinates. We have already included organization complexity based upon the number of business segments as part of our CEO P ower control variable. Hence, we also include a control variable, Complexity, analogous to that used by Cheng et al.(2016) for geographical complexity. We define the complexity dummy is equal to one for firm-year observations with above the median industry first principle component of the following three variables: the number of geographical segments, geographical sales concen- CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 120 tration, and the percentage of foreign sales. The methodology we use to compute the geographical sales concentration and percentage of foreign sales is similar to the methodology employed to construct the Herfindahl Index.

Tables 13 and 14 report the regression results when we include the additional variable of CEO P ower and Complexity. For Table 13, the sample for columns

(1) through (3) are those firms with CEOs whose age was less than 56 two years prior to leaving, whereas the sample for columns (4) through (6) are those firms with CEOs whose age was at least 56 two years prior to leaving. As in Table 6, we see that the coefficient for IG is significantly negative, implying that with optimal internal governance, the general trend of decreasing investment rate for the two years prior to the departure of the older CEO is greatly mitigated. In contrast, internal governance seems to be much less important for younger CEOs. Hence, our results are not affected by the inclusion of the CEO P ower and Complexity variables. Interestingly, CEO P ower control variable is significantly negative only for the younger CEOs, for whom internal governance is hardly effective. This result indicates that two years prior to departure, powerful younger CEOs do not reduce the investment rate. The rationale might be that younger CEOs have sufficiently longer investment horizon and therefore their career concerns prevent them from being myopic. Finally, for all departing CEOs, regardless of age, the investment rate decreases with leverage, indicating that these CEOs try to conserve resources in the face of increasing financial obligations. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 121

Table 14 summarizes the determinants of the investment rate for the ﬁrst full year of the new CEO. The sample for columns (1) through (3) are those

ﬁrms with incoming CEOs whose age was less than 56 one year after turnover, whereas the sample for columns (4) through (6) are those ﬁrms with incoming

CEOs whose age was at least 56 one year after turnover. As is summarized in Table

7, internal governance seems to be much less important for new CEOs. This result is consistent with Pan et al.(2016) who suggest that the CEO agency problems are reduced with the arrival of a new CEO because she may not enjoy the same private benefits as her predecessor. In line with the findings of Table 13, powerful younger successors tend to resume the normal level of investment more effectively, illustrating the leadership role of CEOs in investment policy when they are in power and with right executive horizon. We also find that the investment rate for younger CEOs decreases with both the leverage ratio and the rate of investment in R&D. Again, the complexity variable is not significant. Such findings of Tables

13 and 14 are again consistent with the theoretical development of Acharya et al.(2011) that the internal governance mechanism should be eﬀective only if the

CEO is myopic and of short executive horizon.

5 Conclusions

Weisbach(1995) finds that a year or two before the CEO retires, the firm experiences a decrease in total investment. He suggests that this decrease reflects the CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 122 agency costs of the replaced CEO – who is sacrificing future cash flows in order to realize immediate increased cash flows. Pan et al.(2016) argue that older CEOs exhibit agency problems due to private benefits that accrue over her tenure. They

ﬁnd evidence of a CEO’s investment cycle, in which disinvestment decreases over a CEO’s tenure, whereas investment increases – leading to diﬀerences in growth in assets and employment over the CEO’s tenure. A number of papers (for example,

Acharya et al.(2011)) have theorized that the power dynamics between senior executives and the CEO might help the CEO make shareholder wealth-maximizing decisions. Acharya et al.(2011) theorize that internal governance may mitigate the

CEO horizon problem. The potential CEO successor of the company, they argue, should have a longer horizon than the current CEO. If the current CEO puts in place policies that destroy the capital and reputational stock of the firm, then the successor will find herself when named CEO running a diminished firm. Accord- ingly, lower level managers who hope to succeed the current CEO will oppose CEO policies that do not maximize the value of the firm. The current CEO is “forced” to abide by the wishes of the lower management, since the current CEO needs the assistance of the lower level manager to produce current earnings that support the current stock price, which is an important factor in the level of the CEO’s current compensation. On the other hand, if most administrative duties regarding the

firm’s business operation are conducted by the subordinates, and the success of the company is so dependent on the effort exerted by the subordinates, then no one is in charge in coordinating the activities of the subordinate to ensure value CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 123 maximization. Consequently, the optimal internal governance should be such that the responsibilities of the running the firm should be shared as opposed to be run solely by the CEO or only by key top management subordinates.

Accordingly, this paper examines the impact of internal governance on the

CEO’s investment cycle. We ﬁnd that good internal governance helps reduce older CEOs under-investing before their exit, whereas bad governance does not.

For younger CEOs, we do not find any relationship between internal governance and investment. We also find that new incoming CEOs divest these projects profitably. Such findings are consistent with the theoretical development of Acharya et al.(2011) that the internal governance mechanism should be effective only if the CEO is myopic and has a short executive horizon. Our results compliment the

findings of Cheng et al.(2016) who find that the extent of real earnings management decreases with key subordinate executives’ horizon and influence and that of

Jain et al.(2016) who ﬁnd that ﬁrms with better internal governance have lower information asymmetry and higher liquid. These results are robust to: normal

CEO retirements (exclude performance-related turnover), sudden CEO deaths, and controlling for board size, proportion of outsiders on the board, CEO’s pay- performance sensitivity, CEO pay slice, CEO power, ﬁrm complexity and if the

CEO was an outsider or not. Future research might examine if internal governance has an impact on merger returns and managerial disclosure policies. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 124

References

Acharya, V. V., Myers, S. C., & Rajan, R. G. (2011). The internal governance of

ﬁrms. The Journal of Finance, 66 (3), 689–720.

Adams, R. B., Almeida, H., & Ferreira, D. (2005). Powerful CEOs and their

impact on corporate performance. The Review of Financial Studies, 18 (4),

1403–1432.

Aggarwal, R. K., Fu, H., & Pan, Y. (2017). An empirical investigation of internal

governance.

Baker, M., & Wurgler, J. (2013). Behavioral corporate ﬁnance: An updated

survey. In Handbook of the Economics of Finance (Vol. 2, pp. 357–424).

Elsevier.

Bebchuk, L. A., Cremers, K. M., & Peyer, U. C. (2011). The ceo pay slice. Journal

of ﬁnancial Economics, 102 (1), 199–221.

Bertrand, M., & Mullainathan, S. (2001). Are CEOs rewarded for luck? The

ones without principals are. The Quarterly Journal of Economics, 116 (3),

901–932.

Brickley, J. A., Linck, J. S., & Coles, J. L. (1999). What happens to CEOs after

they retire? New evidence on career concerns, horizon problems, and CEO

incentives. Journal of Financial Economics, 52 (3), 341–377.

Chakravarthy, B. S. (1986). Measuring strategic performance. Strategic manage-

ment journal, 7 (5), 437–458. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 125

Cheng, Q., Lee, J., & Shevlin, T. (2016). Internal governance and real earnings

management. The Accounting Review, 91 (4), 1051–1085.

Core, J., & Guay, W. (1999). The use of equity grants to manage optimal equity

incentive levels. Journal of accounting and economics, 28 (2), 151–184.

Coughlan, A. T., & Schmidt, R. M. (1985). Executive compensation, management

turnover, and ﬁrm performance: An empirical investigation. Journal of

accounting and economics, 7 (1-3), 43–66.

Cragun, O. R., Nyberg, A. J., & Wright, P. M. (2016). Ceo succession: what we

know and where to go? Journal of Organizational Eﬀectiveness: People and

Performance.

Dechow, P. M., & Sloan, R. G. (1991). Executive incentives and the horizon

problem: An empirical investigation. Journal of accounting and Economics,

14 (1), 51–89.

Elliott, J. A., & Shaw, W. H. (1988). Write-oﬀs as accounting procedures to

manage perceptions. Journal of accounting research, 91–119.

Fama, E. F., & Jensen, M. C. (1983a). Agency problems and residual claims. The

journal of law and Economics, 26 (2), 327–349.

Fama, E. F., & Jensen, M. C. (1983b). Separation of ownership and control. The

journal of law and Economics, 26 (2), 301–325.

Feng, M., Ge, W., Luo, S., & Shevlin, T. (2011). Why do cfos become involved in

material accounting manipulations? Journal of accounting and economics,

51 (1-2), 21–36. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 126

Finkelstein, S. (1992). Power in top management teams: Dimensions, measure-

ment, and validation. Academy of Management journal, 35 (3), 505–538.

Gibbons, R., & Murphy, K. J. (1992). Optimal incentive contracts in the presence

of career concerns: Theory and evidence. Journal of political Economy,

100 (3), 468–505.

Green, R. C., & Talmor, E. (1986). Asset substitution and the agencycosts of

debt ﬁnancing. Journal of Banking & Finance, 10 (3), 391–399.

Heaton, J. B. (2002). Managerial optimism and corporate ﬁnance. Financial

management, 33–45.

Huson, M. R., Malatesta, P. H., & Parrino, R. (2004). Managerial succession and

ﬁrm performance. Journal of Financial Economics, 74 (2), 237–275.

Jain, P., Jiang, C., & Mekhaimer, M. (2016). Executives’ horizon, internal gover-

nance and stock market liquidity. Journal of Corporate Finance, 40 , 1–23.

Jensen, M. C. (2000). Value maximization and the corporate objective function.

Jensen, M. C., & Meckling, W. H. (1976). Theory of the ﬁrm: Managerial behavior,

agency costs and ownership structure. Journal of ﬁnancial economics, 3 (4),

305–360.

Kale, J. R., Reis, E., & Venkateswaran, A. (2009). Rank-order tournaments

and incentive alignment: The eﬀect on ﬁrm performance. The Journal of

Finance, 64 (3), 1479–1512.

Landier, A., Sauvagnat, J., Sraer, D., & Thesmar, D. (2012). Bottom-up corporate

governance. Review of Finance, 17 (1), 161–201. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 127

Landier, A., Sraer, D., & Thesmar, D. (2009). Optimal dissent in organizations.

The Review of Economic Studies, 76 (2), 761–794.

Malmendier, U., & Tate, G. (2005). Ceo overconﬁdence and corporate investment.

The journal of ﬁnance, 60 (6), 2661–2700.

Matˇejka, M., Merchant, K. A., & Van der Stede, W. A. (2009). Employment

horizon and the choice of performance measures: Empirical evidence from

annual bonus plans of loss-making entities. Management Science, 55 (6),

890–905.

Murphy, K. J., & Zimmerman, J. L. (1993). Financial performance surrounding

CEO turnover. Journal of Accounting and Economics, 16 (1-3), 273–315.

Palia, D. (2000). The impact of regulation on ceo labor markets. The Rand journal

of economics, 165–179.

Pan, Y., Wang, T. Y., & Weisbach, M. S. (2016). CEO investment cycles. The

Review of Financial Studies, 29 (11), 2955–2999.

Parrino, R. (1997). CEO turnover and outside succession a cross-sectional analysis.

Journal of ﬁnancial Economics, 46 (2), 165–197.

Strong, J. S., & Meyer, J. R. (1987). Asset writedowns: Managerial incentives

and security returns. The Journal of Finance, 42 (3), 643–661.

Warner, J. B., Watts, R. L., & Wruck, K. H. (1988). Stock prices and top

management changes. Journal of ﬁnancial Economics, 20 , 461–492.

Weisbach, M. S. (1988). Outside directors and ceo turnover. Journal of ﬁnancial

Economics, 20 , 431–460. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 128

Weisbach, M. S. (1995). Ceo turnover and the ﬁrm’s investment decisions. Journal

of Financial Economics, 37 (2), 159–188.

Worrell, D. L., Davidson III, W. N., Chandy, P., & Garrison, S. L. (1986). Manage-

ment turnover through deaths of key executives: Eﬀects on investor wealth.

Academy of Management Journal, 29 (4), 674–694. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 129

Appendix: Construction of the Internal Gover- nance Measure

As implemented in Aggarwal et al.(2017), δ is calculated by the number of executive titles of CEO scaled by the total number of executive titles carried by the entire top management team of five executives. We split the title string of each executives using four delimiters or words: 1) “,”, 2) “;”, 3) “&”, 4) “and”. Then the number of pieces split from the title string is the number of corporate titles held by the executive. Furthermore, as is done by Aggarwal et al.(2017), we eliminate terms such as “R&D”, “, LLC”, “, U.S.”, and etc., which can cause a bias in counting the titles. However, according to the result of our manual checking, the above data processing procedures still are incapable of generating a clean measure of CEO’s fraction of corporate titles, leading to serious measurement errors and misspecifications. For instance, in fiscal year 2004, the executive title of Mark

McDonald of AAR Corp. is recorded as “group vp-structures & systems, mainte- nance, repair and overhaul”. Rather than mechanically apply the aforementioned method and split the title string into 5 parts, we should in fact consider this manager as the group vice president of a certain division, and with only 1 corporate title. As illustrated in this example, when dealing with our comprehensive sample from 1996 to 2017 with a wide variety of records in the title ﬁeld, the above method would result in quadrupling or quintupling number of titles, resulting in a CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 130 substantial bias on the primary proxy. Faced with such an empirical challenge, we develop a more intelligent framework of title screening, utilizing the features and functions associated with regular expression built in R for string manipulation.

Regular expression or regex, is a special string representation for abstracting and describing a certain common pattern of multiple strings. R, as a powerful statistical computing language, provides us with a suﬃcient capability of string manipulation using regex and thus is chosen as the primary environment to build the screening system in our paper. Based on intensive experiment, sampling and manual checking, we recognize ﬁve most common patterns, which are the building blocks for the more complex titles triggering the miss-counting: 1) “of . . . and

. . . ”, 2) “of. . . ,. . . and”, 3) “ of . . . and . . . and”, 4) “of . . . and . . . of”, 5) “. . . and . . . oﬃcer or head”. Each of above regex describes one pattern in a set of title strings, resulting in a particular type of problem and requiring certain treatments.

The first regex is to identify the title strings in which the string contains “of” following with at least one “and”. For instance, in fiscal year 2003, Alan J. Black of GREAT ELM CAPITAL GROUP INC carried executive title recorded as “senior vp; managing director of Europe, Middle East and Africa”, which is clearly captured by the first pattern. To fix the problem, we need to know whether or not, or if so, how many commas or/and “and” appear in that structure. Thus, we need further utilize the second and third regex to figure out the detailed composition of the first regular expression. Given the fact that the regular pattern of title string in the above example only contains one comma between “of” and “and”, CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 131 the correct number of titles can be calculated programmatically as the number of split parts minus the sum of one and the number of commas, generating the result of 2 titles. Similar to the second and third regular patterns, the fourth pattern is also closely associated with the first regex. The fourth regex flags titles such as “Chairman, Chief Executive Officer, President, Chairman of American Airlines

Inc, Chief Executive Oﬃcer of American Airlines Inc and President of American

Airlines Inc”, held by Gerard J. Arpey of AMERICAN AIRLINES GROUP INC, in ﬁscal year 2009. The fourth pattern identiﬁes title strings in which the word

“and” connects multiple independent corporate titles, such as “Chief Executive

Oﬃcer of American Airlines Inc” and “President of American Airlines Inc.” Thus, in face of this situation, we should stick with the method of pure delimiters which implies that the CEO had 6 distinct titles. The last regular expression captures the corporate titles whose name contains the word “and” or symbol “&”. For example, “executive vp, general counsel, chief ethics & compliance oﬃcer” held by Paul R. Shlanta of SOUTHERN CO GAS in 2005 falls in this last category.

The ﬁfth level regex adjusts the word “and” or ”&” in the title of “chief ethics

& compliance officer” as one corporate title. Thus, working with the above five regular expressions, we could develop the screening system which identifies all the problematic patterns and automatically fix the majority of the miscounting.

Although regex is useful in minimizing misspecification, some highly complex titles can be identified by the system but can’t be resolved programmatically, and CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 132 therefore relies on manual correction8. For instance, D. Bryan Jordan carries the executive title “Chairman, Chief Executive Officer, President, Member of Credit

Policy & Executive Committee, Member of Executive & Risk Committee, Chief

Executive Oﬃcer of First Tennessee Bank, President of First Tennessee Bank and

Director of First Tennessee Bank”, which is a mixture of patterns one, two, three and four. Furthermore, throughout the entire time frame from 1992 to 2017, we observe two distinct styles of recording CEO’s title in general. In early data, especially before 2000, the title ﬁeld of executives commonly uses symbols and abbreviations, such as using “-”, “&”, “oﬀr.” and “vp” to represent “of”, “and”,

“oﬃcer” and “vice president”; in most current data, especially after 2007, the title ﬁeld primarily use full words and expressions to record annual title. Our title screening framework is able to deal with two styles of records simultaneously. We also use the program to identify and eliminate individuals who only held advisory positions, membership of committees and nonexecutive titles such as chairman.

The speciﬁc deﬁnition, example and variation of the above mentioned regular expressions are addressed in details in the following table:

8 We still had to manually check around 1,000 titles CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 133 1 2 6 6 6 3 2 3 2 6 1 2 3 “Senior Vice PresidentCommercial of Operations” Proprietary Pharmaceutical Products and Global “managing director, headfinance of officer” accounting policy and controls & former chief “Chairman, Chief Executivelines Officer, Inc, President, Chief ChairmanAmerican Executive of Airlines Officer American Inc” of Air- American Airlines Inc and President of “Senior Vice President ofSecretary” Legal & Government Affairs, General Counsel and “President of Western Region HealthHealth Plan, Net Health of Net, California, Inc. Inc.” and President of “Executive Vice President of Legal, GovernmentPresident Affairs of and Sustainability Cliffs and China” “Chief Compliance Officer,Compliance Officer Senior of Vice Americancan Airlines President, Airlines Inc, Inc Senior General and Vice Counsel, General President“Chairman, Counsel of Chief Chief of Ameri- Executive American Airlines Officer,Service Inc” Company, President, Chief Executive Chairman Officer ofand of President Arizona Arizona of Public Public Arizona Service Public Company Service Company” “Senior Vice President oftration, Employment Law Government Affairs & & Litigation, Contracts Risk“Executive Adminis- Management Vice and President, General Counsel” Generaland Manager General of Manager Business Productment Line Operations & Management & Service & P&L Enablement” Design, Network Enable- “Executive Vice President, Secretary, Generaldent of Counsel, CNA Executive Insurance Vicepanies Companies, Presi- and General Secretary Council of of CNA CNA“Global Insurance Insurance Brand Companies” Com- President ofcessories” Luxury, Women’s Collections, and World of Ac- “Chief Technology Officer, Senior& Vice Technology and President Member of of Operations, Executive Engineering Council” Fiscal Year CEO name Company Name Title String # Title 199620042012 John P. Jones, III Gary F. Kennedy,2003 Esq. Carlos2013 AMERICAN Alban AIRLINES GROUP AIR INC PRODUCTS2017 & CHEMICALS INC Alan J. Black2006 Paul “senior H. vp, Grazewski “exec. general counsel v-p-gases & Thomas & chief P. equip.” Gibbons compliance ABBOTT2007 officer-AMR LABORATORIES and American” Susan AMERICAN L. SCIENCE Decker ENGINEERING GREAT BANK ELM OF CAPITAL GROUP NEW INC YORK Steven E. MELLON 3 “Senior Buller, CORP Vice CPA President of “Vice Product2009 Chairman Management, & Marketing “senior BLACKROCK CEO & INC vp;managing of ALTABA Strategy” INC director Clearing, of Markets Europe, and Middle Client Management” East and Africa” Gerard J. Arpey 1 2 AMERICAN AIRLINES GROUP INC 2 200020072014 “head Robert of R. advertiser Herb and Publisher Paul group R. &2012 Shlanta chief finance officer” David W. Meline 1 ADVANCED MICRO DEVICES D. Bruce Sewell SOUTHERN CO GAS AMGEN INC APPLE 2 “executive INC vp,2011 chief sales and marketing officer” Steven Jackson Sell “executive vp, general counsel & chief ethics HEALTH and NET compliance INC officer”2012 “Executive VP, CFO and Principal Financial & Accounting Officer” P. Kelly Tompkins 3 2 CLEVELAND-CLIFFS INC 3 2012 Guy H. Kerr2013 BELO CORP -SER Gary2016 A F. COM Kennedy, Esq. AMERICAN AIRLINES GROUP INC Donald E. Brandt, CPA PINNACLE WEST CAPITAL CORP “Executive Vice President of Law & Government and Secretary”20142017 Peter W. Quigley 2 Susan Louise Spradley KELLY SERVICES INC -CL A VIAVI SOLUTIONS INC 20132015 Jonathan David Kantor CNA FINANCIAL CORP Valrie Hermann2013 RALPH LAUREN CORP John J. Tracy BOEING CO [1] “of . . . and . . . ” [2] “of. . . [,]. . . and” [3] “ of . . . and . . . and” [4] “ of . . . and . . . of” [5] “. . . and . . . officer or head” [1]+[2]+[3] [1] + [2] + [4] [1]+[2]+[3]+[4] CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 134

Table 1: Variable Deﬁnitions This table reports the variables used in our empirical analysis and their deﬁnitions.

Variable Description

Industry-adjusted ROA is deﬁned as ROA minus the industry median level ROA. The median level is calculated ROA at the two-digit SIC industry level using the entire Com- pustat universe

The industry adjusted market-to-book ratio is deﬁned as the ﬁrm’s market-to-book ratio minus the industry’s me- M/B dian market-to-book ratio . The median level is calculated at the two-digit SIC industry-year using the Com- pustat universe

Denotes the fraction of executive titles held by the CEO and proxies for the relative contribution of the CEO to the entire cash flow of the firm. It is calculated as the number of executive titles of CEO scaled by the total δ number of titles carried by the top management team of top five managers, including the CEO. The number of titles is calculated using our screening method built upon regex.

Number of executive titles carried by the CEO including f chair and membership of board and executive committees.

Number of executive titles carried by the top four non- g CEO executives ranked by total compensation.

Indicator of effective internal governance, and is defined as a dummy variable that takes the value of one if δ falls within the optimal range. The optimal range is defined IG as optimal value minus a quarter of sample standard deviation of δ, and optimal value plus a quarter of sample standard deviation of δ.

Capital expenditures rate (capital expenditures/ begin- Investments ning of period assets) + acquisition rate (acquisitions/ beginning of period assets) CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 135

Sppe Property sales/ beginning of period assets

Gains or losses of property sales / beginning of period Sppiv assets

One-year lagged values of: (long term debt + debt in Leverage current liabilities)/ beginning of period assets

Size One-year lagged values of: natural logarithm of assets

One-year lagged values of: research and development ex- R&D penditures/ beginning of period assets

Directors Total number of directors serving on the board

Outsiders Percentage of outside directors serving on the board

Pay performance sensitivity measured as logarithmic PPS CEO total portfolio delta (in thousands).

Pay slice of total CEO compensation out of the total com- CPS pensation for the whole management team.

A new CEO is deﬁned as an outsider successor if she had less than two years of employment as an executive in the Outsider Successor ﬁrm before being appointed (as opposed to serving on the board of directors).

An index variable to proxy for the overall power of CEO is defined as the sum of a sequence of dummies that identifies three aspects of CEO power: prestige power, expertise power and ownership power. The P restige P ower dummy takes a value of one if the CEO is also the founder of the firm. Expertise P ower is a dummy that takes the CEO P ower value of 1 if the number of the segments of business of the firm is higher than industry median by 2-digit SIC. Ownership P ower is the sum of two dummy additional variables: Share takes the value of 1 if the percentage of shares owned by the CEO is higher than industry median; and P ay takes the value of 1 if the pay slice of the CEO is higher than industry median. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 136

A proxy for geographical complexity is defined as a dummy variable that is equal to one for firm-year observations with above the median industry first principle Complexity component of the following three variables: the number of geographical segments, geographical sales concentration, and the percentage of foreign sales. CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 137

Table 2: Descriptive Statistics This table reports the descriptive statistics of our sample for the period 1996 to 2017. See Table 1 for variable deﬁnitions.

Mean Median p25 p75 STD Skewness Kurtosis

δ 0.263 0.250 0.222 0.300 0.069 0.563 3.842 f 2.639 2 2 3 1.035 2.135 10.822 g 10.086 10 8 11 2.788 1.938 10.774

Age 55.624 56 51 60 7.108 0.244 3.779

CPS 0.387 0.391 0.321 0.455 0.120 -0.042 4.557

CEOP ower 1.169 1.000 1.000 2.000 0.861 0.296 2.522

Complexity 0.390 0.000 0.000 1.000 0.488 0.450 1.203

PPS 5.213 5.221 4.226 6.215 1.612 -0.075 4.441

ROA 0.062 0.034 -0.001 0.102 0.111 1.330 5.649

M/B 1.126 0.415 -0.197 1.585 2.787 2.825 16.458

Leverage 0.246 0.223 0.073 0.359 0.247 15.519 931.618

Size 7.776 7.670 6.547 8.904 1.727 0.350 3.228

R&D 0.028 0.000 0.000 0.029 0.066 6.750 103.122

Directors 0.717 0.778 0.600 0.875 0.196 -1.024 3.322

Outsiders 9.505 9.000 8.000 11.000 2.499 0.969 6.396

Investments 0.102 0.058 0.026 0.117 0.172 8.588 160.441

Sppe 0.004 0.000 0.000 0.002 0.033 75.747 8372.471

Sppiv -0.003 0.000 -0.001 0.000 0.048 -73.912 7332.092 CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 138

Table 3: Regressions of Firm Performance on Internal Governance for Younger and Older CEOs This table summarize the results of regressing firm performance variables (current year’s industry-adjusted ROA and market-to-book ratios) on the internal governance variable δ and other control variables. See Table 1 for variable definitions. Panel A presents the results for younger CEOs (defined as those whose age is less than the median CEO age of 56 years), and Panel B presents the results for older CEOs (defined as those whose age is greater than or equal to the median CEO age of 56 years. Our sample period is 1996 to 2017. t-statistics are given in parentheses and all standard errors are two-way clustered by firm and year. ***, ** and * denotes statistical significance at the 1%, 5% and 10% levels, respectively.

(1) (2) (3) (4) ROA ROA M/B M/B

Panel A: Younger CEOs

δ 0.104*** 0.085* -0.775 -0.947 (2.76) (1.90) (-0.57) (-0.62) δ2 -0.155** -0.115 1.946 2.175 (-2.43) (-1.52) (0.86) (0.85) ROA 0.498*** 0.507*** (34.91) (30.80) LaggedM/B 0.425*** 0.479*** (20.01) (17.55) Size -0.044*** -0.026*** -1.194*** -0.257 (6.15) (2.96) (3.55) (-0.61) Leverage 0.033*** 0.033*** -0.116 -0.304 (5.47) (4.38) (-0.38) (-0.95) R&D 0.045 0.127*** 0.469 1.875 (1.35) (2.59) (0.51) (1.41) Directors -0.000 -0.012 (-1.02) (-0.86) Outsiders 0.003 0.044 (0.71) (0.28) Year fixed-effects yes yes yes yes Firm fixed-effects yes yes yes yes Adj.R2 0.286 0.287 0.243 0.272 N 13461 9341 13543 9536 CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 139

(1) (2) (3) (4) ROA ROA M/B M/B

Panel B: Older CEOs

δ 0.083** 0.048 1.255 3.912*** (2.41) (1.26) (1.09) (2.87) δ2 -0.129** -0.055 -1.482 -6.079*** (-2.24) (-0.90) (-0.76) (-2.70) ROA 0.532*** 0.530*** (38.13) (32.57) LaggedM/B 0.492*** 0.532*** (19.99) (19.06) Size -0.021*** -0.013** -0.486*** -0.133 (-4.31) (-2.20) (-2.86) (-0.65) Size2 0.001*** 0.001* 0.003 -0.020* (3.74) (1.66) (0.29) (-1.69) Leverage 0.027*** 0.022*** -0.372** 0.243 (4.71) (2.93) (-2.28) (0.90) R&D 0.057 0.030 3.160** 3.553 (0.77) (0.62) (2.35) (1.55) Directors -0.000 -0.027** (-0.96) (-2.36) Outsiders -0.010** 0.305** (-2.33) (2.17)

Year fixed-effects yes yes yes yes Firm fixed-effects yes yes yes yes Adj.R2 0.301 0.299 0.282 0.330 N 13783 10276 14039 10588 CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 140

Table 4: Changes in Investment Rates Around CEO Turnover This table presents differences in the average investment rates surrounding CEO turnover. The year the incoming CEO leads the firm is designated as year zero. Time interval (in years) in which the test of difference is performed is indicated in the headers. Investments is defined in Table 1. The table presents means of differences in Investment 2 years prior to the turnover and year t, where t = -1, 0, 1 and 2. Our sample period is 1996 to 2017. t-statistics are given in parentheses. ***, ** and * denotes statistical significance at the 1%, 5% and 10% levels, respectively.

(-2,-1) (-2,0) (-2,1) (-2,2)

Investments 0.009* 0.016*** 0.013** 0.007 (1.71) (3.15) (2.46) (1.21)

Figure 1: CEO’s Investment Cycle Surrounding CEO Turnover CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 141

Table 5: Changes in Investment Rates Around CEO Turnover For Older/Younger CEOs and Firms with Good/Bad Internal Governance This table presents differences in the average investment rates surrounding CEO turnover. The year the incoming CEO leads the firm is designated as year zero. Time interval (in years) in which the test of difference is performed is indicated in the headers. CEO age is measured two years before CEO turnover, and Investments and IG are defined in Table 1. The table presents means of differences in Investment 2 years prior to the turnover and year t, where t = -1, 0, 1 and 2. Our sample period is 1996 to 2017. t-statistics are given in parentheses. ***, ** and * denotes statistical significance at the 1%, 5% and 10% levels, respectively.

(-2,-1) (-2,0) (-2,1) (-2,2)

Older CEOs and Good Internal Governance (IG=1)

Investment 0.003 0.019 0.005 0.012 (0.225) (1.592) (0.405) (1.117)

Younger CEOs and Bad Internal Governance (IG=0)

Investment 0.026** 0.039*** 0.030** 0.030** (2.069) (3.199) (2.223) (2.103)

Figure 2: CEO’s Investment Cycles Surrounding CEO Turnover by Age and Good/Bad Internal Governance CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 142

Table 6: Regressions of Changes in Investment Rates Before and in the Year of CEO Turnover (-2,0) on Internal Governance for Older/Younger CEOs This table reports the results of the regression whereby the dependent variable is the change in investments rate from two years before CEO turnover compared to the turnover year. The main explanatory variable is the Internal Governance dummy variable (IG) which equals one if the internal governance is within the optimum range as determined by Table 3, column (4) for the M/B performance variable. Younger (older) CEOs are those whose age two years before the CEO turnover year is less than (greater than or equal to) the median age of 56 years. All other variables are defined in Table 1. Our sample period is 1996 to 2017. t-statistics are given in parentheses and all standard errors are two-way clustered by firm and year. ***, ** and * denotes statistical significance at the 1%, 5% and 10% levels, respectively.

Change in Investments (-2,0)

Younger CEOs Older CEOs

(1) (2) (3) (4)

IG -0.044 0.036 -0.117*** -0.120*** (-1.25) (0.81) (-3.38) (-2.92) LaggedM/B 0.014** 0.004 0.001 0.002 (2.57) (0.74) (0.54) (0.80) Size 0.064 0.053 0.079*** 0.097** (1.30) (0.67) (2.64) (2.28) Leverage 0.299 0.814*** 0.323** 0.340*** (1.16) (2.81) (2.22) (2.89) R&D 0.213 -2.470 -0.264 0.711 (0.84) (-1.49) (-0.28) (1.01) Directors 0.049** -0.023** (2.17) (-2.36) Outsiders 0.316 0.147 (1.43) (1.28)

Year fixed-effects yes yes yes yes Firm fixed-effects yes yes yes yes Adj.R2 0.289 0.358 0.146 0.174 N 569 388 1024 754 CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 143

Table 7: Regressions of Changes in Investment Rates After CEO Turnover (1,+2) on Internal Governance for Older/Younger CEOs This table reports the results of the regression whereby the dependent variable is the change in investments rate from the year after CEO turnover to two years after the turnover year. The main explanatory variable is the Internal Governance dummy variable (IG) which equals one if the internal governance is within the optimum range as determined by Table 3, column (4) for the M/B performance variable. Younger (older) CEOs are those whose age one year after the CEO turnover year is less than (greater than or equal to) the median age of 56 years. All other variables are defined in Table 1. Our sample period is 1996 to 2017. t-statistics are given in parentheses and all standard errors are two-way clustered by firm and year. ***, ** and * denotes statistical significance at the 1%, 5% and 10% levels, respectively.

Change in Investments (1,+2)

Younger CEOs Older CEOs

(1) (2) (3) (4)

IG -0.026 -0.026 -0.014 -0.009 (-1.07) (-0.80) (-0.75) (-0.43) M/B -0.001 0.000 0.001 0.003 (-0.62) (0.01) (0.65) (1.10) Size 0.145*** 0.220*** 0.013 -0.015 (5.35) (5.18) (0.86) (-0.72) Leverage 0.179*** 0.267*** -0.010 -0.110 (3.29) (3.34) (-0.12) (-0.85) R&D 2.628*** 3.204*** -0.507 0.140 (2.63) (3.16) (-1.24) (0.53) Directors -0.002 0.001 (-0.33) (0.15) Outsiders -0.076 -0.093 (-0.79) (-1.61)

Year fixed-effects yes yes yes yes Firm fixed-effects yes yes yes yes Adj.R2 0.216 0.319 0.115 0.283 N 1223 865 631 459 CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 144

Table 8: Regressions of Property Sales in the Year of CEO Turnover on Internal Governance for Older/Younger CEOs This table reports the results of the regression whereby the dependent variable is the ratio of dollar property sales to beginning period assets (Sppe) in the year of CEO turnover. The main explanatory variable is the Internal Governance dummy variable (IG) which equals one if the internal governance is within the optimum range as determined by Table 3, column (4) for the M/B performance variable. Younger (older) CEOs are those whose age two years before the CEO turnover year is less than (greater than or equal to) the median age of 56 years. All other variables are defined in Table 1. Our sample period is 1996 to 2017. t-statistics are given in parentheses and all standard errors are two-way clustered by firm and year. ***, ** and * denotes statistical significance at the 1%, 5% and 10% levels, respectively.

Sppe at t=0

Younger CEOs Older CEOs

(1) (2) (3) (4)

IG -0.002 -0.001 0.014 0.023** (-0.93) (-0.69) (1.53) (2.09) M/B -0.000 -0.000* 0.001** 0.001 (-0.46) (-1.75) (2.13) (1.54) Size 0.004 0.001 0.002 0.004 (1.32) (0.33) (0.48) (0.60) Leverage 0.006 0.022** 0.035*** 0.050* (0.92) (2.45) (2.68) (1.68) R&D 0.011 0.024 -0.037 -0.008 (0.55) (0.91) (-0.57) (-0.10) Directors 0.000 -0.003 (0.16) (-1.53) Outsiders 0.040** -0.020 (2.57) (-1.22)

Year fixed-effects yes yes yes yes Firm fixed-effects yes yes yes yes Adj.R2 0.177 0.503 0.137 0.207 N 450 307 746 556 CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 145

Table 9: Regressions of Gains or Losses on Property Sales in the Year of CEO Turnover on Internal Governance for Older/Younger CEOs This table reports the results of the regression wherein the dependent variable is the ratio of dollar gains or losses on property sales to beginning period assets (Sppiv) in the year of CEO turnover. The main explanatory variable is the Internal Governance dummy variable (IG) which equals one if the internal governance is within the optimum range as determined by Table 3, column (4) for the M/B performance variable. Younger (older) CEOs are those whose age two years before the CEO turnover year is less than (greater than or equal to) the median age of 56 years. All other variables are defined in Table 1. Our sample period is 1996 to 2017. t-statistics are given in parentheses and all standard errors are two-way clustered by firm and year. ***, ** and * denotes statistical significance at the 1%, 5% and 10% levels, respectively.

Sppiv at t=0

Younger CEOs Older CEOs

(1) (2) (3) (4)

IG -0.007 -0.011 -0.004 -0.008** (-1.52) (-1.51) (-1.58) (-2.43) M/B 0.000 -0.000 -0.000 -0.000 (0.17) (-0.83) (-0.31) (-0.26) Size -0.001 -0.005 0.005*** 0.009*** (-0.41) (-1.26) (3.11) (3.36) Leverage -0.002 0.011 -0.005 -0.003 (-0.27) (0.61) (-0.88) (-0.25) R&D 0.006 0.104 0.049 -0.034 (0.52) (1.10) (0.90) (-0.57) Directors -0.000 -0.001 (-0.06) (-1.34) Outsiders 0.021 0.022* (0.72) (1.83)

Year fixed-effects yes yes yes yes Firm fixed-effects yes yes yes yes Adj.R2 0.160 0.222 0.057 0.094 N 565 389 1059 790 CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 146

Table 10 (Robustness Test 1): Regressions of Changes in Investment Rates Around Sudden CEO Deaths (-2, 0) For Older CEOs This table reports regression results of the relationship wherein the dependent variable is the difference in long term investment from two years before to the end of the fiscal year in which the sudden CEO death event is announced. The main explanatory variable is the Internal Governance dummy variable (IG) which equals one if the internal governance is within the optimum range as determined by Table 3, column (4) for the M/B performance variable. All other variables are defined in Table 1. Our sample period is 1996 to 2017 and only includes older CEOs (age greater than or equal to 56 years two years before their sudden death is announced) in 1996. t-statistics are given in parentheses. ***, ** and * denotes statistical significance at the 1%, 5% and 10% levels, respectively.

Robust Regressions

CAPX Acquisitions Investments

(1) (2) (3)

IG -0.061 -0.016 -0.120 (-2.745)** (-3.215)*** (-10.779)*** Size -0.001 0.006 0.015 (-0.140) (3.816)*** (4.382)*** Leverage 0.071 -0.004 -0.012 (1.074) (-0.292) (-0.364) R&D -0.006 -0.001 0.015 (-0.025) (-0.028) (0.129)

Adj.R2 0.300 0.523 0.895 N 17 17 17 CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 147

Table 11 (Robustness Test 2): Regressions of Investment Policy Variables Before and in the Year of CEO Turnover Controlling for Performance-Related Turnover, Hiring of Outsider CEO, and CEO Pay- Performance Sensitivities This table reports the results of the regression whereby the dependent variable is the change in investments rate from two years before CEO turnover compared to the turnover year. The main explanatory variable is the Internal Governance dummy variable (IG) which equals one if the internal governance is within the optimum range as determined by Table 3, column (4) for the M/B performance variable. Outsider Successor is a dummy variable that is set to unity if the new CEO had less than two years of employment as an executive in the firm before being appointed CEO. PPS is the pay for performance sensitivity of the CEO’s compensation. All other variables are defined in Table 1. Our sample period is 1996 to 2017 and only includes older CEOs (age greater than or equal to 56 years two years before CEO turnover). t-statistics are given in parentheses and all standard errors are two-way clustered by firm and year. ***, ** and * denotes statistical significance at the 1%, 5% and 10% levels, respectively.

Voluntary Turnover (1) (2) (3) Change in Sppe at t=0 Sppiv at t=0 Investments (-2,0) Outsider Successor -0.059 0.004 -0.050*** (-1.27) (1.50) (-3.39) IG -0.061** 0.009*** -0.015*** (-2.31) (3.08) (-3.27) M/B -0.005*** -2.42e-04 -1.11e-04 (-3.11) (-1.20) (-0.26) Size 0.110*** -0.005** 0.018*** (4.81) (-2.19) (3.06) Leverage 0.160** 0.025 0.026* (2.07) (1.65) (1.73) R&D 1.607*** 0.053 0.374** (3.62) (0.96) (2.49) Directors -0.026*** 0.001 -0.000 (-3.82) (1.38) (-0.28) Outsiders 0.022 -0.001 0.098*** (0.26) (-0.45) (3.11) PPS -0.006 0.001 -0.003 (-0.66) (0.90) (-1.16) Year fixed-effects yes yes yes Firm fixed-effects yes yes yes Adj.R2 0.654 0.479 0.654 N 391 298 402 CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 148

Table 12 (Robustness Test 3): Regressions of Firm Performance on In- ternal Governance and CEO’s Pay Slice for Younger and Older CEOs This table summarize the results of regressing firm performance variables (current year’s industry-adjusted ROA and market-to-book ratios) on the internal governance variable δ and CEO’ pay slice (CPS). CPS is the fraction total CEO compensation out of the total compensation for the whole management team. See Table 1 for variable definitions of other controls. Panel A presents the results for younger CEOs (defined as those whose age is less than the median CEO age of 56 years), and Panel B presents the results for older CEOs (defined as those whose age is greater than or equal to the median CEO age of 56 years. Our sample period is 1996 to 2017. t-statistics are given in parentheses and all standard errors are two-way clustered by firm and year. ***, ** and * denotes statistical significance at the 1%, 5% and 10% levels, respectively.

(1) (2) (3) (4) (5) (6) ROA ROA ROA M/B M/B M/B Panel A: Younger CEOs δ 0.085* 0.065 -0.947 -1.569 (1.90) (1.43) (-0.62) (-1.03) δ2 -0.115 -0.087 2.175 3.050 (-1.52) (-1.14) (0.85) (1.19) CPS 0.150*** 0.157*** 2.490* 2.691** (3.80) (3.91) (1.94) (2.11) CPS2 -0.152** -0.164*** -1.183 -1.609 (-2.49) (-2.66) (-0.64) (-0.87)

Controls yes yes yes yes yes yes Year fixed-effects yes yes yes yes yes yes Firm fixed-effects yes yes yes yes yes yes Adj.R2 0.287 0.290 0.291 0.272 0.277 0.275 N 9341 9513 9340 9536 9710 9535 Panel B: Older CEOs δ 0.048 0.047 3.912*** 3.914*** (1.26) (1.24) (2.87) (2.87) δ2 -0.055 -0.058 -6.079*** -6.139*** (-0.90) (-0.95) (-2.70) (-2.73) CPS -0.005 -0.009 -0.634 -0.781 (-0.17) (-0.33) (-0.63) (-0.75) CPS2 0.087** 0.094** 2.248 2.442* (2.25) (2.40) (1.63) (1.72)

Controls yes yes yes yes yes yes Year fixed-effects yes yes yes yes yes yes Firm fixed-effects yes yes yes yes yes yes Adj.R2 0.121 0.125 0.124 0.172 0.171 0.172 N 8348 8542 8348 8252 8442 8252 CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 149

Table 13 (Robustness Test 4): Regressions of Investment Policy Vari- ables Before and in the Year of CEO Turnover on Internal Governance for Older/Younger CEOs Controlling for CEO Power and Geographical Complexity This table reports the results of the regression whereby the dependent variable is the change in investments rate from two years before CEO turnover compared to the turnover year. The main explanatory variable is the Internal Governance dummy variable (IG) which equals one if the internal governance is within the optimum range as determined by Table 3, column (4) for the M/B performance variable. CEO P ower is defined as the sum of a sequence of dummies that identifies three aspects of CEO power: prestige power, expertise power and ownership power. Complexity is a dummy variable that is equal to one for firm-year observations with above the median industry first principle component of the following three variables: the number of geographical segments, geographical sales concentration, and the percentage of foreign sales. Our sample period is 1996 to 2017. t-statistics are given in parentheses and all standard errors are two-way clustered by firm and year. ***, ** and * denotes statistical significance at the 1%, 5% and 10% levels, respectively.

Change in Investments (-2, 0)

Younger CEOs Older CEOs

(1) (2) (3) (4) (5) (6)

IG 0.022 0.052 0.048 -0.121*** -0.120*** -0.121*** (0.47) (1.18) (1.06) (-2.93) (-2.92) (-2.93) CEO P ower -0.086*** -0.092*** 0.004 0.004 (-2.65) (-2.72) (0.36) (0.35) Complexity 0.053 0.092 0.013 0.013 (1.06) (1.60) (0.54) (0.53) M/B 0.003 0.003 0.001 0.002 0.002 0.002 (0.61) (0.55) (0.28) (0.83) (0.81) (0.84) Size 0.046 0.046 0.032 0.098** 0.094** 0.095** (0.62) (0.58) (0.44) (2.31) (2.11) (2.14) Leverage 0.833*** 0.835*** 0.870*** 0.340*** 0.338*** 0.338*** 3.08 (2.88) (3.26) (2.89) (2.83) (2.84) R&D -2.561 -2.504 -2.628 0.698 0.674 0.662 (-1.57) (-1.52) (-1.63) (0.98) (0.92) (0.89) Directors 0.028 0.052** 0.031 -0.023** -0.023** -0.023** (1.22) (2.24) (1.37) (-2.37) (-2.35) (-2.35) Outsiders 0.617** 0.296 0.602** 0.148 0.150 0.150 (2.17) (1.36) (2.16) (1.28) (1.29) (1.29) Year fixed-effects yes yes yes yes yes yes Firm fixed-effects yes yes yes yes yes yes Adj.R2 0.402 0.358 0.408 0.173 0.173 0.173 N 388 388 388 754 754 754 CHAPTER 2: INTERNAL GOVERNANCE AND INVESTMENT 150

Table 14 (Robustness Test 5): Regressions of Investment Policy Variables After CEO Turnover (1,+2) on Internal Governance for Older/Younger CEOs Controlling for CEO Power and Geographical Complexity This table reports the results of the regression whereby the dependent variable is the change in investments rate from the year after CEO turnover to two years after the turnover year. The main explanatory variable is the Internal Governance dummy variable (IG) which equals one if the internal governance is within the optimum range as determined by Table 3, column (4) for the M/B performance variable. CEOP ower is defined as the sum of a sequence of dummies that identifies three aspects of CEO power: prestige power, expertise power and ownership power. Complexity is a dummy variable that is equal to one for firm-year observations with above the median industry first principle component of the following three variables: the number of geographical segments, geographical sales concentration, and the percentage of foreign sales. Our sample period is 1996 to 2017. t-statistics are given in parentheses and all standard errors are two-way clustered by firm and year. ***, ** and * denotes statistical significance at the 1%, 5% and 10% levels, respectively.

Change in Investments (1, +2) Younger CEOs Older CEOs (1) (2) (3) (4) (5) (6) IG -0.023 -0.026 -0.022 -0.004 -0.009 -0.004 (-0.72) (-0.79) (-0.71) (-0.16) (-0.44) (-0.17) CEO P ower -0.047** -0.046** -0.010 -0.009 (-2.34) (-2.34) (-0.54) (-0.53) Complexity -0.034 -0.024 -0.008 -0.004 (-0.81) (-0.58) (-0.39) (-0.23) M/B 0.001 0.000 0.001 0.003 0.003 0.003 (0.34) (0.10) (0.40) (0.98) (1.08) (0.97) Size 0.223*** 0.224*** 0.226*** -0.014 -0.016 -0.014 (5.29) (5.25) (5.35) (-0.67) (-0.74) (-0.69) Leverage 0.215*** 0.274*** 0.220*** -0.106 -0.106 -0.104 (2.67) (3.40) (2.72) (-0.82) (-0.80) (-0.79) R&D 3.300*** 3.200*** 3.296*** 0.173 0.142 0.173 (3.29) (3.15) (3.28) (0.68) (0.53) (0.67) Directors 0.001 -0.003 0.000 0.001 0.000 0.001 (0.10) (-0.41) (0.05) (0.19) (0.07) (0.15) Outsiders -0.098 -0.079 -0.099 -0.091 -0.093 -0.091 (-1.02) (-0.82) (-1.04) (-1.62) (-1.61) (-1.62)

Year fixed-effects yes yes yes yes yes yes Firm fixed-effects yes yes yes yes yes yes Adj.R2 0.339 0.320 0.339 0.284 0.281 0.282 N 865 865 865 459 459 459 CHAPTER 3: DELEGATION, INFORMATION AND MARKET 151

Chapter 3: To Delegate or Not to Delegate? On the Quality of Voluntary Corporate Financial Disclosure and Its Market Impacts

Yankuo Qiao†

This study investigates the impact of delegation structure of the top management team upon the quality of corporate voluntary disclosure on ﬁnancial outcomes.

The paper develops two competing hypotheses pertaining to the functional relationship between the degree of delegation and the management forecast accuracy.

On the one hand, as indicated by the literature on internal governance, the ef-

ﬁcacy of the top management team is optimized when neither the CEO nor the subordinate managers are dominant. On the other hand, an extensive literature has documented the importance and centrality of the CEO as well as the relevance of the subordinate managers to the voluntary disclosure activities. The empirical ﬁndings are in support of an inverted hump-shaped relationship between the degree of delegation and the quality of voluntary information provision, suggesting that an internal optimality of responsibility sharing between the CEO and

†Rutgers Business School–Newark and New Brunswick CHAPTER 3: DELEGATION, INFORMATION AND MARKET 152 her immediate subordinates does not exist for information production, transmission and dissemination. Partial delegation and mixed executive duties lead to deteriorating quality of voluntary disclosure. In particular, the paper analyzes several aspects of managerial earnings forecasts (MFs), the most influential type of voluntary financial disclosure. The documented curvilinear forms are generally persistent across multiple quality metrics of MFs. Consistent with the literature on executive horizon and risk propensity, the curvilinear relation is more significant when the top management team is led by an older CEO. The paper utilizes an identification strategy of structural equations, which controls for selection bias and reverse causality. To theoretically underpin empirical findings, a model of internal information production is developed in the framework of Bayesian Nash

Equilibrium. The paper further documents that when the delegation structure is clear, namely either the CEO or subordinates are in charge, the liquidity of the company’s stock improves. The empirical evidence also suggests that the variation of liquidity driven by delegation structures is not actively incorporated in stock prices.

1 Introduction

Albeit seemingly production irrelevant, disclosure policy is an important component of the strategic management of the ﬁrm. To guide investor expectations, managers would like to voluntarily provide information in addition to the amount of disclosure mandated by market regulations (see, for example, Diamond 1985; CHAPTER 3: DELEGATION, INFORMATION AND MARKET 153

Diamond and Verrecchia 1991). This form of voluntary disclosure is usually referred to as management guidance or management earnings forecasts (MFs) and contains speciﬁc forward-looking information for the upcoming event of earnings announcement. Managers are naturally insiders and their guidance is expected to be insightful. The extant literature has documented the importance of voluntary and transparent disclosure by management. For example, such disclosure can lower cost of capital (which in turn can be value enhancing), facilitate price discovery and reduce litigation risk (see, for example, Amihud 2002; Amihud and Mendelson

1986, 1989; Biddle and Hilary 2006; Botosan 1997; Brennan and Subrahmanyam

1996; Bushman and Smith 2001; Francis, Philbrick, and Schipper 1994; Goodman,

Neamtiu, Shroﬀ, and White 2013; Jensen 2005; Kwak, Ro, and Suk 2012; Nelson and Pritchard 2007; Skinner 1994, 1997). Voluntary disclosure is considered as an eﬀective communication vehicle between managers and investors to reduce information risk assigned by the market (see, for example, Graham, Harvey, and

Rajgopal 2005). Moreover, Balakrishnan, Billings, Kelly, and Ljungqvist(2014) identify the causal relationship between MFs and market liquidity and its impact on the valuation of a company’s stock.

It would seem that the amount of disclosure transparency the firm will provide depends upon CEO expectations of the future. In particular, if the managers perceive that the market is under-valuing her firm, such managers choose to provide additional forecasting information to influence market expectations. Even if the CEO believes that investors are over-valuing the stock, the CEO should strive to generate accurate forecasts since inaccurate forecasts would damage the CHAPTER 3: DELEGATION, INFORMATION AND MARKET 154 manager’s reputation and the firm’s ability to access capital markets in the future

(see, for example, Jensen 2005). This would imply that the decision to voluntarily disclose information to the market should also depend upon the risk proﬁle of the

ﬁrm. CEOs of high risk ﬁrms will choose to disclose less even knowing that the

firm will be punished by the market with higher cost of capital, because not only management guidance requires costlier efforts but also they are less likely to make accurate predictions. The disclosure practices of the firm are thus dependent on the CEO’s incentive and ability to generate accurate information about the future prospects of the firm. Under this overarching premise, the purpose of this study is to investigate the impact of the structure of the CEO-led top management team upon the quality of voluntary disclosure activities. In particular, I look at the degree of delegation responsibilities or sharing executive duties between CEO and her subordinates, a proxy for the relative centrality of the top management team.

I measure the extent of delegation responsibilities based upon the fraction of executive titles carried by the CEO as compared to the other members of the top management team. The measure is drawn from the theoretical model of internal governance in Acharya, Myers, and Rajan(2011), and empirically operationalized in Aggarwal, Fu, and Pan(2017). This proxy is further improved in Brick,

Palia, and Qiao(2019) using a text mining technique, regular expression ( regex), to account for the various patterns embedded in the title strings from Execucomp.

Building upon the extant literature, I develop two competing hypotheses with regard to whether there exists an internal optimality of sharing executive duties for voluntary disclosure quality. It should be noted that the purpose of the paper CHAPTER 3: DELEGATION, INFORMATION AND MARKET 155 is not to measure the number of points of contact between the management and the investment community. Rather, the paper is examining the optimal delegation structure that generates the information for improving management forecasts. As indicated by the literature on internal governance (see, for example, Acharya et al. 2011; Landier, Sauvagnat, Sraer, and Thesmar 2013; Landier, Sraer, and Thes- mar 2009), the efficacy of the top management team is optimized when neither the CEO nor subordinate managers are dominant. Self-interested CEOs who face sufficient challenges from their subordinates are more likely to make policy decisions in the best interest of the firm. Indeed, if the CEO and her subordinates could collaborate and coordinate together effectively as a team in the process of information production, better business insights could be formulated as a common knowledge of the team, thereby improving the quality of voluntary disclosure through which significant amount of forward-looking information is shared with investors. Moreover, the career and reputation concerns of subordinate managers, who have longer executive horizons than the CEO, would mitigate the concern that the CEO might manipulate the channel of voluntary information provision.

Thus, there should exist an internal optimality of the degree of sharing executive duties for voluntary disclosure activities. In other words, the above strand of the literature predicts a hump-shaped relationship between the degree of delegation and the quality of voluntary disclosure.

On the other hand, there is an extensive literature on the importance and centrality of the CEO (see, for example, Buyl, Boone, Hendriks, and Matthyssens

2011; Chaganti and Sambharya 1987; Francis and Armstrong 2003; Gupta and CHAPTER 3: DELEGATION, INFORMATION AND MARKET 156

Govindarajan 1984; Hambrick and Mason 1984; Heaton 2002; Stinchcombe, McDill, and Walker 1968; Thomas, Litschert, and Ramaswamy 1991; Thomas and Simerly

1994) as well as the relevance of the subordinate managers to the voluntary disclosure activities (see, for example, Bamber, Jiang, and Wang 2010; Dyck, Morse, and Zingales 2010; Feng, Ge, Luo, and Shevlin 2011; Ge, Matsumoto, and Zhang

2011; Jollineau, Vance, and Webb 2012; Jollineau et al. 2012; Koo and Lee 2017).1

If there exists conflicts of interests or heterogeneous preferences across executives, namely agency problem among agents, then the internal optimality of delegation executive duties for voluntary disclosure activities predicted in the first hypothesis might not be true. The amount of disclosure transparency and the accuracy of forward looking information could be better when the CEO is in total control or fully delegates the responsibilities to her subordinates. Specifically, when CEOs are dominant and they have a unique informational advantage (Harris and Raviv

2005), they could better solicit information from the operations and fully utilize their superior managerial skills and quality to utilize the information. However, when subordinates are not sufficiently compensated or given appropriate hierarchical status, subordinates have less incentive to cooperate with the CEO especially if the CEO and subordinates have different preferences and views. If the CEO could not accurately infer precise information from her subordinates because of agency conflicts, the CEO can increase the accuracy by providing more indepen- dence to her subordinates. As such, on the one hand, the CEO could infer the

1 Actually there is a large literature on the inefficient communication between CEO and subordinates and this literature finds delegation is a good substitute for communication. More detail is given in the literature review section. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 157 private information and knowledge of her subordinates from their operational and strategic choices; one the other hand, given a bigger role of running the firm, subordinates would more proactively contribute their private information to organizational learning. In essence, the CEO can obtain more accurate information because the subordinates achieve a higher social hierarchy status in the firm (see, for example, Bamber et al. 2010; Bunderson and Reagans 2011; Dewatripont and

Tirole 2005; Gautier and Paolini 2007; Haleblian and Finkelstein 1993; Landier et al. 2009). Thus, this rationale predicts an inverted hump-shaped relationship between sharing executive duties and disclosure quality. The internal optimality of shared executive duties doesn’t exist.

The paper documents empirical evidence in alignment with the second hypothesis. In particular, when the top management team is headed by older CEOs, I

find a statistically significant inverted hump-shaped relationship between responsibility sharing in the top management team and disclosure quality. Moreover, the statistical significance of the relationship becomes more significant in the sample of even older CEOs and becomes less significant when the sample includes younger

CEOs. The empirical results are consistent with two strands of the literature centered on the age of the CEO, a readily observable CEO characteristic. One strand of studies shows that CEO age is related to the risk proﬁle of the ﬁrm and aging

CEOs tend to run a low risk firm (see, for example, Holmström 1999; Prendergast and Stole 1996; Serfling 2014). Given that managerial attention is limited (see, for example, Barkema and Schijven 2008; Bettman, Johnson, and Payne 1986;

Haleblian and Finkelstein 1993; Payne, Bettman, and Johnson 1988), a firm of CHAPTER 3: DELEGATION, INFORMATION AND MARKET 158 high risk profile naturally would not proactively engage in voluntary disclosure activities, since it is costly to generate necessary information to provide quality forecast, while inaccurate disclosure may hurt valuation and management reputation. In contrast, firms of relatively low risk profile are more likely to strive to provide accurate forward-looking information and deploy resources accordingly.

Hence, we would expect to find more significant relationship between the delegation structure of the top management team and voluntary disclosure quality when the risk profile of the firm is low. In other words, the effect of responsibility sharing becomes more salient when the firm is led by an older CEO who maintains a low risk profile and considers the disclosure policy as an important aspect of the

ﬁrm strategy to guide the market expectation. The other strand of the studies is focused on the relation between age and executive horizon. Age has been widely used as a proxy for the executives’ employment horizon (see, for example, Brickley,

Linck, and Coles 1999; Dechow and Sloan 1991; Gibbons and Murphy 1992; Jain,

Jiang, and Mekhaimer 2016; Matˇejka, Merchant, and Van der Stede 2009). An aging CEO is usually of shorter executive horizon than her younger subordinates, leading to significant heterogeneity in managerial preferences and risk propensities. As such, due to conflicts of interest, it is more difficult for an aging CEO to collaborate with her subordinates effectively, and for the management team as whole to collectively form accurate forward-looking information, further making the premise for the existence of internal optimum less valid.

The empirical evidence reconciles two strands of literature. The ﬁrst strand emphasizes the centrality of the CEO. Due to the superior role, detached posi- CHAPTER 3: DELEGATION, INFORMATION AND MARKET 159 tion from the executive ladder and the extraordinary managerial quality, the CEO is more likely to gather inside information, develop insightful business ideas and become more informed relative to other executives in the top management team.

The other strand of studies demonstrates the importance of non-CEO executives such as CFO, chief marketing oﬃcer (CMO) or general counsel (GC) in the information production and voluntary disclosure activities. The ﬁndings indicate that it is detrimental to the quality of voluntary disclosure only if the top team is neither centralized at the CEO nor decentralized by subordinate managers or, in other words, the executive duties between the CEO and her subordinates are mixed together.

To empirically examine the eﬀect of shared executive duties upon the corporate disclosure practices, the paper is focused on the most important and widely recognized voluntary ﬁnancial disclosure by practitioners and scholars – management earnings forecasts (MFs). According to Hirst, Koonce, and Venkataraman

(2008), MFs is characterized as “one of the key voluntary disclosure mechanisms by which managers establish or alter market earnings expectations, preempt litigation concerns and inﬂuence their reputation for transparent and accurate reporting”.

Specifically, the paper constructs four metrics, each of which measures one aspect of voluntary disclosure quality of MFs: bias, error, accuracy and optimism. In order to account for the asymmetric and qualitative nature of those measures, I use nonlinear estimators of Generalized Linear Model (GLM) in addition to OLS regression to improve the efficiency of model estimation. However, estimating one equation alone in the regression analysis may result in spurious statistical infer- CHAPTER 3: DELEGATION, INFORMATION AND MARKET 160 ence because of selection bias since it is completely the discretionary choice of the management to publish any MFs or not. Therefore, I conduct a Heckman correction whereby estimating a Probit model in the first stage prior to the main equation. In addition to selection bias, one might argue that there might be endogenous feedback from voluntary disclosure quality to responsibility sharing in such a way that when the firm chooses to improve the quality of voluntary financial reporting, there might be more disclosure relevant tasks and positions created for managers. To ameliorate such concerns, I propose to use instrumental variable of CEO power index (P owerIndex), which measures the overall degree of power of the CEO. According to Finkelstein(1992), there are four dimensions of CEO power that are inter-related: structural power, ownership power, expert power and prestige power. The degree of responsibility sharing mainly measures the structural power and is positively correlated with other aspects of executive powers.

P owerIndex is a valid and strong instrument for the key variable of interest, degree of delegation in the top management team. As such, the main identiﬁcation strategy of the paper is to estimate a system of equations, which ameliorate the concerns of selection bias and reverse causality. The empirical evidence of the paper suggests that the responsibility sharing in the top management team would causally aﬀect the quality of voluntary disclosure through the channel of internal information production and external information dissemination.

The paper further investigates the corresponding market impacts of the improved quality of voluntary disclosure. Specifically the paper is focused on the change of stock liquidity in face of certain delegation structures. When the infor- CHAPTER 3: DELEGATION, INFORMATION AND MARKET 161 mational efficacy of the top management team is clear and deemed good, namely, when to a large extent, either the CEO takes the total control or fully delegates to her subordinates, the liquidity of the company’s stock shall improve. In particular, I find that when the degree of delegation falls in certain range away from the interior minimal point, the stock liquidity, as measured by Turnover and Amihud

Price Impact improves accordingly. Consistent with the age-varying functional relationship between delegation and voluntary disclosure quality, the salutary effect of clear delegation structure upon stock liquidity strengthens as CEO ages, while the eﬀect diminishes as the sample includes younger CEOs in the sample.

Prior literature (see, for example, Balakrishnan et al. 2014) have documented the causal linkage between stock liquidity and the amount of voluntary information provision by the management. Focusing on the quality of voluntary disclosure, my paper further extends the literature by showing that certain delegation structure of the top management team impacts the stock liquidity through the channel of voluntary information provision quality.

At last, the paper examines the value relevance of the variation of liquidity driven by delegation structures. The literature has shown that expected liquidity improvement decreases ex-ante stock return while unexpected liquidity is positively related to stock returns over time. The results of the portfolio tests demonstrate that the future stock return of informationally efficient regimes is higher than that of informationally inefficient regimes. The results indicate that the liquidity effect of delegation structure is largely unexpected and the corresponding liquidity variation is not actively incorporated in asset prices. The finding is CHAPTER 3: DELEGATION, INFORMATION AND MARKET 162 consistent with the empirical implications of Brick, Palia and Qiao (2019). The mispricing of unexpected liquidity improvement provides alternative explanation for the internal governance literature. Further studies are to be done to untie the

Gordian Knot of ﬁrm value, information and management structure.

This paper is the first study focused on the impact of responsibility sharing in the top management team upon the quality of voluntary disclosure. The paper looks into the specific structure of the top management team and identifies the nonlinear relationship between the degree of delegation responsibilities and the quality of voluntary information provision by the management. The paper is also a direct extension to Balakrishnan et al.(2014) which document that managers can causally improve liquidity of their firm’s shares through voluntary disclosure in face of the exogenous variation of external information environment. Balakrish- nan et al.(2014) focus on the subsequent liquidity change resulting from the active disclosure activities of managers. This paper focuses on the channel within the top management team that increases the effectiveness of voluntary information provision. Taken together, the two papers complement the link of voluntary disclosure activities stemming from the structure of the top management team to the resultant market liquidity of the firm’s stock. The mispricing of unexpected liquidity improvement driven by delegation structure further aligns the seeming discrepancy between information and governance roles of delegation responsibilities. Finally, the relationship between responsibility sharing in the top management team and disclosure quality contributes to the reconciliation of the two strands of studies in the literature. The quality of information provision is better either when the CHAPTER 3: DELEGATION, INFORMATION AND MARKET 163

CEO is of total control or when the CEO fully delegates the tasks to other executives. The empirical evidence suggests that the disclosure quality is inferior only when the responsibility and power allocation in the top management team is inconsistent with both strands of literature. In other words, it is detrimental to the communication between the ﬁrm and investors when the delegation structure of the top management team is unclear and executive duties between the CEO and her subordinates are mixed together. It is the partial involvement of the CEO that truly deteriorates the internal information production, knowledge sharing and organizational learning.

2 Theoretical Background and Literature Review

2.1 Why Firms Disclose?

One might argue that firms and executives are motivated to disseminate information manipulatively, so as to signal information opportunistically and to obscure indications of inferior performance. In fact, there is a vast stream of literature on the real and material earnings management as summarized by review papers such as Dechow, Ge, and Schrand 2010; Dechow and Skinner 2000; Fields, Lys, and Vin- cent 2001. However, voluntary financial disclosure by management is considered an effective communication vehicle between managers and investors that reduces information risk assigned by the market (see, for example, Graham et al. 2005).

Such voluntary disclosure activities can be value enhancing because it lowers cost of capital (see, for example, Amihud 2002; Amihud and Mendelson 1986, 1989; CHAPTER 3: DELEGATION, INFORMATION AND MARKET 164

Botosan 1997; Brennan and Subrahmanyam 1996). Balakrishnan et al.(2014) identify the causal relationship between management earnings forecasts and market liquidity. Through their identification strategy, Balakrishnan et al.(2014) find that in face of the exogenous loss of public information, managers seek to actively improve the liquidity and valuation of their firm’s shares by voluntarily disclosing more information than the amount mandated by market regulation. As the business world becomes increasingly complex and litigious, it is necessary for firms to deal with legal and regulatory compliance, and to manage litigation risk. Empir- ical evidence suggests that voluntarily and objectively disclosing information is a good practice of mitigating litigation risk (see, for example, Francis et al. 1994;

Kwak et al. 2012; Nelson and Pritchard 2007; Skinner 1994, 1997). Moreover, firm could utilize the opportunity of voluntary disclosures, such as management earnings forecasts (MFs), to signal relative performance, guide investors’ expectation and facilitate price discovery, thereby facilitating fair evaluation and ameliorating the cost of over-and under-valued equity (see, for example, Jensen 2005; Kwak et al. 2012). In sum, the studies on top managers and disclosure practices recognize the financial importance and the value-relevance of voluntary disclosure policy. Voluntary disclosure activities could be strategically utilized by the management together with other firm policies such as financing and investment for value-enhancing.

In addition, managers can establish strong personal reputation through reporting truthfully and forecasting accurately (Stocken 2000). Empirical evidence indicates that managers are greatly concerned about building the strong credibility CHAPTER 3: DELEGATION, INFORMATION AND MARKET 165 of communicating financial information transparently with investors’ community, especially when it comes to forward-looking information during high uncertainty period (see, for example, Buyl et al. 2011; Graham et al. 2005). One benefit for managers of forecasting accuracy is to increase their credibility and reputation for capital markets (see, for example, Yang 2012). Moreover, by trading off short-term payoff for long term credibility gains, managers would receive better assessment from managerial labor market, leading to higher reservation wage (see, for example, Hui and Matsunaga 2014). As such, managers, especially non-CEO executives, are rewarded for high disclosure quality and thus are motivated to communicate effectively with investors. Furthermore, extant studies document the relationship between the management forecasts and investment decisions, suggesting that forecast accuracy is indicative of managerial competence and is positively related to investment efficiency (see, for example, Biddle and Hilary 2006; Bushman and

Smith 2001; Goodman et al. 2013).

2.2 Internal Governance

The literature suggests that failure by management to provide accurate and timely information can increase information asymmetry in the market by which ﬁrm is punished with high cost of capital, acute risk of litigation and impaired management credibility. Hence, there may exist a governance mechanism that could mitigate the agency problem. Acharya et al.(2011) theorize that internal governance, a monitoring mechanism from within the management team, can reduce agency cost, and complements other governance mechanisms such as board of di- CHAPTER 3: DELEGATION, INFORMATION AND MARKET 166 rectors and large shareholders. They develop a theoretical model that delineates the relationship between responsibility sharing and ﬁrm performance. Acharya et al.(2011) predict that the relationship between delegation responsibilities and

firm performance is hump-shaped, suggesting that optimal governance is when neither the CEO nor subordinate managers are dominant. With optimal internal governance in place, corporate decisions and policy outcomes would improve through the checks and balances among top managers, mitigating the idiosyncratic biases of individual managers. Adams, Almeida, and Ferreira(2005) demonstrate that the financial performance is more variable for firms led by powerful CEOs as measured by title, ownership and status, who are in general less disciplined by subordinate managers and other governance mechanisms. In contrast, the risk profile of the firm is lower in face of good internal governance, leading to more certain profitability and stable financial performance. Hence, if the CEO and subordinate managers could work collaboratively and effectively together as a coherent team, they are more likely to generate accurate and insightful forward-looking information pertaining to the financial and economic prospects of the firm. In addition, as a bottom-up control mechanism, internal governance stems from the degree of delegation in the top management team, which could be also beneficial for external information dissemination.

The leader of the management team, the CEO, generally has relatively shorter executive horizon compared to others and is likely to manipulate information for equity incentive (see, for example, Feng et al. 2011; Jain et al. 2016). The literature documents that subordinate managers are motivated to discipline the CHAPTER 3: DELEGATION, INFORMATION AND MARKET 167

CEO’s myopic behavior of disclosing opportunistically for a variety of reasons.

Specifically, Yang(2012) and Hui and Matsunaga(2014) demonstrate that by disclosing transparently and forecasting credibly, managers gain managerial reputation. Moreover, Jensen(2004, 2005) recognize and theorize the detrimental effect of overvalued equity on firm’s long term performance. Given the relative longer executive horizon of subordinate managers compared to that of the CEO, inflating equity value through earnings management and information manipulation is not in the best interest of non-CEO executives. According to Cheng, Lee, and Shevlin(2015), subordinate managers usually care about the long term performance, especially those who are potential candidates for CEO in the future, rather than short term gain from information manipulation. In sum, literature shows that subordinate managers usually stand in the first line of defense against biased financial reporting and act as whistle blower on corporate fraud. Clearly, when internal governance is optimal and the implementation of corporate policy demands collective team work between the CEO and her subordinate managers, the CEO is more likely to abide the wishes of subordinate managers.

2.3 Importance of CEO

The CEO, as the substantive and symbolic leader of the firm, is responsible for coherently managing the internal relationships of different stakeholder groups of interest and sailing the firm through the complexity and turbulence of external market conditions (Thomas and Simerly 1994). The centric and dominating role of the CEO in determining firm policy and influencing firm performance is widely CHAPTER 3: DELEGATION, INFORMATION AND MARKET 168 documented both in finance and economics literature and in organizational management literature (see, for example, Buyl et al. 2011; Chaganti and Sambharya

1987; Francis and Armstrong 2003; Gupta and Govindarajan 1984; Hambrick and

Mason 1984; Heaton 2002; Stinchcombe et al. 1968; Thomas et al. 1991; Thomas and Simerly 1994). Although specific tasks associated with the policy implementation are carried out by lower level managers, the CEO, as the head of the top management team, plays a key role in structuring, exchanging and integrat- ing asymmetric information from subordinate managers of diversified functional tracks. From an information-processing prospective, Hambrick(1994, 1995) reveal that the leadership of the CEO is of utmost importance in unleashing the full capacity of a functionally diversified top management team through integrat- ing the perspectives from subordinate managers to avoid team fragmentation. In a similar vein, extensive studies are focused on the unique and decisive role of

CEOs that diﬀerentiate them from other managers in the top management team

(Arendt, Priem, and Ndofor 2005; Haleblian and Finkelstein 1993; Jaw and Lin

2009; Kisfalvi and Pitcher 2003; Minichilli, Corbetta, and MacMillan 2010; Pa- padakis and Barwise 2002). Buyl et al.(2011) document that the impact of the

CEO’s expertise and characteristics upon the integration and exchange of information and knowledge among functionally diversified managers. In fact, the CEO is usually hired inside as the winner of the tournament within the top management team or outside as the savior to turn the firm around. Due to the superior position and managerial skills, the CEO is more likely to gain access to inside information and develop insightful business views pertaining to uncertain market conditions CHAPTER 3: DELEGATION, INFORMATION AND MARKET 169 and prospects of firm performance.

2.4 Inﬂuential Top managers and Disclosure Activities

A large strand of literature is focused on theoretically and empirically analyzing the relationship between top managers and the firm disclosure practices. Since the corporate management is shared effort, responsibility and wisdom of all top managers, examining only the role of the CEO doesn’t provide the full picture of disclosure quality. Focusing either on the characteristics of individual top managers or the composition of the top management team, this strand of studies extends the scope of neoclassical firm theory (see, for example, Aguilar 1967; Jensen and Meckling 1976; Modigliani and Miller 1958) and conventional management science (see, for example, Lieberson and O’Connor 1972), which assume that firm optimally respond to the external business environment and conclude that the internal monitoring mechanism and idiosyncratic influence of individual managers should have no economically significant impact on the financial outcomes of corporations. The theoretical foundation of this strand of studies is the upper echelon theory by Hambrick and Mason(1984), which predicts that due to the heterogeneous characteristics among managers such as demographic variables, education and functional background, and the invisible values and qualities, managers are not interchangeable and the composition of management team largely shape the

firm policy and financial outcomes. Bamber et al.(2010) identify the economically significant idiosyncratic influence of five highest paid managers on the quality of voluntary disclosure. The findings also indicate that, even though seemingly some CHAPTER 3: DELEGATION, INFORMATION AND MARKET 170 executives, such as president or COO, are not directly participating in disclosure practices as their main executive duty, they are involved in the process of disclosures by sitting in the disclosure committee or by upholding a corporate culture of transparency.

Influential individual managers with different functional tracks unequivocally contribute to improving various aspects of disclosure quality. Using various measures, such as tenure, presence in the top management team (five highest paid) and rank of executive titles, to identify influential managers, the extant studies are focused on a wide range of top managers such as Chief Financial Officer (CFO),

Chief Marketing Oﬃcer (CMO), General Council (GC) and Chief Operating Of-

ficer (COO). Specifically, Ge et al.(2011) document a CFO fixed effects upon accounting practices and financial reporting, and the idiosyncratic influence of individual CFO is moderated by the effect of the CFO’s discretionary powers. The salutary effect of CFO is more manifested in financial reporting quality when the

CFO has great discretionary powers. Kwak et al.(2012) document that the presence of a powerful GC in the top management team (ranked in five highest paid) significantly improves disclosure quality, particularly the accuracy of management earnings forecast (MFs). They suggest that when the GC is among the five highest paid managers and regarded as member of top management team, she usually becomes the chair or active member of the disclosure committee and scrutinize the practices of mandatory and voluntary disclosure. The empirical evidence shows that through the strong connection with board and disclosure committees, the

GC eﬀectively monitors the unusual disclosure practices that hurt shareholder’s CHAPTER 3: DELEGATION, INFORMATION AND MARKET 171 interest. As a result, the empirical evidence suggests that the prominence of the

GC is positively associated with MFs that are of higher frequency, less oppor- tunistic, and more likely to reveal bad news. Koo and Lee(2017) document the salutary effect of a influential CMO upon management revenue forecasts. Ac- counting for tenure, relative compensation and executive titles, the study uses a composite score to measure the executive power of CMO and demonstrates that a powerful CMO increases the firm’s likelihood of issuing revenue forecasts and the accuracy of the forecasts. The empirical evidence indicates that when usually considered lower-ranked CMOs become powerful, they are more motivated to employ their expertise on market forecasting and contribute to revenue guidance that should have been in hands of higher-ranked executives such as CEO, thereby leading to better communication between the firm and investors in terms of the future operation and profitability. In sum, leveraging unique functional track and expertise, individual managers play a strong information role in various aspects of disclosure practices, enhancing the overall level of firm transparency. In fact, it is widely adopted in the literature almost as an premise that in order to formulate transparent reporting and reliable forecasting, firm needs expert inputs and fresh perspectives from subordinate managers of diversified functional background (see, for example, Haleblian and Finkelstein 1993).

2.5 Delegation and Internal Information Production

Numerous theoretical and empirical papers illustrate the information role of subordinate managers through the channel of delegation responsibilities. Haleblian CHAPTER 3: DELEGATION, INFORMATION AND MARKET 172 and Finkelstein(1993) document that the overall eﬃcacy of top management team is dependent on its size, composition, centrality and dominance of CEO. The empirical evidence shows that an exclusive management team with a dominant CEO performs worse in face of a turbulent external business environment, suggesting that the information processing capability of centralized top team is inferior to a well-diversiﬁed and decentralized one. The theory papers of delegation pay particular attention to the information advantage of power and responsibility sharing.

Gautier and Paolini(2007) ﬁnd that delegation serves as a revelation mechanism that the CEO could learn and gather information from the actions and choices of her subordinates. Delegation is regarded as a tradeoﬀ between control power and information advantage. The theoretical work in Dewatripont and Tirole(2005) suggests that delegation could enhance interest congruence and communication among managers. In short, the power and responsibility sharing could in its own right reduce the cost of information production and signal revelation.

The review paper of Bunderson and Reagans(2011) states that the proactive inputs from field managers, who are exposed to different sets of tasks and interact with various groups of stakeholders, enhance the firm’s capability of information processing, knowledge sharing and collective learning, and leads to more effective error corrections and informed actions. Moreover, such mechanism of information sharing is moderated by the effect of social hierarchy—heterogeneity in power and social status among managers. Clearly, when the power distribution of the top management team is significantly skewed to the CEO, the inputs from individual managers are not given equal and fair hearing which will in turn discour- CHAPTER 3: DELEGATION, INFORMATION AND MARKET 173 age low-ranked managers from sharing insights and perspectives. In a nutshell, they conclude that centralized power allocation in the organizational structure is detrimental to internal information production in that insights and perspectives of higher-ranked managers, such as the CEO, are given disproportionate weight whereas inputs and contributions of her subordinate managers are usually ne- glected—they merely defer to the tone set from the top. In contrast, a team of subordinate managers upon which important executive duties are bestowed would no longer acquiesce the preference of CEO and is more motivated to proactively contribute fresh perspectives. In their analytical work, Landier et al.(2009) illustrate that an informed CEO is more willing to use objective information optimal to firm value in face of likely dissenting subordinates. Their work suggests that even when the CEO is in a superior position of information, overlooking the insights and contributions of subordinate managers is still detrimental to the internal information production and firm value.

2.6 Aging CEO

2.6.1 Heterogeneity in Executive Horizons

Hambrick and Mason(1984) argue that significant attention must be paid to managerial demographic characteristics. This paper has triggered numerous studies on relationship between firm performance and demographic composition of the organization’s dominant coalition, namely the top management team. The experience and idiosyncratic preference of managers could be collectively represented CHAPTER 3: DELEGATION, INFORMATION AND MARKET 174 by age, a readily observable managerial characteristic. Age consists of various sets of elements that dynamically and constantly shape identity and behavior of managers. It is widely used in the literature as a proxy for employment horizon of executives (see, for example, Brickley et al. 1999; Dechow and Sloan 1991; Jain et al. 2016; Matˇejka et al. 2009; Pan, Wang, and Weisbach 2016). As the CEO ages, her executive horizon becomes shorter and she is more focused on boosting short term profitability. Accordingly, older CEOs tend to downsize risky investments and initiate projects that could generate a large lump sum of current cash flows.

Moreover, it is more acute that the executive horizon of aging CEO is different from that of her subordinate managers who are likely much younger and have career and reputation concern for the long term. A top management team of significant different executive horizons, especially between CEO and her immediate subordinates, is less likely to achieve interest alignment and to forster effective information exchange, and therefore is of high risk of group fragmentation.

2.6.2 Heterogeneity in Risk Propensities

In a similar vein, age, as a readily observable demographic variable that consol- idates life and professional experience, cohort eﬀect and values, is the crux of scholarly studies on the risk-taking behavior of CEO in corporate decisions. In the prior theoretical works, the prediction on how age shapes a CEO’s attitudes towards risk is rather equivocal. In particular, theoretical models that incorporate career concerns indicate that younger CEOs are more risk-averse since managerial labor market assessed their capabilities intensely in early years. (see, for exam- CHAPTER 3: DELEGATION, INFORMATION AND MARKET 175 ple, Hirshleifer and Thakor 1992; Holmstr¨om 1999; Scharfstein and Stein 1990;

Zwiebel 1995). Younger CEOs with less records of achievements will be punished by the market for inferior performance more harshly than those well-established senior managers of good reputation. To gradually build up their reputation of high managerial quality and maximize their potential for future career development, younger CEOs would adopt conservative investment policies and run firms of low risk profile. In contrast, supported by anecdotal evidence, Prendergast and Stole(1996) develop a signaling model and illustrates younger CEOs tend to invest more aggressively to demonstrate superiority of talent and managerial ability. Consistent with Prendergast and Stole(1996), a more recent empirical work by Serfling(2014) verifies the negative causal relationship between CEO age and risk-taking propensity. Aging CEOs expend more effort on risk management and deploy less risky investment policy.

2.7 Voluntary Disclosure, Liquidity and Stock Returns

Early theoretical works by Diamond(1985) and Diamond and Verrecchia(1991) show that ﬁrms disclose voluntarily to level the playground and to reduce information asymmetry among their investors. A subsequent survey by Graham et al.

(2005) suggest that managers commit to voluntary disclosure in order to “reduce the information risk that investors assign to our work”. A more recent analytical work by Lester, Postlewaite, and Wright(2011) aims to theoretically establish the linkage between financial information provisions and asset liquidity. Balakrishnan et al.(2014) examine whether or not managers can indeed actively influence the CHAPTER 3: DELEGATION, INFORMATION AND MARKET 176 information environment and the liquidity of their shares. Despite the fact that liquidity is often modeled as the interaction between market makers and investors in exogenously specified information environment, Balakrishnan et al.(2014) find that managers are able to actively influence stock liquidity by supplying more information through voluntary disclosure than regulation would otherwise man- date, thereby underpinning the linkage between information disclosure and stock liquidity. In face of exogenous losses in external information environment, Bal- akrishnan et al.(2014) document that active response of managers in the form of management guidance is shown beneficial to the liquidity of shares, which in turn improves firm value.

Since investors value stocks by returns net of trading costs due to market friction, information asymmetry and adverse selection, they naturally demand higher return of illiquid stocks to compensate various forms of trading cost incurred

(Amihud and Mendelson 1986, 1989). The notion that asset returns increase in illiquidity or the positive price-liquidity relationship has been widely veriﬁed and well accepted in the previous studies (see, for example, Amihud 2002; Amihud and Mendelson 1986; Brennan, Chordia, and Subrahmanyam 1998; Chordia, Sub- rahmanyam, and Anshuman 2001). One major empirical discrepancy among the studies in this strand of literature is about how fast the variation of liquidity will be incorporated in asset prices. For example, the novel results of Chordia et al.

(2001) show that the volatility of liquidity is negatively related to stock returns, contrary to the well-established fact that agents care about the risk associated with liquidity. According to Amihud(2002), expected illiquidity is incorporated in asset CHAPTER 3: DELEGATION, INFORMATION AND MARKET 177 prices by investors which increase ex-ante stock return and decrease contemporaneous stock price while unexpected illiquidity is negatively related to stock returns overtime. It indicates that the positive linkage between stock return and illiquidity is moderated by how market participants react to the variation of liquidity.

3 Hypothesis Development

In this section I develop two competing hypotheses with regard to the relationship between the degree of delegation responsibilities in the top management team and the quality of voluntary disclosure activities.

On the one hand, the literature on the internal governance mechanisms (see, for example, Acharya et al. 2011; Landier et al. 2009) indicates an internal optimality of sharing executive duties between CEO and her subordinates for voluntary disclosure practices. Internal governance is a governance mechanism stemming from within the top management team. Eﬀective internal governance demands

CEO facing sufficient challenges and heterogeneity in preferences from her subordinates. It is considered optimal when the shared executive duties between CEO and her subordinate managers are balanced to a certain degree that neither party is dominant in the top management team. Good internal governance would op- timize the investment policy of the firm, which in turn lower the risk profile of the firm and improve the prospects in the long run (Adams et al. 2005). It is well known that managers have limited attention (see, for example, Barkema and

Schijven 2008; Bettman et al. 1986; Payne et al. 1988), and disclosures of inferior CHAPTER 3: DELEGATION, INFORMATION AND MARKET 178 quality would hurt ﬁrm valuation and management reputation (see, for example,

Biddle and Hilary 2006; Bushman and Smith 2001; Goodman et al. 2013). Hence, managers, who run a low risk firm and believe that the perceived risk by the capital market is higher than the actual risk, will choose to disclose more, because not only it is beneficial for the valuation but also they are capable of giving accurate predictions. In contrast, managers who run a high risk firm will choose to disclose less even knowing that the firm will be punished by the market with higher cost of capital, because not only management guidance requires efforts but also they are less likely to make accurate predictions. As such, managers of firms that maintain a set of risky investments would care less about providing accurate forward-looking information to the market and expend less managerial effort on voluntary disclosure activities, resulting in inferior quality of voluntary information provision. Therefore, if good internal governance could effectively lower the risk profile of the firm, the amount of disclosure transparency and the quality of management guidance should be optimized when the responsibility sharing in the top management team is proportionate to some degree. Moreover, providing accurate forward-looking information for earnings is a costly process, which requires managerial effort and the collective wisdom of the top management team as whole. If the CEO and her subordinates communicate and collaborate with each other effectively, the firm tends to provide high quality disclosure when the

CEO shares a portion of executive duties with her subordinates. In the context of a collaborative and collective management team, the CEO is more likely to get a better sense of the information generated by ﬁeld managers, and to develop CHAPTER 3: DELEGATION, INFORMATION AND MARKET 179 deeper insights of the business, which in turn leads to more accurate and valuable

MFs. With regard to the internal information production and risk proﬁle of the

ﬁrm, this argument predicts an internal optimum or a hump-shaped relationship between sharing executive duties and the quality of voluntary disclosure.

Hypothesis 1A: There exists a hump-shaped relationship between the degree of delegation in the top management team and the quality of voluntary disclosure practices. The inﬂection point represents the optimally shared executive duties between CEO and her subordinates for voluntary information provision.

The underlying assumption of Hypothesis 1A is that CEO and her immediate subordinates could collaboratively work together as a coherent team in the process of information and knowledge sharing. However, as documented in the literature of strategic information transmission, such a premise is not a common theme (see, for example, Crawford and Sobel 1982; Harris and Raviv 2005). In contrast, due to the agency problem among executives, neither CEO nor subordinate managers are likely to voluntarily communicate their private information fully to the others. If this is more prevalent empirically, a contrary functional relationship between degree of delegation and quality of voluntary disclosure to the one in Hypothesis 1A could very well be true. Below, I present the rationale for a competing hypothesis to Hypothesis 1A.

There are extensive studies that focus on the centrality of CEO (see, for example, Buyl et al. 2011; Chaganti and Sambharya 1987; Francis and Armstrong

2003; Gupta and Govindarajan 1984; Hambrick and Mason 1984; Heaton 2002; CHAPTER 3: DELEGATION, INFORMATION AND MARKET 180

Stinchcombe et al. 1968; Thomas et al. 1991; Thomas and Simerly 1994). CEO is at the top of the executive ladder, and is usually the symbolic as well as the actual leader of the firm. The centrality and importance of the CEO is manifested in every aspect of the firm policy and in each process of firm operations, including disclosure policy and voluntary information provision activities. There is a reason for the incumbent CEO to be appointed to the position and remain in power. CEO is usually hired as either an insider who is the winner of the tournament among managers or as an outsider who is regarded as the savior to turn firm around. In either case, CEO is usually superior to her subordinate managers in terms of experience, managerial ability and skills. As such, it is relatively easier for the CEO than the subordinate managers to develop deeper understanding of the business and to provide insightful guidance. Following this rationale, it could be argued that CEO should take on more executive duties in order to collect more information from different tracks of operations, thereby formulating reliable forecast for future earnings and financial performance.

On the other hand, there is a strand of the literature that documents the unique contribution of non-CEO executives to internal information production and voluntary disclosure practices (see, for example, Bamber et al. 2010; Dyck et al. 2010; Feng et al. 2011; Ge et al. 2011; Jollineau et al. 2012; Koo and Lee

2017; Kwak et al. 2012). If the field managers have their own inside information and knowledge which is not known by the CEO, those self-interested subordinates are not necessarily revealing the information to the CEO because of conflict of interests. For example, if there are multiple potential projects and the CEO is CHAPTER 3: DELEGATION, INFORMATION AND MARKET 181 in total charge of decision making, subordinate managers may intentionally hide information and mislead CEO to choose the one that costs themselves less effort or allow them to consume more non-pecuniary consumption. As such, the internal information production would be better if the CEO delegates more duties and responsibilities to her subordinates for two reasons. First, it improves the social and executive hierarchy of the non-CEO executives, motivating them to work harder and be more honest with the firm (see, for example, Bunderson and Reagans 2011;

Landier et al. 2009). Subordinate managers no longer merely defer to the tone set from the top but they become motivated to contribute more insightful inputs in the decision making process. Second, the CEO could infer more information and knowledge from the actions of ﬁeld managers, which enhances communication, organizational learning and interest congruence (see, for example, Bamber et al. 2010; Dewatripont and Tirole 2005; Gautier and Paolini 2007; Haleblian and

Finkelstein 1993). While trading oﬀ control power for inside information, the position of the CEO will become stronger overtime in terms of information advantage through collectively gathering information and knowledge from her subordinates.

Furthermore, subordinate managers sometimes have a much longer career ahead than the incumbent CEO. They may care more about their professional development, reputation and personal impact on the market. As such, they may actively and honestly disclose information and mitigate the concern that the CEO might manipulate the channel of voluntary information provision.

Harris and Raviv(2005) demonstrate that the more important the information that the subordinates have, the more likely the CEO will delegate responsibilities CHAPTER 3: DELEGATION, INFORMATION AND MARKET 182 to the subordinates. The CEO has the incentive to delegate more, since the subordinates are responsible for creating current earnings (Acharya et al. 2011).2 In sum, we expect two informational regimes depending on the relative ability of the CEO to obtain accurate information from subordinates: centralization and delegation. The quality of voluntary disclosure could improve in either regime but deteriorate in the middle, indicating an inverted hump-shaped or U-shaped relationship.

Hypothesis 1B: There exists an inverted hump-shaped relationship between the degree of delegation in the top management team and the quality of voluntary disclosure practices. The internal optimality of sharing executive duties between

CEO and her subordinates doesn’t exist for voluntary disclosure quality. A theoretical basis for this hypothesis can be found in the Appendix A.

The validity of the underlying assumption for Hypothesis 1B, namely, the prevalence of agency problem within the top management team, is determined by the level of interest congruence and dynamism of power distribution among senior executives. Hence, the functional form of the relationship between sharing executive duties and quality of disclosure activity should be dependent on the characteristics of the CEO. In particular, I posit that age, as a readily observable demographic characteristic, is an important attribute of CEO that would shape the functional form of the relationship of interest, since CEO age is deeply related

2 Note that Acharya et al.(2011) assume no information asymmetry between the CEO and her subordinates. Hence, while it may be optimal for the CEO to co-share responsibilities among the subordinates, it may be more imperative when there is information asymmetry to give even more responsibilities to the subordinates in order to motivate them to produce information. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 183 to the executive horizon, the risk propensity, and managerial style of the CEO.

The premise of Hypothesis 1A is that CEO and subordinates are able to collaborate and coordinate eﬀectively as a team in the process of information production and knowledge sharing. I propose that such a premise is less realistic when

CEO is older. An aging CEO has a shorter executive horizon than her younger subordinates, leading to significant heterogeneity in managerial preferences (Landier et al. 2009). The psychological and social gap exaggerated by CEO age make the CEO and her subordinates increasingly difficult to coherently learn from each other’s private information and knowledge through responsibility sharing. Thus, the conflict of interests between the aging CEO and her subordinates invalidates the premise of Hypothesis 1A, and is indicative of a functional relationship of inverted hump-shaped relation between delegation and disclosure quality.

Hypothesis 2: When the top management team is led by an older CEO, the functional form of the relationship between responsibility sharing and voluntary disclosure quality is more likely to be inverted hump-shaped.

Another strand of studies shows that CEO age is related to the risk profile of the firm and aging CEO tends to run a low risk firm (see, for example, Holm- ström 1999; Prendergast and Stole 1996; Serfling 2014). Aging CEO should care more about the quality of disclosure practices in the sense that low risk firms can more effectively implement a transparent disclosure policy and issue high quality management guidance to reduce the risk of information asymmetry in the market, thereby lowering discount rate and upholding valuation. In other words, albeit seemingly production-irrelevant, disclosure policy becomes a key aspect of the CHAPTER 3: DELEGATION, INFORMATION AND MARKET 184 strategic management of the firm under the helm of an aging CEO. Hence, we expect the inverted hump-shaped relationship between delegation responsibilities and voluntary disclosure quality to be even more pronounced as the CEO ages.

Hypothesis 3: When the top management team is led by an older CEO, the curvilinear form of the relationship between responsibility sharing and voluntary disclosure quality is likely to be more signiﬁcant.

Prior literature (see, for example, Balakrishnan et al. 2014) has documented the causal linkage between stock liquidity and the amount of voluntary information provision by the management. This study is a direct extension to Balakrishnan et al.(2014) by looking at the channels of delegation structure for information production, through which the quality of voluntary disclosure is determined. Thus, it is natural and interesting to examine the direct impact of certain information regimes, as identiﬁed by delegation structures, on stock liquidity. According to the two competing hypothesis, Hypothesis 1A and Hypothesis 1B, I propose the following two hypotheses:

Hypothesis 4A: If there exists an internal optimality of shared executive duties between CEO and her subordinates for voluntary information provision, then stock liquidity increases as the degree of delegation approaches the internal optimum.

Hypothesis 4B: If the optimal informational eﬃciency of delegation structure should lie in either of the two regimes in which either CEO has all the responsibilities or the entire delegation of responsibilities is given to the subordinates, then CHAPTER 3: DELEGATION, INFORMATION AND MARKET 185 stock liquidity improves as the degree of delegation approaches either zero or one.

Managers strive to generate accurate forecasts in corporate voluntary ﬁnancial disclosure to reduce information risks and to foster eﬀective communication with the investors’ community (Balakrishnan et al. 2014; Diamond 1985; Graham et al.

2005), which will improve the liquidity of the company’s shares and eventually up- hold a premium valuation. The eﬀect of liquidity upon security prices and returns has been well-recognized in the literature. Investors would demand high return for illiquid assets to compensate trading cost (Amihud and Mendelson 1986, 1989). In this regard, albeit production irrelevant, voluntary disclosure policy indeed matters for ﬁrm value. However, it is not clear yet how fast the variation of liquidity will be incorporated in asset prices. Contrary to the widely-documented fact that investors care about the risk associated with liquidity, Chordia et al.(2001) show that the volatility of liquidity is negatively related to stock returns. A plausible reconciliation is given in Amihud(2002). The study shows that expected illiquidity is incorporated in asset prices by investors which increase ex-ante stock return and decrease contemporaneous stock price while unexpected illiquidity is negatively related to stock returns overtime. Therefore, the linkage between stock return and illiquidity is moderated by how market participants react to certain variation of liquidity. Accordingly, I propose the following hypothesis:

Hypothesis 5A: If market participants actively incorporate the liquidity effect of informationally efficient delegation structures in a timely manner or even overestimate the salutary effect on liquidity, the future stock returns of firms of infor- CHAPTER 3: DELEGATION, INFORMATION AND MARKET 186 mationally efficient regimes will be lower.

Hypothesis 5B: If it takes a long time for market participants to correctly incorporate the informational eﬃciency driven by the delegation structures or they simply underestimate the salutary eﬀect on liquidity, the future stock returns will be higher.

4 Methodology and Sample

4.1 Measures and Methodology

4.1.1 Voluntary Disclosure Quality

The measure of responsibility and power sharing in the top management team, as theorized by Acharya et al.(2011) is first empirically operationalized in Aggarwal et al.(2017) as the fraction of executive titles carried by the CEO. This proxy is further improved by Brick et al.(2019) who use a text mining technique of regular expression (regex) to account for the various patterns embedded in the title strings from Execucomp. Please refer to Brick et al.(2019) for detailed explanation and caveats of this procedure. To be consistent with the prior studies, I use δ to denote the proxy for responsibility sharing and power distribution of the top management team. δ is calculated as the number of executive titles of the CEO (f) deflated by the total number of executive titles (f + g) carried by the entire top five managers identified by the executive compensation.

To empirically examine the effect of delegation responsibilities upon the cor- CHAPTER 3: DELEGATION, INFORMATION AND MARKET 187 porate disclosure practice, the paper is focused on the most important and widely recognized voluntary financial disclosure by practitioners and scholars – management earnings forecasts (MFs). According to Hirst et al.(2008), MFs is characterized as “one of the key voluntary disclosure mechanisms by which managers establish or alter market earnings expectations, preempt litigation concerns and influence their reputation for transparent and accurate reporting.” As such, albeit

MFs is only a tip of the information iceberg, its discretionary nature and huge market impacts make MFs a ideal channel through which to examine the impact of power and responsibility sharing in the top management team on corporate disclosure policy.

Building upon the previous studies (see, for example, Ajinkya, Bhojraj, and

Sengupta 2005; Baginski, Hassell, and Kimbrough 2002; Bamber et al. 2010), I use four metrics to measure the accuracy of MFs. I measure forecast bias (F AbsBias) by comparing the EPS of MFs to that of the prevailing analysts’ consensus forecasts.3 F AbsBias is calculated as the absolute diﬀerence between EPS of MFs and EPS of analyst consensus forecasts. I measure the inverse of forecast accuracy

(F AbsError) as the absolute value of the diﬀerence between the management

EPS forecast and ex-post realized actual EPS. Because the metric of responsibility delegation is an annual measure, the MFs metrics should be annual as well.

I construct the above measures, F AbsBias and F AbsError for each ﬁrm and

ﬁscal year by averaging all the MFs in a ﬁscal year. Notice that I do not measure

3 I use MEAN AT DATE from I/B/E/S management guidance, which is defined as the consensus forecast available at the time when management guidance was captured CHAPTER 3: DELEGATION, INFORMATION AND MARKET 188 the quality for each individual forecast because if the voluntary disclosure policy is sticky during the fiscal year (Bamber et al. 2010) then the lack of variation across quarters may mechanically reduce standard errors of the regression, leading to statistically significant parameter estimates as shown in Appendix B. Notice that the number of MFs issued by the firm in itself is a representation of information abundancy. Thus, I also take the sum of the above inverse quality measures to gauge the cumulative performance of forecast accuracy throughout the fiscal year.

The values for the above two measures are given as follows.

N=# MF s 1 F AbsBias = EP S of Analyst Consensus EP S of MF s it N | itn − itn| n X

N=# MF s 1 F AbsError = Actual EP S EP S of MF s it N | itk − itn| n X

Above F AbsErrorit and F AbsErrorit are average measures for forecast bias and forecast error for firm i in year t; subscript n denotes each individual forecast in a fiscal year and k denotes the actual announcement for the corresponding fiscal period.

Alternatively, as shown in Table 1.B, I code whether or not the actual EPS matches that of MFs. Speciﬁcally, the dummy variable Accuracy takes the value of unity if the ex-post realized EPS matches the forecasted EPS by the management

(for point forecast) or falls within the range of management forecasts (for range or open-ended forecast). In addition, to measure the level of optimism in MFs, I code whether or not MFs beat the corresponding consensus of analyst forecasts CHAPTER 3: DELEGATION, INFORMATION AND MARKET 189 for the same ﬁscal period. Optimism takes the value of unity if the management forecasted EPS (point forecast) or its lower bound (range or open-ended forecast) is greater than the analyst forecasted EPS. I aggregate the metrics Accuracy and

Optimism to the firm-year level by taking the average of Accuracy and Optimism for every firm in each fiscal year. More specifically:

N 1 unity if MF s matches Actuals Accuarcy = { } it N Litn n X

N 1 unity if MF s beats Analyst Consensus Optimism = { } it N Litn n X Similar to F AbsBias and F AbsError, I also take the sum of Accuracy and

Optimism for every firm in each fiscal year. Notice that the two main proxies are constructed using the information of both analyst forecast and management guidance from I/B/E/S while the additional proxies Accuracy and Optimism are constructed solely based on the data file of management guidance from I/B/E/S.

Moreover, Optimism is the only one of the four that diﬀerentiate upward or downward bias while the other three measures are purely about the quality of information production.

In order to avoid spurious statistical inference due to endogeneity and to clearly identify the causal relationship between responsibility sharing in the top management team and the quality of voluntary ﬁnancial disclosure, I propose to use the a comprehensive framework of a system of regression equations, which alleviates the empirical concerns about selection bias and reverse causality. The ﬁrst and CHAPTER 3: DELEGATION, INFORMATION AND MARKET 190 main equation of the system is:

2 0 ˆ MF s Inverse quality = β0 +β1δ +β2δ +β3xit 1 +β4δ +β5imr +γk +λt +εit (1) −

The first equation is main equation of interest for this study and the other two equations which I present below are supporting equations to mitigate the potential confounding effects of reverse causality and selection bias. The dependent variables of the first equation are metrics of MFs quality: F AbsBias, F AbsError,

Accuracy and Optimism. xit represents a vector of control variables of firm characteristics and board characteristics at the beginning of the fiscal year. Specifically, the first firm control variable is Size which is the logarithm of the total assets

(Log(assets)). I include the variable Leverage defined as the sum of long-term debt plus short-term debt in current liabilities divided by beginning period total assets. Acharya et al.(2011) show that the level of responsibility sharing ( δ ) is related to corporate governance. Hence, I control for corporate governance by including board characteristics such as the number of board directors (Directors) and the percentage of outside directors (Outsiders). The model specification also includes time (λk) and industry fixed effects (γk) defined at two-digit SIC level.

In addition, econometric model is estimated using heteroscedasticity consistent standard error (HCSE) for statistical inference. Notice that the measures for disclosure quality are constructed based either on the absolute diﬀerence or on the count data. According to prior econometrics research (see, for example, Abrevaya

1999; O’Hara and Kotze 2010), applying log-transformation to mitigate highly CHAPTER 3: DELEGATION, INFORMATION AND MARKET 191 skewed dependent variables often leads to biased estimates. To properly address asymmetrically distributed dependent variables, Cameron and Trivedi(2013) rec- ommend using Generalized Linear Models (GLM). In fact, extensive studies in the literature of ﬁnance and economics use GLM in a variety of settings where the dependent variables can hardly be assumed normal and are typically of qualitative nature (see, for example, Huizinga and Voget 2009; Lerner 2006; Lerner,

Sorensen, and Str¨omberg 2011). Therefore, in addition to OLS regression, I use maximum likelihood estimation for GLM to improve the eﬃciency of equation 1.

In particular, I conjecture that the dependent variables of quality metrics follow a

Gamma or Poisson distribution depending on the metric construction, and apply log link function to the GLM estimator to ameliorate the concern of left censorship and asymmetric distribution. The choice of distribution family and link function is supported in the subsequent empirical tests. The reason I do not include fixed effects at firm level in the model specification is that firstly, both disclosure quality and power sharing measures are sticky by firm across years; secondly, high dimensional fixed effects typically should not be included in Probit model and

GLM.

Estimating the ﬁrst equation alone may result in spurious statistical inference because of selection bias since it is completely the discretionary choice of the management to provide MFs. Therefore, I conduct a Heckman correction by estimating the following Probit model in the ﬁrst stage prior to the main equation.

In particular, CHAPTER 3: DELEGATION, INFORMATION AND MARKET 192

0 0 F orecast = φ0 + φ1xit 1 + φ2zit+k + ρt + uit + γk + λt + εit (2) −

In the ﬁrst stage Heckman model, I control for an inclusive set of control variables

(z) that aﬀect the likelihood of ﬁrms issuing MFs. (see, for example, Ajinkya et al. 2005; Ajinkya and Gift 1984; Baginski et al. 2002; Bamber et al. 2010;

Lang and Lundholm 1993). Speciﬁcally, since the prior literature has shown that management is less likely to issue forecasts when earnings are more uncertain and thus more diﬃcult to predict, I control for the absolute value of the change in

EPS scaled by lagged stock price (” ∆EPS ” ) and the standard deviation of | | analysts forecast deflated by the median forecast (Disp). It has been shown in the literature that when firms either miss their previous target EPS or report losses are less likely to issue MFs. Accordingly, I include a dummy variable Loss that takes the value of unity if the firm reports a loss. I control for the analyst coverage of the firm by including the number of analysts (#Analysts) following the firm. Given that some industries have higher litigation risk and require higher demand on transparency by investors, I include industry fixed effects in the Probit regression of Heckman model.

In addition to selection bias, one might argue that there might be endogenous feedback from voluntary disclosure quality to responsibility sharing in the sense that when the ﬁrm chooses to improve the quality of voluntary ﬁnancial reporting, there might be more disclosure relevant tasks and positions created for managers.

Therefore, driven by the potential unobservable and time variant ﬁrm choice, the parameter estimates might be aﬀected by the endogenous feedback from disclosure CHAPTER 3: DELEGATION, INFORMATION AND MARKET 193 quality to responsibility sharing measure (δ). To ameliorate such concerns, I propose to use instrumental variable of CEO power index (P owerIndex) in equation

(3), which measures the overall degree of power of the CEO besides δ: prestige power, expertise power and ownership power.

0 δ = ϕ0 + ϕ1P owerIndex + ϕ3xit 1 + γk + λt + εit (3) −

The speciﬁc construction of P owerIndex is delineated in Table 1C. According to Finkelstein(1992), there are four dimensions of CEO power that are inter- related: structural power, ownership power, expert power and prestige power.

The degree of responsibility sharing (δ) mainly measures the structural power and is positively correlated with other aspects of executive powers. Moreover, there is no economic reason to believe that a powerful CEO would direct a better or worse disclosure policy. Therefore, P owerIndex is in theory a valid and strong instrument for responsibility sharing measure (δ) in equation 1. To support the argument statistically, I conduct a battery of tests to examine the validity and relevance of the instrumental variable, P owerIndex. Since I only have one instrument for the single endogenous variable δ , the Hansen-Sargan test (J-test) for system identiﬁcation is not applicable. Thus, I choose to use a rule of thumb test according to the deﬁnition of validity (see, for example, Wooldridge 2003) whereby testing the correlation between P owerIndex and the residuals from equation (1).

The correlation between P owerIndex and residuals for various quality metrics are in general very low (less than 1 %). For instance, the correlation between

P owerIndex and residuals of the regression on aggregate F AbsError is -0.0002 CHAPTER 3: DELEGATION, INFORMATION AND MARKET 194

(p-value= 0.986), suggesting the validity of P owerIndex. I follow Stock, Wright, and Yogo(2002) and Stock and Yogo(2005), and use the minimum eigenvalue of

Cragg-Donald statistic to test the relevance and strength of the instrument (also known as Stock, Wright and Yogo F-test). When there is only one instrumental variable, the minimum eigenvalue of Cragg-Donald statistic is F-statistic in the

ﬁrst stage regression. In the section of Results and Analyses, I show that the value of F-statistic (315) is much larger than the threshold value (10), indicative of a very strong instrument. Please see the results section for further details.

According to Wooldridge(2003), the instrumental bias compared to OLS is inversely proportional to the strength and validity of the instrument. Thus I can conclude that P owerIndex is an eﬀective instrumental variable which mitigates reverse causality while introducing limited estimation bias.

Notice that in our framework of empirical investigation, the main equation

(equation 1) of interest is essentially nonlinear. The nonlinear structure of equation 1 is of twofold. Firstly, the relationship between responsibility sharing and disclosure is quadratic and secondly, provided that the dependent variable is left censored and asymmetric, the functional form of the model is nonlinear in parameter (GLM) and is estimated using maximum likelihood estimation. Hence, I adopt the method of control function (CF), which is widely used in nonlinear models (see, for example, Wooldridge 2015). To implement CF, I need to estimate equation (3) in the ﬁrst stage and include the predicted responsibility sharing measure (δ) in the equation 1, the main model of interest. Applying this comprehensive framework of empirical investigation, I could mitigate the concerns of potential sample CHAPTER 3: DELEGATION, INFORMATION AND MARKET 195 selection bias and reverse causality, thereby identifying the relationship between responsibility sharing and quality of voluntary ﬁnancial disclosure.

Provided that the estimations of the above individual equations are interre- lated, I could organize the above three equations together in a coherent system as follows. Specifically, as explained above, the first equation at the top of the list is the main equation of interest which is estimated in the second stage, while the latter two are complementary equations estimated in the first stage to account for the potential confounding effects of the selection bias and reverse causality.

2 0 ˆ MF s Inverse Quality = β0 + β1δ + β2δ + β3xit 1 + β4δ + β5imr  −  MF MF MF  +γk + λt + εit      (4)   0 0 F F F  F orecast = φ0 + φ1xit 1 + φ2zit+k + ρt + uit + γk + λt + εit −      0 δ δ δ  δ = ϕ0 + ϕ1P owerIndex + ϕ3xit 1 + γk + λt + εit  −   

4.1.2 Stock liquidity

To examine the relationship between certain delegation structures of informational eﬃciency and the liquidity of the company’s shares as posited by Hypotheses

4a and 4b, I adopt two metrics for stock liquidity which are widely used in the literature of market microstructure. The ﬁrst is average daily stock turnover speciﬁed as follows, CHAPTER 3: DELEGATION, INFORMATION AND MARKET 196

Nit 1 Daily T rading V olumeikt T urnoverit = 100% Nit Shares Outstandingikt × Xk in which subscripts i and t denotes firm and fiscal year, respectively; k denotes trading days in a fiscal year and Nit is the number of trading days for the corresponding firm in the fiscal year. Turnover measure is widely used in the studies for cross-sectional analyses of stock and bond liquidity (see, for example, Dick-

Nielsen, Feldhütter,and Lando 2012; Jain et al. 2016; Mahanti, Nashikkar, Sub- rahmanyam, Chacko, and Mallik 2008). The daily turnover measure is expressed in percentage and calculated as daily trading volume (number of shares) scaled by shares outstanding as of the trading day. Consistent with the firm-year unit of analysis in this study, I take the annual average of daily turnover measures to formulate the first liquidity metric, T urnover. To account for the confounding effect of extremely large or small trades during special periods of market movements, I also take the median of daily turnover measure as a comparison in the regression analysis. The second liquidity measure is the well-known Amihud Price Impact initially developed in Amihud(2002),

Nit 1 Daily Stock Returnikt Amihudit = | | Nit Daily Dollar T rading V olumeikt Xk in which subscript i and t denotes firm and fiscal year, respectively; k denotes the trading day and Nit is the number of trading days for the corresponding firm in the fiscal year. Consistent with the unit of analyses in this paper, Amihud Price

Impact liquidity measure (Amihud) is calculated by taking the annual average CHAPTER 3: DELEGATION, INFORMATION AND MARKET 197 of absolute daily stock return scaled by daily dollar trading volume (in millions).

As a comparison, I also use annual median value to account for the potential abnormal trading activities during special market periods of the year. Notice that trading volume is the important component for both of the above two liquidity metrics. However, the trading volume of the NASDAQ listed companies is arguably overstated due to prevailing trading activities among dealers. Thus, following the literature (see, for example, Atkins and Dyl 1997; Jain et al. 2016;

Lipson and Mortal 2009), I divide trading volume of NASDAQ stocks by 2 to adjust for dealers’ trading activities.

Next, I identify the range of informationally eﬃcient degrees of delegation for regression analysis with respect to two informational regimes: centralization and delegation. In particular, I use the parameter estimates of the regression models for

F AbsError to identify the informationally eﬃcient regimes, since F AbsError captures the absolute diﬀerence between the MFs and the earnings of the actual announcement and is considered the most direct measure for the capability of internal information production. If the empirical evidence is in support of Hypothesis

1A, the model-suggested optimal degree of delegation is δ∗ = β /(2β ), and the − 1 2 optimal range is as follows:

1 1 (δ∗ σ , δ∗ + σ ) − 2 δ 2 δ

where σδ is the sample standard deviation of δ. However if the empirical evidence in support of Hypothesis 1B, the range for information eﬃciency is as follows: CHAPTER 3: DELEGATION, INFORMATION AND MARKET 198

(δ , δ∗ 2σ ) (δ∗ + 2σ , δ ) min − δ ∪ δ max

Thus, the econometric model for examining the impact of the underlying mechanism upon stock liquidity is structured as follows,

0 StockLiquidity = β0 + β1Info + β3 xit 1 + γi + λt + εit (5) − in which the dependent variables are liquidity metrics for stock; the variable of interest is a dummy variable info which equals unity if δ is within the optimal range; x is a vector of controls of firm and board characteristics at the beginning of the fiscal year; γi and λt are firm and year fixed effects, respectively. In addition to the covariates from the previous specifications, I incorporate additional variables that have been shown explaining significant portion of cross-sectional and time-series variation in stock liquidity in the previous studies (see, for example, Chung, Van Ness, and Van Ness 1999; McInish and Wood 1992; Stoll 2000).

Specifically, I include stock average daily closing price (P rice), annualized return volatility (V olatility), average daily dollar trading volume (V olume), S&P 500 return (S&P ) and dividend per share (Dividend). Since the market liquidity is unlikely to causally affect the delegation structure of the top management team, the regression analysis is less subject to the confounding effects of reverse causality. Provided that liquidity metrics vary significantly within firm across years, I include granular firm fixed effects in the linear regression model, which further mitigates the concern of missing unobserved endogenous variables. To account CHAPTER 3: DELEGATION, INFORMATION AND MARKET 199 for correlation of residuals within pairs of firm and year, the standard errors of the regression model are two-way clustered at the firm and year level. Since the dependent variable is highly asymmetric and not normally distributed, in addition to OLS, I also utilize estimators of GLM by maximum likelihood estimation to improve efficiency.

4.2 Data, Sample Construction and Summary Statistics

The sample construction begins by obtaining executive titles of CEO and the other subordinate managers of S&P 1500 firms from annual compensation dataset of Execucomp from 1996 to 2017. I only keep the observations of executives with no missing titles. The time horizon of the sample starts from 1996, the starting point for ISS board of directors data. It is necessary to control board characteristics, because the degree of delegation responsibilities is endogenous to corporate governance mechanisms (see, for example, Acharya et al. 2011). The initial sample of executive title strings consists of 201,900 observations, and is of manager-fiscal year level. Next, I calculate the main variable of interest, δ, a metric for the degree of delegation responsibilities between the CEO and her subordinates in the top management team. Per firm-fiscal year, I sort the executives by total compensation (Execucomp item: TDC1) and select the top five well-paid managers. I drop the management teams that have more than one CEO and report executive titles for less than five executives. The resulting sample of δ consists of 31287 firm-fiscal year observations. δ is a proxy for the power allocation between CEO and subordinate managers and relative centrality of the management team. To exercise CHAPTER 3: DELEGATION, INFORMATION AND MARKET 200 caution, I further drop the observations in which CEO identifier (Execucomp item

CEOANN) and/or date of CEO appointment (Execucomp item BECAMCEO) are missing, thereby removing the management teams without clearly identiﬁed lead

CEO from the sample.

Next, I merge the sample of δ with I/B/E/S ticker using the Link Table for

Compustat and I/B/E/S provided by Wharton Research Data Services (WRDS).

I/B/E/S contains both analyst forecasts and management guidance. I drop the

firm-fiscal year observations whose linking information is missing in the Link Ta- ble. The study is focused on the quality of corporate voluntary financial disclosure.

The metrics of voluntary disclosure quality specified in the prior section are based on the measure of EPS. Thus, I merge the data of EPS of MFs, EPS of analysts’ consensus forecasts and the actual EPS of announcement in every fiscal quarter from I/B/E/S with the sample of delegation responsibilities, δ. The joint sample also includes variables for the attributes of managerial forecast used in Table 1.B4 and Table 1.C, such as RANGE DESC and GUIDANCE CODE. The resultant sample consists of 102,120 observations and is at firm-quarter level. Provided that the unit of this study is firm-year, I aggregate the data over the four quarters to calculate (annual) metrics specified in the prior section. The consolidated sample of δ and disclosure quality consists of 28,816 firm-fiscal year observations5. Next, in order to formulate the regression analysis of structural equations, I merge the sample with Compustat, CRSP and ISS to incorporate control variables of firm

4 Note that there is no RANGE DESC code of 5 or 7 in my sample. 5 It is possible that firms do not provide management forecasts in every quarter. In order to remove the possibility of selection bias, we use Heckman analysis to remove the bias. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 201 fundamentals, security prices and board characteristics, such as total assets, leverage, market to book ratio, research and development, number of directors and so on, as proxies for size, market valuation, capital structure, CEO power, corporate governance and likelihood of voluntary information provision. To construct liquidity measures, I download daily stock return, closing price and volume from CRSP for S&P 1500 firms in the sample. The detailed variable definitions and sample construction procedure are specified in Table 1.A and Table1.D, respectively.

The descriptive statistics are summarized in panels of Table 2. The unit of the data is at the firm-fiscal year level. Specifically, Table 2.A reports the summary statistics of the whole sample. Table 2.B provides the results of the test of difference between control variables of the two subsets grouped by whether nor not firms participate in voluntary disclosure activities in a fiscal year.The average value of δ is 0.263, which is quite close to the median value of δ at 0.250, indicating a symmetric bell shape distribution of δ. Provided that the standard deviation of

δ is 0.068 and the distribution of δ is presumably normal, the majority of observations of δ (roughly 95%) should fall into the value range between 0.1 and 0.4.

The sample distribution of δ is generally consistent with that of δ in Aggarwal et al.(2017). The slight discrepancy might be attributed to the long-term structural change of delegation responsibilities in the top team since the 2008 financial crisis that is less represented in the sample of Aggarwal et al.(2017), or to the improved operationalization of the internal governance measure (δ) in this study using the text mining technique. The empirical results of test of difference using either independent group t-test or Wilcoxon z-test evidently indicate that whether or CHAPTER 3: DELEGATION, INFORMATION AND MARKET 202 not firms choose to engage in voluntary information provision is largely correlated with a variety of economic variables of firm characteristics, board characteristics, and external information environment conditions. Albeit the paper constructs the sample in accordance with δ and utilizes data over a much longer time horizon, the documented patterns are consistent with the findings in the prior literature.

Speciﬁcally, ﬁrms which participate in voluntary disclosure tend to have higher valuation measured by M/B (1.291 compared to 1.075) due to lower information risk and cost of capital (see, for example, Botosan 1997; Graham et al. 2005).

Consistent with the negative relationship between risk profile and propensity of voluntary disclosure, firms that voluntarily provide financial information tend to bear lower financial risk measured by Leverage (0.233 compared to 0.248). In line with the fact that larger, more heavily publicized firms issue more voluntary disclosures (see, for example, Ajinkya et al. 2005; Lang and Lundholm 1993),

ﬁrms that issue MFs are on average lager (Size, 7.809 compared to 7.610) and are exposed to more analyst earnings forecasts (#Analyst, 14.992 compared to

12.096). In non-tabulated results, I check the heterogeneity in R&D, which arguably measures proprietary information cost, between the two groups for years before Regulation Fair Disclosure (Reg FD) in effect. The test of difference result shows that firms that choose not to participate in voluntary financial disclosure on average have more R&D expenditures (R&D, 0.052 compared to 0.041), which is consistent with the findings in Wang(2007). Consistent with Ajinkya et al.(2005), the average value of the fraction of outsiders on the board of voluntary disclosure group (0.731) is higher than that of control group (0.707), indicative of a positive CHAPTER 3: DELEGATION, INFORMATION AND MARKET 203 relationship between propensity of voluntary information provision and quality of corporate governance. Analogous to prior research (see, for example, Ajinkya et al. 2005; Ajinkya and Gift 1984; Baginski et al. 2002), firms that conduct voluntary disclosures face less earnings uncertainty as suggested by the differences in

∆EPS , Disp and Loss. Consistent with the literature on ﬁnancial information | | provision and asset liquidity (see, for example, Balakrishnan et al. 2014; Bardos

2011; Lester et al. 2011), liquidity, as measured by T urnover, of firms that engage in voluntary information provision is higher than those do not. Analogously, illiquidity, as measured by Amihud, of firms that engage in voluntary information provision is lower than those do not participate in voluntary disclosure. All the above differences between the group that engages in voluntary disclosures and the one that doesn’t are statistically significant at 1%.

Interestingly, the average value of δ for firms that conduct voluntary disclosure is 0.261 while the corresponding figure for firms that do not issue MFs is 0.265.

In contrast, CEO of voluntary disclosure ﬁrms are more powerful as measured by

P owerIndex (2.229 compared to 2.178). Taken together, the descriptive statistics suggest that ﬁrms of high propensity of voluntary information provision are likely led by powerful CEOs who hold the helm strongly, but also eﬀectively delegate executive duties to her subordinates. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 204

5 Results and Analyses

5.1 Quality of Voluntary Disclosure

The empirical results pertaining to the functional form between the quality of corporate voluntary financial disclosure and the degree of delegation, the key research question of the study, are summarized in the following sections. Four metrics for the quality of management earnings forecasts (MFs) are used in the empirical tests. As introduced in the section of Methodology and Sample, two main metrics are forecast bias (F AbsBias) and forecast error (F AbsError) and the two additional metrics are Accuracy and Optimism. Because firms could decide on whether or not to issue guidance, the economic meanings of the average and aggregate measures are different. Average measure is better than aggregate measure in the sense that it alleviates the measurement bias when a firm issues more guidance than another. For example, a firm (A) that issues 3 guidance may end up with higher accumulated bias than another firm (B) that only issues 1 in a fiscal year. However, we cannot assert the information production of B is superior to

A if the average bias of A is lower than that of B. Conversely, aggregate measure is better than average measure in the sense that it captures firms’ overconfidence for accurate prediction, which itself is a bad sign for information production. In this regard, a firm that issues multiple guidance and accumulates a considerable amount of errors, despite similar average mistakes, might be inferior than the firm with less accumulated errors. Therefore, due to the discretionary nature of the CHAPTER 3: DELEGATION, INFORMATION AND MARKET 205 management guidance and the different economic meanings of the average and aggregate measures, I use both for complementary purpose.

In general, the empirical results are in support of Hypothesis 1B of the two competing hypotheses. The parameter estimates indicate a hump-shaped relationship between executive duties sharing in the top management team (δ) and inverse quality of voluntary disclosure. The forecast quality is inferior for an interior level of δ as opposed to either δ = 0 or δ = 1. Thus, there does not exist an internal optimality of δ for voluntary disclosure quality. The quality of voluntary disclosure is better when the CEO is in total control or fully delegates the responsibilities to her subordinate managers. Moreover, consistent with the prediction made in

Hypothesis 2 and Hypothesis 3, as CEO ages, the relationship between responsibility sharing and inverse voluntary disclosure quality is more inclined to a hump-shaped curve.

5.1.1 Graphical Analysis

Without regard to any econometric models and identification strategies, I first conduct a graphical analysis to demonstrate the empirical implications in a figu- rative manner. The key link on which the paper investigates functional relation between quality of voluntary disclosure and the degree of delegation responsibilities between the CEO and her subordinates. The graphical analysis delineates the curvilinear form of the relationship between δ and information efficiency and as a result helps specify the econometric specification. Hence, I plot the measures of voluntary disclosure quality against the degree of delegation δ in figure 1 through CHAPTER 3: DELEGATION, INFORMATION AND MARKET 206

4. Each ﬁgure consists of three panels, A, B and C, which represents the whole sample and the subsamples by age, respectively. Y-axis of the plots denotes a certain measure of voluntary disclosure quality and X-axis denotes the percentile of δ ranging from 1% to 100%.

In particular, Figure 1 is focused on the quality of voluntary disclosure as measured by forecast bias. In Figure 1.A, the box plot is produced whereby plotting measures of forecast bias, aggregate F AbsBias and average F AbsBias, against percentiles of δ. The red line plot connects the median level of forecast bias for each group classiﬁed by the percentile of δ. The signals embedded in the data are subtle. To further tune-out noises embedded in the data and identify the relationship between disclosure quality and responsibility sharing, I employ the widely used curve-smoothing method of Local Polynomial Regression Fitting.

The black dash line is the fitted curve produced using the predicted median values of each group. For visualization purpose, the scales on the Y-axis of the two line plots are different from that of box plot. To separate one plot from the other, the Y-axis values of red solid plot and black dash plot are tripled and quintupled, respectively. As indicated in Figure 1.A, albeit generally vague, the forecast error increases in δ at the beginning and then decreases after passing a certain inflection point. Such a pattern suggests a hump-shaped relationship between forecast bias and δ, or, in other words, an inverted hump-shaped relationship between voluntary disclosure quality and delegation responsibilities in the top management team.

The functional form of the curvilinear relation indicates that an internal optimum of responsibility sharing for information production and organizational learning CHAPTER 3: DELEGATION, INFORMATION AND MARKET 207 does not exist. In Figure 1.B, I produce the same plots of aggregate F AbsBias against groups of δ percentiles for older CEOs and younger CEOs, respectively.

Consistent with the prediction in Hypothesis 2, a much stronger signal for the underlying hump-shaped relationship is revealed by the sample of older CEOs. As shown in the left picture of Figure 1.B, the curvature of dash line of predicted median values is concave downward to a much larger extent than that in Figure

1.A, suggesting that forecast bias is diminished when the CEO is in total control or fully delegates responsibilities to her subordinates. Partial delegation could be detrimental to the internal information production and knowledge sharing for management teams led by aging CEOs. Interestingly, in contrast to the scenario of older CEO, there seemingly exists upward concave relationship between forecast bias and the degree of delegation for management teams led by younger CEOs.

Similar results consistent with Hypothesis 2 are also obtained in Figure 1.C, using the metric, average F AbsBias for voluntary disclosure quality.

Similar patterns, if not even more discernible, can be observed in the three panels of Figure 2, in which the quality metric is F AbsError. However, the hump- shaped functional form becomes less discernible for additional quality measures,

Accuracy and Optimism. Due to the qualitative nature of alternative metrics for disclosure quality, I use decile instead of percentile to group degree of delegation,

δ. For the same reason, in addition to median, I use solid diamond in red to denote group mean for each box to show the variation of quality metrics across quantile categories. The signal revealed by the alternative metrics in box plot is less clear compared to that of forecast bias and error. Such heterogeneity is CHAPTER 3: DELEGATION, INFORMATION AND MARKET 208 driven by the qualitative nature as well as the economic meaning of the additional quality measures. Especially, Optimism is diﬀerent from other measures in that it disregards downward bias, while others are pure quality measures, indiﬀerent between upward and downward bias. In this regard, the two main measures are superior to the additional ones for the purpose of this study.

5.1.2 Simple Quadratic Regression Analysis

Accompanying with the specification-free graphical analysis, I report in Table 3 the results of regressing degree of delegation, δ, on the four voluntary disclosure quality metrics in the simple quadratic form with year and industry fixed effects. Panels A,

B and C of Table 3 summarizes the regression results for the whole sample, sample of older CEOs (greater than or equal to the sample median age, 56) and the sample of younger CEOs (less than the sample median age, 56), respectively. In each panel, I use both average and aggregate quality metrics for comparison. In general, the results are consistent with the implications of graphical analysis as well as the predictions in Hypotheses 1.B. The hump-shaped functional relation between δ and inverse quality of voluntary disclosure modeled by the quadratic specification is statistically significant (i.e., the coefficient for the linear term of δ is significantly positive and the coefficient for the quadratic term of δ is significantly negative) for the two main metrics, F AbsBias and F AbsError. Consistent with Hypothesis

2 and Hypothesis 3, the statistical signiﬁcance of the simple quadratic regression improves in panel B of Table 3 for older CEOs and diminishes in panel C of Table

3 for younger CEOs. In unreported sensitivity test, the patterns become more CHAPTER 3: DELEGATION, INFORMATION AND MARKET 209 significant for even older CEOs and less significant when the sample includes younger CEOs. Notice that regressions on Optimism, as the only metric that captures the directional bias, has generally statistically insignificant parameter estimates. For brevity, the reported results are estimated by Generalized Linear

Model (GLM) only. The results are similar using other estimation methods such as OLS and Tobit.

5.1.3 Multivariate Regression Analysis

In addition to the graphical analysis and simple quadratic regression analysis, I conduct multivariate analysis of regressing degree of delegation, δ, on the four voluntary disclosure quality metrics in the quadratic specification with firm and board covariates as defined in the section of Methodology and Sample. The multivariate regression is to isolate the effect of δ and to improve the explanatory power of the econometric model. The empirical results are summarized in panel A, B and C of Table 4 for the whole sample, the sample of older CEOs (age greater or equal to the sample median, 56) and the sample of younger CEOs (age less than the sample median, 56). In each panel, I use both average and aggregate quality metrics for comparison purpose. Similar to the simple quadratic analysis, the results are consistent with the predictions in Hypotheses 1B. The hump-shaped functional form between δ and inverse quality of voluntary disclosure modeled by the quadratic specification with firm and board controls is statistically significant for the two main metrics, F AbsBias and F AbsError. Moreover, the statistical significance of the quadratic regression improves in Panel B of Table 4 for older CHAPTER 3: DELEGATION, INFORMATION AND MARKET 210

CEOs and diminishes in panel C of Table 4 for younger CEOs, illustrating an evolving pattern aligned with Hypothesis 2 and Hypothesis 3. Overall, due to higher explanatory power of multivariate econometric model, the statistical results reported in Table 4 are stronger than those in Table 3, further verifying that the eﬀect of δ is not merely a manifestation of other ﬁrm and board fundamentals.

In general, regressing degree of delegation, δ, on additional metrics, Accuracy and Optimism, results in less signiﬁcant parameter estimates. For example, in

Panel A of Table 4, when the empirical tests are performed on the whole sample, the parameter estimates of δ for both Accuracy and Optimism turns statistically insigniﬁcant as shown for the simple quadratic regression. It indicates that the mechanism that underlies the relationship between quality of voluntary disclosure and degree of delegation is not about internal monitoring or control but purely about the capability of information production determined by the delegation structures. For brevity, the reported results are estimated by Generalized Linear Model

(GLM) only. The results are similar using other estimation methods introduced in the section of Methodology and Sample such as OLS and Tobit.

5.1.4 System of Structural Equations

In order to identify the causal relationship between voluntary disclosure quality and the degree of responsibility sharing in the top management team, I estimate a system of structural equations in additional to multivariate regression analysis.

The empirical results of multivariate regression are subject to the selection bias due to the discretionary nature of the voluntary disclosure of managerial forecasts. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 211

Therefore, I conduct a Heckman correction whereby estimating a Probit model in the ﬁrst stage prior to the main equation. In addition to selection bias, I use

P owerIndex as the instrumental variable and apply the method of Control Func- tion (CF) to mitigate endogenous feedback from voluntary disclosure activities to

δ.

The results are summarized in Panels A, B and C of Table 5. For brevity, the empirical analysis of structural equations is focused on the sample of older CEOs, for which the hump-shaped relationship is shown statistically signiﬁcant. The

Panels A, B and C of Table 5 summarize the empirical results after controlling for selection bias, endogeneity, and both, respectively. Speciﬁcally, Table 5.A reports the ﬁrst stage Probit regression of Heckman correction using a set of economic determinants in the literature. The Pearson chi-squared statistic for goodness of

fit test is not statistically significant, indicating the validity of the Probit model for predicting the likelihood of firms issuing MFs. Columns (1) – (8) reports the regression results in the second stage for the aggregate and average quality metrics of earnings forecast using GLM with the inclusion of inverse mills ratio. The coefficients of the variables of interest, δ and δ2 for two major measures, F AbsBias and F AbsError, are statistically significant at 1% for both average and aggregate metrics. As expected, the statistical significance of two additional metrics,

Accuracy and Optimism, is weaker than the major quality metrics. Especially, the quadratic functional form becomes insignificant for Optimism, which captures only the upward bias of MFs. Similarly, Table 5.B reports the first stage regression for Control Function (CF) and the regression results including the endogeneity CHAPTER 3: DELEGATION, INFORMATION AND MARKET 212 control attained from the first stage. Notice that the coefficient of P owerIndex is positive and significant at 1%, suggesting a strong instrument. Moreover, the overall explanatory power of the first stage regression is large, as indicated by the

F-statistic (315.5). The predicted δ (δ) by the ﬁrst stage regression is then used as the endogeneity control in models of columnb (1) through (8). Analogous to results in Table 5.B, the statistical signiﬁcance of δ and δ2 is at 1% for F AbsBias and

F AbsError, while it diminishes for Accuracy and Optimism. Table 5.C reports the regression results in combination of Heckman correction and Control Function method, the results are stronger in general. For brevity, the reported results are estimated by Generalized Linear Model (GLM) only. The results are similar using other estimation methods such as OLS and Tobit.

5.2 Stock Liquidity

Next, the paper conducts empirical tests to investigate the impact of the improved quality of voluntary disclosure driven by informationally efficient delegation structures on stock liquidity. In other words, the study strives to find out whether the liquidity of company’s shares increase in face of certain degrees of delegation which are deemed efficient informationally. According to the section of Hypothesis

Development, the theoretical model in Appendix A, and the above empirical results, the functional relation between degree of delegation and quality of voluntary disclosure is of hump-shaped (U-shape) curvilinear form. Thus, as predicted in

Hypothesis 4B, the stock liquidity should improve in either of the informationally eﬃcient regimes, namely, either the CEO or her subordinates are in control, CHAPTER 3: DELEGATION, INFORMATION AND MARKET 213 but deteriorate around the interior inﬂection point. According to the section of

Methodology and Sample, I choose to use the estimated inﬂection point drawn from the parametrization of the quadratic model for forecast error. Interestingly, the estimated inﬂection point (δ∗) is largely approximate to the sample mean (for

CEOs whose age is 56 or greater) of degree of delegation (δ). According to Cheby- shev’s Inequality (Chebyshev 1867), the method statistically identiﬁes about 25% observations of the sample as informationally eﬃcient delegation structure.

Speciﬁcally, Panels A, B and C of Table 6 summarizes the empirical results of regressing liquidity measures (T urnover, Amihud) on the dummy variable of interest (Info ) and other controls for the whole sample, older CEOs and younger

CEOs, respectively. As a proxy for trading intensity, T urnover measures how many shares a company’s stock are traded daily relative to its shares outstanding, which is positively related to stock liquidity. In contrast, as a proxy for illiquidity, Amihud measures how much price changes as a result of trading, which is negatively related to stock liquidity. In general, the statistical significance of the parameter estimates of Info across three panels of Table 6 is analogous with the variation of the documented relationships between degree of delegation and quality of voluntary disclosure by age cohort. In particular, Table 6.B reports the regression results for the sample of CEO age greater than or equal to 56. In column (1) through (4), Table 6.B, the coefficients of Info are positively and significant at 1% for T urnover and negative and significant at 1% for Amihud, consistent with the predications in Hypothesis 4B. The results indicate that stock liquidity improves when CEO is fundamentally different from his subordinates in executive horizon CHAPTER 3: DELEGATION, INFORMATION AND MARKET 214 and managerial preferences and when either the CEO takes the total control or fully delegates to her subordinates. The statistical significance of the coefficient of Info is qualitatively similar in Table 6.A for the whole sample. However, the coefficient of Info becomes statistically insignificant for the sample of younger

CEOs, which is consistent with the finding that the curvilinear relation diminishes as including younger CEOs in the sample. The statistical significance of the coefficients for T urnover and Amihud varies consistently by CEO age cohort. For brevity, the reported results are estimated by Generalized Linear Model (GLM) only. The results are similar using other estimation methods such as OLS and

Tobit.

5.3 Value Relevance of the Liquidity Eﬀect

Albeit production irrelevant, voluntary disclosure policy is value relevant for the

ﬁrm. In the previous section, I have shown that when the delegation structure falls in certain ranges of the two informationally eﬃcient regimes, the liquidity of the

firm’s shares improves accordingly. Thus it is interesting to examine whether or not the liquidity variation attributable to delegation structures would be correctly priced by the investors. According to Hypothesis 5A and 5B, if the realized superior liquidity is unexpected to investors or they underreact to the liquidity variation, the firms managed by informationally efficient regimes will be under- valued and thus generate higher future stock returns; if investors overreact to the liquidity variation, the firms managed by informationally efficient regimes will be overvalued and thus generate lower future stock returns. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 215

5.3.1 Portfolio Analysis

Panels A, B and C of Table 7 report the results of the portfolio analyses. I create δ portfolios by ranking the value of δ at the beginning of the ﬁscal year into quintiles. Thus the portfolios are formed every year. I use monthly data of ﬁrms led by older CEOs to calculate both equal-weighted and value-weighted portfolio returns (in percentage) every month over the period of 1996 to 2017.

Following Shumway(1997), the stock returns are adjusted for delisting to avoid survivorship bias. Value-weighted portfolio is used to rule out the effect of small- cap stocks. The weightings of value-weighted portfolio are determined using the market capitalizations of constituent stocks at the beginning of the month. I sum the shares outstanding of dual class stocks to calculate market capitalization. To further control the effect of size, which is potentially correlated with both δ and liquidity, I stratify the sample into five size-based quintiles to double sort stocks by both market capitalization and δ. Following the original Fama and French(1993),

I use NYSE size breakpoints to ensure consistency. The returns of the formed portfolios are then regressed on the Fama-French three factors, controlling for the well-known eﬀects of market, size, and valuation.

The portfolio test results by Fama-French three-factor model for equal-weighted, value-weighted and double-sorted portfolios are summarized in Panels A and B of

Table 7. In general, the value and statistical significance of Fama-French three- factor alphas for δ quantile portfolios indicates that the market doesn’t actively incorporate the variation of liquidity driven by delegation structures. In other CHAPTER 3: DELEGATION, INFORMATION AND MARKET 216 words, the contemporaneous improvement of liquidity due to clear delegation structures is largely unexpected to or underestimated by market participants, resulting in higher future stock returns and lower contemporaneous valuation. In particular, for both value-weighted and equal-weighted δ quantile portfolios, there is a U-shaped relationship between values of alpha and the degree of delegation, as shown in the accompanying figures. It suggests that informationally efficient regimes could predict higher future stock returns. Similar patterns could also be observed by the double sorted portfolios, especially for those large and small size portfolios. The empirical results indicate that there exists mispricing of liquidity improvement driven by clear delegation structures. Such a finding is important because it shows the consistency of the study with the empirical works in the literature on internal governance. Essentially, it could serve as a alternative explanation for the salutary effect of internal governance. In addition to the value adding effect of internal governance stemming from the checks and balances among managers, the mispricing on the unexpected variation of liquidity could also sup- ports the contemporaneous valuation premium of firms with a delegation scheme of proportionally responsibility sharing.

5.3.2 Meditation Analysis

To further examine the potential consistency underlies the curvilinear relationship documented in this paper and the one implied by the theory of internal governance

(Acharya et al. 2011), I follow Baron and Kenny(1986) and perform a meditation analysis to investigate whether or not information efficiency mediates the hump- CHAPTER 3: DELEGATION, INFORMATION AND MARKET 217 shaped relationship between firm value and δ. If information content is not an effective channel through which exerts an impact on firm value and financial performance, the variation of stock liquidity driven by delegation structures is not closely value relevant and is not actively priced in stock prices, which in turn rein- force the channel of internal governance through which shareholder value improves due to checks and balances amongst top executives.

According to Baron and Kenny(1986), it includes three steps to examine whether or not a variable of interest, X, exerts an impact on the dependent variable, Y, through a mediator M. Firstly, the dependent variable Y is regressed on

X to check if Y is related to X. If the relation between X and Y is statistically different from zero, then the potential meditator variable M is regressed on the key variable of interest X. If M is significantly related to X in a statistical sense, the last step is to regress Y on X controlling M for the potential meditating effect.

The magnitude and the level of signiﬁcance of the coeﬃcient of X in the last regression is expected to reduce, if M is at least a partial meditator. The empirical results are summarized in Table 8. In particular, I use the average level of Amihud

Price Impact (Amihud) for the fiscal year as the proxy for the efficiency of a firm’s information environment, in that Amihud is widely used in the literature for price informativeness (see, for example, Brennan, Huh, and Subrahmanyam 2013; Jain et al. 2016). I use market to book ratio (M/B) as a performance metric to capture the salutary effect of internal governance, since it reflects the success of strategic management and the firm’s long term adaption to the dynamic business environment (Chakravarthy 1986). I use the sample of older CEOs for the meditation CHAPTER 3: DELEGATION, INFORMATION AND MARKET 218 analysis since internal governance should be effective only if the CEO is old and of short executive horizon. The results are summarized in Table 8. The coefficients of linear and quadratic term of δ in columns (1) and (2) are statistically significant, and their signs are consistent with the findings of this paper for stock liquidity and with the literature on internal governance, respectively. The coefficient of

Amihud is negative and significant in column (3), suggesting that information asymmetry and illiquidity hurts valuation. In column (4), I conduct the last step of the meditation analysis by regressing M/B on δ , δ2 and the potential meditator Amihud. However, the results show that the magnitude and the significance of the coefficients of δ and δ2 doesn’t reduce at all, suggesting that Amihud is not a valid meditator for the curvilinear relation between δ and shareholder value.

In other words, the ﬁndings of this paper regarding stock liquidity and delegation structures do not contrast the value implications of internal governance. The value relevance of δ is stemming from the checks and balances among top executives, but not from the information eﬃcacy and stock liquidity.

Additional tests are reported in Panel A and Panel B of Table 9, where I examine the predictability of δ for future capital appreciation of a company’s share using the whole sample. The dependent variable in Panel A is raw stock return (SReturn) for the following ﬁscal year, whereas the dependent variable in

Panel B is annual cumulative abnormal return by Fama-French three-factor Model and rolling estimation in the past 24 months (CAReturn). Following the above procedure of meditation analysis, I ﬁnd that Amihud largely does not meditate the curvilinear relation between δ and future capital appreciation as measured by CHAPTER 3: DELEGATION, INFORMATION AND MARKET 219

SReturn and CAReturn. For example, in Panel A of Table 9, the coefficients of the linear and quadratic term of δ in column (2) is as statistically significant as those in column (4). In sum, the liquidity effect of delegation structures is not actively priced in stock prices and doesn’t contradict the value relevance of internal governance through the channel of checks and balances among top executives.

5.4 Robustness Check

5.4.1 Alternative Instrument

The construction of P owerIndex includes unitary values of certain titles such as

Chair, Vice Chair and sole executive Director which measure either prestige power or expertise power. One might argue that those components in P owerIndex are highly correlated with the variable of interest, risking the validity of the instrument. However, as explained in the section of Methodology and Sample, the channel of the endogenous feedback is through communication or information col- lection relevant titles, while the title components captured in P owerIndex are more about power in general rather than information or communication oriented.

Hence, P owerIndex has no direct association with the dependent variable and sat- isﬁes the exclusion restriction. Nevertheless, I construct an alternative instrument,

P owerIndex0, whereby excluding all the titles, as a robustness check. Notice that such a practice will artiﬁcially reduce the correlation between P owerIndex and

δ, resulting in weak instrument. Weak instrument in turn may reduce estimation eﬃciency leading to biased and less signiﬁcant results (Bound, Jaeger, and Baker CHAPTER 3: DELEGATION, INFORMATION AND MARKET 220

1993, 1995; Chao and Swanson 2005; Wooldridge 2003).

The results are summarized in Panels A and B of Table 10. For brevity, the empirical analysis of structural equations is focused on the sample of older CEOs, for which the hump-shaped relationship has been shown statistically signiﬁcant.

The Panel A of Table 10 summarizes the empirical results of the regression including the new endogeneity control estimated by the alternative instrument. In

Panel B of Table 10, I include inverse mills ratio and endogeneity control to mitigate selection bias and endogeneity. Specifically, Table 10.A reports the first stage regression for Control Function (CF) and the empirical results of the regression model including the endogeneity control estimated from the first stage. Albeit much weaker than the original instrument, the coefficient of P owerIndex0 is positive and significant at 1%. The predicted δ (δ) by the first stage regression is then used as the endogeneity control in models of columnb (1) through (8). Analogous to results in Table 5.B using original instrument, the statistical significance of δ and

δ2 is at 1% for F AbsBias and F AbsError, while it diminishes for Accuracy and

Optimism. Table 10.B reports the regression results obtained through Heckman correction and Control Function method. Albeit using a much weaker instrument, the empirical results of the quadratic model speciﬁcation are still consistent with those using the original instrument, indicative of the robustness of the statistical inference. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 221

5.4.2 Alternative Delegation Measure

The original measure for the degree of delegation, δ, is constructed using the executive titles of top five well-paid executives. However, some firms may report titles and compensation packages of more than five executives in Form 10-K. One might argue that the inclusion of additional executives in itself is a sign for lifted executive status for those relatively low ranked managers that should not have been covered by the report. If this is true, according to the hypothesis development and the theoretical basis, it is reasonable to not only include top five well-paid, but all the executives listed in annual report to measure the degree of delegation. This modification can be risky and introduce noises to δ since the reasons for firms that include less important executives in their annual report might be heterogeneous.

Nevertheless, I implement an alternative measure for the degree of delegation, δ0 , for which I use titles of all the reported subordinates.

The results are summarized in Panel A, B and C of Table 11. For brevity, the empirical analyses are focused on the sample of older CEOs, for which the hump shaped relationship is statistically signiﬁcant. The Panel A of Table 11 summarizes the empirical results of the multivariate regression. The Panel B of

Table 11 summarizes the empirical results of regression controlling for the reverse causality. The results of the regression model that mitigate both selection bias and reverse causality are reported in Panel C of Table 11. Overall the statistical significance of the quadratic functional form is not affected by the modified measure of delegation δ0 . The results of aggregate quality metrics are in general statisti- CHAPTER 3: DELEGATION, INFORMATION AND MARKET 222 cally more significant than those of average quality metrics. The results become stronger after controlling for inverse mills ratio and endogeneity control. Notice that forecast error, F AbsError, is the only metric of the four that is persistently significant across all the model specifications for the modified delegation measure,

δ0 . F AbsError captures the absolute diﬀerence between the MFs and the earnings of the actual announcement and is considered the most direct measure for the capability of internal information production. The informationally eﬃcient regimes for stock liquidity are also determined based on the parameterization of the regression models for F AbsError.

5.4.3 Regression in δ Quantiles

According to the hypothesis development and the theoretical basis of information production (Appendix A), there are two informationally eﬃcient regimes: centralization and delegation, each of which represents an extrema of δ where either the

CEO or subordinates are in charge. The theory and empirical evidence suggests that given the fundamental diﬀerence between the CEO and her subordinates, the internal optimality of information production by δ doesn’t exist, thereby de- lineating a hump-shaped relationship between δ and inverse quality of voluntary disclosure. However, albeit the econometric model of the study speciﬁes the relationship in a quadratic form, little is known about the true functional form for the hump-shaped relationship. This is one of the reasons that the study starts with graphical analysis. For the last robustness check, I implement a relatively nonparametric test to further quantify the functional relationship between disclosure CHAPTER 3: DELEGATION, INFORMATION AND MARKET 223 quality and δ. I sort δ into percentiles and run regressions in certain quantiles of

δ as illustrated at the top of each column in Table 12. If there is indeed a nonlinear relationship between the quality of voluntary disclosure and the degree of delegation, the coeﬃcients of δ shall change accordingly in regressions performed over diﬀerent ranges of δ. For brevity, the robustness check is focused on the two main quality metrics, F AbsBias and F AbsError, which has been shown most robust for the hump-shaped relationship by the graphical analysis and regression tests.

The results are summarized in Panel A and B of Table 12. Panel A of Table

12 reports the results of average and aggregate F AbsBias, respectively. Panel

B of Table 12 summarizes results of average and aggregate F AbsError. For each quality metric, I also control for selection bias and reverse causality. In particular, I look at the following ranges of δ: less than or equal to Q1 ( 25% ≤ ), greater than or equal to Q3 ( 75%), less than median, greater than or equal ≥ to median, between Q1 and Q3, results of which are summarized in column (1) through (5), respectively. I expect the coefficient of δ is positive and significant in column (1) and column (3) where the functional form is generally upward slopping, the coefficient of δ is negative and significant in column (2) and (4) where the functional form is generally downward slopping, and the coefficient of δ in column

(5) is statistically insigniﬁcant due to nonlinearity.

The results of the two main quality metrics in Table 12 are consistent with the predictions despite that the statistical signiﬁcance is less than that of quadratic speciﬁcation due to smaller sample size and the nonlinearity in each subset of δ. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 224

The empirical results again are consistent with the hypothesis development and the theoretical basis of internal information production (Appendix A). I hereby conclude the hump-shaped functional form between the degree of delegation and the inverse quality of voluntary disclosure given the agency problem between the

CEO and her subordinates.

6 Conclusion and Discussion

The paper developed two competing hypothesis regarding the relationship between the degree of delegation in the top management team and the quality of voluntary ﬁnancial disclosure. Centered on the pair of competing hypotheses, the paper investigates the eﬀect of responsibility sharing in the top management team

(δ) on various aspects of corporate voluntary disclosure of management earnings forecasts. The results are in alignment with the hypothesis that the quality of voluntary disclosure is improved when CEO is in total control or fully delegates responsibilities to her subordinates. The empirical results indicate that the partial involvement of the CEO in specific operations of the firm is not helpful to the information production, knowledge sharing and organizational learning. In particular, when the top management team is headed by aging CEO, I find a stronger hump-shaped relationship between responsibility sharing in the top management team and the inverse of disclosure quality. Moreover, the statistical significance of the hump-shaped relationship is robust to the sensitivity check in such a way that it becomes more significant in the sample of even older CEOs and becomes CHAPTER 3: DELEGATION, INFORMATION AND MARKET 225 less significant when the sample includes younger CEOs. Age is widely used as a proxy for risk taking propensity, executive horizon and managerial preference.

The pattern that the significance of the relationship evolves in age further validates the argument that due to heterogeneity in preference and social hierarchy, it is even harder for older CEOs to coherently collaborate with her subordinate managers for generating insightful forward-looking information. From another perspective, older CEOs tend to run firms of lower risk and expend more effort to guide the market expectations. Voluntary disclosure of high quality becomes an important strategic management vehicle to reduce information risks. Provided that the corporate policy is largely determined by the wishes and tastes of the

CEO, and managerial attention is limited, the eﬃcacy of the top team structure is manifested in the quality of voluntary disclosure only when voluntary disclosure becomes a key part of the corporate policy.

The paper is focused on the most important and widely recognized voluntary

ﬁnancial disclosure by practitioners and scholars – management earnings forecasts

(MFs). Speciﬁcally, the paper constructs four metrics, each of which measures one aspect of voluntary disclosure quality of MFs: bias, error, accuracy and optimism.

In order to account for the asymmetric and qualitative nature of those measures,

I use nonlinear estimators of Generalized Linear Model (GLM) to improve the ef-

ficiency of model estimation. Thus, the parameter estimates should be on average more closely clustered around the true central tendency. Moreover, estimating one equation alone in the regression analysis may result in spurious statistical inference because of selection bias since it is completely the discretionary choice CHAPTER 3: DELEGATION, INFORMATION AND MARKET 226 of the management to publish any MFs or not. Therefore, I conduct a Heckman correction by estimating a Probit model in the first stage prior to the main equation. The reported Pearson statistic indicates that the Probit model is of strong goodness of fit. In addition to selection bias, I implement Control Function (CF) method and use instrumental variable of CEO power index (P owerIndex) that measures the overall degree of CEO power. P owerIndex is a strong instrument as indicated by the statistics of the first stage regression. The main identification strategy of the paper is to estimate a system of structural equations, which ameliorate the concerns of selection bias and reverse causality. The regression results estimated in a system of equations that combines Heckman correction and

Control Function are even stronger, indicating that the simple regression analysis only captures the lower bond of the relationship. The empirical evidence of the paper suggests that the responsibility sharing in the top management team would causally aﬀect the quality of voluntary disclosure through the channel of internal information production.

Another interesting finding is that across various metrics of voluntary disclosure quality, model specifications and estimation methods, the inflection points suggested by the parameterizations are persistently around 0.25, which coincides with the sample median. Given the symmetric bell-shape distribution of δ, this indicates that a significant proportion of large firms has not yet optimized the delegation structure of the management team for the purpose of insightful information provision to investors of the capital market.

The paper further extends the linkage between degree of delegation and qual- CHAPTER 3: DELEGATION, INFORMATION AND MARKET 227 ity of voluntary disclosure by examining the market impacts of the information efficiency. The paper finds that when the delegation structure lies in certain range that is deemed to be effective for communication between managers and investors’ community, stock liquidity, as measured by Turnover and Amihud Price Impact, improves accordingly. The finding is also consistent with the variation of the inverted hump-shaped relationship by CEO age cohort, thereby further reinforcing the theory of informational regimes developed in the paper. The empirical findings of the study complete the linkage between delegation structure and voluntary information provision quality, and further extends the linkage to the corresponding impacts upon stock liquidity.

At last, the paper checks the value relevance of the improvement of liquidity driven by delegation structures to examine whether or not the market actively incorporate the variation of liquidity into stock prices. The empirical results indicate that there exists mispricing of unexpected liquidity improvement driven by clear delegation structures. Namely, it takes the market a relatively long period of time to adjust the price accordingly. This provides an alternative explanation to the empirical literature on internal governance. Aside from the value implications of internal governance, the contemporaneous valuation premium of firms with “balanced” power and responsibility distribution could also be attributable to the mispricing of the unexpected variation of liquidity. Hence, the theory of information production and transmission developed in this paper could be complementary to the theory of internal governance. The two may play important roles side by side in determining different aspects of managerial efficacy and could CHAPTER 3: DELEGATION, INFORMATION AND MARKET 228 both be partially attributable to the nonlinear relationship between firm value and responsibility sharing theorized in Acharya et al.(2011). Further studies are to be done to disentangle the nexus of firm value, information and management structure. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 229

References

Abrevaya, J. (1999). Leapfrog estimation of a ﬁxed-eﬀects model with unknown

transformation of the dependent variable. Journal of Econometrics, 93 (2),

203–228.

Acharya, V. V., Myers, S. C., & Rajan, R. G. (2011). The internal governance of

ﬁrms. The Journal of Finance, 66 (3), 689–720.

Adams, R. B., Almeida, H., & Ferreira, D. (2005). Powerful CEOs and their

impact on corporate performance. The Review of Financial Studies, 18 (4),

1403–1432.

Aggarwal, R. K., Fu, H., & Pan, Y. (2017). An empirical investigation of internal

governance. In AFA 2011 Denver Meetings Paper.

Aguilar, F. J. (1967). Scanning the business environment. New York: Macmillan.

Ajinkya, B., Bhojraj, S., & Sengupta, P. (2005). The association between outside

directors, institutional investors and the properties of management earnings

forecasts. Journal of Accounting Research, 43 (3), 343–376.

Ajinkya, B., & Gift, M. J. (1984). Corporate managers’ earnings forecasts and

symmetrical adjustments of market expectations. Journal of Accounting

Research, 425–444.

Amihud, Y. (2002). Illiquidity and stock returns: Cross-section and time-series

eﬀects. Journal of Financial Markets, 5 (1), 31–56.

Amihud, Y., & Mendelson, H. (1986). Asset pricing and the bid-ask spread. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 230

Journal of Financial Economics, 17 (2), 223–249.

Amihud, Y., & Mendelson, H. (1989). The eﬀects of beta, bid-ask spread, residual

risk, and size on stock returns. The Journal of Finance, 44 (2), 479–486.

Arendt, L. A., Priem, R. L., & Ndofor, H. A. (2005). A CEO-adviser model of

strategic decision making. Journal of Management, 31 (5), 680–699.

Atkins, A. B., & Dyl, E. A. (1997). Market structure and reported trading volume:

Nasdaq versus the nyse. Journal of Financial Research, 20 (3), 291–304.

Baginski, S. P., Hassell, J. M., & Kimbrough, M. D. (2002). The eﬀect of le-

gal environment on voluntary disclosure: Evidence from management earn-

ings forecasts issued in US and Canadian markets. The Accounting Review,

77 (1), 25–50.

Balakrishnan, K., Billings, M. B., Kelly, B., & Ljungqvist, A. (2014). Shaping liq-

uidity: On the causal eﬀects of voluntary disclosure. the Journal of Finance,

69 (5), 2237–2278.

Bamber, L. S., Jiang, J., & Wang, I. Y. (2010). What’s my style? The inﬂuence of

top managers on voluntary corporate ﬁnancial disclosure. The Accounting

Review, 85 (4), 1131–1162.

Bardos, K. S. (2011). Quality of ﬁnancial information and liquidity. Review of

Financial Economics, 20 (2), 49–62.

Barkema, H. G., & Schijven, M. (2008). Toward unlocking the full potential of

acquisitions: The role of organizational restructuring. Academy of Manage-

ment Journal, 51 (4), 696–722. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 231

Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinc-

tion in social psychological research: Conceptual, strategic, and statistical

considerations. Journal of Personality and Social Psychology, 51 (6), 1173.

Bettman, J. R., Johnson, E. J., & Payne, J. W. (1986). Cognitive Eﬀort and De-

cision Making Strategies: A Componential Analysis of Choice. (Tech. Rep.).

DUKE UNIV DURHAM NC.

Biddle, G. C., & Hilary, G. (2006). Accounting quality and ﬁrm-level capital

investment. The Accounting Review, 81 (5), 963–982.

Botosan, C. A. (1997). Disclosure level and the cost of equity capital. Accounting

Review, 323–349.

Bound, J., Jaeger, D. A., & Baker, R. (1993). The cure can be worse than the

disease: A cautionary tale regarding instrumental variables (Tech. Rep.).

National Bureau of Economic Research.

Bound, J., Jaeger, D. A., & Baker, R. M. (1995). Problems with instrumental

variables estimation when the correlation between the instruments and the

endogenous explanatory variable is weak. Journal of the American Statistical

Association, 90 (430), 443–450.

Brennan, Chordia, T., & Subrahmanyam, A. (1998). Alternative factor spec-

iﬁcations, security characteristics, and the cross-section of expected stock

returns. Journal of Financial Economics, 49 (3), 345–373.

Brennan, Huh, S.-W., & Subrahmanyam, A. (2013). An analysis of the amihud

illiquidity premium. The Review of Asset Pricing Studies, 3 (1), 133–176. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 232

Brennan, & Subrahmanyam, A. (1996). Market microstructure and asset pricing:

On the compensation for illiquidity in stock returns. Journal of Financial

Economics, 41 (3), 441–464.

Brick, I., Palia, D., & Qiao, Y. (2019). The impact of internal governance on a

ceo’s investment cycle. In AFA 2020 San Diego Meetings Paper.

Brickley, J. A., Linck, J. S., & Coles, J. L. (1999). What happens to CEOs after

they retire? New evidence on career concerns, horizon problems, and CEO

incentives. Journal of Financial Economics, 52 (3), 341–377.

Bunderson, J. S., & Reagans, R. E. (2011). Power, status, and learning in orga-

nizations. Organization Science, 22 (5), 1182–1194.

Bushman, R. M., & Smith, A. J. (2001). Financial accounting information and

corporate governance. Journal of Accounting and Economics, 32 (1-3), 237–

333.

Buyl, T., Boone, C., Hendriks, W., & Matthyssens, P. (2011). Top management

team functional diversity and ﬁrm performance: The moderating role of

CEO characteristics. Journal of Management Studies, 48 (1), 151–177.

Cameron, A. C., & Trivedi, P. K. (2013). Regression analysis of count data

(Vol. 53). Cambridge university press.

Chaganti, R., & Sambharya, R. (1987). Strategic orientation and characteristics

of upper management. Strategic Management Journal, 8 (4), 393–401.

Chakravarthy, B. S. (1986). Measuring strategic performance. Strategic Manage-

ment Journal, 7 (5), 437–458. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 233

Chao, J. C., & Swanson, N. R. (2005). Consistent estimation with a large number

of weak instruments. Econometrica, 73 (5), 1673–1692.

Chebyshev, P. L. (1867). Des valeurs moyennes. J. Math. Pures Appl, 12 (2),

177–184.

Cheng, Q., Lee, J., & Shevlin, T. (2015). Internal governance and real earnings

management. The Accounting Review, 91 (4), 1051–1085.

Chordia, T., Subrahmanyam, A., & Anshuman, V. R. (2001). Trading activity

and expected stock returns. Journal of Financial Economics, 59 (1), 3–32.

Chung, K. H., Van Ness, B. F., & Van Ness, R. A. (1999). Limit orders and the

bid–ask spread. Journal of Financial Economics, 53 (2), 255–287.

Crawford, V. P., & Sobel, J. (1982). Strategic information transmission. Econo-

metrica: Journal of the Econometric Society, 1431–1451.

Dechow, Ge, W., & Schrand, C. (2010). Understanding earnings quality: A

review of the proxies, their determinants and their consequences. Journal of

Accounting and Economics, 50 (2-3), 344–401.

Dechow, & Skinner, D. J. (2000). Earnings management: Reconciling the views of

accounting academics, practitioners, and regulators. Accounting Horizons,

14 (2), 235–250.

Dechow, & Sloan, R. G. (1991). Executive incentives and the horizon problem:

An empirical investigation. Journal of accounting and Economics, 14 (1),

51–89.

Dewatripont, M., & Tirole, J. (2005). Modes of communication. Journal of CHAPTER 3: DELEGATION, INFORMATION AND MARKET 234

Political Economy, 113 (6), 1217–1238.

Diamond, D. W. (1985). Optimal release of information by ﬁrms. The Journal of

Finance, 40 (4), 1071–1094.

Diamond, D. W., & Verrecchia, R. E. (1991). Disclosure, liquidity, and the cost

of capital. The Journal of Finance, 46 (4), 1325–1359.

Dick-Nielsen, J., Feldh¨utter,P., & Lando, D. (2012). Corporate bond liquid-

ity before and after the onset of the subprime crisis. Journal of Financial

Economics, 103 (3), 471–492.

Dyck, A., Morse, A., & Zingales, L. (2010). Who blows the whistle on corporate

fraud? The Journal of Finance, 65 (6), 2213–2253.

Fama, E. F., & French, K. R. (1993). Common risk factors in the returns on

stocks and bonds. Journal of Financial Economics, 33 , 3–56.

Feng, M., Ge, W., Luo, S., & Shevlin, T. (2011). Why do CFOs become involved in

material accounting manipulations? Journal of Accounting and Economics,

51 (1-2), 21–36.

Fields, T. D., Lys, T. Z., & Vincent, L. (2001). Empirical research on accounting

choice. Journal of Accounting and Economics, 31 (1-3), 255–307.

Finkelstein, S. (1992). Power in top management teams: Dimensions, measure-

ment, and validation. Academy of Management Journal, 35 (3), 505–538.

Francis, & Armstrong, A. (2003). Ethics as a risk management strategy: The

Australian experience. Journal of Business Ethics, 45 (4), 375–385.

Francis, Philbrick, D., & Schipper, K. (1994). Shareholder litigation and corporate CHAPTER 3: DELEGATION, INFORMATION AND MARKET 235

disclosures. Journal of Accounting Research, 32 (2), 137–164.

Gautier, A., & Paolini, D. (2007). Delegation and information revelation. Journal

of Institutional and Theoretical Economics, 163 (4), 574–597.

Ge, W., Matsumoto, D., & Zhang, J. L. (2011). Do CFOs have style? An empir-

ical investigation of the eﬀect of individual CFOs on accounting practices.

Contemporary Accounting Research, 28 (4), 1141–1179.

Gibbons, R., & Murphy, K. J. (1992). Optimal incentive contracts in the presence

of career concerns: Theory and evidence. Journal of Political Economy,

100 (3), 468–505.

Goodman, T. H., Neamtiu, M., Shroﬀ, N., & White, H. D. (2013). Management

forecast quality and capital investment decisions. The Accounting Review,

89 (1), 331–365.

Graham, J. R., Harvey, C. R., & Rajgopal, S. (2005). The economic implications of

corporate ﬁnancial reporting. Journal of accounting and economics, 40 (1-3),

3–73.

Gupta, A. K., & Govindarajan, V. (1984). Business unit strategy, managerial

characteristics, and business unit eﬀectiveness at strategy implementation.

Academy of Management Journal, 27 (1), 25–41.

Haleblian, J., & Finkelstein, S. (1993). Top management team size, CEO dom-

inance, and ﬁrm performance: The moderating roles of environmental tur-

bulence and discretion. Academy of management Journal, 36 (4), 844–863.

Hambrick, D. C. (1994). Top management groups: A conceptual integration and CHAPTER 3: DELEGATION, INFORMATION AND MARKET 236

reconsideration of the” team” label. Research in Organizational Behavior,

16 , 171–171.

Hambrick, D. C. (1995). Fragmentation and the other problems ceos have with

their top management teams. California Management Review, 37 (3), 110–

127.

Hambrick, D. C., & Mason, P. A. (1984). Upper echelons: The organization as

a reﬂection of its top managers. Academy of Management Review, 9 (2),

193–206.

Hardaker, J. B., Richardson, J. W., Lien, G., & Schumann, K. D. (2004, 6).

Stochastic eﬃciency analysis with risk aversion bounds: a simpliﬁed ap-

proach. The Australian Journal of Agricultural and Resource Economics,

48 (2), 253–270.

Harris, M., & Raviv, A. (2005). Allocation of decision-making authority. Review

of Finance, 9 (3), 353–383.

Harsanyi, J. C. (1968). Games with incomplete information played by “bayesian”

players part ii. bayesian equilibrium points. Management Science, 14 (5),

320–334.

Heaton, J. B. (2002). Managerial optimism and corporate ﬁnance. Financial

Management, 33–45.

Hirshleifer, D., & Thakor, A. V. (1992). Managerial conservatism, project choice,

and debt. The Review of Financial Studies, 5 (3), 437–470.

Hirst, D. E., Koonce, L., & Venkataraman, S. (2008). Management earnings CHAPTER 3: DELEGATION, INFORMATION AND MARKET 237

forecasts: A review and framework. Accounting Horizons, 22 (3), 315–338.

Holmstr¨om,B. (1999). Managerial incentive problems: A dynamic perspective.

The Review of Economic Studies, 66 (1), 169–182.

Hui, K. W., & Matsunaga, S. R. (2014). Are CEOs and CFOs rewarded for

disclosure quality? The Accounting Review, 90 (3), 1013–1047.

Huizinga, H. P., & Voget, J. (2009). International taxation and the direction and

volume of cross-border m&as. The Journal of Finance, 64 (3), 1217–1249.

Jain, P., Jiang, C., & Mekhaimer, M. (2016). Executives’ horizon, internal gover-

nance and stock market liquidity. Journal of Corporate Finance, 40 , 1–23.

Jaw, Y.-L., & Lin, W.-T. (2009). Corporate elite characteristics and ﬁrm’s inter-

nationalization: CEO-level and TMT-level roles. The International Journal

of Human Resource Management, 20 (1), 220–233.

Jensen, M. C. (2004). The agency cost of overvalued equity and the current state

of corporate ﬁnance. European Financial Management, 10 (4), 549–565.

Jensen, M. C. (2005). Agency costs of overvalued equity. Financial management,

34 (1), 5–19.

Jensen, M. C., & Meckling, W. (1976). Theory of the ﬁrm: managerial behavior,

agency costs and ownership structure. Journal of Finance Economics, 3 .

Jollineau, S. J., Vance, T. W., & Webb, A. (2012). Subordinates as the ﬁrst

line of defense against biased ﬁnancial reporting. Journal of Management

Accounting Research, 24 (1), 1–24.

Kisfalvi, V., & Pitcher, P. (2003). Doing what feels right: The inﬂuence of CEO CHAPTER 3: DELEGATION, INFORMATION AND MARKET 238

character and emotions on top management team dynamics. Journal of

Management Inquiry, 12 (1), 42–66.

Koo, D. S., & Lee, D. (2017). Inﬂuential Chief Marketing Oﬃcers and Management

Revenue Forecasts. The Accounting Review, 93 (4), 253–281.

Kwak, B., Ro, B. T., & Suk, I. (2012). The composition of top management with

general counsel and voluntary information disclosure. Journal of Accounting

and Economics, 54 (1), 19–41.

Landier, A., Sauvagnat, J., Sraer, D., & Thesmar, D. (2013). Bottom-up corporate

governance. Review of Finance, 17 (1), 161–201.

Landier, A., Sraer, D., & Thesmar, D. (2009). Optimal dissent in organizations.

The Review of Economic Studies, 76 (2), 761–794.

Lang, M., & Lundholm, R. (1993). Cross-sectional determinants of analyst ratings

of corporate disclosures. Journal of Accounting Research, 31 (2), 246–271.

Lerner, J. (2006). The new new ﬁnancial thing: The origins of ﬁnancial innova-

tions. Journal of Financial Economics, 79 (2), 223–255.

Lerner, J., Sorensen, M., & Str¨omberg, P. (2011). Private equity and long-run

investment: The case of innovation. The Journal of Finance, 66 (2), 445–

477.

Lester, B., Postlewaite, A., & Wright, R. (2011). Information and liquidity.

Journal of Money, Credit and Banking, 43 , 355–377.

Lieberson, S., & O’Connor, J. F. (1972). Leadership and organizational per-

formance: A study of large corporations. American Sociological Review, CHAPTER 3: DELEGATION, INFORMATION AND MARKET 239

117–130.

Lipson, M. L., & Mortal, S. (2009). Liquidity and capital structure. Journal of

ﬁnancial markets, 12 (4), 611–644.

Mahanti, S., Nashikkar, A., Subrahmanyam, M., Chacko, G., & Mallik, G. (2008).

Latent liquidity: A new measure of liquidity, with an application to corpo-

rate bonds. Journal of Financial Economics, 88 (2), 272–298.

Matˇejka, M., Merchant, K. A., & Van der Stede, W. A. (2009). Employment

horizon and the choice of performance measures: Empirical evidence from

annual bonus plans of loss-making entities. Management Science, 55 (6),

890–905.

McInish, T. H., & Wood, R. A. (1992). An analysis of intraday patterns in bid/ask

spreads for NYSE stocks. The Journal of Finance, 47 (2), 753–764.

Minichilli, A., Corbetta, G., & MacMillan, I. C. (2010). Top management teams

in family-controlled companies:‘familiness’,‘faultlines’, and their impact on

ﬁnancial performance. Journal of management studies, 47 (2), 205–222.

Modigliani, F., & Miller, M. (1958). The cost of capital, corporation ﬁnance and

the theory of ﬁnance. American Economic Review, 48 (3), 291–297.

Nelson, K. K., & Pritchard, A. C. (2007). Litigation risk and voluntary disclosure:

The use of meaningful cautionary language. In 2nd Annual Conference on

Empirical Legal Studies Paper.

O’Hara, R., & Kotze, J. (2010). Do not log-transform count data. Nature Preced-

ings, 1–1. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 240

Pan, Y., Wang, T. Y., & Weisbach, M. S. (2016). CEO investment cycles. The

Review of Financial Studies, 29 (11), 2955–2999.

Papadakis, V. M., & Barwise, P. (2002). How much do CEOs and top managers

matter in strategic decision-making? British Journal of Management, 13 (1),

83–95.

Payne, J. W., Bettman, J. R., & Johnson, E. J. (1988). Adaptive strategy selection

in decision making. Journal of Experimental Psychology: Learning, Memory,

and Cognition, 14 (3), 534.

Prendergast, C., & Stole, L. (1996). Impetuous youngsters and jaded old-timers:

Acquiring a reputation for learning. Journal of political Economy, 104 (6),

1105–1134.

Scharfstein, D. S., & Stein, J. C. (1990). Herd behavior and investment. The

American Economic Review, 465–479.

Serﬂing, M. A. (2014). CEO age and the riskiness of corporate policies. Journal

of Corporate Finance, 25 , 251–273.

Shumway, T. (1997). The delisting bias in crsp data. The Journal of Finance,

52 (1), 327–340.

Skinner, D. J. (1994). Why ﬁrms voluntarily disclose bad news. Journal of

Accounting Research, 32 (1), 38–60.

Skinner, D. J. (1997). Earnings disclosures and stockholder lawsuits. Journal of

Accounting and Economics, 23 (3), 249–282.

Stinchcombe, A. L., McDill, M. S., & Walker, D. R. (1968). Demography of CHAPTER 3: DELEGATION, INFORMATION AND MARKET 241

organizations. American Journal of Sociology, 74 (3), 221–229.

Stock, J., Wright, J., & Yogo, M. (2002). A survey of weak instruments and

weak identiﬁcation in generalized method of moments. Journal of Business

& Economic Statistics, 20 (4), 518–529.

Stock, J., & Yogo, M. (2005). Asymptotic distributions of instrumental variables

statistics with many instruments. Identiﬁcation and inference for economet-

ric models: Essays in honor of Thomas Rothenberg, 109–120.

Stocken, P. C. (2000). Credibility of voluntary disclosure. The RAND Journal of

Economics, 359–374.

Stoll, H. R. (2000). Presidential address: Friction. The Journal of Finance, 55 (4),

1479–1514.

Thomas, A. S., Litschert, R. J., & Ramaswamy, K. (1991). The performance im-

pact of strategy-manager coalignment: An empirical examination. Strategic

Management Journal, 12 (7), 509–522.

Thomas, A. S., & Simerly, R. L. (1994). The chief executive oﬃcer and corporate

social performance: An interdisciplinary examination. Journal of Business

Ethics, 13 (12), 959–968.

Wang, I. Y. (2007). Private earnings guidance and its implications for disclosure

regulation. The Accounting Review, 82 (5), 1299–1332.

Wooldridge, J. M. (2003). Reasoning about rational agents. MIT press.

Wooldridge, J. M. (2015). Control function methods in applied econometrics.

Journal of Human Resources, 50 (2), 420–445. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 242

Yang, H. I. (2012). Capital market consequences of managers’ voluntary disclosure

styles. Journal of Accounting and Economics, 53 (1-2), 167–184.

Zwiebel, J. (1995). Corporate conservatism and relative compensation. Journal

of Political Economy, 103 (1), 1–25. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 243

Appendices

A A Theory of Corporate Internal Information Produc-

tion

To clearly and systematically articulate the hypotheses postulated based on the literature, I develop a theoretical model of internal information production and strategic information transmission with mathematical specializations and derivations. The model is based on the theoretical work of Crawford and Sobel(1982) for strategic communication between Sender and Receiver, and is an extension to the model of authority allocation between CEO and manager in Harris and

Raviv(2005). Note that the model is by no means a full-ﬂedged theoretical development but rather a heuristic one with innocuous implications for characterizing the strikingly straightforward equilibrium.

My model is focused on the information eﬃcacy of the top management team consisting of a CEO and other top executives. In particular, the potential signals obtained by the CEO and managers arep ˜ anda ˜, respectively. Both the random signals of the CEO and the managers are assumed to be uniformly distributed on [0,P ] and [0,A], respectively. Since for the “partition equilibrium” introduced later, it is the lengths, but not the locations, of the support ofp ˜ anda ˜ play a vital role, thus assuming both intervals that start at zero is without loss of generality. Note that the simpliﬁcation of uniformity is merely for the comparative static analysis, while according to Crawford and Sobel(1982), the existence of CHAPTER 3: DELEGATION, INFORMATION AND MARKET 244 equilibrium conditions does not rely on the distribution forms of random signals.

The two parameters P and A of uniformity can be interpreted as the knowledge, skills or expertise of CEO and her subordinate managers, respectively, which determines the lengths of information interval. The relative values of P and A characterize the relative importance of the knowledge or expertise possessed by

CEO and other top managers. Namely, as indicated in the CEO centric literature, the smaller is A, the larger is the CEO’s informational advantage over her subordinate managers with respect top ˜ . Clearly, knowledge and information capacity could be developed by learning and thus P and A are increasing functions of δ, i.e., P (δp) and A (δa), in which δp and δa are the respective fractions of executive responsibilities and tasks undertaken by the CEO and her subordinate managers.

I use an exogenously determined parameter b to represent how nearly the preferences and interests of CEO and her subordinate managers coincide. Given the widely documented agency problem and the non-contractual communication issue illustrated in Crawford and Sobel(1982), parameter b, as the proxy for preference similarity, is manifested in the heterogeneous utility functions between the CEO and her subordinates. Following Crawford and Sobel(1982), preference similarity parameter b is utilized to diﬀerentiate the utility function of Sender and

Receiver, namely, U S ( , b) versus U R ( ). Without loss of generality, b 0, where · · ≥ b = 0 represents the ﬁrst best condition when the preferences and interests of the

CEO and her subordinates are perfectly aligned.

Regardless of the conﬂicts of interests among agents, the true or benchmark level of information that a ﬁrm should have been collected, analyzed and absorbed CHAPTER 3: DELEGATION, INFORMATION AND MARKET 245 in the form of organizational knowledge is denoted as y. Intuitively, the true level of information is determined by the sum of signals obtained by the CEO and her subordinate managers, i.e., y = p+a. Since random signalsp ˜ anda ˜ are in support of uniform distributions on [0,P ] and [0,A], lengths of which are determined by the degree of delegation parameter, δ, the above expression is equivalent to more general form of linear combination function, such as, y = λp + µa, in which λ and

µ indicates the relative importance of signals. I deﬁne a quadratic loss function for falsiﬁed information production to characterize the optimization problem as follows.

2 min E (L (y, y∗ (p, a))) = min E (y y∗) (A.A.1) y∗ y∗ − e e e It is equivalent to maximizing the opposite quadratic objective function as follows,

2 max E (O (y, y∗ (p, a))) = max E (y y∗) y∗ y∗ − − e e e in which y is the benchmark information level and y∗ is determined by the equilibrium conditions articulated in the following sections. The above optimization program serves as the overarching framework for the later theoretical developments and mathematical derivations. Note that I assume quadratic objective function for closed form solutions. The main results of equilibrium hold for any forms of twice continuously diﬀerentiable von Neumann-Morgenstern utility function. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 246

A.1 The First-Best Case

When there is no agency problem among agents, the preferences and interests of the CEO and her subordinate managers are perfectly aligned. In the absence of conﬂicts of interest, the CEO and other top managers share the homogenous utility function. Namely, the exogenously given preference similarity parameter, b, is equal to zero. In this case, both CEO and her subordinate managers are sharing their private information and knowledge truthfully, voluntarily and proactively, and there are no frictions in the communication caused by lack of understanding, varied idiosyncratic preferences and divergent ideologies.

With all the above assumptions in place, the equilibria information production of the ﬁrst-best case is as follows,

y∗ = p + a + s (A.A.2) in which p and a are respective signals of the CEO and her subordinates from the uniformly distributed random signal generating processes supported on [0,P ] and [0,A], respectively. s represents the synergistic gain of information driven by the complementary knowledge and expertise between CEO and her subordinates whose interests are perfectly aligned. Without loss of generality, s 0. As indi- ≥ cated by the literature on the internal governance mechanisms (see, for example,

Landier et al.(2009); Acharya et al.(2011)), there should exist an hump-shaped relationship between the quality of voluntary disclosure practices and sharing executive duties between CEO and her subordinates. According to the Hypothesis CHAPTER 3: DELEGATION, INFORMATION AND MARKET 247

1.A, s should be determined by the degree of delegation δ and s (δ) has an in-

ds ∂2s ternal optimality, i.e. dδ = 0 and ∂δ2 < 0. Note that given non-negative synergy term, the equilibria level of information production exceeds the benchmark level, which signiﬁcantly depart from the literature of strategic information transmission.

Thus, the above functional form that incorporates the synergy term is deemed hy- pothetical and is only structured for the ﬁrst-best case, in which the fundamental premise of agency problem for principal-agency models in the literature of information and bargaining is not in place.

A.2 Bayesian Nash Equilibrium Cases

In this section, I deﬁne and solve for equilibrium strategies when the agency problem between CEO and her subordinate managers is in place, i.e. b > 0. Consistent with Crawford and Sobel(1982) and Harris and Raviv(2005), the equilibrium con- cept I employ for the three informational regimes: centralization, delegation and partial delegation is Bayesian Nash Equilibrium by Harsanyi(1968), which is essentially a Nash Equilibrium that conditions agents’ actions on their information and on states that they ﬁnd themselves.

A.2.1 Centralization

The ﬁrm is of centralization regime if the CEO is in total control of business operations. The disproportionately large executive duties and authorities are allocated to the CEO compared to her subordinate managers. As such, CEO maintain a centric role in every aspects of managing the ﬁrm from decision making, policy CHAPTER 3: DELEGATION, INFORMATION AND MARKET 248 implementation to monitoring operations. The other subordinate managers are more like implementers who follow the lead of the CEO and report to the CEO per her request. As an analogy to the setting of Crawford and Sobel(1982), the CEO is considered as Receiver and the subordinate managers is treated as

Sender in the centralization regime. Their utility functions are diﬀerentiated by the preference similarity parameter, b. I use r to denote the message reported by the subordinate managers (Sender). According to Crawford and Sobel(1982) and Harris and Raviv(2005), the formal statement of Bayesian Nash Equilibrium could be expressed as follows,

(1) for each a [0,A] , if r∗ is in support of q (r m), and ∈ |

q (r a) dr = 1, | ZR

where the Borel set R is the set of feasible signals, then r∗ solves

max U S ( , b); and r R ∈ ·

A (2) for each r, p, y∗ (p, r) arg min O (y (p, r) , a) g (a r) da, where ∈ y 0 | R

A g (a r) = q (r a) f (a) /P (r) = q (r a) f (a) / q (r t) f (t) dt, | | | | Z0

according to Bayesian Theorem and Total Expectation Theorem.

In particular, when the assumptions of uniformity and quadratic objective function are in place, the above equations can be expressed as follows. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 249

A A min O (y (p, r) , a) g (a r) da = min (a + p y (p, r))2g (a r) da y | y − − | Z0 Z0

A A g (a r) = q (r a) f (a) / q (r t) f (t) dt = q (r a) / q (r t) dt | | | | | Z0 Z0 in which q (r a) is the reporting rule of the subordinate managers given the ob- | tained value a ofa ˜, and g (a r) is posterior belief of the CEO for the true signal of | her subordinate managers conditional on each feasible report. The formal statement of the Bayesian Nash Equilibrium is centered on the optimal reporting rule, q (r a), which would induce the equilibria y∗ (p, r). As stated in Crawford and | Sobel(1982), “This represents [the subordinate managers’] optimal compromise between including enough information in the signal to induce [the CEO] to respond to it and holding back enough so that his response is as favorable as possible.”

According to Hardaker, Richardson, Lien, and Schumann(2004), the above formal equilibrium could be written in the reduced form as follows,

y∗ = p + E (a r) (A.A.3) |

e in which p is the signal obtained by the centralized CEO and E (˜a r) is the poste- | rior expectation ofa ˜ given the managers’ report r. Same as Crawford and Sobel

(1982), the reduced form expression could be characterized by “partition equilibria”, in which driven by divergent preferences and interests, subordinate managers CHAPTER 3: DELEGATION, INFORMATION AND MARKET 250 introduce noise not only by not diﬀerentiating as ﬁnely as possible among the information states but also by the randomization within each steps of the partition.

Namely, they only reveal which step of the partition the signal actually lies in. For- mally, let a (N) (a (N) , . . . , a (N)) denote a partition of [0,A] with N steps, ≡ 0 N where 0 = a0 (N) < a1 (N) < . . . < aN (N) = 1. To be concise and without loss of clarity, I shall suppress the dependence of N and write ai rather than ai (N).

Clearly, if a [ai 1, ai], the subordinate managers choose to report a uniform ∈ − distribution on [ai 1, ai], which in turn results in the posterior belief of the CEO − given managers’ report uniformly supported on [ai 1, ai]. Consequently, E (a r) is − |

(ai 1,ai) the midpoint of the interval in which r lies, i.e., E (a r [ai 1, ai]) = − . The | ∈ − 2 e reduced form equilibrium is thus expressed as, e

ai + ai+1 y∗ (p, r) = y (a , a ) = p + , r (a , a ) . (A.A.4) i i+1 2 ∀ ∈ i i+1

I follow Crawford and Sobel(1982) and characterize the partition equilibrium as follows, for an equilibrium condition y∗ (p, r), where r is in support of q (r a) | distributed on [ai, ai+1],

S S (1) O (y (ai, ai+1) , ai, b) = O (y (ai 1, ai) , ai, b), i = 1,...,N 1, − −

(2) y∗ (p, r) = y (a , a ) for all n (a , a ), i i+1 ∈ i i+1

(3) a0 = 0 , and

(4) aN = 1 in which OS ( , b) is the objective function of Sender (subordinate managers), · CHAPTER 3: DELEGATION, INFORMATION AND MARKET 251 which is divergent from the ﬁrst best objective function of the ﬁrm by the preference parameter b; y∗ (p, r) is the reduced-form equilibrium, (1) is an arbitrage condition for boundary-value, (3) and (4) are the initial and terminal conditions, respectively. Applying the quadratic objective function and uniformity assumption, the arbitrage condition (1) is specialized to

ai+ai+1 2 ai 1+ai 2 a b = − a b − 2 − i − − 2 − i − 1 1 ai+ai+1 2 2 ai 1+ai 2 2 a b = − a b 2 − i − 2 − i − ai+ai+1 a b = ai+ai+1 a b 2 − i − 2 − i −

Provided that ai in the partition of [0 ,A] is monotonically increasing, the above equation only holds if

ai+ai+1 ai+ai 1 a b = − a b 2 − i − − 2 − i − ai+1 = 2ai ai 1 + 4b − − For i = 1,...,N 1 , the above second-order linear diﬀerence equation could be − solved via successive calculation whereby iterating backward for each value of i, given a0 = 0 :

ai = 2ai 1 ai + 4b ai ai 1 = ai 1 ai + 4b − − ⇒ − − − −

Clearly,

ai ai 1 = ai 1 ai 2 + 4b = ai 2 ai 3 + 4b + 4b − − − − − − − − (1) = ai (i 1) a0 + 4b + ... + 4b = a1 + 4 (i 1) b − − − − i 1 items − | {z } CHAPTER 3: DELEGATION, INFORMATION AND MARKET 252

ai 1 ai 2 = ai 2 ai 3 + 4b = ai 3 ai 4 + 4b + 4b − − − − − − − − − (2) = ai (i 1) a0 + 4b + ... + 4b = a1 + 4 (i 2) b − − − − i 2 items − | {z } ai 2 ai 3 = ai 3 ai 4 + 4b = ai 3 ai 4 + 4b + 4b − − − − − − − − − (3) = ai (i 1) a0 + 4b + ... + 4b = a1 + 4 (i 3) b − − − − i 3 items − | {z }

···

ai (i 1) a0 = a1 + 4 (i i) b = a1 (i) − − − −

Take a sum of above equations from (1) to (i), the solution of the second-order diﬀerence equation parametrized by a1 is:

a + a = a + ... + a + 4 (i 1) b + 4 (i 2) b + ... 4 (i i) b i 0 1 1 − − − iitems arithematic progression i i(4(i 1)b+0) | {z } − ai = ia1 + 4 (|i k) b = ia1 + {z 2 } k − P = ia + 2i (i 1) b 1 −

Given that aN = A, where A denotes the relative importance of the knowledge and expertise of subordinate managers, I can compute

a = a N + 2N (N 1) = A a = (A 2N (N 1) b) /N N 1 − ⇒ 1 − −

Substituting for the value of ai, it yields CHAPTER 3: DELEGATION, INFORMATION AND MARKET 253

iA a = + 2bi (i N)(i = 0,...,N) i N − and

A ai ai 1 = + 2b (2i N 1) − − N − − in which N is the largest positive integer that i could take, such that

a = ia + 2i (i 1) b A 2i (i 1) b < A i 1 − ≤ ⇒ −

Therefore,

1 1 1 2A 2 N(A, b) = + 1 + 1 (A.A.5) 2 2 b − * + in which operator takes the smallest integer greater than or equal to the h·i operand. According to Crawford and Sobel(1982), the above expression, N (A, b)

, is considered as the Pareto best equilibria, representing the ﬁnest partition of

[0,A]. Clearly, according to the theoretical background, N is a function of A and b, in which A is positively determined by the fraction of executive tasks assigned to the managers, δa.

With the Bayesian Nash Equilibrium fully characterized above, it follows the derivation of the ex-ante expected information loss of the centralization regime:

According to the Total Expectation Theorem, CHAPTER 3: DELEGATION, INFORMATION AND MARKET 254

ai E (E (L(y∗(p, r), a) r)) = E L (y∗ (p, r) , a)p (a r) da (i = 1,...,N) | ai 1 | Z −

According to Bayesian Theorem and Total Probability Theorem, when the signal uniformly distributed in the step (ai 1, ai), −

q (r a) f (a) q (r a) f (a) f (a) p (a r) = | = | = | P (r) ai+1 q (r t) f (t) dt ai+1 f (t) dt ai | ai R R Thus,

ai E L (y∗ (p, r) , a)p (a r) da ai 1 | Z − ai L (y∗ (p, r) , a)f (a) da ai 1 = E − ai f(t)dt R ai 1 ! − N ai ai f(t)dt L (y∗ (p, r) , a)f (a) da ai 1 R ai 1 = − − A × ai f(t)dt i R f(a)da R ai 1 ! 0 − XN 1 R ai R = L (y∗ (ai, ai+1) , a) da A ai 1 i Z − XN 1 ai (a + a ) 2 = i i+1 a da A 2 − i ai 1 X Z − N 3 3 1 1 (ai + ai 1) 1 (ai + ai 1) = − ai − ai 1 A 3 2 − − 3 2 − − i ! X 1 N 1 1 1 3 1 1 1 3 = ai 1 ai ai ai 1 A 3 2 − − 2 − 3 2 − 2 − i ! X 1 1 3 = (ai ai 1) 12 A − − CHAPTER 3: DELEGATION, INFORMATION AND MARKET 255

Substituting the value of the ﬁrst diﬀerence of ai, the closed form solution of the ex-ante expected loss of information is,

1 1 A 3 E(L(y∗(p, r), a)) = + 2b(2i N 1) 12 A N − − 1 A 2 b2(N 2 1) = + − 12 N 3

Whereby explicitly parameterizing A and N by latent variables, the above equation becomes

2 2 2 1 A (δa) b (N(b, A (δa)) 1) L∗ (b, δ ) = + − . (A.A.5) C a 12 N (b, A (δ )) 3 a

Namely, the ex-ante equilibrium loss of information, Lc∗ (b, δa), is a function of two main variables, the relative importance of subordinate managers’ expertise or knowledge, A (δa), and the Pareto best partition number, N (b, A (δa)), which are determined by the latent preference parameter b and the exogenously given parameter, δa, the fraction of executive tasks fulﬁlled by the subordinate managers.

Comparative statics analysis will be performed on the closed form equilibrium solution in the later section.

A.2.2 Delegation

The equilibrium condition of the delegation regime is purely symmetric to that of the centralization derived in the above section. The only diﬀerence is the organizational background. In the case of delegation regime, disproportionally large CHAPTER 3: DELEGATION, INFORMATION AND MARKET 256 amount of executive responsibilities and tasks are undertaken by the subordinate managers. In other words, subordinate managers are fully in charge of the business operations throughout while the CEO is taking on an advisory role. In an extreme case, the CEO is largely in an emeritus status and relinquish executive authorities to other top executives, becoming merely a symbolic leader.

In the case of the delegation regime, opposite to the centralization regime, CEO is sharing her private knowledge and information to the subordinate managers and is thus considered as Sender. With all other settings that follow through symmetrically, the ex-ante expected information loss of the delegation regime is as follows,

2 2 2 1 P (δp) b N(b, P (δp)) 1 L∗ (b, δ ) = + − (A.A.6) D p 12 N (b, P (δ )) 3 p in which b is the latent preference parameter and δp is the fraction of executive duties taken by the CEO. Similarly to the centralization regime, the expected information loss for the delegation regime, Ld∗ (b, δp), is a function of the relative importance of the CEO’s expertise or private knowledge, P (δp), and the Pareto- best partition number, N (b, P (δp)), which are determined by the exogenous parameters b and P (δp).

A.2.3 Partial Delegation

Partial delegation is the middle ground between the above two information regimes, in which the CEO partially involve in the business operations and the delegation structure between CEO and her subordinates are not clear. In case of partial CHAPTER 3: DELEGATION, INFORMATION AND MARKET 257 delegation, neither the CEO nor her subordinates are in charge. In contrast, both of them take part in the various ﬁrm activities. The executive responsibilities are shared jointly and carried out partially by the each party. The notion of partial delegation departs from the traditional literature of strategic information transmission in that both parties, the CEO and her subordinates, are information

Senders. Likewise, they are also Receivers by whom the information level of the

ﬁrm is revealed.

To ﬁt the case of partial delegation to the framework of strategic information literature (see, for example, Crawford and Sobel 1982; Harris and Raviv 2005), I assume they disseminate information simultaneously instead of modeling a repeated game for the sake of tractability and closed form equilibria. I further assume that the random signalsp ˜ anda ˜, uniformly supported on the private information of the CEO and her subordinates, i.e., [0,P ] and [0,A], and are independent of each other. The ex-ante expected loss of information production in the case of partial delegation is modeled as follows,

E (L (y∗ (p, a) , p, a))

N(A,P ) 2 ai 1 + ai pi 1 + pi = − a + − p f (a, p) d(a, p) 2 − 2 − i X APZ

N N ai 1+ai 2 pi 1+pi 2 A P ai pi − − 1 1 2 a + 2 p =  − −  dadp A P ai 1 pi 1 ai 1+ai pi 1+pi i j Z − Z − +2 − a − p X X  2 2   − −  N N A P ai pi  2  1 1 ai 1 + ai = − a dadp A P 2 − i j ai 1 pi 1 ! X X Z − Z − CHAPTER 3: DELEGATION, INFORMATION AND MARKET 258

NA NP ai pi 2 1 1 pi 1 + pi + − p dadp A P 2 − i j ai 1 pi 1 ! X X Z − Z − NA NP ai pi 1 1 ai 1 + ai pi 1 + pi + 2 − a − p dadp A P 2 − 2 − i j ai 1 pi 1 X X Z − Z − NA NP ai 2 pj 1 1 ai 1 + ai = − a da 1 dp A P 2 − i j ai 1 pj 1 X X Z − Z − NA NP pj 2 pj 1 1 pi 1 + pi + − p da 1 dp A P 2 − i j pj 1 pj 1 X X Z − Z − Np NA ai pj 1 1 ai 1 + ai pi 1 + pi + 2 − a da − p dp A P 2 − 2 − i j ai 1 pj 1 X X Z − Z − NA ai 2 NP 1 ai 1 + ai 1 = − a da (pj pj 1) A 2 − P − − i ai 1 j X Z − X NP pi 2 Na 1 pi 1 + pi 1 + − p dp (ai ai 1) P 2 − A − − j pi 1 i X Z − X Np NA ai pj 1 ai 1 + ai 1 pi 1 + pi + 2 − a da − p dp A 2 − P 2 − i ai 1 j pj 1 X Z − X Z − NA ai 2 NP pi 2 1 ai 1 + ai 1 pi 1 + pi = − a da + − p dp A 2 − P 2 − i ai 1 j pi 1 X Z − X Z − 1 1 1 2 1 1 2 + ai 1 ai ai ai 1 A2 2 − − 2 − 2 − 2 − ! 1 1 1 2 1 1 2 pi 1 pi pi pi 1 · P 2 − − 2 − 2 − 2 − !

1 1 3 1 1 3 = (ai ai 1) + (pi pi 1) + 0 12 A − − 12 P − −

in which p and a are signals obtained by the CEO and the subordinate managers in support of uniformly distributed variables on [0,P ] and [0,P ], which denote the relative importance of the knowledge or expertise of CEO and her subordinates. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 259

Substituting for the value of the ﬁrst-diﬀerence of ai and pi yields

2 2 2 2 1 A (δa) b N(b, A (δa)) 1 1 P (δp) L∗ (b, δ ) = + − + PD a 12 N (b, A (δ )) 3 12 N (b, P (δ )) a p (A.A.7)

2 2 b N(b, P (δp)) 1 + − 3

in which b is the predetermined latent parameter for preference similarity, and δa and δp are respective fraction of tasks preassigned to CEO and her subordinate managers in the case of partial delegation. Both b and δ are considered exogenous parameters to the modeling framework. The above mathematical derivation illustrates a strikingly simple equilibrium condition that given independent signal and simultaneous dissemination, the ex-ante expected information loss of partial delegation, LPD∗ (b, δa), could perfectly separate the contribution of the CEO and the subordinates to the information transmission noise. Namely, the functional form of LPD∗ is the direct sum of that of LC∗ and that of LP∗ . Note that albeit seemingly apparent relations by functional structures, the delegation parameters,

δa and δp, are diﬀerent in the three information regimes. To compare the equilibria in diﬀerent regimes, I conduct comparative statics analysis in the next section.

A.3 Comparative Statics Analysis

The ex-ante expected losses of information production in the regimes of centralization, delegation and partial delegation are listed as follows (via equations (A.A.5), CHAPTER 3: DELEGATION, INFORMATION AND MARKET 260

(A.A.6) and (A.A.7)),

2 2 C 2 C b N b, A δa 1 C 1 A δa L∗ b, δ = + − C a C 12 N (b, A (δa ))! 3

2 2 D 2 D b N b, P δp 1 D 1 P δp L∗ b, δ = + − D p D 12 N b, P δp ! 3

2 2 PD PD PD 1 A δa P δp L∗ b, δ = + PD a 12 N (b, A (δPD)) PD  a ! N b, P δp !  2   b 2 2 + N b, A δPD 1 + N b, P δPD 1 3 a − p − PD PD = LC∗ b, δa + LD∗ b, δp .

C D PD in which δa , δa and δa are preassigned fractions of executive tasks of subordinate managers for the three information regimes: centralization, delegation and partial

C D P delegation, respectively; δa , δa and δa D are preassigned respective fractions of

D PD C executive tasks of the CEO for the three regimes. By deﬁnition, δa > δa > δa

D PD C and δp > δp > δp . To compare the equilibria information loss across three

∂Lx∗ regimes, I need to ﬁrst compute the comparative static, x , where x = C,D ∂δy { } x and y = a, p . To ensure L∗ is diﬀerentiable at every point of δ , I replace { } x y N(b, δx) with N(b, δx) 6, where N(b, δx) = N(b, δx) 1 . The comparative stat- y y y y − D E ics analysis is asb follows: b

6 This is a arguably good approximation to make given Lx∗ (b, A ( )) is continuous and monotonically increasing in A (see, for example, Crawford and Sobel 1982· ) CHAPTER 3: DELEGATION, INFORMATION AND MARKET 261

2 2 x x 2 x ∂L∗ b, δ ∂ 1 A δ b N b, A δy 1 x y = y + − x x x ∂δ y ∂δy 12 N b, A δy ! 3 

 2 2 x 2 x ∂ 1 A δ ∂ b N b, A δy 1 = y + − x x x ∂δy 12 N b, A δy !  ∂δy  3  1 ∂ A δx  ∂  A δx ∂A δx = 2 y y y x x x x x 12 ∂δy N b, A δy ! · ∂A δy N b, A δy ! · ∂δy ! x 2 2 x ∂N b, A δy + b N b, A δy x 3 ∂δy where,

x ∂ A δy x x ∂A δy N b, A δy ! 1+ 1+2A δx b x x ( ( y ) ) ∂A(δy ) 1 x 1 2b ∂A(δy ) x ∂A δ x q 2 ∂δy 2 y 2 x ∂δy · − (1+2A(δy )b) · = 2 q x 1+ (1+2A(δy )b) q 2

1+ 1+2A δx b x x ( ( y ) ) ∂A(δy ) 1 1 ∂A(δy ) x x q 2 ∂δy  2  ∂δy · − 1 · 2+ x A(δy )b s =  2  x  1+ (1+2A(δy )b) q 2

1+ 1+2A δx b x ( ( y ) ) 1 ∂A(δy ) x  q 2   ∂δy − 1 × 2 2+ x A(δy )b s =   2    x  1+ (1+2A(δy )b) q 2

x x 1+ (1+2A(δy )b) 1 1 1 1+ (1+2A(δy )b) Clearly, q 2 > 2 , and   < 2 , thus q 2 1 − 2 2+ x A(δy )b  s    CHAPTER 3: DELEGATION, INFORMATION AND MARKET 262

1 x > 0. According to the theoretical background, b > 0 , A δy ,  1  2 2+ x A(δy )b s   x x x  x  ∂A(δy ) ∂N(b,A(δy )) ∂ A(δy ) N b, A δy > 0, ∂δx > 0, and ∂δx > 0. Therefore, x x > y y ∂A(δy ) N(b,A(δy )) x ∂Lx∗ (b,δy ) 0 and x > 0. To this end, I have proved that the ex-ante expected informa- ∂δy

x x tion loss ∂Lx∗ b, δy monotonically decrease in the degree of delegation δy , where x = C,D and y = a, p , all else equal. { } { } Given the result of comparative statics, I have the relations as follows,

PD C LC∗ b, δa > LC∗ b, δa

PD D LD∗ b, δp > LD∗ b, δp PD C PD D because by deﬁnition, δa > δa and δp > δp . Therefore, the relations for the expected information loss between regimes are determined as follows.

PD PD PD C D LPD∗ b, δa = LC∗ b, δa + LD∗ b, δp > LC∗ b, δa + LD∗ b, δp

PD PD PD C LPD∗ b, δa = LC∗ b, δa + LD∗ b, δp > LC∗ b, δa

PD PD PD D LPD∗ b, δa = LC∗ b, δa + LD∗ b, δp > LD∗ b, δa

In sum, given the existence of heterogeneous preferences and interests between CHAPTER 3: DELEGATION, INFORMATION AND MARKET 263

CEO and her subordinate managers (b > 0), the ex-ante information loss of partial delegation is greater than the other two regimes combined, illustrating the informationally inefficient condition when the delegation structure is not clear and the partial involvement of the CEO is predominantly in place for various firm activities. In contrast, when there is no agency problem among agents, and the preferences and interests of the CEO and her subordinate managers are perfectly aligned (b = 0, first-best case), the information production level of partial delegation is at least same as that of centralization regime and delegation regime, given non-negative synergistic gain, i.e., y = p + a + s >= p + a.

Note that although this is a well-developed theoretical framework, it is ad- mittedly a heuristic one with innocuous simpliﬁcations. One is that I assume the latent parameter of preferences similarity, b, is predetermined exogenously and will not change in any variables of interest, such as degree of delegation, δ. Relaxing such assumption might musk the clearly-tractable closed form relationship among the three informational regimes. Therefore, rigorous empirical tests are needed for further veriﬁcation of the model implications. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 264

B An Econometric Reasoning

To illustrate my argument, I employ the following econometric modeling. Note that this is by no means a full-ﬂedged model but an organized econometric representation in support of my empirical strategy. I will start with most general form of multiple linear regression model,

y = Xβ + ε (A.B.1) in which

y x x x 1 11 12 ··· 1K     y x x x  2   21 22 ··· 2K  y =   ,X =   = x1 x2 xK ,  .   . . .. .  ··· N K  .   . . . .  ×              yN   xN1 xN2 xNK   N 1  ··· N K   ·   × x1i β1 ε1       x β ε  2i   2   2  xi =   i = 1, 2, . . . k, β=  , ε=  .  .   .   .   .   .   .                     xNi   βK   εN   N 1  K 1  N 1   ×   ×   ×

y y1N 1 To simplify the derivation, I standardize y and x so that that y = − × , x = i Sy k

xk xk1N 1 − × . To be concise, I still denote the standardizedy ˜ andx ˜k as y and xi. Sxk e e Again, in order to illustrate the point in a clean cut way, I assume no endogeneity and homogenous standard error that is not clustered at any level. The plain OLS CHAPTER 3: DELEGATION, INFORMATION AND MARKET 265 estimator is characterized as follows,

1 1 1 1 β = (X0X)− XY = ( X0X)− X0Y (A.B.2) N N

V ar(β Xb) = E((β E(β X))(β E(β X))0 X) | − | − | | 1 1 b = (X0Xb )− X0bE(ε0ε Xb )X0(Xb0X)− |

2 1 1 2 1 1 = σ (X0X)− = σ ( X0X)− N N

101 1 1 1 xi01 N xi 1 N N N N in which, when K = 2, N X0X =   =  . 1 x 1 1 x x 1 x 1 Px2  N i0 N i0 i   N i N i      Clearly when N is suﬃciently large (large sample analysis),  accordingP toP asymp- totic Law of Large Numbers (LLN),

1 1 x 1 0 X0X   =   N → 2  x σx   0 1          provided that xi is standardized; When K > 2, for the sake of tractability, I further assume that x is orthogonal to each other, i.e. x 0x = 0 for i = j. Note that for i i j 6 standardized xi , such restriction is equivalent to no signiﬁcant multicollinearity among independent variables. As such,

101 x201 xK 01 N N ··· N   x201 x20x2 0 1  N N ···  X0X =   N  . . ..   . . . 0       xK 01 xK 0xK   0 0   N N    CHAPTER 3: DELEGATION, INFORMATION AND MARKET 266

As N is suﬃciently large (large sample analysis),

1 x x 1 0 0 1 ··· K ···     2 2 x1 σ 0 0 σ 0 1  x1 ···   x1 ···  N X0X   =   →  . . ..   . . ..   . . . 0   . . . 0           2   2   xK 0 0 σxk   0 0 0 σxk         

Provided that y and xi are standardized, I could derive the following expression,

1 0 0 1 0 0 ··· ···     1 1 0 σ2 0 0 1 0 1  x1 ···   ···  ( X0X)− =   =   = IN (A.B.3) N  . . ..   . . ..   . . . 0   . . . 0           1     0 0 0 σ2   0 0 0 1   x1        Analogously, when N is suﬃciently large,

1 N x10y σx1,y     1 x 0y σ 1  N 2   x2,y  X0Y =     . (A.B.4) N  .  →  .   .   .           1     xK 0y   σxK ,y   N        With all the above setting being clear, we could examine the statistical signiﬁcance of βi as follows.

b CHAPTER 3: DELEGATION, INFORMATION AND MARKET 267

β [β] 1 2 i,1 − i,1 − V ar(xi) Cov(xi, y) βi σxi,y βi tβ = × − = − i h i → 1 1 1 − 2 2 bV ar(β X) V ar(xi) ( N σε ) N σε b | i,i rh i q q (A.B.5) b Provided that xi, the variable of interest (δ) does not change by fiscal year and the disclosure policy (y) is very sticky, when I inflate the sample by using MFs for each fiscal period, the components of equation (A.B.5) shall change accordingly as

1 2 1 2 2 2 follows: N N M, N σε N M σε , σxi,y M σxi,y. Therefore, asymptoti- ↑→ × → × → × cally, t , the magnitude of statistical signiﬁcance, will increase dramatically in βi

b 3 2 M. The speed is as large as M assuming the true coeﬃcient,βi, is zero. Hence, taking into the consideration of the data and the research topic of interest, I choose to use ﬁrm-year as the unit of analysis for this study. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 268

C A Diagram of Hypotheses

Hypothesis 1A:

Hypothesis 1B: CHAPTER 3: DELEGATION, INFORMATION AND MARKET 269

Table 1.A: Variable Deﬁnitions This table provide the list of variables in the empirical analysis and their deﬁni- tions.

Variable Description Forecast bias is defined as the absolute difference between EPS of MFs and EPS of analysts’ consensus forecasts. In particular, the F AbsBias aggregate measure is the sum of the absolute differences for each fiscal quarter and the average measure is the arithmetic mean of the absolute differences in the fiscal year. Forecast error is defined as the absolute difference between EPS of MFs and actual EPS of announcement. In particular, the F AbsError aggregate measure is the sum of the absolute differences for each fiscal quarter and the average measure is the arithmetic mean of the absolute differences in the fiscal year. It’s a dummy variable that takes the value of unity if EPS of MFs matches actual EPS of announcement, and zero, otherwise. The specific rules that determine whether forecast value matches actual value are shown in Table 1B. Analogous to forecast error Accuracy and forecast bias, forecast accuracy is aggregate to firm-fiscal year level. In particular, the aggregate measure is the sum of the value of the dummy variables for each fiscal quarter and the average measure is the arithmetic mean of the values. It’s a dummy variable that takes the value of unity if the forecasted EPS (point forecast) or its lower bound (range or open- ended forecast) is greater than the analyst forecasted EPS as indicated by guidance code in I/B/E/S. The forecast optimism Optimism is aggregate to firm-fiscal year level. In particular, the aggregate measure is the sum of the value of the dummy variables for each fiscal quarter and the average measure is the arithmetic mean of the values. It denotes the fraction of executive titles held by the CEO relative to her subordinates. The metric is calculated as the number of executive titles of the CEO scaled by the total number of δ titles carried by the top management team of top five managers, including the CEO. The number of titles is calculated using our screening method built upon regex. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 270

Table 1.A.Cont.: Variable Deﬁnitions This table provide the list of variables in the empirical analysis and their deﬁni- tions.

Variable Description It is an instrumental variable that measures the overall degree of CEO power. It is defined as the sum of dummies that identifies P owerIndex the other three aspects of CEO power in addition to δ: prestige power, expertise power and ownership power. The detailed specification is shown in Table 1C. Daily turnover measure (in percentage) is daily trading volume T urnover divided by shares outstanding. In particular, Turnover is either annual average or median of daily turnover measure. Amihud Price Impact is constructed according to Amihud Amihud (2002). In particular I use both annual average and median for comparison. The industry adjusted M/B is defined as market-to-book ratio minus the industry median level market-to-book ratio. The me- M/B dian level is calculated at the level of 2-digit SIC industry-year for the Compustat universe. The sum of long term debt and debt in current liabilities divided Leverage by book value of assets at the beginning of the fiscal year. Size Natural log of total assets The research and development expenditures divided by book R&D value of assets at the beginning of the fiscal year. Directors Total number of directors serving on the board Outsiders Percentage of outsider directors Annual absolute difference in EPS (Compustat item EPSFI) ∆EPS | | scaled by stock price at the beginning of the period (lagged) Positive standard deviation of EPS of analyst’s forecasts scaled Disp by the absolute value of median EPS of analyst’s forecasts. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 271

Table 1.A.Cont.: Variable Deﬁnitions This table provide the list of variables in the empirical analysis and their deﬁni- tions.

Variable Description A dummy variable takes the value of unity if firms reports a loss, Loss i.e., net income (Compustat item NI) is less than zero, and zero, otherwise. #Analyst The number of analysts following the firm in the fiscal year. P rice Stock’s average daily closing price V olume Stock’s average daily dollar trading volume (in billions) S&P Average daily return of S&P 500 index Annualized standard deviation of daily stock returns (in percent- V olatility age) Dividend Dollar dividend paid per share outstanding CHAPTER 3: DELEGATION, INFORMATION AND MARKET 272 1*(1 val ≤ ≤ actual actual ≤ ≤ 2, then Accuracy = 1; zero 1*(1 + 0.05) 1*(1 - 0.05) val ≤ 0, If val 0, If val < > actual actual, then Accuracy = 1; zero otherwise. actual, then Accuracy = 1; zero otherwise. 1 1 ≤ ≤ ≥ 1 1 1 1*(1 - 0.05), then Accuracy = 1; zero otherwise. If val If val If val Given val val + 0.05), then Accuracy = 1; zero otherwise. otherwise. Given val desc Code Text Rules 149 Comfortable with Break even 0103 Between 10 More than Less than 0211 About May exceed Accuracy = 0 0608 High end of Low end of 121613 Slightly more15 than Significantly more17 than Slightly less than Significantly less than Not to exceed Type of Forecast Range Range Forecast Open-ended forecast Point forecast General impression Table 1.B: Coding RulesThis for table MFs reports Accuracy the specific rules of determining whether or not actual EPS matches different types of MFs. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 273 higher than industry median by 2-digit sic; zero otherwise. than industry median; zero otherwise. zero otherwise. Chair/Vice ChairFounder Takes the value of 1 if CEODirector is the chair or the vice chair; zeroExpert otherwise. Takes the value of 1 if CEO is theShare founder; Takes zero the otherwise. value of 1 if CEO is the only Takes directorpay the on value the of board; 1 zero if otherwise. the number of the segments of business of Takes the the value firm of is 1 if the percentage of shares owned by the Takes CEO the value is of higher 1 if the pay slice of the CEO is higher than industry median; Dimension of power VariablePrestige power Expertise power Value Ownership power Overall power PowerIndex Sum of above variables together and formulate the index. Table 1.C: Construction ofThis CEO table Power reports Indexpower. of the Prestige specific Power, components Expertise of Power and the Ownership overall Power power of the CEO relative her industry peers besides structural CHAPTER 3: DELEGATION, INFORMATION AND MARKET 274

Table 1.D: Data Merging and Sample Selection Procedures This table reports the detailed steps of sample construction whereby combing data from various databases such as Execucomp, I/B/E/S, CRSP and Compustat.

Sample Construction Procedure # of Observations

Obtain executive title strings from Execucomp for S&P 201900 1500 companies from 1996 through 2017

Drop if missing TDC1 (total compensation) 199900

Sort executives by TDC1 and select only top 161840 five executives per firm-fiscal year Drop if the management team per firm-fiscal year doesn’t clearly identify CEO in the title 156435 string, has dual CEO or reports less than five top executives.

Aggregate the data from executive-ﬁscal year 31287 level to ﬁrm-fsical year level by calculating δ

Drop if missing CEOANN and/or BECAM- 29323 CEO

Merge the sample of δ from Execucomp with I/B/E/S 29323 ticker using the link table between Compustat and I/B/E/S provided by WRDS.

Drop if missing I/B/E/S ticker in the link 28816 table

Merge the sample with the data ﬁle of DET GUIDANCE 102120 in I/B/E/S for EPS of MFs, EPS of analysts’ consensus forecasts, and actual EPS in each ﬁscal quarter.

Aggregate the quarterly data to firm-fiscal year level by 28816 calculating metrics for the quality of corporate voluntary financial disclosure.

Merge the sample with data from daily stock ﬁle of CRSP 7194483

Aggregate and merge with ﬁrm level fundamentals from 28816 Compustat

Merge with Board characteristics from ISS 28816 CHAPTER 3: DELEGATION, INFORMATION AND MARKET 275

Table 2.A: Summary Statistics of Total Sample Please refer to Panel A of Table 1 for variable deﬁnitions. The sample consists of S&P 1500 ﬁrms and ranges from 1996 to 2017. The process of sample selection and construction is exhibited in Panel D of Table 1 and explained in detail in Section III.

Mean Median p25 p75 Std. Dev. Skewness Kurtosis

δ 0.263 0.250 0.222 0.300 0.068 0.544 3.761 F AbsBiasAvg 0.099 0.050 0.025 0.101 0.174 5.133 34.314 F AbsBiasSum 0.501 0.262 0.119 0.560 0.753 3.809 20.595 F AbsErrorAvg 0.354 0.183 0.070 0.397 0.575 4.249 24.924 F AbsErrorSum 2.045 0.980 0.285 2.445 3.212 3.666 19.714 AccuracyAvg 0.352 0.250 0.000 0.556 0.338 0.667 2.232 AccuracySum 1.972 1.000 0.000 3.000 2.306 2.110 10.752 OptimismAvg 0.127 0.000 0.000 0.200 0.213 2.116 7.606 OptimismSum 0.796 0.000 0.000 1.000 1.360 2.742 15.224 P owerIndex 2.240 2.000 1.000 3.000 1.164 0.140 2.671 M/B 1.192 0.430 -0.228 1.682 4.166 2.160 38.502 Leverage 0.240 0.219 0.070 0.355 0.211 2.515 28.247 Size 7.713 7.604 6.488 8.828 1.713 0.376 3.209 R&D 0.034 0.000 0.000 0.034 0.080 7.037 100.009 Directors 9.499 9.000 8.000 11.000 2.511 0.962 6.331 Outsiders 0.718 0.778 0.600 0.875 0.196 -1.032 3.341 ∆EPS 5.624 0.019 0.007 0.056 580.464 139.106 20876.570 Disp 0.492 0.362 0.137 0.675 1.918 108.193 14752.560 Loss 0.179 0.000 0.000 0.000 0.383 1.674 3.801 #Analyst 13.595 11.000 6.000 19.000 9.978 1.083 4.184 P rice 36.286 28.772 16.585 45.398 42.891 15.740 582.129 V olume 59.837 13.825 4.196 48.899 188.162 17.829 601.089 S&P 0.000 0.000 0.000 0.001 0.001 -1.303 12.037 V olatility 42.417 35.982 26.043 51.162 27.015 12.293 701.273 Dividend 0.519 0.160 0.000 0.750 1.103 17.543 669.260 T urnoverAvg 0.963 0.721 0.445 1.179 0.915 6.980 178.138 T urnoverMed 0.781 0.594 0.362 0.962 0.739 9.120 360.813 AmihudAvg 0.110 0.002 0.000 0.008 6.898 154.445 25012.350 AmihudMed 0.021 0.001 0.000 0.006 0.260 60.526 5321.284 CHAPTER 3: DELEGATION, INFORMATION AND MARKET 276

Table 2.B: Test of Difference Between Firm-Year Observations With and Without Voluntary Disclosure This table reports the means/ medians of firm characteristics and liquidity metrics for the group that engages in voluntary disclosure and the group that doesn’t. It also reports the results for the test of difference in mean and median by t-test and Wilcoxon z-test, respectively. The sample consists of S&P 1500 firms and ranges from 1996 to 2017. ***, ** and * denote significance levels at 1%, 5% and 10%, respectively.

Voluntary Disclosure: Voluntary Disclosure: Test of Diﬀerence

Yes No P-value

Mean Median Mean Median t-test Wilcoxon z-test

δ 0.261 0.250 0.265 0.250 0.000*** 0.000***

P owerIndex 2.299 2.000 2.178 2.000 0.000*** 0.000***

M/B 1.291 0.509 1.075 0.335 0.000*** 0.000***

Leverage 0.233 0.226 0.248 0.211 0.000*** 0.333

Size 7.811 7.701 7.605 7.492 0.000*** 0.000***

R&D 0.034 0.002 0.034 0.000 0.806 0.0000***

Directors 9.439 9.000 9.575 9.000 0.000*** 0.087*

Outsiders 0.707 0.750 0.731 0.800 0.000*** 0.000***

∆EPS 0.545 0.017 11.271 0.024 0.143 0.000***

Disp 0.458 0.359 0.531 0.365 0.003*** 0.621

Loss 0.146 0.000 0.214 0.000 0.000*** 0.000***

#Analyst 14.992 13.000 12.096 9.000 0.000*** 0.000***

T urnoverAvg 0.967 0.759 0.958 0.673 0.618 0.000***

T urnoverMed 0.792 0.631 0.768 0.544 0.016** 0.000***

AmihudAvg 0.025 0.001 0.203 0.003 0.035** 0.000***

AmihudMed 0.010 0.001 0.033 0.002 0.000*** 0.000*** CHAPTER 3: DELEGATION, INFORMATION AND MARKET 277 δ , and the degree AbsBias F δ and AbsBias F . Figure 1.A is for the whole sample, whereas Figure 1.B and Figure δ Figure 1.A Figure 1.B Figure 1.C This figure delineates the relationship between the quality of voluntary disclosure, as measured by Figure 1: Graphical Analysis of the Curvilinear Relation Between of delegation responsibilitiesranging between from CEO 1 and toforecast her 100. metric subordinates against for Y-axis1.C the the denotes exhibit groups whole aggregate the of sample. relationships oragainst percentiles in average the X-axis of the quality group denotes samples median. metric. of theLocal Based older Polynomial percentile The on and Regression the of box younger Fitting. redrespectively, plot CEOs. line, for The is the The visualization vertical dash constructed solid purpose. values line line of whereby in in median plotting black red plot is shows and the the the fitted plot predicted smooth of curve curve forecast are produced bias tripled by the and method quintupled, of CHAPTER 3: DELEGATION, INFORMATION AND MARKET 278 , and the AbsError F δ and AbsError F . Figure 2.A is for the whole sample, whereas Figure 2.B and Figure δ Figure 2.A Figure 2.B Figure 2.C ranging from 1 to 100. Y-axis denotes aggregate or average quality metric. The box plot is constructed whereby plotting δ This figure delineatesdegree the of relationship delegation between responsibilitiesof the between CEO quality and offorecast her voluntary subordinates metric disclosure, for against as the2.C the whole measured exhibit groups sample. the by of relationship X-axisagainst percentiles in denotes the of group the the median. percentile sampleLocal Based of Polynomial on older Regression the and Fitting. redrespectively, younger line, for The CEOs. the visualization vertical dash purpose. values The line of solid in median line black plot is in and the red the fitted shows predicted smooth the curve curve plot are produced tripled of by the and forecast method quintupled, bias of Figure 2: Graphical Analysis of the Curvilinear Relation Between CHAPTER 3: DELEGATION, INFORMATION AND MARKET 279 δ , and the degree Accuracy δ and Accuracy Figure 3.C . Figure 3.A is for the whole sample, whereas Figure 3.B and Figure δ Figure 3.A Figure 3.B This figure delineates the relationship between the quality of voluntary disclosure, as measured by Figure 3: Graphical Analysis of the Curvilinear Relation Between of delegation responsibilitiesranging between from CEO 1 and toforecast her 100. metric subordinates against for Y-axis3.C the the denotes exhibit groups whole aggregate the of sample. relationships oragainst percentiles in average the X-axis of the quality group denotes samples median. metric. of theLocal Based older Polynomial percentile The on and Regression the of box younger Fitting. redrespectively, plot CEOs. line, for The is the The visualization vertical dash constructed solid purpose. values line line of whereby in in median plotting black red plot is shows and the the the fitted plot predicted smooth of curve curve forecast are produced bias tripled by the and method quintupled, of CHAPTER 3: DELEGATION, INFORMATION AND MARKET 280 δ , and the degree Optimism δ and Optimism Figure 4.C . Figure 4.A is for the whole sample, whereas Figure 4.B and Figure δ Figure 4.A Figure 4.B This figure delineates theof relationship between delegation the responsibilities quality ofranging between voluntary from CEO disclosure, 1 and as measured toforecast her by 100. metric subordinates against for Y-axis4.C the the denotes exhibit groups whole aggregate the of sample. relationships oragainst percentiles in average the X-axis of the quality group denotes samples median. metric. of theLocal Based older Polynomial percentile The on and Regression the of box younger Fitting. redrespectively, plot CEOs. line, for The is the The visualization vertical dash constructed solid purpose. scales line line whereby of in in plotting median black red is plot shows the and the fitted the plot smooth predicted of curve curve forecast produced are bias by tripled the and method quintupled, of Figure 4: Graphical Analysis of the Curvilinear Relation Between CHAPTER 3: DELEGATION, INFORMATION AND MARKET 281 remains consistent for samples of even older AbsError Accuracy Optimism δ δ AbsBias F are in the parentheses and ***, **, * denote levels of significance at 10%, Panel A: The whole Sample Panel C: CEO age less than 56 t-statistics Panel B: CEO age greater than or equal to 56 AbsError Accuracy Optimism F Average Forecast Metrics Aggregate Forecast Metrics (1) (2) (3) (4) (5) (6) (7) (8) 0.122 0.154 0.266 0.347 0.109 0.262 0.255 0.244 0.139 0.217 0.307 0.348 0.090.1920.151 -0.166 0.141 1.0330.177 1.552 0.275 -2.721 0.619 0.295 0.293 -2.565 0.184 -0.442 1.758 0.382 -1.009 2.857 0.391 0.097 -0.535 -0.483 0.338 -0.401 0.308 0.339 (3.46) (2.82) (-2.53) (1.54) (3.91) (2.97) (-2.20) (1.79) (2.37) (2.30) (-0.77) (0.71) (2.38) (1.91) (-1.44)(0.12) (0.16) (0.06) (-0.12) (0.43) (1.32) (-1.30) (0.36) (-0.87) (-0.31) (0.69) (-0.65) (1.05) (0.31) (-0.24) (-0.30) (-0.14) AbsBias F (-3.43) (-2.78) (2.78) (-1.68) (-3.92) (-2.73) (2.38) (-2.40) (-2.18) (-1.84) (0.95) (-0.83) (-2.12) (-1.36) (1.64) (-0.75) 2.578** 2.219** -0.588 0.956 2.361** 1.995* -1.395 0.211 -4.143** -3.115* 1.274 -1.943 -3.668** -2.534 2.825 -1.676 4.803*** 3.806*** -2.611** 3.207 5.124*** 4.678*** -3.069** 3.733* -8.227*** -6.559*** 4.922*** -6.181* -8.940*** -7.745*** 5.896** -8.561** F 2 2 2 2 2 2 δ Year FEsIndustry FEsPseudo R yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes δ δ Year FEsIndustry FEsPseudo R N yes yesδ yes yes 10,091 yes 11,795 yesN 12,286 yesδ yes 12,286δ 5,363Year FEs yesIndustry yes FEs 10,091Pseudo R N 6,078 yes yes 11,795 yes yes 6,349 12,286 yes yes yes 6,349 yes 12,286 4,728 yes yes yes yes 5,717 5,363 yes yes 5,937 6,078 5,937 6,349 yes yes 6,349 4,728 yes yes 5,717 yes yes 5,937 yes 5,937 yes ) in the simple quadratic specification for the whole sample and subsamples by CEO age using GLM estimator over the δ Table 3: Simple QuadraticThis table Regressions reports of the empirical Disclosuremeasure results Quality of ( Metrics regressing the on fourperiod voluntary disclosure from quality 1996 measures respectively tosensitivity on check 2017. responsibility of sharing CEO OLS ages.CEOs results and The are hump-shaped samples qualitatively relationship that similar5% between include and and forecast younger thus 1%, bias CEOs. respectively. and are not reported. The statistical inference is robust to the CHAPTER 3: DELEGATION, INFORMATION AND MARKET 282 AbsError Accuracy Optimism are in the parentheses and ***, **, * denote levels for the Whole Sample δ AbsBias F t-statistics The whole Sample AbsError Accuracy Optimism F Average Forecast Metrics Aggregate Forecast Metrics (1) (2) (3) (4) (5) (6) (7) (8) 0.009 0.113*** 0.030*** -0.001 0.076*** 0.179*** 0.109*** 0.038*** 0.0040.147 -0.023** 0.000 -0.1000.302 -0.025** 0.039 0.461 0.876*** 0.008 0.364 -0.010 -0.012 0.426 -0.162 0.008 -0.204** -0.023** 0.365 0.624*** 0.353 0.341 0.375 (0.62)(3.14) (8.25) (0.99) (3.30)(0.36) (-1.09) (-0.04)(1.05) (-2.57) (-3.33) (-0.80) (5.41) (0.07) (4.41) (0.51) (-2.52) (13.05) (9.24) (3.40) (10.86) (0.82) (2.62) (0.54) (-0.07) (-1.30) (-5.62) (-1.28) (1.24) (-2.40) (-2.32) (6.24) (2.12) (2.40) (-0.58) (1.14) (1.56) (1.47) (-0.80) (0.35) -0.616 -0.342 -0.331* -0.609** -0.119 -0.245 -0.389* -0.578* AbsBias F (-4.00) (-1.28) (-0.14) (-1.66) (-2.08) (0.09) (2.64) (-0.63) (-2.16) (-2.28) (0.59) (-1.54)(-1.63) (-0.96) (-1.65) (-1.75) (-1.22) (-2.09) (0.84) (-1.09) (-0.35) (-0.82) (-1.75) (-1.91) 2.239** 2.426** -0.443 1.547 1.627 1.656 -0.775 0.474 -3.946** -3.947** 0.787 -3.608 -2.972* -2.431 1.430 -2.525 0.420*** 0.116 -0.082 -0.415*** 0.576*** 0.390*** 0.047 -0.680*** -0.015*** -0.005 -0.000 -0.008* -0.007** 0.000 0.007*** -0.003 F 2 D ) in the multivariate quadratic specification for the whole sample. The unit of the analysis is firm-year. The sample consists & 2 δ Size Leverage R Outsiders Year FEsIndustry FEsPseudo R N yes yes yes yes 7,650 yes yes 9,067 yes yes 9,426 9,426 yes yes 7,650 yes yes 9,067 yes yes 9,426 yes 9,426 yes M/B Directors δ δ Table 4.A: Multivariate Regressions of Disclosure Quality Metrics on of significance at 10%, 5% and 1%, respectively. This table reports the empiricalmeasure results ( of regressing the fourof voluntary disclosure S&P quality measures 1500 respectivelyGeneralized firms on responsibility Linear and sharing Models thevariable. (GLM) period OLS to ranges results improve are from qualitatively the 1996 similar estimation and to thus efficiency 2017. are for not reported. left-censored The and results asymmetrically are distributed obtained dependent using Maximum Likelihood Estimation for CHAPTER 3: DELEGATION, INFORMATION AND MARKET 283 remains consistent δ AbsError Accuracy Optimism are in the parentheses and ***, **, * denote levels for the Sample of Older CEOs δ AbsBias F t-statistics CEO age greater than or equal to 56 AbsError Accuracy Optimism F Average Forecast Metrics Aggregate Forecast Metrics (1) (2) (3) (4) (5) (6) (7) (8) 0.0210.123 -0.031** 0.003 -0.2460.260 0.015 0.100 0.229 -0.247 0.035*** 0.332 -0.012 -0.086 0.447 0.024*** -0.331* 0.016 0.390 -0.238** -0.248 0.329 0.312 0.298 (3.23) (3.09) (-2.24) (2.20) (3.45) (3.36) (-1.84) (1.78) (3.01) (1.75) (-1.83) (-1.73) (3.43) (3.19) (1.15) (-2.62) (1.64)(0.71) (-2.45) (-1.40) (0.38) (0.94) (1.11) (-1.30) (2.85) (-0.51) (-0.94) (-1.90) (2.78) (-2.01) (1.21) (-1.26) -0.001 0.134*** 0.029** -0.049** 0.059*** 0.184*** 0.087*** -0.024 -0.163 -1.355*** -0.129 0.497 0.777 -0.565 0.608* 0.405 AbsBias F (-3.65)(-2.74) (-3.06) (-0.53) (2.40) (0.46) (-2.26) (-1.26) (-3.99) (-1.63) (-3.20) (-0.08) (1.91) (1.19) (-2.27) (-0.76) (-0.03) (7.25) (2.36) (-2.51) (3.15) (9.42) (6.53) (-1.28) (-0.29) (-2.71) (-0.40) (1.16) (1.44) (-1.00) (1.78) (0.85) 4.225*** 4.443*** -2.337** 4.448** 4.579*** 5.147*** -2.598* 3.779* 0.480*** 0.251* -0.188* -0.270* 0.534*** 0.500*** 0.135 -0.405*** -8.077*** -7.603***-0.013*** 4.274** -0.003 -8.002** 0.002 -9.035*** -8.685*** -0.008 4.735* -0.007 -8.250** -0.000 0.005 -0.004 F 2 D & 2 δ M/B Size δ Leverage R Directors N 4,074 4,819 5,028 5,028 4,074 4,819 5,028 5,028 Outsiders Year FEsIndustry FEsPseudo R yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes ) in the multivariate quadratic specification for CEOs older than or equal to 56 (the sample median age). The unit of the δ Table 4.B: Multivariate RegressionsThis of table Disclosure reports the Quality empiricalmeasure Metrics results of ( on regressing the fouranalysis voluntary disclosure is quality measures firm-year. respectivelyusing on responsibility The Maximum sharing sample Likelihood consistsand Estimation asymmetrically of for distributed Generalized S&P dependentinference Linear 1500 variable. is Models firms robust OLS (GLM) and to results to the the are sensitivity improve period check qualitatively the of ranges similarof estimation CEO from and significance age. efficiency at 1996 thus for The 10%, are to hump-shaped left-censored 5% not relationship 2017. and reported. between 1%, forecast The respectively. The bias results and statistical are obtained for samples of even older CEOs and samples that include younger CEOs. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 284 are in the parentheses and ***, **, * AbsError Accuracy Optimism t-statistics for the Sample of Younger CEOs δ AbsBias F CEO age less than 56 AbsError Accuracy Optimism F Average Forecast Metrics Aggregate Forecast Metrics (1) (2) (3) (4) (5) (6) (7) (8) 0.3110.039 -0.082 0.451 1.6010.009 -3.135 -0.073 0.091*** -1.600 0.029** -0.864 -0.039* 2.4260.208 -0.978 0.071*** 2.476 0.175*** 1.427 0.0310.363 0.132*** -0.720 -2.570 0.007 0.017 0.339 0.011 0.086 0.403 0.044 0.436 -0.074 0.292 -0.134 0.404 -0.164 0.378 0.403 (1.90) (0.10) (0.43) (1.09) (3.25) (1.46) (0.00) (-0.71) (0.20)(0.01) (-0.06) (0.18) (1.35)(0.39) (-0.04) (-1.49) (4.49) (-0.53) (-0.55) (2.14) (0.87) (-0.61) (-1.83)(1.04) (0.88) (1.14) (3.50) (0.19) (-0.42) (-1.18) (0.07) (9.10) (0.00) (0.44) (8.46) (0.87) (0.21) (-0.43) (-1.09) (-0.83) -0.788 -0.135 -0.424* -0.074 -0.529 -0.179 -1.099*** 0.073 -0.010 -0.014 -0.005 -0.013 -0.004 -0.008 -0.009 -0.012 AbsBias F 0.403* 0.017 0.046 0.198 0.654*** 0.235 0.000 -0.122 (-1.60) (-0.37) (-1.75) (-0.19) (-1.21) (-0.52) (-3.09) (0.20) (-3.62) (-2.11) (-0.84) (0.67)(-0.65) (-1.64) (-1.10) (-0.57) (0.22) (-0.85) (2.61) (2.33) (-0.29) (-0.62) (-0.92) (-0.82) -0.019*** -0.010** -0.003 0.004 -0.010 0.001 0.010*** 0.015** F 2 D & 2 R Directors δ M/B Leverage Year FEsIndustry FEsPseudo R N yes yes yes yes 3,576 yes yes 4,248 yes yes 4,398 4,398 yes yes 3,576 yes yes 4,248 yes yes 4,398 yes yes 4,398 δ Size Outsiders ) in the multivariate quadratic specification for CEOs younger than 56 (the sample median age). The unit of the analysis is firm-year. δ Table 4.C: Multivariate RegressionsThis of table Disclosure reports Quality the empirical( Metrics results on of the fourThe voluntary sample disclosure consists quality of measuresEstimation S&P respectively for 1500 on Generalized firms responsibility Linear and sharing Modelsdependent the measure (GLM) variable. period to ranges OLS improve from results thedenote 1996 estimation are levels to efficiency qualitatively of for 2017. similar significance left-censored The and at and results thus 10%, asymmetrically are are 5% distributed obtained and not using 1%, reported. Maximum respectively. Likelihood CHAPTER 3: DELEGATION, INFORMATION AND MARKET 285 AbsError Accuracy Optimism AbsBias F are in the parentheses and ***, **, * denote levels of t-statistics for the Sample of Older CEOs With Heckman Correction ) in the multivariate quadratic specification for CEOs older than or δ δ AbsError Accuracy Optimism F Average Forecast Metrics Aggregate Forecast Metrics 0.281 0.193 0.351 0.462 0.439 0.313 0.334 0.322 0.223 -0.315* 0.117 -0.220 -0.009 -0.376** -0.242** -0.312 -0.006 0.001 0.002 -0.007 -0.004 -0.000 0.003 -0.005 AbsBias F -1.176** -1.933*** -0.081 0.137 0.086 -0.933 0.900** 0.996* 4.867*** 5.097*** -2.475** 3.1560.182***0.589*** 0.226*** 4.801*** 0.298** 0.021 4.910*** -0.196* -0.034 -3.014** -0.000 1.813 0.166*** 0.632*** 0.221*** 0.532*** 0.013 0.095 -0.075** 0.100 0.037*** -0.0193.115*** 0.004 1.579*** 0.012 -0.066 0.112 0.044*** -0.005 1.805*** 0.020** 0.690** -1.239*** 0.012 -1.013*** -9.262*** -8.616*** 4.506** -5.933* -9.546*** -8.244*** 5.426** -4.795 F 2 D & 2 δ δ M/B Size Leverage R Directors Outsiders Imr 0.001 0.150 Pseudo R 0.044 (0.53)(5.39) (6.80) (3.75) (9.27) (2.07) (1.31) (-1.86) (-1.16) (-0.00) (6.74) (4.01) (8.04) (3.32) (0.71) (0.80) (-2.55) (0.61) (6.94)(4.30) (-1.35) (0.11) (0.48) (-2.20) (-1.19) (-3.82) (-0.97) (-0.25) (0.24) (-0.06) (0.60) (0.16) (-1.01) (-1.59) (2.57) (1.75) (0.54) (9.83) (5.91) (-0.36) (0.33) (6.33) (2.16) (-5.82) (-2.93) (2.97) (1.32) (-1.81) (1.08) (-1.12) (-0.05) (-2.12) (-2.01) (-1.55) -0.015 (-5.37) (1) (2) (3) (4) (5) (6) (7) (8) (-6.93)(-1.64) (3.58) (-4.05) (3.67) (-3.58) (-2.31) (2.46) (1.53) (-1.65) (3.39) (-3.96) (3.12) (-2.97) (-2.10) (2.17) (0.85) (-1.31) (-4.69) (2.88) (-1.56) (0.49) (0.84) (3.49) (-0.35) (2.21) (0.83) 0.010*** 0.060*** 0.271*** 0.017*** F orecast -0.283*** -0.211*** -0.961*** 2 | D EPS Analysts & ∆ Heckman First Stage| loss disp # Size Leverage CEO age greater than or equal to 56 M/B R N 18,265 N 3,873 4,600 4,791 4,791 3,873 4,600 4,791 4,791 Directors Year FEsIndustry FEsPearson chi2Gof p-valuePseudo R yes 18357.22 yes Year 0.365 FEs Industry FEs yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes Outsiders Table 5.A: Regressions of Disclosure Quality Metrics on equal to 56 (thethe sample model median specification age)significance to using at mitigate GLM 10%, estimator the 5% over concern and the 1%, of period respectively. selection from bias. 1996 to 2017. The inverse mills ratio is included in This table reports thequality measures first respectively stage on results responsibility of sharing Heckman measure correction ( and the empirical results of regressing the four voluntary disclosure CHAPTER 3: DELEGATION, INFORMATION AND MARKET 286 ) is b δ are in the parentheses and AbsError Accuracy Optimism t-statistics AbsBias F ) in the multivariate quadratic specification for CEOs older δ for the Sample of Older CEOs With Endogeneity Control δ AbsError Accuracy Optimism F Average Forecast Metrics Aggregate Forecast Metrics 0.006 0.135*** 0.026** -0.040* 0.056*** 0.177*** 0.076*** -0.031 0.0210.156 -0.031** 0.003 -0.240 0.016 0.0880.260 -0.208 0.036*** 0.231 -0.011 -0.099 0.332 0.025*** -0.373** 0.017 0.446 -0.290** -0.283 0.389 0.289 0.313 0.298 (1.59)(0.89) (-2.48) (-1.36) (0.41) (0.82) (1.14) (-1.10) (2.89) (-0.57) (-0.85) (-2.11) (2.93) (-2.44) (1.24) (-1.44) -0.243 -1.370*** -0.102-2.060 0.236 -0.330 0.810 0.907 -0.454 -0.130 0.728** 0.775 0.450 2.274 3.976*** 2.435 AbsBias F (-1.13) (-0.18) (0.72) (-0.06) (0.43) (1.23) (2.66) (1.11) 4.466*** 4.480*** -2.410** 3.667* 4.483*** 4.892*** -2.920** 3.475* 0.445*** 0.247* -0.174* -0.017 0.547*** 0.531*** 0.198 -0.370** -8.396*** -7.652***-0.013*** 4.363** -0.003 -6.592* 0.001 -8.907*** -0.007 -8.335*** 5.131** -0.007* -7.819** -0.001 0.004 -0.005 F 2 D & 2 b δ δ M/B Size Leverage R Directors Outsiders δ δ 0.427 Pseudo R (1.82) (2.23) (2.70) -0.000-0.000 (0.28) (2.73) (7.07) (1.69) (2.05) (-1.66) (-1.77) (-0.09) (2.93) (3.43) (8.80) (3.38) (5.38) (1.64) (-1.60) (-2.35) 0.000* (3.39) (3.10) (-2.30) (1.81) (3.36) (3.16) (-2.07) (1.65) (-0.46) (-2.69) (-0.02) (17.76) 0.003** (-3.77) (-3.07) (2.44) (-1.86) (-3.92) (-3.04) (2.08) (-2.17) 0.009*** (1)0.011*** (2) (3) (4) (-0.43) (-2.72) (5) (-0.31) (0.44) (6) (1.47) (7) (-0.79) (8) (2.12) (0.94) -0.013*** (-2.62) (-0.51) (0.40) (-1.24) (-1.67) (-0.22) (0.99) (-0.84) 2 CF First Stage CEO age greater than or equal to 56 D & M/B Size Leverage R Directors P owerIndex Outsiders Year FEsFirm FEsF-statisticAdj.R yes yes 315.5 Industry Year FEs FEs yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes N 15,408 N 4,074 4,819 5,028 5,028 4,074 4,819 5,028 5,028 included in the model***, specification **, to mitigate * the denote confounding levels effects of of significance potential at endogeneity. 10%, 5% and 1%, respectively. Table 5.B: Regressions ofThis Disclosure table Quality reports Metrics thedisclosure on first quality stage measures results respectively of onthan responsibility Control or sharing Function equal measure (CF) ( to regression 56 and the (the empirical sample results median of age) regressing using the GLM four estimator voluntary over the period from 1996 to 2017. The endogeneity control ( CHAPTER 3: DELEGATION, INFORMATION AND MARKET 287 ) are included in the model specification b δ AbsError Accuracy Optimism AbsBias F for Older CEOs With Heckman Correction and Endogeneity δ are in the parentheses and ***, **, * denote levels of significance at 10%, 5% and CEO age greater than or equal to 56 t-statistics AbsError Accuracy Optimism F Average Forecast Metrics Aggregate Forecast Metrics (1) (2) (3) (4) (5) (6) (7) (8) 0.242 -0.324* 0.106 -0.296 -0.029 -0.414** -0.290** -0.311** 0.281 0.192 0.351 0.462 0.438 0.313 0.335 0.322 (3.67) (3.61) (-2.35) (1.85) (3.26) (2.94) (-2.29) (2.10) (6.93)(3.54) (9.19) (2.08)(2.87) (1.16) (-1.71) (-1.98) (-1.54) (-1.61) (0.51) (6.53) (4.06) (0.93) (7.78) (3.47) (3.53) (0.21) (1.23) (-2.64) (-0.27) (-2.79) (2.34) (1.10) (1.41)(9.81) (-1.85) (5.90) (0.97) (-0.34) (-1.50) (2.58) (-0.17) (6.36) (-2.31) (-2.39) (2.20) (-2.31) (-5.72) (-5.38) -0.006 0.000 0.002 -0.008 -0.004 -0.001 0.002 -0.006 -1.201 0.534 0.802 1.316 1.162 2.056 3.526** 1.874 AbsBias F (-4.12) (-3.54) (2.50) (-2.00) (-3.86) (-2.83) (2.31) (-2.83) (-1.28) (0.08) (0.43) (-1.19) (-1.03) (-0.17) (0.44) (-1.39) (-2.24) (-3.73) (-0.18) (1.19) (0.24) (-1.41) (2.84) (0.61) (-0.65) (0.30) (0.63) (0.64) (0.62) (1.10) (2.33) (1.22) -1.214** -1.911*** -0.059 0.526 0.133 -0.836 0.999*** 0.215 5.004*** 5.036*** -2.539** 3.884* 4.657*** 4.673*** -3.297** 3.075** 0.185***0.569*** 0.224*** 0.306** 0.0190.036*** -0.184* -0.040** -0.019 -0.259 0.163*** 0.004 0.215*** 0.652*** 0.004 0.560*** 0.013 -0.038*** 0.151 0.044*** -0.320*** -0.004 0.021** 0.010 3.114*** 1.583*** -0.062 0.300*** 1.812*** 0.706** -1.218*** -0.471*** -9.446*** -8.534*** 4.586** -7.290** -9.352*** -7.920*** 5.771** -7.318*** F 2 ) in the multivariate quadratic specification for CEOs older than or equal to 56 (the sample median age) using GLM δ D & 2 b δ δ M/B Size Leverage R Outsiders Directors δ Imr Year FEsIndustry FEsPseudo R N yes yes yes yes 3,873 yes yes 4,600 yes yes 4,791 4,791 yes yes 3,873 yes yes 4,600 yes yes 4,791 yes 4,791 yes to mitigate the concern of selection bias. 1%, respectively. Table 5.C: Regressions of DisclosureControl Quality Metrics on This table reportssharing the measure ( empirical resultsestimator over of the period regressing from the 1996 to four 2017. voluntary The disclosure inverse mills quality ratio and measures endogeneity respectively control ( on responsibility CHAPTER 3: DELEGATION, INFORMATION AND MARKET 288

Table 6.A: Regressions of Info on Stock Liquidity Metrics for the Whole Sample This table reports the empirical results of regressing the stock liquidity, as measured by turnover (T urnover) and Amihud Price Impact (Amihud), on informational efficiency dummy (Info) for the whole sample using GLM estimator over the period from 1996 to 2017. OLS results are qualitatively similar and thus are not reported. The model specification includes additional explanatory variables for stock liquidity. To be consistent with the firm-year unit of study, the dependent variables are either fiscal year average or median level of liquidity measures. t-statistics are in the parentheses and ***, **, * denote levels of significance at 10%, 5% and 1%, respectively.

The whole Sample

T urnoverAverage T urnoverMedian AmihudAverage AmihudMedian (1) (2) (3) (4)

Info 0.061*** 0.060*** -0.180*** -0.175*** (3.75) (3.71) (-5.41) (-5.47) M/B 0.002** 0.004*** -0.065*** -0.061*** (2.17) (3.42) (-16.22) (-16.29) Size 0.013** 0.031*** -0.778*** -0.753*** (2.31) (5.74) (-51.90) (-54.30) Leverage 0.166*** 0.123*** 1.013*** 0.898*** (6.11) (4.52) (8.94) (9.15) R&D 1.168*** 1.274*** -2.864*** -2.664*** (9.43) (9.95) (-9.43) (-9.80) Directors -0.036*** -0.037*** 0.021*** 0.017** (-15.43) (-15.61) (2.58) (2.23) Outsiders 0.036 0.038 0.217 0.217 (1.16) (1.19) (1.42) (1.62) P rice 0.001*** 0.001*** -0.002*** -0.002*** (3.39) (3.69) (-5.92) (-5.94) V olume -0.000*** -0.000*** 0.000 -0.000 (-3.56) (-4.03) (1.03) (-0.11) S&P 90.672*** 69.353*** -320.770** -312.904** (4.48) (3.33) (-2.21) (-2.16) V olatility 0.017*** 0.015*** 0.021*** 0.017*** (23.27) (20.49) (18.19) (15.15) Dividend -0.031*** -0.035*** 0.057 0.047 (-4.84) (-5.49) (1.63) (1.35) #Analysts 0.021*** 0.023*** -0.070*** -0.065*** (32.74) (34.64) (-39.25) (-39.28) Year FEs yes yes yes yes Industry FEs yes yes yes yes Pseudo R2 0.479 0.410 0.189 0.299 N 15502 15502 15502 15502 CHAPTER 3: DELEGATION, INFORMATION AND MARKET 289

Table 6.B: Regressions of Info on Stock Liquidity Metrics for Older CEOs This table reports the empirical results of regressing the stock liquidity, as measured by turnover (T urnover) and Amihud Price Impact (Amihud), on informational efficiency dummy (Info) for CEOs older than or equal to 56 (the sample median age) using GLM estimator over the period from 1996 to 2017. OLS results are qualitatively similar and thus are not reported. The model specification includes additional explanatory variables for stock liquidity. To be consistent with the firm-year unit of study, the dependent variables are either fiscal year average or median level of liquidity measures. t-statistics are in the parentheses and ***, **, * denote levels of significance at 10%, 5% and 1%, respectively.

CEO age greater than or equal to 56

T urnoverAverage T urnoverMedian AmihudAverage AmihudMedian (1) (2) (3) (4)

Info 0.085*** 0.090*** -0.722*** -0.230*** (4.11) (4.32) (-5.86) (-4.87) M/B 0.000 0.002 -0.074*** -0.055*** (0.20) (1.13) (-10.67) (-10.78) Size 0.020*** 0.037*** -0.708*** -0.729*** (2.58) (5.08) (-7.83) (-42.63) Leverage 0.219*** 0.162*** 0.558 0.755*** (5.73) (4.19) (1.32) (6.72) R&D 0.747*** 0.811*** -4.575*** -2.674*** (4.46) (4.58) (-8.60) (-6.99) Directors -0.033*** -0.034*** -0.059 0.010 (-11.06) (-11.10) (-0.93) (1.22) Outsiders 0.083* 0.085* -1.146*** 0.282* (1.92) (1.95) (-3.16) (1.84) P rice 0.001* 0.001* -0.002*** -0.002*** (1.77) (1.91) (-6.36) (-5.14) V olume -0.000*** -0.000*** -0.001*** -0.000*** (-5.46) (-6.12) (-3.48) (-3.50) S&P 68.607** 48.774* -44.684 -351.143* (2.47) (1.71) (-0.32) (-1.82) V olatility 0.018*** 0.016*** 0.019*** 0.018*** (28.97) (25.58) (4.46) (14.55) Dividend -0.037*** -0.041*** 0.191* 0.050 (-3.64) (-4.20) (1.71) (1.56) #Analysts 0.021*** 0.023*** -0.078*** -0.065*** (22.99) (25.30) (-16.29) (-34.93) Year FEs yes yes yes yes Industry FEs yes yes yes yes Pseudo R2 0.439 0.360 0.163 0.277 N 8329 8329 8329 8329 CHAPTER 3: DELEGATION, INFORMATION AND MARKET 290

Table 6.C: Regressions of Info on Metrics of Stock Liquidity for Younger CEOs This table reports the empirical results of regressing the stock liquidity, as measured by turnover (T urnover) and Amihud Price Impact (Amihud), on informational efficiency dummy (Info) for CEOs younger than 56 (the sample median age) using GLM estimator over the period from 1996 to 2017. OLS results are qualitatively similar and thus are not reported. The model specification includes additional explanatory variables for stock liquidity. To be consistent with the firm-year unit of study, the dependent variables are either fiscal year average or median level of liquidity measures. OLS results are qualitatively similar and thus are not reported. t-statistics are in the parentheses and ***, **, * denote levels of significance at 10%, 5% and 1%, respectively.

CEO age less than 56

T urnoverAverage T urnoverMedian AmihudAverage AmihudMedian (1) (2) (3) (4)

Info 0.035 0.030 -0.003 -0.010 (1.38) (1.18) (-0.06) (-0.25) M/B 0.005*** 0.007*** -0.071*** -0.067*** (3.55) (4.48) (-15.44) (-15.31) Size 0.016** 0.036*** -0.777*** -0.751*** (2.03) (4.51) (-48.14) (-49.10) Leverage 0.087** 0.051 1.075*** 1.041*** (2.38) (1.40) (9.59) (9.63) R&D 1.307*** 1.439*** -2.326*** -2.272*** (7.44) (7.91) (-7.01) (-7.71) Directors -0.040*** -0.041*** 0.010 0.010 (-10.92) (-11.08) (1.50) (1.52) Outsiders -0.014 -0.011 0.175 0.211** (-0.30) (-0.23) (1.58) (2.09) P rice 0.001*** 0.001*** -0.004*** -0.004*** (3.53) (3.91) (-10.26) (-10.29) V olume -0.000*** -0.000*** 0.000*** 0.000*** (-3.49) (-3.93) (4.90) (4.39) S&P 105.475*** 81.690*** -200.857* -173.167 (3.53) (2.64) (-1.84) (-1.62) V olatility 0.016*** 0.014*** 0.022*** 0.018*** (11.78) (10.35) (15.19) (13.27) Dividend -0.025*** -0.029*** -0.017* -0.023** (-3.03) (-3.34) (-1.65) (-2.25) #Analysts 0.022*** 0.023*** -0.064*** -0.060*** (24.36) (24.61) (-35.21) (-35.61) Year FEs yes yes yes yes Industry FEs yes yes yes yes Pseudo R2 0.513 0.517 0.185 0.337 N 7,173 7,173 7,173 7,173 CHAPTER 3: DELEGATION, INFORMATION AND MARKET 291 at the beginning of the δ are in the paratheses and ***, **, * denote t-statistics by Fama-French 3 Factor Model Delegation Quintiles Alpha Portfolio δ Figure 6.A Figure 6.B portfolio are regressed on Fama-French three factors controlling for market excess return, SMB δ Equally Weighted Monthly Portfolio Returns Value Weighted Monthly Portfolio Returns (3.95) (3.68)(6.18) (2.46) (4.75) (2.93) (6.28) (3.07) (6.38) (3.95) (6.15) (3.68) (6.18) (2.46) (4.75) (2.93) (6.28) (3.07) (6.38) (6.15) (-9.84) (-7.60) (-9.02) (-8.85) (-8.68) (-9.84) (-7.60) (-9.02) (-8.85) (-8.68) (26.50) (26.30) (25.66) (26.71) (28.51) (26.50) (26.30) (25.66) (26.71) (28.51) Smallest 2 3 4 Largest Smallest 2 3 4 Largest 0.566*** 0.561***0.908*** 0.389** 0.958*** 0.403***0.269*** 0.969*** 0.425*** 0.220*** 0.878*** 0.302*** 0.940*** 0.566*** 0.267*** 0.561*** 0.258*** 0.908*** 0.389** 0.958*** 0.269*** 0.403*** 0.969*** 0.220*** 0.425*** 0.878*** 0.302*** 0.940*** 0.267*** 0.258*** -0.283*** -0.233*** -0.287*** -0.245*** -0.241*** -0.283*** -0.233*** -0.287*** -0.245*** -0.241*** Alpha MarketExcessReturn SMBReturn HMLReturn ﬁscal year. The monthlyreturn, and returns HML of return. The stock returns are adjusted for delisting correction. Table 7.A: Equal-Weighted andThis Value-Weighted table reports the resultsportfolio, of respectively. time series The portfolio test. portfolios Panel are A formed and by B summarizes sorting the stocks results of intolevels equally quintiles of weighted according signiﬁcance and value at to weighted 10%, the 5% value and of 1%, respectively. CHAPTER 3: DELEGATION, INFORMATION AND MARKET 292

Table 7.B: Double Sorted δ Portfolio Alpha by Fama-French 3 Factor Model This table reports the results of time series portfolio test. The portfolios are formed by double stratifying sample by market capitalization (Size) and degree of delegation (δ). Size has been adjusted for dual class stocks and is sorted according to NYSE breakpoints for consistency. The weights of portfolio are determined by Size at the beginning of each month. The monthly returns of value-weighted portfolio are regressed on Fama-French three factors model controlling for market excess return, SMB return, and HML return. The stock returns are adjusted for delisting correction. t-statistics are in the paratheses and ***, **, * denote levels of signiﬁcance at 10%, 5% and 1%, respectively.

Size Quintiles Smallest 2 3 4 Largest Delegation Alpha 0.583** 0.551** 0.34 0.263 0.393** (2.33) (2.08) (1.61) (1.26) (2.32) MarketExcessReturn 0.852*** 1.00*** 0.907*** 0.850*** 0.904*** (14.29) (15.87) (17.96) (16.99) (22.32) SMBReturn 0.661*** 0.237** 0.174** -0.037 -0.333*** (8.70) (2.94) (2.70) (-0.587) (-6.46) HMLReturn -0.392*** -0.224*** -0.192*** -0.180*** 0.110*** (-7.81) (-4.22) (-4.57) (-4.31) (3.28) 2 Alpha 0.649** 0.531** 0.031 0.734** 0.317 (2.26) (2.30) (0.130) (2.81) (1.36) MarketExcessReturn 0.825*** 0.914*** 0.957*** 0.959*** 0.937*** (12.04) (16.61) (16.67) (15.40) (16.80) SMBReturn 0.604*** 0.320*** 0.114 -0.057 -0.384*** (6.91) (4.59) (1.55) (-0.72) (-5.40) HMLReturn -0.305*** -0.157*** -0.199*** -0.120** 0.001 (-5.29) (-3.40) (-4.13) (-2.30) (0.029) 3 Alpha 0.275 0.378* 0.335 0.353 -0.134 (1.051) (1.71) (1.44) (1.71)* (-0.61) MarketExcessReturn 1.010*** 0.894*** 0.980*** 0.853*** 1.028*** (16.19) (16.98) (17.66) (17.33) (19.56) SMBReturn 0.629*** 0.429*** -0.007 0.078 -0.390*** (7.91) (6.34) (-0.10) (1.27) (-5.82) HMLReturn -0.418*** -0.251*** -0.105** -0.102** 0.088** (-8.04) (-5.66) (-2.25) (-2.47) (2.01) 4 Alpha 0.588** 0.458** 0.524** 0.116 0.075 (2.52) (2.18) (1.99) (0.55) (0.413) MarketExcessReturn 0.880 0.893 0.873 0.943 0.844 (15.80) (17.79) (13.81) (18.83) (19.47) SMBReturn 0.833 0.174 0.184 -0.070 -0.362 (11.79) (2.72) (2.28) (-1.09) (-6.56) HMLReturn -0.350 -0.302 -0.312 -0.119 0.006 (-7.50) (-7.21) (-5.93) (-2.83) (0.153) Centralization Alpha 0.888*** 0.594** 0.178 0.162 0.192 (3.21) (2.62) (0.830) (0.79) (0.91) MarketExcessReturn 1.059 0.967 1.015 0.817 0.797 (16.06) (17.88) (19.84) (16.64) (15.85) SMBReturn 0.774 0.354 -0.0460 -0.167 -0.123 (9.19) (5.14) (-0.71) (-2.68) (-1.92) HMLReturn -0.424 -0.247 -0.0620 -0.140 -0.123 (-7.62) (-5.48) (-1.44) (-3.40) (-2.94) CHAPTER 3: DELEGATION, INFORMATION AND MARKET 293

Table 8: Meditation Analysis of δ’s Impact on Valuation Through Informa- tion Efficiency This table reports the results of meditation analysis. The dependent variable of model (1) is average level of Amihud Price Impact (Amihud) for the fiscal year. The dependent variable of models (2) through (4) is market-to-book ratio (M/B). Model (1) is estimated using Maximum Likelihood Estimation for Generalized Linear Model (GLM) with industry fixed effects dummies, whereas models (2) through (4) are estimated by OLS with firm fixed effects. The sample consists of S&P 1500 firms and the period ranges from 1996 to 2017. t-statistics are in the parentheses and ***, **, * denote levels of significance at 10%, 5% and 1%, respectively.

CEO age greater than or equal to 56 Amihud Average M/B (1) (2) (3) (4) δ 5.807*** 6.502*** 6.606*** (5.26) (2.79) (2.83) δ2 -11.045*** -10.182*** -10.337*** (-6.27) (-2.62) (-2.66) Amihud -0.328** -0.348** (-2.09) (-2.17) M/B -0.058*** 0.388*** 0.388*** 0.388*** (-10.78) (5.27) (5.27) (5.26) Size -0.759*** -0.944*** -0.949*** -0.947*** (-41.78) (-5.28) (-5.28) (-5.29) Leverage 0.917*** -0.507 -0.477 -0.484 (6.85) (-0.69) (-0.65) (-0.66) R&D -3.030*** 0.128 0.107 0.105 (-7.46) (0.07) (0.06) (0.06) Directors 0.013 -0.030 -0.031 -0.031 (1.51) (-1.24) (-1.27) (-1.25) Outsiders 0.368** 0.028 0.059 0.033 (2.20) (0.10) (0.21) (0.12) P rice -0.002*** 0.012*** 0.012*** 0.012*** (-5.05) (3.60) (3.63) (3.60) V olume -0.000** 0.001* 0.001* 0.001* (-2.23) (1.73) (1.77) (1.74) S&P -342.865* 380.088*** 381.807*** 378.665*** (-1.82) (3.42) (3.43) (3.40) V olatility 0.023*** 0.001 0.001 0.001 (17.64) (0.28) (0.35) (0.35) Dividend 0.058* 0.203** 0.205** 0.203** (1.82) (2.29) (2.32) (2.29) #Analysts -0.068*** 0.020** 0.021** 0.020** (-35.03) (2.20) (2.25) (2.18) Year FE yes yes yes yes Industry FE or Firm FE yes yes yes yes Pseudo R2 or Adjusted R2 0.082 0.202 0.202 0.202 N 8329 8186 8186 8186 CHAPTER 3: DELEGATION, INFORMATION AND MARKET 294

Table 9.A: Meditation Analysis of the Predictability of δ for Future Raw Return Through Information Efficiency This table reports the results of meditation analysis. The dependent variable of model (1) is average level of Amihud price impact (Amihud) for the fiscal year. The dependent variable of model (2) through (4) is annual stock return (SReturn) in the following fiscal year. Model (1) is estimated using Maximum Likelihood Estimation for Generalized Linear Model (GLM) with industry fixed effects dummies, whereas models (2) through (4) are estimated by OLS with firm fixed effects and standard errors clustering at firm and year level. The sample consists of S&P 1500 firms and the period ranges from 1996 to 2017. t-statistics are in the parentheses and ***, **, * denote levels of significance at 10%, 5% and 1%, respectively.

The Whole Sample Amihud Average SReturn (1) (2) (3) (4) δ 4.691*** 0.621** 0.603** (6.09) (2.23) (2.17) δ2 -8.589*** -0.988** -0.962** (-6.97) (-2.14) (-2.08) Amihud 0.115** 0.114** (2.36) (2.33) SReturn -0.169*** -0.169*** -0.169*** (-14.79) (-14.75) (-14.77) M/B -0.065*** -0.008*** -0.008*** -0.008*** (-16.26) (-5.85) (-5.81) (-5.80) Size -0.776*** -0.112*** -0.111*** -0.111*** (-51.68) (-8.92) (-8.84) (-8.85) Leverage 1.007*** 0.125*** 0.118*** 0.120*** (8.93) (2.90) (2.75) (2.78) R&D -2.870*** -0.337** -0.333** -0.334** (-9.49) (-2.25) (-2.23) (-2.23) Directors 0.021** -0.005* -0.005* -0.005* (2.55) (-1.81) (-1.84) (-1.82) Outsiders 0.224 -0.018 -0.018 -0.019 (1.46) (-0.54) (-0.55) (-0.58) P rice -0.002*** -0.001*** -0.001*** -0.001*** (-5.92) (-3.58) (-3.52) (-3.56) V olume 0.000 -0.000* -0.000* -0.000* (1.12) (-1.80) (-1.80) (-1.83) S&P -322.258** -75.158*** -74.322*** -74.768*** (-2.24) (-4.44) (-4.39) (-4.42) V olatility 0.021*** 0.002*** 0.002*** 0.002*** (18.20) (3.68) (3.68) (3.69) Dividend 0.055 0.003 0.003 0.003 (1.61) (0.71) (0.74) (0.70) #Analysts -0.070*** -0.008*** -0.007*** -0.008*** (-39.20) (-8.56) (-8.45) (-8.49) Year FE yes yes yes yes Industry FE or Firm FE yes yes yes yes Pseudo R2 or Adjusted R2 0.054 0.233 0.234 0.234 N 15502 14864 14864 14864 CHAPTER 3: DELEGATION, INFORMATION AND MARKET 295

Table 9.B: Meditation Analysis of the Predictability of δ for Future Abnor- mal Return Through Information Efficiency This table reports the results of meditation analysis. The dependent variable is cumulative abnormal return (CAReturn) in the following fiscal year. The models are estimated by OLS with firm fixed effects and standard errors clustering at firm and year level. The sample consists of S&P 1500 firms and the period ranges from 1996 to 2017. t-statistics are in the parentheses and ***, **, * denote levels of significance at 10%, 5% and 1%, respectively.

The Whole Sample CAReturn (1) (2) (3) δ 0.493* 0.479* (1.93) (1.87) δ2 -0.838** -0.816* (-2.01) (-1.96) Amihud 0.112** 0.111** (2.31) (2.29) SReturn -0.345*** -0.345*** -0.345*** (-31.75) (-31.69) (-31.71) M/B -0.008*** -0.008*** -0.008*** (-6.49) (-6.47) (-6.45) Size -0.039*** -0.038*** -0.038*** (-3.50) (-3.41) (-3.42) Leverage 0.038 0.033 0.034 (0.97) (0.84) (0.86) R&D -0.208 -0.204 -0.205 (-1.40) (-1.38) (-1.38) Directors -0.001 -0.001 -0.001 (-0.27) (-0.31) (-0.29) Outsiders 0.009 0.009 0.009 (0.32) (0.31) (0.30) P rice -0.001*** -0.001*** -0.001*** (-6.31) (-6.28) (-6.30) V olume -0.000 -0.000 -0.000 (-0.63) (-0.65) (-0.67) S&P 127.939*** 128.052*** 127.908*** (8.75) (8.77) (8.75) V olatility 0.000 0.000 0.000 (1.35) (1.31) (1.31) Dividend -0.001 -0.000 -0.001 (-0.14) (-0.12) (-0.15) #Analysts -0.003*** -0.003*** -0.003*** (-4.06) (-3.95) (-3.98)

Year FE yes yes yes Firm FE yes yes yes Adjusted R2 0.143 0.143 0.143 N 14838 14838 14838 CHAPTER 3: DELEGATION, INFORMATION AND MARKET 296 AbsError Accuracy Optimism AbsBias F ) is included in the model specification to mitigate the 0 ) for CEOs older than or equal to 56 (the sample median δ with Alternative Endogeneity Control δ P owerIndex AbsError Accuracy Optimism F Average Forecast Metrics Aggregate Forecast Metrics AbsBias F F 4.445*** 4.532***-8.379*** -7.721***-0.007 -2.346** 3.612* 4.274**0.064* 0.000 -6.506*0.148 0.168*** 4.720*** -9.229*** -0.001 5.207*** 0.003 0.087 -8.767*** -0.008 -2.640* -0.057 4.754* -0.057 3.713* -0.003 -8.154** 0.071 0.099*** 0.209*** 0.002 0.3270.259 0.021 -0.001 0.381* -0.049 0.231 -0.006 0.473*** 0.332 -0.278 0.445 0.389 0.289 0.312 0.298 -0.3650.011 -1.470***(0.80)0.411*(1.96) -0.060 -0.036*** (-2.79)-21.009** -0.101(-2.08) 0.292 0.007 -10.670 (-0.49) (0.83) (-1.10) 0.019 -0.009 8.429 0.643 (-0.06) (1.23) -0.282 (1.23) (-1.17) 5.480 0.029** -0.651 (2.17) (0.46) 0.096 -0.016 (0.45) -13.013 0.784** (-1.18) (-1.29) -0.224 (-1.03) 0.034*** 0.452 -7.670 0.020 (3.72) (-0.74) -0.515*** (-3.41) 21.542*** (1.41) -0.357 8.313 (-1.46) (2.75) (0.65) 2 are in the parentheses and ***, **, * denote levels of significance at 10%, 5% and 1%, respectively. D & 2 0 b δ δ δ M/B Size Leverage R Directors Outsiders by alternative measure of CEO power ( t-statistics δ 0.000**(1.97) 0.003**(2.38) -0.015***(-3.03) -0.001(-0.09) -0.000 (3.39)(-1.24) (-3.77) (3.16) (-1.38) (-3.12) (0.03) (-2.25) (1.76) (2.40) (1.80) (0.65) (-0.20) (4.81) (-1.85) (-1.32) (0.42) (3.57) (0.11) (-4.09) (-0.66) (3.41) (-0.38) (-1.34) (-3.24) (0.27) (0.31) (-1.87) (2.72) (1.93) (1.76) (1.45) (-0.20) (5.58) (-2.25) (-0.99) (1.74) (0.76) (2.72) (-1.17) (-1.12) δ (3.14) 0.414 Pseudo R 0.013***(3.19) (-0.67) (-2.94) (-0.18) (0.54) (1.18) (-1.14) (2.27) (0.94) 0.002*** (1) (2) (3) (4) (5) (6) (7) (8) 0 2 CF First Stage CEO age greater than or equal to 56 D & M/B Size Leverage R Directors N 15408 N 4,074 4,819 5,028 5,028 4,074 4,819 5,028 5,028 Outsiders Year FEsFirm FEsF-statistic yes Adj.R yes 9.86 Year Industry FEs FEs yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes P owerIndex Table 10.A: Regressions ofThis Disclosure table Quality reports Metrics thedisclosure on first quality stage measures results respectively ofage). on Control responsibility Function The (CF) sharing regression measure predicted concern and ( of the potential empirical endogeneity. resultsfrom The of 1996 unit regressing to the of 2017. four thethe The estimation voluntary analysis results efficiency is for are firm-year. left-censored obtainedthus and The using are asymmetrically Maximum sample not distributed Likelihood consists dependent reported. Estimation variables. of for OLS S&P Generalized results 1500 are Linear firms qualitatively Models and similar (GLM) and the to period improve ranges CHAPTER 3: DELEGATION, INFORMATION AND MARKET 297 are included in the 0 P owerIndex AbsError Accuracy Optimism are in the parentheses and ***, **, * denote ) estimated by b δ AbsBias F t-statistics with Alternative Endogeneity Control and Heckman δ CEO age greater than or equal to 56 AbsError Accuracy Optimism F Average Forecast Metrics Aggregate Forecast Metrics (1) (2) (3) (4) (5) (6) (7) (8) 0.350 0.209 -0.064 -0.062 0.487** 0.442** 0.416** -0.249 0.281 0.195 0.352 0.461 0.461 0.313 0.334 0.322 (9.82) (5.88) (-0.32) (2.60) (6.31) (2.14) (-5.76) (-5.40) (2.03) (-1.13) (0.05) (-1.89) (0.55) (-1.34) (-3.30) (-2.11) (2.16) (-1.73) (0.92) (1.29) (2.94) (-0.54) (3.10) (1.25) (3.68) (3.71)(5.63) (-2.32)(1.53) (1.89) (6.50) (1.00) (-0.17) (3.47) (-0.42) (-1.94) (3.16) (-0.25) (4.78) (-2.13) (2.12) (2.23) (5.72) (1.96) (-1.68) (2.36) (-1.78) (-1.45) -0.002 0.002 -0.001 -0.012 -0.001 0.001 -0.003 -0.008 AbsBias F (-1.48) (-0.60) (1.22) (1.19) (-0.88) (-0.55) (2.57) (0.73) (-4.14) (-3.61) (2.47)(-2.43) (-2.03) (-3.91) (-4.03) (-0.04) (-3.00) (1.31) (2.18) (-0.01) (-2.93) (-1.68) (3.00) (0.61) (-0.35) (0.38) (-0.17) (-1.58) (-0.27) (0.23) (-0.65) (-1.51) -15.120 -5.863 8.509 14.156 -9.173 -5.759 20.456** 6.272 0.430** -0.236 0.008 -0.463* 0.120 -0.295 -0.507*** -0.368** 0.030** -0.022* 0.008 0.019 0.040*** -0.008 0.029*** 0.013 -1.303** -1.992*** -0.014 0.585 -0.005 -0.997* 1.061*** 0.219 3.117*** 1.573*** -0.059 0.302*** 1.799*** 0.684** -1.228*** -0.472*** 5.019*** 5.147*** -2.484**0.228*** 3.923* 0.244*** 4.904*** -0.005 4.959*** -0.080* -3.052** 3.242** 0.194*** 0.239*** -0.050* -0.053* -9.479*** -8.683*** 4.506** -7.343** -9.690*** -8.313*** 5.440** -7.552*** F 2 D & 2 0 ) in the multivariate quadratic specification for CEOs older than or equal to 56 (the sample median age) using GLM estimator b δ Year FEsIndustry FEsPseudo R yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes Imr N 4,074 4,074 4,074 4,074 3,873 4,600 4,791 4,791 Outsiders M/B Directors δ Leverage R δ Size δ Correction This table reports the empiricalmeasure results ( of regressing the fourover voluntary the disclosure quality period measures from respectivelymodel on 1996 specification responsibility to sharing to 2017. mitigate Thelevels the of inverse concerns mills significance of ratio at selection and 10%, bias endogeneity 5% and control and endogeneity. ( 1%, respectively. Table 10.B: Regressions of Disclosure on Quality Metrics CHAPTER 3: DELEGATION, INFORMATION AND MARKET 298 AbsError Accuracy Optimism AbsBias F CEO age greater than or equal to 56 AbsError Accuracy Optimism F Average Forecast Metrics Average Forecast Metrics (1) (2) (3) (4) (5) (6) (7) (8) ) in the multivariate quadratic specification for CEOs older than or equal to 56 (the sample median age). 0 1.701 4.991*** -1.981** 2.8090.183 2.681** -0.1020.012 6.367***0.081 0.026 -1.360* -0.031** -0.270* 3.553* -0.199 0.0060.259 0.015 0.213 0.062 0.234 -0.251 -0.175 0.020 0.331 0.184*** -0.007 -0.411*** -0.010 0.447 -0.190 0.033*** 0.017 -0.208** 0.388 -0.262 0.289 0.312 0.298 (1.16) (3.99) (-2.04)(1.67)(1.22) (1.57) (7.14) (-0.70) (1.97)(0.83) (3.42) (0.29)(0.41) (-2.58) (-2.43) (4.44) (-1.73) (-1.08) (0.82) (-1.94) (3.99) (1.44) (0.59) (1.92) (1.15) (7.97) (-1.32) (-1.19) (1.42) (9.48) (2.78) (-0.03) (-1.38) (-0.77) (-2.66) (-1.00) (6.01) (-2.57) (1.24) (-1.33) AbsBias F 0.032* 0.124*** 0.038*** -0.050*** 0.074*** 0.146*** 0.077*** -0.026 (-2.01)(-2.93) (-4.21) (-0.87) (2.21) (0.27) (-1.54)(-4.26) (-1.27) (-5.43) (-2.74) (-2.05) (-0.91) (-4.34) (1.15) (0.26) (2.26) (2.15) (-2.23) (-3.68) (-0.78) (-4.40) (1.04) (0.85) δ -5.128** -9.662*** 3.880** -5.042 -6.775*** -11.891*** 2.820** -7.373** -0.015*** -0.005 0.001-2.013*** -2.313*** -0.008 -0.244 -0.009** 0.490 0.001 -1.631*** 0.005** -1.978*** -0.004 0.231 0.402 F 2 D are in the parentheses and ***, **, * denote levels of significance at 10%, 5% and 1%, respectively. & 2 0 0 M/B Size Directors δ δ Leverage R Outsiders Year FEsIndustry FEsPseudo R N yes yes yes yes 4,074 yes yes 4,819 yes yes 5,028 5,028 yes yes 4,074 yes yes 4,819 yes yes 5,028 yes 5,028 yes Table 11.A: Regressions ofThis Disclosure table Quality reports Metrics thesibility on empirical sharing results Alternative measure of Delegation regressing ( ResponsibilitiesThe the Measure unit four of voluntary disclosure theare quality analysis obtained measures is respectively using firm-year. onleft-censored Maximum alternative The Likelihood and respon- sample Estimation asymmetrically consistst-statistics for distributed of Generalized dependent S&P Linear variables. 1500 Models firms OLS (GLM) and results to the are improve period the qualitatively ranges similar estimation from efficiency and 1996 for thus to 2017. are not The reported. results CHAPTER 3: DELEGATION, INFORMATION AND MARKET 299 AbsError Accuracy Optimism AbsBias F ) in the quadratic specification for CEOs older 0 δ ) is included in the model specification to mitigate the b δ AbsError Accuracy Optimism F Average Forecast Metrics Aggregate Forecast Metrics 0.0700.659 -0.216 1.219 0.0520.259 0.975 -0.260 0.235 0.863 -0.065 0.331 2.885 -0.251 0.447 -0.235*** 3.462** -0.281 0.387 2.583*** 1.676 0.289 0.312 0.298 1.591 4.756*** -2.129**0.032 2.6640.193 0.123*** 0.037*** 2.141 -0.0810.013 -0.051 5.685*** 0.039 -0.030** -1.749** 0.072*** -0.258 0.006 3.255* 0.142*** 0.075*** 0.016 0.266 -0.026 -0.112 0.022 0.220*** -0.007 -0.388** 0.034*** 0.018 (0.29)(0.29) (-1.17) (0.84) (0.49) (-0.78) (1.08) (0.31) (-0.24) (1.27) (-1.30) (-2.89) (2.31) (-1.44) (3.74) (1.01) (0.75) (-2.37) (0.88) (0.67) (1.17) (-0.51) (6.18) (1.29) AbsBias F -4.984* -9.330*** 4.083** -4.834 -6.038** -10.924*** 3.353*** -6.943** -0.015** -0.005 0.001 -0.009 -0.009* 0.000 0.005** -0.005 -1.993*** -2.268*** -0.225 0.512 -1.526* -1.837*** 0.284 0.446 F 2 D 2 0 0 b δ δ & δ Size M/B R are in the parentheses and ***, **, * denote levels of significance at 10%, 5% and 1%, respectively. Leverage Directors Outsiders t-statistics δ 0.410 Pseudo R 0.001 (-1.83) (-4.01) (2.31) (-0.84) (-2.15) (-3.90) (2.68) (-2.12) (1.98) (1.78) (0.65) -0.003-0.000 (1.10) (0.50) (7.12) (-0.56) (3.36) (0.43) (-1.49) (-0.93) (2.78) (0.89) (7.85) (-0.79) (9.32) (3.28) (-1.41) (-2.48) 0.000* (1.09) (3.73) (-2.17) (0.84) (1.45) (3.85) (-2.48) (1.77) (-1.01) (-2.75) (-0.22) (23.63) 0.008** (-2.64) (-5.32) (-0.84) (0.60) (-1.77) (-4.09) (1.28) (0.93) 0.013*** (1) (2) (3) (4) (5) (6) (7) (8) -0.014*** (-2.53) (-0.94) (0.22) (-0.76) (-1.87) (0.09) (1.97) (-0.82) 2 CF First Stage CEO age greater than or equal to 56 D & Year FEsFirm FEsF-statisticAdj.R yes N 558.20 yes Industry FEs Year FEs yes 15408 yes yes N yes yes 4,074 yes yes 4,819 yes 5,028 yes 5,028 yes yes 4,074 yes yes 4,819 yes yes 5,028 yes 5,028 M/B R Directors Outsiders P owerIndex Size Leverage Table 11.B: Regressions ofThis Disclosure table Quality reports Metrics thedisclosure on first quality stage Alternative measures results Delegation respectively of on Measure Control alternative with Function responsibility Endogeneity (CF) sharing regression Control measureconfounding and ( effects the empirical of results potentialperiod of endogeneity. ranges regressing from the The 1996 fourto unit to voluntary improve of 2017. the the estimation The efficiency resultsare analysis for are not is asymmetrically obtained reported. firm-year. distributed by dependent Maximum The variables. Likelihood sample OLS Estimation consists results for are Generalized of qualitatively Linear S&P similar Models 1500 and (GLM) thus firms and the than or equal to 56 (the sample median age). The endogeneity control ( CHAPTER 3: DELEGATION, INFORMATION AND MARKET 300 ) are included in the model specification b δ AbsError Accuracy Optimism AbsBias F are in the parentheses and ***, **, * denote levels of significance t-statistics CEO age greater than or equal to 56 AbsError Accuracy Optimism F Average Forecast Metrics Aggregate Forecast Metrics (1) (2) (3) (4) (5) (6) (7) (8) 1.761 1.9690.281 1.018 0.196 0.925 0.350 2.377* 0.463 2.202 0.436 2.111*** 0.313 1.127 0.334 0.322 0.225 -0.1100.0150.117 0.036 -0.022* -0.264* -0.278 0.007 0.252* 0.078 0.013 -0.295 -0.083 0.019 0.248*** 0.294** -0.338*** -0.009 -0.297 0.027*** -0.210** 0.011 -0.313** 1.718 4.600*** -2.254** 2.256 2.366* 5.513*** -1.771** 2.873** (9.49)(1.30) (3.35) (1.40) (-0.00) (1.11) (2.54) (0.62) (3.58) (1.65) (-2.28) (-15.12) (1.44) (-5.49) (3.01) (1.00) (1.54) (-0.77)(1.11) (0.39)(0.58) (-1.71) (-1.65) (-1.47) (0.97) (1.72) (0.73) (0.97) (-1.51) (-0.56) (1.38) (3.60) (2.10) (-0.68) (-2.96) (-1.50) (4.81) (-2.51) (1.14) (-2.34) (3.49) (7.36) (3.03) (-1.97) (4.91) (7.40) (6.53) (-2.57) (1.22) (3.44) (-2.21) (1.22) (1.66) (3.59) (-2.42) (2.13) AbsBias F (-3.32) (-5.24) (-0.74) (1.18) (-3.51) (-4.56) (-0.22) (0.58) (-2.09)(-2.15) (-3.64) (-0.69) (2.35) (0.25) (-1.27) (-1.20) (-2.47) (-1.90) (-3.57) (-0.30) (2.73) (0.79) (-2.59) (-1.37) -5.200**-0.010** -8.898*** 4.299** -0.004 -4.256 0.001 -6.406** -0.008 -10.307*** 3.535*** -0.008* -6.410*** -0.002 0.002 -0.006 1.114*** 0.362*** -0.000 0.296** 0.411*** -0.343** -0.835*** -0.480*** 0.065*** 0.128*** 0.034*** -0.039** 0.092*** 0.138*** 0.055*** -0.036** -1.617*** -2.258*** -0.201 0.519 -1.642*** -2.084*** -0.050 0.206 F 2 ) in the multivariate quadratic specification for CEOs older than or equal to 56 (the sample median age) using GLM δ D & 2 0 0 b Year FEsIndustry FEsPseudo Adj. R yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes δ N 3,873 4,600 4,791 4,791 3,873 4,600 4,791 4,791 R Outsiders Imr Leverage Directors δ δ M/B Size at 10%, 5% and 1%, respectively. to mitigate the concerns of selection bias and endogeneity. Table 11.C: RegressionsHeckman of Correction DisclosureThis Quality table on reports Alternativesharing the measure Delegation ( empirical Measure resultsestimator with over of the Endogeneity period regressing from Control the 1996 to and four 2017. voluntary The disclosure inverse mills quality ratio and measures endogeneity respectively control ( on responsibility CHAPTER 3: DELEGATION, INFORMATION AND MARKET 301 ) in δ 75%) , 50% (25% ≥ AbsBias F 50% < Aggregate ) on responsibility sharing measure ( 75% ≥ AbsBias 25% F ≤ are in the parentheses and ***, **, * denote levels of significance 75%) , t-statistics 50% (25% as indicated at the top of each column. The unit of the analysis is firm-year. OLS ≥ δ Quantiles δ AbsBias 50% F on < Average AbsBias 75% ≥ F 25% (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) 0.187 0.4530.026 0.344* 0.667*** 0.007 0.590*** 0.030* 0.039** 0.281 0.037*** 0.611* 0.506*** 0.013 0.806*** 0.768** 0.048** 0.034** 0.059*** 0.057* 0.321 0.251 0.313 0.238 0.269 0.483 0.301 0.414 0.363 0.422 (1.70) (-4.35)(2.62) (2.32) (-3.78)(0.74) (6.69) (-1.18) (1.40) (3.79)(1.44) (1.81) (6.54) (2.17)(2.05) (0.31) (2.94) (6.81) (-3.40)(4.08) (2.65) (1.93) (3.69) (1.78) (3.55) (6.44) (1.20) (2.30) (-4.35) (1.41) (5.01) (7.26) (1.44) (2.94) (-0.79) (1.87) (3.97) (7.72) (1.20) (0.65) (5.06) (2.60) (9.77) (0.24) (2.06) (3.45) (2.00) (1.17) (0.49) (2.07) (2.27) (4.36) (3.42) (0.35) (-0.69) (4.45) (1.92) (5.20) (-1.05) (3.13) -0.009 0.003 -0.014** -0.001 -0.006 -0.010 -0.003 -0.002 -0.006 0.001 1.984* -3.195*** 1.691** -2.075*** -0.405 1.910** -2.785*** 1.401* -2.325*** -1.132 ≤ (-1.06) (0.38) (-2.24) (-0.14)(-1.96) (-1.27) (-0.78) (-1.53) (-1.46) (-1.52) (-0.45) (-2.23) (-0.23) (-1.16) (-3.60) (-0.55) (0.12) (0.58) (-0.05) (0.78) (-3.70) (-0.57) (-1.65) (0.60) (-0.24) (-0.96) (0.39) (-1.76) (2.35) (1.18) 0.472** 0.798*** 0.248 0.362 0.206 0.064 0.166 0.079 -0.167 -0.348 -1.431** -0.814 -0.913 -1.236 -1.206** -1.855*** -0.519 0.374 -0.046 0.978 0.101*** 0.372*** 0.124*** 0.254*** 0.181*** 0.095*** 0.288*** 0.130***1.815*** 0.183*** 4.808*** 0.111** 2.609*** 3.943*** 3.092*** 0.182 3.240*** 1.624*** 2.181*** 1.457*** -10.727*** -2.029 -3.940* 1.528 -0.432 -2.609 1.392 -4.223* 6.278** 5.195 2 D & b M/B Size Leverage R Directors Outsiders Imr δ δ Year FEsIndustry FEsPseudo R N yes yes yes yes 1,221 yes yes 920 yes yes 1,932 yes yes 1,941 3,873 yes yes 1,221 yes yes 920 yes yes 1,932 yes yes 1,941 yes 3,873 yes subsamples for CEOs older thancolumn or represents a equal subsample to of 56results certain (the are quantile sample of qualitatively median similar age) andat using thus 10%, GLM are estimator 5% not over and reported. the 1%, period respectively. from 1996 to 2017. Each This table reports the empirical results of regressing the forecast bias measure ( Table 12.A: Regressions of CHAPTER 3: DELEGATION, INFORMATION AND MARKET 302 ) in δ 75%) , 50% (25% ≥ AbsError F 50% < Aggregate ) on responsibility sharing measure ( 75% ≥ AbsError 25% F are in the parentheses and ***, **, * denote levels of ≤ 75%) , t-statistics as indicated at the top of each column. The unit of the analysis is firm-year. 50% (25% δ ≥ Quantiles δ AbsError on 50% F < Average AbsError 75% ≥ F 25% (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) 0.013 -0.008 0.003 -0.002 -0.000 0.012 -0.011 0.007 -0.012* -0.002 0.026 0.638** 0.093 0.536** 0.322** -0.104 0.855** 0.2970.295 0.127 0.297 0.686*** 0.313 0.342 0.208 0.285 0.285 0.334 0.273 0.313 (2.36)(1.46) (-2.62) (-1.07) (2.29) (-1.93) (0.37) (-0.34) (0.59) (-0.03) (1.78) (1.17) (-2.07) (-1.47) (1.92) (0.97) (-1.69) (-1.80) (-0.29) (-0.25) (6.75)(0.11) (5.27) (2.07) (7.09) (0.49) (5.99) (2.53) (8.87)(2.38) (2.18) (5.22) (3.58) (-0.45) (4.08) (3.04) (2.43) (5.72) (4.59) (1.45) (3.66) (5.76) (0.59) (4.14) (-4.00) (2.79) (1.93) (0.54) (-0.41) (0.84) -0.026 -0.075***-0.329 -0.030** -0.016 0.382 -0.487** -0.018 -0.129 -0.348** -0.015 -0.027 -0.069 -0.005 -0.143 -0.587*** 0.009 -0.177 0.009 -0.554** (-2.83) (-2.03)(-1.41) (-4.04) (-2.89)(-1.29) (-1.74) (-1.99) (1.18) (-0.86) (-3.76) (-2.25)(-3.50) (-1.41) (-0.52) (-0.09) (-4.05) (-1.99) (-0.72) (-1.66) (-0.72) (0.77) (-2.49) (-0.91) (-0.21) (-1.21) (0.62) (-0.28) (-0.38) (0.48) (0.16) (-2.59) (-1.12) (-0.63) (0.47) (-0.16) (-2.27) (-0.04) (1.74) (2.17) ≤ 2.732** -2.113*** 1.733** -1.072* 0.181 2.270* -1.983** 1.550* -1.190* -0.392 1.020** 2.463*** 1.168*** 1.699*** 1.557*** -0.735*** 1.599* 0.215 -0.059 0.400 0.250*** 0.296*** 0.221*** 0.209*** 0.218*** 0.182*** 0.276*** 0.199*** 0.104*** 0.170*** -2.279*** -2.030** -2.527*** -1.326* -1.924*** -2.655*** -1.903*-9.789*** -1.775** -0.327 -0.894 -1.711 0.143 1.909 1.112 -3.425 -0.649 -0.102 5.372* 6.385** 2 D & b M/B Size δ Leverage R Directors Outsiders Imr δ Year FEsIndustry FEsPseudo R N yes yes yes yes 1,477 yes yes 1,049 yes yes 2,343 2,257 yes yes 4,600 yes yes 1,477 yes yes 1,049 yes 2,343 yes 2,257 yes yes 4,600 yes yes significance at 10%, 5% and 1%, respectively. OLS results are qualitatively similar and thus are not reported. subsamples for CEOs older thancolumn or represents equal a to subsample 56 (the of sample certain median quantile age) of using GLM estimator over the period from 1996 to 2017. Each This table reports the empirical results of regressing the forecast bias measure ( Table 12.B: Regressions of