tan et al - does stronger corporate governance improve financial reporting quality

This empirical strategy allows us to overcome the endogeneity problem of governance mechanisms and provide evidence on the causal effect of corporate governance on financial reporting qu

Does Stronger Corporate Governance Improve Financial Reporting Quality? Evidence from a Regression Discontinuity Analysis of Shareholder-Sponsored

Governance Proposals

Liang Tan

George Washington University

Ph: 202-994-7481 liangtan@gwu.edu

Yanfeng Xue

George Washington University

Ph: 202-994-6747 yxue@gwu.edu

Yong Yu

University of Texas at Austin

Ph: 512-471-6714 Yong.Yu@mccombs.utexas.edu

Abstract

This study examines whether stronger corporate governance leads to higher quality financial reporting We use a regression discontinuity method to analyze the effect on financial reporting quality of shareholder-sponsored governance proposals that pass or fail by a small margin of votes in annual meetings This empirical strategy allows us to overcome the endogeneity problem of governance mechanisms and provide evidence on the causal effect of corporate governance on financial reporting quality We find that passing proposals related to board of directors and executive compensation leads to an improvement in the quality of financial reporting We also find that this positive effect on financial reporting is more pronounced for firms with lower quality financial reporting prior to the voting We find no evidence that passing proposals related to antitakeover provisions affects financial reporting quality

Preliminary draft - Please do not cite or quote Comments welcome

et al 2010; Brickley and Zimmerman 2010) As Brickley and Zimmerman (2010) conclude,

"Upon more careful reflection, however, most of these results are also consistent with alternative explanations Moreover, causality often is impossible to infer." The difficulty of establishing a causal relation is further exacerbated by the lack of comprehensive theories that allow researchers to classify a specific governance feature as "good" or "bad."

The objective of this study is to provide evidence on the causal effect of corporate governance on financial reporting by applying a regression discontinuity method to analyze a threshold event - shareholder proposals related to corporate governance Our empirical strategy focuses on examining the effect on financial reporting quality of shareholder-sponsored

governance proposals that pass or fail by a small margin of votes in annual meetings (i.e.,

close-call proposals) For such close-close-call proposals, the treatment firms (e.g., those with a proposal that just passes with 50.1% of the votes) are, on average, the same as the control firms (e.g., those with a proposal that just fails to pass with 49.9% of the votes) on virtually every dimension (Lee, 2008) However, a proposal is significantly more likely to be implemented and thus lead to changes in corporate governance for the treatment firms than for the control firms (Ertimur et al 2010) Thus, for the close-call proposals, passing a proposal is a "locally" exogenous event The estimated effects of passing a proposal on financial reporting quality should not be affected by any observed or unobserved confounding factors, as long as their effects are continuous around

the majority threshold (which we show is indeed the case for our setting) Therefore, our empirical strategy allows us to overcome the endogeneity problem of governance mechanisms and provide clear evidence of a causal effect of corporate governance on financial reporting

Our research design also allows us to overcome the limitation of having to classify a governance feature as "good" or "bad" arbitrarily To the extent that shareholders have better information and abilities to detect the weakness of a firm' governance structures, they are likely

to request changes that represent improvements in the firm's governance Consistent with the shareholder proposals improving governance, Cunat et al (2012) find that passing governance-related shareholder proposal triggers significant positive market reactions

We use the regression discontinuity method to analyze a large sample of 3,775 governance-related shareholder proposals from 1997 to 2010 We gauge financial reporting quality using both the accrual-based measure from the modified Dechow and Dechiv (2002) model and the measures based on accounting restatements and internal control deficiencies We find that the passing of governance-related shareholder proposals is related to an improvement in the accrual-based measure of reporting quality and a reduction in the likelihood of having internal control deficiencies, suggesting that stronger corporate governance results in higher quality financial reporting

Next, we examine which types of governance mechanisms drive the positive effect on financial reporting quality We classify all shareholder-initiated governance proposals into three groups that are related to executive compensation, boards of directors, and antitakeover provisions (included in the G-index constructed by Gompers et al 2003), respectively.1 We find strong evidence that passing proposals related to board of directors improves the accrual-based

1 We also attempt to identify proposals directly related to auditing However, there are too few proposals related to auditing issues, making it difficult to estimate the impact of these proposals precisely using our research design

measure of reporting quality, and reduces the likelihood of having accounting restatements and internal control deficiencies We also find some evidence that passing proposals related to executive compensation improves the accrual-based measure of reporting quality However, we find no evidence that passing proposals related to antitakeover provisions affects financial reporting quality

Further, we conduct cross-sectional analyses to examine whether the effects of passing governance proposals on financial reporting depend on the sponsorship of the proposals and the quality of financial reporting prior to the voting Prior research shows that shareholder governance proposals are more likely to be implemented when they are sponsored by institutions rather than individuals (Gillian and Starks 2007; Ertimur et al 2010) Further, improvements in governance mechanisms are likely to lead to a larger increase in financial reporting quality for firms with relatively poorer financial reporting prior to the voting We find some evidence that passing proposals related to executive compensation and boards of directors leads to a larger increase in financial reporting quality for firms with poorer reporting quality prior to the voting

We find no evidence that the effects of passing shareholder proposals on reporting quality vary with institution sponsorship of the proposals Furthermore, we find no evidence in any subsample that passing proposals related to antitakeover provisions affects financial reporting quality Overall, our results suggest that stronger corporate governance leads to an improvement

in financial reporting quality, and this positive effect is concentrated in governance mechanisms related to board of directors and, to a lesser extent, executive compensation

Our paper makes two primary contributions First, it contributes to our understanding of the relation between corporate governance and financial reporting by providing clear evidence on the causal effect of corporate governance on financial reporting Using the regression

discontinuity design and the threshold event of shareholder-sponsored governance proposals, we are able to overcome the problem of the joint endogeneity of governance and accounting system

as well as the difficulty of lacking theories to classify a governance feature as good or bad Our results suggest that stronger governance mechanisms, particularly those related to board of directors and executive compensation, lead to improved financial reporting Second, our study also adds to our understanding of the economic consequences of shareholder-initiated governance proposals Cunat et al (2012) examine market reactions to governance proposals and find that passing governance proposals triggers significant positive market reactions, reduces acquisitions and capital expenditures, and improves long-term firm performance We extend Cunat et al (2012) by demonstrating that the governance-related shareholder proposals have an important, and previously unidentified, benefit of improving firms' financial reporting

Our results are also relevant to the standing debate on shareholder-sponsored governance proposals While proponents argue that such proposals can improve governance and enhance firm value, critics raise the concern that shareholders who initiate the proposals can be self-serving and the shareholder voting could be uninformed and disruptive to firm operation (Bainbridge 2006) This debate is especially heated by the recent enactment of the Dodd-Frank Act, which requires firms to put their executive compensation policy under shareholder advisory vote at least once every three years Our study shed light on the economic consequence of shareholder-sponsored governance proposals by providing evidence on the role of those proposals in improving firms' financial reporting quality

The rest of the paper proceeds as follows Section 2 reviews prior literature and discusses institutional background Section 3 describes the regression discontinuity research method, and

Section 4 describes our sample and empirical measures Section 5 presents the empirical findings Second 6 concludes

2 Prior research and institutional background

2.1 Prior research

Although a large body of research has examined whether financial reporting quality is influenced by various governance mechanisms, the empirical evidence is generally mixed (see Armstrong et al (2010) for a comprehensive review of this literature) For example, research examining how managerial compensation influences financial reporting fails to find consistent results On one hand, some studies find a negative relation between managerial equity incentives and the quality of financial reporting (e.g., Cheng and Warfield 2005; Bergstresser and Philippon 2006; Efendi et al 2007); on the other hand, other studies find no relation between the two (e.g., Erickson et al 2006; Baber et al 2007) or even a positive relation (e.g., Warfield et al 1995; Armstrong et al 2010) The results from the stream of research examining the effects of various features of boards of directors on financial reporting are also mixed For example, while some studies find that more independent boards are associated with higher quality reporting (e.g., Klein 2002; Efendi et al 2007), others find little evidence that board independence has a significant impact on financial reporting quality (e.g., Agarawal and Chadha 2005; Larcker et al 2007) The evidence from examining anti-takeover provisions (G-index) is also inconclusive For example, Larcker et al (2007) find no relation between G-index and measures of reporting quality, whereas Baber et al (2012) document a negative relation between G-index and restatement in part of their sample period and no relation in the rest of the sample period

While many factors may contribute to the mixed findings in the extant literature, the joint endogeneity of corporate governance and financial reporting represents perhaps the biggest challenge (Armstrong et al 2010; Brickley and Zimmerman 2010) A firm’s choice of a particular governance mechanism is endogenously determined by many factors including firm fundamental characteristics, information environments, other governance mechanisms, and managerial attributes And many of those factors are also likely to be important determinants of the firm's financial reporting This endogeneity problem raises concerns that any association found between corporate governance and financial reporting can be driven by observable or unobservable confounding factors Recent reviews of the literature (Armstrong et al 2010; Brickley and Zimmerman 2010) conclude that the extant literature has provided very little evidence on the causal relation between corporate governance and financial reporting and call for more research that better addresses the endogneity problem and provides clearer causal evidence

In addition to this endogeneity problem, the difficulty in providing causal evidence between corporate governance and financial reporting is further exacerbated by the lack of comprehensive theories that allow researchers to classify a feature of corporate governance as

"good" or "bad." (Brickley and Zimmerman 2010) A particular governance feature (e.g., more outside directors or smaller board size) is unlikely to be optimal for all firms Rather, whether a specific governance feature is good or bad for a particular firm likely depends on the firm's fundamental characteristics, information environments, other governance mechanisms, shareholder structure, etc As such, the same governance feature may be good for some firms, but bad for others This difficulty forces researchers to rely on some ad hoc classification (e.g., having more outside directors on boards is assumed to be a good governance feature for all

firms), which can lead to mixed findings, depending on the sample of firms examined, and also make it difficult to interpret the findings

In this study we examine the causal effect of corporate governance on financial reporting

by applying a regression discontinuity method to analyze a threshold event - shareholder proposals related to corporate governance One key advantage of our empirical strategy is to overcome the endogeneity problem of governance and provide clear evidence of a causal effect

of corporate governance on financial reporting In addition, our research design also overcomes the limitation of having to classify a governance feature as "good" or "bad" arbitrarily

2.2 Institutional background

According to Rule 14a-8 of the Securities Exchange Act of 1934, shareholders who continuously own more than $2,000 or 1% of a company for at least one year can initiate a proposal The shareholder proposal must be placed on the agenda of the next annual meeting and included in the proxy distributed to the shareholders.2 Shareholder proposals have been used to address all kinds of issues including removing anti-takeover provisions, monitoring management compensation, strengthening shareholder voting rights, promoting corporate social or environmental responsibilities, and advocating corporate charitable contributions Because we are interested in how corporate governance affects financial reporting, we focus on the shareholder proposals that are directly related to firms’ corporate governance mechanisms, which comprise the vast majority of all shareholder proposals

Although shareholder proposals are advisory in nature (i.e, the voting results are not binding and the implementation of the proposals are subject to the boards’ and managers’

2 One exception is that if the proposal violates certain conditions, such state law, the company may ask the SEC to approve their exclusion of the proposal from the meeting agenda.

discretion), the voting results on the proposals have been demonstrated to have significant influences on firm policies, especially when the proposals receive majority support from shareholders Ertimur et al (2010) show that among S&P 1500 firms, 40% of proposals receiving majority votes were implemented in 2004, and the proposals receiving majority votes are significantly more likely to be implemented than the ones without majority support Ferri and Sandino (2009) find that firms that are targeted by shareholder proposals in support of expensing employee stock options are more likely to adopt the accounting practice and this likelihood increases with the voting support for the proposals Ertimur et al (2011) document that the likelihood of implementing executive compensation-related proposals increases substantially after receiving majority votes

The shareholder proposals offer a threshold event that allows us to address the endogneity of governance mechanisms and provide causal evidence on the effect of corporate governance on financial reporting Specifically, we apply a regression discontinuity (RD) method to examine changes in firm’s financial reporting quality in response to proposals that pass or fail the shareholder votes by a small margin The main idea behind the design is that the governance rules adopted can be considered random in those closed-call firms whose proposals pass or fail by a very small margin (Lee 2008) In other words, firms whose proposal passes by a small margin are essentially the same as firms whose proposal fails by a small margin in all dimensions, except for the significant difference in the likelihood of implementing the governance changes requested by the proposal Thus, a positive association between the voting results and changes in financial reporting quality in the subsequent periods should not be affected

by any observed or unobserved confounding factors, as long as their effects are continuous around the majority threshold (which we show is indeed the case) Therefore, this positive

association can be interpreted as the causal evidence that improvements in corporate governance lead to higher financial reporting quality

Examining shareholder proposals also allows us to overcome the difficulty in classifying

a governance feature as good or bad Instead of classifying a change in a governance mechanism

as an improvement or a deterioration arbitrarily, we assume that shareholders are in better position (than researchers) to access a firm's governance structure and they make value-maximizing decisions in their initiation of and voting on governance proposals, and thus the changes required by the shareholder proposals represent improvements in governance on average This assumption is supported by recent evidence Cunat et al (2012) show that passing governance-related proposal leads to significant positive market reactions, suggesting that shareholder proposals represent governance improvements that enhance firm value

We first examine all governance-related proposals to provide evidence on the average effect of corporate governance on financial reporting We then examine the effect of specific categories of governance mechanisms Because there are only limited numbers of shareholder proposals related to each specific governance mechanism, we follow Cunat (2011) to group all governance-related proposals into different categories: 1) proposals related to executive compensation (e g., putting a cap on executive bonus); (2) proposals related to board of directors (e.g., requiring majority vote for director elections); and (3) proposals related to anti-takeover provisions included in the G-index developed by Gompers, Ishii, and Metrick (2003) (e.g., proposals to eliminate classified board or golden parachute)

3 Identification Strategy – A Regression Discontinuity Design

We use a regression discontinuity (RD) design to examine changes in firms’ financial reporting quality in response to governance-related shareholder proposals that pass or fail by a small margin (i.e., closed-call proposals).3 A RD design is a quasi-experimental data design where treatment assignment depends upon an observed assignment variable around certain thresholds (Hahn, Todd, and van der Klaauw, 2001; Lee and Lemieux 2010) Lee (2008)

demonstrates that as long as the treatment assignment process cannot be precisely manipulated,

all outcome related factors other than the treatment will evolve continuously with respect to the assignment and variations in the treatment will be locally randomized around the cutoff The treatment effect can, therefore, be causally identified by controlling for a smooth function of the assignment variable and focusing on discontinuous changes in outcomes around the cutoff In other words, from a regression perspective, even if there exist omitted correlated variables, as long as they do not exhibit precisely the same discontinuity as the treatment does, the treatment effect can be consistently estimated Such a mild identification condition makes the RD analysis more credible than typical “natural experiment” strategies, such as difference-in-differences or instrumental variables (Lee and Lemieux, 2010)

In the context of governance-related shareholder proposals, the treatment is the impact of the proposed change in a firm’s governance structure and the assignment variable is the vote shares for the proposals Two important features in a shareholder proposal are critical to our identification First, passing a proposal results in a discrete jump in the probability that this

3 The RD design was first invented in the psychology literature by Thistlethwaite and Campbell (1960) to estimate the impact of merit awards on future academic outcomes It has gained great popularity in economics and finance research since late 1990s to address endogeneity because its relative mild assumption for causal inference compared

to other non-experimental approaches Lee and Lemieux (2010) provide an excellent introduction and user guide to this design Related to our study, in financial reporting literature, Tan (2013) adopts the RD design in the covenant violation setting and studies creditors’ demand for conservatism reporting In shareholder proposal literature, two papers have adopted the RD design Cunat et al (2012) use the RD design to examine market reaction on shareholder proposals that pass or fail by a small margin Ertimur et al (2013) focus on a specific type of shareholder proposals – adopting majority voting for directors and analyze market reactions to such change in the director election system

proposal will be implemented Ertimur et al (2010) document that 31.1% of passed proposals were implemented, while only 3.2% of failed proposals were implemented and most of the jump occurs around the passing cutoff (20.7%) Ertimur et al (2011) show that the likelihood of implementing executive compensation-related proposals increases substantially after the proposals receive majority votes This discrete jump in the probability of implementation ensures that our quasi-experiment gets a legitimate treatment at the majority threshold.4

Second, no discontinuity should be observed on vote share distribution around the majority threshold Recall that the critical identification assumption for a RD design is that the

treatment assignment process is not precisely manipulated A precise manipulation will result in

a discontinuous distribution of the assignment variable around the cutoff First of all, in practice, the vote outcome of a shareholder proposal is not revealed until the meeting date This mitigates the possibility that someone has superior information about the likely voting outcome and influences the vote to change the result of a close-call proposal Second, using a shareholder proposal sample similar to ours, Cunat at al (2012) do not find any discontinuity in distribution around the threshold5

The standard RD regression model applying to our shareholder proposals setting is the following:

Financial Reporting Quality i,t+1 = α 0 + α 1 Pass i,t + P l (X it , β l

where Financial Reporting Quality it+1 is one of our three proxies of financial reporting quality

for a firm i in year t+1 after the proposal meeting year t Pass it is an indicator variable equal to 1

if the shareholder proposal is passed in year t, 0 otherwise X it is the vote share for the proposal P(X it ,β) is a polynomial in vote shares and represents the smooth function of the assignment

variables We allow the coefficients to differ on either side of the voting cutoff, indicated by superscript/subscript l and r for the polynomial functions Similar to Cunat et al (2012), we present our results using a polynomial of order four, though our findings are robust to higher-order polynomials We correct the standard errors using two-way clustering by firm and by year (Petersen, 2009; Gow, Ormazabal and Taylor, 2010)

One complication in the application of the standard RD regression model (i.e model (1))

to analyze shareholder proposals is that shareholders of a firm may vote on more than one proposal on a particular meeting date That is, we could have one outcome but multiple

treatments Theoretically, we can create and estimate a Pass dummy and a set of polynomials in

vote shares for each individual proposal However, there are 72 distinct types of related shareholder proposals according to Risk Metrics’ classification The limited number of observations for each type prevents us precisely estimating the governance effect for every proposal, and requires the aggregation of shareholder votes

governance-Let’s assume there are J types of proposals voted in a shareholder meeting Following

Cunat et al (2012) and assuming for all J, α J =α, P J l = P l and P J r = P r, then we can rewrite model (1) as

Financial Reporting Quality i,t+1 = α∑ N

j=1 Pass j i,t + P l (∑ N

j=1 X j it , β j,l

) + P r (∑ N

j=1 X j it , β j,r ) + ε it (2)

α thus captures the average causal effect of passing a proposal on financial reporting quality We

first estimate the average causal effect among all proposals using model (2) Then, we relax the

restriction and let α vary by three major types of proposals (i.e., proposals related to boards of

directors, executive Compensation, and anti-takeover provisions (G-index), respectively) and estimate their individual effects jointly using the following regression model:

Financial Reporting Quality i,t+1 = α 0 + α 1 Pass G i,t + α 2 Pass B i,t + α 3 Pass C i,t + P G l (X G it , β G,l

quality, we expected the three coefficients α 1, α 2 and α 3 to be negative, consistent with the notion

that passing of governance-related shareholder proposals leads to higher reporting quality

4 Data and Variable Measurement

4.1 Data

We obtain shareholder proposals data from Risk Metrics’ shareholder proposals database, which contains shareholder proposals from all S&P 1500 companies plus an additional 500 widely held firms It provides information such as company names, meeting dates, a description

of the resolution and categorization of the proposal type, the voting percentage in favor of the proposal, an indicator variable showing whether the proposal has received a majority vote and the name of shareholder proponents We keep only proposals classified as governance-related by the database for nonfinancial firms with available meeting dates and vote share data Our sample

contains 3,775 governance-related shareholder proposals from 1997 to 2010 Further, we obtain accounting data from Compustat and stock price data from CRSP We obtain SOX 404 internal control weakness data from the Audit Analytics database, and the accounting restatement data from the Audit Analytics database and the U.S government accountability office (GAO) database.6

Figure 1 draws the distribution of vote shares The figure plots both histograms and kernel density estimation of the distribution Close-call proposals are not uncommon In fact, they are more frequent than proposals that are supported overwhelmingly For example, the number of proposals with supporting vote shares between 40% and 60% (26.68%) are 7.5 times greater than that with supporting vote shares between 80% and 100% (3.58%) We also notice that vote shares are generally distributed smoothly There is no noticeable kink around the 50% threshold (see formal statistical tests in Section 5.1)

Table 1 reports summary statistics of the sample Panel A shows summary statistics at the proposal level On average, proposals receive around 37% of the supporting votes and are passed 29% of the time G-index, Board and Executive Compensation related proposals are the three major types of proposals in the sample 38.7% of the sample is G-index proposals and they receive the highest shareholder approval (56.36%) than any other types of proposals These observations are consistent with Cunat et al (2010) There are only 57 proposals that are related

to Audit and 343 other types of proposals Audit and Other categories are approved in only 5%

of the cases Given the low approval rate for these two types, there are very few observations

6 Unlike Audit Analytics, the GAO data only provide the date when the restatements were issued but not the restatement periods We examine restatement data from Audit Analytics and find that the median number of days between the start and end dates of the restatement periods are 455 days and the median number of days between the end dates of restatement period and the restatement dates are 167 days We apply these statistics to estimate the restatement periods for restatements in the GAO database That is, we assume that the restatement period is from

622 days (455+167) to 167 days prior to the restatement date

around the majority threshold, which prevents us from estimating their individual effects precisely As a result, we exclude Audit and Other categories from our analyses

Panel B of Table 1 presents summary statistics at the firm-year level for key variables used in this study The number of observations varies according to the availability of the data On average, 44.6% of the sample firms have proposals sponsored by institutional investors The average number of restatements happening in the three year window after the proposal voting year is 0.117 and average incidences of internal control weakness in the same window is 0.082, consistent with statistics documented in prior studies

4.2 Financial Reporting Metrics

We use three measures to proxy for financial reporting quality Our first measure of financial reporting quality is the accrual-based Dechow and Dichev (2002) model modified by

McNichols (2002) (MDD) Following Kothari et al (2005), we also control for performance in

the model

ΔWC i,t = α +β 1 CFO i,t-1 + β 2 CFO i,t + β 3 CFO i,t+1 + β 4 ΔREV i,t + β 5 PPE i,t + β 6 ROA i,t-1 +ε (4)

where ΔWC i,t is the change in working capital from year t-1 to year t and calculated as ΔAR + ΔInventory - ΔAP - ΔTP +ΔOther Assets (net), where AR is accounts receivable, AP is accounts payable, and TP is taxes payable CFO’s are operating cash flows from fiscal year t-1, t, and t+1 ΔREV i,t is the change of sales from year t-1 to t All variables are scaled by average total assets of year t-1 and t ROA i,t-1 is the net income before extraordinary items in year t divided by average total assets of year t-1 and t

Dechow and Dichev (2002) estimate firm-specific time-series regressions and use standard deviation of the residuals over eight years as measures of accrual quality Our research design makes it hard to estimate accrual quality over a time window as long as eight years We therefore estimate model (1) cross-sectionally for each industry (two-digit SIC code) and year, and use the absolute value of the residuals as our measure of reporting quality Jones et al (2008) shows that the residuals from the cross-sectional regressions based on the Dechow and Dichev (2002) model modified by McNichols (2002) is the best at predicting accounting fraud compared

to other measures of accrual quality

Our second measure of financial reporting quality is number of internal control

weaknesses (# of ICW) reported under SOX 404 within three years after the proposal meeting

year Sarbanes-Oxley Act Section 404 requires management and auditors to disclose any material internal control weaknesses over financial reporting if detected Finally, we use number

of restatement (# of Restate) within three years after the proposal meeting year as our third

measure of financial reporting quality

5 Empirical results

5.1 Density Discontinuity and Pre Proposal Covariates

Before formally testing our hypotheses, we first examine the validity of our RD setting

As discussed before, the local randomization assumption is valid as long as the economic agents

cannot precisely manipulate the assignment variable, i.e., vote shares for the proposals A direct

test of it is to examine the density of vote shares If voting results are not manipulated, then the aggregate distribution of vote shares should be continuous We use the two-step procedure described in McCrary (2008) to formally test this assumption In the first step, we partitioned the