Corporate Governance and Costs of Equity: Theory and Evidence

Transcription

1 Corporate Governance and Costs of Equity: Theory and Evidence Di Li Erica X. N. Li October 8, 2013 Abstract We propose an alternative explanation for the existence during 1990s and disappearance during 2000s of the governance-return relation. Using a real options model with managerial agency problem, we show that corporate governance mitigates investment distortions, either over- or under-investment, so that well governed firms have more valuable investment options, present mainly during booms, and divestiture options, present mainly during busts, than poorly governed firms. Since investment options are riskier and divestiture options are less risky than assets-in-place, the mitigation of investment distortions generates the procyclical governance-return relation. Using governance index (G-index) and entrenchment index (E-index) as the measures of governance quality, we show that during the period of , well governed firms earn significantly higher returns during booms and lower returns during busts than poorly governed firms, controlling for various risk factors and firm characteristics. The results are robust to both aggregate and industry-specific business cycle classifications, to industry-median adjusted returns, and to product market competition as alternative measure of governance. JEL Classification: G1 G32 D92 E32 Keywords: Corporate governance, managerial agency problem, stock returns, investment, business cycles This is a preliminary draft. Please do not cite or circulate without consent. The previous version of this paper is circulated under the title Does corporate governance affect the cost of equity capital?. We thank Mike Barclay, John Long, Wei Yang, and especially Lu Zhang for their invaluable guidance. Comments and suggestions from Greg Bauer, Frederick Bereskin, Yangchun Chu, Fangjian Fu, Laura Liu, Jerold Warner, Charles Wasley, Yaxuan Qi, Chen Lin and seminar participants at London School of Economics, Pennsylvania State University, Stanford University, University of Rochester, University of Washington, University of Michigan, University of California at San Diego, and the Wharton School at the University of Pennsylvania are also gratefully acknowledged. All errors are our own. George State University, J. Mack Robinson College of Business Georgia State University, 35 Broad Street, Suite 1201, Atlanta, GA dili@gsu.edu; Tel: (404) Cheung Kong Graduate School of Business, Beijing, China, ; xnli@ckgsb.edu.cn; Tel: (86) exit 3075.

2 1 Introduction Does corporate governance affect the costs of equity capital? Gompers, Ishii, and Metrick (2003) (GIM) show that firms with stronger corporate governance earn higher average returns from 1990 to Core, Guay, and Rusticus (2006), however, find that this positive relation between governance and returns is reversed from 2000 to Recent paper by Bebchuk, Cohen, and Wang (2013) concludes that the association between governance and return disappears for the sample from 2000 to They argue that the the disappearance is due to the over-valuation of poorly governed firms before 2001 and the subsequent correction after market participants fully recognized the negative effects of poor governance on firm value by the end of In this paper, we propose an alternative explanation for the aforementioned existence and disappearance of the governance-return relation. Using a real options model, we show that the effect of corporate governance on stock returns is procyclical. In particular, strong governance leads to higher stock returns during booms, but leads to lower stock returns during busts. Our empirical evidences are consistent with the model predictions. In the model, a manager is either an empire-builder or a shirker. In either case, the investment or divestiture decision is distorted: The former type tends to over-invest and is reluctant to disinvest; the latter type avoids any effortful decisions, either investment or divestiture. Stronger corporate governance makes any of these suboptimal investment behaviors costly to the manager. The stronger the governance, the less distorted investment decisions made by the manager. Therefore, the model predicts that, all else equal, a firm with stronger governance has more valuable investment and divestiture options and hence higher firm value. The paper s most important insight is that the effect of corporate governance on stock returns can be positive or negative, depending on economic fundamentals. Investment options allow a firm to expand when the profitability is sufficiently high; thus, they are call options and are riskier than the underlying assets. On the contrary, divestiture options are put options and are less risky than the underlying assets because they give a firm options to scale down production when profits are too low. Because a firm s value consists of assets-in-place, investment options, and divestiture options, its beta is a value-weighted average of the betas of the aforementioned three parts. Therefore, all else equal, higher value of investment options, relative to the total 1

3 firm value, leads to higher expected stock returns, but higher value of divestiture options leads to lower expected stock returns. As demonstrated in Figure 1, during boom periods when a firm s value consists mainly of investment options and assets-in-place, a well governed firm, in contrast to a poorly governed firm, has more valuable investment options and consequently higher expected returns. On the contrary, during busts when a firm s value consists mainly of divestiture options and assets-in-place, a well governed firm has more valuable divestiture options and lower expected returns. This intuition leads to the paper s main hypothesis: corporate governance affects stock returns positively during booms and negatively during busts. In other words, the effect of corporate governance on the costs of equity capital is procyclical. To verify the model s predictions, we conduct a series of empirical tests. The key of the tests are the measure of corporate governance strength and the classification of business cycles. The two most used measures in the studies of the governance-return relation are governance index (G-index) and entrenchment index (E-index). We follow the literature and use these two as our main measures of governance strength. For robustness, we also repeat the tests using product market competitiveness, measured by Herfindahl-Hirschman index (Herfindah, 1950; Hirschman, 1945), as an alternative proxy for corporate governance. Classifying business cycles in our context means identifying periods with high investment opportunities, i.e., boom periods, and periods with high divestiture opportunities, i.e., bust periods. Since the pioneer work by Tobin (1969), Tobin s average Q has been widely used as a proxy for investment opportunities. It can also be shown that the relative importance of investment and divestiture options to firm value is an increasing function of Tobin s average Q in our model. Therefore, we classify business cycles based on economy-wide Tobin s average Q. We rank all the quarters during by their Tobin s Q s. The top 20% quarters with highest Tobin s Q are labeled as boom, the bottom 20% as bust, and the rest as normal. This procedure gives us a classification of aggregate business cycles. Past literature (see, Harford, 2005; Hoberg and Phillips, 2010, for example) has found that investment opportunities faced by firms are not only affected by economy-wide conditions, but also by industry-specific shocks. Industry-specific business cycles, although correlated, do not exactly synchronize with each other. In extreme cases, boom for one industry can be bust for an- 2

4 other. 1 To measure investment opportunities more accurately, we also classify industry-specific business cycles based on industry-level Tobin s average Q. The governance-return relation along business cycles is examined using both the time-series portfolio approach and the cross-sectional characteristics approach for the period of , during which both G-index and E-index are available. The portfolio approach examines the returns of governance hedge portfolio based on G(E)-index, which longs the value-weighted portfolio of firms with G(E)-index smaller than or equal to 5(0) and shorts the value-weighted portfolio of firms with G(E)-index greater than or equal to 14(5). On average, return of governance hedge portfolio based on G-index is 0.71% per month (t-statistic = 3.22) during booms and 1.08% per month (t-statistic = 1.08) during busts, controlling for market, size, value, and momentum factors and with aggregate business cycles classification. The results remain similar quantitatively and qualitatively for governance portfolio based on E-index and for returns adjusted for industrial median. The statistically insignificant governance-return relation during busts might be due to the inaccurately classified booms/busts. To overcome this problem, we take the cross-sectional characteristics approach advocated by Brennan, Chordia, and Subrahmanyam (1998) and regress individual stock returns on governance indicator (one for strong governance and zero for weak governance) interacting with boom, normal, and bust indicators, respectively, controlling for firm characteristics. Here, boom (normal or bust) indicator can be based either on aggregate Tobin s Q or on industry-specific Tobin s Q because it is defined as firm level. To implement the characteristics approach, we use the clustered ordinary least squares assuming that variance of error terms is clustered in time (month). Consistent with the model predictions, we find that on average, a well governed firm outperforms (underperforms) an otherwise identical poorly governed firm during booms (busts). In general, the results are more pronounced and more statistically significant if we classify business cycles with finer industry definitions. With G-index as the governance measure and aggregate business cycles classification, a well governed company s stock earns a monthly premium (discount) of 26 (149) basis points (t-statistic = 0.67 (1.71)) over a poorly governed firm s stock during booms (busts), controlling for an extensive set of firm characteristics used in Bren- 1 The end of 1990s was a golden period for the IT industries but a difficult time for consumer non-durables industry. 3

5 nan, Chordia, and Subrahmanyam (1998). However, when business cycles are classified in Fama and French (1997) 48-industry level, the monthly premium (discount) of the strong governance stock becomes 69 (63) basis points (t-statistic = 2.03 (2.63)). The results are similar in magnitude as well as statistical significance when E-index is used as the governance measure and when returns are adjusted for industry median. With industry-specific business cycle classifications, we also conduct the characteristics regressions using Fama and MacBeth (1973) method, 2 and the results remain quantitatively similar. We take several approaches to check the robustness of our empirical findings. First, we utilize an alternative sample period and alternative cutoffs for the classification of business cycles. Specifically, we identify Tobin s Q cutoff points in a longer period from 1970 to 2012 and repeat our empirical analysis. The results are essentially the same. We then changes the cutoffs of boom/bust to top/bottom 30% and again obtain similar results to the benchmark ones. Second, following Giroud and Mueller (2010, 2011), we use the product market competition, proxied by the Herfindahl-Hirschman index (Herfindah, 1950; Hirschman, 1945) (HHI), as an alternative measure of corporate governance. Because HHI is available in the longer period of , we test our hypothesis in periods of and using HHI as governance measure and find supportive results in both sample periods. Finally, we address the possibility that the effective governance level is higher for firms with higher G(E)-index during bust periods because those firms usually face harsher economic conditions, which can serve as an external governance mechanism. We run cross-sectional regressions of Tobin s Q s of equity on governance indicator interacting with boom, normal, and bust indicators, controlling for firm characteristics. The results show that well governed firms (the ones with low G(E)-index) are valued higher than poorly governed firms (the ones with high G(E)-index) during both booms and busts. Therefore, the negative governance-return relation during busts in our sample is not driven by the reversal of effective level of governance during busts. Our paper belongs to the literature that studies the governance-return relation. We propose and test an alternative explanation for the positive governance-return relation in the 1990s and its subsequent disappearance. Gompers, Ishii, and Metrick (2003) and Cremers, Nair, and 2 With aggregate business cycle classification, there are only 2 bust quarters (8 bust months) and the results from Fama and MacBeth (1973) method have little power. 4

6 John (2009), among others, provide some explanations for the positive governance effect in the 1990s. Bebchuk, Cohen, and Wang (2013) documents the gradual recognition of the effects of governance on firm value before 2002 by market participants, which, they argue, connects the disappearance of the governance-return relation during The misplacing of poorly governed firms may have contributed to the positive governance-return relation found during the 1990s. However, our study shows that even after market participants fully recognize the effects of governance on firm value, one should still observe positive (negative) governance-return relation during boom (bust) periods. Therefore, this paper complements previous studies and provide an alternative explanation. This paper is related to Dow, Gorton, and Krishnamurthy (2005), who study the effect of governance on bond pricing and term structure, and Albuquerque and Wang (2008), who study the effect of country-level investor protection on equity risk premium and risk free rate. In diverging from those studies, this paper builds on the studies of managerial agency problems (Bertrand and Mullainathan, 2003; Hicks, 1935; Jensen, 1986, 1993) and focuses on the effect of within-country firm-level governance on the cross-sectional stock returns. Finally, the paper is related to the literature that studies the impact of corporate policies on cross-sectional stock returns (e.g., Berk, Green, and Naik, 1999; Carlson, Fisher, and Giammarino, 2004; Zhang, 2005) but with a focus on distorted investment policies. The remainder of this paper is organized as follows. Section 2 presents a simple real options model to provide intuition on the model s main hypothesis. Sections 3 and 4 explain the data sample and how to classify booms and busts. Section 5 presents empirical evidences on the hypothesis. Section 6 concludes. Appendices A, B and C give the proofs of the lemmas. 2 Model This section presents a real options model, following Dixit and Pindyck (1994), to illustrate the impact of corporate governance on investment policies, firm value, and expected stock returns. Assume that the Capital Asset Pricing Model (CAPM) holds in the economy and the market price of risk is a constant, defined as φ. Consider a firm with N units of capital and the cash flow y t 5

7 generated by each unit of capital at time t follows a Geometric Brownian motion dy t = πy t dt + σy t dz t, (1) where π is the constant drift, σ is the variance parameter, and dz t is the increment of a standard Wiener process. Given that the CAPM holds, the risk-adjusted discount factor for the cash flows generated by the assets-in-place is given by r a = r f + φσρ ym, where r f is the constant risk-free rate and ρ ym is the coefficient of correlation between cash flow y and the market portfolio, which is assumed to be a positive constant. 3 In addition to assets-in-place, the firm has a investment option to increase its cash flow to (N + 1)y by making a fixed amount of investment I and a divestiture option to sell one unit of installed capital at price I, reducing the cash flow of the firm to (N 1)y. 4 Therefore, the value of the firm, denoted as V, consists of the values of assets-in-place, investment option, and divestiture option, denoted as V a, V g, and V d, respectively. Appendix A shows that the expected return on firm value is a value-weighted average of return on assets-in-place, investment option, and divestiture option and can be written as [( ) Va r s = r f + φσρ ym + V ( Vg V ) β 1 + ( ) ] Vd β 2, (2) V where β 1 > 1 and β 2 < 0 are two constants. Investment option is riskier than assets-in-place, indicated by β 1 > 1, and divestiture option is less risky than assets-in-place, indicated by β 2 < 0. In fact, divestiture option earns a return lower than the risk-free rate. As illustrated in Carlson, Fisher, and Giammarino (2006), investment option is call option, whose value grows faster than assets-in-place as economic fundamentals gets better and drops to zero more quickly when conditions get worse. On the contrary, the value of divestiture option drops to zero when economic 3 Merton (1990) provides the proof. 4 The resale price of capital can be different from its purchasing price for reasons argued in Shleifer and Vishny (1992). For simplicity, we assume that the two prices are the same. Assuming different prices for purchasing and selling capital do not affect the qualitative results of the model. 6

8 fundamental is strong and goes up as it gets worse. Therefore, divestiture option serves as a hedge for adverse economic conditions and earns lower expected returns. Lemma 1 The expected return on firm value is positively related to the fraction of investment option value to total firm value, V g /V, and negatively related to the fraction of divestiture option value to total firm value, V d /V. Lemma 1 builds on two basic results in the model: (1) investment option is call option and is riskier than assets-in-place; (2) divestiture option is put option and is less risky than assetsin-place. The value of investment option is positively correlated with the underlying economic fundamentals and goes up faster than the value of assets-in-place when the fundamentals get better. Because the return on firm value is the value-weighted average of returns on assets-inplace, investment options, and divestiture options, all else equal, the expected return of the firm is higher if a larger fraction of the firm value comes from its investment option. On the contrary, the value of divestiture option is negatively correlated with fundamentals and has a negative beta. All else equal, the expected return of the firm is lower if a larger fraction of the firm value comes from its divestiture option. Assume that the manager decides on the investment policies of the firm but his (her) incentive is not perfectly aligned with that of outside shareholders. We consider two main agency problems discussed in the literature, empire-building (Albuquerque and Wang, 2008; Jensen, 1986, 1993) and shirking (Bertrand and Mullainathan, 2003; Hicks, 1935). We assume that for one unit of investment made by the firm, the manager gains, per share of his ownership, additional B units personal benefits net costs imposed by corporate governance. Stronger governance makes suboptimal investment decisions more costly to managers and, all else equal, leads to lower absolute value of B. A manager with a positive value of B is a empire-builder and the one with a negative value of B enjoys quiet life. A manager with B = 0 has incentive perfectly aligned with outside shareholders and there are no agency costs. To focus on the impact of suboptimal investment decisions on firms stock returns, the manager s private benefits are assumed to be non-pecuniary and do not come from no outright appropriation of the firm s cash flows. The reduction in the firm s value, therefore, comes entirely from the distorted investment policies. 7

9 The model also assumes that the manager and outside shareholders are subject to the same stochastic discount factor, which may not be true for several reasons as studied in Chen, Miao, and Wang (2010). This simplification hence ignores other agency problems such as compensation with undiversifiable risk, which is beyond the scope of this paper. The following lemmas are shown to hold under this simple setup. The proofs are relegated to Appendices A and B. Lemma 2 Tobin s average Q, defined as V NI, and values of investment and divestiture options decrease with the absolute value of B. Lemma 2 is quite intuitive. In the model, managers with empire building incentives tend to advance investments and delay divestitures, while managers who prefer quiet life tend to delay both investments and divestitures. Any deviation from optimal investment, either overor under-investment, reduces the values of investment/divestiture options and hence Tobin s average Q. Moreover, the larger the deviation, the lower the values of investment/divestiture options and Tobin s average Q. The absolute magnitude of B is positively related to the deviation from optimal investment, therefore, negatively related to Tobin s average Q and values of investment/divestiture options. Based on lemmas 1 and 2, we derive our main hypothesis below. Hypothesis 1 During periods when investment options are more important for firm value relative to divestiture options, all else equal, firms with strong governance have higher expected returns than firms with weak governance; during periods when divestiture options are more important, firms with weak governance have higher expected returns than firms with strong governance. The intuition behind the above hypothesis is as follows. As lemmas 2 and 1 indicate that stronger governance leads to higher value of investment option and hence higher expected return on firm value, all else equal. However, stronger governance also leads to higher value of divestiture option and hence lower expected return on firm value. The net effect of stronger governance on firm s stock return is positive when the value of investment option dominates that of divestiture option and vice versa. Because the value of investment (divestiture) option is positively (negatively) related to economic fundamentals, we hypothesize that all else equal, firms with stronger governance have higher expected returns than firms with weaker governance during booms, but lower expected returns during busts. 8

10 To summarize, the real options model implies that even though a firm with stronger governance always has a higher average Tobin s Q than a firm with weaker governance, its expected return is higher only when investment options are the important part of its firm value and lower when divestiture options are the important part. In the next section, we empirically test this hypothesis. 3 Classification of Business Condition To test our hypothesis, we need to find an empirical proxy to measure the relative importance of investment and divestiture options to firm value. The most widely used proxy in the literature for investment opportunities is Tobin s average Q. Tobin (1969) first proposes to use Tobin s average Q to measure firm s incentive to invest in capital. Abel (1983) shows that the optimal rate of investment depends on the marginal Q and Hayashi (1982) presents conditions under which average Q and marginal Q are equal. Given that marginal Q is not observable, average Q becomes widely used to measure firm s growth opportunities. In our model, the values of assets-in-place and investment options are positively related to and the value of divestiture options is negatively related to economic fundamentals. It can be shown that the ratio of investment to divestiture options is positively related to Tobin s average Q under the conditions that the risk premium of the firm is positive, or in another words, the value of the firm positively co-moves with economic fundamentals. Lemma 3 The ratio of investment and divestiture options V g /V d is an increasing function of Tobin s average Q as long as the firm has positive risk premium. The proof of Lemma 3 is provided in Appendix C. Since majority of the firms in our sample co-move positively with the market, we can use Tobin s average Q to proxy for the relative importance of investment and divestiture options, with high Tobin s average Q indicating that investment options are more important for firm value and low Tobin s average Q indicating that divestiture options are more important. To be concise, we omit the word average and use Tobin s Q to refer to Tobin s average Q hereafter. 9

11 For each quarter, we calculate the total market value of assets for all companies in the COM- PUSTAT/CRSP population at the quarter end. This market value of assets is particularly computed as the market valuation of stocks plus the value of debt obtained from the companies balance sheet. We then define the aggregate Tobin s Q for the COMPUSTAT/CRSP population as the ratio of the market value of assets to the total book value of assets of these companies in the quarter. We further classify each quarter into three categories of business condition boom, normal, and bust by comparing the value of Tobin s Q with its historical value: A quarter is labeled as boom (bust) if the Tobin s Q of this quarter is above (below) the 80th (20th) percentile of the ranked quarterly Q s during the period , and a quarter is labeled as normal if it is neither a boom nor bust. Figure 2(a) illustrates the dynamics of the aggregate Tobin s Q from 1990 to The graph shows that during the 1990s, Tobin s Q is relatively high, especially in the second half of the 1990s; Tobin s Q experiences a sharp drop at the beginning of the 2000s and remains at a low level in the first decade of the 2000s (relative to the 1990s), especially during the beginning and the end of the 2000s. It is well known that firm s investment decisions are affected not only by broad economic fundamentals but also by industry-specific technological/regulatory shocks (see Harford, 2005, for example). Hoberg and Phillips (2010) show that industry-specific business cycles, although correlated, do not perfectly synchronize with each other. Therefore, for a specific firm, the aggregate Tobin s Q might not be an accurate indicator for the amount of its investment opportunities. To better capture the investment opportunities faced by firms, we classify firms into 10 and 48 industries, respectively, based on Fama and French (1997) and compute Tobin s Q at the industry level. Specifically, we calculate the total market value of assets for all companies in an industry for each quarter as the market value of stocks plus the debt value from balance sheets of the companies in that quarter. Then we compute Tobin s Q of the industry by dividing the total market value of assets in that industry by the total book value of assets of the industry at the quarter end. Figures 2(b) and 2(c) plot the time series of industry-level Tobin s Q for the Fama and French 10 (FF10) industries and 48 (FF48) Industries. Consistent with previous evidences, the graphs 10

12 show that although Tobin s Q s of different industries exhibit similar cyclical movements, they do not reach peaks and troughs at the exactly same quarters. In some extreme cases, the periods that some industry experiences peak are the periods of trough for other industries. For example, the end of 1990s was a golden period for the IT industries while the Tobin s Q of consumer non-durables industry reached its bottom. To better understand the variations of aggregate and industry-level Tobin s Q s overtime, Table 2 reports the number and the percentages of quarters labeled as boom and bust within each year, respectively, between Columns under Pool report statistics on boom/bust classified based on the aggregate Tobin s Q while columns under FF10 and FF48 report the ones based on FF10 and FF48 industry classifications, respectively. We can see that based on the aggregate Tobin s Q, all the booms cluster within the period , during which 75% of the quarters are classified as boom and none is classified as bust. On the contrary, for the same period only 32.5% (25.35%) of the industry-quarters are booms and 7.5% (17.19%) are busts based on the FF10 (FF 48) industry classification. 5 Similarly, for the period of 2008 to 2012, 75% of the quarters are busts and none is boom based on the aggregate Tobin s Q, while 36% (23.02%) of the industry-quarters are busts and 8.5% (15.31%) are booms based on the FF10 (FF48) industry classification. Based on the above observation, we conduct our empirical tests using both the aggregate boom/bust classification and the industry-specific boom/bust classifications. 4 Data Our sample includes all the companies for which corporate governance information from Risk- Metrics is available. RiskMetrics publishes provisions of investor rights and takeover protection, initially based on survey results from the Investor Responsibility Research Center (IRRC) that was acquired by by ISS in IRRC/ISS has released eight volumes of such surveys (September 1990; July 1993; July 1995; February 1998; November 1999; February 2002; January 2004; and January 2006). The surveys cover large companies in S&P 500 and the largest corporations listed by Fortune, Forbes, and BusinessWeek, and starting 1998 the sample expanded to include 5 For the FF10 (FF48) industry classifications, there are 4 10 (4 48) industry-quarters within each year. 11

13 smaller firms and firms with high institutional ownership. In 2007 RiskMetricks changed the survey methodology to meet ISS specifications and some variables needed for the construction of governance indices were no long collected. Therefore our main sample ends in December 2007, the last month before the new version of survey was conducted. We use two indices to measure the effectiveness of corporate governance: governance index (G-index) following Gompers, Ishii, and Metrick (2003), and entrenchment index (E-index) introduced by Bebchuk, Cohen, and Ferrell (2009). Gompers, Ishii, and Metrick (2003) first introduce the governance index using twenty-four provisions published by IRRC on investor rights and takeover protection. This index is constructed by counting the number of such provisions that apply to the company. Bebchuk, Cohen, and Ferrell (2009) investigate the twenty-four provisions and identify six provisions that matter the most: staggered boards, limits to shareholder bylaw amendments, poison pills, golden parachutes, and supermajority requirements for mergers and charter amendments. They construct the E-index as the number of these six provisions that apply to the firm. We retrieve the G-index directly from RiskMetricks and construct the E-index following Bebchuk, Cohen, and Ferrell (2009). In Table 1 we report the summary statistics of governance index and entrenchment index. The average G-index is 9.02 and the median is 9. For E-index, its mean is 2.31 and median is 2. We also report the distributions of these indices for each published volume. There are small variations across releases but the distribution is overall very stable. Our summary statistics of G- index and E-index are close to those reported by Gompers, Ishii, and Metrick (2003) and Bebchuk, Cohen, and Ferrell (2009). Because RiskMetrics does not publish volumes for each year, we follow Gompers, Ishii, and Metrick (2003) and Bebchuk, Cohen, and Ferrell (2009) to assume that the corporate governance measures of the covered companies remain unchanged between two consecutive releases. To test our hypotheses, we match the sample with monthly stock return data from the Center for Research in Security Prices (CRSP) and financial data from COMPUSTAT in annual and quarterly frequencies. 12

14 5 Empirical Tests In this section, we first present summary statistics on the governance-return relation and then conduct rigorous empirical tests on Hypothesis?. We use two different approaches to evaluate the the governance-return relation along business cycles: the portfolio approach and the characteristics approach. The first approach forms governance hedge portfolio based on E(G)-index and measures the abnormal return of the governance portfolio along business cycles, controlling for market, size, value, and momentum factors. The second approach examines the effect of governance on individual firm returns along business cycles, controlling for various firm characteristics. 5.1 Summary Statistics To obtain a rough idea about the governance-return relation, we following Gompers, Ishii, and Metrick (2003) and construct democracy portfolios, which contains firms with G(E)-index smaller than or equal to 5(0) and dictatorship portfolios, governance hedge portfolios that contain firms with greater than or equal to 14(5). The governance hedge portfolio, referred as G(E)-portfolio, is then contracted by longing the G(E)-index based democracy portfolio and shorting the corresponding dictatorship portfolio. For each month, we calculate the value-weighted returns of G- and E-portfolios, the time series of which are plotted in Figures 3(a) and 3(b) (in solid lines), respectively. Consistent with our hypotheses, the figures show that during the second half of the 1990s when investment opportunities are ample, the governance hedge portfolios yield more positive returns than other periods; while in the beginning of the 2000s when it turns into busts for investment, they experience more negative returns than other periods. In the same figures, we also plot the series of returns of the governance hedge portfolios adjusted for the contemporaneous industry median (in dashed lines), with industries classified following Fama and French (1997) 48-industry definition. The pattern of the industry-adjusted return series is the same. The equal-weighted returns of the G- and E-portfolios are computed and plot them in Figures 3(c) and 3(d). They exhibit similar pattern as the series of returns to the governance hedge portfolio. The return differences between strong and weak governance stocks are more pronounced 13

15 at the end of the 1990s and at the beginning of the 2000s, more likely to be positive in the former period and negative in the latter period. The industry-adjusted average return differences between strong and weak governance stocks exhibit the same pattern. In Table 3, we summarize average returns of the governance hedge portfolios during boom/bust/normal periods, using both raw returns and returns adjusted by the corresponding Fama and French (1997) 48-industry median returns. Results based on the aggregate boom/bust classification are included in Panel A. We find that in the boom periods, the governance hedge portfolio yields positive returns on average, 86 (107) basis points per month for raw returns of the G(E)-portfolio and 149 (136) basis points per month for industry-adjusted returns of the G(E)-portfolio. The average value-weighted return of the G-portfolio is negative during busts, being 60 basis points (bps) per month for raw returns and 26 bps for industry-adjusted returns. However, the average return of the E-portfolio during busts is positive even though with a much smaller magnitude compared to the average return during booms, being 24 bps per month for raw returns and 61 bps for industry-adjusted returns. The equal-weighted returns of the governance hedge portfolios exhibit similar patterns as the value-weighted ones. Table 2 shows that there are only 2 bust quarters (8 bust months) during the period based on the aggregate business cycle classification, which gives low power to the statistics for the bust periods. In Panels B and C of Table 3, we report the average return differences between strong and weak governance stocks, based on FF10 and FF48 industry-specific business cycle classifications, respectively. Here the industry-adjusted returns in Panels B and C are based on 10- and 48-industry classifications correspondingly. Under FF10 industry classification (Panel B), the statistics show that strong governance stocks on average earn higher returns than weak governance stocks in the boom periods. With industry median adjustment, this difference is 54 basis points per month. For the bust periods, the average return difference between strong and weak governance is negative for both governance measures and for both raw and industry-adjusted returns, though the difference is smaller when E-index is used as the governance measure. Under FF48 industry classification, the average return difference between strong and weak governance stocks is positive during booms and the magnitude is large. For example, with G-index as the governance measure, the difference in industry-adjusted returns is on average 97 bps each month. 14

16 In the bust periods, the average return difference is in general negative. The study above does not consider the effects of other risks factors and firm s characteristics on returns. Next, we conduct rigorous statistical tests on the governance-return relation using both the portfolio approach and the characteristics approach. 5.2 Portfolio Approach In this subsection, we follow Gompers, Ishii, and Metrick (2003) and Bebchuk, Cohen, and Ferrell (2009) taking a portfolio approach to test whether the risk-adjusted abnormal return based on the governance trading strategy depends on business condition. Specifically, we employ the standard four-factor model, that is, Fama and French (1993) three-factor model augmented by the momentum factor of Carhart (1997). 6 Without the dependence of the risk-adjusted abnormal return on business condition, the standard four-factor model is estimated as R t = α + β 1 RMRF t + β 2 SMB t + β 3 HML t + β 4 UMD t + ɛ t, (3) where R t is the value-weighted excess return to the governance hedge portfolio in month t, RMRF t is value-weighted market excess return in month t, SMB t (small minus big), HML t (high minus low), and UMD t (up minus down) are Fama and French (1993) three factors plus the momentum factor in month t, measuring returns on zero-investment factor-mimicking portfolios constructed to capture size, book-to-market (B/M), and momentum effect, respectively. Gompers, Ishii, and Metrick (2003) use a sample up to 1999 and find that the governance hedge portfolio (based on G-index) earns a positive abnormal return. We use the standard fourfactor model in equation (3) and the subsample up to 1999 to confirm this finding. The result in reported in column 1 of Table 4. In Panel A, we run the test using raw returns, and in Panel B, we use returns adjusted for contemporaneous median and industries classified following Fama and French (1997) 48-industry definition. We find that indeed there is a positive abnormal return (statistically significant at 10%) associated with the governance hedge portfolio. The abnormal 6 We obtain the factors from Kenneth French s website. 15

17 return is also economically significant. The governance hedge portfolio earns an abnormal return of 51 basis point per month for the raw-return specification and 46 basis points per month for the industry-adjusted-return specification. Bebchuk, Cohen, and Ferrell (2009) measure governance using E-index based a subset of the provisions used by Gompers, Ishii, and Metrick (2003), and they also find that the governance hedge portfolio constructed using E-index earns positive abnormal returns. Similarly, we estimate the standard four-factor model (3) using a subsample up to 2003 that is used by Bebchuk, Cohen, and Ferrell (2009) and report the results in column 4 of Table 4. With E-index as the governance measure, we find that there is still a positive abnormal return associated with the governance hedge portfolio and it has a larger magnitude as well as more statistical significance. The finding of superior abnormal returns earned by strong governance stocks has been challenged in the recent literature (Bebchuk, Cohen, and Wang, 2013; Core, Guay, and Rusticus, 2006). Using a more updated sample, Bebchuk, Cohen, and Wang (2013) show that the superior performance of the strong governance no longer exits especially after 2001, which the authors argue to be attributed to the learning by investors of the mispricing. We run the standard four-factor model (3) with the whole sample from 1990 to The results are reported in columns 2 (for G-index) and 4 (for E-index) of Table 4. We find that using G-index as the governance measure, the abnormal return on the governance hedge portfolio becomes much smaller and is no longer statistically significant; using E-index as the governance measure, there is still a positive and statistically significant abnormal return associated with the governance hedge portfolio. These results are close to the findings of Bebchuk, Cohen, and Wang (2013). 7 Our hypothesis states that the strong governance stocks are riskier than weak governance stocks during boom periods while less risky during busts. Therefore, we hypothesize that the governance hedge portfolio earns higher (lower) risk-adjusted abnormal returns during boom (bust) periods. In other words, the abnormal return α varies in business condition. To test this 7 In Table 2 of Bebchuk, Cohen, and Wang (2013), they also find a positive and significant abnormal return on the governance hedge portfolio for 1990 to The magnitude (69 basis points) is close to our finding of 66 basis points. It is only in the subsample after 2001 do they find that strong governance stocks no longer outperform weak governance stocks. 16

18 hypothesis, we model α to depend on business condition: R t = α BM I BM t + α NM I NM t + α BT I BT t + β 1 RMRF t + β 2 SMB t + β 3 HML t + β 4 UMD t + ɛ t, (4) where I BM t, It NM, and It BT are binary variables indicating if month t is boom, normal, or bust business condition, respectively. For example, if month t is bust, I BT t bust, I BT t = 1; and if month t is not = 0. 8 Our hypotheses hence predict α BM > 0 and α BT < 0. Since we have specifically directional hypotheses, we conduct the statistical inference mainly using the p-value of one-sided tests. In particular, for α BM being significantly different from zero, we care about whether it is from the positive side instead of being from the negative side. We run this test with the whole sample from 1990 to 2007 and report the results in columns 3 (for G-index) and 6 (for E-index) of Table 4. Because of our specifically directional hypotheses, we report the p-values of the one-sided tests for α BM and α BT. In additional to the asterisks for significance levels of the two-sided test, we use the bold fond to highlight the estimates of α BM and α BT if they are statistically significant at 10 percent level for the one-sided tests. Using G-index as the governance measure, in the raw-return specification, we find that strong governance stocks outperform weak governance stocks during booms by 71 basis points per month and underperform weak governance stocks during busts by 1.08 percentage points monthly. The former is statistically significant at a one-sided p-value of 2.4 percentage points while the latter is not as quite significant with a one-sided p-value of 14.1 percentage points. From Table 2 we note that there are only two quarters in our sample period (September 1990 to December 2007) that are bust. The lack of enough bust months may undermine the power of the hypothesis test for α BT. 9 With returns adjusted for contemporaneous median (using Fama and French (1997) 48-industry definition), the estimated results are essentially the same. During booms, strong governance stocks on average outperform weak governance stocks by a positive 8 Another way to test our hypotheses is to run the standard four-factor model (3) in boom, normal, and bust periods separately. However, in our sample period (September 1990 to December 2007) there are only two quarters (eight months) that are bust. This makes the regression in the bust periods almost impractical. 9 One possibilility for the insignificance of α BT might be that during busts the agency conflict is not as severe as during booms so that the governance does not make much difference in returns. However, the estimate of α BT is large in magnitude, which does not seem to support this argument. 17

19 and statistically significant abnormal return (95 basis points per month), while during busts, weak governance stocks earn a superior abnormal return of 105 basis points per month though not statistically significant. The estimation results based on E-index as the governance measure are almost the same as those based on G-index. Strong governance stocks outperform weak governance stocks during boom periods by 119 basis points (96 basis points with industry adjustment) per month, and the difference is highly statistically significant. During bust periods, though not statistically significant, weak governance stocks on average outperform strong governance stocks by an abnormal return of large magnitude 1.15 percentage points (87 basis points with industry adjustment) per month. Furthermore, the results above exhibit one interesting pattern for abnormal returns of governance hedge portfolio in the business cycles abnormal returns of the governance hedge portfolio are procyclical, that is, they are decreasing in the order of boom, normal, and bust periods. This pattern echoes our theoretical predictions on the effect of governance on stock returns. In all, the results based on the governance hedge portfolio and factor models are not inconsistent with our argument on the relation between governance and cost of equity. During boom periods, strong governance increases the correlation between firms cash flow and business condition. Therefore strong governance stocks require higher returns during boom periods. During bust periods, strong governance reduces the association between cash flows and business condition so that strong governance stocks earn lower returns. When the sample period consists mostly of booms (i.e., Gompers, Ishii, and Metrick, 2003), it would seemingly suggest a positive superior return associated with strong governance stocks. When boom and bust periods are both included in the sample of study or if the investigation is not conditioned on business condition (i.e., Bebchuk, Cohen, and Wang, 2013), the two opposite effects partly offset each other and it would seemingly suggest no relation between governance and stock returns. 5.3 Characteristics Approach Despite its popularity in empirical asset pricing studies, the factor approach is subject to some caveats in our context. The portfolio approach can only be applied on the sample with boom/bust 18

20 classified based on aggregate business condition. With the economy-wide business cycle classification, the sample period covers only two bust quarters, which leads to low power for the test of the governance-return relation during busts. Moreover, we show in Section 3 that the timings of boom/bust periods for different industries exhibit variations in time. Therefore, it can imply imprecise classification of business condition faced by companies and cause failure in detecting the procyclical governance-return relation. In this section, we adopt the characteristics approach pioneered by Brennan, Chordia, and Subrahmanyam (1998) and conduct tests at the firm level. This approach is more suitable for tests based on industry-specific business cycles because business condition indicators can now be defined at the firm level. Moreover, including a large set of firm characteristics addresses the concern of omitted variables bias since Gompers, Ishii, and Metrick (2003) find that firm characteristics are correlated with the governance index. The basis model for the characteristics approach is as follows r it = a + bg it + cx it + e it, (5) where, for firm i in month t, r it is the stock return, G it is the governance measure variable, X it is a vector of stock characteristics, and e it is the error term. To test our hypothesis that the effect of governance on stock returns depends on business condition, we augment the basis model by allowing the effect of governance, b, to vary along business cycles. Particularly, we assume that b is a function of business condition, b = b BM I BM it + b NM I NM it are business condition indicators for stock i in month t. Indicator I BM it a boom period for stock i and zero otherwise. Indicators I NM it normal and bust periods, respectively. + b BT Iit BT, where I BM and I BT it it, Iit NM, and Iit BT equals one if month t is are defined similarly for Tests of asset pricing models with firm characteristics are usually conducted following Fama and MacBeth (1973) two-pass procedure. In the first stage, equation (5) is estimated crosssectionally each month; in the second stage, the average of the coefficients and their standard errors are obtained using the estimated time series (of boom, normal, and bust periods, respectively) from the first stage. It is clear that Fama and MacBeth (1973) approach will have no power 19

21 for the tests during bust periods because there are only 2 bust quarters (8 bust months) in the sample with aggregate business cycle classification. Petersen (2009) shows that both the clustered ordinary least squares regression and Fama and MacBeth (1973) approach generate unbiased estimates of coefficients and standard errors if the sources of bias is cross-sectional dependence, which applies here since stock returns exhibit little persistence in time but strong co-movement in the cross section. More importantly, the clustered ordinary least squares regression suffers less the problem of low power during busts in the aggregate business cycles classification. Therefore, we choose the clustered ordinary least squares regression as our main method and repeat the tests using Fama and MacBeth (1973) approach as a robustness check. The empirical model of the characteristics regression is hence given by ( ) ( r it = a + γ t + b BM G it Iit BM + b NM G it Iit NM ) + b BT ( G it I BT it ) + cx it + e it, (6) where γ t is the fixed effect for month t and the error term e it is assumed to be clustered at the monthly level, both of which are used to control for the cross-sectional dependence among stocks. For stock characteristics, X it, we use the same set of characteristics used by Brennan, Chordia, and Subrahmanyam (1998) and Gompers, Ishii, and Metrick (2003), and the definition and construction of these variables can be found in Appendix two of Gompers, Ishii, and Metrick (2003). Our hypotheses state that b BM > 0 and b BT < 0. We employ the panel sample from September 1990 to December 2007 and regress monthly stock returns on interactions of governance measure SG and business condition controlling for stock characteristics as shown in the model (6), where SG is a dummy indicating strong governance that equals one if G(E)-index is smaller than or equal to 5(0) and zero if G(E)-index is greater than or equal to 14(5). In the regressions, we include monthly fixed effects and allow the error term to be cluster in time (month). The results are reported in Table 5. We start from business cycles classified using COMPUSTAT/CRSP population and then we repeat the regression using business cycles classified on the industry level using Fama and French 10- and 48-industry definitions. For each specification, we conduct the regression with raw returns (reported in Panel A) and repeat the regression using stock returns adjusted for contemporaneous industry median 20