Peer Influence on Dividend Policies

Transcription

1 Peer Influence on Dividend Policies Binay Kumar Adhikari Culverhouse College of Commerce, The University of Alabama Tuscaloosa, AL ( ) Abstract This study finds that one of the most important determinants of a firm s dividend policy is the policy of its peers. Using peer firms idiosyncratic stock returns as instruments to identify the variation in peer firms dividend policies, I show that firms tend to mimic their peers decisions to pay dividends, the dividend yield, decisions to substantially increase or decrease dividends, or to initiate or omit dividends. Consistent with the valuation-as-yardstick concept, mimicking occurs mostly in smaller and younger firms which follow other smaller and younger firms in the same industry. JEL: G35 Key words: Dividend policy, Peer Firms, Mimicking ===Under Major Revision== January 2013

2 Peer Influence on Dividend Policies Whether firms decide to increase, cut or initiate a dividend, we believe that a best-in-class dividend policy should be based on the five CLIMB dimensions: Capital planning, Long term sustainability, Investor preferences, Materiality and Benchmarking to peers. Virtually all board and senior management analysis related to dividend decisions starts with in-depth peer benchmarking. - Dividends: The 2011 guide to dividend policy trends and best practices (J. P. Morgan) I. INTRODUCTION Economic theory suggests that individuals and firms often have incentives to imitate each other. For instance, Scharfstein and Stein (1990) show that managers frequently ignore their own private information and imitate the decisions of others to avoid negative reputation. By imitating, managers send signals to others about their own quality. This behavior can be rational from the perspective of the mangers who are concerned about their reputation in the labor market. Mimicking behavior in financial decisions is common among individuals. Duflo and Saez (2002) find that investing behaviors of people are affected by their peer group. At the firm level, a host of anecdotal evidence shows that firms tend to mimic each other s investment decisions such as new product development, international expansion, R&D, etc. Lieberman and Asaba (2006) review the literature on theories of business imitation and conclude that there are two broad categories of theories: information based theory, where firms follow others that are perceived to have superior information, and rivalry based theory, where firms imitate others to maintain competitive parity or limit rivalry. Evidence from a corporate finance survey also suggests that CFOs regard peer firms decisions as an important factor in making their own firm s policies (Graham and Harvey (2001)). Despite ample theoretical support and anecdotal evidence of the existence of firms imitative behavior, the empirical finance literature is relatively silent on the issue of firms following their peer firms in making financial decisions. Most research on corporate financial policies implicitly assumes that firms make their decisions in isolation. Even though empirical methods try to recognize the importance of peer firms by including industry fixed effects and averages, the causal effect of peer firms policies on a firm s financial decisions has rarely been systematically investigated. 1

3 One of a few recent papers on peer influence on a firm s financial decisions is by Leary and Roberts (2012), who show that corporate capital structure choices are highly interdependent. The authors rather intriguing findings suggest that the most important observable determinant for a firm s capital structure is the capital structure of its peers, and that firms make financial decisions in large by responding to financial decisions of other firms, as opposed to changes in firm-specific characteristics. Consistent with the information based theory of learning and reputation models (e.g. Scharfstein and Stein, 1990), they find that smaller, more financially constrained firms with lower paid and less experienced CEOs are more likely to mimic their peers in capital structure decisions. A parallel study by Fracassi (2012) investigates corporate financial policy decisions, using a matrix of social ties, and concludes that socially connected companies have similar levels of investments. Another similar study by Patnam (2011), who uses the data from publicly listed companies in India, examines whether peer effects exist in corporate market investment, executive compensation and expenditure on R&D. She finds positive industry peer effects for market investment and R&D but not for executive compensation. In this paper I investigate whether firms imitate their peer firms dividend policies, such as the decision to pay dividends each year, the amount of the dividend (yield), dividend initiation, dividend increase, dividend cut and omission. Payout policy constitutes one of the most important decisions made by a firm, and cash dividends have been the primary payout method for centuries. Empirical analysis of peer effect on dividend policies is important because the extant literature gives ample explanation for why firms are likely to imitate each other s dividend policies. Admittedly, the objective of this paper is rather modest in the sense that I attempt to identify the existence of mimicking behavior of firms in dividend decisions without strictly delving into its underlying causes. However, it is important to point out different motivations firms might have for following their peers in dividend policies. Perhaps the strongest support of mimicking comes from the inertia-based explanation of dividends 1. The original purpose of dividends was to make stocks easy to price by making them comparable to debt. Dividends provide the most direct statistic for firm valuation so that investors do not have to concern themselves with accounting practices and more indirect ways of valuation. Sheer popularity of dividend discount models in finance text-books speaks strongly for the 1 See Ben-David (2010) for a review of this literature. 2

4 importance of dividends as a valuation tool. Moreover, empirical evidence suggests that dividends are perceived valuable by retail investors (Graham and Kumar (2006)) as well as by sophisticated institutional investors such as hedge funds (Ben-David, Glushkov, and Moussawi (2010)). Not surprisingly, practitioners and analysts commonly cite dividend yields as a measure of valuation. As dividends have become a common means of payout and a social norm, it has put pressure on managers to conform to it (Frankfurter and Wood (1997)). This valuation-as-yardstick concept proposes that firms manage their dividends in order to help investors value their stream of cash flows and make them comparable to other firms, especially those in the same industry. Consequently, if firms in the same industry compete in the capital markets for high valuations, they have strong motivation to react to each other s dividend policies. In fact, Lintner (1956) observes that managers consider the amount of payout relative to the benchmark of the existing rate of dividends paid by their firms. Simply put, deviating too much from the industry in terms of dividend policy is likely to result in negative market reaction. Thus, managers simply play it safe by conforming to the industry practice. This should be especially true for younger and smaller firms that do not have long histories of operation and hence are in greater need of being comparable to peers. A prediction from the valuation-as-yardstick concept is that dividend changes are correlated within industries. If investors use the same dividend yield to price firms within an industry and if firms are interested in having high valuations, a change in dividend payout by one firm is expected to be followed with payout changes in the same direction by peer firms. Indeed, empirical evidence by Firth (1996) supports this conjecture by showing that dividend announcement by a firm also affects the valuation of other firms in the same industry. In a related way, signaling motivations also make a plausible explanation for mimicking behavior. Lintner (1956) stresses that a dividend increase signals management s perception that earnings are going to increase. Signaling theory argues that owing to the asymmetric information, dividends are explicit signals about future earnings, sent intentionally by management to the shareholders (Bhattacharya (1979, Miller and Rock (1985), John and Williams (1985)). Empirical evidence does have support, albeit with limited conclusiveness, that firms pay dividends in order to convey information to shareholders (e.g. DeAngelo, De Angelo and Skinner (1996), Benartzi, Michely and Thaler (1997)). Clearly, to the extent dividends are effective signals that attract 3

5 investors attention and that firms compete with each other for this positive attention, firms have substantial motivation to mimic peer firms dividend policies. Again, this should be more prevalent among younger and smaller firms who have higher need to signal their quality. To my knowledge, this is the first study that attempts to parse out the impact of peers firms dividend policy on the own firm s dividend policy in a causative sense. I find evidence, with varying degrees of success, that firms mimic their peer firms in their decision to pay annual dividend, the amount of dividend, the decision to substantially increase or decrease dividends and the decision to initiate and omit dividends. Consistent with the valuation-as-yardstick concept, mimicking is predominant in smaller and younger firms which follow the decisions of other smaller and younger peer firms in the same industry. For more established (older and larger) firms, the only strong evidence of mimicking is for large dividend increases, and some evidence of mimicking in annual dividend payments. This work is related to Firth (1996) who finds that the dividend change of one firm is related to stock price performance of other firms in the same industry, Leary and Roberts (2012) who find capital structure choices of firms in the same industry are highly interdependent, Fracassi (2012) and Patnam (2012) who show that interrelated firms have correlated financial decisions. As a sideproduct, my results help explain the dividend puzzle - the largely unresolved question of why firms pay dividends - and also contribute to the understanding of the disappearance of dividends among small payers and concentration among larger payers. II. IDENTIFICATION AND INSTRUMENTAL VARIABLES Identifying mimicking behavior is empirically challenging because of obvious simultaneity issues. The issue of simultaneity arises from the fact that rather than one firm mimicking the other, all firms in the same industry are responding to a common shock to the industry. For example, firms in the same industry might change their dividend policies together because of change in investment opportunity or tax laws pertaining to that industry. This creates positive correlation in their policies without any causality flowing from one firm to another. This essentially poses an endogeneity problem brought about by unobservable omitted variables. 4

6 To address the issue of such endogeneity, I follow Leary and Roberts (2012) and instrument the peer firms dividend policies with the idiosyncratic component of peer firms stock returns. Following Leary and Roberts, I define peer firms as all other firms in the same three-digit SIC code in a given year. The strategy here is to exploit heterogeneity in dividend policies among peer firms brought about by their idiosyncratic equity shock. While Leary and Roberts use one lag of average idiosyncratic abnormal return of peer firms as an instrument for a firm s book and market leverage, I find that idiosyncratic returns and their appropriate lags also serve as strong and valid instruments for peer firm s dividend payment, yield, initiation, omission, increase and decrease decisions. The rationale follows. The first rationale is for the relevance of instruments. I posit that my instruments are relevant in the sense that they are strong predictors of the firms dividend decisions. Several studies have found that profitable firms are more likely to pay cash dividends. Fama and French (2001) find that likelihood of paying divided increases with profitability. Likewise, DeAngelo, DeAngelo and Skinner find that firms tend to increase their dividends in a period of earnings growth (1996). Benito and Young (2001), Ferris, Sen, and Yui (2006), and Renneboog and Trojanowski (2005) obtain similar evidence in the UK firms, and Von Eije and Megginson (2006) find the similar results among European Union firms. Moreover, using a sample of six countries around the world, Denis and Osobov (2008) find that dividends are concentrated among the largest and most profitable payers. Perhaps the strongest support for my instruments comes from Benartzi, Michaely and Thaler (1997) who show that firms that increase dividends in year 0 have experienced significant earnings increases in years -1 and 0 and firms that cut dividends in year 0 have experienced a reduction in earnings in year 0 and -1. Consistent with all these findings, I find that two lags of idiosyncratic stock returns are positively and significantly related to the propensity to pay dividends both for individual firms and for peer firms. In particular, idiosyncratic stock returns are positively and significantly correlated with dividend payment, initiation and increase decisions and are negatively correlated with dividend cut and omission decisions. I employ two lags of idiosyncratic returns, unlike Leary and Roberts (2012) who use just one, for two important reasons. First, dividend decisions tend be based on the history of profitability rather just last year s profitability. Therefore, it is likely that a few lags of equity shock might jointly explain dividend decisions. Second, using multiple instrumental variables, as opposed to just one, allows me to perform the econometric test of overidentification restriction (Hansen s J) which is a crucial test to establish the validity of instruments. For dividend omission 5

7 and decrease decisions, I also include a contemporaneous equity shock as an instrument because negative dividend changes are likely to be driven by more recent financial distress. Certainly, my econometric tests confirm that these instruments are not weak by showing that they explain peer firms dividend decisions in a highly significant way. The second argument is for the validity of instruments. For the instruments to be valid they must satisfy the exclusion restriction. In this context, this means peer firms idiosyncratic returns should affect the own firm s dividend policies only through their correlation with peers dividend policies. I obtain the idiosyncratic part of stock returns using Fama-French and Carhart model which is well-established in the asset pricing literature to isolate firm-specific shock component. Moreover, my model goes a step further and removes any industry specific commonalties in stock returns by including excess industry return as an additional factor in the model. Therefore, theoretically, peer firms idiosyncratic shock should not contain any information about the own firms. These shocks actually have some nice statistical properties that speak for their suitability as instruments. In particular, these shocks are serially uncorrelated and serially cross-uncorrelated which means firms shocks do not forecast future shocks for themselves or their peers. Therefore, the (lagged) idiosyncratic stock returns are plausible valid instruments. Sure enough, my econometric tests show that these instruments strongly pass the test of overidentification which further endorses their suitability. III. DATA AND SUMMARY STATISTICS My primary data come from CRSP-COMPUSTAT merged database from 1965 to Following previous studies, I exclude financial (6000=<SIC=<6999), utilities (4900=<SIC=<4999) and government entities (SIC>9000) from the sample in order to avoid firm policies dictated by regulatory restrictions. I define and explain the construction of various variables in the appendix. Table 1 presents the summary statistics of my final sample of 65,815 firm-year observations for peer firms and 67,582 observations for individual firms. I lost some data in the process of averaging for peer firms. The final sample consists of 7,881 unique firms over the 46 year period. There are 242 industries identified by three digit SIC codes in my sample. A typical industry has a median of 14 firms and a mean of 32 firms. In order to mitigate unwanted influence of extreme outliers, all continuous variables used in this study are winsorized at upper and lower 0.5%. Separate summary statistics are provided for individual firm variables and peer firm average variables. I find that slightly 6

8 less than half of the firm years correspond to positive divided payments with average dividend yield of around 1.5%. Similarly, about 1.7% of firm years are dividend initiation years and about 2.3% of firm years entail dividend omissions. Nearly 14% of firm years had large dividend increases whereas about 5.5% of firm years had large dividend cuts. By construction, peer firm averages for all variables are similar to individual firm averages but standard deviations of peer firm averages are consistently lower than those of individual firm variables. Not going into the details of summary statistics, I note other variables are similar to those found in previous literature. IV. EMPIRICAL METHODOLOGY I employ the following empirical model which is similar to the one used by Leary and Roberts (2012) and is a generalization of the ones used in many past studies. (1) The indices i, j and t correspond to firm, industry and year respectively. Peer firms are defined as all firms in the same three-digit SIC code, except firm i, in a given year. The outcome variable is one of the six dividend related policies of firm i in industry j and year t. The variable denotes peer firm average outcome (average of the all the firms in industry j in year t except firm i). Following Leary and Roberts, I use a contemporaneous measure as opposed to a lagged measure, to focus on identification. A contemporaneous measure is appropriate because the effect of peer firms can be more clearly identified if there is not enough time lag for other variables to make an influence. and vectors contain peer firm averages and firm-specific characteristics, respectively, as control variables. represents year fixed effects and is the firm specific error term that is assumed to be correlated within the firm and heteroskedastic. Therefore, all my regression specifications have heteroskedasticity robust standard errors clustered within firms (Perterson (2009)). Construction of the Instruments 7

9 To parse out the idiosyncratic component of stock returns, I follow Leary and Roberts (2012) but also augment the model with size, book to market and momentum factors as follows: Where refers to the total return for the firm i in industry j over the month t. is the excess market return, is size factor, is the book to market factor, is the momentum factor, and is the excess return on an equally weighted industry (three-digit SIC codes) portfolio, excluding firm i's return. Notice, the last factor, is not a risk factor recognized in the literature, but it is included to remove any common variation in returns across the same peer group. I estimate equation (2) for each firm on a rolling annual basis using historical monthly returns. I require at least 24 months of historical data and use up to 60 months of data in the estimation. Most of my sample uses 60 months of data. As an example, in order to obtain expected and idiosyncratic return for COKE from January 2000 to December 2000 I first estimate equation (2) using monthly returns from January 1995 to December Then, using the estimated coefficients from equation (2) and monthly factor returns from January 2000 to December 2000, I use equation (3) to compute the expected and idiosyncratic returns as follows: Here the letters with hats indicate estimated parameters. To obtain the expected and idiosyncratic risk of 2001, I repeat the same process by updating the estimation sample from January 1996 to December 2000 and use the estimated betas for 2001 returns. This process generates time varying betas but they are constant for a calendar year. 8

10 Understanding the importance and relevance of this instrument is crucial for appreciating the results and inferences of this paper. The idiosyncratic return obtained from the above model,, is the return of the firm after removing all known sources of systematic variation (i.e., exposure to market, size factor, book to market factor and momentum factor). If well-founded asset pricing theories were to be trusted, the residuals obtained from the Fama French and Carhart model should be purely firm specific and hence should be free from any commonalities across the firms. Additionally, my model also includes the industry s excess return in order to purge remaining systematic variation of returns, if any, across the firms in the same industry. Therefore, in addition to being strongly validated by rigorous econometrics tests, my instruments are also theoretically wellfounded. Table 2 presents the summary statistics for the estimated factor regressions. On average, there are 55 observations (the median being 60) per rolling regression over the year. The majority of the regressions have a full five year (60 month) window. The R-squared seems reasonable with a mean of.307 and a median of.288. The factor regressions load positively on market, size and bookto-market factors and negatively on the momentum factor. Industry beta has the smallest load in absolute term which suggests that the four factors are successful in purging most of the systematic variation in the asset returns. The average idiosyncratic return is less than 10 basis points which arise from rounding errors and observations dropped in the data cleaning process. I construct the final instrument by taking the geometric average of the monthly idiosyncratic returns obtained this way. Then, I average the compounded returns across the peer firms in the industry excluding firm i. V. PEER FIRM EFFECTS ON DIVIDEND POLICIES In this section I present and discuss the results obtained from my regression analysis using instrumental variable techniques to identify peer firms influence on different dividend decisions. Table 3 presents the results from regression analysis that explores whether peer firms decision to pay a dividend in a year affects the own firm s decision. The first model is a linear probability model (LPM), estimated by OLS, where the dependent variable is a dummy variable indicating whether the firm pays a cash dividend in the given fiscal year. Probit models obtain similar 9

11 results but I chose to use LPM because of computational ease, especially with different subsamples. My variable of interest is the fraction of peer firms who are dividend payers. The control variables include several factors identified in the literature to predict dividend decisions. For example, I control for future growth opportunities by including size, sales growth, market to book ratio and R&D expenses (see e.g Grullon and Michaely (2002), Fama and French (2002) and Grullon, Paye, Underwood, and Weston. (2009)). Likewise, I include the ratio of retained earnings to book equity (RE/BE) to control for the life cycle stage of the firm (DeAngelo, DeAngelo and Stulz (2006)). Similarly, I include the firm s cash flow volatility to control for cash flow risk. Finally, I add two lags of the firm s idiosyncratic stock returns to the regression. Consistent with my hypothesis, the fraction of dividend paying firms in the peer group has a significant and positive impact on the own firm s dividend paying decision even after controlling for an array of other factors found in previous literature. Most other variables also take their expected signs. For instance, older and larger firms, and firms with higher earned to contributed capital ratio (RE/BE) are more likely to pay dividends. On the other hand, firms with higher sales growth are less likely to pay dividends. Similarly, firms with risky cash flows and higher leverage are less likely to pay dividends. Noticeably, consistent with my rationale for the choice of instruments, the two lags of the own firm s idiosyncratic returns are positive and highly significant in explaining the decision to pay dividends. Even if OLS obtains a significant and positive coefficient on peer firms dividend payout policy in explaining the own firms dividend paying decisions, one cannot establish the causality because of endogeneity issues discussed earlier. Therefore, I employ a two-stage least squares (2SLS) technique using the two lags of peer firms average idiosyncratic equity shock as my instruments for peer firms dividend paying decisions. Model 2 presents results of the first stage regression where the dependent variable is the fraction of peer firms which paid cash dividend in a given year. Apart from other control variables similar to those in OLS, this model includes my instruments, namely the two lags of peer firm average idiosyncratic equity shock, as explanatory variables. The results reveal that my instruments positively predict peer firms decisions to pay dividends in a highly significant way. Both coefficients on lagged average peer firm idiosyncratic returns are individually significant at 1% level and the underidentification test shows that they are also jointly significant at 1%. This suggests that my instruments are relevant in explaining the fraction of peer firms paying cash dividends each year. Moreover, the overidenfication test obtains a small and insignificant chi-squared statistic 10

12 meaning that these instruments are exogenous to the system and hence are valid for the 2SLS analysis. In model 3 of table 3, the second stage regression shows that the (instrumented) fraction of dividend paying peer firms has a positive and significant coefficient in explaining the own firm s dividend payment decisions. Since the second stage is a linear probability model, the coefficient of on peer firms policies implies that a 10% increase in the fraction of dividend paying firms in the peer group increases the probability of the own firm paying dividend by about 5.3%. This is an economically significant impact which is also statistically highly significant. This effect of peer firms can be considered a causal effect because it is estimated from a valid instrumental variable approach. All other control variables retain values similar in magnitude and significance to those obtained from OLS estimation. Additionally, some peer firm average variables also turn significant in interesting ways. In particular, peer firm leverage has a positive impact on dividend paying decisions and peer firm average size has a significantly negative impact in the own firm s dividend paying decisions. This suggests that a firm which has more leveraged and smaller peers than itself is more likely to pay dividends. The next analysis is for dividend yield. Whereas the first analysis deals with whether or not firms own dividend payment decisions react to the dividend payment decision of peer firms in a given year, analysis of dividend yield will show whether they react to the amount of dividends paid. Analysis of mimicking in dividend yield is especially interesting because practitioners, as well as academics (e.g. Graham and Kumar (2006)) regard dividend yield as a yardstick for valuation. Table 4 presents regression results on dividend yield of the firm. Analogous to dividend payment decisions, OLS obtains a positive and highly significant coefficient on peer firm average dividend yield. Other variables take expected signs as suggested by previous literature. Although the positive coefficient on peer firm average yield is consistent with my hypothesis, it does not necessarily suggest a causal relationship because of endogeneity issues. Therefore, in order to establish causality, I again employ 2SLS, using the two lags of idiosyncratic returns as instruments for peer firm dividend yield. Model 2 depicts the first stage regression and obtains positive and highly significant coefficients on both lags of average idiosyncratic returns in explaining peer average dividend yield. Once again, the instruments pass the test of overidentification meaning that these are valid instruments for the 2SLS estimation. Peer firm level control variables take similar signs to individual 11

13 firm level control variables in OLS. Finally, the second stage estimate (model 3) also obtains a positive and significant coefficient on peer firm s average dividend yield in explaining the individual firm s dividend yield. Specifically, the coefficient estimate suggests that a 10% increase in peer firms average dividend yield cause the firm to increase its own dividend yield by about 4%. Clearly, in addition to being statistically significant, this estimate is economically substantial. Next, table 5 presents results from the analysis where the dependent variable takes the value of 1 for firm i in year t if the percentage increase in dividends is at least 15% and zero otherwise. The rationale is that if mimicking in dividends exists because of valuation or signaling motivations, a large increase in peer firms dividends should be followed by a large dividend increase by the own firm. Since the goal is to test whether firms react to large dividend increases by peer firms, a cut off of 15% is chosen following Li and Zhao (2008) to make sure that the increase is perceivably large. Once again, the results are similar to those from dividend payment and dividend yield decisions. In short, both OLS and 2SLS obtain a positive and significant coefficient on the fraction of peer firms who increase dividends substantially in explaining the own firm s large dividend increase decision. My instruments easily pass both underidentification and overidentification tests so the estimates from 2SLS can be considered causal. 2SLS obtains a coefficient of on the fraction of peer firms increasing dividends. In economic terms, this implies that a 10% change in the fraction of peer firms with a large increase in dividends changes the probability of the own firm s large dividend increase by about 5.4%. Again, this estimate is both statistically and economically significant. Finally, I conduct analysis on three more dividend related decisions: dividend initiation, dividend omission and the decision to significantly cut dividends. I define large dividend cuts as a decrease of dividends by at least 5%. Since dividend cuts are rare, choosing a higher threshold would have resulted in too small a sample of dividend cuts to do a meaningful analysis. Tables 6, 7 and 8 show the results of dividend initiations, large cuts, and omissions, respectively. The summary of the results is the following. OLS estimates obtain positive and statistically significant coefficients on peer firm averages in explaining individual firms decisions of dividend initiation, omission and dividend cut. To establish causality, I run 2SLS regressions. The first stages show that my instruments significantly explain peer firms dividend initiation, dividend cut and dividend omission decisions. In particular, peer firms average idiosyncratic returns and their lags positively and significantly predict peer firms dividend initiation decisions, and negatively and significantly predict dividend cut and 12

14 omission decisions. Next, the second stages obtain positive and significant coefficients of peer firm dividend initiation and peer firms large dividend cuts in explaining the own firms dividend initiation and dividend cuts, respectively. This suggests that firms are more likely to initiate dividends and cut dividends if their peers do so. However, for the dividend omission decision, the second stage does not obtain a significant coefficient on peer firms decision to omit dividends. So, in the full sample analysis, I do not find significant evidence of mimicking in dividend omission decisions. Admittedly, statistical significance of peer firms influence in each of the last three decisions is weaker than others. This is not surprising since these decisions are far less frequent than the ones discussed earlier. Most firms initiate their dividends once in their lifetime and they hardly ever cut or decrease dividends. Considering the rarity of these types of decisions, a 10% level of statistical significance is quite meaningful. Overall, I find strong evidence that firms tend to mimic their peer firms decisions to pay dividends, the amount of dividends and the decision to substantially increase dividends. There is somewhat weaker evidence for peer group influence on dividend initiation and dividend cut decisions. From full sample analysis, I do not find statistically significant evidence of mimicking in dividend omission. However, as shown later, evidence of mimicking in some of these decisions turn out to be significant in some subsamples. VI. WHO MIMICS AND WHO IS MIMICKED? Leary and Roberts (2012) find that younger and smaller firms tend to mimic the decisions of older and larger firms. They attribute these results to observational learning and reputational hypothesis suggesting that managers of young and small firms, who have the greatest reputational concerns, are likely to mimic those who seem to have greater expertise. While their findings are plausible for capital structure decisions, this might not necessary hold for dividend decisions. In fact, if the valuation-as-yardstick theory is true, firms should mimic the dividends of other firms with similar size and age to facilitate comparability among own types. Moreover, for more mature and bigger firms, dividends become a norm and a sticky decision so there might not be much time-series variation in their dividend decisions. Therefore, mimicking behavior might exist only in smaller and younger firms who have more flexibility in these decisions. In order to delve into these issues I strive to test if younger and smaller firms mimic dividend decisions of older and larger firms or they mimic their own age and size cohorts. 13

15 Size Cohorts Table 9 examines whether firms follow their larger counterparts in the industry or whether they follow their own size cohorts in dividend payment, dividend yield, large increase, initiation, decrease and omission decisions. To answer this, I partition the sample firms into three terciles of firm size within an industry year. Then, following Leary and Roberts (2012), I define size followers as the firms in bottom two terciles of firm asset size and size leaders as the firms in the top tercile of asset size distribution. In column A of table 9, I exclude the top third of the size distribution (size leaders) from the analysis so that the estimation is done using only the subsample for smaller (size follower) firms. Peer firm average variables are also created only using the subsample of the follower firms. The dependent variables are the different dividend decisions of the follower firms. In essence, column A shows whether smaller firms mimic other firms in small size cohort. All results shown in table 9 are key statistics from 2SLS regression analysis. Next, in column B of table 9, I exclude the bottom two thirds of the size distribution (size followers) from the analysis so that the estimation is done only for the top third subsample of the largest firms. Again the peer firm average is the average of the top third size cohort of the sample only. The dependent variables are the different dividend related policies of the firms in the top size cohort. In essence, column B shows whether the largest firms mimic dividend policies of other firms in the largest group. Finally, column C shows whether firms in the smaller size cohort follow those in the larger size cohort. This analysis is similar to that of Leary and Roberts (2012) who find that smaller firms mimic capital structure choices of the larger ones. Here the dependent variables are the different dividend related decisions of the smaller firms. However, the independent variables are the corresponding dividend decisions of the largest third of the firms. In essence, I am testing whether smaller firms respond to dividend policies of larger firms in the same industry. For dividend payment decisions, column A of table 9 shows that the coefficient on peer firm average payment decisions (instrumented with 2SLS) is positive and highly significant. The implication is that smaller firms tend to mimic the decisions of other firms in the same size cohorts. 14

16 Specifically, a 10% increase in smaller peer firms deciding to pay dividends increases the chance of a firm in the same size cohort to pay dividends by about 12.1%. Obviously, this figure is economically large on top of being highly statistically significant. Likewise, there is also significant evidence of mimicking among the size leaders. In terms of economic magnitude, a 10% increase in the fraction of the peer firms paying dividends among the size leaders increases the probability of a size leader firm to pay dividend by 10.2%. For larger firms, the coefficient is smaller and slightly less significant than that obtained for size followers. However, more interestingly, there is no evidence that smaller firms mimic the dividend decisions of lager firms as the estimates presented in column C of table 9 show an insignificant coefficient on (instrumented) peer firm averages of size leaders in explaining the dividend policies of size followers. Overall, it appears that firms mimic dividend payment decisions of their own size cohorts in the industry and the magnitude of mimicking is larger among smaller firms. The bigger firms dividend decisions have no significant influence on the smaller firms decisions. The latter contrasts with the results of Leary and Roberts regarding their capital structure decisions. For dividend yield, the coefficient on peer firm average is significant only for the size follower subsample. The estimated coefficient of 1.03 on peer firm averages among smaller firms (column A) suggests that size followers tend to change their dividend yield approximately one to one when their peers in the same size group change their yield. There is no statistically significant evidence of mimicking among the size leader tercile (column B). Once again, there is no evidence that firms in the size followers mimic the dividend yield of size leaders (column C). My analysis of large dividend change indicates that mimicking in large dividend increase happens in firms in both size groups. Interestingly, the magnitude of mimicking in dividend increases is much larger among bigger firms. Column A suggests that among smaller firms, a 10% change in the fraction of peer firms increasing the dividend changes the probability of the own firm s dividend increase by about 10.2%. On the other hand, among larger firms it changes the probability of another large firm doing the same by 27.1%. This suggests larger firms are more pressured to increase dividends than smaller firms when their counterparts do so. Additionally, I also find evidence that smaller firms mimic larger firms decision to substantially increase dividends (column C). Large dividend increases are the only decisions where I find evidence of mimicking 15

17 across cohorts. However, the magnitude of cross-cohort mimicking is smaller compared to withincohort mimicking. The final three analyses pertain to dividend initiation, omission and large dividend decrease by firms of different size sub-samples. Recall, in the full sample analysis I do not find strong evidence of mimicking in any of these decisions. For dividend initiation decisions, I find statistically significant evidence of mimicking behavior among size followers (column A) and no evidence of mimicking among size leaders (column B). In particular, among small firms a 10% increase in the fraction of peer firms initiating dividends increases the chance of a firm in that cohort of doing the same by 12%. As before, there is no evidence that smaller firms mimic decisions of larger firms (column C). Once more, for large dividend cut decisions, I find significant evidence of mimicking among small firms only. The coefficient on peer firm dividend cut suggests that a 10% change in the fraction of firms cutting dividends changes the likelihood of the own firm s dividend cut by 9.4%. There is no evidence of mimicking in dividend omission in any size group although the small size group is the closest to being significant. Age Cohorts Table 10, which is analogous to table 9, presents similar analyses on age leader (top age tercile) and age follower (bottom two age terciles) firms. The idea is to examine if firms in any age group predominantly exhibit mimicking behavior and, as found by the previous study, whether younger firms mimic the decision of older firms. To summarize, the results from different age groups are similar to those from different size groups. Mimicking of dividend decisions mainly exists among younger firms and younger firms follow other younger firms in the industry (column A). Once again there is very little evidence of younger firms mimicking the decisions of older firms (column C). Only younger firms tend to mimic annual dividend payment decisions of their other younger counterparts (column A) and there is no strong evidence of mimicking among older firms (column B). As usual, there is no evidence that young firms mimic the decisions of their older counterparts in the same industry (column C). Once again, for dividend yield decision there is statistically significant evidence of mimicking among younger firms only. Firms in both age cohorts mimic the decision of 16

18 their cohorts (column A and B) to increase dividends. Just as in the size terciles, the magnitude is much larger and statistical significance is much stronger for the older firm cohort in mimicking dividend increase decisions. This again supports that the established firms are more pressured to increase dividends when their peers do so. There is also somewhat weaker evidence that smaller firms mimic larger firms decision to increase dividends. As before, the evidence of mimicking of dividend cut and omission is significant, each at 5% level, for younger age cohorts. There is no significant evidence of mimicking dividend initiation across any age group although both groups come close to being significant. To recap this section, rather than smaller and younger firms mimicking older and larger ones, firms tend to mimic dividend decisions of peers of similar age and size. It seems that most action in mimicking happens among younger and smaller firms. This is rather surprising given that most dividend payers are older and larger firms. Even though mimicking of dividend payment decisions and initiation decisions were significant mostly for younger and smaller firms, older and larger firms seem to be more pressured to mimic major positive changes in dividend (e.g. initiation and large increase). It appears younger and smaller firms tend to have the liberty to follow their counterparts in routine dividend decisions as well as both negative and positive changes (i.e. large increase and large decrease). This is consistent with theory and empirical evidence that for more established firms, dividend payment becomes a norm over time. So, these firms avoid negative changes in dividends, even if it might sometimes be efficient, due to the fear of adverse market reactions. The fact that younger and smaller firms tend to mimic both positive and negative changes in dividends but older firms mimic positive changes only might partially explain disappearance of dividends among small payers and concentration among large payers originally documented by Fama and French (2001) and later refined by DeAngelo, DeAngelo and Skinner (2004). Moreover, supporting the valuation-as-yardstick concept, firms tend to follow their own size and age cohorts and mimicking is predominantly driven by smaller and younger groups. 17

19 VII. CONCLUSION This work contributes to an interesting domain of the literature that investigates whether peer firms matter in making financial decisions. While intuition and theoretical findings support the idea that peers must influence a firm s financial decisions, empirical evidence is somewhat scarce. This paper adds to this literature by showing that decisions of peers have a significant effect on a firm s dividend decisions. This work also contributes to the overall dividend policy literature. Why firms pay dividends has always baffled financial economists. Many firms religiously pay dividends even though it may be an inefficient way to distribute returns to stockholders. I come up with a partial explanation of this phenomenon by showing that firms pay dividends because their peers do so, in order to keep their valuation comparable. Interestingly enough, most mimicking in dividends occurs in smaller and younger firms even though most of the dividend payers are larger and older firms. The fact that smaller and younger firms mimic the dividend decisions of their counterparts but older and larger firms mostly do not is largely consistent with the valuation-as-yardstick explanation of dividend decisions. Incidentally, I uncover a few interesting findings that help partially explain the phenomenon of disappearance of dividends and concentration of dividends among large payers. I find that larger and older firms respond more strongly to the increase in dividends of their larger and older counterparts in the same industry. They do not respond to peers dividends cuts, perhaps because negative changes in dividends, regardless of the reason, are likely to face negative market reactions. On the other hand, younger and smaller firms mimic both negative and positive dividend changes of their peers. Over time, this is likely to concentrate dividends among the larger and older firms. I hope that this work will motivate researchers to acknowledge the importance of peer firms in empirical analysis of dividend policies, and also help advance the research on the effect of peer firms on a variety of other corporate decisions. 18

20 References Benartzi, Shlomo, Roni Michaely, and Richard Thaler, 1997, Do changes in dividends signal the future or the past?, Journal of Finance 52, Ben-David, Itzhak, 2010, Dividend policy decisions, in H. Kent Baker and John R. Nofsinger, eds., Behavioral Finance (John Wiley & Sons, Inc.), Robert W. Kolb Series in Finance, chapter 23, Ben-David, Itzhak, Denys Glushkov, and Rabih Moussawi, 2010, Direct evidence for the behavior of hedge funds when arbitrage is costly, Working Paper, the Ohio State University. Benito, Andrew and Garry Young, 2001, Hard times of great expectations? Dividend omissions and dividend cuts by UK firms, Working paper, Bank of England. Bhattacharya, Sudipto, 1979, Imperfect information, dividend policy, and the bird-in-hand fallacy, Bell Journal of Economics 10, DeAngelo Harry, Linda DeAngelo and René M. Stulz, 2006, Dividend policy and the earned/contributed capital mix: A test of the lifecycle theory, Journal of Financial Economics, 81, DeAngelo, Harry, Linda De Angelo, and Douglas Skinner, 1996, Reversal of fortune: Dividend policy and the disappearance of sustained earnings growth, Journal of Financial Economics 40, DeAngelo, Harry, Linda DeAngelo, and Doug Skinner, 2004, Are dividends disappearing?, Dividend concentration and the consolidation of earnings, Journal of Financial Economics, 72, Denis, D., and I. Osobov, 2008, Why do firms pay dividends? International evidence on the determinants of dividend policy, Journal of Financial Economics 89, Duflo, Esther, and Emmanuel Saez, 2002, Participation and investment decisions in a retirement plan: The influence of colleagues' choices, Journal of Public Economics 85, Fama, Eugene F. and Kenneth R. French, 2001, Disappearing dividends: Changing firm characteristics or lower propensity to pay?, Journal of Financial Economics, 60,

21 Fama, Eugene F., and Kenneth R. French, 2002, Testing tradeoff and pecking order predictions about dividends and debt, Review of Financial Studies 15, Ferris, S., Sen, N., Yui, H., God save the queen and her dividends: corporate payouts in the UK, Journal of Business 70, Firth, Michael, 1996, Dividend changes, abnormal returns, and intra-industry firm valuations, Journal of Financial and Quantitative Analysis 31, Fracassi, Cesare, 2012, Corporate Finance Policies and Social Networks, Working paper, University of Texas at Austin. Frankfurter, George M., and Bob G.Wood, 1997, The evolution of corporate dividend policy. Journal of Financial Education 23, Graham, John R., and Campbell Harvey, 2001, The practice of corporate finance: Evidence from the field, Journal of Financial Economics, 60: Graham, John R., and Alok Kumar. 2006, Do dividend clienteles exist? Evidence on dividend preferences of retail investors. Journal of Finance 61:3, Grullon, Gustavo, and Roni Michaely, 2002, Dividends, share repurchases, and the substitution hypothesis, Journal of Finance 57, Grullon, Gustavo, Bradley Paye, Shane Underwood, and James P. Weston, 2009, Has the propensity to pay out declined?, Journal of Financial and Quantitative Analysis, 46, John, Kose, and Joseph Williams, 1985, Dividends, dilution, and taxes: A signalling equilibrium, Journal of Finance 40, Leary, Mark T., and Michael R. Roberts, Do peer firms affect corporate financial policy?, Working paper, University of Pennsylvania and Washington University in St. Louis. Li, Kai, and Xinxeli Zhao, 2008, Asymmetric Information and Dividend Policy, Financial Management 37, Lieberman, Marvin and Shigeru Asaba, 2006, Why do firms imitate each other?, Academy of Management Review 31,