Asymmetric Information and Dividend Policy

Transcription

1 See discussions, stats, and author profiles for this publication at: Asymmetric Information and Dividend Policy Article in Financial Management November 2008 Impact Factor: 1.36 DOI: /j X x CITATIONS 59 READS authors: Kai Li University of British Columbia - Va 88 PUBLICATIONS 1,664 CITATIONS SEE PROFILE Xinlei Zhao Kent State University 10 PUBLICATIONS 190 CITATIONS SEE PROFILE All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately. Available from: Xinlei Zhao Retrieved on: 08 May 2016

2 Asymmetric Information and Dividend Policy Kai Li and Xinlei Zhao We examine how informational asymmetries affect firms dividend policies. We find that firms that are more subject to information asymmetry are less likely to pay, initiate, or increase dividends, and disburse smaller amounts. We show that our main results are not driven by our sample and that our results persist after accounting for the changing composition of payout over the sample period, the increasing importance of institutional shareholdings, and catering incentives. We conclude that there is a negative relation between asymmetric information and dividend policy. Our results do not support the signaling theory of dividends. In this paper, we study how informational asymmetries affect firms dividend policies by examining the relation between a firm s dividend policy and the quality of its information environment. Dividends have long puzzled financial economists. Miller and Modigliani (1961) prove that dividend policy is irrelevant to share value in a perfect and efficient capital market. However, the observation that share prices typically rise when firms increase dividend payments suggests that, on the contrary, dividends do matter after all. Various studies have proposed various explanations for firms dividend behavior (see Allen and Michaely, 2003, for a comprehensive review of the literature). Among them, the dividend signaling theory is one of the dominant explanations. Under the signaling models of Bhattacharya (1979), John and Williams (1985), and Miller and Rock (1985), managers know more about the firm s true worth than do its investors and use dividends to convey information to the market. Thus, these models suggest a positive relation between information asymmetry and dividend policy. Other studies have developed tests to examine the dividend signaling models. However, our study may be the first to specifically examine the testable implications of the signaling models in the context of the relation between information asymmetry and firms dividend policies. To conduct our research, we ask the following questions: Are corporate dividend policies affected by the degree of information asymmetry that firms face? Is the relation consistent with the signaling view of asymmetric information? Given that information asymmetry is a We thank Xia Chen for her help in obtaining the analyst coverage data, Bill Christie (the editor), an anonymous referee, Nalinaksha Bhattacharyya, Laurence Booth, Jason Chen, Qiang Cheng, Ming Dong, Charles Gaa, Ron Giammarino, Rob Heinkel, Harrison Hong, Alan Kraus, Rafael La Porta, Ranjan D Mello, Hernan Ortiz-Molina, Gordon Phillips, Antoinette Schoar, Carina Sponholtz, John Thornton, seminar participants at Kent State University, University of British Columbia, and participants of the Northern Finance Association Meetings in Vancouver, the FMA European Conference in Stockholm, and the FMA Annual Meetings in Salt Lake City for valuable comments. We gratefully acknowledge the contribution of Thomson Financial for providing analyst data, available through the Institutional Brokers, Estimate System. These data have been provided as part of a broad academic program to encourage earnings expectations research. Li acknowledges the financial support from the Social Sciences and Humanities Research Council of Canada. Li also wishes to thank the MIT Sloan School of Management for its hospitality and support when this paper was initially written. All errors are our own. Kai Li is the W.M. Young Professor of Finance at the University of British Columbia in Vancouver, BC, Canada. Xinlei Zhao is an Associate Professor of Finance at Kent State University in Kent, OH. Financial Management Winter 2008 pages

3 674 Financial Management Winter 2008 major market imperfection and that dividend policies are among the most important corporate decisions, these are important questions. We use analyst earnings forecast errors and the dispersion in analyst forecasts to gauge the degree of information asymmetry between managers and investors. We find that both analyst earnings forecast errors and the dispersion in forecasts are negatively, and very often significantly, associated with a firm s likelihood of paying dividends, initiating or increasing dividends, and with the level of dividends paid. Overall, our findings suggest that firms with more transparent information environments pay out more dividends. This evidence does not support the signaling theory of dividends. We also examine the relation between the quality of a firm s information environment and measures of total payout that include both dividends and repurchases. We do not find a positive association between information asymmetry and repurchase activities. Signaling theory predicts a stronger positive relation between asymmetric information and dividends than between asymmetric information and repurchases. Our finding of a stronger negative relation between asymmetric information and dividends confirms our evidence on the lack of support for the signaling theory. Our results are not broadly consistent with the dividend signaling models. The paper is organized as follows. In Section I, we discuss the signaling theory, describe the sample and variables, and provide summary statistics. Section II presents our empirical results on firms dividend policies. Section III provides robustness checks on our main findings. Section IV presents our investigation of repurchase activities and total payout policies, and Section V concludes. I. Variable Construction and Sample Characteristics Since dividends provide a costly way of resolving asymmetric information, we examine the relation between information asymmetry and firms dividend policies under the signaling models. Because the resolution of asymmetric information is valuable, firms with greater asymmetric information should be more active dividend payers. Therefore, after controling for other dividend determinants, if the signaling theory of dividends is valid, we would observe a positive relation between information asymmetry and firm dividend policy. Further, because dividends imply a firm commitment and are also historically tax disadvantaged relative to repurchases, dividends constitute a more costly signal and investors should perceive them as having stronger information content. Thus, the signaling theory predicts a stronger positive relation between asymmetric information and dividends than between asymmetric information and repurchases. 1 Following earlier studies, we use Compustat and CRSP to examine dividend policy in industrial firms. We exclude utilities (SIC ) and financial firms (SIC ). We note that doing so does not change our main conclusions (results available on request). To construct measures of asymmetric information, we merge our initial sample with Institutional Brokers Estimate System (IBES). Due to the availability of Detailed History Files from IBES, our sample period is from 1983 to Our final sample is an unbalanced panel comprising 22,413 firm-year observations. A. Measures of Dividend Policies To explore the role of asymmetric information in dividend policy, we focus on quarterly regular dividends to common shareholders, the dividends with the greatest possible information content. 1 We thank an anonymous referee for pointing this out to us.

4 Li & Zhao Asymmetric Information and Dividend Policy 675 The dividend items in Compustat (e.g., data items 21 and 26) include nonregular dividend payments, such as special dividends and liquidation dividends, and thus they may not carry the same information content as predicted in the models of Bhattacharya (1979), Miller and Rock (1985), and John and Williams (1985). We note that using the dividend variables in Compustat to define our dependent variables has no material effect on our main conclusions. We follow Grullon, Michaely, Benartzi, and Thaler (2005) and Amihud and Li (2006) by using the CRSP database to identify dividend and nondividend payers. We collect all regular quarterly dividends on ordinary common stocks in the CRSP daily file (CRSP distribution code first digit = 1 (ordinary dividend); second digit = 2 (cash, US dollars); third digit = 3 (quarterly dividend); fourth digit = 2 (normal taxable at same rate as dividend)). After adjusting for changes in number of shares outstanding, we aggregate the quarterly dividends into an annual dividend amount. We set our first dividend variable, the payer dummy, equal to one for firm i in year t if the annual amount of dividends paid is positive, and zero otherwise. Our second dividend variable captures the initiation decisions of nondividend payers. For firm i in year t, we set the initiation dummy equal to one if this is the first time firm i pays dividends, and zero for all the years prior to year t. The dividend initiation sample includes only the firm-years until the non-dividend-paying firm makes its first dividend payment, or when the sample period ends, whichever comes earlier. Over the sample period, if a firm omits and then resumes dividend payments, our initiation dummy variable captures only the very first time that the firm initiated dividend payment. The decision that dividend payers have to make on a regular basis is whether or not to increase dividends. Lintner (1956) shows that dividends are sticky and firms usually are reluctant to cut or omit dividends. Thus, our next measure of dividend policy examines dividend increases by dividend payers. We set the increase of dummy equal to one for firm i in year t if the percentage increase in dividends is greater than 15%, and zero otherwise. To exclude any minor changes in the sample, we use a cutoff point of 15% when we define dividend increases. Our rationale is that if the signaling models hold, then we are more likely to find a negative relation between the quality of a firm s information environment and a large increase in dividends. Our main results are not sensitive to the level of cutoff used in defining the dividend increase dummy. We obtain our fourth dividend variable, dividend payout, by scaling the annual dividend amount by total assets. To ensure that our results are not driven by price variation or affected by the fact that a significant proportion of firms with negative earnings are paying dividends, we normalize the amount of dividends by book assets, instead of market capitalization or earnings following Allen and Michaely (2003). Table I provides summary statistics of our dividend policy variables. Column (1) shows that the proportion of dividend payers declines steadily over the sample period, starting at 80.0% in 1983 and reaching 30.7% in 2002, with a slight rebound in We note that the proportion of dividend payers is higher in our sample than in the one used by Fama and French (2001), suggesting that our sample firms are on average larger and more mature than the general population of firms covered in Compustat/CRSP. (We note that as a robustness check, we examine the effect of our sample selection criterion on our main results.) This difference is due to our sample requirement for the availability of analyst forecast data. Nonetheless, the same declining trend in the propensity to pay dividends is evident throughout most of our sample period. Column (2) in Table I reports the fraction of first-time payers in year t among surviving nondividend payers from year t 1. In our sample, the fraction of firms that initiate dividends starts at 7.3% in This measure drops steadily throughout most of the sample period and then rises again beginning in Column (3) shows that there is no apparent time trend in the fraction of dividend payers increasing dividends. Column (4) shows that the average dividend

5 676 Financial Management Winter 2008 Table I. Time Series Characteristics of Dividend Policy The sample period is from 1983 to We obtain accounting information from Compustat, dividend information from CRSP, and analyst forecasts from IBES. We define dividend payers as firms that pay quarterly dividends to common shareholders (CRSP four-digit distribution code = 1232) in year t. We define nondividend payers as firms that do not pay dividends in year t. Dividend initiation takes the value of one if the firm makes its first dividend payment in year t, and zero for all the years prior to t. Dividend increase takes the value of one if the percentage increase in dividends is greater than 15%. We present frequency counts for dividend payers, nondividend payers, and payers that increase dividends. Dividend payout is the ratio of annual aggregation of quarterly dividends paid to common shareholders to total assets at the end of year t measured in percentages. We present annual averages for this measure. Year (1) (2) (3) (4) Proportion Proportion Proportion Dividend of Dividend of Nondividend of Payers Payout Payers Payers Initiating Increasing Dividends Dividends payout appears to decline steadily over the sample period, from 2.12% in 1983 to 0.54% in 2002, before rising in The increasing use of dividends as cash payout toward the end of our sample period is probably partly due to the tax reform in 2003, after which most dividends were taxed at a lower 15% rate. B. Measures of Firms Information Environments We use analyst earnings forecast errors and the dispersion in analyst earnings forecasts to capture the quality of a firm s information environment. Elton, Gruber, and Gultekin (1984) show that a large fraction of analyst forecast error is attributable to misestimation of firm-specific factors rather than to misestimation of economy or industry factors. Their finding suggests that analyst forecast errors are a reasonable proxy for the degree of information asymmetry about the firm.

6 Li & Zhao Asymmetric Information and Dividend Policy 677 The dispersion in analyst earnings forecasts represents the dispersion among analysts about a consensus estimate of the forecast. Since disagreement among analysts is an indication of a lack of available information, we use this standard deviation as another metric of the degree of information asymmetry for a firm. We define analyst earnings forecast error as the absolute value of the difference between the mean earnings forecast and actual earnings, divided by the absolute value of actual earnings. Dispersion of analyst earnings forecast is the standard deviation of the earnings forecast scaled by the absolute value of the mean earnings forecast. We require our sample firms to have both of these measures available. We have one reservation regarding our use of analyst forecast errors and forecast dispersion as measures of asymmetric information, which is that forecast errors and dispersion might not so much capture asymmetries in information as levels of uncertainty that are common to both managers and outside investors. For example, our measures might pick up a more risky environment for the firm, implying a greater deviation than what we expected (and a larger variance of such deviations). We argue that this concern does not pose any serious problems for our analysis. First, this is because other studies show that our measures for information asymmetry do capture dimensions beyond firm risk. Ajinkya, Atiase, and Gift (1991) and Lang and Lundholm (1993, 1996) show that as firms enhance information disclosure, analyst earnings forecast accuracy increases while forecast dispersion decreases. Bowen, Davis, and Matsumoto (2002) show that conference calls improve analyst forecast precision and reduce forecast dispersion, and Chen and Matsumoto (2006) find that better access to management is associated with more accurate analyst forecasts. Second, the concern about risky environments does not pose serious problems for our analysis because the positive correlation between firm risk and our two measures for asymmetric information is quite low (to be shown later). And to further lessen this concern, we control for firm risk in all of our regression specifications. Thus, our results are not contaminated by the commonality between information asymmetry and uncertainty, which is captured by firm risk. Panel A of Table II reports summary statistics for our two measures of asymmetric information. The mean (median) analyst earnings forecast error is 21.7% (3.8%) of actual earnings, but the mean (median) analyst forecast dispersion is 14% (3.2%) of the mean earnings forecast. The large difference between mean and median values suggests that the distributions of these two measures are highly skewed. Panel B presents summary statistics grouped by firm dividend policies. We find that both measures of asymmetric information are significantly lower for dividend payers than are those observed for nondividend payers. In addition, nondividend payers who initiate dividends and dividend payers with above-median payouts have lower forecast errors and forecast dispersion than do nondividend payers who do not initiate dividends and dividend payers with below-median payout, respectively. The univariate results suggest a negative association between the degree of information asymmetry and dividend policies. C. Other Firm Characteristics We also control for other firm characteristics that may affect a firm s dividend policy: size, growth potential (the market-to-book ratio (M/B ratio), and asset growth), profitability, and firm risk. Fama and French (2001) show that firms paying dividends are usually larger, with lower growth potential and higher cash flows. We add firm risk because Grullon, Michaely, and Swaminathan (2002), Hoberg and Prabhala (2008), and Bulan, Subramanian, and Tanlu (2007) suggest that firms pay dividends as a signal of firm maturity and declining risk.

7 678 Financial Management Winter 2008 We follow Fama and French (2001) in the construction of our first four variables that describe firm characteristics. We define profitability as earnings before extraordinary items (data 18) + interest expense (data 15) + income statement deferred taxes (data 50, if available)/total assets (data 6). We use both the M/B ratio and asset growth as growth opportunity measures. We define the M/B ratio as the ratio of the market value of total assets to the book value of total assets. We define the market value of total assets as the market value of equity plus the book value of total assets minus the book value of equity, and the book value of equity is defined as stockholders equity (data 216) or common equity (data 60) + preferred stock par value (data 130) or total assets (data 6) total liabilities (data 181), plus balance sheet deferred taxes and investment tax credit (data 35, if available) and postretirement benefit liabilities (data 330, if available), minus the book value of preferred stocks (estimated in the order of the redemption (data 56), liquidation (data 10), or par value (data 130), depending on availability). We define firm size as the annual percentile of market capitalization and use NYSE firms to calculate cutoff points. We do so to neutralize any effects of the growth in typical firm size through time, with the largest (smallest) firm taking the value of one (0.01). For firm risk, we follow Hoberg and Prabhala (2008) by using the standard deviation of residuals from a regression of firm daily stock returns on returns of the market portfolio. Our main results remain the same if we use the standard deviation of daily returns or the standard derivation of residuals from a regression of daily excess returns on the three Fama and French (1992) factors. Panel A of Table II presents the summary statistics for firm characteristics. We show that the mean (median) profitability of our sample firms is about 7.1% (9.6%), and the mean (median) M/B ratio is 1.99 (1.49). The mean (median) growth rate of assets is 23.6% (10.1%), suggesting a highly skewed distribution for asset growth among sample firms. The mean (median) firm risk is 2.79% (1.37%). Summary statistics of firm size suggest that on average, our sample firms are slightly smaller than the median NYSE firm but larger than the average publicly traded firm. In terms of the risk measure, our sample firms are less risky than an average public firm as examined in Hoberg and Prabhala (2008). The standard deviations indicate that there are large variations across firms. In Panel C, Table II, we report the pairwise correlations between firm characteristics and the asymmetric information measures. The two asymmetric information measures have a correlation of 0.37, suggesting that when analysts cannot agree on a firm s earnings forecast, they are less likely to provide accurate forecasts. Neither of the asymmetric information measures is highly correlated with the firm characteristics that we find are important determinants of dividend policy. In particular, the correlations between firm risk and the two measures of information asymmetry are below This result confirms that there is some overlap between firm risk and our measures of information asymmetry. However, it also indicates that the extent of overlap is limited, which suggests that our two measures do pick up aspects of a firm s information environment that are not captured by firm risk. Thus, our two measures are more likely to be exogenous proxies for asymmetric information, implying that our model specification should be a relatively clean test of the relation between information asymmetry and dividend policy. II. Main Results Given that most of our analyses involve panel data, our estimates are based on robust standard errors. We estimate these errors by assuming independence across firms, but we account for possible autocorrelation within the same firm. The robust standard errors are frequently much larger than conventional estimates, which assume independence among firm-year observations,

8 Li & Zhao Asymmetric Information and Dividend Policy 679 Table II. Summary Statistics The sample period is from 1983 to We obtain accounting information from Compustat, dividend information from CRSP, and analyst forecasts from IBES. We define profitability as earnings before extraordinary items (data 18) + interest expense (data 15) + income statement deferred taxes (data 50, if available)/total assets (data 6). The market-to-book (M/B) ratio is the ratio of the market value of total assets to the book value of total assets. Asset growth is the rate of growth of total assets. Firm size is the NYSE market capitalization percentile. Firm risk is the standard deviation of residuals from the market model measured in percentages. We define forecast error as the absolute value of the difference between mean analyst earnings forecasts and actual earnings, divided by the absolute value of actual earnings. We define forecast dispersion as the standard deviation of analyst earnings forecast scaled by the absolute value of the mean earnings forecast. Panel A presents summary statistics of firm characteristics and measures of firms information environment. Panel B presents summary statistics of measures of firms information environment for firms with different dividend characteristics. Panel C presents a correlation matrix of firm characteristics and measures of firms information environment. p-values appear in parentheses. Panel A. Firm Characteristics and Information Environment Mean Median Standard 25th 75th Deviation Percentile Percentile Profitability M/B ratio Asset growth Firm size Firm risk Forecast error Forecast dispersion Panel B. Measures of Firms Information Environment Grouped by Dividend Policy Mean Standard 25th Median 75th Deviation Percentile Percentile Forecast error Dividend payers Nondividend payers Difference p-value <0.001 Nondividend payers initiating dividends Other nondividend payers Difference p-value Dividend payers with above-median payouts Dividend payers with below-median payouts Difference p-value <0.001 Forecast dispersion Dividend payers Nondividend payers Difference p-value <0.001 Nondividend payers initiating dividends Other nondividend payers Difference p-value 0.001

9 680 Financial Management Winter 2008 Table II. Summary Statistics (Continued) Panel B. Measures of Firms Information Environment Grouped by Dividend Policy (Continued) Mean Standard 25th Median 75th Deviation Percentile Percentile Dividend payers with above-median payouts Dividend payers with below-median payouts Difference p-value <0.001 Panel C. The Correlation Matrix Profitability M/B Asset Firm Firm Forecast Ratio Growth Size Risk Error M/B ratio [0.003] Asset growth [0.011] [<0.001] Firm size [<0.001] [<0.001] Firm risk Forecast error [<0.001] [<0.001] [0.499] [<0.001] [<0.001] Forecast dispersion [<0.001] [<0.001] [0.033] so our significance tests are not inflated by the large number of firm-year observations in our sample. To address any potential concerns that the fraction of firms paying dividends exhibits a strong time trend and might be influenced by industry-specific factors, we include year and industry dummies in our estimation. For each measure of dividend policy, we present results that use three specifications involving different combinations of our measures for asymmetric information: analyst earnings forecast errors only, the dispersion in analyst forecasts only, and both measures together. We include the same set of firm characteristics in all specifications. Our main model specification is as follows: Dividend Policy it = α 0 + f ind + f t + β 1 Profitability it + β 2 M/B Ratio it + β 3 Asset Growth it + β 4 Firm Size it + β 5 Firm Risk + β 6 Information Asymmetry it + e it, (1) where the dependent variable can be any of our dividend policy measures: payer dummy, initiation dummy, increase dummy, and dividend payout. For the first three measures, we use the logistic regression; for the last measure, we run panel data OLS tests. We note that running the payout regression under the tobit model in a panel data setting gives qualitatively similar results. In Table III, Panel A, we present the logistic regression results from our examination of the likelihood of making dividend payments. This panel shows that our results on firm characteristics support the findings in Fama and French (2001): larger firms with higher profitability and lower

10 Li & Zhao Asymmetric Information and Dividend Policy 681 growth potential are more likely to pay dividends. Moreover, we show that risky firms are less likely to pay dividends. This finding confirms the result in Grullon et al. (2002) and Hoberg and Prabhala (2008). More importantly, after controlling for the usual determinants of a firm s propensity to pay dividends, our results show negative coefficients on both measures of information asymmetry, which suggests that firms in a poorer information environment are less likely to pay dividends. This evidence does not support the signaling models of dividends. In Panel B, we report the results from the logistic regressions that we use to examine the nondividend payers decisions to initiate dividends. Similar to our results on the decision to pay, we find that larger, more profitable firms are more likely to initiate dividend payments. The propensity to initiate dividends is negatively associated with the M/B ratio. After we control for the M/B ratio, we find that the asset growth rate is positively associated with the propensity to initiate dividends. Also, risky firms are less likely to initiate dividends, a result that is consistent with the maturity and risk argument. Both measures of asymmetric information are negatively associated with the likelihood of dividend initiation. Panel C presents regression results from our examination of the decision to increase dividends among payers. We find that more profitable firms are more likely to increase dividends. The positive effect of M/B ratios on the likelihood of increasing dividends appears to contradict the growth opportunity argument. However, this result can be explained by the dual role played by the M/B ratio. Fama and French (2002) suggest that the M/B ratio is a measure of both profitability and growth potential. It is likely that the M/B ratio is more a measure of profitability than a measure of growth opportunities among dividend-paying firms. Both measures of asymmetric information are negatively associated with the likelihood of increasing dividends. In Panel D, we examine the determinants of the level of dividend payout. We show that larger, more profitable firms with lower risk pay more cash dividends. Consistent with the findings from the other panels, both asymmetric information measures are negatively related to the amount of dividends paid. Our findings lead us to conclude that there is a negative relation between asymmetric information and measures of dividend policy. Our results do not support the signaling theory of dividends. We note that using insider returns as a proxy for information asymmetry, Khang and King (2006) show that the amount of dividends is negatively related to returns to insider trades across firms. They thus conclude that their results do not support the signaling theory of dividends either. III. Additional Investigation Here, we address other possibilities that may lead to our results. First, we ask if our sample construction, which requires firms to have data available on analyst earnings forecasts, could systematically bias our findings. Second, we ask if a significant part of our results could be explained by the increasing use of share repurchases as a form of payout. Third, we ask how sensitive are our results to other factors that have been suggested in the literature to explain dividend policy, such as institutional monitoring and catering. A. Sample Selection As mentioned before, our sample firms are different from the general population covered in Compustat/CRSP as examined in Fama and French (2001). So the important question is, does this sample difference drive the results? 2 2 We thank an anonymous referee for suggesting this analysis to us.

11 682 Financial Management Winter 2008 Table III. Explaining Dividend Policy The sample period is from 1983 to We obtain accounting information from Compustat, dividend information from CRSP, and analyst forecasts from IBES. We define profitability as earnings before extraordinary items (data 18) + interest expense (data 15) + income statement deferred taxes (data 50, if available)/total assets (data 6). The market-to-book (M/B) ratio is the ratio of the market value of total assets to the book value of total assets. Asset growth is the rate of growth of total assets. Firm size is the NYSE market capitalization percentile. Firm risk is the standard deviation of residuals from the market model measured in percentages. We define forecast error as the absolute value of the difference between mean analyst earnings forecasts and actual earnings, divided by the absolute value of actual earnings. We define forecast dispersion as the standard deviation of analyst earnings forecast scaled by the absolute value of the mean earnings forecast. The dependent variable in Panel A is the payer dummy set equal to one for firm i in year t if the annual amount of dividends paid is positive, and zero otherwise. The dependent variable in Panel B is the initiation dummy set equal to one if this is the first time firm i pays dividends, and zero for all the years prior to year t. The dependent variable in Panel C is the increase dummy set equal to one for firm i in year t if the percentage increase in dividends is greater than 15%, and zero otherwise. The dependent variable in Panel D is dividend payout, which we define as the ratio of annual aggregation of quarterly common dividends obtained from CRSP to total assets measured in percentages. The estimation includes industry and year dummies. We base the reported p-values on White (1980) heteroskedasticity-consistent standard errors, adjusted to account for possible correlation within a (firm) cluster. Panel A. The Decision to Pay Dividends (1) (2) (3) Profitability M/B ratio Asset growth [0.018] [0.018] [0.018] Firm size Firm risk Forecast error [0.088] [0.200] Forecast dispersion [0.075] [0.151] Intercept Number of observations 22,413 22,413 22,413 Pseudo R Panel B. The Decision to Initiate Dividends (1) (2) (3) Profitability M/B ratio Asset growth [0.020] [0.021] [0.029] Firm size [0.021] [0.015] [0.019]

12 Li & Zhao Asymmetric Information and Dividend Policy 683 Table III. Explaining Dividend Policy (Continued) Panel B. The Decision to Initiate Dividends (Continued) (1) (2) (3) Firm risk Forecast error [0.025] [0.060] Forecast dispersion [0.082] [0.182] Intercept [0.004] [0.005] [0.007] Number of observations 10,642 10,642 10,642 Pseudo R Panel C. The Decision to Increase Dividends (1) (2) (3) Profitability M/B ratio [0.003] [0.002] [0.002] Asset growth [0.276] [0.278] [0.281] Firm size [0.155] [0.134] [0.152] Firm risk Forecast error [0.054] [0.213] Forecast dispersion [0.016] [0.040] Intercept Number of observations 10,631 10,631 10,631 Pseudo R Panel D. The Decision on the Amount of Dividends (1) (2) (3) Profitability [0.004] [0.006] [0.007] M/B ratio [0.014] [0.015] [0.016] Asset growth [0.243] [0.238] [0.238] Firm size Firm risk

13 684 Financial Management Winter 2008 Table III. Explaining Dividend Policy (Continued) Panel D. The Decision on the Amount of Dividends (Continued) (1) (2) (3) Forecast error [0.001] [0.146] Forecast dispersion [<0.001] [<0.001] Intercept Number of observations 22,413 22,413 22,413 Adjusted R We add to the baseline model in Table III, a dummy variable, no analyst coverage, which we set equal to one for firms without analyst coverage, and zero otherwise. By doing so, we can run regressions on the entire Compustat/CRSP population, including firms with no analyst following. Since Alford and Berger (1999) and Hong, Lim, and Stein (2000) show that greater analyst coverage may be associated with several firm characteristics, such as lower firm risk and less information asymmetry, we opt not to use analyst coverage as the measure for asymmetric information in our main analysis. We report the results in Table IV. We find that the no analyst coverage dummy is significantly and negatively associated with a firm s decision to pay dividends and a nondividend payer s decision to initiate dividends. However, the regression results still fail to show a positive relation between dividend policies and the two information asymmetry measures. In fact, there is a significant negative association between our information asymmetry measures and firms decisions to initiate dividends, to increase dividends, and how much to pay. Thus, we conclude that the negative relation between asymmetric information and dividend policy is unlikely to be driven by the different composition between our sample of firms and the Compustat/CRSP population. Our evidence does not support the signaling theory of dividends. B. Share Repurchases So far, we have not considered other forms of cash payout that firms might use. However, we could argue that a firm s dividend policy may be affected by its repurchase activities. To investigate this conjecture further, we use two alternative specifications. First, we examine the relation between measures of asymmetric information and dividends based on the following model, where we control for the amount of repurchases: Dividend Policy it = α 0 + f ind + f t + β 1 Profitability it + β 2 M/B Ratio it + β 3 Asset Growth it + β 4 Firm Size it + β 5 Information Asymmetry it + β 6 Repurchase Amount it + e it. (2) We compute the amount of share repurchases using the measure suggested by Fama and French (2005). The repurchase amount is the product of the change in the split-adjusted number of shares and the average of split-adjusted share prices at the beginning and the end of the year, normalized by total assets:

14 Li & Zhao Asymmetric Information and Dividend Policy 685 Table IV. Sample Selection The sample period is from 1983 to To assess the impact of sample selection criterion on our main results, we expand the sample to include all firms from the Compustat/CRSP merged file. For firms without information on analyst forecasts, we assign zero to their forecast errors and forecast dispersion. We also add to the regression model the no analyst coverage dummy, which we set equal to one for firms without any analyst coverage, and zero otherwise in year t. We define profitability as earnings before extraordinary items (data 18) + interest expense (data 15) + income statement deferred taxes (data 50, if available)/total assets (data 6). The market-to-book (M/B) ratio is the ratio of the market value of total assets to the book value of total assets. Asset growth is the rate of growth of total assets. Firm size is the NYSE market capitalization percentile. Firm risk is the standard deviation of residuals from the market model measured in percentages. We define forecast error as the absolute value of the difference between mean analyst earnings forecasts and actual earnings, divided by the absolute value of actual earnings. We define forecast dispersion as the standard deviation of analyst earnings forecast scaled by the absolute value of the mean earnings forecast. The dependent variable in Column (1) is the payer dummy, set equal to one for firm i in year t if the annual amount of dividends paid is positive, and zero otherwise. The dependent variable in Column (2) is the initiation dummy, set equal to one if this is the first time firm i is paying dividends, and zero for all the years prior to year t. The dependent variable in Column (3) is the increase dummy, set equal to one for firm i in year t if the percentage increase in dividends is greater than 15%, and zero otherwise. The dependent variable in Column (4) is dividend payout, which we define as the ratio of annual aggregation of quarterly common dividends obtained from CRSP to total assets measured in percentages. The estimation includes industry and year dummies. We base the reported p-values on White (1980) heteroskedasticity-consistent standard errors adjusted to account for possible correlation within a (firm) cluster. (1) (2) (3) (4) Decision Decision Decision Decision on to Pay to Initiate to Increase the Amount Dividends Dividends Dividends of Dividends Profitability [0.007] M/B ratio [0.660] Asset growth [<0.001] [0.210] [0.026] [0.270] Firm size [<0.001] [<0.001] [0.576] [<0.001] Firm risk Forecast error [0.126] [0.024] [0.034] [<0.001] Forecast dispersion [0.488] [0.073] [0.053] [<0.001] No analyst coverage [<0.001] [<0.001] [0.491] [0.196] Intercept Number of observations 70,874 46,788 19,981 70,874 Pseudo/adjusted R Repurchase Amount it = [(Adj. Shares it Adj. Shares it 1 ) (Adj. Price it 1 + Adj. Price it )/2]/Total Assets it. (3) After controlling for the amount repurchased, we report the results in Table V, whose four panels correspond to the panels in Table III.

15 686 Financial Management Winter 2008 Table V. Repurchase, Institutional Ownership, and Catering The sample period is from 1983 to We obtain accounting information from Compustat, dividend information from CRSP, and analyst forecasts from IBES. We define profitability as earnings before extraordinary items (data 18) + interest expense (data 15) + income statement deferred taxes (data 50, if available)/total assets (data 6). The market-to-book (M/B) ratio is the ratio of the market value of total assets to the book value of total assets. Asset growth is the rate of growth of total assets. Firm size is the NYSE market capitalization percentile. Firm risk is the standard deviation of residuals from the market model measured in percentages. We define forecast error as the absolute value of the difference between mean analyst earnings forecasts and actual earnings, divided by the absolute value of actual earnings. We define forecast dispersion as the standard deviation of analyst earnings forecast scaled by the absolute value of the mean earnings forecast. The repurchase amount is the product of the split-adjusted change in shares outstanding and the average of the split-adjusted stock price at the beginning and the end of the year, normalized by total assets and measured in percentages. Institutional ownership is the fractional share ownership by institutions. Dividend premium is the difference between log(m/b ratio) for dividend payers and the same measure for nondividend payers. The dependent variable in Panel A is the payer dummy, set equal to one for firm i in year t if the annual amount of dividends paid is positive, and zero otherwise. The dependent variable in Panel B is the initiation dummy, set equal to one if this is the first time firm i is paying dividends, and zero for all the years prior to year t. The dependent variable in Panel C is the increase dummy, set equal to one for firm i in year t if the percentage increase in dividends is greater than 15%, and zero otherwise. We define the dependent variable in Panel D, dividend payout, as the ratio of annual aggregation of quarterly common dividends obtained from CRSP to total assets measured in percentages. The estimation has industry and year dummies included. We base the reported p-values on White (1980) heteroskedasticity-consistent standard errors adjusted to account for possible correlation within a (firm) cluster. Panel A. The Decision to Pay Dividends (1) (2) (3) (4) Profitability M/B ratio Asset growth [0.066] [0.005] [0.018] [0.052] Firm size Firm risk Forecast error [0.151] [0.450] [0.943] [0.581] Forecast dispersion [0.150] [0.187] [0.406] [0.189] Repurchase amount [0.009] [0.010] Institutional ownership [0.002] [0.001] Dividend premium [<0.001] [<0.001] Intercept Number of observations 22,413 22,413 19,295 19,295 Pseudo R

16 Li & Zhao Asymmetric Information and Dividend Policy 687 Table V. Repurchase, Institutional Ownership, and Catering (Continued) Panel B. The Decision to Initiate Dividends (1) (2) (3) (4) Profitability M/B ratio Asset growth [0.052] [0.192] [0.003] [0.022] Firm size [0.017] [0.002] [0.388] [0.065] Firm risk Forecast error [0.063] [0.075] [0.100] [0.090] Forecast dispersion [0.190] [0.175] [0.227] [0.211] Repurchase amount [0.392] [0.143] Institutional ownership [0.119] [0.005] Dividend premium [0.806] [0.719] Intercept [0.006] [<0.001] [<0.001] [0.003] Number of observations 10,642 10,642 8,703 8,703 Pseudo R Panel C. The Decision to Increase Dividends (1) (2) (3) (3) Profitability M/B ratio [0.005] [0.001] [0.001] [0.002] Asset growth [0.373] [0.262] [0.315] [0.384] Firm size [0.166] [0.088] 0.417] [0.250] Firm risk Forecast error [0.241] [0.371] [0.545] [0.503] Forecast dispersion [0.045] [0.019] [0.031] [0.026] Repurchase amount [0.004] [0.015] Institutional ownership [0.278] [0.106] Dividend premium [0.494] [0.514] Intercept Number of observations 10,631 10,631 9,648 9,648 Pseudo R

17 688 Financial Management Winter 2008 Table V. Repurchase, Institutional Ownership, and Catering (Continued) Panel D. The Decision on the Amount of Dividends (1) (2) (3) (4) Profitability [0.018] [0.001] [<0.001] [<0.001] M/B ratio [0.007] [0.061] [0.037] [0.021] Asset growth [0.315] [0.210] [0.365] [0.471] Firm size Firm risk Forecast error [0.103] [0.231] [0.470] [0.708] Forecast dispersion [<0.001] [<0.001] [0.018] [<0.001] Repurchase amount [0.037] [0.052] Institutional ownership [<0.001] [<0.001] Dividend premium [0.410] [0.072] Intercept Number of observations 22,413 22,413 19,295 19,295 Adjusted R Column (1) of Table V presents the results based on the specification in Equation (2). The results still show a negative association between the measures of asymmetric information and dividend policy. Moreover, we observe a significant, positive relation between firms decisions to pay dividends and the amount of repurchases, and between the amount of dividends and the amount of repurchases. This finding confirms previous evidence in Fama and French (2001) and Grullon and Michaely (2002) that repurchases are primarily made by dividend-paying firms, and that as a result, repurchasing firms are more likely to be dividend payers. It also confirms that such firms pay more cash dividends. Unlike previous studies, we show that there is a negative relation between the amount of repurchases and the likelihood that dividend payers will increase dividends. That is, when firms repurchase more, they are less likely to increase dividends (see Column (1) of Panel C). When we replace the repurchase amount with the repurchase dummy, which we set equal to one when the repurchase amount is positive, and zero otherwise, our main results on the relation between information asymmetry and dividend policy are unchanged. Second, we use a seemingly unrelated regressions (SUR) model in which we jointly estimate the amount of dividends paid and the amount repurchased. In unreported results, we find that our earlier results do not change: firms that are less subject to the problem of information asymmetry pay a larger amount of dividends. Thus, after accounting for firms contemporaneous repurchase activities, we conclude that the negative relation between measures of asymmetric information and dividend policy is robust. This finding does not support the signaling theory of dividends.

18 Li & Zhao Asymmetric Information and Dividend Policy 689 C. Institutional Shareholdings Allen, Bernardo, and Welch (2000) present a model in which they use dividends to attract betterinformed, monitoring, institutional shareholders. Their theory predicts a positive correlation between dividends and institutional shareholdings. To explore whether our results are driven by institutional monitoring, we control for institutional holdings in our regression specifications, and report the results in Table V, Column (2). We find that measures of asymmetric information remain negatively related to firms dividend policies. Further, contrary to the monitoring argument in Allen et al. (2000), but mostly consistent with the empirical findings in Grinstein and Michaely (2005), we show that, after we control for firm risk, firms with higher institutional shareholdings are less likely to pay dividends and are associated with lower dividend payouts. It is clear that the negative relation between asymmetric information and dividend policy is not explained by the presence of (monitoring) institutional shareholders. Again, our results fail to lend support to the signaling theory of dividends. D. The Catering Theory of Dividends Using aggregate data, Baker and Wurgler (2004) develop their catering theory of dividends. They find that investor demand for dividend-paying stocks is time-varying. Managers cater to investor demand for dividends by paying dividends when investors place a premium on dividendpaying stocks, and vice versa. We use the dividend premium measure provided in Baker and Wurgler (2004), which they define as the difference in the value-weighted average M/B ratio of payers and the value-weighted average M/B ratio of nondividend payers. We add this measure to Equation (1). Column (3) of Table V presents the results. (We note that because the sample in Baker and Wurgler s, 2004, study ends in 2000, the sample size with the catering measure is smaller.) We find that adding the dividend premium into our model specification has no material effect on the role of asymmetric information in dividend policy. Moreover, the coefficient estimate of dividend premium contradicts the argument in Baker and Wurgler (2004). We show that this finding is mainly due to our inclusion of the year dummies and firm risk. Once we remove these dummies and the risk variable, the coefficient on dividend premium is significant and positive. This result is consistent with Baker and Wurgler s argument that the dividend premium primarily captures the temporal variation in market sentiment. Column (4) of Table V presents our results when we use all additional dividend factors. It is clear that our main results on asymmetric information do not change with this expanded model specification. Thus, we conclude that the negative relation between information asymmetry and dividend policy is not driven by other factors that may affect a firm s dividend policy. And again, our evidence does not support the signaling theory of dividends. IV. Repurchase and Total Payout Although our main focus is on the relation between information asymmetry and firms dividend policies, we also examine whether information asymmetry is an important consideration for repurchases. Vermaelen (1984), Ofer and Thakor (1987), and McNally (1999) extend the models in Bhattacharya (1979) and Miller and Rock (1985) to repurchases, suggesting that the signaling motive may also determine firms repurchase decisions. However, the inherent inflexibility in dividends implies that dividends have stronger informational content than do repurchases. Thus, if the signaling models are valid, we expect to find a weaker (less positive or more negative)