# Does dividend policy relate to cross-sectional variation in information asymmetry? Evidence from returns to insider trades.

We examine the relation between dividends and information asymmetry by using insider returns as a proxy for information asymmetry. We find that dividends are negatively related to returns to insider trades across firms. Firms that pay consistently high dividends have lower insider returns than do firms that pay consistently low dividends. These results do not support traditional dividend signaling models. Rather, they are consistent with the proposition that firms with the highest dividends have the lowest levels of information asymmetry.Why do some firms pay a higher level of dividends than others? Although past research provides us with a greater understanding of dividend policy, there is ample evidence we still have a long way to go (Brav, Graham, Harvey, and Michaely, 2005).

The purpose of our article is to add to earlier research by examining the relation between the level of dividend payout and information asymmetry. To do so, we use returns to insider trades as a proxy for the level of information asymmetry between insiders and outside investors. Examining this question is important because it provides evidence on whether asymmetric information motivates dividend policy.

To date, the evidence on this question is mixed and related mainly to tests of dividend signaling models. Several studies examine short-term stock price reactions to dividend changes and find that stock prices initially react as predicted by dividend signaling models (Aharony and Swary, 1980; Asquith and Mullins, 1983; Healy and Palepu, 1988; and Michaely, Thaler, and Womack, 1995). However, other studies show that stock prices continue to drift for two to three years after the dividend change (Charest, 1978; Michaely, et al. 1995; and Grullon, Michaely, and Swaminathan, 2002), indicating that dividends may contain information, but are not an effective signal. Other studies examine whether dividend changes are positively related to future earnings changes. Again the evidence is mixed. Some studies find evidence supporting a relation (Watts, 1973; Gonedes, 1978; Healy and Palepu, 1988; and Brooks, Charlton, and Hendershott, 1998). Other studies find evidence to the contrary (Penman, 1983; DeAngelo, DeAngelo, and Skinner, 1996; Benartzi, Michaely, and Thaler, 1997; Grullon, et al. 2002; and Grullon, Michaely, Benartzi and Thaler, 2005).

Using our approach, we find an inverse relation between dividend payout and returns to insider trades. This relation provides evidence that is not consistent with traditional dividend signaling models, because firms with the highest dividend payout also have the lowest levels of information asymmetry to begin with. If firms use dividend policy as a signal, then we should expect that those firms with the greatest information asymmetry will try to ameliorate the problem by paying dividends. Instead, we find that the information asymmetry between firm insiders and their outside investors appears to be lower for firms with greater dividend payouts. Thus, our evidence is consistent with recent studies that indicate that managers may set dividend policy for other reasons, such as agency issues, dividend clienteles, etc. (Koch and Shenoy, 1999; Grullon, et al. 2002; and Grullon, et al. 2005). Finally, our evidence is also consistent with research that shows that the firms with the highest dividend payout also tend to be large, mature firms with lots of free cash flow and few growth opportunities (Allen and Michaely, 2004). These firms are the least likely to face substantial information asymmetry.

To the best of our knowledge, our measure, which uses returns to insider trades as a measure of the level of information asymmetry between managers and their outside investors, is unique. It is a widely accepted notion that corporate insiders often possess and trade on information that is not available to outside investors. This information asymmetry gives insiders the ability to identify and take advantage of mispricing in their own firms' shares.

We focus on measuring the level of information asymmetry that allows insiders to benefit through their trading. Using returns to insider trades as a metric for asymmetric information has the advantage that these returns measure the per share gain to insider trades. Consequently, we view returns to insider trades as a measure of the economic content of the information asymmetry between inside management and outside investors. If there is a perfectly elastic demand curve for shares, then the greater the change in share price to new information, the greater is the change in value of the firm's equity.

Other finance and accounting studies have used other proxies for information asymmetry. For example, some papers use the size of the bid-ask spread or the size of the information asymmetry component of the spread (e.g. Demsetz, 1968, and Brennan and Subrahmanyam, 1995). However, the spread reflects information asymmetry among investors, not between inside managers and outside investors. Other studies use the number of financial analysts, the dispersion in analyst forecasts, or accounting measures, such as the book-to-market ratio or the ratio of intangible assets to total assets. Huddart and Ke (2004) present evidence that institutional ownership, the number of analysts following a firm, the book-to-market ratio, and the frequency with which firms report losses may not be good proxies for information asymmetry. We use some of these factors as control variables and others as additional explanatory variables.

The article proceeds as follows: Section I describes the sample. Section II presents the empirical results examining whether the level of dividends is related to asymmetric information. Section III concludes the article.

I. Data Sample

Our source of the insider trading data is the Ownership Reporting System of the Securities and Exchange Commission (SEC). The data set summarizes all insider transactions in publicly held firms that were reported to the SEC between January 1982 and December 2003. The data items include the date of each transaction, the classification of the insider, the type of transaction, and the number of shares traded. We use CRSP for stock returns and Compustat for annual dividend and firm characteristics.

Based on the information in the data set, we designate the following as insiders: officer (O), officer and beneficial owner (OB), officer and director (OD), officer and treasurer (OT), president (P), chairman of the board (CB), chief executive officer (CEO), chief financial officer (CFO), director (D), director and beneficial owner (DO), and officer, director and beneficial owner (H). The SEC provides the titles and codes. We include all transactions that are open market purchases and sales prior to May 1991. From May 1991 onward, the SEC does not distinguish between public and private purchases and sales, so we use both. We exclude other transactions, such as employees' obtaining shares through employee stock option exercises, because they are more likely to be motivated by liquidity and portfolio diversification considerations than by information.

To measure insider trading returns, we examine excess returns over the 21-trading-days following an insider transaction. We assume the transaction occurs on the morning of the transaction date, so the 21-trading-day period includes the day of the transaction. We use this horizon because the degree of asymmetry should be the greatest in the interval between the insider trade and when the trade becomes public information, assuming markets are semi-strong efficient. Although the SEC now requires that transactions be reported by the end of the second business day after the trade occurs, throughout most of our sample period, the SEC only required insiders to report their trades within ten days subsequent to the last day of the month in which the trade occurred. Thus, the 21-trading-day horizon represents a close approximation of this interval. We note that when we repeat all analyses by replacing the 21-trading-day horizon with a 28-trading-day horizon, we obtain similar results. The 28-trading-days corresponds to 40 days in calendar time, which is the maximum time between an insider trade and the date of filing with the SEC. This horizon represents the longest time until an insider trade becomes public record.

Table I reports summary statistics for our sample of insider transactions, using the measure of insider returns in excess of the CRSP value- and equal-weighted indexes. To be included, the transactions must meet two criteria. First, we must be able to match the firm CUSIP with a CRSP permanent number. Second, Compustat must contain annual total assets, total market value, cash dividends to common stock, long-term debt, R&D expense, and earnings per share for the firm.

In addition to using two indexes, we use three separate weighting schemes for the observations. Doing so allows us to examine whether the results are robust to reasonable alternative weightings of the observations when we calculate average insider returns. Weighted-by-shares weights each trade by the natural log of the number of shares traded relative to the natural log of the total number of shares traded by insiders on that particular day. In this and all tables we denote the market value of trade as MV. Weighted-by-MV weights each trade by the natural log of the market value of the trade relative to the natural log of the total MV of trades by insiders on that day. Note that we separate purchases from sales when creating the weights. Unweighted weights each trade equally. Thus, the first two weighting schemes give greater weight to larger trades.

Panel A reports summary statistics for the distribution of insider returns from purchase and sales transactions. Overall, the distributions of insider returns are higher using the value-weighted index compared with the distributions using the equal-weighted index. Using the value-weighted index, the medians for the purchase returns are 0.76% to 0.94%, and the medians for the sales returns are -0.21% to -0.67%. Using the equal-weighted index, the medians are close to zero for the purchase returns, but the median for the sales returns are between -1.58% and -1.94%.

Panels B and C display the results when we sort insider returns from purchases and sales by the natural log of the market value of equity (ME) and book-to-market ratio (BM). We sort the sample of insider transactions into ME and BM quintiles independently. Panel B shows the results using the value-weighted index. Panel C presents the results using the equal-weighted index. The results suggest there is a size (ME) and book-to-market (BM) effect in insider returns. For purchases, we see some evidence of a pattern similar to Fama and French (1992), who show that small stocks and high BM stocks have higher returns. This pattern in insider returns is especially true for ME, although it is not as evident in BM, which displays deviations from this pattern. The pattern is most evident for the unweighted results. For sales, the pattern is more mixed. Regardless, the results suggest we should control for size and book-to-market in subsequent analyses.

Table II presents summary statistics for our sample of insider transactions, which we sort into two groups based on the level of the firm's dividend prior to the time of the transaction. We rank the insider purchases and sales separately according to the firm's average dividend yield over the previous five years. We define the dividend yield in a given year as the total cash dividends paid over the market value of equity as of the end of the fiscal year.

We then partition the sample into two halves. We define high-dividend-paying transactions as those when the firm has a dividend yield above the sample median. Low-dividend-paying transactions are those when the firm has a dividend yield below the median. We note that among the low-dividend-paying transactions, there are 45,864 purchase observations and 123,273 sales observations where the firm did not pay any dividends in the previous five years. Panel A reports the number of insider purchases and sales. Panel B reports the number of firms.

Consistent with past studies on insider trading, the number of sales transactions (248,995) exceeds the number of purchase transactions (99,637), although the number of firms whose stock is being bought (6,305) is nearly the same as those whose stock is being sold (6,207). The larger number of sales transactions makes sense, because insiders have a greater variety of reasons to sell. For example, many employees receive shares through employee stock options and then sell those shares in open market transactions to rebalance or diversify their portfolios, or both. (We note that if an employee obtains shares through the exercise of employee options and then sells them in the open market, we include these open market sales in the sample, but we do not include obtaining shares through options exercises.) This greater number of reasons to sell also suggests that insider sales transactions may often not be motivated by information, but by other considerations. Jeng, Metrick, and Zeckhauser (2003) and Lakonishok and Lee (2001) find that insider trading gains are evident for insider purchase transactions, but not sales transactions. Thus, insider returns from purchases should be a less noisy proxy for information asymmetry than returns from insider sales. For this reason, we examine purchases and sales separately in the following analyses.

II. High- Compared to Low-Dividend-Paying Firms and Insider Returns

We now compare high- to low-dividend-paying insider returns. In Section A, we examine insider return differences when we divide the sample of transactions into those from high-dividend-paying firms and those from low dividend-paying firms. In Section B, we report differences in abnormal insider returns using the Fama-French three-factor model to control for the market, size, and book-to-market effects. In Sections C and D, we show the results from cross-sectional tests. In Section E we report time-series test results that use a monthly sampling frequency.

A. Insider Returns

Table III reports insider returns for high-dividend-paying transactions compared to low-dividend-paying transactions. As in Tables I and II, insider returns are the average market-adjusted insider return (raw stock return minus the CRSP value- or equal-weighted index return) from the date of the insider trade, where the averages are weighted-by-shares, weighted-by-MV, and unweighted.

Panel A displays the results using the CRSP value-weighted index. Panel B uses the CRSP equal-weighted index. The results indicate that on average, excess returns are higher following insider purchases than insider sales. These results are consistent with insiders having information about their firms' prospects that outsiders do not have. The stock price tends to rise by more after an insider purchase than after an insider sale. Further, this difference in average excess returns between insider purchases and sales is smaller for high-dividend-paying transactions, which is consistent with high-dividend-paying firms having lower information asymmetry.

For the 21-trading-day horizon, the average weighted-by-shares excess return over the CRSP value-weighted (equal-weighted) index following insider purchases exceeds that for insider sales by 1.95% (1.98%) for high-dividend-paying transactions compared to 3.98% (3.96%) for low-dividend-paying transactions. The weighted-by-MV and unweighted results are qualitatively similar. If insiders had no superior information, then the difference in average excess return between purchases and sales would be zero for both high- and low- dividend-paying transactions. The results suggest that the advantage of being an insider in a high-dividend-paying firm may be smaller relative to being an insider in a low-dividend-paying firm.

Another way we can compare the advantage of insiders in low- compared to high-dividend-paying firms is to examine the gains from insider purchases and sales separately. Table III shows that average insider returns on purchases are smaller for transactions from high-dividend-paying firms. For example, the weighted-by-shares average excess return over the CRSP value-weighted (equal-weighted) index from high-dividend-paying firms is 2.26% (1.82%) less than from low-dividend-paying firms. This difference has a t-statistic of-21.14 (-17.33). These results suggest that after purchasing stock, insiders in high-dividend-paying firms experience smaller gains than do those in low-dividend-paying firms. These results are consistent with high dividend-paying firms having lower information asymmetry. We find similar results when we use the weighted-by-shares and unweighted averages.

Similar comparisons for the sales transactions in Table III provide mixed results. When we use the CRSP equal-weighted index, the results are consistent with those of the purchase transactions, in that insiders in low-dividend-paying firms appear to have a greater advantage. We note that for sales, a lower or more negative return represents a greater gain to the seller who sells ahead of the poor return. However, we find the opposite result when we use the CRSP value-weighted index. The differing results between the two benchmarks may be due to size and book-to-market effects, and to the fact that the transactions are spread out across the sample period, as we discuss below. Regardless, for sales transactions there does not appear to be an overriding difference between average excess returns to high- and low-dividend-paying firms. This lack of difference is consistent with the proposition that insider sales may often be motivated by factors other than information.

The comparisons in Table III show a negative relation between average insider returns and past dividend yields for purchases, but no consistent relation for sales. Table I suggests that these results may be influenced by differences in size and book-to-market. In addition, the transactions in Table III are spread out across the sample period, which could present a problem if the CRSP value- or equal-weighted indexes are not appropriate benchmarks for measuring abnormal returns to insiders. For example, consider hypothetically the excess return relative to the CRSP value-weighted index. If insider transactions tend to be in smaller-capitalization stocks, then insider purchases and sales might be followed by positive excess returns over the early 1980s when small-cap stocks had higher average returns than large-cap stocks, but negative excess returns in the 1990s when large-cap stocks had higher average returns. If the ratio of purchases to sales varies over time, we may be comparing excess returns on transactions that took place in the 1980s, when there tend to be positive excess returns, with transactions that took place in the 1990s, when there tend to be negative excess returns. Thus, we consider the results in Table III to be descriptive.

B. Fama-French (1993) Abnormal Returns

To control for the market, size, and book-to-market factors, we calculate abnormal returns using the Fama-French (1993) three-factor model. For each transaction, we estimate the parameters to the Fama-French (1993) three-factor model by using consecutive 21-trading-day returns over the three years prior to the transaction. (1) Thus, for each transaction we estimate:

[R.sub.it] - [R.sub.ft] = [alpha] + [beta] ([R.sub.mt] - [R.sub.ft]) + [gamma] SM[B.sub.t]) + [lambda] HM[L.sub.t] + [[epsilon].sub.t], (1)

where [R.sub.it], is the 21-trading-day insider return; [R.sub.ft], is the 21-trading-day risk-free rate; a is the intercept; [beta] is the sensitivity to the market factor; [R.sub.mt], is the market return; [gamma] is the sensitivity to the size factor; SMB, is the Fama-French (1993) factor-mimicking portfolio for size; [lambda] is the sensitivity to the book-to-market factor; HM[L.sub.t] is the Fama-French (1993) factor-mimicking portfolio for the book-to-market factor; and [[epsilon].sub.t] is the regression error.

We then use the parameter estimates to calculate an expected return for the 21-trading-day period following the transaction. We obtain the abnormal return by subtracting the expected return from the actual return. Doing so yields:

[R.sup.ab.sub.it+1] = [R.sub.it+1] - [[R.sub.ft+1] + [beta] ([R.sub.mt+1] - [R.sub.ft+1]) + [gamma]SM[B.sub.t+1] + [lambda]HM[L.sub.t+1], (2)

where [R.sup.ab.sub.it+1] is the 21-trading-day insider abnormal return. The rest of the variables are analogous to those in Equation (1). We use the time subscript t+1 to emphasize that the estimation period for the parameters in Equation (1) lies immediately before the transaction.

Panel A of Table IV reports the results. We note that in Table IV, the reduced number of observations is due to the requirement that for the Fama-French (1993) estimation there must be three years of returns prior to the insider trade.

For purchases, the results in Table IV are qualitatively similar to those reported in Table III. For the 21-trading-day horizon, the weighted-by-shares average excess return from high-dividend-paying firms is 2.26% less than that from low-dividend-paying firms with a t-statistic of -17.4. The results are similar when we use the weighted-by-shares and unweighted averages and are consistent with high dividend-paying firms having lower information asymmetry. For sales transactions, the results are consistent with the purchase transactions in that high-dividend-paying firms appear to have lower information asymmetry. The weighted-by-shares average excess return from high-dividend-paying firms is 0.15% less than that from low-dividend-paying firms with a t-statistic of 2.11. As before, a lower or more negative return represents a greater gain to the seller who sells ahead of the poor return.

C. Cross-Sectional Tests

To analyze the difference in insider returns between high- and low-dividend-paying firms in a cross-sectional framework, we use the Fama and MacBeth (1973) method to examine whether the level of dividends is related to average insider returns. We run a set of monthly regressions for purchases and one for sales in which the dependent variable is weighted insider excess returns. We note that higher (lower) weighted insider returns on purchase (sales) transactions are consistent with greater information asymmetry.

In this analysis, we measure insider returns using the returns in excess of the CRSP value-weighted index over the 21-trading-days following the trade. (We note that as a robustness check, we also use raw returns and returns in excess of the CRSP equal-weighted index. The results are unchanged.) As before, we calculate the averages using weighted-by-shares, weighted-by-MV, and unweighted. We assign insider transactions to each month. All insider transactions that occur after the 20th day of the prior month and before the 21st day of the current month are assigned to the current month. We then run the two cross-sectional regressions each month to produce two time series of coefficients. We test the hypothesis that the expected coefficient is zero by forming t-statistics from the time series average of the monthly cross-sectional coefficients divided by the standard error of the mean. We calculate the standard error from the time series of coefficient estimates.

To test whether dividends are related to insider returns, we include the dividend yield D[Y.sub.i] as the first independent variable. This is our variable of interest and is defined as the average annual dividend yield (dividend over stock price) over the previous five years. In addition to the dividend yield, we include two sets of variables that other studies show may be related to average returns, dividends, or information asymmetry. We include these variables to hold "all else equal." (2)

The first set comprises the control variables of the book-to-market ratio for firm i from the previous fiscal year, BM; the natural log of the market value of equity as of the previous year's fiscal year-end, In(M[E.sub.i]); the variance of annual earnings per share over the previous five years, E[V.sub.i]; the debt ratio (long-term debt to total assets), D[R.sub.i]; and the natural log of the number of shareholders from the previous fiscal year, ln(S[H.sub.t]).

The second set comprises seven additional explanatory variables for information asymmetry. Although these variables may be unrelated to dividends, we include them as a precaution in case some are correlated with dividends. The addition of these variables ensures that omitted variable bias does not drive our results. If they are not correlated with dividends, the cost of including these additional variables is that they will reduce the efficiency of our estimators. However, the inclusion of these variables will provide evidence of robustness to the extent that the dividend yield maintains its significance even with their addition.

The first of the seven variables is R[D.sub.i], which is the average of the ratio of R&D expense to sales from the previous fiscal year. The next two, which capture differences in asymmetry due to intangible assets, are the ratio of intangible assets (other than goodwill) over total assets from the previous fiscal year, I[T.sub.i], and the ratio of tangible assets (defined as the ratio of plant, property, and equipment) over total assets, [T.sub.i]. I[T.sub.i] captures those cases in which firms capitalize the costs associated with intangible assets. However, since accounting standards recognize intangible assets only if they are purchased, but virtually always recognize tangible assets, we include [T.sub.i] to capture intangible assets for firms with unrecognized intangible assets. The fourth and fifth additional variables are E[X.sub.i], which is the sum of extraordinary items, discontinued operations, and special items over total sales from the previous fiscal year; and R[V.sub.i], which is the variance of daily stock returns over the previous calendar year. The last two variables are indicator variables. L[S.sub.i] is equal to one if earnings per share is negative in the previous fiscal year, and zero otherwise. [E.sub.i] is equal to one if the stock trades on the NYSE or Amex in the previous fiscal year, and zero otherwise.

We estimate three specifications. The first uses only the dividend yield. The second includes the dividend yield along with the five control variables. The third includes the dividend yield along with the control and additional explanatory variables. Thus, our cross-sectional regressions are as follows:

[R.sub.ki] = [alpha] + [[lambda].sub.dy] D[Y.sub.i] + [[epsilon].sub.ki] (3)

[R.sub.ki] = [alpha] +[[lambda].sub.dy] D[Y.sub.i] + [[lambda].sub.bm] B[M.sub.i] + [[lambda].sub.mv] ln(M[E.sub.i]) + [[lambda].sub.ev] E[V.sub.i] + [[lambda].sub.dr] D[R.sub.i] + [[lambda].sub.sh] ln(S[H.sub.i]) + [[epsilon].sub.ki] (4)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (5)

where [R.sub.ki] is the month t weighted insider return for day k. k is a day on which an insider transaction occurred for firm i. To classify the day as a purchase or sales day, we net the transactions on day k. Thus, if a firm has many days on which transactions take place, then there will be multiple observations for that firm in a given month. This method of netting and classifying the observations allows us to measure insider excess returns over the 21-trading-days following the transaction and to separate purchase days from sales days.

Panels B through G of Table IV present the results of our cross-sectional tests. Panels B and C report the weighted-by-shares results for all firms and the subsample of dividend-paying firms, respectively. Weighted-by-shares weights each trade by the natural log of the number of shares traded relative to the natural log of the total number of shares traded by insiders for that particular month. Panels D and E report the weighted-by-MV results for all firms and the subsample of dividend-paying firms, respectively. Weighted-by-MV weights each trade by the natural log of the market value of the trade relative to the natural log of the total market value of trades by insiders for that month. Note that we separate purchases from sales when creating the weights. Panels F and G report the unweighted results for all firms and the subsample of dividend-paying firms, respectively. Unweighted weights each trade day equally.

For purchase transactions, the mean coefficient on dividend yield is significant at the 1% or 5% level for each combination of weighting, sample, and model. For example, Panel B uses weighted-by-shares observations for the sample of all firms. The mean coefficient on dividend yield for Model 3 is -0.0011 with a t-statistic of -2.78, indicating that higher dividends are correlated with lower insider returns. This result is consistent with information asymmetry being negatively related to the level of dividends. Panel C shows that the results are not driven by non-dividend-paying firms, because the results are similar for the subsample of dividend-paying firms. In this case, Model 3 has a mean coefficient of **0.0029 with a t-statistic of -2.35.

For sales transactions, the mean coefficient on dividend yield is positive in all cases, but seldom significant. We note that we would expect a positive coefficient for sales transactions. Lower dividend yields should be associated with lower insider returns following a sale, because the insider return is what the insider avoids by selling his stock. The weak results for the sales regressions are consistent with insider sales being a poorer proxy for information asymmetry.

D. Other Cross-sectional Tests

We analyze two additional specifications to examine robustness. The first is that we control for the percentage of insider ownership, because insider gains may be closely related to the percent of the firm held by insiders. For example, firms with a greater percentage held by insiders are likely to experience more insider trading. (3) To account for this, we use the percentage of insider ownership from Compact Disclosure from 1990 through 2000 and re-estimate our regressions. We note that requiring insider ownership data results in the loss of about 20% of our sample transactions for the sample period. Using the sample of dividend-paying firms, the results show that insider ownership is not related to insider returns, although the mean coefficient on dividend yield for each model is still significant at the 10% level or better. (4)

The second specification accounts for possible endogeneity between dividend decisions made by management and insider gains. For example, if part of management's motivation is to maximize their own trading gains, then their dividend decisions may be influenced by their expected trading gains. Using the sample of dividend-paying firms, we use two-stage least squares to re-estimate our regressions in Equation (5). The results are similar and are available on request.

Overall, our results are consistent with a negative relation between dividends and information asymmetry. The results are also consistent with sales transactions often being motivated by considerations other than inside information.

E. Time Series Tests

Our final analysis is a time series test using monthly data and a monthly sampling frequency to construct four calendar-time portfolios. For a given month, the first portfolio HDP contains high-dividend-paying firms whose insiders are net purchasers of shares in the previous month; LDP is a portfolio of low-dividend-paying firms whose insiders are net purchasers in the previous month; HDS is a portfolio of high-dividend-paying firms whose insiders are net sellers of shares in the previous month; and LDS is a portfolio of low-dividend-paying firms whose insiders are net sellers of shares in the previous month.

We calculate the average returns to each portfolio for each month from January 1982 to December 2003 as follows. To calculate the return for a given month, we net the insider trades for each firm in the previous month and classify it as having insiders who are net purchasers or sellers. Then we separate the net purchase and net sales firms and rank them based on dividend yield. Those that appear above the median are high-dividend-paying firms, and those below the median are low-dividend-paying firms. In this way, we assign the firms to one of the four portfolios described above.

The monthly return for each portfolio in a given month is an average of the monthly excess returns for the firms in the portfolio. To obtain the average, we use the three weighting schemes described in Section I. To analyze the extent to which insider returns for high-dividend-paying firms are smaller than those for low-dividend-paying firms, we perform a time series test that uses the monthly version of the Fama-French (1993) three-factor model that we also use in Section IIB. The dependent variable in the regression for net purchases (sales) is the zero-cost, or long-short, portfolio that is long the HDP (HDS) portfolio and short the LDP (LDS) portfolio. Thus:

[R.sub.HDPt] - [R.sub.LDPt] = [alpha] + [beta]([R.sub.mt] - [R.sub.ft]) + [gamma]SM[B.sub.t] + [lambda]HM[L.sub.t] + [[epsilon].sub.t] (6)

[R.sub.HDSt] - [R.sub.LDSt] = [alpha] + [beta]([R.sub.mt] - [R.sub.ft]) + [gamma]SM[B.sub.t] + [lambda]HM[L.sub.t] + [[epsilon].sub.t], (7)

where [R.sub.XDYt], is the month t weighted return to the XDY portfolio (in which X = H,L and Y=P,S); a is the intercept, or abnormal return; [beta] is the sensitivity to the market factor; [R.sub.mt] is the market return; [R.sub.ft] is the risk-free rate; [gamma] is the sensitivity to the size factor; SM[B.sub.t] is the Fama-French (1993) factor-mimicking portfolio for size; [lambda] is the sensitivity to the book-to-market factor; HM[L.sub.t] is the Fama-French (1993) factor-mimicking portfolio for the book-to-market factor; and [[epsilon].sub.t] is the regression error. By using these regressions, we can examine the difference between insider returns to high-dividend-paying compared to low-dividend-paying firms.

We estimate both purchase and sales regressions over the time series of 264 monthly observations from January 1982 to December 2003. If we assume the Fama-French (1993) factors are risk factors, then [alpha] represents the average risk-adjusted return disadvantage of being an insider in a high-dividend-paying firm compared to a low-dividend-paying firm. (5) For example, if high-dividend-paying firms have lower insider returns than low-dividend-paying firms because they are larger in size, then using the size factor will control for this size effect. Similarly, we can capture the effects related to book-to-market ratios and the market factor by the corresponding factor-mimicking portfolios in the regressions. We note that the disadvantage, or lower returns, to high-dividend-paying firms would be captured as a negative a in the purchase regression and a positive a in the sales regression.

In Table V, for the three weighting methods, Panel A shows the average monthly insider return in excess of the CRSP value-weighted index. When we examine the weighted-by-shares results, we find that the average insider excess return for high-dividend-paying firms is 1.43% per month compared to 2.89% for low-dividend-paying firms. The difference of -1.46% is significantly different from zero with a t-statistic of -5.41. We also find that the proportion of months in which the insider excess returns of high-dividend-paying firms is smaller than low-dividend-paying firms is 63.3%. When we apply a binomial test for whether this proportion is greater than 50%, we obtain a test statistic of 4.31. The weighted-by-MV and unweighted results are similar. Thus, it appears that insiders in high-dividend-paying firms have less of an informational advantage than do their counterparts in low-dividend-paying firms.

For each weighting method the insider sales transactions show no difference between high- and low-dividend-paying firms. The difference in average excess return is not significantly different from zero. Again, this difference is consistent with the proposition that proportionately fewer of the insider sales transactions are information-based relative to insider purchases.

Panel B in Table V presents the Fama-French (1993) regression results for Equation (6). The Panel shows that the alphas for the purchase regressions are negative and significant. For example, when we examine the weighted-by-shares result, we find that [alpha] for the purchase regression is consistent with insiders in high-dividend-paying firms having less of an informational advantage. The monthly alpha for the disadvantage of high-dividend-paying insiders compared to low-dividend-paying insiders is -0.0141, or -1.41% per month. This difference has a t-statistic of -6.06.

The weighted-by-MV and unweighted results are similar. As expected the alphas for the sales regressions are nonsignificant. The results provide further evidence that these differences are not the result of the Fama-French (1993) factors of market, size, and book-to-market, and that the results are robust to reasonable weightings of the observations.

Finally, the coefficients on the market factor, SM[B.sub.t] and HM[L.sub.t] suggest that high-dividend-paying firms have lower betas, are larger, and are more value-oriented (higher BM) than low-dividend-paying firms. These results are consistent with the assertion that those firms with the highest dividend payout tend to be large mature firms with few growth opportunities.

III. Conclusion

In this study, we examine the relation between dividends and average insider returns to draw inferences about dividends and information asymmetry. To the best of our knowledge, our use of returns to insider trades as a measure of the level of information asymmetry between managers and their outside investors is unique in this context.

Our empirical evidence supports the proposition that the level of dividends is negatively related to information asymmetry. We find that firms that pay consistently high dividends have lower insider returns on purchases than do those that pay consistently low dividends. This finding supports the proposition that dividend policy is not a signal meant to reduce information asymmetry, because those firms that pay the highest level of dividends tend to have the lowest information asymmetry already. The lack of significant results for insider sales is consistent with insider sales transactions not being a good proxy for information asymmetry because these transactions are often not motivated by information.

Our findings are consistent with much of the recent evidence from empirical tests of dividend signaling models. Our tests imply that dividends may convey information, but they are not an effective signal that reduces information asymmetry. Further, our results are also consistent with studies showing that those firms with the highest level of dividend payout tend to be large mature firms with ample free cash flow and few growth opportunities. Overall, this article provides evidence that is not consistent with traditional dividend signaling models and suggests that dividend policy is set for other reasons, such as agency issues, dividend clienteles, etc.

We thank Brent Ambrose, Wayne Ferson, Richard C. Green, Bradford Jordan, Yong-Cheol Kim, David Mauer, David Myers, Lilian Ng, and Qinghai Wang for their helpful comments. We also received valuable comments from workshop participants at Southern Methodist University, the University of Kentucky, the University of Texas at Arlington, the University of Wisconsin-Milwaukee, the 2002 Financial Management Association Meeting, and the 2003 American Finance Association Meeting.

References

Aharony, J. and I. Swary, 1980, "Quarterly Dividends and Earnings Announcements and Stockholders' Returns: An Empirical Analysis," Journal of Finance 35, 1-12.

Allen, F. and R. Michaely, 2004, "Payout Policy," in G. Constantinides, M. Harris, and R. Stulz, Ed., Handbook of the Economics of Finance Volume 1A, Amsterdam, Elsevier North-Holland.

Asquith, P. and D. Mullins, 1983, "The Impact of Initiating Dividend Payments on Shareholder' Wealth," Journal of Business 56, 77-96.

Benartzi, S., R. Michaely, and R. Thaler, 1997, "Do Changes in Dividends Signal the Future or the Past?" Journal of Finance 52, 1007-1034.

Berk, J., 1995, "A Critique of Size-Related Anomalies," Review of Financial Studies 8, 275-286.

Brav, A., J. Graham, C. Harvey, and R. Michaely, 2005, "Payout Policy in the 21st Century," Journal of Financial Economics 77, 483-527.

Brennan, M. and A. Subrahmanyam, 1995, "Investment Analysis and Price Formation in Securities Markets," Journal of Financial Economics 38, 361-381.

Brook. Y., W. Charlton, and R. Hendershott, 1998, "Do Firms Use Dividend Increases to Signal Large Future Cash Flow Increases?" Financial Management 27, 46-57.

Charest, G., 1978, "Dividend Information, Stock Returns and Market Wfficiency--II," Journal of Financial Economics 6, 297-330.

DeAngelo, H., L. DeAngelo, and D. Skinner, 1996, "Reversal of Fortune: Dividend Policy and the Disappearance of Sustained Earnings Growth," Journal of Financial Economics 40, 341-371.

Demsetz, H., 1968, "The Cost of Transacting," Quarterly Journal of Economics 82, 33-53.

Fama, E. and K. French, 1992, "The Cross-Section of Expected Stock Returns," Journal of Finance 47, 427-465.

Fama, E. and K. French, 1993, "Common Risk Factors in Returns on Stocks and Bonds," Journal of Financial Economics 33, 3-56.

Fama, E. and J. MacBeth, 1973, "Risk, Return and Equilibrium: Empirical Tests," Journal of Political Economy 81, 607-636.

Ferson, W. and C. Harvey, 1999, "Conditioning Variables and the Cross Section of Stock Returns," Journal of Finance 54, 1325-1360.

Gonedes, N., 1978, "Corporate Signaling, External Accounting, and Capital Market Equilibrium: Evidence on Dividends, Income, and Extraordinary Items," Journal of Accounting Research 16, 26-79.

Grullon, G., R. Michaely, S. Bernartzi and R. Thaler, 2005, "Dividend Changes Do Not Signal Changes in Future Profitability," Journal of Business 78, 1659-1682.

Grullon, G., R. Michaely, and B. Swaminathan, 2002, "Are Dividend Changes a Sign of Firm Maturity?" Journal of Business 75, 387-424.

Healy, P. and K. Palepu, 1988, "Earnings Information Conveyed By Dividend Initiations and Omissions," Journal of Financial Economics 21, 149-175.

Huddart, S. and B. Ke, 2004, "Information Asymmetry and Cross-Sectional Variation in Insider Trading," Pennsylvania State University Working Paper.

Jeng, L., A. Metrick, and J. Zeckhauser, 2003, "Estimating the Returns to Insider Trading: A Performance-Evaluation Perspective," Review of Economics and Statistics 85, 453-471.

Koch, P. and C. Shenoy, 1999, "The Information Content of Dividend and Capital Structure Policy," Financial Management 28, 16-35.

Lakonishok, J. and I. Lee, 2001, "Are Insider Trades Informative?" Review of Financial Studies 14, 79-111.

Michaely, R., R. Thaler and K. Womack, 1995, "Price Reactions to Dividend Initiations and Omissions: Overreaction or Drift?" Journal of Finance 50, 573-608.

Penman, S., 1983, "The Predictive Content of Earnings Forecasts and Dividends," Journal of Finance 38, 1181-1199.

Watts, R., 1973, "The Information Component of Dividends," Journal of Business 46, 191-211.

(1) Please see Fama and French (1993) for their discussion on the construction of these long-short factor-mimicking portfolios. We obtain both daily and monthly HML and SMB returns along with the excess market return and risk-free rate from Kenneth French's website: http://mba.tuck.dartmouth.edu/pages/faculty/kenfrench.

(2) Since past studies indicate that these five control variables are related to both dividends and information asymmetry, the issue of multicollinearity may arise. We do not believe this is a critical issue because the average correlations across all months for the purchase and sales samples between dividend yield and the other five variables of BM, ln(MV), EV, DR, and ln(SH) are 0.03, 0.28, 0.04, 0.13, and 0.39, respectively. However, as a check, we estimate the regressions after eliminating these five control variables and including the others to check for robustness. The results are unchanged.

(3) We would like to thank an anonymous referee for pointing this out.

(4) For brevity, we omit these results. They are available from the authors on request.

(5) Whether these represent risk factors is subject to considerable controversy, but this method has become a standard way of adjusting for risk in the finance studies. See Fama and French (1993), Berk (1995), and Ferson and Harvey (1999).

Kenneth Khang and Tao-Hsien Dolly King *

* Kenneth Khang is an assistant professor of finance at Idaho State University in Pocatello, ID. Tao-Hsien Dolly King is an associate professor of finance at the University of North Carolina--Charlotte in Charlotte, NC.

Table I. Distribution of Market-Adjusted Insider Returns Table I presents summary statistics for the sample of insider returns for insider purchases and sales made between 1982 and 2003. Insider returns are the average insider excess returns over the 21-trading-day period following the insider trade. To calculate the weighted-by-shares average, we weight each transaction by the natural log of the number of shares traded relative to the natural log of the total number of shares traded by insiders for that day. In the table, we denote the market value of trade as MV To calculate the weighted-by-MV average, we weight each transaction by the natural log of the dollar amount of the trade relative to the natural log of the total dollar amount of the trade by insiders for that day. The unweighted average weights each trade equally. Panel A presents the distribution Panels B (C) present the distribution by ME (natural log of the market value of equity) and BM (book-to-market) for excess returns relative to the CRSP value weighted (equal-weighted) index. The number of observations is below each percentage figure. Panel A. Excess Returns Distributions Excess Returns Relative to CRSP Value-Weighted Index Weighted-by- Shares Purchases Sales 95th percentile 29.21% 95th percentile 19.56% 75th percentile 8.11% 75th percentile 5.46% median 0.82% median -0.67% 25th percentile -5.26% 25th percentile -6.51% 5th percentile -17.40% 5th percentile -19.06% Weighted-by-MV Purchases Sales 95th percentile 27.20% 95th percentile 19.04% 75th percentile 7.66% 75th percentile 5.38% median 0.76% median -0.62% 25th percentile -5.11% 25th percentile -6.34% 5th percentile -16.80% 5th percentile -18.44% Unweighted Purchases Sales 95th percentile 31.70% 95th percentile 20.32% 75th percentile 8.81% 75th percentile 5.99% median 0.94% median -0.21% 25th percentile -5.71% 25th percentile -6.15% 5th percentile -18.91% 5th percentile -18.24% Excess Returns Relative to CRSP Equal-Weighted Index Weighted-by- Shares Purchases Sales 95th percentile 27.39% 95th percentile 18.48% 75th percentile 7.23% 75th percentile 4.70% median 0.03% median -1.58% 25th percentile -6.29% 25th percentile -7.71% 5th percentile -18.46% 5th percentile -19.97% Weighted-by-MV Purchases Sales 95th percentile 25.45% 95th percentile 17.98% 75th percentile 6.89% 75th percentile 4.63% median -0.05% median -1.53% 25th percentile -6.15% 25th percentile -7.55% 5th percentile -17.86% 5th percentile -19.49% Unweighted Purchases Sales 95th percentile 29.55% 95th percentile 18.53% 75th percentile 7.82% 75th percentile 4.46% median -0.02% median -1.94% 25th percentile -6.90% 25th percentile -8.11% 5th percentile -20.04% 5th percentile -20.01% Panel B. Excess Returns Relative to CRSP Value-Weighted Index Weighted by Shares BM Insider 1 5 MV Purchases (low) 2 3 4 (high) Quintiles 1 (small) 4.08% 5.30% 4.61% 5.33% 4.60% 6.06% 1970 2181 2921 4774 9080 20926 2 5.03% 3.49% 1.91% 2.21% 4.00% 4.28% 2885 2735 3855 5573 4877 19925 ME 3 1.75% 2.22% 2.55% 2.25% 3.15% 3.19% 4050 4080 4505 4268 3029 19932 4 2.86% 1.88% 1.90% 1.80% 4.42% 3.46% 4787 5612 4892 3146 1490 19927 5 (large) 0.42% 1.14% 0.41% 2.13% 1.76% 1.19% 7232 5321 3771 2144 459 18927 BM quintiles 3.10% 3.89% 3.15% 4.40% 5.06% 3.94% 20924 19929 19944 19905 18935 99637 Weighted by MV 1 (small) 3.79% 4.99% 4.33% 5.15% 4.33% 5.42% 1970 2181 2921 4774 9080 20926 2 4.90% 3.32% 1.78% 2.16% 3.85% 3.40% 2885 2735 3855 5573 4877 19925 ME 3 1.59% 2.12% 2.38% 2.16% 3.02% 2.72% 4050 4080 4505 4268 3029 19932 4 2.72% 1.74% 1.76% 1.66% 4.33% 2.86% 4787 5612 4892 3146 1490 19927 5 (large) 0.41% 1.04% 0.36% 1.92% 1.72% 1.01% 7232 5321 3771 2144 459 18927 BM quintiles 2.71% 2.55% 2.18% 2.90% 5.43% 3.13% 20924 19929 19944 19905 18935 99637 Unweighted 1 (small) 4.78% 5.69% 3.92% 4.64% 6.37% 5.41% 1970 2181 2921 4774 9080 20926 2 4.60% 3.23% 2.87% 2.21% 4.76% 3.45% 2885 2735 3855 5573 4877 19925 ME 3 2.54% 2.32% 2.12% 2.89% 4.08% 2.71% 4050 4080 4505 4268 3029 19932 4 3.94% 2.44% 1.91% 1.98% 5.87% 2.85% 4787 5612 4892 3146 1490 19927 5 (large) 0.69% 1.23% 0.54% 2.27% 1.51% 1.01% 7232 5321 3771 2144 459 18927 BM quintiles 2.72% 2.56% 2.18% 2.91% 5.43% 3.13% 20924 19929 19944 19905 18935 99637 BM Insider 1 5 MV Sales (low) 2 3 4 (high) Quintiles 1 (small) -0.37% -0.37% 0.34% -0.26% -0.08% 0.07% 4995 5557 7944 11406 22381 52283 2 -0.66% 0.03% -0.60% -0.16% -0.73% 0.15% 6635 8845 9551 13137 11643 49811 ME 3 0.32% 0.03% 0.29% -0.62% 0.64% 0.27% 8101 11755 12061 10940 6908 49765 4 1.28% -0.21% -0.22% -0.13% 0.20% 0.12% 12899 11611 11815 9295 4212 49832 5 (large) 0.73% 0.06% -0.47% -0.92% 0.10% 0.46% 19662 12023 8433 5022 2164 47304 BM quintiles 0.73% 0.47% -0.17% 0.01% -0.09% 0.20% 52292 49791 49804 49800 47308 248995 Weighted by MV 1 (small) -0.56% -0.26% 0.19% -0.30% -0.10% 1.12% 4995 5557 7944 11406 22381 52283 2 -0.58% -0.04% -0.56% -0.11% -0.72% 0.37% 6635 8845 9551 13137 11643 49811 ME 3 0.34% 0.03% 0.30% -0.64% 0.60% 0.26% 8101 11755 12061 10940 6908 49765 4 1.35% -0.22% -0.22% -0.13% 0.17% 0.40% 12899 11611 11815 9295 4212 49832 5 (large) 0.69% 0.04% -0.48% -0.86% 0.07% 0.04% 19662 12023 8433 5022 2164 47304 BM quintiles 0.63% 0.22% 0.33% 0.16% 0.91% 0.45% 52292 49791 49804 49800 47308 248995 Unweighted 1 (small) 0.95% 0.50% 1.60% 0.70% 1.38% 1.13% 4995 5557 7944 11406 22381 52283 2 0.45% 0.88% -0.31% 0.64% 0.18% 0.37% 6635 8845 9551 13137 11643 49811 ME 3 0.01% 0.09% 0.71% -0.31% 0.89% 0.25% 8101 11755 12061 10940 6908 49765 4 1.30% -0.36% 0.23% 0.19% 0.80% 0.41% 12899 11611 11815 9295 4212 49832 5 (large) 0.43% 0.29% -0.56% -1.18% 0.18% 0.04% 19662 12023 8433 5022 2164 47304 BM quintiles 0.63% 0.22% 0.33% 0.18% 0.91% 0.45% 52292 49791 49804 49800 47308 248995 Panel C. Excess Returns Relative to CRSP Equal-Weighted Index Weighted by Shares BM Insider 1 5 ME Purchases (low) 2 3 4 (high) Quintile 1 (small) 2.72% 4.30% 3.43% 4.36% 3.25% 3.61% 1970 2181 2921 4774 9080 20926 2 3.84% 2.63% 1.15% 1.29% 2.45% 2.05% 2885 2735 3855 5573 4877 19925 ME 3 0.76% 1.52% 1.72% 1.30% 1.59% 1.38% 4050 4080 4505 4268 3029 19932 4 2.12% 1.12% 0.90% 0.99% 3.05% 1.40% 4787 5612 4892 3146 1490 19927 5 (large) -0.63% 0.20% -0.61% 1.41% 0.88% -0.10% 7232 5321 3771 2144 459 18927 BM quintile 1.28% 1.63% 1.31% 2.09% 2.76% 1.81% 20924 19929 19944 19905 18935 99637 Weighted-by-MV 1 (small) 2.41% 4.02% 3.14% 4.18% 3.01% 3.37% 1970 2181 2921 4774 9080 20926 2 3.69% 2.46% 1.02% 1.24% 2.29% 1.92% 2885 2735 3855 5573 4877 19925 ME 3 0.60% 1.39% 1.54% 1.18% 1.45% 1.24% 4050 4080 4505 4268 3029 19932 4 1.95% 0.97% 0.76% 0.84% 2.98% 1.24% 4787 5612 4892 3146 1490 19927 5 (large) -0.67% 0.06% -0.70% 1.18% 0.84% -0.20% 7232 5321 3771 2144 459 18927 BM quintile 1.03% 1.37% 1.07% 1.88% 2.54% 1.55% 20924 19929 19944 19905 18935 99637 Unweighted 1 (small) 3.50% 4.37% 2.49% 3.44% 4.80% 4.00% 1970 2181 2921 4774 9080 20926 2 3.25% 2.22% 1.88% 1.13% 3.03% 2.20% 2885 2735 3855 5573 4877 19925 ME 3 1.45% 1.32% 1.19% 1.76% 2.60% 1.61% 4050 4080 4505 4268 3029 19932 4 2.80% 1.61% 0.76% 0.91% 4.36% 1.78% 4787 5612 4892 3146 1490 19927 5 (large) -0.52% 0.14% -0.67% 1.31% 0.76% -0.13% 7232 5321 3771 2144 459 18927 BM quintile 1.52% 1.54% 1.06% 1.81% 3.86% 1.93% 20924 19929 19944 19905 18935 99637 BM Insider 1 5 Sales (low) 2 3 4 (high) 1 (small) -1.37% -1.32% -0.60% -0.95% -1.00% 4995 5557 7944 11406 22381 2 -1.94% -0.93% -1.38% -0.89% -1.80% 6635 8845 9551 13137 11643 ME 3 -0.66% -1.13% -0.76% -1.64% -0.51% 8101 11755 12061 10940 6908 4 -0.14% -1.17% -1.27% -1.11% -0.43% 12899 11611 11815 9295 4212 5 (large) -0.84% -1.03% -1.54% -2.25% -0.57% 19662 12023 8433 5022 2164 BM quintile -0.93% -1.11% -1.08% -1.18% -1.04% 52292 49791 49804 49800 47308 Weighted-by-MV 1 (small) -1.60% -1.23% -0.76% -0.98% -1.02% 4995 5557 7944 11406 22381 2 -1.86% -0.99% -1.34% -0.83% -1.78% 6635 8845 9551 13137 11643 ME 3 -0.65% -1.13% -0.73% -1.65% -0.52% 8101 11755 12061 10940 6908 4 -0.09% -1.17% -1.27% -1.09% -0.45% 12899 11611 11815 9295 4212 5 (large) -0.86% -1.02% -1.53% -2.18% -0.58% 19662 12023 8433 5022 2164 BM quintile -0.93% -1.10% -1.10% -1.18% -1.04% 52292 49791 49804 49800 47308 Unweighted 1 (small) -0.74% -0.95% 0.09% -0.64% -0.16% 4995 5557 7944 11406 22381 2 -1.49% -0.79% -1.96% -0.88% -1.57% 6635 8845 9551 13137 11643 ME 3 -1.66% -1.92% -1.17% -2.12% -0.92% 8101 11755 12061 10940 6908 4 -0.83% -2.31% -1.74% -1.54% -0.76% 12899 11611 11815 9295 4212 5 (large) -1.84% -1.50% -2.53% -3.23% -1.16% 19662 12023 8433 5022 2164 BM quintile -1.41% -1.60% -1.48% -1.46% -0.72% 52292 49791 49804 49800 47308 Table II. Sample Statistics for Insider Transactions Table 11 presents summary statistics from the sample of insider transactions from 1982 to 2003. Panel A presents the number of insider transactions segmented by dividend policy. We rank the firms by average dividend yield over the previous five years and partition the sample into two halves, high-dividend-paying firms and low-dividend-paying firms. Panel B reports the number of firms by dividend policy. Panel A. Number of Transactions Insider Insider 21-trading-days Purchases Sales Total High-dividend-paying firms 49,822 124,512 174,334 Low-dividend-paying firms 49,815 124,483 174,298 Total 99,637 248,995 Panel B. Number of Firms Insider Insider 21-trading-days Purchases Sales Total High-dividend-paying firms 2,614 2,882 5,496 Low-dividend-paying firms 3,691 3,325 7,016 Total 6,305 6,207 Table III. Market-Adjusted Insider Returns for High-Dividend-Paying Firm Transactions Compared to Low-Dividend-Paying Firm Transactions Table III reports the average insider excess returns over a 21-trading-day period. We rank the firms by their average dividend yields over the previous five years and partition the sample of transactions into two halves, high-dividend-paying firms and low-dividend-paying firms. To calculate the weighted-by-shares average, we weight each transaction by the natural log of the number of shares traded relative to the natural log of the total number of shares traded by insiders for that day. In the table, we denote the market value of trade as MV. To calculate the weighted-by-MV average, we weight each transaction by the natural log of the dollar amount of the trade relative to the natural log of the total dollar amount of the trade by insiders for that day. The unweighted average weights each transaction equally. Panel A displays the two average excess returns relative to the CRSP NYSE/Amex/Nasdaq value-weighted index. Panel B displays the two average excess returns relative to the equal-weighted index. The t-statistics apply to the differences in the mean returns between high-dividend-paying firms and low-dividend-paying firms. Panel A. Excess Returns Relative to CRSP Value-Weighted Index Insider Purchases Insider Sales (N=99,637) (N=248,995) Weighted-by-Shares High-Dividend-paying 1.79% -0.16% firms (N=174,298) Low-dividend-paying 4.05% 0.07% firms (N=174,334) Difference -2.26% *** -0.23% *** t-statistic -21.14 -3.53 Weighted-by-MV High-Dividend-paying 1.63% -0.16% firms (N=174,298) Low-dividend-paying 3.82% 0.07% firms (N=174,334) Difference -2.19% *** -0.22% *** t-statistic -21.14 -3.72 Unweighted High-Dividend-paying 1.58% 0.23% firms (N=174,298) Low-dividend-paying 4.68% 0.67% firms (N=174,334) Difference -3.10% *** -0.44% *** t-statistic -27.1 -7.37 Panel B. Excess Returns Relative to CRSP Equal-Weighted Index Weighted-by-Shares High-Dividend-paying 0.97% -1.01% firms (N=174,298) Low-dividend-paying 2.79% -1.16% firms (N=174,334) Difference -1.82% *** 0.16% ** t-statistic -17.33 2.42 Weighted-by-MV High-Dividend-paying 0.80% -1.00% firms (N=174,298) Low-dividend-paying 2.55% -1.18% firms (N=174,334) Difference -1.75% *** 0.18% *** t-statistic -17.21 2.96 Unweighted High-Dividend-paying 0.65% -1.30% firms (N=174,298) Low-dividend-paying 3.21% -1.39% firms (N=174,334) Difference -2.56% *** 0.08% t-statistic -22.62 1.43 Panel A. Excess Returns Relative to CRSP Value-Weighted Index Difference t-statistic Weighted-by-Shares High-Dividend-paying 1.95% *** 35.62 firms (N=174,298) Low-dividend-paying 3.98% *** 35.87 firms (N=174,334) Difference t-statistic Weighted-by-MV High-Dividend-paying 1.79% *** 33.57 firms (N=174,298) Low-dividend-paying 3.75% *** 36.48 firms (N=174,334) Difference t-statistic Unweighted High-Dividend-paying 1.35% *** 24.21 firms (N=174,298) Low-dividend-paying 4.01% *** 38.47 firms (N=174,334) Difference t-statistic Panel B. Excess Returns Relative to CRSP Equal-Weighted Index Weighted-by-Shares High-Dividend-paying 1.98% *** 35.95 firms (N=174,298) Low-dividend-paying 3.96% *** 36.30 firms (N=174,334) Difference t-statistic Weighted-by-MV High-Dividend-paying 1.80% *** 33.57 firms (N=174,298) Low-dividend-paying 3.73% *** 36.96 firms (N=174,334) Difference t-statistic Unweighted High-Dividend-paying 1.95% *** 34.70 firms (N=174,298) Low-dividend-paying 4.59% *** 44.78 firms (N=174,334) Difference t-statistic *** Significant at the 0.01 level. ** Significant at the 0.05 level. * Significant at the 0.10 level. Table IV. Fama-French (1993) Abnormal Returns and Fama-MacBeth (1973) Regressions Panel A reports the average 21-trading-day Fama-French (1993) three-factor model abnormal returns. For each transaction, we estimate the Fama-French (1993) three-factor model using consecutive 21-trading-day returns over the previous three years. We use these estimates to calculate an expected return for each 21-trading-day period following a transaction. We obtain the abnormal return by subtracting the expected return from the actual. We then rank the firms by their average dividend yields over the previous five years and partition the sample of transactions into two halves, high-dividend-paying firms and low-dividend-paying firms. To calculate the share-weighted average, we weight each transaction by the natural log of the number of shares traded relative to the natural log of the total number of shares traded by insiders for that month. In the table, we denote the market value of trade as MV. To calculate the MV-weighted average, we weight each transaction by the natural log of the dollar amount of the trade relative to the natural log of the total dollar amount of the trade by insiders for that month. Panels B, C, D, E, F, and G present the average coefficients and the associated t-statistics for monthly Fama-MacBeth (1973) regressions of insider excess returns over the 21-trading-days following an insider transaction day (k). Panels B and C report the weighted-by-shares results. Panels D and E report the weighted-by-MV results. Panels F and G report the unweighted results. For the weighted-by-shares results, we weight each transaction by the number of shares bought (sold) relative to the total number of shares bought (sold) by insiders for that month. For the weighted-by-MV results, we weight each transaction by the market value of the shares bought (sold) relative to the total market value of the shares bought (sold) by insiders for that month. The unweighted results weight each transaction equally. We regress these insider excess returns on the following for firm is DY is the average dividend yield over the previous five years. BM is the firm's book-to-market ratio from the previous fiscal year. Ln (ME) is the natural log of the firm's market value of equity from the previous fiscal year-end. EV is the annual earnings-per-share variance over the previous five years. DR is the ratio of the firm's long-term debt over total assets from the previous fiscal year. RD is the ratio of R&D expense over sales from the previous fiscal year. IT is the ratio of intangible assets over total assets from the previous fiscal year. T is the ratio of plant, property, and equipment over total assets from the previous fiscal year. Ln(SH) is the natural log of the number of shareholders from the previous fiscal year-end. EX is the sum of extraordinary items, discontinued operations, and special items over total sales from the previous fiscal year. RV is the variance of daily returns over the previous calendar year. LS is equal to one if earnings per share was negative during the previous fiscal year, and zero otherwise. E is equal to one if the firm was traded on the NYSE or Amex during the previous fiscal year, and zero otherwise. There are three specifications for the Fama-MacBeth (1973) regressions: [R.sub.ki] = [alpha] + [[lambda].sub.dy] D[Y.sub.i] + [[epsilon].sub.ki] [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] Panel A. Excess Returns Relative to the Fama-French 3-factor Model Insider Purchases Insider Sales (N=99,365) (N=248,605 Weighted-by-Shares High-Dividend-paying 1.73% -0.05% firms (N=174,092) Low-dividend-paying 3.99% -0.20% firms (N=173,878) Difference -2.26% *** 0.15% ** t-statistic -17.40 2.11 Weighted-by-MV High-Dividend-paying 1.57% -0.05% firms (N=174,092) Low-dividend-paying 3.75% -0.21% firms (N=173,878) Difference -2.18% *** 0.16% ** t-statistic -17.02 2.35 Unweighted High-Dividend-paying 1.51% -0.07% firms (N=174,092) Low-dividend-paying 3.80% -0.25% firms (N=173,878) Difference -2.29% *** 0.18% ** t-statistic -18.06 2.48 Panel A. Excess Returns Relative to the Fama-French 3-factor Model Difference t-statistic Weighted-by-Shares High-Dividend-paying 1.78% *** 30.94 firms (N=174,092) Low-dividend-paying 4.19% *** 32.49 firms (N=173,878) Difference t-statistic Weighted-by-MV High-Dividend-paying 1.62% *** 28.91 firms (N=174,092) Low-dividend-paying 3.96% *** 31.74 firms (N=173,878) Difference t-statistic Unweighted High-Dividend-paying 1.58% *** 27.90 firms (N=174,092) Low-dividend-paying 4.04% *** 31.98 firms (N=173,878) Difference t-statistic Table V. Returns for Zero-cost Portfolios: High-Dividend-Paying Firms Compared to Low-Dividend-Paying Firms Panel A presents mean monthly insider excess return relative to the CRSP value-weighted index on four portfolios. We rank firms by their average dividend yields over the previous five years and partition the sample into two halves, high-dividend-paying firms and low-dividend-paying firms. We calculate a portfolio's excess return for each month as follows. First, we net the number of shares for insider transactions by firm to yield a net purchase or sales amount for the previous month. We then calculate the monthly insider excess return for each firm as the excess return relative to the CRSP value-weighted index in the subsequent month. Next, for each month we calculate the average excess return across firms within the four categories use to form the portfolios. These categories are high-dividend-paying firms where the insiders are net purchasers of shares, high-dividend-paying firms where insiders are net seller of shares, low-dividend-paying firms where insiders are net purchasers, and low-dividend-paying firms where insiders are net sellers. The weighted-by-shares results weight each firm-month observation by the natural log of the monthly net shares purchased (sold) relative to the natural log of the total number of shares net purchased (sold) for that month in the respective category. In the table, we denote the market value of trade as MV. The weighted-by-MV results weight each firm-month observation by the natural log of the market value of shares net purchased (sold) relative to the natural log of the market value of shares net purchased (sold) for that month in the respective category. The unweighted results weight each observation equally. In Panel B we use the Fama-French (1993) three-factor model on the returns to two zero-cost portfolios. The first zero-cost portfolio is long the portfolio of high-dividend-paying firms, where insiders are net purchasers, and short the portfolio of low-dividend-paying firms, where insiders are net purchasers. The second zero-cost portfolio is the same as the first, except we use firms that are net sellers. Thus, for the zero-cost portfolio regressions are: [R.sub.HDPt] - [R.sub.LDPt] - [alpha] + [beta][R.sub.mt] + [gamma]SM[B.sub.t] + [lambda]HM[L.sub.t] + [[epsilon].sub.t] [R.sub.HDSt] - [R.sub.LDSt] - [alpha] + [beta][R.sub.mt] + [gamma]SM[B.sub.t] + [lambda]HM[L.sub.t] + [[epsilon].sub.t] where [R.sub.HDPt] is the month t return to the portfolio of high-dividend-paying firms where insiders are net purchasers; [R.sub.LDPt] is the month t return to the portfolio of low-dividend-paying firms where insiders are net purchasers; [R.sub.HDSt] is the month t return to the HDP portfolio where insiders are net sellers; [R.sub.LDSt] is the month t return to the LDP portfolio where insiders are net sellers; [alpha] is the intercept term, or the abnormal return; [beta] is the sensitivity to the market factor; [R.sub.mt] is the excess return on the market, which in this case is the CRSP value-weighted index; [gamma] is the sensitivity to the size factor; SM[B.sub.t] is the Fama-French (1993) factor-mimicking portfolio for size; [gamma] is the sensitivity to the book-to-market factor; HM[L.sub.t] is the Fama-French (1993) factor-mimicking portfolio for the book-to-market factor; and [e.sub.t] is the regression error. Panel A. Mean Monthly Excess Returns Relative to CRSP Value-Weighted Index Weighted-by-Shares Net Insider Net Insider Purchases Sales High-dividend- 1.43% -0.17% paying firms Low-dividend- 2.89% -0.18% paying firms excess [R.sub.HDPt] - -1.46% *** 0.01% excess [R.sub.LDPt] t-statistic -5.41 0.05 Proportion 63.3% *** 44.7% * Negative Months Binomial Test 4.31 -1.72 Weighted-by-MV Net Insider Net Insider Purchases Sales High-dividend- 1.29% -0.17% paying firms Low-dividend- 2.66% -0.17% paying firms excess [R.sub.HDPt] - -1.37% *** 0.01 excess [R.sub.LDPt] t-statistic -5.33 0.01 Proportion 62.1% *** 44.0% * Negative Months Binomial Test 3.94 -1.85 Unweighted Net Insider Net Insider Purchases Sales High-dividend- 1.20% -0.16% paying firms Low-dividend- 2.63% -0.17% paying firms excess [R.sub.HDPt] - -1.43% *** 0.00% excess [R.sub.LDPt] t-statistic -5.45 0.02 Proportion 62.9% *** 44.3% * Negative Months Binomial Test 4.19 -1.85 Panel B. Regressions of Zero-cost Portfolio Returns using Fama-French 3-Factor Model Excess Alpha Mkt Return SMB Weighted-by- shares Insider Purchases [R.sub.HDPt]- -0.0141 *** -0.1616 *** -0.4372 *** [R.sub.LDPt] t-statistic -6.06 -2.81 -6.00 Insider Sales [R.sub.HDSt]- -0.0002 -0.1526 *** -0.5462 *** [R.sub.LDSt] t-statistic -0.14 -3.92 -11.08 Weighted-by- MV Insider Purchases [R.sub.HDPt]- -0.0130 *** -0.1785 *** -0.4629 *** [R.sub.LDPt] t-statistic -6.17 -3.42 -6.99 Insider Sales [R.sub.HDSt]- -0.0004 -0.1525 *** -0.5361 *** [R.sub.LDSt] t-statistic -0.27 -4.06 -11.28 Unweighted Insider Purchases [R.sub.HDPt]- -0.0134 *** -0.1883 *** -0.4725 *** [R.sub.LDPt] t-statistic -6.29 -3.56 -7.04 Insider Sales [R.sub.HDSt]- -0.0003 -0.1494 *** -0.5417 *** [R.sub.LDSt] t-statistic -0.20 -3.93 -11.26 R-SQR HML (adjusted) Weighted-by- shares Insider Purchases [R.sub.HDPt]- 0.2983 *** 0.3154 [R.sub.LDPt] t-statistic 3.43 Insider Sales [R.sub.HDSt]- 0.4711 *** 0.6357 [R.sub.LDSt] t-statistic 8.01 Weighted-by- MV Insider Purchases [R.sub.HDPt]- 0.2894 *** 0.3769 [R.sub.LDPt] t-statistic 3.66 Insider Sales [R.sub.HDSt]- 0.4862 *** 0.6543 [R.sub.LDSt] t-statistic 8.57 Unweighted Insider Purchases [R.sub.HDPt]- 0.2859 *** 0.3789 [R.sub.LDPt] t-statistic 3.57 Insider Sales [R.sub.HDSt]- 0.4634 *** 0.6406 [R.sub.LDSt] t-statistic 8.07 *** Significant at the 0.01 level. ** Significant at the 0.05 level. * Significant at the 0.10 level.

Printer friendly Cite/link Email Feedback | |

Author: | Khang, Kenneth; King, Tao-Hsien Dolly |
---|---|

Publication: | Financial Management |

Date: | Dec 22, 2006 |

Words: | 11384 |

Previous Article: | A comparison of syndicated loan pricing at investment and commercial banks. |

Next Article: | Who wants to trade around ex-dividend days? |

Topics: |