Ex-dividend day price and volume: the case of 2003 dividend tax cut.

Transaction costs play an important role in trading behavior. Trading around the ex-dividend day may be unprofitable when traders face large transaction costs. Since low dividend yield stocks have relatively higher transaction costs than high dividend yield stocks, stocks with a high dividend yield are more likely to show a significant amount of tax-induced trading around the ex-dividend day. Hence, we expect the decrease in the ex-dividend day trading volume to be more significant for high dividend yield stocks. Risk, just like transaction costs, discourages trading around the ex-dividend day. As implied by the dynamic trading model of Michaely and Vila (1995), trading volume is a decreasing function of transaction costs and risk. Therefore our last hypothesis is as follows.

Hypothesis 4: Trading volume is positively related to dividend yield and negatively related to risk and transaction costs.

Given that the 2003 Act only lowers dividend tax rates for individuals, any significant changes on the ex-dividend day price behavior before and after the tax cut would suggest that the ex-dividend day price reflects individual investors' relative tax rates on dividends and capital gains. This would suggest that individual investors' preferences have a measurable impact on ex-dividend day price formation and trading behavior.

DATA AND METHODOLOGY

We collect data from the Center for Research in Security Prices (CRSP). We examine firms that pay taxable regular cash dividends (distribution codes 1222, 1232, 1242, and 1252) to their common stockholders (share codes 10,11, and 12). (6) We exclude closed-end funds, unit investment trusts, ETFs, ADRs, and REITs because of their different tax treatment and more complex distributions. We eliminate observations where the dividend was less than one cent, (7) where there was no trade on the ex-dividend date, or the price was lower than five dollars. (8) By the end of January 2001, all New York Stock Exchange (NYSE) stocks had converted their price quotations to decimals. (9) Thus, we choose February 2001 as the starting point of our study to avoid the possible effect of the change in the market microstructure. We eliminate year 2003 as the transition year since there may be a lag in investors' restructuring of their portfolios. (10) Therefore we use 2001 (excluding January) and 2002 to represent a distinct pre-Act period and 2004 and 2005 to represent a distinct post--Act period.

To estimate the PDR, we adopt a methodology similar to Michaely (1991). We adjust the ex-dividend day closing price by the expected daily return of the stock. (11) We estimate the expected return using the market model over the window -45 to -2 and +2 to +45, where day zero is the ex-dividend day and the market return is the value-weighted market portfolio (including distributions). Our event window is the same that is used by Michaely and Vila (1995)

[5] [PDR.sub.i] = [P.sub.i,cum] - [P.sub.i,ex]/1 + E(r.sub.i)/[D.sub.i]

where

[P.sub.i,cum] is the cum-dividend day closing price for stock i,

[P.sub.i,ex] is the ex-dividend day closing price for stock i,

E([r.sub.i]) is the expected daily return of stock i, estimated by the market model, and

[D.sub.i] is the dividend amount for stock i.

Due to the large variation in our dividend distribution sample, we winsorize the data at the 2.5 percent level to reduce the effect of outliers. (12) Furthermore, following Michaely (1991), we correct two sources of heteroskedasticity: the security's variance and the dividend yield effect. The stock variance is calculated from the market model and dividend yield is calculated as the amount of the cash dividend divided by the cum-dividend price. As shown in Michaely (1991), the covariance matrix of the disturbances of ex-dividend day PDR is

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [[sigma].sup.2] is the variance of stock i, estimated from the market model, and d is the dividend yield on stock i.

So the mean PDR is calculated using weighted least squares where:

[6] [bar.PDR] = [n.summation over (i=1)(d.sup.2.sub.i / [[sigma].sup.2.sub.i] x [PDR.sub.i]/[n.summation over (i=1] ([d.sup.2.sub.i] / [[sigma].sup.2.sub.i])

In addition, we use the median of the PDR since the price drop ratio may not be normally distributed. The Wilcoxon sign test is used to test the median difference from one and Wilcoxon--Mann--Whitney test is used to test the difference in the PDR before and after the 2003 Act. The two--sided p-value is reported in the tables.

We also analyze the ex-dividend day excess returns to further examine the ex-dividend day price behavior. We calculated the ex-dividend day excess return on stock i as

[7] [P.sub.i,ex] + [D.sub.i] - [P.sub.i,cum]/[P.sub.i,cum] - E([r.sub.i]).

Besides the ex-dividend day, we examine returns of ten days surrounding the ex-dividend day. Our study window is [-5, +5], where day zero is the ex-dividend day. We use [-45, -6] and [6, 45] as our estimation window to calculate the expected return by the market model. We adopt the weighted least squares with the inverse of the standard deviation of the estimation period return as the weight to correct for heteroskedasticity.

Following Michaely and Vila (1995), risk is measured by the variance of the security's return scaled by the variance of the market returns during the estimation period of the market model. Similar to Michaely and Vila (1996), we use market capitalization on the cum-dividend day as a proxy for transaction costs. Our assumption is that large--capitalization stocks are more liquid and have lower transaction costs. To reduce the effect of outliers, we specify the variable as the natural logarithm of market capitalization.

Similar to Michaely and Vila (1995) and Graham et al. (2003), we measure trading volume by stock turnover. Daily turnover is defined as the ratio of daily shares traded to shares outstanding. We compute the average turnover from the average daily turnover for days -45 to -6 and +6 to +45

[8] [ATO.sub.i] = [[summation].sub.t[member of][-45,-6][union][+6,+45] [TO.sub.it]/T

where [TO.sub.it] is the daily turnover for security i on day t, and T is the number of days with valid volume observations in the estimation period. Then for each day in the event window [-5, +5], we calculate the AV as the difference between the stock's actual to average turnover, relative to the average turnover:

[9] [AV.sub.it] = [TO.sub.it] - [ATO.sub.i]/[ATO.sub.i]

The mean daily AV for the sample is computed as

[10] [AV.sub.t] + [N.summation over i=1] [AV.sub.it]/N t [member of] [-5, +5]

We also compute the cumulative abnormal volume (CAV) of 11 days around the ex-dividend day. The mean of CAV is calculated in the same way as the mean of AV. T-statistics for the volume are calculated using the cross-sectional estimates of the variance of abnormal volume. Given that the daily turnover data is highly skewed, we winsorize the daily AV data at the 2.5 percent level. Data with missing or negative trading volume are deleted from the sample.

EMPIRICAL RESULTS

We begin by providing some overall statistics regarding the dividend behavior of firms (share codes 10-12) tracked by CRSP (distribution codes 1222, 1232, 1242, and 1252) between the period 2001 and 2005 in Table 1. (13) We eliminate firms paying less than a one-cent regular dividend and firms with a stock price lower than five dollars. Based on these restrictions, the number of firms paying regular dividends has increased from 1,751 firms in 2001 to 1,898 firms in 2005. The number of dividend initiations and the number of firms continuing to pay dividends has similarly increased over the period with a corresponding decline in the number of firms discontinuing dividends. Most dramatically, the number of firms initiating dividends more than doubled between 2002 and 2003 (102 to 215) and the number of firms discontinuing dividends went from 310 in 2001 to 119 in 2003. (14) We also document an increasing average annual dividend per share which during the pre-Act period was 0.536 and 0.546 in 2001 and 2002, respectively, while the average annual dividend per share during the post-Act period was 0.560 and 0.583 in 2004 and 2005, respectively.

Table 1 shows that the composition of firms paying dividends across time may be changing substantially. Chetty and Saez (2006) also find that the 2003 Act induced many firms to initiate dividend payments. To avoid introducing a composition bias into the sample, we restrict the remainder of our analysis to firms that pay dividends in all four years of the pre- and post-Act periods that results in a sample of 1,278 firms. Firms may change the amount or frequency of distribution of dividends during the pre- and post-Act periods, however. (15)

PDR AND EXCESS RETURN ANALYSIS