CAPITAL GAINS TAXES: Empirical Analysis 5

All control variables are collected from Compustat.20 To avoid extreme observations, values of Debt/Assets, Return on Assets and Book/Market are trimmed at the 1st and 99th percentiles. Inferences are insensitive to this trimming.

Table 1 presents descriptive statistics for the control variables, split between dividend-paying and non-dividend-paying stocks. Not surprisingly, dividend-paying stocks are larger, on average, more highly levered (median debt/asset ratios of 0.42 versus 0.64), more profitable (median ROA of 13.76 percent versus 10.8 percent respectively), and have higher book/market ratios (0.33 versus 0.18) implying lower expected earnings growth. t-tests of the mean control variables for the two groups are significantly different at conventional levels; however, for each control variable, the two distributions overlap substantially.

Table 2, Column B, shows results including the control variables. Two facts are noteworthy. First, although the coefficient estimate on the dividend indicator variable drops from -6.24 to -5.72, it remains strongly negative suggesting that differences in returns are not driven primarily by the control variables. Second, only debt/assets and book/market are significantly related to returns (both negatively).
We are happy to offer you only best and very speedy payday loans deposited directly into your bank account in just a few hours. This is faster than you can believe, but it’s very much real. Visit us at add comment and request a loan today, we will make sure it reaches you as soon as possible, this is what we are good at.
To our knowledge, the impact of the budget agreement and its capital gains tax rate cut should not have varied substantially across industrial sectors. However, to ensure that the results are not caused by differences across industries, Table 2, Column C reports results using industry indicator variables. Specifically, firms are split into financial, extractive, manufacturing, utilities and services sectors based on their Datastream level 3 industry codes. Again, inferences are unaffected. Only the indicator variable for the extractive industries is statistically significant.

To determine whether the results are robust across industries, the simple regression was reestimated for each industry separately. The relation between stock returns and the dividend indicator variable is significantly negative in each industry, except the extractive industry where it is insignificant. To ensure that results are not driven by the coarseness of the industry partition, firms are split into finer partitions, up to Datastream level 6 industry codes that divide firms into eighty-three groupings. Results are always consistent.

Finally, unreported regressions show that inferences are unaltered when the explanatory variables include beta or the percentage change in stock prices over various combinations of the months preceding the event period. In addition, excluding firms that went ex-dividend during the examination period does not qualitatively affect the results.