Printer Friendly

The effect of shareholder-level dividend taxes on stock prices: evidence from the Revenue Reconciliation Act of 1993.

I. INTRODUCTION

The effect of shareholder-level taxes on stock prices is a fundamental issue for both tax policy and stock valuation. Shareholder-level taxes may affect share prices because these taxes are imposed on distributions of earnings (dividend taxes) and on the sale of shares through trades, share repurchases, and liquidating distributions (capital gains taxes), thus reducing after-tax cash flows to investors. The effect of shareholder-level taxes on share prices, however, is complicated because the amount of explicit shareholder-level taxes depends on the type of income realized (dividend or capital gain) and the timing of income recognition. Furthermore, the tax rate also depends on whether the shareholder is an individual, a corporation, or a tax-exempt entity (e.g., a pension fund). Not surprisingly, the extent to which shareholder-level taxes affect stock prices is a difficult empirical question that has generated decades of research without reaching consensus (e.g., Zodrow 1991; Erickson and Maydew 1998).

This study investigates the effects of shareholder-level dividend taxes on share value by analyzing stock price reactions to the increase in the maximum individual income tax rate (i.e., the dividend tax rate) from 31.0 to 39.6 percent enacted in the Revenue Reconciliation Act of 1993 (Public Law 103-66; hereafter RRA93). In particular, we investigate whether firm dividend policy and the tax status of the marginal investor influence the stock price reaction to an increase in the dividend tax rate. We expect that the higher the firm's dividend yield, the more negative the firm's stock price reaction to this tax increase. However, because RRA93 primarily increased the tax rates of individual investors, we expect the negative effect of the firm's dividend yield on event-period returns to be mitigated by the level of institutional ownership, a proxy for the likelihood that a stock's marginal investor is not an individual taxpayer. We test these hypotheses by regressing cumulative daily abnormal stock returns surrounding the passage of RRA93 on firm dividend yield, institutional ownership, the interaction of these two variables, and control variables. Consistent with our expectations, we find that the higher the firm's dividend yield, the more negative the stock price reaction over the five-day period that Congress enacted the increase in individual income tax rates in August 1993. In addition, we find that these returns are more negative for stocks in which the marginal investor is likely to be an individual taxpayer. Our results are robust to controls for risk, profitability, leverage, size, industry, and stock return differences related to dividend policy in nonevent weeks.

Our evidence that RRA93 adversely affected stock prices is consistent with earlier research reporting investor-level tax effects on the pretax returns of assets such as preferred stock (Erickson and Maydew 1998; Engel et al. 1999), municipal bonds (Poterba 1986; Mankiw and Poterba 1996), and Treasury bills (Guenther 1994). Our findings extend this literature by demonstrating that the tax-induced price effect on common stocks depends on both the firm's dividend policy and shareholder-level tax characteristics. Specifically, this study makes two contributions to the literature investigating the stock price effects of shareholder-level taxes in general, and of dividend taxes in particular.

First, our finding that the magnitude of the adverse price effect of RRA93 increases in dividend yield suggests that the extent to which changes in shareholder-level dividend taxes are impounded into stock prices depends on the market's expectations of a firm's dividend policy. This helps distinguish between the two competing theories of shareholder-level tax effects on share prices. (1) In brief, the "tax-capitalization" view (e.g., Auerbach 1979) predicts that a firm's dividend policy has no effect on the relation between shareholder tax rates and stock price because firm value is eventually distributed as taxable dividends. In contrast, the traditional or "tax-penalty" view posits that a firm's dividend policy affects the relation between shareholder tax rates and stock price because dividend distributions trigger shareholder taxes at ordinary income tax rates, whereas other distributions (such as share repurchases) are taxed at capital-gain tax rates. Although both views predict that increases in dividend tax rates will decrease share prices, they differ with respect to whether the firm's dividend policy affects the magnitude of the price decrease. To date, empirical evidence regarding the relation among tax rate changes, dividend policy, and share values has been inconsistent. Our finding that the adverse price effect of RRA93 increased in dividend yield supports the traditional view.

Second, our evidence demonstrates that the effect of shareholder-level dividend taxation on the value of a particular stock also depends on the tax status of the marginal investor in that stock. Although theoretical models of tax effects on stock prices recognize the importance of the marginal investor, empirical studies generally do not investigate or control for this factor because of the lack of shareholder-level information. For example, although Lang and Shackelford (2000) speculate that the price effects of shareholder-level capital gains taxes on common stocks will depend, in part, upon whether the marginal investor for a particular stock is an individual taxpayer, they do not provide evidence on this point. Indeed, tax research commonly assumes that the marginal investor is subject to the highest applicable tax rate (e.g., Collins and Kemsley 2000). To our knowledge, this is the first empirical study to use an exogenous proxy for the tax status of the marginal investor to demonstrate that a dividend tax rate change affects prices in a predictable manner. (2)

Our study is closely related to a concurrent working paper, Dhaliwal et al. (2002) (hereafter DLT), which examines long-window stock returns to test whether returns are higher for dividend-paying firms, consistent with the tax-penalty (i.e., traditional) view. Like our study, DLT use the level of institutional ownership as a proxy for the tax rate or tax status of the marginal investor, reasoning that institutional owners face, on average, lower tax rates on dividend income than do individual investors. Consistent with the tax-penalty view (and our findings in the control period), DLT find that dividend yield is positively associated with long-term stock returns and that this effect decreases in the level of institutional ownership. Nonetheless, a potential weakness of long-window returns studies (such as DLT) and recent price-level studies (e.g., Harris and Kemsley 1999) is that the results are susceptible to confounding factors such as risk. Our event-study methodology isolates the price effects of dividend policy and institutional ownership during a short event window around an unexpected increase in tax rates, thereby minimizing the effects of nontax factors. Hence, our research design triangulates both existing event studies (e.g., Lang and Shackelford 2000) and long-window return studies (e.g., DLT) by simultaneously incorporating a proxy for the tax status of the marginal investor and demonstrating the effect of dividend yield in a setting less susceptible to confounding factors.

We organize the remainder of this study as follows. Section II describes our hypotheses and research method. Section III presents results. In the final section we explain the implications of our results for the competing theories that explain how shareholder-level taxes influence share values and describe the limitations of our study.

II. RESEARCH METHOD

Although both the tax-capitalization and traditional views predict that increases in dividend tax rates will decrease share prices, only the traditional view predicts that the firm's dividend policy affects the magnitude of the price decrease. A series of studies by Harris and Kemsley (1999), Collins and Kemsley (2000), and Harris et al. (2001) analyze the market values (price levels) of pooled cross-sectional samples and conclude that dividend taxes on retained earnings are fully impounded into share values, irrespective of a firm's dividend policy. Other studies criticize this conclusion (Shackelford and Shevlin 2001; Dhaliwal et al. 2001). A potential weakness of studies using price levels is that the observed effects of dividend policy might reflect differential returns for factors associated with dividends such as risk or earnings persistence (Hanlon et al. 2001). To minimize these potential confounding factors, we test for price effects associated with dividend policy during a short event window when tax rates unexpectedly increase.

If the traditional view is descriptive of how dividend tax rate increases influence stock prices, we would expect the market to react more negatively to the tax rate increase, the higher the firm's dividend yield. Because RRA93 primarily increased the tax rates of individual investors, we expect the negative association between dividend yield and returns to be concentrated in those firms owned primarily by individuals? We use the level of institutional ownership as a proxy for the likelihood that a stock's marginal investor is not an individual taxpayer. Hence, we test the following hypothesis (stated in the alternative):

H1: The share price reaction to the dividend tax rate increase enacted in RRA93 is negatively related to dividend yield for stocks with low institutional ownership.

To test whether high institutional ownership mitigates the effect of the tax rate change, we posit the following conditional hypothesis (stated in the alternative):

H2: The negative relation between dividend yield and the share price reaction to the dividend tax rate increase enacted in RRA93 is mitigated by the level of institutional ownership.

We estimate the following regression model to test our hypotheses:

(1) [CAR.sub.it] = [[gamma].sub.0] + [[gamma].sub.1] [EVENT.sub.t] + [[gamma].sub.2] [DIV.sub.i] + [[gamma].sub.3][INST.sub.i] + [[gamma].sub.4][DIV.sub.i] x [INST.sub.i]

+ [[gamma].sub.5][DIV.sub.i] x [EVENT.sub.t] + [[gamma].sub.6][DIV.sub.i] x [INST.sub.i] x [EVENT.sub.t] + [[gamma.sub.k][X.sub.ki] + [[epsilon.sub.it]

where:

[CAR.sub.it] = cumulative abnormal returns for sample firm i determined using a standard market model cumulated over the nine five-day periods beginning four weeks prior to August 3, 1993, and ending four weeks after August 9, 1993; (4)

[EVENT.sub.t] = an indicator equal to 1 if period t is August 3-9, 1993 (i.e., the RRA93-event period), and 0 for the control periods (four weeks preceding August 3, 1993, and four weeks following August 9, 1993);

[DIV.sub.i] = an indicator equal to 1 for high dividend-yield firms and 0 for no-dividend firms or, in an alternative specification, a continuous dividend-yield variable calculated as a firm's common stock dividends for the fiscal year ending prior to January 1993, deflated by firm market value at December 31, 1992; (5)

[INST.sub.i] = the percentage of firm i's common stock owned by institutional investors; and

[X.sub.ki] = a vector of other variables controlling for firm attributes (e.g., firm size, profitability, leverage, and book-to-market ratio) that may be associated with stock returns during the sample period.

To test whether the change in the dividend tax rate affected stock prices, we first estimate daily abnormal returns for each sample firm using a standard market model and stock returns obtained from the Center for Research in Security Prices (CRSP). We then cumulate abnormal returns, [CAR.sub.t], over the five-day event period commencing on August 3, 1993, and ending August 9, 1993. This five-day period begins the day before the tax rate increase cleared the Joint Conference Committee and includes the days both houses of Congress narrowly approved the legislation (by a House vote of 218-216 on August 5 and a Senate vote of 51-50 on August 6). Although Congress debated this legislation for some time prior to the enactment of RRA93, ample anecdotal evidence indicates substantial doubt of passage. (6) Nonetheless, market anticipation in advance of our estimation period would tend to bias our tests against finding a significant effect of the tax rate change on stock prices. We control for the relation between dividend policy and stock returns outside our event period by cumulating abnormal returns over four five-day control periods on both sides of the event period. In other words, our regression includes nine cumulative abnormal return estimates for each sample firm, with four control observations preceding and another four control observations following the RRA93-event observation.

We use two alternative specifications of DIV to represent the dividend policy of sample firms. In the first specification DIV is an indicator variable equal to 1 if firm i is ranked in the top three deciles of dividend yield for the fiscal year ending prior to January 1993, and 0 for non-dividend-paying firms. To increase the distinction between the two groups identified by this indicator and thereby increase the power of our tests, we exclude from this analysis dividend-paying firms that are not ranked in the top three deciles of dividend yield. (7) Our alternative specification includes all sample firms and DIV is a continuous dividend-yield variable that we calculate as a firm's common stock dividends for the fiscal year ending prior to January 1993, deflated by firm market value at December 31, 1992.

INST represents the likelihood that the marginal investor in a firm's stock is not an individual. We obtain institutional ownership data as of December 31, 1992, for each sample firm using the CDA Spectrum database. This database reports institutions' common stock ownership at the end of each calendar quarter, based on SEC form 13(f) filings. Institutions include banks, broker-dealers and other investment advisers, college endowments, mutual funds, corporate investors, insurance companies, and pension and retirement funds. In general, institutional investors face lower tax rates on dividend income than individual investors because they are either tax exempt (e.g., college endowments, pension and retirement funds) or eligible for the corporate-dividends-received deduction (e.g., any corporate investor, including banks, broker-dealers, and insurance companies organized as corporations). INST is an imperfect proxy for the likely tax status of the firm's marginal investor because it includes (1) shares owned by individuals but held in the street name of their brokers and (2) shares owned by mutual funds that are, in turn, owned by individual investors in taxable accounts (as opposed to tax-deferred pension accounts). Noise in this measure would tend to bias our tests against the expected results.

Equation (1) allows us to estimate the effects of DIV and DIV x INST during the control weeks when expectations of future tax rates are presumably constant. Our analysis controls for these differences because, according to the traditional view described above, dividend-paying firms should earn higher pretax returns than non-dividend-paying firms during the control weeks to compensate taxable investors for the tax penalty created by dividend payments. The estimated coefficient for DIV represents the average difference in abnormal returns between dividend-paying and non-dividend-paying firms with no institutional ownership throughout the control period. The coefficient on DIV x INST represents the effect of increasing institutional ownership on the overall relation between DIV and abnormal returns during the control period. Consistent with DLT, we expect a positive coefficient on DIV and a negative coefficient on DIV x INST.

We test our first hypothesis with the estimated coefficient on DIV x EVENT, which we expect to be negative. This coefficient represents the incremental difference in abnormal returns between dividend- and non-dividend-paying firms with no institutional ownership during the RRA93-event period compared to the control periods. We test our second hypothesis with the estimated coefficient for DIV x INST x EVENT This coefficient represents the effect of institutional ownership on the incremental difference in abnormal returns for dividend- vs. non-dividend-paying firms during the RRA93-event period compared to the control periods. If the level of institutional ownership represents the likelihood that the marginal investor will not be subject to the tax rate increase, then the estimated coefficient on this three-way interaction should be positive.

This research design controls for DIV and DIV x INST during the nonevent period (when expected tax rates are constant), so the coefficients on DIV x EVENT and DIV x INST x EVENT reflect the incremental effects of the RRA93 tax rate increase. This reduces the likelihood that our tests are confounded by an omitted variable that is correlated with dividend policy or institutional ownership (e.g., risk). Although we cannot completely eliminate the possibility that a nontax factor confounds our results, we know of no plausible nontax explanation for the predicted pattern of coefficients.

Our design includes several control variables. The regression includes EVENT to capture any difference in abnormal returns during the event period that is unrelated to dividend policy. We include INST as a separate variable to capture any differential abnormal returns associated with the level of institutional ownership. Following Lang and Shackelford (2000), we include additional control variables to verify that differences in firm profitability, leverage, risk, size, or industry membership do not influence our results. We represent firm profitability as the firm's return on value (ROV), defined as the firm's net income before extraordinary items for the fiscal year ended prior to January 1993, deflated by firm market value at December 31, 1992. We represent firm leverage (LEV) as the firm's total liabilities for the fiscal year ended prior to January 1993, deflated by firm market value at December 31, 1992. We use firm book-to-market ratios (BK/MKT) to control for risk and expected growth differences across sample firms, and define BK/MKT as the firm's book value of equity for the fiscal year ended prior to January 1993, deflated by firm market value at December 31, 1992. We use the log of firm market value at December 31, 1992 (SIZE), to control for size differences across sample firms. Finally, we control for industry membership using indicator variables for firm one-digit SIC code industry membership.

III. RESULTS

We estimate Equation (1) using a sample of firms with financial and stock price information available on Compustat and CRSP for the estimation period. Of the 1,755 firms in our initial sample, we eliminate 109 firms that report both losses and negative retained earnings for the fiscal year ending prior to January 1993. We impose this restriction because shareholder distributions not paid out of current or accumulated earnings and profits are not subject to the dividend tax rate. We eliminate an additional 334 firms for which institutional ownership data are not available on the CDA Spectrum database. The final sample of 1,312 firms is distributed across nine (one-digit SIC code) industries, with firms in the rubber, metal, and machines industry representing the largest portion of the sample, approximately 27.0 percent. Firms in the agriculture, forestry, and fishing industry represent the smallest portion of the sample, approximately 0.2 percent.

Table 1 presents descriptive statistics for sample firms in the aggregate, as well as partitioned by dividend policy using the dichotomous definition of DIV. Approximately 71 percent of sample firms paid some amount of dividends to common shareholders during the fiscal year ending prior to January 1993. (8) The 29 percent of firms that paid no dividends comprise the "no-dividend" (DIV = 0) group (n = 385). The top 30 percent of firms ranked on dividend yield comprise the "high-dividend" (DIV = 1) group (n = 393).

Univariate comparisons in Table 1 indicate that high-dividend firms have a higher percentage of institutional ownership than firms that do not pay dividends. Although there could be other explanations for differences in institutional ownership levels, this finding is consistent with a clientele effect for dividend-paying firms (Allen et al. 2000). The existence of tax-induced clienteles, however, biases our tests against finding support for our predictions because high-tax-rate individuals are not the natural clientele for high-dividend-yielding stocks (Dhaliwal et al. 1999; Seida 2001). Table 1 also indicates that high-dividend firms are more profitable (ROV) and larger (SIZE) than no-dividend firms. These differences highlight the importance of controlling for these firm characteristics in our regression analyses.

Table 2 presents our regression results using both the dichotomous (Column [3]) and continuous (Column [4]) specifications of DIV. (9) Although the F-statistics for both regressions are statistically significant (p = 0.00 in Column [3] and p = 0.00 in Column [4]), we caution that the overall explanatory power of the estimated regressions is quite modest (adjusted [R.sup.2] of 0.015 in Column [3] and 0.007 in Column [4]). Because the results are consistent across both specifications, we discuss only the results using the dichotomous DIV measure. To aid in the interpretation of key effects, Figure 1 plots fitted values of the dependent variable, CAR, during the control (Panel A) and RRA93-event (Panel B) periods, as a function of dividend policy and institutional ownership. Consistent with DLT, we find that abnormal returns of firms with no institutional ownership are, on average, positively related to dividend yield during the control period (i.e., the coefficient on DIV is significantly positive), and institutional ownership mitigates this positive effect (i.e., the coefficient on DIV x INST is significantly negative). Panel A of Figure 1 illustrates that during the control period with constant tax rates on dividend income, high-dividend firms with low institutional ownership earn higher abnormal returns than no-dividend firms with low institutional ownership, on average, but this difference decreases as the level of institutional ownership rises. We determine the range (in INST) over which this difference in CAR across high-dividend vs. no-dividend firms remains statistically significant. Specifically, we test whether the sum of the coefficients for DIV and DIV x INST is significantly greater than zero at varying levels of INST. The predicted CAR for a high-dividend firm is significantly higher than the predicted CAR for a no-dividend firm (p < 0.05, one-tailed test) for any value of institutional ownership less than 60 percent. These results support the view that when expected tax rates are constant, positive abnormal returns compensate individual investors for the dividend tax penalty.

[FIGURE 1 OMITTED]

Our main results concern the incremental effects of DIV and DIV x INST during the RRA93-event period vs. the control period, as indicated by the interaction of these variables with EVENT Consistent with our expectations, the coefficient for DIV x EVENT is negative and statistically significant. Also as expected, the coefficient for DIV x INST x EVENT is positive and statistically significant. These results indicate that the share price reaction to the dividend tax rate increase implemented by RRA93 is negatively related to dividend yield, and that this negative effect is mitigated by institutional ownership (i.e., the likelihood that the marginal investor was not subject to the increased individual tax rate). In particular, for any value of institutional ownership less than 56 percent, high-dividend firms experienced significantly lower abnormal returns than no-dividend firms during the event period (relative to the control period)--i.e., the sum of the coefficients for DIV x EVENT and DIV x INST x EVENT for any value of INST less than 56 percent is significantly negative (p < 0.05, one-tailed test).

Panel B of Figure 1 illustrates the interactive effect of dividend yield and institutional ownership during the RRA93-event period when the tax rate on dividend income for individual taxpayers increased. Consistent with the three-way interaction, a comparison of Panels A and B indicates that the interactive effect of dividend policy and institutional ownership differs across the control and RRA93-event periods. Because share prices of dividend-paying firms are adjusting to higher shareholder-level tax rates during the RRA93-event period, high-dividend firms with low institutional ownership earned lower abnormal returns during this period, on average, than did no-dividend firms with low institutional ownership. (10) Consistent with this difference in abnormal returns being attributable to the tax penalty on individual investors' dividend income, the difference in abnormal returns decreases as the level of institutional ownership rises (just as during the control period).

To assess the reasonableness of our regression estimates, we first derive a theoretical prediction of the abnormal returns that should occur during the event period, and then compare this predicted effect to the magnitude of the effect implied by our regression results. For simplicity, we illustrate this analysis for a stock with zero institutional ownership. We assume that a share's value equals the present value of an annuity of its annual dividend amount received in perpetuity, plus some liquidation value. Because RRA93 should affect only the portion of share value attributable to expected dividends, we do not attempt to explain how liquidation value is determined. Consequently, we compute the predicted share price effect of RRA93, [DELTA]PRICE, as follows:

(2) [DELTA]PRICE = (AnnualDividendAmount/DiscountRate) x ([[td.sub.before] - [td.sub.after]]/[1 - tg]) x [DELTA]PROB.

The first factor on the right-hand side is simply the present value of the dividend stream, which varies with the assumed discount rate. We use the average annual dividend payment of high-dividend firms ($1.43) for the numerator of this term, and assume a discount rate of 10 percent. The second factor recognizes that the shareholder tax discount on dividend income is a function of both the dividend tax rate (td) and the capital gains tax rate (tg) (Poterba and Summers 1985, 232). The numerator of this factor is the increase in the dividend tax rate. Using [td.sub.before] = 31.0 percent, [td.sub.after] = 39.6 percent, and tg = 28.0 percent (i.e., the maximum long-term capital gains tax rate in 1993), this factor implies that the portion of share value attributable to expected dividends should decline in value by 11.94 percent as a result of RRA93. The third factor on the right-hand side recognizes that, in expectation, some part of the RRA93 tax rate increase had already been impounded into share values prior to the event period. Because RRA93 passed by such a narrow margin, we assume a prior probability of 0.5 such that the change in probability during the event period (i.e., [DELTA]PROB) is also 0.5.

Under these assumptions, the predicted mean price decrease ([DELTA]PRICE) for high-dividend firms is $0.85 per share [($1.43/0.10) x 11.94% x 0.5], and the predicted event-period abnormal return for these firms is -2.77 percent (-$0.85/$30.85 average share price). In comparison, the estimated coefficient on DIV x EVENT in the dichotomous specification is -2.16 percent. Given these basic assumptions, this analysis suggests that the estimated coefficient is reasonable. (11)

With respect to our control variables, the coefficient for INST is significantly positive. Unlike the interaction terms involving dividend policy, we do not interpret this direct effect of INST as tax related. Possible nontax explanations of the positive effect of INST on abnormal returns are that institutional investors are better able to pick stocks that outperform the market due to informational advantages (Ali et al. 2002), or that management oversight by institutional investors enables firms to outperform the market. The significantly positive coefficient on EVENT indicates that, on average, during the RRA93-event period no-dividend firms realized abnormal returns that were more positive than those earned during the control period. The positive effect of EVENT could reflect other anticipated economic consequences of RRA93, or other economic events that occurred during the same week. The estimated coefficients for book-to-market ratio (BK/MKT) (in Column [4]) and the transportation and utilities industry (SIC 4000) indicator (not reported in Table 2) are both significantly positive (p = 0.00), suggesting that it is important to control for these firm characteristics when analyzing stock returns over this event period. (12)

IV. CONCLUSIONS

This study investigates the effect on share prices of an increase in the individual income tax rate for dividend income. Specifically, we examine abnormal returns on common stocks during the five-day period in 1993 when Congress enacted the Revenue Reconciliation Act of 1993, which increased the top tax rate for individuals from 31.0 percent to 39.6 percent. We predict that the tax-related effect on share prices is a joint function of firm dividend policy and the tax status of the firm's marginal investor. Consistent with our predictions, we find that the higher the firm's dividend yield, the more negative the firm's abnormal returns during the RRA93-event period. Moreover, because the 1993 tax rate increase affected only individuals, we predict and find that the level of institutional holdings, which represents the likelihood that the marginal investor in a particular stock is not an individual taxpayer, mitigates the negative effect of the firm's dividend yield on RRA93-event-period abnormal returns. Importantly, we find that the tax-related share price effects of dividend yield and institutional ownership during the RRA93-event period (in which expected future tax rates are increasing) are exactly the opposite of these effects during the control period (in which tax rates are constant).

Our study makes two contributions to the literature investigating the extent to which share values impound shareholder-level dividend taxes. First, we find that the effect of shareholder dividend taxes on the firm's share value depends, in part, on the tax status of the firm's marginal investor. Although both the tax-capitalization and traditional perspectives recognize that the relevant shareholder tax rate is that of the marginal investor, there is scant empirical evidence that the marginal investor's tax status varies across stocks. We argue that institutional ownership is a reasonable proxy for the likelihood that the marginal investor in a particular firm is not an individual taxpayer. Our conclusions conflict with those of Harris and Kemsley (1999) and the assumption of Collins and Kemsley (2000, 415) that the marginal investor for all stocks is a top-tax-rate individual taxpayer. In turn, our results reinforce the need to control for differences in the tax status of marginal investors across firms in future research investigating shareholder-level tax effects on share values.

Second, we find that firm dividend policy matters. The negative effect of dividend yield on abnormal returns during the event period in which RRA93 increased dividend tax rates is consistent with the traditional view that dividend policy determines the extent to which dividend taxes are reflected in share prices. Likewise, the positive effect of dividend yield during the control period when expectations of future tax rates were constant is also consistent with the traditional view that individual shareholders of dividend-paying firms require higher returns to compensate for the dividend tax penalty. Both results are inconsistent with a strict interpretation of the tax-capitalization view that dividend policy is irrelevant (e.g., Harris and Kemsley 1999, 281).

Finally, our results are also important from a tax-policy perspective because they demonstrate that the effect of changes in shareholder-level taxes on stock prices depends on both the firm's dividend policy and the tax status of its shareholders. Our evidence enables policymakers to better anticipate the consequences (whether intentional or not) of tax rate changes for firm market values.

Readers should interpret the results of this study in light of three limitations. First, the fluidity of dividend tax rates over time makes it difficult to know whether investors view a tax rate change as temporary or permanent, and this has implications for distinguishing between the tax-capitalization and traditional perspectives. In particular, both the traditional and tax-capitalization views predict that dividend policy affects the extent to which share values impound temporary tax rate changes (Poterba and Summers 1985, 261). For example, although the tax-capitalization view assumes that all distributions are subject to dividend taxes, only those distributions occurring during the effective period of a temporary tax increase, which is a function of dividend policy, would be discounted at the temporarily higher tax rate. Thus, if investors viewed the RRA93 tax rate increase as temporary rather than permanent, then the significant negative effects of dividend policy we document during the event period would be consistent with both theories. Nonetheless, our research design also captures the significantly positive effect of dividend policy on returns during control periods on both sides of the event period, and this result is entirely inconsistent with the tax-capitalization perspective. On balance, therefore, we interpret our evidence as supporting the traditional view that the extent to which share values impound dividend taxes depends on the firm's dividend policy.

Second, because the tax legislative process spans a time frame longer than our event window, our design likely does not capture the complete price effect of the tax rate increase. Thus, our estimates of the price effects associated with the tax rate increase likely underestimate the true effects. Finally, although the magnitude of estimated coefficients appear in line with theoretical expectations, the model's overall explanatory power is quite modest.
TABLE 1
Sample Descriptive Statistics and Tests for Differences in Means
across No-Dividend and High-Dividend Firms (a)

                           Significant              Standard
Variables                Difference (b)     Mean    Deviation

DIV YIELD
  Overall sample                            0.021     0.033
  No-Dividend Firms                         0.000     0.000
  High-Dividend Firms                       0.049     0.048

INST
  Overall sample                            0.404     0.238
  No-Dividend Firms          HD > ND        0.342     0.240
  High-Dividend Firms                       0.390     0.233

ROV
  Overall sample                            0.043     0.095
  No-Dividend Firms          HD > ND        0.023     0.150
  High-Dividend Firms                       0.052     0.064

LEV
  Overall sample                            1.738     3.163
  No-Dividend Firms     HD [approximately
                          equal to] ND      1.684     3.498
  High-Dividend Firms                       2.115     3.340

BK/MKT
  Overall sample                            0.624     0.465
  No-Dividend Firms     HD [approximately
                          equal to] ND      0.688     0.665
  High-Dividend Firms                       0.698     0.352

SIZE
  Overall sample                            5.988     1.913
  No-Dividend Firms          HD > ND        4.855     1.677
  High-Dividend Firms                       6.294     1.883

                          25%                 75%
Variables               Quartile   Median   Quartile

DIV YIELD
  Overall sample         0.000     0.016     0.031
  No-Dividend Firms      0.000     0.000     0.000
  High-Dividend Firms    0.033     0.041     0.056

INST
  Overall sample         0.203     0.407     0.591
  No-Dividend Firms      0.134     0.305     0.543
  High-Dividend Firms    0.184     0.375     0.572

ROV
  Overall sample         0.033     0.053     0.074
  No-Dividend Firms      0.017     0.043     0.070
  High-Dividend Firms    0.044     0.064     0.079

LEV
  Overall sample         0.288     0.727     1.525
  No-Dividend Firms

                         0.219     0.571     1.467
  High-Dividend Firms    0.530     1.066     1.805

BK/MKT
  Overall sample         0.332     0.556     0.793
  No-Dividend Firms

                         0.276     0.544     0.927
  High-Dividend Firms    0.520     0.645     0.828

SIZE
  Overall sample         4.653     5.957     7.318
  No-Dividend Firms      3.743     4.736     5.940
  High-Dividend Firms    5.092     6.271     7.535

(a) Our overall sample consists of 1,312 firms. The "no-dividend
firms" subsample consists of the 385 firms that did not pay dividends
in the fiscal year ending prior to January 1993. The "high-dividend
firms" subsample consists of the 393 firms ranked in the top 3 deciles
of dividend yield for the fiscal year ending prior to January 1993.

(b) Significant differences between the no-dividend and high-dividend
subsamples are based on t-tests of means (p < 0.05).

DIV YIELD = common stock dividends for the fiscal year ending prior to
            January 1993, deflated by firm market value at December
            31, 1992;
INST = the percentage of common stock at December 31, 1992, held
       by institutional investors;
ROV = net income before extraordinary items for the fiscal year
      ending prior to January 1993, deflated by firm market
      value at December 31, 1992;
LEV = total liabilities for the fiscal year ending prior to
      January 1993, deflated by firm market value at December
      31, 1992;
BK/MKT = book value for the fiscal year ending prior to January
         1993, deflated by firm market value at December 31, 1992;
         and
SIZE = log of the firm's market value at December 31, 1992.

TABLE 2
Pooled, Cross-Sectional OLS Regressions of Cumulative Abnormal Returns
on Independent and Control Variables Using Dichotomous and Continuous
Measures of Dividend Yield (a)

                                Dichotomous DIV (b)  Continuous DIV (c
Variables           Prediction  Parameter Estimates (t-statistics) (d)
(1)                    (2)              (3)                 (4)

Intercept               ?             -0.0156             -0.0118
                                     (-4.77)             (-4.97)
DIV                     +              0.0133              0.1379
                                      (6.58)              (4.70)
DIV x INST              -             -0.0184             -0.3292
                                     (-4.47)             (-4.57)
DIV x EVENT             -             -0.0216             -0.2540
                                     (-5.48)             (-4.61)
DIV x INST x EVENT      +              0.0294              0.5040
                                      (4.14)              (3.70)
EVENT                   ?              0.0075              0.0022
                                      (3.09)              (1.75)
INST                    ?              0.0093              0.0054
                                      (2.47)              (2.21)
ROV                     ?             -0.0102             -0.0054
                                     (-1.91)             (-0.78)
LEV                     ?              0.0001              0.0001
                                      (0.01)              (0.14)
BK/MKT                  ?              0.0017              0.0030
                                      (1.54)              (3.01)
SIZE                    ?              0.0002              0.0004
                                      (0.65)              (1.83)
Adjusted [R.sup.2]                     0.015               0.007
Sample Size                            6,706              11,177

(a) Cumulative daily abnormal returns (CAR) are the sum of each sample
firm's daily market model abnormal returns for each five-day event or
control period. We estimate the market model parameters for each
sample firm by regressing firm daily returns for calendar year 1992 on
the CRSP value-weighted market index. EVENT is equal to 1 for the
week of 8/3/93-8/9/93, and 0 otherwise (i.e., 0 for the four control
weeks preceding and following the event week). Table 1 defines all
other variables.

(b) In the dichotomous specification, DIV is equal to 1 for firms
ranked in the top three deciles of dividend yield for the fiscal year
ending prior to January 1993, and 0 for non-dividend-paying firms.
We exclude dividend-paying firms not ranked in the top three deciles
of dividend yield from this specification.

(c) In the continuous specification, we include all sample firms in
the regression and define DIV as the dividends the firm paid to common
shareholders during the fiscal year ending prior to January 1993,
deflated by firm market value at December 31, 1992.

(d) Reported results are after removing the effects of influential
observations identified by Belsley et al. (1980) procedures (i.e.,
absolute R-student value exceeding 2). We calculate t-statistics using
the Froot (1989) adjustment that controls for cross-sectional
dependence across sample observations. We estimate both regressions
with indicator variables (coefficient estimates suppressed) to control
for (one-digit SIC code) industry effects.


The authors appreciate the constructive comments of Ken Klassen, two anonymous reviewers, Dennis Chambers, Merle Erickson, Deen Kemsley, Teresa Lightnet, Ed Maydew, James Myers, Mary Margaret Frank, Lil Mills, Sue Porter, Terry Shevlin, Connie Weaver, Dave Ziebart, and seminar participants at The University of Oklahoma, The University of Texas at Austin, and Virginia Polytechnic Institute and State University. Professor Ayers acknowledges the financial support of the J. M. Tull School of Accounting, the Terry College of Business, and the Sanford-Terry Research Grant Program. Professor Robinson acknowledges the financial support of the Red McCombs School of Business.

Submitted September 2001

Accepted May 2002

(1) Poterba and Summers (1985) and Zodrow (1991) provide thorough comparisons of the tax-capitalization and tax-penalty perspectives.

(2) Prior studies investigate whether the yield spread for municipal bonds reflects the tax rate for an individual or corporate marginal investor (Poterba 1986; Fortune 1988; Mankiw and Poterba 1996) and whether there is an institutional tax clientele for dividend-paying stocks (Dhaliwal et al. 1999; Seida 2001).

(3) RRA93 also increased the top corporate tax rate from 34 to 35 percent. After accounting for the standard 70 percent corporate-dividends-received deduction, this 1 percent tax rate increase corresponds to a maximum increase in tax rates on dividends of 0.3 percent for corporate investors. This change would tend to bias our tests against H2.

(4) We estimate the market model parameters for each sample firm by regressing firm daily returns for calendar year 1992 on the value-weighted market index. We use market-adjusted abnormal returns despite the inclusion of many dividend-paying firms in the market return. Consequently, our tests should be biased against rejection because the market adjustment could remove part of the price effect we seek to estimate.

(5) Investors might expect firms to change their dividend policies as a result of the higher tax rate on dividends paid to individual investors. We also estimated our regressions using dividends paid during the fiscal year ending after December 1993 as a proxy for a firm's expected dividend yield. Results are essentially identical to those presented later in Table 2.

(6) Prior to the Joint Conference Committee, the Senate version of RRA93 passed by the narrowest of margins (i.e., 51-50) on June 25, 1993, relying on the tie-breaking vote of Vice President Gore. Passage of the Joint Conference Committee version remained doubtful until the final Senate vote (e.g., Calmes 1993). In sensitivity analysis, we repeated our analyses considering legislative dates prior to August 3, 1993, as event dates. Consistent with speculation that the passage of the tax rate increase was uncertain prior to the congressional votes in August 1993, we find no evidence of abnormal returns surrounding the legislative events prior to August 3, 1993. Nonetheless, it is possible that the market gradually impounded the promised tax rate increase during the legislative process or during the preceding presidential campaign. Indeed, in their investigations of political campaign rhetoric, Slemrod and Greimel (1999) and Ayers et al. (2002) present evidence that market prices reflect the tax rhetoric of opposing presidential candidates.

(7) In sensitivity analysis, we include all sample firms in our analysis and define DIV as an indicator variable equal to 1 if firm i pays dividends during the fiscal year ending prior to January 1993, and 0 for non-dividend-paying firms. Although the results of this analysis are less pronounced than those presented later in Table 2, Column (3), conclusions are the same.

(8) In comparison, 65 percent of the 1,755 firms with available Compustat and CRSP data paid dividends in the fiscal year ending prior to January 1993.

(9) Statistical tests are based on Froot (1989) adjusted standard errors to control for cross-sectional dependence arising for sample observations sharing the same event date.

(10) Event period CARs for high-dividend firms are significantly lower than those for no-dividend firms at all levels of institutional ownership below 34 percent (p < 0.05, one-tailed test). We determine this range of significance by testing whether the sum of the coefficients for DIV, DIV x INST, DIV x EVENT, and DIV x INST x EVENT is significantly less than zero at varying levels of INST.

(11) The -2.16 percent value for the DIV x EVENT coefficient implies that this coefficient captures approximately 78 percent of the predicted effect. Using a similar approach, the coefficient estimate based on the continuous DIV measure (Table 2, Column [4]) explains approximately 43 percent of the predicted effect.

(12) Collinearity diagnostics verify that the correlation between firms' dividend policy and institutional ownership does not significantly affect our results. The highest condition indices for each analysis range from 18 to 20, well below the upper bound of 30 that Belsley et al. (1980) suggest as problematic. In sensitivity analyses we included an INST x EVENT interaction term in each of the Table 2 regression analyses. Although including INST x EVENT induces high levels of multicollinearity, conclusions are unaffected.

REFERENCES

Ali, A., C. Durtschi, B. Lev, and M. Trombley. 2002. Changes in institutional ownership and subsequent earnings announcement abnormal returns. Working paper, The University of Arizona, Tucson, AZ.

Allen, F., A. E. Bernaldo, and I. Welch. 2000. A theory of dividends based on tax clienteles. The Journal of Finance 55 (December): 2499-2536.

Auerbach, A. J. 1979. Wealth maximization and the cost of capital. Quarterly Journal of Economics 93 (August): 433-446.

Ayers, B. C., C. B. Cloyd, and J. R. Robinson. 2002. "Read my lips ...": Does the market believe the tax rhetoric of presidential candidates? Working paper, University of Georgia, Athens, GA.

Belsley, D. A., E. Kuh, and R. E. Welsch. 1980. Regression Diagnostics: Identifying Influential Data and Sources of Collinearity. New York, NY: John Wiley & Sons.

Calmes, J. 1993. Deficit plan survives the House, 218-216, in hard-won vote uncertain until end. Wall Street Journal (August 6): A3.

Collins, J., and D. Kemsley. 2000. Capital gains taxes and dividend taxes in firm valuation: Evidence of triple; taxation. The Accounting Review 75 (October): 405-427.

Dhaliwal, D., M. Erickson, and R. Trezevant. 1999. A test of the theory of tax clienteles for dividend policies. National Tax Journal 52 (June): 179-194.

--, --, M. M. Frank, and M. Banyi. 2001. Are shareholder dividend taxes on corporate retained earnings impounded in equity prices? Additional evidence and analysis. Working paper, University of Chicago, IL.

--, O. Z. Li, and R. Trezevant. 2002. Is a dividend tax penalty incorporated into common stock return? Working paper, The University of Arizona, Tucson, AZ.

Engel, E., M. Erickson, and E. Maydew. 1999. Debt-equity hybrid securities. Journal of Accounting Research 37 (Autumn): 249-274.

Erickson, M. M., and E. L. Maydew. 1998. Implicit taxes in high dividend yield stocks. The Accounting Review 73 (October): 435-458.

Fortune, P. 1988. Municipal bond yields: Whose tax rates matter? National Tax Journal 41 (June): 219-233.

Froot, K. 1989. Consistent covariance matrix estimation with cross-sectional dependence and heteroskedasticity in financial data. Journal of Financial and Quantitative Analysis 24 (September): 333-355.

Guenther, D. 1994. The relation between tax rates and pretax returns: Direct evidence from the 1981 and 1986 tax rate reductions. Journal of Accounting and Economics 18 (November): 379-393.

Hanlon, M., J. Myers, and T. Shevlin. 2001. Dividend taxes and firm valuation: A re-examination. Working paper, University of Washington, Seattle, WA.

Harris, T. S., and D. Kemsley. 1999. Dividend taxation in firm valuation: New evidence. Journal of Accounting Research 37 (Autumn): 275-292.

--, R. G. Hubbard, and D. Kemsley. 2001. The share price effect of dividend taxes and tax imputation credits. Journal of Public Economics 79 (March): 569-596.

Lang, M., and D. A. Shackelford. 2000. Capitalization of capital gains taxes: Evidence from stock price reactions to the 1997 rate reduction. Journal of Public Economics 76 (April): 69-85.

Mankiw, N., and J. Poterba. 1996. Stock market yields and the pricing of municipal bonds. NBER Working Paper No. 5607. Cambridge, MA: National Bureau of Economic Research.

Poterba, J., and L. H. Summers. 1985. The economic effects of dividend taxation. In Recent Advances in Corporate Finance, edited by E. Altman, and M. Subrahmanyan, 227-284. Homewood, IL: R. D. Irwin.

--. 1986. Explaining the yield spread between taxable and tax exempt bonds. In Studies in State and Local Public Finance, edited by H. Rosen, 5-48. Chicago, IL: University of Chicago Press.

Seida, J. A. 2001. Evidence of tax-clientele-related trading following dividend increases. The Journal of the American Taxation Association 23 (Supplement): 1-21.

Shackelford, D., and T. Shevlin. 2001. Empirical tax research in accounting. Journal of Accounting and Economics 31 (September): 321-387.

Slemrod, J., and T. Greimel. 1999. Did Steve Forbes scare the municipal bond market? Journal of Public Economics 74 (October): 81-96.

Zodrow, G. 1991. On the "traditional" and "new" views of dividend taxation. National Tax Journal 44 (December): 497-509.
Benjamin C. Ayers
University of Georgia

C. Bryan Cloyd
University of Illinois at Urbana-Champaign

John R. Robinson
The University of Texas at Austin
COPYRIGHT 2002 American Accounting Association
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2002 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Ayers, Benjamin C.; Cloyd, C. Bryan; Robinson, John R.
Publication:Accounting Review
Date:Oct 1, 2002
Words:7722
Previous Article:How informative are value-at-risk disclosures?
Next Article:The relation between cost shifting and segment profitability in the defense-contracting industry.
Topics:

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |