Jennifer L. Blouin University of North Carolina. Jana Smith Raedy University of North Carolina

Transcription

1 &DSLWDO*DLQV7D[HVDQG(TXLW\7UDGLQJ(PSLULFDO(YLGHQFH Jennifer L. Blouin University of North Carolina Jana Smith Raedy University of North Carolina Douglas A. Shackelford 1,2 University of North Carolina University of Maastricht (visiting) NBER Abstract Individual investors have an incentive to sell appreciated (depreciated) stock after (before) it qualifies for long-term capital gains treatment. This paper presents evidence that these incentives affect stock returns and trading volume around two disclosures, quarterly earnings announcements and additions to the S&P 500 index. Specifically, we find equity constricts (expands) and prices rise (drop) in our estimate of the incremental taxes (tax savings) generated, if appreciated (depreciated) property is sold before it qualifies for long-term treatment. The price pressure is temporary, reversing in subsequent trading days. &ODVVLILFDWLRQFRGHV: H22, H24, G12, G14, M41.H\ZRUGV: Capital gains taxes, stock price, trading volume, quarterly earnings announcements, S&P 500 index 1 Douglas A. Shackelford Kenan-Flagler Business School University of North Carolina Campus Box 3490, McColl Building Chapel Hill, NC (phone) (fax) doug_shack@unc.edu ( ) 2 We appreciate the helpful comments from Jeff Abarbanell, Mary Barth, Robert Bushman, Merle Erickson, Robert Lipe, Ed Maydew, Kevin Raedy, and workshop participants at the College of William & Mary, Duke University, London Business School, University of North Carolina, and Southeastern Summer Accounting Research Conference. We also acknowledge the contribution of I/B/E/S International Inc. for providing earnings per share forecast data available through the Institutional Brokers Estimate System. These data have been provided as part of a broad academic program to encourage earnings expectations research.

2 &DSLWDO*DLQV7D[HVDQG(TXLW\7UDGLQJ(PSLULFDO(YLGHQFH,QWURGXFWLRQ The purpose of this paper is to determine the extent to which share prices and trading volume are affected by individual tax incentives to shift sales around the long-term capital gains holding period. Under current U.S. law, individuals generally have an incentive to hold appreciated stock for more than one year because long-term capital gains are taxed at no more than 20 percent. Sales of shares held for a shorter period (short-term capital gains) are taxed at up to 39.1 percent. Conversely, individuals have an incentive to sell depreciated shares before one year lapses, creating short-term capital losses that offset the tax-disfavored short-term capital gains, recovering 39.1 cents per dollar of loss. Depreciated securities sold after one year of ownership are deducted initially against long-term capital gains; thus refunding only 20 cents for each dollar of loss. 1 Shackelford and Verrecchia [2001] analyze the impact of such tax incentives on equity prices and trading volume around disclosures. Using a stylized model of trade, they show that the holding period incentive can constrict the supply of equity and bid up share prices around the release of good news to the market. In short, tax incentives to defer sales of appreciated property until long-term qualification restrain the portfolio rebalancing that would occur in the absence of taxes. To induce selling, buyers must compensate sellers through higher share prices for the additional taxes associated with short-term capital gains. Opposite effects are predicted for depreciated shares. That is, bad news is associated with equity expansion, increased trading, and falling prices. 1 Complex rules govern the netting of capital gains and losses. The general rules apply (i.e., short-term losses offset short-term gains and long-term losses offset long-term gains) as long as an individual s long-term capital gains equal or exceed long-term capital losses and his short-term capital gains equal or exceed short-term capital losses (Shackelford [2000]). Data from individual tax returns discussed below suggest that this presumption is not unreasonable. On the other hand, if total personal capital losses exceed total personal capital gains during a year, then the distinction between short-term and long-term is irrelevant for that year and the tax deduction is limited to $3000, a constraint that Poterba [1987] and Auerbach, Burman and Siegel [2000] seldom find binding. Capital losses in excess of $3000 are carried forward indefinitely and included in future years calculations of capital gains and losses. These future deductions retain their short-term and long-term character; however, the importance of this distinction diminishes with the time value of money.

3 Guided by Shackelford and Verrecchia [2001], this study investigates equity trading around two unrelated public disclosures that are known to trigger substantial portfolio rebalancing and thus accentuate the possibility of tax-motivated trading: quarterly earnings announcements and changes to the Standard & Poor s 500 index. We find evidence consistent with personal holding period incentives impacting equity trading in both settings. In particular, we find equity constricts (expands) and prices rise (drop) in our estimate of the incremental taxes (tax savings) generated if appreciated (depreciated) property is sold before it qualifies for long-term treatment. The price pressure is temporary, reversing in subsequent trading days, which implies that preferential treatment for long-term capital gains increases stock market volatility. An incremental contribution of this study is that it documents a more general influence for personal holding period incentives on equity trading than previously reported. Prior studies had documented that holding period incentives could affect share prices and trading volume under special conditions. For example, Poterba and Weisbenner [2001] link holding period incentives to the January effect, implying that year-end tax planning may be important enough to move prices, but leaving open the question of shareholder tax effects during the rest of the year. 2 In this paper, the quarterly earnings announcements results show that the effect occurs throughout the year. Similarly, Reese [1998] reports price pressure when the initial public shareholders (i.e., those who buy at the initial public offering) first qualify for long-term treatment. 3 However, firms in their first months of operation typically differ from other companies, with disproportionate ownership by individuals and illiquid trading, both of which bias in favor of finding tax-induced price pressure. In this paper, the S&P 500 index results document tax- 2 Poterba and Weisbenner [2001 find that turn-of-the-year returns for depreciated firms were greatest from 1970 to 1976 and in 1985 and 1986, years when half of any net long-term capital losses expired unused while short-term capital losses could be fully deducted. They interpret this result as consistent with price reversal following a taxinduced, year-end sell-off intended to ensure short-term capital loss treatment. 3 Reese [1998] reports that from trading volume increased and prices fell for appreciated firms when their initial public shareholders first qualified for long-term capital gains tax treatment. This is consistent with a surge in selling pressure (when the lower rates first apply) that could not be met at the current market price. Conversely, for depreciated firms he finds that investors sold disproportionately immediately before qualification to ensure deductions at the high short-term rate. He adds that the relations do not hold for IPOs from , consistent with the incentive weakening after the Tax Reform Act of 1986 narrowed the spread between long-term and short-term capital gains tax rates. 2

4 motivated price pressure for some of the largest, most closely followed, and most efficiently priced U.S. companies. More generally, this paper contributes to the current debate in accounting circles about the extent to which equity prices impound shareholder taxes (see Landsman and Shackelford [1995], Erickson [1998], Erickson and Maydew [1999], Guenther and Willenborg [1999], Harris and Kemsley [1999], Collins and Kemsley [2000], Lang and Shackelford [2000], Dhaliwal, et al. [2001], Hanlon, Myers and Shevlin [2001], Shackelford and Shevlin [2001], among others). By documenting price pressure from a tax provision that only affects individuals, this paper provides evidence that capital gains tax planning by individual shareholders can move prices. The remainder of the paper develops in the following manner: Section 2 develops hypotheses that follow from Shackelford and Verrecchia [2001]. Section 3 lists additional necessary conditions for the holding period to affect equity trading. Section 4 and 5 present the empirical findings. Concluding remarks follow. +\SRWKHVLV'HYHORSPHQW Shackelford and Verrecchia [2001] develop a three-period model of trade with two groups of investors and two investments: a risky, taxed asset and a riskless, tax-free asset. The investors are identical, except that they are initially endowed with different shares of each asset. In the first period, both groups await a public disclosure. When the information is disclosed in the second period, investors rebalance. In the third period, all assets are liquidated and consumed. Gains on sales during the second period are taxed (short-term capital gains) at a higher rate than gains on sales in the third period (longterm capital gains). The disclosure is assumed to lead to homogeneous expectations about the uncertain value of the risky asset. In the absence of taxes, all investors would gravitate toward an equilibrium in which the risks associated with holding the risky asset were optimally shared. Investors who were overweighted in the risky asset, compared with the optimal risk-sharing amount, would unwind their positions by selling 3

5 shares to underweighted investors. After a good news disclosure, this selling would ensure a certain profit and eliminate the risk of being overweighted in an asset whose future value is uncertain. If the gain is taxed, then the overweighted investor must choose between selling the shares at disclosure and paying the higher short-term capital gains tax on the certain profits or retaining the shares and paying the lower long-term tax on the profits, if any, at liquidation. Thus, at disclosure, investors choose between an optimal risk strategy and an optimal tax strategy. 4 Shackelford and Verrecchia [2001] show that under these circumstances investors will sell less at the disclosure that they would in the absence of taxes. The greater the appreciation, the greater the additional tax, the greater the incentive to defer selling, and the greater the reduction in selling. Furthermore, the wider the spread between shortterm and long-term capital gains tax rates, the greater the incentive to defer selling. To entice sellers, buyers must provide compensation in the form of higher sales prices. A critical assumption throughout the model is that all investors are subject to the same tax. If the tax only applies to some investors, then the outcome becomes unclear. For example, if the overweighted investors are subject to tax, but the underweighted investors are not, then trading volume will fall and share price rise, as described above, because the selling shareholders are taxed. However, if the overweighted investors are not subject to tax, but the underweighted investors are, then the holding period incentive will not affect trading volume or share prices because the selling shareholders are untaxed. Currently in the U.S., the personal tax is the only one with different capital gains tax rates depending on the holding period. The personal tax applies to the direct holdings of individuals and their investments in entities that flow-through capital gains and losses, including mutual funds on personal account, partnership units, shares in S or limited liability corporations, and trusts in which the individual is a beneficiary. The personal tax does not apply to shares held by C corporations, pensions (including 4 An anecdote in 7KH :DOO6WUHHW-RXUQDO (July 10, 2000, A1) illustrates this dilemma, detailing the fateful sell/hold decision of a shareholder in a highly appreciated, Internet IPO.... Mr. Seiff realized that if he sold the bulk of his shares before April 13 a year and a day after he d first exercised his options he would have to pay the far higher tax rates for short-term capital gains. That would mean an additional $100,000 in taxes. Mr. Seiff would still be ahead so long as Scient s shares didn t fall far below $100 in the next two weeks. I was gambling, he says, 4

6 IRAs and 401(k) investments in mutual funds), tax-exempt organizations and foreign investors. Thus, it is an empirical question whether the personal holding period incentives are important enough in the aggregate to affect equity trading. 5 The purpose of this paper is to provide empirical documentation by testing the following hypothesis, provided in alternative form: H 1 : The incremental taxes (tax savings) from selling appreciated (depreciated) stock, which arise from the different tax treatment accorded short-term and long-term stock, increase (decrease) stock returns and decrease (increase) trading volume around public disclosures. 2WKHU1HFHVVDU\&RQGLWLRQV Before proceeding to the empirical tests, it is important to note two relevant features of the U.S. taxation of individual capital gains and losses that Shackelford and Verrecchia [2001] exclude from their model. First, the model imposes conditions that force a tax-risk tension around the holding period. In reality, the current holding period is not a binding constraint for many individual investors. Some investors buy and hold shares for many years. Others churn much more rapidly (e.g., day traders). On the other hand, Burman and Ricoy [1997] report that 49 percent of the stock sold in 1993 had been held for more than one year. In other words, the current holding period of one year approximates the median holding period for stocks. Thus, it seems plausible that the current holding period is an important consideration for a substantial number of investors. Second, as noted above, the marginal tax rate applied to a capital gain and loss depends on complex rules concerning the netting of gains and losses. In his detail of these provisions (which are Wouldn t you wait two weeks for $100,000? Unfortunately, as this investor waited for the favorable long-term capital gains tax rates, his stock tumbled from $132 to $30 per share. 5 Enough shares are potentially subject to personal taxes that it is not infeasible to conjecture that the holding period incentives affect share returns and trading volume. Among our sample firms for the earnings announcement (S&P 500) tests, we find that, on average, 38 (48) percent of their shares are held by institutions filing form 13-F: 24 (29) percent by mutual funds, 8 (11) percent by banks, 3 (5) percent by insurers and 3 (3) percent by pensions. Except for trades by pensions, which are never subject to personal taxation, the gains and losses for the other institutional holdings may be subject to individual capital gain and loss treatment. Individual capital gains and losses include sales of personal holdings, street-name holdings, mutual funds, partnerships, S corporations, limited liability corporations, trusts, and other entities that pass-through taxable gains and losses. Thus, the taxation of many shares held by financial institutions may flow directly to individual tax returns. At the same time, individuals also hold shares through institutional accounts that are unaffected by the personal taxation, such as closely-held C corporations, individual retirement accounts, 401(k) retirement accounts, and other defined contribution plans with 5

7 beyond the scope of this paper), Shackelford [2000] proves that, under current law, distinctions between long-term and short-term are relevant if and only if the selling shareholder s long-term capital gains equal or exceed his long-term capital losses and his short-term capital gains equal or exceed his short-term capital losses. Some individuals will not meet these conditions and thus face the same marginal tax rate, regardless of the holding period. However, the bull market during the 1980s and 1990s and inflation in earlier years (recall taxes are assessed on nominal, not real, profits) have caused these conditions to hold for many investors in recent years. For example, the Internal Revenue Service [1999D, 1999E] reports that in 1997 individuals in the maximum tax bracket (39.6 percent), who account for 61 percent of all net capital gains, reported $169 billion of long-term capital gains and only $5 billion of long-term capital losses and $16 billion of short-term capital gains and only $8 billion of short-term capital losses. 6 In summary, these additional two factors (that the holding period approximates investment horizons and that individuals gains and losses mix in a way that creates a tax incentive) are additional necessary conditions for the holding period incentive to affect equity trading. Recent data suggest that it is conceivable that both hold for a preponderance of individual investors. However, one explanation for a failure to detect a relation between the holding period incentive and equity trading would be that either condition does not hold for enough investors. 4XDUWHUO\(DUQLQJV$QQRXQFHPHQWV The hypothesis that holding period incentives affect equity trading is tested using two, unrelated public disclosures: quarterly earnings releases and changes to the Standard and Poor s 500 index. We focus on these disclosures because each releases (non-tax) information that is known to trigger extensive portfolio rebalancing. The unusually high level of trading activity associated with these disclosures individual investments in mutual funds being particularly problematic because these accounts can be either taxable (e.g., personal account) or non-taxable (e.g., IRA). 6 As an aside, the magnitude of these capital gains taxes serve as prima facie evidence rejecting claims that widespread tax evasion (see discussions in Poterba [1987] and Landsman, Shackelford and Yetman [2001], among others) or tax avoidance (see discussions in Constantinides [1983, 1984], Stiglitz [1983], Shackelford [2000], Scholes, et al. [2001], among others) renders capital gains taxes irrelevant for individual equity trading. 6

8 should create the conditions where individual investors must choose between an optimal risk strategy and an optimal tax strategy. 7 Our empirical analysis begins with the earnings announcements, where we regress abnormal returns (abnormal volume) on a measure of the additional taxes under short-term treatment, compared with long-term treatment, and control variables, including unexpected earnings. A positive (negative) coefficient on the incremental tax measure will be interpreted as evidence that the holding period incentives increase (decrease) stock returns (trading volume) when earnings are released. 6DPSOH All 126,270 firm-quarters from 1983 to 1999 on CRSP, IBES, and Compustat s industrial annual, full coverage, and research files are examined. Firms are deleted from the final sample if data are missing (13,646), unexpected earnings are zero or the firm experienced no price change (12,446), or earnings are negative (15,262). 8 Tests are conducted on the remaining sample of 84,916 observations. When additional controls are included in the regression, data are missing for 10,599 firms, leaving a subset of 74,317 observations. 9DULDEOHV 7 An alternative approach would be to select a sample of trading days based on high volume and presume that information was released to the markets on those days. An advantage of this strategy would be that it ensures traders were active on the sample days and thus more likely to encounter risk/tax tradeoffs. A disadvantage of this selection process is that it would introduce subjectivity. For example, the definition of high volume could bias the findings. If high volume means the number of shares traded, then the sample could include a disproportionate number of days with single, large institutional trades, biasing against findings that individual tax incentives matter. If high volume means the number of trades, then the sample could include a disproportionate number of days with small, individual trades, biasing in favor of rejecting the null. To avoid these problems associated with identifying the sample based on ex post criteria, we choose the objective, ex ante disclosure events. 8 The rationale for the screens is as follows: (1) Firms with zero unexpected earnings (defined as reported quarterly earnings less the median IBES forecast within the 60 days preceding the release date, scaled by the firm s share price at the end of the quarter including the earnings announcement) or zero price change should have no price response and thus no tax distortion. (2) Hayn [1995] documents that profitable and loss firms have different earnings response coefficients and thus perhaps should not be evaluated together. However, results are qualitatively unaltered if these screens are ignored. 7

9 The dependent variable in the stock returns tests is a conventional measure in tests of priceearnings relations, the cumulative, buy-and-hold, market-adjusted return (&$5) for trading days t-1 and t, where day t is the earnings announcement date. 9 Table 1, which presents descriptive statistics for several regression variables, shows the mean (median) &$5 is 0.38 (0.16) percent. The dependent variable in the trading volume tests is abnormal volume ($9), based on measures in Ajinkya and Jain [1989] and Lynch and Mendenhall [1997]. We measure a firm s volume as the natural logarithm of dollar volume on day t divided by the natural logarithm of the dollar value of outstanding shares on day t, relying on prior reports that this transforms raw trading volume data to a distribution closer to normal. We regress this ratio on an identical measure of total market volume over 100 consecutive days, ending on day t-2. The intercept and slope coefficients from this estimation are then used to establish an expected two-day average volume (beginning on day t-1), based on actual volume for the entire market during the period. The difference between actual and expected volume is the dependent variable. Mean (median) $9 is 1.55 (1.25) percent. The primary control variable is unexpected earnings (8(). Unexpected earnings are computed as reported quarterly earnings less the median IBES forecast within the 60 days preceding the release date, scaled by the firm s share price at the end of the quarter preceding the announcement. 10 Mean (median) 8( is 0.05 (0.04) percent. A positive regression coefficient estimate is anticipated on 8( in the returns regression, consistent with the well-documented positive correlation between abnormal returns and unexpected earnings. 7D[0HDVXUH The ideal tax measure would capture the change in capital gains taxes, if any, that individual shareholders encounter if they sell shares when earnings are announced and face the short-term tax rate 9 Results are qualitatively unaltered when we employ an alternative measure as the dependent variable, the two-day, cumulative, buy-and-hold abnormal return using a market model where beta is estimated for the 100 days, ending two days before earnings are released. 10 Extreme values of 8( are winsorized at the 1 percent and 99 percent levels to mitigate the influence of data errors. 8

10 rather than sell in the future when the long-term rate applies. Unfortunately, we cannot observe individual investors marginal tax rates, holding periods, or total portfolio of realized gains and losses, all of which are necessary to compute the ideal tax measure. Instead we employ a cruder measure, the product of a rate difference ('5$7() and a taxable base measure ( %$6(). '5$7( is the maximum federal short-term capital gains tax rate less the maximum federal longterm capital gains tax rate at disclosure. 11 '5$7( is 30 percent before 1987, 10.5 percent in 1987, 0 from , 3 percent from , 11.6 percent from 1993 to May 6, 1997, and 19.6 percent since then. %$6( is the percentage change in stock prices during the previous six, 12 or 18 months, ending after trading on day t-2. The applicable duration depends on the long-term capital gains holding period at the time of the earnings announcement. The holding period is one year for all days, except June 23, 1985 through June 30, 1988, when it is six months, and July 29, 1997 through December 31, 1997, when it is 18 months. 12 This duration is selected because the difference between long-term and short-term rates is most relevant for investors who are nearest long-term qualification at the earnings announcement. Thus, %$6(is computed as though the marginal investor is an individual who has held the stock for precisely one day less than necessary to obtain long-term capital gains tax treatment. Such an individual would have the greatest incentive to postpone the sale of an appreciated stock to garner long-term capital gains treatment or accelerate the sale of a depreciated share to ensure short-term capital loss treatment. Mean 11 Assuming individuals face the holding period incentive are in the top tax bracket seems reasonable based on the latest analysis of individual income tax returns (1997 returns) by the Statistics of Income (Internal Revenue Service, [1999a, 1999b]). They find that individuals in the maximum tax bracket (39.6 percent) account for 61 percent of all net capital gains (long-term and short-term capital gains less long-term and short-term capital losses). The percentage increases to 75 percent when individuals in penultimate bracket (36 percent) are also considered. 12 Throughout the investigation period, the long-term capital gains holding period is determined by the date of sale with one exception. The holding period is six months for assets purchased after June 22, 1984 and before January 1, Therefore, it is unclear whether investments sold from December 24, 1984 through June 22, 1985 face the new six-month holding period or the prior 12-month holding period. We assume a 12-month holding period; however, results are qualitatively insensitive to assuming a 6-month holding period. Similarly, sales during the first half of 1988 may have faced either a six-month holding period or a 12-month holding period. Because no sale during the first half of 1988 could have qualified for long-term treatment unless it had been purchased before

11 (median) %$6( is 20 (10) percent with a standard deviation of 60 percent. Sixty-three percent report appreciation during the holding period. The primary variable of interest in this study is the product of '5$7( and %$6(. Its mean (median) is (0.003) with a standard deviation of In other words, 2.4 percent of a stock s value, on average, will be forfeited in higher taxes, if it is sold before long-term qualification. Although its components are measured imprecisely and rely on the various assumptions that have been detailed above, a strength of this tax measure is that it is unlikely to be correlated with any determinant of the price-earnings relation. By interacting tax rate spreads that range across time from zero to 30 percent with firm-level changes in stock prices over the last 6 to 18 months, the tax measure should avoid specification concerns arising from omitted correlated variables. 13 We predict a positive coefficient on '5$7(* %$6( when the dependent variable is abnormal returns. A positive coefficient will be interpreted as evidence that tax considerations boost (decrease) prices when shareholders are selling appreciated (depreciated) stock. A negative coefficient is expected for '5$7(* %$6( when the dependent variable is abnormal volume. A negative coefficient will be interpreted as evidence that equity constricts (expands) when shareholders are selling appreciated (depreciated) stock. Both components of the tax measures ('5$7( and %$6() are included separately in the regression without prediction for the sign of their coefficients. One possibility for a positive coefficient on %$6( would be the incentive to defer selling appreciated property to exploit the time value of money and/or the step-up in tax basis at death (see Holt and Shelton [1962], Landsman and Shackelford [1995], and thus faced a six-month holding period, we assume a six-month holding period for all sales in the first half of 1988; however, results are qualitatively insensitive to assuming a 12-month holding period. 13 To illustrate, suppose a firm s stock price increases one percent monthly throughout the investigation period. Its tax measure would be 3.8 percent [0.3*{(1.01) 12 1}] in 1983 when the spread was 30 percent and the holding period was 12 months; 1.8 percent in 1985 when the holding period fell to six months; zero in 1989 when the long-term and short-terms were identical; 0.4 percent in 1992 when the spread was 3 percent and the holding period was 12 months; 1.5 percent a year later when the spread increased to 11.6 percent; and 3.8 percent in the second half of 1997 when the spread increased to 19.6 percent and the holding period to 18 months. Introducing cross-temporal price variation for each firm introduces further variation in the tax measures, making it even more unlikely for taxes to be correlated with any omitted factor. 10

12 Burman [1999], Klein [1999], among many others). The importance of this lock-in effect should vary with the long-term capital gains tax rate. During the period examined in this study, the long-term capital gains tax rate was 20 percent, except from 1987 through May 6, 1997, when it was 28 percent. Therefore, we predict that, conditional on appreciation, the lock-in effect, if it mattered, was more pronounced during the period when long-term capital gains tax rates were 28 percent than when rates were 20 percent. Accordingly, we add a categorical variable (/2:B/7&*), indicating the 20 percent rate period, and interact it with %$6(. Thirty-two percent of the observations are drawn from periods when the 20 percent rate applied. A negative correlation would be consistent with reduced lock-in distortion when long-term capital gains tax rates are low. 6WRFN5HWXUQ5HVXOWV Table 2, Column A presents summary statistics from regressing cumulative abnormal returns on the tax measure, its components, and 8(. 14 As predicted, the coefficient on '5$7(* %$6( is positive (2.28) and statistically significant (W-statistic of 2.3), albeit modest, given the large sample size. However, given the imprecision in the tax measure (namely, the assumptions about investor tax rates and the appropriate period for computing price changes) and our inability to advance an omitted correlated variable to explain the finding, we infer that these results provide some initial evidence consistent with the holding period incentive affecting stock returns at quarterly earnings announcements. 14 Throughout the paper, if the null hypothesis of correct model specification under White s [1980] test is rejected at conventional levels, the reported standard errors are computed using White s [1980] consistent covariance matrix estimation to correct for an unspecified form of heteroskedasticity. However, the results are qualitatively unaltered if ordinary least squares standard errors are always used. In addition, the empirical results do not appear to suffer from cross-sectional dependence for two reasons. First, by examining returns from two-day windows, we avoid the cross-sectional dependence problems typically associated with long windows, such as one quarter or one year (Bernard [1987]). Second, cross-sectional dependence problems typically cluster in intra-industry analysis as opposed to inter-industry analysis (Bernard [1987]). The sample in this study includes 269 three-digit SICs; only six of which represent more than 2 percent of the sample. Furthermore, when we exclude firms that announce earnings on the same day as three or more firms in their three-digit SIC, leaving 64,376 observations, results are qualitatively unchanged. On the other hand, tests indicate that multicollinearity may be a slight econometric problem, inflating standard errors and biasing against rejecting the null hypothesis that taxes do not matter. 11

13 The regression coefficients enable estimates of economic significance. For example, the product of '5$7(* %$6( s regression coefficient estimate of 2.28 and its mean value of implies that capital gains tax effects account for abnormal returns of percent (7 percent annualized), or 15 percent of the observed two-day &$5, on average. A one standard deviation increase in the tax variable increases two-day, cumulative abnormal returns by 0.21 percentage points (26 percent annualized), a 55 percent increase in abnormal returns for the mean firm. 15 Examining the components of the tax measure, we find the coefficient on /2:B/7&** %$6( is negative, consistent with the lock-in effect being more important when long-term capital gains tax rates are higher. However, the coefficient on %$6( is not positive so we advance this interpretation cautiously. The coefficient on '5$7(, for which no prediction is offered, is negative; suggesting unspecified cross-temporal variation that correlates with changes in the spreads between short-term and long-term rates. As expected, the coefficient on 8(is positive and highly significant. To confirm that the tax variable is independent of other price-earnings effects, Table 2, Column B includes some explanatory variables motivated by extant studies. 16 We find that '5$7(* %$6( s coefficient (2.40) and its significance (W-statistic of 2.4) increase slightly. The additional control variables are detailed below: 121/,1($5 is designed to address concerns that extreme values of unexpected earnings are less value-relevant. It is computed as 8(* 8(, consistent with Lipe, Bryant and Widener [1998]. As expected, its coefficient is negative, implying extreme values are less valuerelevant. 35(',&7 is intended to control for Lipe s [1990] finding that firms with more predictable earnings patterns have higher earnings response coefficients. It is the mean of the absolute values of 8( across the investigation period and constant across all years for a firm. As expected, the coefficient on 8(*35(',&7is negative, consistent with less predictable earnings being less value-relevant; but it is not reliably less from zero percentage points are the product of '5$7(* %$6( s standard deviation of and its regression coefficient estimate of percent is the 0.22 percentage points divided by the mean &$5 of Throughout the paper we disregard behavioral effects concerning trading and security pricing that have been identified in recent finance studies (e.g., Barber and Odean [2000], Shleifer [2000], among many others), including findings, such as Odean [1998] that investors are reluctant to sell depreciated stock despite tax incentives to do so. The fact that we find a relation between equity trading and tax incentives provides some assurance that the deleterious effects of ignoring behavioral considerations are limited. 12

14 3(56,67 is motivated by findings that persistence explains cross-sectional differences in earnings response coefficients (e.g., Easton and Zmijewski [1989]. It is the autoregressive coefficient from Foster s [1977] time-series model and constant across all years for a firm. As expected, the coefficient on its interaction with 8( is positive, consistent with market recognition that earnings shocks manifest themselves in the future; but it is not significantly greater than zero. 07% is motivated by Collins and Kothari s [1989] report that share prices are increasing in anticipated earnings growth. It is market value divided by book value as of the last day of the fiscal quarter. As expected, the coefficient on its interaction with 8( is reliably greater than zero. %(7$ is designed to control for the effects of risk on the price-earnings relation. Consistent with Easton and Zmijewski [1989], it is the market-model beta estimated over the 100 trading days that end two days before the announcement date. Contrary to expectations, the coefficient on its interaction with 8( is positive. 6,=( is included to be consistent with prior research (e.g., Atiase [1985], Easton and Zmijewski [1989]). It is the natural logarithm of the market value of equity at the end of the quarter preceding the earnings release. No sign is predicted for its coefficient because size serves as a proxy for various constructs in the literature, but its coefficient and the coefficient on its interaction with 8( are significantly negative. As noted above, including these additional explanatory variables has nominal effect on the tax coefficient. Furthermore, when we repeat the regression excluding the tax measure and its components, the signs and significance of the additional variables hold. Thus, we infer that the non-tax control variables are independent of the tax variables. Hereafter, we only report results from tests using the more parsimonious model in column A. Robustness checks confirm that inferences are unaffected if the complete set of explanatory variables is used. 3ULFH5HYHUVLRQ If prices move because holding period incentives create a temporary equity constriction or expansion, then prices should revert to original levels at some point. One reason that reversal might not occur immediately is that it takes time for investors to disentangle price movements attributable to capital gains tax distortion from price movements for other reasons. We test for price reversal by reestimating the parsimonious model for one-day abnormal returns through 20 days after the quarterly earnings 13

15 announcement. The evidence is consistent with a price reversal beginning about a week after the earnings release. Table 3 presents summary statistics from daily regressions for the 84,742 observations for firms existing each day. The one-day tax coefficient is only significantly positive on trading day t (W-statistic of 1.7). However, when the dependent variable is the cumulative abnormal returns for the six trading days beginning with day t-1, the tax coefficient is 3.3 with a W-statistic of 3.0. For the subsequent 11 trading days (t+5 through t+15), the cumulative abnormal returns produce a tax coefficient of 3.9 with a W- statistic of 3.6. This period includes nine consecutive negative tax coefficients (trading days t+5 through t+13) with significant coefficients on trading days t+7 (W-statistic of 1.7) and t+13 (W-statistic of 1.8). We conclude that these results are consistent with a reversal during the two weeks following the week of the announcement. 7UDGLQJ9ROXPH5HVXOWV Table 2, Column C (D) presents estimated coefficients from the trading volume regression using the parsimonious (complete) model. 17 As expected and consistent with the stock return regression results, the coefficient on '5$7(* %$6(is negative and highly significant (W-statistics of 4.7 in column C and -3.7 in column D). This finding is consistent with holding period incentives causing individual holders of appreciated (depreciated) stock to restrict (expand) the supply of equity. The statistical strength of the volume results increases our confidence that the regressions are detecting the influence of holding period incentives. The regression coefficient estimate implies that a one standard deviation increase in '5$7(* %$6( increases two-day cumulative abnormal volume by 19 percent for the mean firm The price-volume relation is not as well understood as the price-earnings relation. Consequently, we generally rely on the price-earnings relation to guide our tests of trading volume. However, to test whether asymmetric volume responses to price increases versus price decreases (Karpoff [1987], Bamber and Cheon [1995]) affect the conclusions in this paper, a categorical variable indicating the sign of the two-day raw return surrounding the earnings announcement was added as an explanatory variable to trading volume tests. Inferences are unaltered. 14

16 6HQVLWLYLW\7HVWV We conduct several sensitivity tests to assess the robustness of the price and volume regressions. One, to provide an alternative specification, we segregate the sample into three periods based on the spread between short-term and long-term capital gains tax rates: (a) when the '5$7( is zero or three ( ), (b) when '5$7( is greater than 10 and less than 20 (1987, ), and (c) when '5$7( is 30 ( ) and limit the explanatory variables to %$6( and 8(. If capital gains tax incentives affect returns (volume), then the coefficients on %$6( should be greater (smaller) in regimes with larger '5$7(. Consistent with these predictions, the coefficient on %$6( in the returns regressions is largest when '5$7( is 30. For instance, when '5$7( is 30, the coefficient on %$6( is five-fold its measure when '5$7( is zero or three. Moreover, as expected, when the dependent variable is $9, the coefficient on %$6( is largest when '5$7( is zero or three. Two, to test whether the findings hold for both depreciated and appreciated shares, we reestimate the returns and volume models, segregating the sample on the sign of %$6(. The signs of the coefficients are always consistent with predictions, but the significance levels vary. When the dependent variable is &$5, the tax coefficient is marginally significant for both appreciated and depreciated subsamples. When the dependent variable is $9 and the sample is appreciated (depreciated) firms, the tax coefficient is highly significant (insignificant). Three, an individual s marginal tax rate for capital gains and losses is determined annually. Thus, tax planning could become more precise as individuals near year-end. However, we find no such evidence. Inferences are qualitatively unchanged when disclosures in December are deleted from the study and when disclosures in October, November, and December are deleted from the study. Finally, to ensure that our tax measure is not capturing some unspecified variation in corporate taxes, we repeat the regression including the maximum federal corporate income tax rate in the model. Results hold percent is the product of '5$7(* %$6( s regression coefficient estimate of 3.18 from column C and its 15

17 6 3,QGH[&KDQJHV 2YHUYLHZ The remainder of the paper reports results from tests of the holding period incentives on equity trading using the disclosure that a firm is joining or exiting the Standard and Poor s 500 Stock Index. Because the S&P 500 includes the largest U.S. corporations, finding that tax incentives affect equity trading around this disclosure should rule out possible concerns that the prior results are driven by liquidity constraints at small, thinly traded firms. Furthermore, finding that individual holding period incentives matter with index changes should address any concerns that some unspecified information at earnings announcements accounts for the prior findings. Harris and Gurel [1986] and Lynch and Mendenhall [1997] document that both share prices and trading volume soar when firms join the S&P 500 with positive abnormal returns of 3-4 percent typical on the first trading day after the disclosure. Their explanation for these price increases is upward price pressure created by a surge in demand from passive funds and other portfolio managers who track the 500 index. 19 When the rebalancing is completed, prices return to prior levels. This paper focuses on the shareholders who sell to the index funds. If enough are individuals affected by the holding period incentives, then personal tax considerations may play a role in the stock returns following an S&P disclosure. To test this possibility, we conduct return regressions, similar to those reported above. As expected, we find that share returns for firms joining the S&P 500 index are positively correlated with our tax measure ('5$7(* %$6(. Because index fund managers are contractually required to purchase shares at any price (or appear to behave accordingly), we expect trading volume to increase without regard to the effect that taxes might have on prices. Consistent with this expectation, when we do conduct trading volume tests for the S&P 500 sample, the coefficient on the tax variable is insignificant, consistent with fund managers standard deviation of divided by mean $9 of Other explanations include long-run downward sloping demand curves (e.g., Shleifer [1986]) and information effects arising from the implied certification of a firm s financial strength associated with inclusion, (e.g., Jain [1987], Dhillon and Johnson [1991]). See further discussion in Kaul, Mehrotra, and Morck [2000]. 16

18 rebalancing their portfolios around index changes without regard to share prices. Similarly, when a firm leaves the index, fund managers should sell their shares without regard to the tax implications. If so, the returns for firms exiting the S&P 500 index should not reflect any tax considerations. Consistent with this prediction, we find that share returns for firms leaving the index do not vary with the tax measure. 5HJUHVVLRQPRGHO The dependent variable is the one-day abnormal returns ($5) on day t+1 using a market model, where beta is estimated for the 100 days preceding day t and t is the day of the announcement. 20 Abnormal returns are computed on the day following the announcement because S&P announces changes to the index after the market closes. The tax measure ('5$7(* %$6() is computed identically with its components continuing as explanatory variables. 21 As in the earnings announcements tests, a positive coefficient is predicted on the tax measure. For appreciated firms, such a finding will be interpreted as evidence that mutual funds compensate individual shareholders based on the incremental taxes associated with selling appreciated stock before long-term qualification. For depreciated firms, a positive coefficient will be interpreted as evidence that mutual funds and individual shareholders share in the latter s tax savings from selling depreciated stock before long-term qualification. The only other regression variable in the primary tests is designed to control for changes in the demand by index funds over the investigation period. '(0$1' is the percentage of equity mutual fund assets held in index funds, as shown in Bogle s [1999] Exhibit V, 0RQH\ [April 1999, p.102], and Brennan [1999]. 22 Consistent with a dramatic increase in the number and holdings of S&P 500 index 20 Results are qualitatively unaltered if beta is estimated using the 100 days following day t+5 or if market-adjusted returns are employed. 21 For parsimony, /2:B/7&* and its interaction are excluded from this regression model because its inclusion has no effect on the tax variable of interest. 22 Results for the variable of interest are qualitatively unaltered if '(0$1' is dropped from the regression or alternative measures of demand from index funds are used, including Vanguard s number of index funds, Vanguard s percentage of assets in index funds, and natural logarithm of Vanguard s index fund assets, all as reported in Bogle [1999]. 17

19 funds during the investigation period, '(0$1'increases steadily from 0.2 percent in 1978 to 8.0 percent in The percentage decreased in only two years, 1983 and As index funds have become more active in the equity markets, the demand for shares when firms join the S&P 500 should have increased accordingly. If increased demand for shares creates upward price pressure, a positive coefficient is expected on '(0$1'. 6DPSOH We purchased from Standard & Poor s a list of the 518 changes to the S&P 500 from January 1, 1978 to December 31, From the Standard & Poor s list, we drop 70 additions attributable to restructurings of existing S&P 500 firms and 17 additions for which data are missing. The final sample includes 431 S&P 500 additions. Five firms are included twice in the sample. Annual additions range from six in 1992 to 35 in both 1998 and Before October 1989, S&P 500 announcement and addition dates were identical. Since then, the announcement has preceded the addition by seven days, on average. Table 4 shows that the $5 on the first trading day following the announcement for the 329 appreciated additions (those with positive %$6(, i.e., have increased in value during the holding period) range from 4.93 percent to percent with a mean (median) of 3.68 (3.29) percent. The 102 depreciated additions have similar price increases with a mean (median) $5 of 4.13 (3.85) percenton day t+1. Of the 518 deletions, 270 cannot be examined because they merge out of existence immediately following their removal from the index. Data are also missing for another 38 deletions (primarily because of leveraged buyouts), leaving 134 appreciated deletions and 76 depreciated deletions to investigate. Table 4 shows that the mean and median $5 on the first trading day following the announcement is 23 Besides capturing the intended increase in demand from S&P 500 index funds over time, the coefficient on '(0$1' may capture other unspecified intertemporal changes. The only intertemporal institutional change that we are aware is that before October 1989, announcements coincided with changes to the index. Now announcements precede changes. 18

20 negative for both groups of deletions, consistent with a temporary decline in demand when mutual funds unload their holdings. 3ULPDU\)LQGLQJV The first column of Table 5, Panel A provides summary statistics for the 329 appreciated additions when abnormal returns are computed for the first trading day following announcement of an inclusion to the S&P 500 (trading day t+1). 24 As predicted, the regression coefficient estimate on '5$7(* %$6( is significantly greater than zero (W-statistic of 2.1), consistent with the holding period incentives for individual investors affecting returns when S&P announces an addition to its 500 index. The tax coefficient enables rough estimates of economic significance. For example, the product of '5$7(* %$6( s regression coefficient estimate of 6.52 and its mean value of 0.11 (from Table 4) implies that the tax premium accounts for one-day abnormal returns of 0.72 percent (181 percent annualized), on average. This compares with a mean abnormal return of 3.68 (from Table 4) on the day following the S&P announcement. In other words, the regression coefficient estimate on '5$7(* %$6( implies that compensation for incremental capital gains taxes accounts for 20 percent of the abnormal return on the announcement day. The tax coefficient is not significantly different from zero for the next nine trading days. In fact, during the nine trading days, only one coefficient ( %$6( on trading day t+8) is significantly different from zero. Review of the other coefficients on day t+1 reveals that the '5$7( coefficient, for which no prediction is offered, is insignificant. The coefficient on %$6( is negative (W-statistic of 2.7), consistent with investors speculating on future additions to the index (1HZ<RUN7LPHV, May 21, 1986). 25 As 24 To mitigate the impact of outliers, we drop from each regression in the paper any observation with a studentized residual exceeding five. Conclusions, however, are unchanged if no outliers are deleted. 25 Consistent with this explanation, the coefficient on %$6( is less negative if the %$6( measurement period concludes one month preceding the S&P 500 announcement and insignificant if years before 1990, when the S&P added firms on the announcement day, are excluded from the study. 19