Capital Gain Tax Overhang and Price Pressure. Li Jin * Harvard Business School This draft: 6/2/04. Key words

Transcription

1 Capital Gain Tax Overhang and Price Pressure Li Jin * Harvard Business School ljin@hbs.edu This draft: 6/2/04 Key words capital gain tax, institutional investor, price pressure, limits to arbitrage, earnings announcement * I thank Malcolm Baker, Dan Bergstresser, Long Chen, Josh Coval, Mihir Desai, Ben Esty, Robin Greenwood, SP Kothari, Blake LeBaron, Josh Lerner, Jay Light, Bob Merton, Randall Morck, Andre Perold, Hank Reiling, Rick Ruback, David Scharfstein, Anna Scherbina, Erik Stafford, Laura Starks, Jeremy Stein, Peter Tufano, Luis Viceria and Bin Zhou, and participants at workshops at several universities for offering valuable suggestions. I acknowledge the excellent research assistance from Jenn Chu, Bryan Lincoln, Steven Oliveira, Michael Sorell, Deborah Strumsky and Zhijie Xue, and editorial assistance by Jenn Chu and John Simon. I thank I/B/E/S for providing analyst forecasts data, and SEC for providing data on the investment advisor profiles. I am grateful to the extensive effort by Harvard Business School Special Projects Group, led by Toni Wegner, in cleaning up the data used in this project. I also benefited from talks with David Granger and Melvyn Gonzalez at Thompson Financials to understand the CDA/Spectrum data better. Research support from Harvard Business School Division of Research is gratefully acknowledged. All remaining errors are my own.

2 Abstract Capital gains tax can impose potentially large cost on investors selling stocks. This cost can sometimes be an order of magnitude larger than conventional transaction costs. This paper addresses the question of whether capital gains taxes serve as an impediment to selling and if so, to what degree this delayed selling by investors correspondingly affects stock prices. Using a database of large U.S. institutions stock holdings with data on institutions client profiles, two main results are obtained. First, selling decisions by institutions serving tax-sensitive clients are shown to be sensitive to their cumulative capital gains, which is not the case for institutions with predominantly tax-exempt clients. In particular, both the likelihood and magnitude of selling by institutions that serve taxsensitive clients are negatively related to the cumulative capital gains. Second, tax-related underselling appears to significantly impact stock prices during negative earnings announcements, for stocks held by a large number of tax-sensitive investors. Specifically, following a negative quarterly earnings surprise, tax-sensitive investors sell less aggressively a stock that has large capital gains; thus for a stock held primarily by taxsensitive investors, the corresponding price reaction is less negative if it has accumulated large capital gains. Further analysis shows that the price reaction pattern is more severe when arbitrage is more costly.

3 When investors sell stocks at a profit they realize capital gains and pay tax that they could defer indefinitely if they did not sell. There is a large literature that shows that barriers to trading affect investors willingness to trade, and thus asset prices. Given that the capital gains taxes pose a cost of selling that can be an order of magnitude larger than conventional transaction costs like commissions and bid-ask spreads, we want to investigate whether capital gains taxes serve as an impediment to selling and if so, to what degree this delayed selling by investors correspondingly affects stock prices. The first half of the paper explores whether institutions serving tax-sensitive clients are sensitive to their cumulative capital gains when selling. Matching a database of large U.S. institutions stock holdings with data on institutions client profiles, I find that institutions that predominantly serve tax-sensitive clients exhibit different trading patterns than institutions with predominantly tax-exempt clients. When institutions in the aggregate are selling a stock, both the likelihood and magnitude of selling by institutions that serve taxsensitive clients are negatively related to cumulative capital gains. Large capital gains discourage, large capital losses encourage, selling, ceteris paribus. Consistent with the hypothesized impact of the capital gains tax, the behavior of institutions with predominantly tax-exempt clients does not seem to be related to cumulative capital gain. To test this I control for capital losses on other holdings that institutions could potentially use to offset capital gains. In addition, I account for the strong possibility that taxable investors might try to minimize their tax liabilities. The result-- that tax-sensitive institutional investors seem to react to capital gains taxes in a sensible way--is robust to alternative ways of constructing the capital gains measure. The second half of the paper looks for evidence of the impact of tax-related underselling on stock prices. I examine price reactions following negative quarterly earnings surprises. Ball and Brown (1968) and many subsequent researchers document that earnings surprises affect stock prices both during the announcement period and afterwards. For negative earnings surprises, stock prices decrease both during the announcement period and afterwards. The tax-sensitive selling hypothesis predicts that the tax-sensitive investors might be inclined to selling less aggressively a stock that has large capital gains than a 1

4 stock that has no capital gains. Thus, for a stock held by a large number of tax-sensitive investors (a tax-sensitive stock ), we might expect the price reaction to be less negative (more negative) if the stock has accumulated large capital gains (losses). Indeed, I find empirically that during the three-day announcement window such tax-sensitive stocks with larger capital gains exhibit less negative returns. Such a pattern does not exist for stocks held predominantly by tax-exempt investors. Further analysis suggests that the pattern is stronger when arbitrage force is weaker. In Section I, I review the related literature and outline the theoretical foundation for the empirical tests reported in this paper. Explanations of the data and of the procedure for constructing the measures used in the empirical tests are provided in Section II. Section II also contains a detailed discussion about the criteria to select the a priori tax-sensitive and non-tax-sensitive institution subsamples, to make sure that the definitions are unlikely to be subject to mis-classification 1. In Section III, I present empirical evidence that taxsensitive institutions react to capital gains tax in a sensible way, deferring the sale of stocks that have incurred large capital gains, and that a similar pattern is not detected for non-taxsensitive institutions. Evidence of price pressure following negative quarterly earnings announcements for stocks held by large numbers of tax-sensitive investors is presented in Section IV. Investors might be subject to behavioral bias, such as the disposition effect documented by Barber and Odean (2000). In addition, creative financial engineering might allow investors to circumvent some of the capital gains. Section V discusses these considerations in detail, and shows that they are unlikely to significantly affect the tax sensitive trading pattern documented here. Section VI concludes. I. Theoretical foundation and related literatures There are two major reasons for investors to want to delay the realization of capital gains tax 2 : the option to reset capital gains and losses at death, which forgive the tax on any 1 Essentially, to avoid mis-classification, I selected only institutions that I have a high confidence that are either tax-sensitive or non-tax-sensitive. 2 Some other motivations are not elaborated here, among which is the charitable contribution option which might apply to some investors: donating eliminates the capital gains tax liability altogether, but the full 2

5 embedded capital gains or losses at the time of death and requires the tax bases of the inherited assets to be reset to prevailing market prices, and the option to combine capital gains with capital losses. Appendix A contains a simple numerical example illustrating the effect of delaying the capital gains realization on the after-tax proceeds from selling. Given these considerations, a tax-sensitive investor should rationally time the realization of capital gains tax, and such tax timing could result in a substantial lifetime reduction on the capital gains tax. For that reason, Feldstein, Slemrod and Yitzhaki (1980) and Poterba (1987a, 1987b) have argued that the true capital gains tax rate is substantially low, and only in the order of 25% of the statutory capital gains tax rate. Theoretical modeling of investor behavior under capital gains tax can be traced to Constantinides (1983, 1984) and Dammon, Spatt, and Zhang (2001a, 2001b). Dammon, Spatt and Zhang (2001a), for example, characterized optimal portfolio decisions in the presence of capital gains taxes and short-sale restrictions, establishing that the incentive to re-diversify the portfolio is inversely related both to the size of the embedded capital gains and to investor age. What drives these results is the option to reset tax bases at death. 3 Most theoretical models presume stocks to be fairly priced but investors nonetheless want to sell for reasons of diversification or liquidity. In practice, the reason for selling could be to rebalance a portfolio or to meet a liquidity shock, but it could also be that investors disagree on the true value of the stock and some (rightly or wrongly) believe the stock to be overvalued. In the latter case the capital gains tax overhang occasions a potential mispricing as it constitutes a barrier for some pessimistic investors to sell. This might hinder the market s ability to aggregate information (Hong and Stein 2003). At the least, withholding selling might serve to unbalance supply and demand and, absent a perfectly elastic demand for the stock, create price pressure. market value of stock is counted towards the tax deduction. Thus if a donation is expected, the donor might withhold the selling of the stocks with large capital gains. In addition, it is often maintained that some investors will do just about anything to avoid paying tax, even if doing so is not economically beneficial. 3 To maintain tractability, Dammon, Spatt, and Zhang (2001a) focused on the case of a single risky asset and thus ignore another major source of the value of tax timing: by postponing the realization of capital gain, investors preserve the option to combine it with capital loss trading on other risky assets thus greatly enhancing the value of the option to delay realizing capital gain. Dammon, Spatt, and Zhang (2001b) address that issue. 3

6 Prior research has found that price pressures resulting from imbalances in demand and supply can be quite large, despite the presence of arbitragers in the market. Shleifer (1986), Harris and Gurel (1986), and Lynch and Mendenhall (1997) document that both share prices and trading volumes soar when firms join the S&P 500, with positive abnormal returns of 3% to 4% typical on the first trading day after the disclosure. They explain the price increases as upward price pressure created by a surge in demand from indexers. The inability of arbitragers to totally wipe out the price pressure is consistent with the substantial body of empirical evidence on the limits to arbitrage, for example, Mitchell, Pulvino and Stafford (2002). Index funds in the mid-1980s held as much as 3% of the shares outstanding for a stock, as reported in Shleifer (1986). In the context of the taxbased selling considered here, it would not be too surprising to observe price pressure when as much as 50% of a stock s float is affected by tax-sensitive investors trading patterns, especially in the case of stocks with severe capital gains or losses. Indeed, the event study outlined in Section IV, examining negative quarterly earnings announcements for a number of stocks held predominantly by tax-sensitive investors, supports this hypothesis. This paper adds to the literature on the link between taxation and investor behavior 4. The existing works that are particularly relevant to this paper are Blouin, Raedy, and Shackelford (2003) and Ivkovich, Poterba and Weisbenner (2004). Motivated by the theoretical work in Shackelford and Verrecchia (2002), Blouin, Raedy, and Shackelford (2003) show that, for appreciated stocks, share returns rise and trading volume falls with a measure for incremental tax (long term minus short term capital gains rate) around both earnings announcements and additions to the S&P 500 index. They argue that these findings are consistent with the hypothesis that investors defer sales of appreciated stocks until they qualify for long-term capital gains tax treatment, and withdrawal of appreciated 4 There is an extensive and active literature documenting how investors behave under taxes, and the potential implications on asset prices. Representative works include Ayers, Lefanowicz and Robinson (2003), Callaghan and Barry (2003), Gibson, Safieddine and Titman (2000), Graham, Michaely and Roberts (2003), Green and Hollifield (2003), Grinblatt and Keloharju (2004), Grinblatt and Moskowitz (2004), Poterba (2001), Poterba and Weisbenner (2001), Reese (1998) and Sias and Starks (1997), among others. Graham (2003) and Scholes and Wolfson (1992) summarize the impact of taxes on a large body of economic activities. 4

7 securities constrains the supply of equity, temporarily boosting equity prices. Similarly, Ivkovich, Poterba and Weisbenner (2004) compare individual investors' realizations in their taxable and tax-deferred accounts, and find a strong lock-in effect for capital gains in taxable accounts relative to tax-deferred accounts. They also find that tax-loss selling is most pronounced in December, particularly if there is opportunity to combine it with capital gains during the year. The findings in this paper are consistent with these earlier results. However, this paper yields several significant insights, and more specifically addresses the relationship between capital gains tax related incentives and returns. First, this paper establishes a direct link between investor tax-related trading behavior and asset prices: while Blouin, Raedy, and Shackelford (2003) conjecture that investor response to capital gains tax changes drive returns and volumes during the events, and Ivkovich, Poterba and Weisbenner (2004) study the trading pattern of taxable individual investors, this is the first paper to simultaneously address both, and providing concrete evidence that the trading patterns of taxable investors indeed has asset pricing implications. Second, extending the results of Blouin, Raedy, and Shackelford (2003), my results suggest that the capital gains tax can demonstrate itself beyond the consideration of the short-term and long-term capital gains tax rate difference: as explained later, in 5 out of 11 years in this sample the differences between short-term and long-term capital gains tax rates are either zero or 3%. Third, my results show that not only individual investors, but even financial institutions, can exhibit tax-sensitive trading patterns. Given that institutions are often argued to be the marginal investor to set prices, their behavior will likely have larger price impacts. II. Data, measure and empirical methodologies Data Sources The data used in this study come from the Spectrum 13F institutional investor holding database. Institutional investment managers who exercise investment discretion over $100 million or more in Section 13F securities must report their holdings of more than 10,000 5

8 shares or greater than $200,000 in value to the SEC on Form 13F. Section 13F securities generally include equity securities that trade on an exchange or are quoted on the Nasdaq National Market, some equity options and warrants, shares of closed-end investment companies, and some convertible debt securities. Shares of open-end investment companies (i.e., mutual funds) are not included as Section 13F securities. Form 13Fs are publicly available on the SEC s EDGAR database. The data is available from 1980 to 2002, but a number of institutions are improperly classified in 1998 and beyond. Spectrum reports five types of institutional investors classification: bank trust departments, the clients of which can be either tax-sensitive or tax-exempt; insurance companies, which are tax-sensitive overall, but could also have tax-exempt accounts for its clients; investment companies (open-end or close-end mutual funds), the clients of which can be either tax-sensitive or tax-exempt; independent investment advisors, whose clients can be either tax-sensitive or non-tax-sensitive; others, such as foundations, ESOPs, university endowments, and internally managed private and public pension funds, many of which are tax-exempt. A detailed explanation of the classification of institutional investors in 13F data is provided in Appendix B. Institutional holdings data are supplemented with Investment Adviser Public Disclosure (IAPD) data obtained from the SEC. IAPD data contain investment advisors self-reported clientele bases, broken down into 10 categories: individuals (other than high net worth individuals); high net worth individuals; banking or thrift institutions; investment companies (including mutual funds); pension and profit sharing plans (other than plan participants); other pooled investment vehicles (mostly hedge funds); charitable organizations; corporations or other businesses not listed above; state or municipal government entities; and other, such as non-us government entities. Investment advisors are required to report the percentage of business represented by each clientele category. I limit the data to a period that extends from 1987 through In addition to data problems in 1998 and beyond, there are other reasons for focusing on these 11 years of data. One, because this time period lies between two major tax law changes (in 1986 and 6

9 1997) the tax regime is relatively constant. Short-term capital gains statutory tax rates fluctuated but the long-term capital gains tax rate remained unchanged at 28%. 5 Two, a capital gains measure constructed during this period is likely to be more robust to assumptions about initial holding positions. To the extent that institutional holding information prior to 1981 is lacking, various assumptions can be made with respect to these initial stock holdings in constructing the subsequent capital gains measure. Appendix C describes the procedure in detail. The capital gains measures for the earlier years (1981 to 1986) are more sensitive, the capital gains measures during 1987 and 1997 more robust, to these assumptions. Three, during these 11 years capital gains were considerable and thus relevant: the NYSE/AMEX/NASDAQ value weighted market index rose by 400% during this period. Identifying tax-sensitive versus tax-exempt institutional investors The availability of data is one reason for studying institutional rather than individual investors, but there are several other reasons for focusing on the former. First, retail investors might be more subject to the disposition effect documented by researchers such as Barber and Odean (2000), who showed that individual investors of a brokerage firm hold onto losers for too long and sell winners too soon, even if doing so violates efficient tax planning. Professional money managers, being more sophisticated and likely more disciplined, might be less subject to behavioral biases. The capital gains tax is, moreover, a relatively more important component of transaction costs for institutions given that many are able to negotiate better bid-ask spreads and broker commissions. Furthermore, given that institutions hold large amount of stocks, if they act in a coordinated way, the impact on stock price and liquidity will likely be much larger. However, institutions can serve both tax-sensitive and non-tax-sensitive clients, and even institutions that serve primarily tax-sensitive clients may not exhibit tax-sensitive selling behavior if there is an agency problem and managers are not rewarded for being tax-savvy. 5 The short-term capital gain statutory tax rates for individuals was 38.5% in 1987, decreased to 28% from 1988 to 1990, and climbed to 31% during 1991 and 1992, before settling at 39.6% thereafter. 7

10 For an institution to be tax-sensitive, we need (1) the existence of tax-sensitive investors capable of exerting pressure on managers, or (2) managers who care about the tax consequences of their portfolio under management. 6 Among institutional investors that, a priori, might be expected to react to taxes, are investment managers for high net worth individuals and hedge funds. The interests of the former are served by their relatively small number, which facilitates communication and collaboration with and, ultimately, monitoring and collective discipline of fund managers. Managers of hedge funds, for incentive reasons, are typically required to invest a significant portion of their personal wealth in the funds they manage, and thus likely care about their personal tax consequences. Informal conversations with industry experts suggest that managers personal tax considerations often play a significant role in the trading activities of such institutions. Finally, judging from conversations with industry executives, insurance companies (of the type identified in the Spectrum 13F database) also tend to show taxsavvyness. Of the five types of institutions identified by Spectrum, banks and mutual funds are excluded from this study as no concrete information about taxable investor base can be obtained for them 7. Insurance companies are universally treated as tax-sensitive 8. The profiling information from IAPD is used to distinguish independent investment advisors 6 There is still some debate as to whether mutual funds are tax-sensitive. Informal conversations with some mutual fund managers reveal that many do not regard tax as a first order consideration for mutual funds. Two possible explanations are offered. One, mutual fund investors might be largely tax-exempt and so shouldn t care about tax. Perhaps as a consequence, most mutual fund managers compensation contracts don t explicitly reward them for after-tax performance. Two, mutual fund managers are not required to put a significant portion of their personal wealth into the funds they manage, and many do not. They thus have no personal reason to care about tax consequences of fund activities. On the other hand, recent empirical research suggests that even for mutual funds, tax seems to play an important role. See, for example, Bergstresser and Poterba (2002) and Huddart and Narayanan (2002). It is possible that some, but not all, mutual funds will exhibit tax-sensitive trading behavior. This paper s focus on non-mutual fund institutional investors thus complements the results of the foregoing research. 7 Another consideration is that banks (mutual fund families) typically have many separate accounts (funds) under one roof. To the extent that these accounts/funds don t have the same level of tax-savvyness, the documentation of the existence of tax-sensitive trading using aggregate data would be more difficult. The multiple account problems might still exist in other types of institutions, but would be less severe. 8 There is one caveat: while insurance companies invest large amount of money in their own accounts and pay tax on these, they also invest on behalf of their clients through potentially tax-advantaged accounts. The investments through own account and through client accounts are not distinguishable in the data. To the extent tax-exempt client investment might be substantial, I re-do the tests in this paper excluding the insurance companies. The results are not affected qualitatively. 8

11 who serve primarily tax-sensitive clients from those who serve primarily non-tax-sensitive clients. I classify as tax-sensitive those independent investment advisors with a clientele consisting of primarily (>=50%) high net worth individuals and hedge funds. I also include investment advisors that do not file the IAPD forms, but are included in the hedge funds section of the Money Manager Directory. As reported by Brunnermeiser and Nagel (2003), only institutions that conduct non-hedge fund business such as advising mutual funds and pension plans, are requested to file for the IAPD disclosure. That they are classified as hedge funds in the Money Manager Directory and not listed on the IAPD form indicates that the majority of their business is in hedge funds. Investment advisors with more than 50% of their clientele composed of pension funds, state and local governments, and charitable organizations are classified as non-tax-sensitive institutions. This classification of tax-sensitive and non-tax-sensitive institutions is fairly conservative: there are many investment advisors who don t fall into either the tax-sensitive or the non-taxsensitive category, since only five out of ten types of clients are clearly identified as either taxable or non-taxable. For institutions categorized in 13F data as other I conduct a word search for pension, endowment, foundation, and variations thereof in the names, and identify as non-tax-sensitive any for which these words come up. Combining capital gains with capital losses Institutions seldom hold just one stock. At any time some of their holdings might have capital gains, others capital losses. A tax-sensitive institution might want to combine the selling of the two types of securities to exploit the tax shield afforded by the losses. Because current tax code permits capital losses to be carried over for only a short period of time, institutions are motivated to offset with capital gains selling their capital losses as soon as losses are incurred. 9 This has two implications. First, capital gains might be realized promptly to offset an existing or anticipated capital loss. Second, in the absence of an existing or anticipated capital loss, institutions might postpone selling a stock with 9 If an institution s losses exceed its gains, no deduction is permitted currently. Instead the net realized loss is carried back and deducted against net capital gain for the previous three years. Any remaining loss offsets net capital gain in the subsequent five years, after which it expires unutilized. See Rendleman and Shackelford (2003). 9

12 capital gains until it can find an opportunity to combine its sale with that of a stock that has incurred capital losses. Given tax-sensitive institutions inclination to combine capital losses with capital gains, the measure of capital gains used in the tests that follow is the accrued capital gains, the amount that cannot be offset by the current level of accrued capital losses from other securities in the institution s portfolio. More precisely, I measure uncovered capital gains and covered capital losses. If a tax-sensitive investor combines capital gains with capital losses, only the uncovered portion of the gains (that which cannot be offset by losses) might constitute an impediment to selling. An institution, on the other hand, will want to quickly realize the capital losses given sufficient opportunities to use the tax-credit to offset taxes on other realized capital gains. It thus has an incentive to quickly realize capital losses that are covered (or usable right away). A capital loss will be uncovered (i.e., not usable) only if (1) capital gains to date are insufficient to offset it, and (2) the institution perceives the likelihood of utilizing the capital loss tax credit in the next few years to be slim. In this case the institution will hesitate to realize such (uncovered) capital losses. This was a rare occurrence during the period : with rapidly rising stock prices, institutions almost always had sufficient capital gains to offset their capital losses. For brevity, I report only results using uncovered capital gains and unadjusted capital losses (i.e., I assume all capital losses to be covered ). Alternative measures give qualitatively similar results 10. Constructing measures of capital gains overhang The capital gains measure is defined as: 10 There is evidence that capital loss tax credit might not be very important in practice: as reported in Blouin, Raedy and Shackelford (2003), the Internal Revenue Service reports that in 1997 individuals in the maximum tax bracket (39.6%), who accounted for 61 percent of all net capital gain, reported $169 billion of long-term capital gain and only $5 billion of long-term capital losses and $16 billion of short-term capital gain and only $8 billion of short-term capital losses. As a robustness check, I define the capital gain without adjusting for the aggregate capital losses from other securities holdings. The empirical results are qualitatively the same, although some of the coefficients are not always as significant. All coefficients are nevertheless statistically significant at the 5% level. 10

13 Capital gains = ln(current stock price/value-weighted average acquiring price). In the above calculation, I acknowledge that tax-savvy institutions can optimize their selling to reduce taxes. Under current U.S. tax law institutions can designate the tranch of stocks to be sold. Thus, unless doing so would trigger a different tax rate (such as shortterm capital gains versus long-term capital gains), an institution could sell first the shares acquired at the highest price, so as to maximize its capital loss or minimize its capital gain. Thus, the first measure of capital gains I construct incorporates this selling strategy, which assumes an institution sells first the stocks acquired at the highest price, sells second those acquired at the next highest price, and so on. Alternatively, the second measure of capital gains I construct assumes the institution sells existing shares proportionally, that is, all shares regardless of purchase date and tax basis are equally likely to be sold. The detailed construction of capital gain/loss measures is in Appendix C. For each institution that I identify as either tax-sensitive or non-tax-sensitive I collect data on all stock holdings reported in the 13F database and construct capital gains measures accordingly. The observations are unique quarter-institution-stock combinations for the period 1987 to The following empirical tests are further constrained to only stocks for which there are at least ten 13F institutions holding the stock during that quarter, to guard against any spurious statistical effect driven by a very few institutions with abnormal selling patterns. Summary statistics for the sample included in this study are provided in Table I. There are, for example, altogether 49,078 quarterly observations for the tax-sensitive institutions. The mean capital gains measure is 3.00%, the median 1.52%. Also evident in Table I is considerable cross-sectional variation in the measure of capital gains for both types of institutions. [Insert Table I around here] 11

14 Methodology for regressions To be consistent with a large body of literature in finance, whenever possible, the empirical tests that follow use the Fama-MacBeth methodology to control for potential crosscorrelation in data. Specifically, for each quarter, I run regressions across all institutions and all stocks, while controlling for fixed effects at the institution and stock levels. I then make inferences using the time series mean and standard deviation of these quarter-byquarter measures, as in Fama and MacBeth (1973) and Fama and French (2002). To guard against the potential problem of serial correlation in the coefficients in the Fama-MacBeth approach, I further adjust the t-statistics using an approach similar to the one outlined in Brav, Lehavy and Michaely (2003), Appendix A. As a robustness check I re-run the tests using bootstrapping methods. The results corroborate one another. Interested readers can obtain the bootstrapping results from the author. Testing unconditional differences in holding patterns An important and necessary step before examining how the capital gains tax impacts trading is to address the issue of whether, and to what extent, these two groups of investors differ unconditionally in their holding and trading patterns. If there are systematic differences, then it is crucial not to attribute these differences to tax-related trading. I categorize the holding and trading characteristics of the two groups of investors in Table II. I report the average number of stocks held, and the characteristics of these stocks (value versus growth, small versus large). I also report the quarterly portfolio turnover by these two groups of investors, where quarterly portfolio turnover is defined as the ratio of quarterly trading (sum of the absolute value of the dollar amounts) over the overall holding in a quarter. [Insert Table II around here] 12

15 Unconditionally, the tax-exempt institutions tend to hold more stocks, likely reflecting the fact that the average tax-exempt institution is larger than the average taxable institution. However, the two groups of institutional investors have similar preferences over value versus growth and over size. In particular, both groups prefer stocks that have growth characteristics (BE/ME<30 th percentile). The difference, although statistically significant, is economically less meaningful. Tax-exempt institutional investors such as pension funds are more likely to stay with larger stocks. The difference, at 9%, is statistically as well as economically significant, likely due to the fact that some tax-exempt institutions like pension funds face more stringent regulations, such as the prudent-man rule, that favor investments in larger stocks. However, there is no qualitative difference: both groups strongly prefer large stocks to small stocks (81% versus 90% holdings). Interestingly, the unconditional quarterly portfolio turnover characteristics for the two groups of institutions are slightly different, with tax-sensitive institutions turning over their portfolios slightly more rapidly than non-taxable institutions. While the result is significant only at 5% level, due to the fact that both groups exhibit large variations in the turnover measures, the more rapid turnover on the part of taxable institutions implies that any causal relationship found, in which higher-capital gains reduce trading, cannot be attributed to inherent differences between taxable and non-taxable institutions. Furthermore, in Tables III to VI, I document that, conditional on a large capital gains tax overhang, the taxsensitive institutions trade less frequently. Since the unconditional results go against the conditional results, it is hard to attribute the tax-related trading pattern documented below to any hard-wired unconditional pattern. III. How do tax-sensitive and non-tax-sensitive investors react differently to a selling event? The first main objective of this paper is to assess whether tax-sensitive and non-taxsensitive investors react differently to a selling event. Focusing on quarters in which institutions overall are selling a given stock, I evaluate first whether the likelihood that a 13

16 particular institution will sell is related to the cumulative capital gains accrued by the selling institution. Second, if the institution in question does sell in a particular quarter, I examine whether the magnitude of selling is related to the institution s capital gains. I analyze the likelihood of selling as a function of cumulative capital gains by running logit regressions, one for tax-sensitive and one for non-tax- sensitive institutions. These regressions estimate the likelihood of selling a stock conditional on overall institutional selling of the stock. The dependent variable is a dummy variable that takes the value of 1 if the institution sells the stock during the quarter in question, and zero if otherwise. The independent variable is the capital gain/loss that the institution has accumulated on that stock, at the beginning of the quarter. The control variables are: 1) past performance of the institution, measured by the value-weighted return on the institution s portfolio during the last period, 2) the log of the total dollar holdings of equity of the institution at the beginning of the period, 3) the percentage of net money inflow of the institution during the quarter, measured by the total dollar value of the position changes, divided by the total dollar holdings of the institution, and 4) the return on the stock during the previous quarter. The regression takes the form: (1) Probability of selling = a + b* capital gain+c*controls; The unit of observation is each unique {institution, stock, quarter} combination. All stocks for which there is overall institutional selling during a quarter are included in the regression for that quarter. In addition, I require there to be at least 10 institutions holding the stock at the time of selling. The results are reported in Table III. Two types of logit regressions are run: one using the Newey-West adjustment for serial correlation up to 8 quarters, and the other controlling for random effects at both the institution level and the stock level. The overall results are consistent with each other. As can be seen in Table III, for a tax-sensitive institution, conditional on institutions overall selling a stock, the likelihood of selling by a specific institution is negatively related to that institution s cumulative capital gains on the stock. 14

17 Higher capital gains diminish, higher capital losses increase, the likelihood that an institution will sell. For example, in the first column of Table III, the odds ratio for the capital gains measure is Other factors that are significant are the portfolio return from last period (with odds ratio of 1.35) and portfolio inflow during the quarter (with odds ratio of 0.07). These results suggest that, if the other factors are at the median levels reported in Table I, ( for the portfolio return from last period and for the portfolio inflow during the quarter), the probability of selling at a zero capital gains would be p = = 49.9%, whereas a unit increase in the capital gains measure would reduce the selling probability to p = = 38.6%, a 11.3% reduction. On the other hand, no statistically reliable result holds for the control sample of a priori non-tax-sensitive institutions: neither higher nor lower capital gains seem to influence the likelihood that these institutions will sell 11. [Insert Table III around here] To determine whether tax-sensitive and non-tax-sensitive institutions behave differently when facing a selling event, I run a logistic regression with an interaction term on the taxsensitivity of the institution (which takes a value of 1 if the institution is tax-sensitive, 0 otherwise) and cumulative capital gains. (2) Probability of selling = a + b*[ capital gains*dummy tax-sensitive ] + c* capital gains + d* control variables The results are reported in Table IV. Two types of regressions are run as before, the model with the Newey-West adjustment, and the model with random effects. Two regressions are 11 We show only the results for the combined sample of a priori tax-sensitive and non-tax-sensitive institutions. The results also hold for individual types. But there being fewer observations within each of these types, the separate results for the various types of tax-sensitive and non-tax-sensitive institutions are less interesting. 15

18 run and reported for each of the two types. The first does not control separately for the capital gains, the second does. [Insert Table IV around here] In both regressions the interaction term, capital gains*dummy tax-sensitive, is statistically significant at the 1% level. Higher capital gains, interacted with the tax-sensitive investor dummy, will reduce the likelihood of selling. Controlling for capital gains separately reveals the changes in selling likelihood to indeed come from the interaction term, suggesting that capital gains have different impacts on the likelihood of selling of the taxsensitive and non-tax-sensitive investors. Next, conditional on an institution selling a stock in a particular quarter, I study whether the magnitude of selling is related to the amount of capital gains (losses). Again, I focus only on quarters in which institutions overall are selling; however, now in addition I impose the requirement that the institution in question also sells. To ascertain whether the percentage of holdings sold is related to the cumulative capital gains I run the following regression: (3) Percentage sold = a + b* capital gains+c*controls Specifically, for each institution and each stock I regress the current quarter s percentage selling on the beginning-of-quarter cumulative capital gains, measured as the log of the price appreciation ratio. The control variables are the same as in Table II. The results are reported in Table V. [Insert Table V around here] The results confirm that for a priori tax-sensitive institutions the capital gains do indeed affect the amount being sold. This result is statistically as well as economically significant. For tax-sensitive institutions, a capital gains measure of 1 (which means that the current 16

19 price is times the average acquiring price) decreases the amount sold by about 10%. Such a pattern is largely non-existent in the a priori non-tax-sensitive sample. Of course, the relationship between capital gains and percentage selling can only be approximate. The capital gains being the average percentage gains, if the institution can dedicate the highest priced tranch of stock to be sold first the realized capital gains on that tranch could be less than the average capital gains calculated here. Thus, as the institution sells more, the average capital gains on shares sold will increase. This might be one reason that tax-sensitive institutions often sell some, but not all, of their stocks with capital gains: the earlier sales, to the extent that it incurs a lower capital gains tax, can be executed at a lower cost to the seller. To test whether the magnitude of selling differs for tax-sensitive and non-tax-sensitive institutions, I regress current period percentage selling on the beginning-of-period cumulative capital gains and the interaction term between capital gains and tax status, while controlling for other factors that might affect selling. Specifically, I run the regression: (4) percentage selling = a + b*[ capital gains*dummy tax-sensitive ] + c* capital gains + d*controls The results are reported in Table VI. Two regressions are run and reported. As in the earlier tests, the first does not control separately for the capital gain, the second does. [Insert Table VI around here] The interaction term between capital gains and tax-sensitivity is negative and statistically significant. It is also economically significant: a capital gains measure of 1 will reduce the magnitude of selling by about 10%, but only for the tax-sensitive investors. 17

20 The above regression analysis suggests that trading by tax-sensitive and non-tax-sensitive investors differs significantly on the tax front. In particular, for stocks with large capital gains, tax-sensitive investors will be less likely to sell. Moreover, if they do sell, they will sell less. Such a pattern is absent among non-tax-sensitive investors. IV. Evidence of price pressure: event study using a negative earnings surprise In this section, I attempt to determine whether the documented differences in the selling patterns of tax-sensitive and non-tax-sensitive investors might affect stock prices. In particular, I focus on the case of negative quarterly earnings surprises, which cause negative price reactions during the announcement window, and I assess whether the price pressure from a capital gains overhang is manifested in certain tax-sensitive investors decreased willingness to sell. Investors opinions about a stock and its subsequent trading are likely to differ during negative quarterly earnings announcements. Investors who believe a stock to be overvalued will sell. According to the capital gains tax-related selling hypothesis, during such an event, some tax-sensitive investors who would otherwise want to sell might decide to wait until the capital gains on their stock can be better managed. For instance, as earlier mentioned, they may be able to offset their gains with capital losses incurred by other stocks. If their capital gains taxes are sufficiently high, these investors might put off selling indefinitely. This is likely to dampen the speed with which bad news would be incorporated into stock prices during the event. Therefore, to ascertain the possible impact of tax-sensitive investors trading behavior on stock prices, I test for the existence of price pressure for selected stocks that are held by a large amount of tax-sensitive investors. To do that, I select stocks that have the highest concentration of tax-sensitive investors, and see whether the existence of tax-sensitive trading affects prices during announcements. As the goal here is to identify the existence of capital gains-related price pressures, rather than to ascertain the prevalence of such price 18

21 pressure in all stocks, I focus on a sub-sample of the stocks where I have high confidence that a large number of tax-sensitive investors exist. Defining tax-sensitive versus non-tax-sensitive stocks I define a stock as tax-sensitive if, at the beginning of a quarter, a large proportion of its institutional investors can be identified as tax-sensitive. For a particular stock, I use the proportion of 13F holdings held by tax-sensitive institutions to proxy for the proportion of holdings construed to be tax-sensitive. Specifically, I divide the number of shares held by tax-sensitive 13F institutions by the total number of shares held by 13F institutions overall. For each quarter, I rank sort this measure for all stocks reported in 13F and select the stocks with the highest 10% of this proportion measure to be my sample of tax-sensitive stocks. I correspondingly define the sample of non-tax-sensitive stocks symmetrically, picking the 10% of stocks with the highest measure of the proportion of non-tax-sensitive institutional holdings. It should be noted that the ratio of tax-sensitive institutional holdings to total 13F holdings is likely to be a conservative estimate of a stock s true taxable investor base for two reasons. One, the ratio does not take into account other investor types (individuals and insiders) who are largely taxable. Two, because I am able to identify only a portion of the 13F institutional investors as definitely tax-sensitive or definitely non-taxsensitive, the ratio of tax-sensitive to overall institutions could likely be much higher in reality: other potentially tax-sensitive institutions such as mutual funds and banks, which cannot be categorized as either tax-sensitive or non-tax-sensitive in this study, are not included in the numerator, but because they are a part of the overall 13F institutional holdings, they nonetheless count towards the denominator. Because the identification of the two sets of stocks tax-sensitive versus non-tax-sensitive stocks is critical to the analysis, I perform a series of robustness checks to verify that the results are not biased by the criteria for identifying the two groups. First, I repeat the analysis using a threshold of 1% and 5% in the proportion measures in the identification of 19

22 the two samples. 12 Second, I construct an alternative tax-sensitive (non-tax-sensitive) stock sample by requiring that at least 50% (33%, 25%) of the 13F institutional holdings be shown to belong to tax-sensitive institutions. Third, to use proportion of tax-sensitive investors among institutions to proxy for proportion of tax-sensitive investors among all investors might be less than justifiable if institutions in aggregate hold only a tiny fraction of the overall shares outstanding. To address this concern I also try to limit the analysis to stocks for which institutional holdings constitute a significant portion of overall holdings. The limit thresholds are such that at least 10% (20%, 30%) of the total shares outstanding are held by institutional investors. Finally, I check the results where the tax-sensitive stocks are defined as stocks for which a large percentage of the overall shares outstanding (as opposed to institutional holdings) is held by either tax-sensitive institutions defined above, or individual investors. 13 The results of all four robustness tests, not reported for brevity, leave the results qualitatively unchanged. Measuring capital gains Without a way to secure holdings data for all tax-sensitive investors, institutional or individual, the level of capital gains incurred by a representative investor must be approximated. I proxy the capital gains incurred by an average investor using the (splitadjusted) price appreciation during the previous five years, up to the last trading day. Specifically, the natural log of the (split-adjusted) price ratio, ln(p -5 / P 0 ), is used to proxy for capital gains and losses. This measure of capital gains is acknowledged to be 12 A related idea is to develop a continuous measure of percentage tax-sensitive holdings, and ascertain its property in a regression. However, this is problematic. As mentioned before, because we could not identify all tax-sensitive institutions, we lack a precise measure of the tax-sensitive investor base of a stock. Therefore, while we can identify with relative confidence a small sample of stocks as held by extremely large amount of tax-sensitive investors, we are less confident about the exact magnitude of the overall taxsensitivity of a continuum of stocks. For example, while we are confident that a stock with 60% tax-sensitive investors and a stock with 50% of tax-sensitive investors are both significantly held by tax-sensitive investors, we cannot say for sure that the first stock is held by more tax-sensitive investors, because some of the tax-sensitive investors for these stocks might not be identified as such by our method. Therefore, while it makes sense to study whether tax-related selling affect the price of a stock with 50% tax-sensitive investors (and compare the pattern with that of a stock with no tax-sensitive investor holdings), it would be less meaningful to argue that the tax-related price pressure will be more significant for a stock with 2% taxsensitive holdings than for a stock with 1% tax-sensitive holdings: the measure will be too noisy to achieve that goal. 13 I measure the individual investor base as those investors that are neither institutions nor insiders. 20