Closed-End Fund Discounts and Taxes*

Transcription

1 Closed-End Fund Discounts and Taxes* Shishir Paudel, Sabatino (Dino) Silveri, and Mark Wu July 2017 Abstract We revisit the effect of personal taxes on closed-end fund discounts. After documenting CRSP errors in classifying the tax status of distributions, we make use of two important institutional features that the related literature has ignored: (i) tax rate changes, and (ii) non-taxable distributions. We find that taxable distributions lead to discount reductions, consistent with a taxbased explanation of the discount. Moreover, the discount reduction following a taxable distribution is greater during high tax periods than during low tax periods. In contrast, non-taxable distributions have no discernible impact on the discount, again consistent with a tax-based explanation of the discount. Importantly, we are able to rule out a competing short-term trading hypothesis. Taken together, our results highlight the importance of personal taxes in explaining closed-end fund discounts. JEL Classification: G10, G23, H20 Keywords: equity closed-end funds, discounts, personal taxes, return of capital distribution, tax cut, transaction cost * We are very grateful to Murali Jagannathan, Kristian Rydqvist and Steven Schwartz for their invaluable support and comments. We want to thank Hyuna Park, and seminar or conference participants at Alabama A&M University and the Global Finance Association Annual Meeting (2017) for valuable comments. Wu gratefully acknowledges the support form the Foundation to Promote Scholarship at Roger Williams University. Any errors, omissions, or inconsistencies are our own. Corresponding author; College of Business and Public Affairs, Alabama A&M University; 4900 Meridian Street, Normal, AL 35762, shishir.paudel@aamu.edu. Fogelman College of Business and Economics, University of Memphis, 3675 Central Avenue, Memphis, TN 38152, ssilveri@memphis.edu. Mario J. Gabelli School of Business, Roger Williams University; One Old Ferry Road, Bristol, RI 02809, markwu@rwu.edu.

2 1. Introduction One of the enduring anomalies in finance is the so called close-end fund puzzle. A closed-end fund, like the more popular open-end fund, is a mutual fund which typically holds other publicly traded securities. Unlike open-end funds, however, closed-end funds have no obligation to redeem investors shares for cash. The closed-end fund puzzle is the empirical finding that closed-end fund shares often trade at a price below the net asset value (NAV) of the underlying portfolio of securities. 1 Since the value of the assets held by a closed-end fund is observable, fund shares should not deviate from its intrinsic value (i.e., NAV) in the absence of any frictions. The academic literature proposes various explanations, both rational and behavioral, for the discount. Among the rational explanations, the personal tax based explanation is one of the oldest and theoretically prominent. However, empirical evidence on the tax explanation has been mixed. We attempt to isolate the effect of personal taxes on the closed-end fund discount. The personal tax explanation is based on the potential tax liability an investor inherits when she buys closed-end fund shares with embedded gains. The buyer is forced to pay taxes on accumulated gains when the fund realizes (and distributes) those gains, even though the gains were accumulated before she purchased the fund. 2 Although the buyer recoups the tax payments when she sells her investment, she loses the time value of money. The longer the investor plans to hold the fund shares, the larger the potential loss. Thus, the discount reflects an embedded tax liability a shareholder faces. In a seminal paper, Malkiel (1977) shows both theoretically and empirically that taxes can create a closed-end fund discount of up to six percent. Fredman and Scott (1991) also find that the closed-end fund discount is at least partially explained by taxes as do Elton, 1 Net asset value (NAV) refers to the total value of a fund s portfolio net of any expenses/liabilities divided by the number of fund shares outstanding. 2 The Investment Company Act (1940) requires that mutual funds distribute at least 90% of their realized capital gain and 100% of their dividend income to fund shareholders in order to qualify as pass through investment vehicles. 1

3 Gruber, and Blake (2005), Chay, Choi, and Pontiff (2006), and Brennan and Jain (2007). These papers utilize the ex-dividend day price change behavior to isolate the effects of taxes on closedend fund discounts. However, Lee, Shleifer, and Thaler (1991) question the tax explanation by showing that closed-end fund discounts neither increase as the market rises nor decrease as the market falls as the tax explanation would predict. In line with Lee et al. (1991), Brickley, Manaster, and Schallheim (1991) also document a negative correlation between closed-end fund discounts and unrealized capital gains. An underlying problem with the arguments in Lee, Shleifer and Thaler (1991) and Brickley, Manaster and Schallheim (1991) is that there is significant uncertainty regarding the realization of accumulated capital gains by the closed-end funds. If investors do not expect the gain to be realized and paid in the near future, the price should not reflect the discount. The tax argument is also indirectly refuted by Brauer (1984) and Brickley and Schallheim (1985), who show that closed-end fund prices increase to NAV at the time of open-ending or liquidation of the funds. However, this evidence is not necessarily inconsistent with the tax explanation. First, there is a self-selection bias among the funds that open-end. Second, the open-ended or liquidated funds in their sample may not have significant accumulated gains at the time. Third, once the fund decides to open-end (or liquidate), the fund shareholders holding period becomes shorter such that the time value of the prospective tax liability is negligible. The literature investigating the role of taxes on closed-end fund discounts suffers from a fundamental problem, namely, the uncertainty regarding the timing of the realization of the accumulated capital gains or the recouping of benefits from capital losses. A recent paper by Day, Li and Xu (2011) (DLX hereafter) addresses this issue by looking at changes in weekly exdividend closed-end fund discounts. DLX argue that if the closed-end fund discount is related to 2

4 an overhanging tax liability of undistributed (both realized and unrealized) gains, the price drop on the ex-dividend day would be less than the dividend amount because future tax liabilities are reduced. Consistent with the tax hypothesis, DLX find that discounts decline following taxable distributions by closed-end funds. Moreover, they find some (albeit weak) evidence that the reduction in discount is higher for a dividend distribution than for a capital gain distribution. DLX verify and extend the ex-day discount results in Brenan and Jain (2007). Notably, DLX ignore a couple of important institutional features. Firstly, they assume that tax rates remain constant over their sample period ( ). However, the tax rate for both capital gains and dividend income drop significantly during DLX s sample period, especially with the 2003 Bush tax cuts. The fact that DLX assume constant tax rates and also assume that the dividend tax rate is always higher than the capital gains tax rate may explain why some of their results are weak. Secondly, DLX do not distinguish between taxable and non-taxable distributions. Return of capital (ROC) distributions are made by some closed-end funds to maintain a constant dividend policy when the funds do not generate sufficient returns (dividend or capital gains) to distribute. Since ROC distributions return part of the original capital to shareholders, the distributions do not incur any taxes. We address both issues. We separate the overall sample into two tax regimes, a high tax period ( ) and a low tax period ( ). If taxes play a role in shaping the discount, then the discount reduction after a distribution will be higher during high tax periods than during low tax periods. Recognizing that the taxability of distributions is important, we use Morningstar data to distinguish between taxable and non-taxable (ROC) distributions. The benefit of using 3

5 Morningstar distribution data as opposed to CRSP distribution data is its accuracy. 3 If taxes are the driving force behind the discount, then non-taxable distributions will have no effect on the discount. 4 Recognizing changes in the tax rate and including both taxable and non-taxable distributions allows us to test an alternative to the tax hypothesis that DLX were unable to test, namely the short-term trading hypothesis. In the ex-day literature, a price drop of less than the dividend can be explained by trading costs (short-term trading hypothesis), i.e., an ex-day price drop of less than the dividend reflects the trading costs faced by a tax neutral investor. This is because in the absence of trading costs, trading by tax neutral arbitragers can force the price drop to equal the dividend amount (Kalay, 1982; Boyd and Jagannathan, 1994). A similar logic can be applied to closed-end fund distributions as we discuss in the next section. Given that transaction costs should be the same for arbitraging taxable and non-taxable distributions, the short-term trading hypothesis is a plausible alternative hypothesis to the tax hypothesis. We follow the literature and compute the change in discount from two weeks prior to the distribution to the week of the distribution and find that the discount declines by 63 basis points after a taxable distribution. Dividing the sample into two tax regimes, we find that the discount following a taxable distribution declines by 82 basis points during the high tax period and by 43 basis points during the low tax period. For non-taxable distributions, we find no statistically significant change in the discount. If the discount reduction is driven by short-term trading costs then we would expect to find a similar discount reduction for both taxable and non-taxable 3 We discuss this in more detail in Section 3.1. We hand-check inconsistencies between CRSP and Morningstar data by cross-checking the taxability of distributions as reported in a fund s N-30D/CSR EDGAR filing. We find that Morningstar is significantly more accurate than CRSP in assigning the correct taxability status to distributions. 4 DLX use CRSP data to determine the taxability status of distributions but as we note, CRSP often mischaracterizes the taxability of distributions. 4

6 distributions. Given that we find a significant difference in discount reduction between taxable and non-taxable distributions, our results support the tax hypothesis and are inconsistent with the shortterm trading hypothesis. A concern with the methodology used in the literature is that using a two-week window to measure the change in discount is too wide as a host of external factors might impact both the fund price and the NAV. To reduce the effect of confounding factors, we compute ex-day returns (i.e., one day returns) for the full sample and for both the high tax period and the low tax period. The tax hypothesis predicts that taxable distributions will result in positive ex-day returns and that these returns will be larger during the relatively high tax period. We also compare the ex-day returns of non-taxable and taxable distributions. The tax hypothesis predicts taxable distributions will result in positive ex-day returns while non-taxable distributions will have no discernible impact on exday returns. We find that, consistent with the tax hypothesis, the ex-day return for taxable distributions is 48 basis points. Moreover, during the high tax period the ex-day return is 58 basis points compared to 37 basis points during the low tax period and the difference between these exday returns is statistically significant. In contrast, the ex-day returns for non-taxable distributions are statistically indistinguishable from zero, irrespective of the tax regimes. Again, these results from examining ex-day returns are also consistent with the tax hypothesis. We contribute to the literature in several ways. We provide evidence that taxes influence the closed-end fund discount using measures that reduce the impact of confounding factors. We use an exogenous shock tax rate changes and find evidence consistent with the tax hypothesis. We also use non-taxable distributions as an alternative way to test the personal tax hypothesis and find consistent results. Moreover, we rule out the short-term trading hypothesis as a plausible 5

7 alternative. Finally, to the best of our knowledge, we are the first to report CRSP errors in identifying the tax status of closed-end fund distributions. The rest of the paper is organized as follows. Section 2 discusses the related literature and lays out our hypothesis. We describe the data in Section 3 and present the main results in Section 4. Section 5 concludes. 2. Literature Review and Hypothesis Development The basic rationale underpinning the tax explanation of the discount is as follows. When a fund accumulates taxable gains (both realized and unrealized), a discount is created. As a fund makes taxable distributions out of its realized gains, a portion of the embedded tax liabilities are passed directly to fund shareholders and the remaining NAV of the fund will hold less taxable gains. As a result of the taxable distribution, fund shareholders are willing to accept less discount from the closed-end fund, i.e., a reduction in discount. The literature attempting to explain the discount dates back over half a century. Pratt (1966) argues that personal taxes do not matter because an investor will eventually recoup all the taxes she pays on the accumulated gain that she inherited when she purchased the fund shares. The taxes that are paid by the investor will result in tax savings whenever the investor offloads her investment. Malkiel (1977) points out that the investor may not sell fund shares for a long time and that the investor may not have gains to claim the tax shield when she sells the fund shares. He argues that it is the time value of money that a long-term investor will most likely lose if she buys fund shares with an imbedded gain and the fund subsequently realizes and distributes the gain. Malkiel (1977) shows both theoretically and empirically that taxes can create a closed-end fund discount of up to six percent. 6

8 Since Malkiel (1977), there have been many studies that test the relation between the accumulated gain and the discount on closed-end funds. The studies can be divided into two main groups. The first group of papers relates the discount to the general shift in market conditions. Both Lee, Shleifer, and Thaler (1991) and Brickley, Manaster, and Schallheim (1991) find that the closed-end fund discount increases during stock market downswings and decreases during stock market upswings. This negative relation between the closed-end fund discount and the stock market performance is potentially inconsistent with the tax hypothesis. Brickley, Manaster, and Schallheim (1991) associate the phenomenon with the tax timing option investors lose when they hold closed-end fund shares instead of investing in the portfolio of stocks held by the funds. A host of confounding factors and the uncertainty of a fund s distribution amount and timing make estimating the net impact of overhanging tax liabilities on discount levels problematic. The second group of papers relates the discount to changes in the fund price on particular dividend distribution days and generally finds support for the tax story. Elton, Gruber, and Blake (2005) find that the ex-day price drop for a taxable distribution by a corporate bond fund is smaller than the dividend amount while the ex-day price drop for a non-taxable distribution by a municipal bond fund is bigger than the dividend amount, consistent with the tax hypothesis. Chay, Choi, and Pontiff (2006) find that the price on the ex-day drops by 93 cents for every dollar of realized capital gains distributed by equity closed-end funds and interpret this drop as the difference between the actual tax rate on realized capital gains and the effective tax rate on unrealized price appreciation. Brennan and Jain (2007) find that the discount declines for a taxable capital gain distribution both on the announcement day and on the ex-dividend day. They interpret the announcement day behavior as evidence of an agency cost effect and the ex-day behavior as evidence of a tax effect. 7

9 DLX test the tax hypothesis by examining distribution-related changes in discounts rather than discount levels. They argue that the price drop will be less than the dividend amount due to the reduction in the tax liability arising from the realization of the accumulated gain that formed part of the NAV. Therefore, a buyer will be willing to accept a lower discount to buy closed-end fund shares after the distribution of taxable gains. They find that discounts decline following taxable distributions and also find some (weak) evidence that the reduction in discount is larger for dividend distributions than for capital gain distributions. Overall, their evidence is consistent with the tax hypothesis. However, DLX ignore tax rate changes that occur during their sample period ( ). The capital gain tax rate dropped from 28 percent to 20 percent on May 6, 1997 and both the dividend and capital gain tax rates were lowered from 38.6 and 20 percent to 15 and 15 percent respectively on May 5, If taxes matter then the drop in discount following a taxable distribution should be greater during high tax periods than during low tax periods. Accordingly, we divide the sample into a high and a low tax rate period and have the following hypotheses. Hypothesis 1: Taxable distributions lead to a reduction in the closed-end fund discount. Hypothesis 2: For a given taxable distribution, the closed-end fund discount falls by a greater amount during high tax periods than during low tax periods. While the DLX results are consistent with the tax hypothesis, the results are also consistent with the short-term trading hypothesis discussed extensively in the ex-dividend day price drop 8

10 literature. 5 An ex-day price decline less than the dividend amount provides an arbitrage opportunity for a trader who is indifferent to taxes. 6 The implication for closed-end fund discounts can be illustrated by the example in Panel A of Table 1. The reduction in discount before and after the distribution ($5) from the cum-day to the ex-day provides a profit opportunity to short-term investors (arbitrageurs). In the example, the arbitrageurs can make $1 per share by buying the fund shares on the cum-day, collecting the dividend, and selling the fund shares on the ex-day (i.e., = $1). If the two-way transaction cost is at least $1, the reduction in fund discount will stay at 0.5% in the example. As long as the two-way transaction cost is no more than $1, the trades of arbitrageurs will put upward pressure on the cum-day price and downward pressure on the exday price until the arbitrage profit disappears, which translates to a reduction in fund discounts smaller than 0.5%. Hence, the decline in discount on the ex-dividend day is consistent with both the lower overhanging tax liabilities in the fund (the tax hypothesis) and the short-term trading (the short-term trading hypothesis). DLX do not test this alternative hypothesis. [Insert Table 1 about here] In addition, as trading costs have declined during DLX s sample period, the smaller reduction in discounts during the low tax period (i.e., post 2002), as tested in Hypothesis 2, might be the result of lower transaction costs. To separate out the effect of taxes from that of transaction costs, we make use of return of capital (ROC) distributions. These are non-taxable distributions that some funds make to maintain a dividend level when the funds do not have sufficient gains available to distribute. Panel B of Table 1 replicates the example from Panel A for an ROC distribution. Per the tax hypothesis, the prospective buyer will not demand any discount either 5 See, for example, Kalay (1982), Karpoff and Walkling (1988, 1990), Boyd and Jagannathan (1994), Koski (1996), and Michaely and Villa (1996). 6 Either the trader is a tax free entity, e.g., a pension fund, or the trader s long-term and short-term tax rates are the same, e.g., market makers, floor traders (Boyd and Jagannathan (1994, p. 712)). 9

11 before or after the ROC distribution and thus, no tax liabilities are reduced. However, absent any transaction costs, a sizeable reduction in discount (or a small drop in price compared to the dividend) can be arbitraged away for both taxable as well as non-taxable distributions. To ensure that the discount reduction following a taxable distribution is due to taxes, we test the following hypothesis for ROC distributions. Hypothesis 3: For non-taxable distributions, the closed-end fund discount will not change on the ex-dividend day (irrespective of the tax regimes high or low). To test our hypotheses, we first follow DLX s approach by examining the change in discount during a two week window (from week -2 to week 0, where week 0 is the ex-dividend week) to measure the effect of taxes on closed-end fund discount. It is possible that both price and NAV (constituents of the discount) are affected by a host of factors during the two week period. To address this concern, we supplement our analysis by comparing the ex-day price drop to the dividend amount. Since the NAV declines by the dividend amount on the ex-day, the reduction in discount should solely come from a reduction in price that is less than the dividend amount. In other words, since the NAV drops by the amount of the dividend on the ex-day, a price drop smaller than the dividend effectively reflects the drop in the fund discount on the ex-day. Algebraically, this can be illustrated as follows: Δ Discount = Discount ex Discount cum = P ex NAV ex NAV ex P cum NAV cum NAV cum 10

12 adding the dividend to both the price and the NAV on the ex-day (DLX Table 2) yields: 7 Δ Discount = (P ex + Div) (NAV ex + Div) (NAV ex + Div) and noting that NAV ex + Div = NAV cum yields: P cum NAV cum NAV cum Δ Discount = P ex P cum + Div NAV cum (1) This final equation is very close to the ex-day fund return except that the denominator is NAV cum as opposed to P cum. We see that if the price drop (P ex P cum ) is exactly the dividend amount, then change of fund discount is equal to zero. If the price drop on the ex-day is less than the dividend amount, as predicted by the tax hypothesis, then the change in fund discount is positive. Focusing on the ex-day return (i.e., a one day return rather than a two week return) significantly reduces the impact confounding factors may have on the discount. We test the following hypotheses for the closed-end fund ex-day price decline. Hypothesis 4a: The ex-day return for taxable distributions is larger during high tax regime periods than during low tax regime periods. Hypothesis 4b: The ex-day return for ROC distributions is zero irrespective of tax regimes (i.e., the ex-day price drop will equal the dividend amount for ROC distributions). 3. Data and Methodology 7 This adjustment is needed due to, what DLX call, the mechanical impact of adjustments to NAV since the ex-day NAV is used in the denominator when calculating the ex-day discount. 11

13 3.1. Data and CRSP Errors Our data comes from two sources: Morningstar and CRSP. We collect fund price, NAV, dividend, short-term capital gain, medium-term capital gain, long-term capital gain and return of capital distributions data from Morningstar. Following the literature, we focus on equity funds (i.e., funds that invest at least 60% of their assets in equity). Following DLX and to avoid biases reported in the literature about newly established or close-to-obsolete funds trading at a premium (or small discount), we omit observations that are within six months from the commencement of a fund or within six months from an obsolete date. We also exclude funds with less than six months of available data and funds with monthly dividends (DLX). We compute the discount as the natural log of the ratio of the fund price to the NAV (Pontiff (1995); DLX). We use CRSP data when computing ex-day returns (i.e., when testing Hypothesis 4a and Hypothesis 4b). [Insert Table 2 about here] To the best of our knowledge, we are the first to report CRSP errors in identifying the tax status of closed-end fund distributions. DLX use CRSP data and drop 108 distributions from their initial sample because CRSP reports these distributions as non-taxable (p. 582, Footnote 9). However, upon further inspection we find that many more distributions that CRSP reports as taxable are actually non-taxable. To determine the integrity of the data, we cross-check Morningstar ROC distributions with CRSP distributions and a fund s EDGAR filings (N- 30D/CSR). We find that whenever a distribution involves a return of capital, CRSP erroneously classifies it as taxable (either as a regular dividend distribution or a long-term capital gain distribution). In contrast, the Morningstar data is much more accurate in classifying the distribution type. In Table 2, we present two examples of fund distributions to illustrate the data issues that we 12

14 identify. 8 The table is organized by showing, from left to right, distribution dates, amounts and types according to Morningstar, distribution dates, types and amounts according to CRSP, and finally the actual amounts and types (yearly) from the funds SEC filings. For Aberdeen Australia Fund, the annual distributions of $0.68 are all ROC distributions, CRSP erroneously classifies all distributions to be taxable dividends (CRSP distribution type distcd=1232, according to the CRSP Distribution Code Documentation, refers to US cash dividend, quarterly, taxable same rate as dividends ). The example of Nuveen Core Equity Alpha involves both dividend and ROC distributions. Morningstar data ($0.15 of dividend and $1.35 of ROC distributions mostly match with numbers in the EDGAR filings. Again, CRSP mechanically classifies all quarterly distributions to be taxable dividends. These examples are just the tip of the iceberg and the errors of CRSP are non-trivial and not random. Nontaxable distributions account for only 3.79% (=108/2847) in DLX according to CRSP, but are as high as 21.76% according to Morningstar in our similar sample from In fact, when at least a fraction of the distributions consists of ROC, we see that CRSP erroneously identifies those distributions as regular taxable dividend or long-term capital gain distributions. We are yet to find any ROC case which is correctly identified as ROC by CRSP, and we are not aware of any CRSP distribution codes that identify short-term capital gains distributions correctly. The incorrect classification of distribution type by CRSP calls into question the accuracy of the data used in DLX. This may also explain the relatively weak results DLX report when exploring the differential impact on the discount of dividend distributions versus capital gain distributions. As a result, we use Morningstar distribution data which more accurately reports the various components of taxable distributions (regular dividends, short-term capital gains, long-term capital gains) as well as ROC distributions. 8 Aberdeen Australia Fund is an international closed-end fund that was founded in 1985, and Nuveen Core Equity Alpha is an U.S. closed-end fund that was founded in

15 3.2. Descriptive Statistics We provide descriptive statistics in Table 3. Consistent with the literature, funds on average trade at a discount of about 9%. The discount varies over the two sub-sample periods. The average discount is more than 10% during the high tax period ( ) and less than 8% during the low tax period ( ), indicating that taxes may matter. While the majority of our sample contains taxable distributions, the number of non-taxable distributions is non-trivial and increased during the second half of the sample period. The reduction in discount after a taxable distribution averages 63 basis points in the full sample, and is statistically significant. Notably, the discount reduction is almost twice as large during the high tax period than during the low tax period (82 basis points versus 43 basis points, both are statistically significant). For the non-taxable (i.e., ROC) distributions, however, the change in discount is statistically indistinguishable from zero in the overall sample and during both the high and low tax periods. 9 The descriptive statistics provide preliminary support for our hypotheses. [Insert Table 3 about here] To see the impact on fund discount during the event window, in Figure 1 we plot the weekly discount dynamics over a thirteen-week period centered on the ex-day, i.e., starting six weeks before the ex-day week and extending to six weeks after the ex-day week. For taxable distributions, the discount declines prior to the distribution and then remains relatively flat after the distribution. In contrast, for non-taxable distributions the discount does not change much either before or after the distribution. 9 It is worth noting that on a particular ex-dividend day, we may have one of the following three situations in the data: (a) a stand-alone ROC distribution (labeled as Pure ROC in Table 3), (b) an ROC distribution that is combined with other taxable distributions (either dividend or capital gain) on the same ex-day (labeled as Mixed ROC in Table 3), and (c) a taxable distribution. In our following main analysis (Table 4), we treat ROC as a continuous variable (that goes from 0% for a pure taxable distribution to 100% for a pure ROC distribution). In later analysis (Table 5), we treat ROC as an indicator variable depending on the proportion of ROC within the total distributions. Results in this paper are similar whether we treat ROC as an indicator variable or a ratio. 14

16 [Insert Figure 1 about here] 3.3. Methodology We follow DLX to test our first three hypotheses by looking at the change in discount over a two-week window and including the demeaned discount two weeks prior to the distribution to control for clustering. We add to the DLX regressions ROC, a continuous variable that goes from 0% for a pure taxable distribution to 100% for a pure return of capital distribution and HighTax, an indicator variable that takes the value one if the distribution is paid during the period when the personal tax rates on dividends and capital gains were relatively high, and zero otherwise. As funds with a managed dividend policy (MDP) tend to have a lower discount and are more likely to distribute a return of capital (Johnson, Lin, and Song, 2006; Wang and Nanda, 2011), we also add MDP, an indicator variable that takes the value one whenever the fund employs a MDP. 10 We also estimate various versions of the base model by including interaction terms between the regressors, and the regressions take the following general form: Δ Discount = β 0 + β 1 ROC i,t + β 2 DemeanDisc i,t + β 3 HighTax t + β 4 MDP i,t + β 5 DivYield i,t + β j X i,t + ε i,t. n j=6 (2) The DLX analysis uses weekly data to compute the discount and measures the change in discount over two (and four)-week period. We also follow the literature and use weekly discount data to test our main hypotheses. However, as we mentioned in Section 2, the two-week period may bring in a lot of noise to our analysis because many exogenous factors affecting both the fund price and the NAV. To verify that our results are not driven by other potential factors, we also conduct robustness tests (Hypotheses 4a and 4b) using the methods widely used in ex-day price 10 Johnson, Lin, and Song (2006) find that dividend signaling explains why funds employ MDPs while Wang and Nanda (2011) find support for an agency cost explanation. In either case, the motivation to employ MDPs is unrelated to taxes. 15

17 drop literature. We compare ex-day returns (fund price data from CRSP) for taxable and nontaxable distributions to test tax and short-term trading explanations of ex-day price behavior. Assuming that the NAV declines by the amount of distribution on ex-day, the reduction in discount exclusively comes from the change in fund price. Thus the ex-day return is directly related to the change in discount (shown in Section 2) Empirical Results We report the regression results in Table 4 with four model specifications. In the first model, we regress change in discount only on nontaxable proportion (a continuous variable ranges from 0% to 100%). The constant term in the regressions reflects the reduction in the discount attributable to taxes (when the nontaxable proportion equals 0%, Hypothesis 1). We see in Model (1) that the positive and statistically significant constant term corresponds to a 59 basis points reduction in the discount, consistent with the tax hypothesis. If a distribution is purely ROC, then this translates to a discount increase of 29 basis points, but this change is not statistically significant (p-value=0.14). This result is consistent with both Hypothesis 1 and the first half of Hypothesis 3. Overall, for a 10 percent increase (decrease) in the proportion of nontaxable distributions, fund discount would go up (down) by 8.8 basis points. In Model (2) we add various controls to the base model, and get similar results. In Model (3), we add the High Tax indicator variable and find that its coefficient is also positive and statistically significant. When we add the High Tax coefficient with the constant term, we see that the discount reduction is 70 basis points for taxable distributions during the high tax period compared to 23 basis points during the low tax period and the difference between the two periods is statistically significant. These results are consistent with Hypothesis As we noted for Equation (1), since the denominator is NAV cum as opposed to the lower P cum, the fund ex-day return could be a slight overestimate of the actual change in fund discount. 16

18 In Model (4) we test the second half of Hypothesis 3 by adding the ROC and High Tax interaction variable to Model (3). The negative and statistically significant coefficient on ROC offsets the positive constant term so that the net effect for non-taxable distributions is a 24 basis points discount increase but this net effect is still not statistically significant (p-value=0.28) as in all four model specifications. Also, from the coefficient on ROC_H, we see that the effect on ROC is not distinguishably different regardless where the tax level is at. Thus, non-taxable distributions have a statistically indistinguishable from zero impact on the discount, consistent with Hypothesis 3. [Insert Table 4 about here] Overall, our results are consistent with the tax hypothesis highlight the impact personal taxes have on the discount. The discount reduction after taxable distributions is greater during high tax periods than during low tax periods while non-taxable distributions do not impact the discount. Differentiating between high and low tax regimes and taxable versus non-taxable distributions also allows us to test and rule out an alternative hypothesis, namely the short-term trading hypothesis we discussed earlier. If the short-term trading hypothesis were at play, then we would expect a similar reduction in the discount during both the high and low tax regimes and for both taxable and non-taxable distributions. Given that we find a significant difference in discount reduction during the high tax regime versus the low tax regime and that we also find a significant difference in discount reduction for taxable versus non-taxable distributions then we can rule out this alternative hypothesis Additional Tests 12 As trading costs have reduced through time, it can be argued that the high tax period corresponds to one where transaction costs were also relatively high so that the larger reduction in the discount in the high tax period just reflects the higher trading costs of the time (trading in the 1990s cost more than trading in the 2000s). However, the result that taxable and non-taxable distributions have a differential impact on the discount coupled with the fact that most of the non-taxable distributions occurred later in the sample period (see Table 3) is evidence inconsistent with the short-term trading hypothesis but consistent with the tax hypothesis. 17

19 In Table 5, we redefine ROC as an indicator variable and replicate our results from Table 4. In Panel A, ROC equals one if a given distribution contains any nontaxable portion, and in Panel B, ROC equals one if a given distribution is at least 50% nontaxable. From left to right, each panel includes four model specifications as in previous Table 4. Our robustness results are consistent with our main results from Table 4, and again show that in general, taxable distributions have the effect of lowering discount for the two-week period ending in ex-dividend week. For example, in Panel A, model specification (1), when ROC takes on a value of zero, the average reduction in discount for a taxable distribution over the two-week period is 63 basis points (the coefficient of the intercept), and is statistically significant. Dividing 63 basis points by the level of the discount, this reduction of discount is about 0.63/8.48= 7.4% of the average discount in ex-day week during our sample period. On the other hand, when ROC takes on a value of one, the average change in discount becomes = -0.12, which implies that a ROC distribution would increase fund discount by 12 basis points. The effect of ROC distributions is not statistically different form zero (p-value=0.42). Results in the other three models and in Panel B are largely the same as previous results from Table 4. Our results are very robust whether we treat ROC as a continuous or discrete variable. [Insert Table 5 about here] As we mention in Section 2, there are many confounding factors that may impact both the fund price and the NAV during the two-week window that DLX use in their analysis. We thus conduct alternative tests using methods from the ex-day price drop literature. In particular, we compare ex-day returns for taxable and non-taxable distributions to test the tax hypothesis and the short-term trading hypothesis. As we show in Section 2, the ex-day return is directly related to the 18

20 change in discount. Thus, using ex-day returns reduces the impact of confounding factors as these returns are measured on one day (the ex-day) rather than over a two week period. [Insert Table 6 about here] The tax story predicts that ex-day returns for taxable distributions are larger during high tax periods (Hypothesis 4a) and that ex-day returns are not significant for non-taxable distributions (Hypothesis 4b). We report the results in Table 6. We see that for taxable distributions, ex-day returns are a significantly positive 48 basis points in the full sample, 58 basis points during the high tax period and 37 basis points during the low tax period. Moreover, the 21 basis points difference between the high and low tax periods is statistically significant (p-value=0.01). These results provide support for Hypothesis 4a. In contrast, the ex-day returns are statistically indistinguishable from zero for non-taxable distributions in the overall sample as well as in the high tax and the low tax periods, consistent with Hypothesis 4b. By narrowing down our analysis to daily returns, our results are qualitatively similar; in other words, even though our event study has a two-week event window, our earlier results do not seem to be driven by potential confounding factors. Overall, these results are consistent with the tax story and reinforce our regression results. 5. Conclusion We revisit the role taxes have on closed-end fund discounts and extend the literature in a few ways. While DLX find that the discount on closed-end funds declines after a taxable distribution and find some (albeit weak) evidence that the discount is lower for a dividend distribution than for a capital gain distribution, they ignore significant changes in the tax rate over time and ignore non-taxable distributions. As a result, they are unable to test whether the drop in 19

21 discount reflects the long-term investor s marginal gain taxes (tax hypothesis) or the tax neutral investor s transaction costs (short-term trading hypothesis). In addition, papers such as DLX that rely on CRSP to identify taxable distributions need to be aware that CRSP often characterizes the tax status of distributions incorrectly. We address the above issues as follows. As tax rates for both capital gain and dividend income drop significantly after 2002, we separate the sample into two tax regimes the high tax regime ( ) and the low tax regime ( ). In addition, we include non-taxable distributions in our analysis. As CRSP often incorrectly characterizes the tax status of distributions we use Morningstar data and manually check any inconsistencies between Morningstar and CRSP against a closed-end fund s EDGAR filings. We find that the reduction in discount is associated with the extent to which distributions are taxed. Moreover, changes in the discount are more sensitive to overhanging tax liabilities when tax rates are high and that non-taxable distributions have, on average, no discernible impact on the discount, irrespective of the tax regime. Our results support the tax hypothesis and provide evidence inconsistent with the short-term trading hypothesis. Overall, our results highlight the importance of expected future tax liabilities in explaining the closed-end fund discount. 20

22 References Boyd, J.H. and Jagannathan, R., Ex-Dividend Price Behavior of Common Stocks. Review of Financial Studies 7, Brickley, J., Manaster, S., Schallheim, J., The tax-timing option and the discounts on closedend investment companies. Journal of Business 64, Brennan, M and Jain, R, Capital gains taxes, agency costs and closed-end fund discounts. Working paper, Univ. Calif. L.A. Chay, J., Choi, D., Pontiff, J., Market valuation of tax-timing options: evidence from capital gains distributions. Journal of Finance 61, Day. T.E., Li, G.Z., Xu, Y., Dividend distributions and closed-end fund discounts. Journal of Financial Economics 100, Elton E.J., Gruber M.J., Blake C.R., Marginal stockholder tax effects and ex-dividend-day price behavior: evidence from taxable versus nontaxable closed-end funds. Review of Economics and Statistics 87, Fredman, A., Scott, G., Investing in Closed-End Funds: Finding Value and Building Wealth, New York. New York Institute of Finance. Johnson, S., Lin, J.C., Song, K., Dividend policy, signaling, and discounts on closed-end funds. Journal of Financial Economics 81, Kalay, A., The ex-dividend day behavior of stock prices: a re-examination of the clientele effect. Journal of Finance 37, Karpoff, J. F. and Walkling, R. L., Short term trading around ex-dividend days: Additional evidence. Journal of Financial Economics 21, Karpoff, J. F. and Walkling, R. L., Dividend capture in NASDAQ stocks. Journal of Financial Economics 28, Klein, P., The capital gain lock-in effect and long-horizon return reversal. Journal of Financial Economics 59, Koski, J., A Microstructure Analysis of Ex-Dividend Stock Price Behavior Before and After the 1984 and 1986 Tax Reform Acts. Journal of Business 69, Lee, C., Shleifer, A., Thaler, R., Investor sentiment and the closed-end fund puzzle. Journal of Finance 46,

23 Malkiel, B., The valuation of closed-end investment-company shares. Journal of Finance 32, Michaely, R. and Vila, J., Trading Volume with Private Valuation: Evidence from the Ex- Dividend Day. Review of Financial Studies 9, Pontiff, J., Closed-end fund premiums and returns: implications for financial market equilibrium. Journal of Financial Economics 37, Pontiff, J., Costly arbitrage: evidence from closed-end funds. Quarterly Journal of Economics 87, Pontiff, J., Excess volatility and closed-end funds. American Economic Review 87, Pratt, E., Myths associated with closed-end investment company discounts. Financial Analysts Journal 22, Wang, Z.J., Nanda V., Payout policies and closed-end fund discounts: signaling, agency costs, and the role of institutional investors. Journal of Financial Intermediation 20:

24 Table 1 Examples of Taxable and ROC Distributions on Closed-End Fund Discount Timeline Taxable Distributions: IPO Date Cum-date Dividend Amount Ex-date Change (cum to ex) NAV Basis Accumulated gain Tax rate on realized gain 20% 20% Built-in tax liability inherited if the fund realizes the gain immediately (Discount required by the long-term retail investor) Price the long-term retail investor willing to pay Max discount required by long-term retail investor % 9.5% 0.5% Buy fund Collect Sell fund cum to ex Strategy for short-term arbitragers share Dividend share profit Arbitragers cash flow ROC Distributions: NAV Basis Accumulated gain Tax rate on realized gain 20% 20% Built-in tax liability inherited if the fund realizes the gain immediately (Discount required by the long-term retail investor) Price the long-term retail investor willing to pay Max discount required by long-term retail investor Strategy for short-term arbitragers % 0.0% 0.0% N/A This table presents two examples of taxable vs. ROC distributions. In both cases, we assume $5 distribution and tax rate of 20% on realized gain. To magnify the tax implications, we also assume that the buyer plans to keep the fund for an indefinite time. 23

25 Table 2 Closed-End Fund Distributions Data Discrepancy among Morningstar (M star), CRSP and Edgar Few Examples M star M star M star M star M star CRSP CRSP Edgar M star Type CRSP date Date Amount Amount Type Total Type* Amt Amt Fund Name Aberdeen Australia Fund Dec Div 0.19 ROC Aberdeen Australia Fund Mar Div 0.17 ROC Aberdeen Australia Fund Jun Div 0.16 ROC Aberdeen Australia Fund Sep Div 0.15 ROC ROC 0.68 FY Ending 10/31/ Nuveen Core Equity Alpha Mar Div 0.37 ROC Nuveen Core Equity Alpha Jun Div 0.37 ROC Nuveen Core Equity Alpha Sep Div 0.35 ROC Div 0.16 Nuveen Core Equity Alpha Dec Div 0.26 ROC ROC 1.34 FY Ending 12/31/ The table compares the distribution amount and types reported in Morningstar, CRSP and Edgar simultaneously. Div refers to the Investment Income and ROC to the Return of Capital distributions respectively. *According to the CRSP Distribution Code Documentation, distribution type = 1232 (distcd=1232) refers to "US cash dividend, quarterly, taxable same rate as dividends." Edgar Type Yearly Edgar Amount Yearly 24

26 Table 3 Sample Statistics Full Sample High Tax Regime Low Tax Regime ( ) ( ) ( ) Average discount (%) Number of funds Number of distributions Pure ROC Mixed ROC ROC (Total) Taxable Average discount (%) (Week 0) ROC Taxable Average change in discount (%) (Week -2 to Week 0) ROC Taxable 0.63 *** 0.82 *** 0.43 *** Average distribution ($) ROC Taxable Average payout (%) ROC Taxable The sample comprises 170 equity closed-end funds that traded in US exchanges between 1993 and Pure ROC represents the ex-day observations when there are no other taxable distributions (dividend, short-term or long-term capital gain) present for that ex-day as reported by Morningstar. Mixed ROC represents the ex-day observations when in addition to the ROC, there is at least one taxable distribution present as reported by Morningstar. ROC represents the ex-day observations which are either Pure ROC or Mixed ROC. Discount is defined as Discount = ln( Price NAV ). Distribution is the dollar amount paid by the funds on ex-day. The average payout is the ratio of distribution amount over cum-day price. Price, NAV, and Distribution data come from Morningstar. Week 0 is the ex-dividend day week and Week -2 is two weeks prior to the ex-dividend day week. *** Different from zero at 1% level. 25

27 Discount (%) Discount (%) Figure 1 Discount Behavior for Taxable and ROC Distributions around Ex-Distribution Week A. Taxable Distributions Week B. ROC Distributions Week In this figure, we plot closed-end fund discounts around the event of distribution. The graph on the top presents the discount behavior for taxable distributions while the bottom one presents ROC distributions. Week 0 denotes the exdistribution week. 26

28 Table 4 Regression of Change in Discount on Non-Taxable Proportions (Continuous Variable) and Other Controls (1) (2) (3) (4) Constant P > t ROC (NT_Prop) P > t Demeandisc P > t H_regime P > t ROC_H 0.37 P > t 0.63 MDP P > t Tpayout P > t Constant + ROC P > t Constant + H_regime P > t Constant + ROC + H_regime + ROC_H 0.59 P > t 0.66 N In this table, we report the regression results of taxes impact on change of closed-end fund discounts. In each of the estimation, we use the full sample of 3,750 observations. ROC is a continuous variable that ranges from zero, when the distributions is full taxable, to one, when the distribution is fully return of capital. Demeandisc (as defined in DLX Table 3) is the mean discount on the fund two weeks prior to the ex-dividend day week (Week -2) minus the cross sectional mean discount on Week -2 of all the closed-end funds in the sample. H_regime is an indicator variable that takes the value of one for the period and value of zero for ROC_H is the interaction between the variables ROC and H_regime. MDP is another indicator variable that takes the value of one if a fund on the exday has managed distribution policy in place; otherwise MDP takes the value of zero. Tpayout is the ratio of distribution amount over the fund price measured two weeks before the ex-day week (Week -2). P>t is the probability that the estimated coefficient is not statistically different from zero. 27