Payout Policies and Closed-end Fund Discounts: Signaling, Agency Costs, and the Role of Institutional Investors

Transcription

1 Payout Policies and Closed-end Fund Discounts: Signaling, Agency Costs, and the Role of Institutional Investors Z. Jay Wang,VikramNanda University of Illinois at Urbana-Champaign, Georgia Institute of Technology We are especially grateful to James Curtis at the SEC for numerous discussions on the institutional details of the managed distribution policy. We thank Murillo Campello, Martin Cherkes, Prachi Deuskar, Fangjian Fu, Michael Weisbach, Lu Zhang, and seminar participants at the 2008 Financial Intermediation Research Society Conference, the 2008 China International Conference in Finance, and the 2009 American Finance Association Meetings for many helpful comments. We thank Don Cassidy at Lipper Analytical Services for providing data. We thank Marek Jochec and Phil Rosen for excellent research assistance. We acknowledge the support from National Natural Science Foundation of China (Project ). The authors are responsible for all the errors. Corresponding author. Fax: addresses: (Z. J. Wang), (V. Nanda)

2 Payout Policies and Closed-end Fund Discount: Signaling, Agency Costs, and the Role of Institutional Investors Abstract The adoption of a Managed Distribution Policy or Plan (MDP) by closed-end funds appears effective in dramatically reducing, even eliminating, fund discounts. We investigate two possible explanations: the signaling explanation proposed in the literature, that the MDP serves as a positive signal of future fund performance, and an alternative explanation based on agency costs. Our results indicate that signaling is, at best, only part of the explanation and that the evidence is generally more consistent with the agency cost hypothesis. For funds adopting aggressive payout targets of 10% (median target) and above, discounts tend to disappear, though there is no discernible improvement in NAV performance. Consistent with the agency cost hypothesis, it is often pressure from institutions/large shareholders that leads to the adoption of aggressive payout policies. Moreover, aggressive MDPs are associated with a decrease in fund size and managerial fees. Suggestive of their activist role in MDP adoptions and/or informed trading, institutions especially ones that are Value oriented tend to build up their holdings in a fund prior to the adoption of an aggressive MDP, and liquidate their positions once the price rises.

3 1 Introduction The closed-end fund (CEF) discount remains one of the more intriguing anomalies in the finance literature. The anomaly refers to the fact that shares of closed-end funds tend to trade at a discount relative to their net asset value (NAV) the magnitude of the discount varying a good deal across funds and over time. 1 In recent years, some CEFs, often under pressure from outside shareholders, have adopted Managed Distribution Policies or Plans (MDPs) to reduce fund discounts. Under MDPs, fund management prescribes a minimum payout target in any given fiscal year, regardless of the realized performance of the underlying asset portfolio. 2 Typically, the policy involves quarterly ormonthlydistributiontocommonshareholders,eitherasafixed percentage of average NAV or as a fixed dollar amount. For MDP funds in our sample, the payout target ranges from a low of 5% to a high of 20%, with a median around 10% in most years. The payout target is met through investment income and realized short-term/long-term capital gains, with any shortfall being covered by a distribution of fund capital. In our analysis, consistent with industry norms 3 and with previous research (Johnson et al. 2006), we use the 10% payout target as the cutoff for moderate versus aggressive payout policies. The MDPs appear to have had remarkable success in reducing, if not eliminating, fund discounts (see Wang 2004 and Johnson et al. 2006). In our sample, for instance, MDP funds exhibit an average (monthly) median discount of only 0.86% over the 1990 to 2006 period, with the more aggressive MDP funds (payout targets 10%) trading at an average premium of 2.32%. In striking contrast, funds without the MDP (hereafter, non-mdp Funds), have an average (monthly) median discount of 10.19% over the same period. In this paper we seek to understand the popularity of MDPs and their apparent success in reducing fund discounts. We consider two primary hypotheses signaling and agency costs and explore their empirical implications. The signaling hypothesis advanced in the literature (Johnson et al. 2006) is that a MDP can serve as a costly signal of the fund s future NAV performance. The claim, based on somewhat informal arguments, is that MDP funds that perform poorly may be obliged to return capital to 1 As with open-end funds, CEFs typically invest in publicly traded securities and manage their holdings for income and capital appreciation. Unlike open-end funds, which offer and sell their shares continuously, CEFs raise capital by selling publicly traded shares in an IPO and subsequent equity offerings. Open-end funds provide liquidity to their shareholders by redeeming shares at a price based on NAV. Shareholders of CEFs, on the other hand, can sell their shares on the secondary market at prices that may vary significantly from NAV. Lee et al. (1991) provide a detailed description of the characteristics of CEF discounts. 2 While funds can always terminate the payout policy, they hesitate to do so given the likely negative shareholder reaction that would ensue (see, for instance, footnote 5). It is reasonable to assume, therefore, that investors would perceive MDPs as a commitment they expect fund management to honor. 3 The generally accepted view in the industry is that to be effective (in terms of reducing discount), MDPs should involve the distribution of at least 10% of average net asset value on an annual basis. See The Investor s Guide to Closed-End Funds, May 2007 (pg. 27), published by Thomas J. Herzfeld Advisors Inc. The issue also describes the history of the MDP and provides a current list of MDP funds. 1

4 investors. Hence, only funds sufficiently confident of future performance would commit to such a target policy. A key implication of the signaling hypothesis is that MDP adoption signals strong future NAV performance and will, therefore, tend to boost share prices and reduce discounts. We would also expect, as is suggested by most formal signaling models, that a stronger signal in the form of a more aggressive MDP will presage an even stronger performance and induce a larger price boost. A caveat in our approach to testing the signaling hypothesis should be pointed out: we take the arguments made in the literature at their face value and assume that a signaling equilibrium in MDPs can exist. However, as we briefly discuss, the theoretical basis for such a signaling equilibrium is weak. The reason is that better performing funds face greater costs from adopting MDPs and may have little incentive to signal in this fashion. Hence, any evidence supportive of MDP signaling needs to be interpreted with some caution. An alternative approach to understanding MDPs is based on the agency cost hypothesis. In developing the hypothesis, we recognize the somewhat different ways in which agency problems may be manifested and impact fund discounts and performance. For instance, agency problems between managers and investors could lead some funds to become large relative to managerial investment abilities/opportunities. In these cases, inducing funds to shrink their size (or moderate their growth) benefits investors, 4 with the adoption of an aggressive payout policy leading to an improvement in both fund performance and discount. However, MDPs may not necessarily improve fund performance when agency problems are present. The reason is that fund discounts may, for instance, reflect the rent extracted by fund managers in excess of value-added (Berk and Stanton 2007; Cherkes, Sagi and Stanton 2009). The adoption of an MDP can induce a wealth transfer from fund managers to shareholders (Cherkes, Sagi and Wang 2009). In this case, fund discounts will be lower due to reduction of the managerial claim on fund assets though there may not be any discernible improvement in the fund s NAV performance per se. An essential element of the agency cost hypothesis is that managers have little incentive to reduce the assets under their control since their fees increase with fund size. Hence, unlike the signaling hypothesis, the adoption of an MDP, especially one with a high payout level, is expected to be involuntary, associated with intervention by shareholders (actual or anticipated). The two hypotheses give rise to a number of empirical predictions that we test. We begin by testing predictions of the signaling hypothesis and examine whether fund discounts and performance are consistent with a signaling explanation for MDPs. We then proceed with tests of the agency cost hypothesis. Specifically, we examine the impact of MDP adoption on fund size and managerial compensation. This is followed by an investigation of the role of outside (institutional) shareholders in inducing MDP adoptions and the shareholding and trading by institutional shareholders around 4 The notion that performance may decline with fund size has been made by several papers in the context of open-end funds (e.g., Berk and Green, 2004). 2

5 these adoptions. Our tests of the signaling hypothesis are based on the predicted relation between the level of MDP and the improvement in discount and performance of MDP funds. Both the signaling and the agency hypotheses predict that funds adopting a more aggressive payout polices will experience a larger improvement in discount, though the underlying mechanisms are quite different. The signaling hypothesis predicts that more aggressive MDP funds will also exhibit a greater improvement in NAV performance; while the agency cost hypothesis does not make such a clear prediction. Our finding is that, consistent with Johnson et al. (2006), MDP funds exhibit smaller discounts and stronger NAV performance on average, compared to similar non-mdp funds. Within the group of MDP funds, however, the performance is stronger only for funds with moderate payout targets (below 10%), contrary to the predictions of signaling model. Moreover, the improvement in discount is mainly evident among the more aggressive MDP funds. Similar results are found when we use matched sample analysis to examine the changes in NAV performance and discounts around the adoption of MDPs. As an alternative to the matched sample approach, we use a panel specification with style and time fixed-effects and several additional variables such as fund turnovers and expenses. The results confirm that the adoption of an aggressive MDP is particularly effective in reducing a fund s discount. However, there is no reliable evidence that performance improves for the funds adopting MDPs (whether moderate or aggressive payout). These results cast doubt on the robustness of earlier findings on performance changes and suggest that they may be sensitive to the introduction of additional explanatory variables. Overall, these results provide little support for the signaling hypothesis. We next investigate whether the adoption of MDPs, especially ones with aggressive payout, reduces fund size and managerial compensation (typically based on NAV). The signaling hypothesis does not, in general, predict a decrease in fund size. The reason is that funds that adopt MDPs should, on average, be those that expect a strong NAV performance. The agency hypothesis, on the other hand, is more consistent with a reduction in fund size and managerial compensation. Our results indicate that MDPs, especially when accompanied by an aggressive payout target, can have asignificantly negative impact on fund size and, in turn, on managerial compensation. Finally, we study the role of institutional shareholders in adopting MDPs and the nature of shareholding around the adoption. The signaling hypothesis is based on the idea of funds voluntarily adopting certain payout policies as a way to distinguish themselves from other funds. Yet, it appears that in many cases shareholder activists force funds to adopt MDPs, which is more consistent 3

6 with the agency hypothesis. 5 Our analysis of the 13-D filings with the SEC at the time of MDP adoptions supports the notion that outside shareholders play an active role especially in the cases in which a more aggressive payout policy is adopted. We discuss a specific caseofinterventionby an institution to illustrate the potential profitability of such activism. To investigate aggregate patterns, we use an event study approach and find that institutional investors, as a whole, tend to hold significantly more shares of aggressive-mdp funds, relative to comparable non-mdp funds, in the quarters prior to policy implementation. Following policy implementation, institutional investors reduce their (aggregate) holdings as the discount is eliminated or turns into a premium. Further analysis reveals that institutions actively involved in trading around the MDP adoptions are mainly value-style investors. Our research is related to two strands of the literature. First, our paper contributes to the existing research on closed-end discounts. 6 The previous literature has focused on various explanations: tax liabilities (Malkiel 1977), illiquidity of asset portfolios (Malkiel 1977, Lee et al. 1990, Cherkes, Sagi and Stanton 2009), signaling and agency costs (Malkiel 1977, Thompson 1978, Lee et al and 1991, Barclay et al. 1993, Pontiff 1995, Chay and Trzcinka 1999, Dimson and Minio-Kozerski 1999b, Ross 2002, Johnson et al. 2006, Berk and Stanton 2007, Cherkes, Sagi and Stanton 2009), and costly arbitrage (De Long et al. 1990, Pontiff 1996 and 1997, Dimson and Minio-Kozerski 1999b, Spiegel 1999, Gemmill and Thomas 2002). The tax liabilities, asset illiquidity, and agency costs provide rational explanations for the existence of CEF discount, while the costly arbitrage approach is built on mispricing due to either behavioral factors or fundamental shocks (see Spiegel 1999 for a model incorporating random shocks on corporate productivity). The paper closest to ours is Johnson et al. (2006) that proposes a signaling explanation for the impact of MDPs on fund discounts. In addition to signaling, we propose and investigate an agency cost explanation for the adoption and impact of MDPs. The argument that fund discounts might be driven by agency costs and managerial fees has been made in papers such as Ross (2002), Berk and Stanton (2007), and Cherkes, Sagi and Stanton (2009). Our paper shows that the empirical evidence is more consistent with the agency cost hypothesis than with signaling. Second, our paper is related to the extensive literature on the role of activist shareholders, specifically institutional shareholders in affecting corporate policies. Several authors have argued that the involvement of large shareholders in monitoring or control activities has the potential 5 Such shareholder pressure was evident, for instance, in the case of Zweig Total Return Fund (ZTR) where the board discontinued the 10% MDP. Shareholders disagreed, and the issue came to a head when a dissident shareholder group issued its own proxy statement for the ZTR 2004 annual meeting, proposing insurgent nominees for director posts and adding a proposal for reinstituting the 10% annual payout. The dissidents were unsuccessful both in capturing board seats and in gaining enough votes to pass their payout policy proposal. However, the ZTR board subsequently reinstated the 10% MDP. 6 See Dimson and Minio-Kozerski (1999a) for an extensive literature review. 4

7 to limit agency problems (e.g., Shleifer and Vishny 1986, Huddart 1993, Maug 1998, etc.). The empirical findings on institutional activism are mixed in terms of its importance for firm governance and are well summarized in survey papers such as Gillan and Starks (1998) and Karpoff (2001). In the closed-end literature, Bradley et al. (2008) focus on the role of shareholder activists in open ending CEFs. Our paper highlights the role of activist institutional investors in implementing fund payout policies and in reducing or eliminating fund discounts while making sizable profits. The rest of the paper is organized as follows. In Section 2, we develop alternative hypotheses and empirical predictions. Section 3 describes the data sources and summary statistics. We present the empirical results in Sections 4 and 5, and conclude in Section 6. 2 Empirical Predictions In the introduction we outlined the signaling and agency cost hypotheses that we explore to understand the adoption and the impact of MDPs. We discuss below the main empirical predictions of the hypotheses that can be used to test them. The signaling hypothesis proposed in the literature (Johnson et al. 2006) relies on the informal notion that a commitment to pay higher dividends can be a signal of the fund manager s ability. The argument is that higher ability managers are in a better position to commit to a high dividend payout since they expect to generate greater returns. Hence, funds adopting MDPs would be expected to experience a reduction in discount and better NAV performance. In developing the testable predictions of the signaling hypothesis, we follow the hypothesis as proposed in the literature. However, it needs to be pointed out that the basis for the signaling argument is actually quite tenuous since MDPs are expected to impose a greater loss of fees on higher ability managers. Managerial fees are usually based on NAV and higher ability fund managers face a greater loss of fees because: (i) MDPs cause a greater dollar reduction in fees since their funds are expected to grow faster; and (ii) depriving higher ability managers of funds is more costly in terms of future NAV. 7 In developing the agency cost hypothesis and its empirical implications, we recognize that, de- 7 The authors have sought to develop a signaling model in which MDP can serve as a signal of managerial ability. In the model the manager benefits from an improvement in the fund s discount. The benefits of signaling may arise because, for instance, a decrease in discount make it less likely that the fund is opened or the manager is replaced and may be similar for higher and lower ability managers. On the other hand, the signaling cost the manager faces is a reduction in fees, stemming from a reduction in the fund s NAV over time. These costs are larger for higher ability managers for the reasons mentioned and, in general, a signaling equilibrium with MDPs will not exist. There are, however, some extreme assumptions under which a signaling equilibrium might exist: for instance when the higher ability managers, in addition to expecting to generate higher returns, also face greater diseconomies of scale. The reason the diseconomies of scale come into play is because a reduction in asset size is then less costly on the margin for higher ability managers. 5

8 pending on the nature of the agency problem and relation between fund size and performance, there maybesomewhatdifferent ways in which MDPs could impact fund performance and discounts. Managers, with fees linked to NAV may, in general, have an incentive to increase the size of their fund beyond what can be profitably managed and is value maximizing for shareholders. Hence, it is possible that the adoption of MDPs, especially when accompanied with a high payout target, could result in reducing fund size (or at least moderate its rate of growth) leading to an improvement in both fund performance and discounts. In somewhat different versions of the agency problem, fund managers are compensated more than what they deserve, but reducing fund size may not necessarily lead to a significant improvement in NAV performance. Along these lines, formal models of fund discounts, such as in Berk and Stanton (2007) and Cherkes, Sagi and Stanton (2009), make the argument that the source of discount is the present value of rent extracted by fund managers in excess of value-added. In this case, it is shown by Cherkes, Sagi and Wang (2009) that the adoption of MDPs can induce a direct wealth transfer from managers to shareholders. By reducing the present value of managerial claim on future fund assets, there is a reduction in fund discount though there may not be any discernible improvement in the fund s NAV performance. Our first set of predictions concern the relation between the level of payout promised by MDP funds and their NAV performance and discounts. Both hypotheses would predict that, in general, more aggressive payout policies will be associated with lower discounts. However, the hypotheses differ in terms of the relation between the level of payout promised by MDP funds and their NAV performance. According to the signaling hypothesis, fund managers adopting more aggressive payout polices are sending a stronger and potentially more costly signal. Hence, a more aggressive payout should be associated with a stronger performance and a greater reduction in fund discount if the MDPs are, indeed, serving a signaling function. The agency cost hypothesis, on the other hand, does not have clear empirical implications with regard to future NAV performance. We can therefore state: Prediction 1 MDP funds performance and discount: Both the signaling and agency cost hypotheses predict that, ceteris paribus, more aggressive MDPs will be associated with lower discounts. The signaling hypothesis predicts that more aggressive MDPs will be associated with stronger NAV performance. The agency cost hypothesis does not necessarily predict that more aggressive MDP funds will be associated with stronger NAV performance. We next turn to the issue of whether the adoption of MDPs, especially aggressive MDPs, will 6

9 tend to reduce fund size and, in turn, decrease fund manager compensation. We take the position that the signaling hypothesis should not predict a decrease in fund size. The reason is that funds that adopt MDPs should, in general, be those expected to have a stronger performance. We are basing this prediction on the general notion that we would not expect signaling to impose larger costs on the funds that choose to signal. The agency hypothesis, on the other hand, is more consistent with fund size deceasing, especially with aggressive MDPs. Hence, we can state: Prediction 2 MDP adoption and fund size: The agency hypothesis is consistent with a decline in fund assets, while the signaling hypothesis is not. A third set of predictions concern the role of outside shareholders, specifically institutions, in the adoption of MDPs. The signaling hypothesis asserts that the MDP decision is made by fund managers to signal their ability and, hence, is inconsistent with MDPs being adopted under pressure from outside investors. The signaling hypothesis also has no clear predictions in terms of changes in share ownership around the adoption of MDPs. The agency hypothesis, on the other hand, is consistent with large shareholders/institutions forcing funds to adopt MDPs. The opposition from managers and, hence, the need for pressure from outside investors is more likely to be observed when the adopted MDPs are more aggressive. To influence the adoption of MDPs, especially aggressive MDPs, we would expect large shareholders/institutions to build up their holdings in an effort to exert pressure. Institutions that buy shares of non-mdp funds at a discount and then successfully lobby for the adoption of MDPs are likely to make significant profits in the process. Since the adoption of an MDP tends to increase the fund s market value, it is quite possible that some institutional trades may be information driven. In our analysis, we also seek to identify and discuss specific cases of intervention by institutions. Hence, we can state: Prediction 3 MDP adoption and shareholder activists: The agency hypothesis is consistent with pressure from shareholder activists to adopt MDPs. Shareholder activists are more likely to be involved for aggressive MDPs. The signaling hypothesis is not consistent with such outside pressure. The agency hypothesis predicts a build-up of institutional shareholdings leading up to MDP adoption. Institutional shareholders are expected to make significant profits by purchasing non-mdp fund shares at discount and then forcing MDP adoption. We would like to note that the predictions of the agency cost hypothesis are not necessarily 7

10 incompatible with noise trader models, such as the one proposed in De Long et al. (1990) (see Wang, 2004). 8 Inthispaperwehaveconfined our hypotheses to those based on rational investors. We believe this is appropriate since the notion of sentiment risk and tests of its empirical validity are not well settled. 3 Data and Summary Statistics 3.1 Data The data sample used in the analysis consists of 236 closed-end equity funds that have at least one year of performance and discount data available over the 1990 to 2006 period. We include all funds with the following Wall Street Journal classifications: General Equity Funds, Convertible Funds, Preferred Equity Funds, Special Equity Funds and World Equity Funds. The monthly discount 9 data is from Lipper and from ETF Connect. 10 The CRSP Stock Database provides the monthly share prices, distribution amount, and number of shares outstanding. We then use discount and share price data to infer NAV for each fund on a monthly basis. We carefully review fund annual reports and proxy statements to identify MDP funds. A total of 88 funds in our sample had managed distribution policies in effect for at least part of the time over the period. For each MDP fund, we identify the payout target as well as the start and end dates for the policy. The payout target is usually expressed as a percentage of average NAV on an annual basis or else as a flat dollar amount of payout. For ease of comparison, we convert flat dollar payouts into percentage terms by normalizing the annualized dollar commitment by average NAV over the pertinent period. We identify the start and end dates for an MDP as the months in which the Board of Directors publicly adopts or terminates the policy. To examine the role of institutional investors in MDP adoptions, we obtain the quarterly institutional holding data for all CEFs in our sample from Thomson Financial Institutional 13(f). We notice that the reported institutional holdings for some CEFs in our sample are incomplete. We 8 Innoisetradermodels,thediscountreflects the fact that investors have relatively short horizons and are exposed to the effect of shifts in noise trader sentiment on the value of CEF shares when they trade. As is shown in Wang (2004), a commitment to an aggressive payout policy is expected to reduce the discount in such a model. This is because an aggressive dividend policy is expected to gradually reduce the size of the fund s NAV and, thereby, the exposure to future sentiment shocks. However, whatever the source of the discount, the decision to commit to a MDP is not one that fund managers would be expected to choose without external pressures, given the significantly negative impact on their compensation. 9 The discount/premium of a fund s share price relative to NAV is defined in percentage terms as (Share Price - NAV)/(NAV)*100. For ease of exposition, we follow the literature and sign the discount or premium (negative for discount / positive for premium) only when there is the possibility of confusion, as in the tables. 10 ETF Connect is a website ( maintained by Nuveen Investments that provides comprehensive information (e.g., share price and NAV history, premium/discount history, performance history, distribution history, etc.) for closed-end and index ETF funds. 8

11 supplement the missing data to the best we can based on the funds 13D/G filings with the SEC. Institutions are further classified into groups based on their legal types and investment styles. 11 There are altogether five legal types: Bank Trust (type=1), Insurance Company (type=2), Investment Company (type=3), Independent Investment Advisor (type=4), and Pension Funds, University and Foundation Endowments, and others (type=5). For empirical analysis, we lump types 3 and 4 together due to lack of precise distinctions. Regarding investment styles, we distinguish institutions based on investment horizons and investment types. As in Bushee (2001) and Bushee and Noe (2000), institutions are classified into three horizon types: Transient (TRA), Dedicated (DED), and Quasi-indexer (QIX). Transient investors have a short investment horizon and high portfolio turnovers, while the other types represent long term investors. Dedicated investors typically hold a concentrated portfolio characterized by large positions on a small set of firms. Quasi-indexers tend to invest in a highly diversified portfolio with low turnovers. According to Abarbanell, Bushee, and Raedy (2003), institutions are classified into four style types: Large Value (LVA), Large Growth (LGR), Small Value (SVA), and Small Growth (SGR). 3.2 Managed Distribution Policies The history of the MDP dates back to the 1970s. In our sample, Source Capital first implemented the policy in Table 1 documents the substantial increase in the number and proportion of CEFs with MDPs over our sample period. By the end of 2006, the number of MDP funds increased from 12 in 1990 to 75, almost 40% of all existing equity funds at the time. The median payout target, which was 10% from 1990 to 2002, dropped to 9% in 2003, and further decreased to less than 8% in Table 1 also reports the number of MDP funds that were aggressive (payout targets 10%) and less aggressive or moderate (payout targets 10%). The number of aggressive payout funds increased from 8 in 1990 to 16 in 2006, while the number of moderate payout funds expanded from 4 to 59 over the same period. 13 The impact of MDPs on fund discount has been well documented in Wang (2004) and Johnson et al. (2006). Here we provide further support for these findings by expanding the sample period to the end of 2006 and, thereby, including a much larger sample of MDP funds. Moreover, we present 11 All data on type classification is from Brian Bushee s website: 12 A total of 58 closed-end funds in our sample implemented MDPs in the post-2001 period, with a median payout target at 7.60%. Eight of them committed to a payout target of 10% or larger. 13 As noted earlier, while MDPs have a commitment aspect (see footnote 5), fund boards have the discretion to revoke the policy. In our sample we identified only five funds that stopped the payout policy during the sample period and continued to operate independently as closed-end funds. There were another three funds that terminated the payout policy but reinstated it later due to shareholder pressure.there are eight cases of the MDP funds leaving the sample at some point on account of liquidation, change of fund structure (e.g., open-ending), or merger with other funds. 9

12 new evidence that highlights the critical role of the MDP payout level on fund discount. Consistent with both hypotheses (Prediction 1), we find that higher payout levels are associated with lower discounts. In Figure 1, we plot the month-end median discount level separately for aggressive-mdp funds, moderate-mdp funds, and non-mdp funds. For non-mdp funds, the average month-end median discount from January 1990 to December 2006 is 10.19%. In contrast, the average discount for all MDP funds is only 0.86%. The difference is statistically significant at the 1% level. The difference is especially striking for aggressive-mdp funds that traded at an average premium of 2.32% over the sample period. The difference is more modest for moderate-mdp funds that exhibited an average discount of 6.70% over the same period. 14 In Table 2, we analyze the source of annual distribution for a subset of CEFs in our sample: General Equity Funds, Convertible Funds, and Preferred Equity Funds. We collect payout information from the annual reports filed with the SEC. For each fiscal year from 1995 to 2006, we report the median statistics for each of the three components of annual distribution: investment income, realized capital gains (sum of realized short-term and long-term capital gains), and return of capital. We also compute and report the median statistics for two yield measures: Distribution Yield and True Yield. Distribution Yield is defined as the total annual distribution (including the return of capital) per share normalized by the end-of-year share price. True Yield is defined as the total annual distribution per share net of the return of capital, normalized by the end-of-year share price. Hence, True Yield captures the part of payout that is funded by the return on assets generated by fund management. As shown in Panel A of Table 2, aggressive-mdp funds finance their annual distribution mainly through realized capital gains and return of capital. Over the 2001 to 2006 period, return of capital accounted for a significant portion of total annual distribution, as the stock market crash in 2001 and other events caused a steep loss of portfolio values. In particular, in 2002 and 2003, virtually all distributions (more than 95%) came from return of capital. Correspondingly, the median Distribution Yield was 13.52% (2002) and 8.49% (2003), while the median True Yield was only 0.45% in these years. For moderate-mdp and non-mdp funds, Panels B and C show that they relied mainly on investment income to fund annual distributions and that return of capital 14 We also investigate the cases in which CEFs adopt MDPs at fund inception. Among the 214 CEFs that introduced funds from 1985 to 2005, 43 adopted MDPs at inception. More than 85% (37 out of 43) of the inception-mdp funds adopted moderate payout targets ( 10%). To investigate the pattern of discount during the first three years of fund life, we identify for each inception-mdp fund a matched non-mdp fund with the same investment objective and inception year. The median series for the paired discount difference suggest that the inception-mdp funds typically trade at a slightly smaller discount when compared to the matched non-mdp funds for most of the post-inception period. Given that the inception-mdp group is dominated by the adoption of moderate payout targets, the evidence is consistent with the finding that moderate MDPs have some but limited effects in reducing fund discounts. The aboveresultsareavailableuponrequest. 10

13 was rare. We also examined the relation between dividend payout policies and the use of leverage by CEFs. Under the Investment Company Act of 1940, levered CEFs must maintain a minimum 2:1 asset to debt ratio. Dropping below the minimum coverage ratio will automatically trigger a SEC review and force funds to de-lever, possibly by liquidating assets and retiring some outstanding debt. Hence, at least in principle, the combination of an aggressive payout policy and the minimum coverage ratio requirement could signal fund management s confidence about higher expected cash flows. Our analysis indicates, however, that the extent of leverage employed by aggressive-mdp funds is generally quite small significantly lower than that by non-mdp and moderate-mdp funds. 15 Moreover, leverage in CEFs refers primarily to their use of preferred stock (reported as leverage by CEFs). 3.3 Institutional Holdings We investigate the institutional investor base for CEFs in general and examine whether different payout groups attract different types of institutional investors. We first examine whether the clienteles of CEFs vary with payout policies. In Panel A of Table 3, we report the total institutional holdings (as a percentage of shares outstanding), the average holdings per institution, and the number of institutional investors for non-mdp group, aggressive-mdp group, and moderate-mdp group respectively. For each quarter from 1990 to 2006, we first take the median holdings and then average over time for each payout group. We report the average of the median series in the table and conduct paired t-tests. Figure 2 plots the median series for total institutional holdings. The non-mdp group has the largest total institutional holdings (8.86%), followed by the moderate- MDP group (5.56%). The aggressive-mdp group, on the other hand, has the lowest institutional holdings (1.10%). The paired t-tests for between-group differences are all statistically significant at the 1% level. The average holdings per institution appears to be quite small (less than 1%) for all three groups, suggesting that institutions typically do not hold large blocks of shares in CEFs. 16 Panel B of Table 3 presents the median difference in institutional holdings between MDP funds and funds that were terminated during our sample period. This will be discussed in details in Section 5.2 where we examine the role of shareholder activism in MDP adoptions. 15 This weakens to an extent the case for signaling as the main motivation behind the adoption of aggressive MDPs. We collect the total liabilities and source of financing from the annual reports and NSAR filings with the SEC. The detailed statistics on leverage ratios across different payout groups are available upon request. 16 We also separately investigate the pattern of institutional ownership during the first three years of fund life for theinception-mdpfunds(seefootnote14). Boththeinception-MDP group and the matched non-mdp group exhibit low institutional ownership immediately following inception. This is consistent with the observation that CEFs are initially sold to retail investors. The institutional ownership increases over time. The trends diverge in the third year with the non-mdp matched funds attracting a larger presence of institutional ownership, consistent with the overall findings. These results are available upon request. 11

14 The evidence suggests a strong clientele difference between payout groups. The institutional investor base for MDP funds is much smaller than that for the non-mdp funds. Most strikingly, aggressive-mdp funds are simply dominated by retail investors, with a minimal presence of institutions. Why do institutional investors, especially those like dividend paying stocks, shed away from aggressive-mdp funds? We believe the answer lies in the composition of distribution made by these aggressive-mdp funds. One motivation for holding dividend paying stocks is to receive a steady income stream, especially during market downturns. However, as shown in Panel A of Table 2, the main source of distribution for aggressive-mdp funds during the recession was the return of capital. After taking out the return of capital from the total distribution, the true yields were actually minimal less than 1%. Hence, the aggressive-mdp funds were simply returning the capital base back to the investors. This could be unattractive to institutional investors since they would have to worry about reinvesting their capital in a very low interest rate environment. A second explanation is based on the notion that it is not the payout policy per se, but rather the discount that attracts some institutional holding. For instance, institutional investors, particularly those that are value-oriented, may regard discounted CEFs as a worthwhile investment. This may be because they expect the funds to be forced to open or to adopt payout policies (a process that they might actively participate in) that would diminish the discount. Institutional investors may also believe that the discounts are, in part, driven by negative sentiment among retail investors and which the institutions recognize and hope to benefit from by holding the discounted stock till the investor sentiment become more optimistic. Table 4 presents the statistics for total institutional holdings by types for non-mdp funds, aggressive-mdp funds, and moderate-mdp funds. The institutional holdings by Thomson legal types are provided in Panel A. For all three payout groups, the Investment Company and Independent Investment Advisor (Type-3 and -4) are the largest institution type. Consistent with the pattern observed in Table 3, the non-mdp group has the largest presence of Type-3 and -4 investors (4.46%), followed by the moderate-mdp group (3.65%) and the aggressive-mdp group (0.62%). For other legal types, the median holdings are typically low across all payout groups less than 1%. Panel B presents the distribution of institutional holdings by investment horizon. As indicated, for all three payout groups, Quasi-indexer is the largest type, followed by the Transient-type. In contrast, median holdings by the Dedicated-type are minimal. Panels C and D of Table 4 presents the institutional holdings by investment style. Based on Abarbanell, Bushee, and Raedy (2003), an institution is classified into Large (Small) style if it has the classification of LVA or LGR (SVA or SGR). Similarly, we define an institution as Value (Growth) if it has the classification of LVA or SVA (LGR or SGR). As shown in Panel C, Large- and Small-style investors are roughly evenly distributed across all payout groups. We observe in Panel 12

15 D that the Value-style institutions have significantly larger median holdings than the Growth-style institutions for both the non-mdp group and the aggressive-mdp groups. The average holdings across different types of institutions, as discussed above, obscures differences in terms of the trading by the different types of institutions. In a later section, we show that there are substantial differences in terms of trading by different types of institutions around MDP adoption. 4 Signaling In this section, we test the signaling hypothesis (Prediction 1) in alternative ways. Section 4.1 analyzes the discount and NAV performance of MDP funds using non-mdp funds as a control group. In Section 4.2 we examine the changes in discount and NAV performance before and after MDP adoption using a difference-in-difference approach, while Section 4.3 uses a panel regression approach and provides additional robustness tests. 4.1 Performance and Discount Relative to Peer Funds We now investigate whether MDP funds, especially aggressive-mdp funds, are also associated with asignificantly better performance as suggested by the signaling hypothesis (Prediction 1). We begin our analysis by comparing the four-factor adjusted NAV return of MDP funds to that of peer funds. For each MDP fund, we obtain the four-factor alpha by regressing monthly NAV returns on the Fama and French (1993) three factors and the Carhart (1997) momentum factor during the period in which the MDP is in effect. We identify peer funds as those that have the same Lipper investment objective but do not adopt an MDP over the sample period. The four-factor adjusted NAV alphas for the peer funds are then estimated over the same time period as for the MDP fund. We then compare the risk-adjusted performance of the MDP fund relative to the median performance of its peer funds. We require funds to have at least 24 months of NAV return data available to be included in the analysis. This reduces the number of MDP funds to 51. The results are reported in Table Panel A of Table 5 compares the four-factor alpha and discount for all MDP funds to peer funds. The median MDP fund outperforms its peer by 0.23% per month and the difference is statistically significant at the 1% level. In terms of discount, the median monthly discount for MDP funds is 5.15%, compared to 12.32% for the peer funds. The median difference in discount is 6.73% and is statistically significant at the 1% level. These findings are largely consistent with the evidence reported in Johnson et al. (2006) and interpreted as supportive of a signaling explanation. However, 17 As a robustness check, we conduct the same tests for the sub-period from 1993 to 2001 (the data period used in Johnson et al. 2006) and for the sub-sample of General Equity Funds. The results obtained are similar. 13

16 these explanations are less persuasive when the evidence is analyzed more closely. In Panels B and C, we examine whether relative performance and discounts differ across MDP funds with different payout targets. If signaling is the main factor driving the results in Panel A, then the effect of MDP on performance and discount should be more pronounced for funds with more aggressive payout targets. However, the results in Panel B show that for MDP funds with payout targets of 10% or higher, there is no significant improvement in performance relative to peer funds. The median difference in four-factor alpha is 0.04% per month, and is not statistically significant. In stark contrast to the relative performance results, these funds are associated with significantly lower discount levels. The median aggressive-mdp fund is priced at a premium of 0.80%, compared to a discount of 12.70% for the median peer fund. The 12.33% difference in discount levels is statistically significant at the 1% level. The results in Panel C show the reverse pattern for MDP funds with a payout target below 10%. These more moderate payout funds significantly outperform their peers in terms of four-factor alpha. The median outperformance is about 0.37% per month and statistically significant at the 1% level. However, the reduction in discounts experienced by these funds is far less pronounced than that experienced by the aggressive payout funds. The median fund with a moderate payout target trades at a discount of 8.91%, compared to a discount of 11.49% for the median peer fund. Overall, the above findings suggest a mismatch between performance improvement and discount reduction among MDP funds. Funds with aggressive payout targets show no significant improvement in risk-adjusted performance (relative to non-mdp funds) but are associated with substantially lower discounts. In contrast, funds with moderate payout targets show significant improvement in risk-adjusted performance though this result may be specification sensitive, as we show later but experience a much smaller reduction in the level of discount. To address the concern that some funds are dropped from the previous analysis due to the 24-month return data requirement, we compare the NAV performance between different payout groups using a portfolio approach. Since the majority of the moderate MDPs were adopted in the later half our sample period, we focus on the period from 1999 to For each month, we form four equally weighted portfolios based on the payout policy: non-mdp portfolio, MDP portfolio, aggressive-mdp portfolio, and moderate-mdp portfolio. Four-factor alphas are obtained for the monthly portfolio returns and the paired differences. As shown in Table 6, there is no evidence suggesting that the aggressive-mdp portfolio delivers better risk-adjusted return relative to the non-mdp and moderate-mdp portfolios. Overall, the lack of consistent outperformance of MDP funds (especially aggressive-mdp funds) over the non- MDP funds casts doubt on the signaling hypothesis. 14

17 The results so far do not control for fund heterogeneity in terms of discounts and performance prior to the adoption of MDPs. In the discussion that follows we will control for such heterogeneity by using a difference-in-difference approach with matched funds as well as a panel specification with time and style fixed-effects. The findings on MDP discounts, as we show below, are robust under various specifications, while the performance results are more specification sensitive. 4.2 Change in Performance and Discount: Before vs. After Policy Adoption In this section we follow Johnson et al. (2006) and use a matched algorithm to examine the changes in performance and discount for MDP funds during the three-year periods before vs. after policy implementation. For each MDP fund, we identify a matched fund that has the same Lipper investment objective, that experienced a similar level of average discount during the pre-policy period as the MDP fund, and that did not adopt a payout policy over the sample period. We require MDP funds to have a minimum of two years of return and discount data available both before and after the policy implementation to be included in the analysis. This reduces the sample of MDP funds to 29. The results are presented in Table 7. Panel A examines the change in performance and discount when all MDP funds are included. During the three-year period prior to implementing the MDP, the four-factor alpha for the median MDP fund is indistinguishable from that of the matched fund. In contrast, during the three-year period after policy implementation, the median MDP fund significantly outperforms the matched fund by 0.21% per month. Regarding the change in discount level relative to the matched funds, the median MDP fund experiences a reduction of 1.91% in discount following policy implementation. However, the reduction is not statistically significant. Wetheninvestigatewhetherthechangeinrelativeperformanceanddiscountdiffers between aggressive and moderate payout funds. As shown in Panel B, the aggressive payout funds have four-factor alphas that are indistinguishable from those of matched funds both before and after the policy implementation. The difference-in-difference test indicates that the change in four-factor alpha is not statistically significant either. Hence, there is no evidence suggesting that MDP funds with aggressive payout targets experience any improvement in performance as would be predicted by the signaling hypotheses. Despite the lack of performance improvement, these aggressive payout funds experience a meaningful reduction in discount relative to the matched funds. In Panel C, we report the results for MDP funds with moderate payout targets. As before, we observe a pattern that is quite different from the results for aggressive-mdp funds. The moderate payout funds on average significantly outperform the matched funds during both the pre- and post-mdp periods. The difference-in-difference analysis indicates that the median improvement in 15