Understanding the Closed-end Fund Puzzle from the Chinese Experience

Transcription

1 Understanding the Closed-end Fund Puzzle from the Chinese Experience Gongmeng Chen, Oliver Rui, and Yexiao Xu This version: April 2004 Abstract Although many characteristics of Chinese closed-end funds resemble those of the U.S. funds, investors and environment are different. For example, institutions such as insurance companies hold a substantial amount in closed-end funds, there are no capital gains taxes, and there is no private investment by the funds. These differences allow us find that the discounts are negatively related to liquidity, percentage of stock holdings, and past risk-adjusted performance. Discounts are also positively influenced by the R 2 from a market model applied to the underlying asset returns, managerial ownership, and the size of the fund. We further show that investor overconfidence and liquidity are major factors that affect the dynamics of discounts. Chen is at the Department of Accountancy, The Hong Kong Polytechnic University, Rui is with the Faculty of Business Administration of The Chinese University of Hong Kong, and Xu is at the School of Management, The University of Texas at Dallas. We are grateful to Burton G. Malkiel, Jeff Pontiff, Ravi Ravichandran, Z. Jay Wang, and to the seminar participants in the 2002 European Financial Management Association Conference, the 2002 Financial Management Association Conference, the tenth Asia Pacific Finance Association Annual Conference, and the 2002 APFA/PACAP/FMA Finance Conference for their helpful comments. The work described in this paper was fully supported by a grant from the Hong Kong Polytechnic University (Project No. A-PC 27). The address of the corresponding author is: Yexiao Xu, School of Management; The University of Texas at Dallas; Richardson, TX yexiaoxu@apache.utdallas.edu 1

2 Understanding the Closed-end Fund Puzzle from the Chinese Experience Abstract Although many characteristics of Chinese closed-end funds resemble those of the U.S. funds, investors and environment are different. For example, institutions such as insurance companies hold a substantial amount in closed-end funds, there are no capital gains taxes, and there is no private investment by the funds. These differences allow us find that the discounts are negatively related to liquidity, percentage of stock holdings, and past risk-adjusted performance. Discounts are also positively influenced by the R 2 from a market model applied to the underlying asset returns, managerial ownership, and the size of the fund. We further show that investor overconfidence and liquidity are major factors that affect the dynamics of discounts. 2

3 Introduction An enduring puzzle in finance is the so called closed-end fund puzzle. We often observe that shares of closed-end funds sell at prices below the net asset value (NAV) of the underlying portfolio of securities. Although the first closed-end fund, the New York Stock Trust, was offered to the public in 1889, this pricing discrepancy was not formally documented until Pratt (1966). Since then, we have learned from empirical study that the useful factors explaining discounts on closed-end funds include: unrealized capital gains (Pratt, 1966; Vives, 1975; Malkiel, 1977; and Mendelson, 1978), portfolio turnover (Pratt, 1966; and Boudreaux, 1973), distribution policy (Malkiel, 1977; Mendelson, 1978; and Thompson, 1978), private investment by closedend funds (Malkiel, 1977; and Anderson and Born, 1987), and investor sentiment or market conditions (Zweig, 1973; Malkiel, 1977; De Long, Shleifer, Summers, and Waldmann, 1990; and Brauer, 1993). Other factors derived from theoretical models include the open-ending option proposed by Brauer (1988), the tax-timing option of Brickley, Manaster, and Schallheim (1991) and Kim (1994), and the stochastic turnover risk that was modeled by Xu (2000). The dynamics of closed-end fund discounts are also interesting. Various empirical studies (for example, Thompson, 1978; Hardouvlis, La Porta, and Wizman, 1993; and Pontiff, 1995) show that funds with positive premiums provide negative abnormal future returns, but funds with discounts earn positive abnormal future returns. These studies suggest that no single explanation accounts for all of the discounts. Many factors matter. In fact, currently, we can at best explain only fifty percent of the discounts on closed-end funds using rational factors (see Malkiel, 1995), so it is possible that other, unknown, factors are contributing to the discount issue. In other words, we are still unable to conclude whether the remaining unexplained discounts are due to market inefficiency. Most research into closed-end funds has concentrated on funds traded in mature capital markets, such as the U.S. and U.K., including country funds that hold international stocks. Given the divergent conclusions of studies in this field, we might obtain further insights through investigating an alternative set of financial markets, in particular a set of emerging markets. 1

4 Emerging capital markets present an unique investment environment with special institutional details and different groups of investors with distinctive social and culture backgrounds. Such markets have low correlations with more developed markets. Hence, any possible data-snooping biases resulting from focusing only on developed markets are lessened. Although the majority of closed-end funds in the U.S. and U.K. are country funds, studies focused on these funds may not offer the same premise as those actually traded in emerging markets. There are two reasons. First, Cooper and Kaplanis (1994) provide evidence of home bias in investor portfolios, suggesting that the discounts could be due to market segmentation. Second, the majority of investors in these funds are from the U.S. or U.K., and the funds are operated under the same regulatory environment. The Chinese closed-end fund market is both ideal and unique in terms of institutional structure and investors behavior. As an emerging capital market, the Chinese stock markets are much less efficient than those in mature markets. If the majority of investors are individual investors, and if a factor such as investor sentiment is important, then this factor should influence the level of discounts for Chinese closed-end funds more than would be the case for the U.S. funds. Over 95% of closed-end fund investors in the U.S. and U.K. are individual investors. Individual investors were also the majority in China until February 25, 2000, when the authorities began to allow insurance companies and pension management companies to hold closed-end funds. Now, the majority of investors in the Chinese closed-end fund markets are institutional investors. This structural change may shift the importance of the sentiment risk for the Chinese closed-end funds. There are also significant differences in the tax environment in China. Currently, the Chinese government does not levy capital gains taxes. Thus, factors found in U.S. data that are related to capital gains taxes and the way in which the gains are realized and distributed are inapplicable in the Chinese case. Moreover, all of the closed-end funds in China are invested in the publicly traded domestic securities. Since these securities are actively traded in China, the valuation of the underlying holdings is not an issue. These unique differences suggest the usefulness and the importance of examining the Chinese closed-end funds. The first closed-end fund in China was sold to the public in April Since then, the industry has grown steadily. Using the unique data set on the Chinese closed-end funds, we document some of the stylized facts about Chinese closed-end funds. We note that these funds share many characteristics with those in the U.S. market. For example: there are 2

5 substantial and persistent discounts for majority funds; most closed-end funds enjoy high premiums at their IPO stage, which disappear after six months; and discounts can predict future fund returns. However, there are some critical differences in Chinese institutional structure, investors behavior, and investment environment. At the same time, there are factors that are more universal, for example, the investor sentiment risk, the imperfect arbitrage opportunity, the stochastic turnover risk, and the information asymmetric. By comparing the characteristics found in the Chinese closed-end fund market with those of the mature capital markets, we may better understand the behavior of closed-end funds. Our empirical results suggest that the closed-end fund discounts are negatively related to fund s liquidity, percentage of stock holdings, and past risk-adjusted performance. Discounts are positively influenced by the R 2 from a market model applied to the underlying asset returns, managerial ownership, and size of the fund. We also show that discounts fluctuate with the trading volume and the returns of a small-size portfolio over time. Due to the unique features of the Chinese market, we believe that investor overconfidence (not investor sentiment) and liquidity are major factors that affect the dynamics of premiums or discounts in the Chinese closed-end fund market. However, with the increasing presence of institutional investors in the Chinese closed-end fund market, the overconfidence effect is disappearing. Section 1 presents a brief history of Chinese investment companies. In Section 2 we discuss our unique data set and the hypotheses that we test. In Section 3 we establish some of the stylized facts about Chinese closed-end funds. Section 4 presents our empirical tests of various hypotheses. Section 5 studies the time series behavior of discounts and the issue of predictability. Section 6 concludes the paper. 3

6 1 The Development of Chinese Investment Companies The first investment fund in China, the Nanshan Venture Capital Fund, was established in November This marked the beginning of the fund industry in mainland China. At the beginning, most of the funds were established with the approval of the local government or the People s Bank of China. Because the funds were intended to attract capital, their investment covered a wide spectrum, from private equities to real estate, and from government bonds to traded stocks. They were named old funds as opposed to the relatively standardized investment funds. These old funds expanded rapidly. By the end of 1997, there were 75 old funds with more than 5.8 billion RMB in book value and 10 billion RMB in market value. However, many aspects of these old funds were not standardized, including fund initiation, fund operations, information disclosure, supervision, and regulation. Because there were so many problems in their daily operation, there has been a virtual halt in the offering of new funds since On November 14, 1997, the Securities Committee of the State Council (which was later merged into the China Securities Regulatory Commission, CSRC) issued its Interim Regulations on Securities Investment Funds (IRSIF). The CSRC later promulgated detailed rules, elaborating on fund initiation, capital raising and trading, fund trustees and managers, the rights and obligations of fund holders, fund investment operations, and supervision and management. As stipulated by the IRSIF, the percentage of bond and equity investments made by a fund cannot fall below 80 percent of the fund s total asset value; the total stock value of one listed company cannot exceed 10 percent of the fund s NAV; and the percentage of investments in national bonds cannot fall below 20 percent of the fund s NAV. Within three months after the approval of a new fund, the fund must raise 200 million RMB, an amount that must exceed 80 percent of the target size of the fund. These rules have had a major impact on the investment behavior of the investment companies. In March 1998, two new funds founded in line with the IRSIF Fund Kai Yuan and Fund Jintai, were issued nationwide, and publicly listed in April (see Table 1). Since the new funds were only allowed to invest in publicly traded stocks and bonds in the Chinese security markets, they were called securities investment funds (hereafter referred to as the closed-end funds). From the regulatory perspective, such funds are established for two main purposes. One is to exploit the advantages of expert management and to provide individual 4

7 investors with a good investment tool. The other is to nurture institutional investors and to promote the steady, healthy development of the security markets. Insert Table 1 Approximately Here To support the healthy development of those closed-end funds, the Chinese government has issued a series of preferential policies. On October 11, 1998, the CSRC promulgated the Notice on the Distribution of New Issues to the Securities Investment Funds. The Notice gave the funds preferential rights in the distribution of new issues. Under the original rules, shares of the new issues distributed to a fund could not be traded until two months after the IPO. However, this rule was modified on November 11, 1999, allowing a fund to trade half of the allocation on the day of the IPO, while holding the rest for six months. In May 2000, the CSRC abolished this preferential policy. Every fund can now set up its stock accounts just like any other investor. Funds can participate in issuing new shares, new share distributions, and in booking new share distributions. There is no limit on how many shares can be allocated to each investment fund. After years of development, the number of funds was growing and the old funds were gradually being standardized and transformed into new funds. The restructuring of the old funds began in the latter half of 1999, and nearly all the new funds that went public in the firsthalfof2000wereconversionsoftheoldfunds. The motivation behind the restructuring of the old funds was to increase the capital and to revive the operations of the funds by clarifying the assets of several old funds. These restructured funds are operated under the same methods as the newly issued funds. By the year 2001, there were 48 publicly traded closed-end funds in China under the management of 14 fund management companies, with a total billion fund units and billion RMB in net asset value. Among them, 27 were formed by reconstructing old funds. There were three open-end funds with billion fund units and billion in net asset value. In total, there were 51 funds in China, with billion fund units and billion RMB in net asset value. These funds have become a very important force in the Chinese security markets. 5

8 2 Data and Stylized Facts The discount phenomenon on most closed-end funds is the predominant issue in the closedend fund literature. As a first step in studying Chinese closed-end funds, we document some of the stylized facts here. We begin with a discussion of our data source. 2.1 Data We obtain data on Chinese closed-end funds from the only two stock exchanges in China, the Shanghai Stock Exchange and the Shenzhen Stock Exchange. This is a quality data set, since it was directly provided by the two stock exchanges. The data set starts from the inception of the first two closed-end funds in April 1998 and continues to the end of It is a weekly data set of all the funds that have ever existed in China. Out of a total of 48 funds, we excluded 12 funds that offered their IPOs after September 2001 because of insufficient data. Although these funds are traded daily, information on their net asset value (NAV) is available only weekly. As in the U.S., the NAVs are published on either the Wednesday or Friday edition of major Chinese financial newspapers, such as Shenzhen Securities Times, China Securities, and other such publications. The information includes weekly closing prices, the NAVs of fund portfolios, total weekly returns including dividends, weekly trading volume in terms of number of shares traded in a week, percentage of holdings by the funds initiators, and the total market capitalization of each fund. Table 2 reports the summary statistics for those variables over each quarter. Since we know little about the validity of the CAPM in Chinese stock markets, we use a market model to decompose security returns into their market-related and idiosyncratic parts. Although the results are similar when we use the CAPM, the market model is conceptually far more robust. As a proxy for the market portfolio we use a value-weighted composite index portfolio of all the stocks traded on both Shanghai and Shenzhen stock exchanges. This index is available from the 2001 version of the China Stock Market & Accounting Research (CSMAR) database, which is the most reliable and widely used security 6

9 database in China 1. We compound the weekly index returns from the daily index returns. Insert Table 2 Approximately Here The number of funds gradually increases from two to 36 over the four-year sample period. Although relatively small, the total market capitalization of Chinese closed-end funds grew rapidly from 7.61 billion RMB (about $0.89 billion U.S. dollars) to billion RMB (about $6.8 billion U.S. dollars) over the four-year period. There are many ways to define trading volume. We use the ratio between the total number of shares traded in a week and the total number of shares outstanding. We find that the average weekly trading volumes fluctuate between 0.7% and 3%, excluding the starting quarter of the data set. These numbers are close to the trading volume for common stocks in the Chinese equity markets. Therefore, closed-end funds are actively traded securities in China. In contrast to U.S. closed-end funds, the percentage held by the initiators of the fund is relatively small (about 2%). The average share price is around 1.2 RMB, which is comparable to the IPO price of 1RMB. 2.2 Establishing Stylized Facts about Chinese Closed-end Funds To study the characteristics of Chinese closed-end funds, in Figure 1 we plot the average weekly closed-end fund premiums across funds. Because few funds existed in 1998 and most of those were in the IPO stage, we begin our plot in To alleviate the extreme premiums in the beginning, we exclude the first four weeks of data after the IPO. In the first half of 1999, the premiums are mostly positive. The premiums disappear and turn into discounts in the second half of The downward trend continues into the first quarter of 2000, at which point the trend reverses itself. Insert Figure 1 Approximately Here Although premiums or discounts tend to fluctuate over time, this apparent change in the trend indicates the policy change that allowed insurance companies and pension companies 1 The construction of this index is similar to the composite index of NYSE/AMEX/NASDAQ found in the CRSP tape. 7

10 to invest in closed-end funds. Therefore, the majority of investors before and after February 25, 2000 were different. In the second half of 2001, we see moderate premiums instead. We also note from Table 2 that in most of the quarters in our sample period, the aggregate returns from the NAV underperform the market index returns when the average premiums are negative. This finding indicates that investors lose interest in the closed-end funds when they perform poorly relative to the market. On average, Figure 1 shows that the same discount phenomenon exists in the Chinese closed-end fund market. Lee, Shleifer, and Thaler (1991), using U.S. data, show that in the initial public offerings, closed-end funds are issued at premiums of nearly 10%, but the premiums quickly turn into discounts within four months. If the same results are true for the Chinese capital markets, showing aggregate premiums (discounts) over time could be biased towards premiums when the number of newly issued funds increases over time. Therefore, we study the discount issue from the life cycle perspective, especially for newly established markets. Given the uniqueness of our data set, which includes both the IPO and post-ipo periods for each fund, we can also investigate the discount issue using the event study approach. In Figure 2 we plot the average weekly premiums across individual funds after their IPOs. 2 The pattern is surprisingly similar to that of the U.S. closed-end funds documented by Lee, Shleifer, and Thaler (1991). A 35% premium at the IPO for a typical fund disappears in 30 weeks. Therefore, we define the first 30 weeks as the IPO period. In the U.S., there might be a price supporting period during a fund s IPO by underwriters, which is similar to theipoofapublicfirm. There is no such price support for the Chinese closed-end funds. Therefore, the IPO period we define here is purely mechanical. In the post IPO stage, we observe substantial discounts. It takes about a year and a half for the average discount to stabilize at around 10%. This process seems to be longer than it is in the U.S. Consistent with the huge premium at the IPO, aggregate fund returns are abnormally high in the first two weeks. 3 The returns rapidly retreat to around zero. Insert Figure 2 Approximately Here 2 As a convention, Figure 2 shows premiums instead of discounts. Although we have 185 weeks of data, the last 40 weeks of data are not shown in the graph since we have fewer than five funds. 3 This is also consistent with findings in Chen, Firth, and Kim (2000) who studied 277 IPOs in China. They find that the mean market-adjusted initial return for the A-share IPOs is extremely high. The average across the two exchanges is 350%. 8

11 To investigate the dynamics of closed-end fund returns and discounts, we first compute the autocorrelation for an equally weighted index of premiums across all funds starting in In Figure 3 the first line suggests that premiums (discounts) are persistent, as they are in Figure 1, with an autocorrelation of 98.8% at the first lag. The correlation only drops to 95% at the sixth lag. Such persistence is much higher than that found in other studies. Perhaps the most interesting time series property is the negative (positive) correlation between premiums (discounts) and future returns. Using the equally weighted index of fund returns and the index of premiums, in Figure 3 we also plot such cross-correlations between current fund premiums δ t and future returns R t+i as a function of future date i. The crosscorrelations are substantial, with a magnitude of 17% at the firstlagand9% atthesixthlag. In contrast, fund return autocorrelations are relatively small, and fluctuate at around zero at different lags. We note that the autocorrelation is 11% at the first lag. The cross-correlation that we observe for Chinese closed-end funds cannot be explained by the popular dividend taxation effect proposed by Pontiff (1995), because there are no capital gains taxes in China. Insert Figure 3 Approximately Here The qualitative characteristics of Chinese closed-end funds resemble those of the U.S. and UK closed-end fund markets. We examine the summary statistics at the individual fund level. Because Figure 2 suggests that there are substantial differences in a fund s premiums during the IPO period compared to the post-ipo period, in Table 3 we study each subsample period separately. 4 Since we are dealing with very skewed random variables, we prefer to use median statistics rather than mean statistics. The reported distribution suggests that the majority of funds are sold at a premium in the IPO period, with an average premium of 10%. These premiums become discounts in the post-ipo stage, with a magnitude of 8%. In contrast to a positively skewed distribution for premiums, the negative premiums in the post-ipo stage have a negatively skewed distribution. That is, we are likely to observe a large premium in the IPO stage and a large discount in the post-ipo stage. At the same time, 4 For stationary purposes, we ignore the first four weeks of data in computing the summary statistics. Moreover, the reported average discount in the post-ipo period should not be the same as shown in Figure 2, since the number of weeks with available data is different across funds. 9

12 discounts in the IPO stage are more persistent (about 97% compared to 91%). Although the persistence of discounts seems to vary much less across funds in the post-ipo stage than in the IPO stage, the volatilities of the discounts over time are about the same in both the IPO and post-ipo stage (about 7.5%). Therefore, we observe large, stable discounts more ofteninthepost-ipoperiod. Insert Table 3 Approximately Here The average weekly fund return is negative (about 0.235%) in the IPO period and slightly positive (about 0.04%) in the post-ipo period. The average weekly returns from the NAVs of underlying assets during the IPO and the post-ipo periods are just the opposite (about 0.424% compared to 0.032%). This pattern seems to be inconsistent with the premium (discount) pattern over the two different periods, since we might expect to see a higher return for a closed-end fund than the return from its NAVs when the fund is traded at a premium. Our calculations suggest that the pattern indicates nothing more than a decline in premiums in the IPO period and a shrinking of discounts in the post-ipo period. We also note that in the IPO period, the underlying assets of closed-end funds perform well, with an average return of 0.424% relative to that in the post-ipo period. Since new funds come into the markets fairly evenly throughout our sample period (see Table 2), the average returns in the two sample periods are comparable. The difference in returns could not be due to different market conditions. Perhaps a high return in the IPO stage merely reflects the level of risk exposure. Therefore, we also estimate the average beta measure by fitting the market model in Table 3. These average betas show no significant change during the two different periods. Moreover, the risk-adjusted alpha measure also points to the same direction, with 0.22% and 0.09% in the IPO and post-ipo periods, respectively. If performance reflects effort, this performance difference suggests that managers work harder in the early stage of a fund. It could also partly due to the fact that prior to May 2000, many of the funds receive preferential treatment in obtaining IPO stocks. In any case, the difference in alphas is a preliminary indication that the performance of a fund may be an important cause of the discounts in the post-ipo stage. A puzzling observation that appears in Table 3 is that a fund itself is exposed to less market risk than are its underlying assets, regardless of whether it is in the IPO stage or in the post-ipo period. For example, the average betas are 0.55 for the fund returns and

13 for the underlying asset returns over the post-ipo period. This phenomenon can only happen when there are systematic factors unique to closed-end funds that are partially negatively correlated with the market factor. However, the differences are statistically insignificant, as Table 4 shows. In other words, the market risks are similar for both the underlying assets and for the fund as a whole. The average R 2 sfromfitting the market model are over 67% and 41% for the NAV returns and the fund returns, respectively. Since a fund must be exposed to at least the same amount of market risk as its underlying assets in general, the differences in R 2 ssuggest that the Chinese closed-end funds are subject to other risks in an important way. Insert Table 4 Approximately Here Although returns from the underlying assets vary substantially (about 2.46%) in the IPO stage, the fund returns fluctuatemuchmore(about3.26%). If we interpret the net asset value as the fundamental value of the closed-end fund, such a discrepancy in the volatilities is consistent with the so-called excess volatility phenomenon found in stock returns by Shiller (1989) and later confirmed by Pontiff (1995) for closed-end funds. Although both returns fluctuate less in the post-ipo period, Table 4 shows that the excess volatility phenomenon is much more severe than in the IPO period. Given similar market risks in both the fund return and its NAV as discussed above, the difference in total volatilities must mean that the idiosyncratic volatility relative to the market factor for a fund return is larger than for its underlying asset returns. Table 4 confirms this observation. Therefore, although persistent discounts can only be attributed to systematic risk factors, many idiosyncratic factors influence fund returns. In fact, some of the idiosyncratic volatilities could be due to unknown systematic factors other than the market factor. We also observe that on average, betas are well below one, since Chinese closed-end funds hold between one-quarter to onethird of their assets in government bonds. Table 3 shows that cross-correlations between current premium and future returns are mostly negative and substantial (about 26%) in the post-ipo period. However, the crosscorrelation varies substantially across funds and is much smaller (about 16%) during the IPO stage. Such a correlation is unlikely from a correlation between current premiums and future returns of the NAVs. The reason is that, given a large and persistent discount in the post-ipo stage, we expect to see a positive cross-correlation between current premium 11

14 and future returns from the NAVs. On the contrary, Table 3 shows that the correlation is negative in the post-ipo period. Despite the structural and institutional differences between the U.S. and Chinese capital markets, the qualitative characteristics of closed-end funds are similar, but with greater magnitude. Although inefficiency may play a vital role in causing the discounts on most Chinese closed-end funds, many important rational factors such as tax-related effects, the distributional policy on gains, and the liquidity of the underlying assets are irrelevant to understanding the discounts on Chinese funds. Therefore, studying the Chinese closed-end funds not only serves as a robust check for the known rational factors, but also provides opportunities in exploring other factors that may have existed in a different environment. To gain additional insights, we focus on the issue of post-ipo discounts, the issue explored in many other studies. 12

15 3 Hypotheses Since most funds are traded at the negative premium, we use the term discount instead of premium in the rest of the paper. The discount phenomenon is not unique to mature capital markets. In fact, the aggregate behavior of the closed-end funds that are traded in emerging markets, such as the Chinese markets, is similar to that of U.S. closed-end funds. Because we far from fully understand thecauseofdiscountsusingtheu.s.andu.k.data,weexaminethesameissuefroma different perspective, such as a different capital market with a unique structure and different investors. By doing so, we avoid the issue of data snooping, and we can reevaluate the known factors and consider additional factors. Attributinganythingthatwedonotfullyunderstand to irrationality is an easy resolution to the closed-end fund puzzle. It may be that irrationality is partly responsible for the discounts we observe. The important question is how important irrationality actually is. If rational factors can explain a greater portion of discounts in Chinese closed-end funds than in U.S. funds, irrationality or market inefficiency may not be a dominant factor in discounts, because mature capital markets tend to be more efficient. Using variance-ratio tests, Chen, Lee, and Rui (2001) reject the hypothesis that stock returns follow a random walk process in China. Therefore, should market inefficiency be an important factor, we would expect to need fewer explanations for the level of discounts or premiums for the Chinese closed-end funds relative to the U.S. closed-end funds. Many important known determinants of discounts offer few or no insights into discounts on Chinese closed-end funds, since the structure of the Chinese capital markets is so different from that of the U.S. For example, one of the most important factors influencing discounts is unrealized capital gains. Since unrealized capital gains impose tax liabilities on current fund holders even when they are not entitled to such capital gains, closed-end funds should sell at discounts relative to their net asset value when there are large unrealized capital gains. As noted, the Chinese government does not currently levy capital gains taxes. Any factors that are related to capital gains or to the way a fund distributes the gains does not apply to Chinese closed-end funds. In addition, the liquidity of the underlying assets should also have little impact on dis- 13

16 counts. Unlike the U.S. or U.K. closed-end funds that are heavily invested in foreign securities and private assets, there are no foreign investments or restricted securities in the portfolios of Chinese closed-end funds. If its investments are not held in cash, each fund is required to invest in either traded government bonds or traded equities. At the same time, since the majority of Chinese investors can only invest in Chinese securities, home bias is not an issue. One reason for investors to hold an actively managed fund is to have access to management skills and/or private information. Therefore, if the premium is the price that investors pay to gain access to the unique abilities and private information of managers and the discount compensates investors for poor management, then the premium or discount should be related to performance. Despite the fact that all the major assumptions underlying the CAPM model are unlikely to apply in China, we must still adjust for risk when measuring performance. Similar to computing the Jensen s alpha, we use the intercept (α) from a market model as a measure of the risk-adjusted performance of a fund. 5 If a fund manager possesses superior management skills, the returns from the net asset value should have a large alpha. Naturally, these funds will experience high demands, which will drive up prices due to the fixed supply and opaque portfolio holdings of these funds. In other words, Hypothesis 1 the discounts (premiums) are negatively (positively) related to a fund s riskadjusted performance of the net asset value. We note that an important assumption supporting this line of argument is that no fund portfolios can be perfectly replicated at all times, which is reasonable even in the U.S. Further, in addition to management skills, superior performance can also be tied to unique information. When the portfolio held by a fund is easy to arbitrage, the discount will be relatively small. We use a more powerful measure, R 2 from the same regression for the market model using returns from the NAVs. We can consider the R 2 measure as a relative measure of Pontiff s (1996) replication risk 6. The higher the R 2, the easier it is to use a market portfolio 5 Alternatively, one can also subtract the riskfree rate from such an intercept. However, since the weekly interest rate is very stable and small, we do not consider it here. 6 He measures the risk using residual variance. As Malkiel and Xu (1997) have found that residual variance and firm size are negatively correlated, such a measure could be subject to the size effect. 14

17 to replicate a fund portfolio. Therefore, discounts should be low. However, in the Chinese markets, the arbitrage opportunities are hard to realize in practice when there is a discount due to the lack of a short selling mechanism. Atthesametime,thereisasecondeffect. When a fund portfolio is very close to a market portfolio, investors will lose interest, because the fund portfolio is unlikely to offer much in the way of unique investment opportunities and management skills. Therefore, there will be selling pressure, which will increase discounts. This selling pressure is sustainable, since short sales are not allowed in China and there is no option market. This argument also relies on an implicit assumption of being able to hold fully diversified portfolios. Although individual investors might not have enough funds and there is no index fund during our sample period, we can mimic the market performance by holding 20 randomly selected stocks, as shown by Xu (2003). Institutional investors such as pension funds might be interested in funds with large R 2 s, since such investors are not allowed to directly participate in stock markets. We control for this effect by using the portfolio composition variable. Given this effect, institutional investors, just like individual investors, again will prefer funds with unique investment opportunities 7. A caveat is that high fund R 2 s could also be due to the possible holdings of stocks with large R 2 s. Morck, Yeung, and Yu (2000) suggest that high R 2 s for individual stocks in the emerging markets might be related to the degree of transparency or the quality of corporate governance. Investors are certainly interested in firms with good corporate governance, which usually have low R 2 s. This effect is difficult to separate from investment opportunities without examining the floating ratio of the underlying stocks (see Wang and Xu, 2003), for which we do not have information. In any case, since imperfect arbitrage is unlikely to materialize, we believe that the second effect is more important. Therefore, the R 2 s for the underlying asset returns should be positively correlated with the discounts of the funds. A more direct way to test this investment opportunity hypothesis is as proposed by Xu (2000), who examines the composition of the fund portfolio. A Chinese closed-end fund holds only domestic equities and government bonds. The more government bonds it holds, 7 If they prefer, pension funds can hold a portfolio of all the closed-end funds to achieve the market performance. 15

18 the less investment expertise it will offer to investors. This factor has become especially important since February 25, 2000, when insurance companies and pension companies were first given the permission to invest in closed-end funds. Before this date, these institutions were only allowed to hold government bonds. Since these institutional investors are still not allowed to invest directly in the equity market, they are probably low on stock holdings relative to the optimal portfolios between stocks and bonds. Therefore, a fund with smaller stockholdingsislesspreferabletocompletetheir optimal portfolios. As we note earlier, the portfolio composition variable can also serve as a control variable for the R 2 measure. Thus, we have the following joint hypothesis: Hypothesis 2 A closed-end fund discount is positively correlated with the R 2 s of a fund s underlying asset returns with respect to a market model and the relative size of government bond holdings Investor sentiment is a frequently tested factor in the literature. This factor may have played a large role in an immature capital market due to the speculative nature of that market and the short investment horizon of investors. Lee, Shleifer, and Thaler (1991) suggest that investor sentiment will likely affect small stocks and closed-end funds at the same time. If this is the case, there should be a negative relation between the returns from a small stock portfolio and changes in the discounts of closed-end funds. During periods when investors hold positive views or sentiments toward investment, especially when investors favor closed-end funds, discounts should narrow 8. At the same time, anecdotal evidence suggests that Chinese investors are aggressive and more enthusiastic about individual stocks when the markets are high. During stock market slumps, individual investors try to seek professional help by investing in closed-end funds. Thus, investor overconfidence might be more important in affecting the discounts of Chinese closed-end funds, since closed-end funds are held solely by domestic investors and there were no open-end mutual funds in China for our sample period. The overconfidence hypothesis predicts an opposite relationship between a size-portfolio s returns and changes in the discounts of funds. 9 8 Bodurtha, Kim and Lee (1995) tested the investor sentiment hypothesis using closed-end country funds (CECFs). They found that changes in the average discount on CECFs are negatively related to returns on the U.S. stock market, controlling for the returns on the foreign market and exchange rate movements. 9 It may be more accurate to use an overall stock market index instead. However, only relatively large 16

19 Hypothesis 3 The correlation between returns of a small size-portfolio and changes in the closed-end fund s discounts should be positive if the overconfidence effect dominates and negative if the sentiment risk is important. There are differences between sentiment and overconfidence. Investor sentiment refers to the phenomenon of noisy investors holding more (or less) risky assets than do their counterparts. Investor overconfidence results in an investor holding more stocks when market-wide returns are high, since investors incorrectly attribute the gains to their talents or to private information. Overconfidence has been discussed extensively in recent years including by Gervais and Odean (2001) and Kyle and Wang (1997). However, we associate overconfidence with investors beliefs in the value of their private information and investment talents compared to the private information and skills of professionals during different market conditions. Perhaps the most significant difference between U.S. and Chinese closed-end funds is in the type of investors. Over 95% of the investors in closed-end fund in the U.S. are retailoriented individual investors. By contrast, more than 50% of Chinese closed-end funds are now held by institutional investors, including insurance companies and pension management companies. If an institutional investor is more sophisticated than a typical individual, neither the sentiment risk nor the overconfidence effect should matter nowadays. This study provides auniquevenuetoexamtheissueusingdifferent sample periods. Another variable that may affect the discounts on closed-end funds is the liquidity of the fund itself. A liquid asset can be traded rapidly and at a low bid/ask cost. Therefore, if there is a liquidity premium, a more liquid fund will enjoy a higher premium than will an illiquid fund. This liquidity effect suggests the following hypothesis: Hypothesis 4 The trading volume should be negatively related to the closed-end fund discounts. Most current studies of U.S. markets emphasize the liquidity effect of the underlying assets of a closed-end fund. This perspective is irrelevant for Chinese closed-end funds, since these funds are only allowed to hold shares in publicly traded domestic companies. Trading volume and well-established companies are allowed to go public in China. Therefore, we use the same proxy to test both hypotheses since potential differences are small. 17

20 may also be highly correlated with fund size. Similar to large stocks, large funds are more actively traded than small funds. Therefore, we also control for the size effect. For several reasons the size variable itself offers explanatory power for the cross-sectional differences in fund discounts. First, large funds tend to hold more shares in large companies. Since there is a size effect for Chinese stocks as shown by Wang and Xu (2003), smaller stocks may offer better investment opportunities. Apart from the size effect, the majority of shares in a publicly traded company are owned by the state, which affects corporate governance. Size may also serve as a proxy for the efficient management of a company, since state ownership is highly correlated with the size variable. It is also possible that the performance of a large fund might be similar to a market portfolio. Thus, such a fund is less likely to offer unique investment opportunities. In any case, we should observe: Hypothesis 5 There is a positive relation between the discount of a closed-end fund and its market capitalization (size). The first hypothesis (performance hypothesis) has been rejected in most empirical studies, except for the study by Chay and Trzcinka (1999), who found that discounts are negatively related to the future alpha of underlying assets. The second hypothesis (investment opportunity hypothesis) is new to this study. Most recent studies on closed-end funds have focused on the investor sentiment risk. Instead, in hypothesis 3 we suggest that investor overconfidence might also play a role. However, both effects depend on the types of investors, which we test in this study. The liquidity hypothesis (hypothesis 4) has been studied extensively by Datar (2001). Here, we reexamine the issue from the Chinese market perspective. The last hypothesis (size effect) is new and may offer additional insight into the discount issue. In addition, we study the informational content of trading volumes and prices in predicting future discounts. We test these five hypotheses by using both a time series analysis and cross-sectional regressions. Although any empirical findings may also be consistent with other theories, the related evidences by themselves are important. 18

21 4 Understanding Discounts A Cross-sectional Perspective Due to the unique features of the Chinese capital market, many useful factors, including unrealized capital gains, income distribution policies, and illiquid assets cannot be applied. We face both challenges and opportunities to investigating new factors as formulated in the above five hypotheses. Similar to most empirical studies on U.S. and U.K. closed-end funds, we also focus on the post-ipo period by adding a fund to our sample 30 weeks after its IPO. Further, we require that at least 12 funds be available in the cross-sectional regression at any given point in time. Therefore, our sample actually starts on February 25, 2000, which coincides with the event of allowing institutional investors to invest in closed-end funds. To test the five hypotheses discussed in the previous section, we apply Fama and MacBeth s (1973) regression technique for each regression equation. The dependent variable is the current week s discount. The independent variables are the previous week s fund turnover (η t 1 ) computed by dividing the week (t 1) s trading volume by the fund s total outstanding shares, the previous week s stock to bond holding ratio (Composition t 1 ), the percentage of the fund initiator s ownership (OS t 1 ), the log of a fund s net asset value (ln(size t 1 )), and the intercept (α NAV,t 1 )and the coefficient of determination (RNAV,t 1 2 )fromthemarketmodelfitted to the previous 23 weeks of the NAV returns. Table 5 reports these cross-sectional regression results. Insert Table 5 Approximately Here Fund managers may possess unique information or management skills in choosing the underlying securities of their portfolios. Since the composition of the portfolio cannot be perfectly replicated, a low demand for under performed funds creates discounts if performance persists. Past returns do not seem to explain the cross-sectional difference in the U.S. closed-end fund discounts. Perhaps what is more relevant is the risk-adjusted measure (α) for the underlying asset returns noted in Hypothesis 1. Equation (5) in Table 5 shows that the relation is negative and statistically significant at the 1% level. Since we estimate those alphas at any given week by using the NAV return data for the previous 23 weeks, excluding the current week s 19

22 data, the regression is predictive. Chinese investors do seem to rely on past performance to pick a fund, as measured by alpha. However, such a relation does not necessarily imply that funds with successful past performance will continue to outperform in the long run. It is also useful to separate the imperfect arbitrage argument of Pontiff (1996) from the investment opportunity effect of Xu (2000). The imperfect arbitrage argument predicts a positive relation between discounts and residual variance. The investment opportunity argument predicts an opposite sign or a positive relation by using a more powerful measure RNAV,t 1, 2 as stated in Hypothesis 2. Equation (6) in Table 5 suggests that discounts are positively related to the RNAV,t 1 2 variable, which inclines us toward the investment opportunity argument. The relation is significant at the 1% level, with an average explanatory power of 14% for a typical individual cross-sectional regression at a given time 10. Such a positive relation continues to be strong after controlling for the size effect, as shown in equation (9). Moreover, the past-performance effect seems to disappear when we use the RNAV,t 1 2 variable in equation (10). This evidence suggests that investment opportunity plays a more important role than perceived performance in explaining the cross-sectional difference in discounts. In equation (2) of Table 5 we use the Composition t 1 measure to test the investment opportunity effect. The relation is negative and significant at a 1% level. The RNAV,t 1 2 measure does not seem to be a substitute for the Composition t 1 variable, as shown in equation (11). Both variables are now statistically very significant, with an average explanatory power of 18%. We tried to control for the potential size effect in equation (12). All variables continue to be verysignificant and have the correct signs. In general, investors prefer funds with a greater unique investment opportunity and a large stock position. This conclusion is especially true for institutional investors with large bond positions. Therefore, investment opportunities do seem to have an effect on cross-sectional differences in discounts. Although the liquidity issue for the underlying assets is much less important for Chinese closed-end funds than for U.S. or U.K. funds, the liquidity in the trading of a fund itself could affect the fund s discounts, as discussed in Hypothesis 4. To capture this effect, we use the relative weekly trading volume η, which is a ratio between weekly trading volume 10 We are aware of the problem of interpreting this average cross-sectional R 2 as summary statistics. This measure is only intended to provide an idea of the tightness of the relation. 20