“Hot Hands” in bond funds

https://doi.org/10.1016/j.jbankfin.2007.04.023Get rights and content

Abstract

We investigate persistence in the relative performance of 3549 bond mutual funds from 1990 to 2003. We show that bond funds that display strong (weak) performance over a past period continue to do so in future periods. The out-of-sample difference in risk-adjusted return between the top and bottom decile of funds ranked on past alpha exceeds 3.5 percent per year. We demonstrate that a strategy based on past fund returns earns an economically and statistically significant abnormal return, suggesting that bond fund investors can exploit the observed persistence. Our results are robust to a wide range of model specifications and bootstrapped test statistics.

Introduction

Despite the enormous size of the market for actively managed bond funds, surprisingly little is known about whether active portfolio management contributes to bond investment returns. A priori, we might expect that the value added by active bond management would be only marginal. The returns of a fixed-income portfolio are almost fully driven by nondiversifiable processes that we know are very hard to predict (see, e.g., Litterman and Scheinkman, 1991, Knez et al., 1994, Gultekin and Rogalski, 1995). These studies suggest that only a few factors account for bond returns.

There are also very few studies that provide empirical evidence to support the existence of skilled bond fund managers. For example, Blake et al. (1993) suggest that return spreads between actively managed bond portfolios can be explained either by differences in the maturity range or by differences in the risk premiums of the securities that are held. The absence of any predictability of risk-adjusted bond performance supports the oft-cited claim that none of the cross-sectional differences in bond fund returns are attributable to fund management skills.

If there is one variable that researchers can use to predict future bond fund performance, it is the fund’s expenses. Bond funds with relatively high expenses generally underperform funds with lower expenses (see, e.g., Blake et al., 1993, Detzler, 1999). Skeptism on managerial skill in the bond market combined with these empirical findings makes a strong case against active bond fund management. The investment implications seem clear: buy shares of bond index funds.

We demonstrate that this argument is not necessarily true. In this study, we show that we can predict future bond fund performance by using historical excess returns. By applying dynamic fund sorts in the tradition of Hendricks et al. (1993) on a large and survivorship-bias free bond fund sample, we show strong evidence of relative out-of-sample predictability. We find that after we control for multiple benchmark return sensitivities, deciles of bond funds with high historical alphas outperform deciles of funds with lowest alphas out-of-sample by more than 3.5 percent per year.

To investigate whether investors can exploit the observed persistence pattern to earn abnormal returns, we simulate an investment strategy by applying modern portfolio theory on past returns. Even after taking the sales load into account, we find that our simulated portfolio of funds strongly outperforms a strategy that invests in passive indexes by more than 1.79 percent per year. Our evidence that bond funds can deliver positive abnormal returns tells an important story: active bonds funds can have incremental economic value.

Since research on bond funds is scarce and not well developed, this paper fills several gaps in the literature. First, to our knowledge, our study is the first to analyze the full universe of more than 3500 bond funds in the CRSP survivorship-bias free mutual fund database over the period 1990–2003. This large sample helps us to overcome the small-sample problems that plague earlier studies on bond fund performance. Second, earlier bond fund studies use only a subset of all common approaches that were originally developed in research on equity funds to test for persistence. We show that these and other methods produce a consistent story in this study on performance persistence. Examples of persistence tests are the cross-sectional regression of current fund alphas on prior-period alphas, where the focus is on the significance of the regression’s slope coefficient (see Blake et al., 1993), and the allocation of funds to one of four cells in a (two-by-two) current-past performance contingency matrix, where persistence is proven when the frequency by which past winners (losers) repeat their performance exceeds a threshold probability (see, e.g., Kahn and Rudd, 1995). We complement prior studies by introducing variants of the methods used by Hendricks et al., 1993, Elton et al., 1996, Carhart, 1997, which enable us to investigate the economic significance of strategies based on short-run persistence in bond fund performance. In doing so, we provide new insights into long-running debates on the benefits of actively managed funds vis-à-vis passive portfolios. Although Hendricks et al. (1993) find that equity fund managers with “hot hands” in the past continue to outperform managers with “icy hands” in the near future, their top-performing fund portfolio does not outperform standard benchmark indexes. Equivalently, previous studies in the bond area suggest that bond index funds are a superior alternative compared to actively managed funds, once we take expenses into account. In contrast to earlier studies, we offer strong evidence of a “hot hands” phenomenon in the bond fund market that translates into strategies that yield both economically and statistically significant excess returns.

We also perform a bootstrap analysis to cover the possibility that our results are driven by distributional features of the data that could make tests of performance persistence prone to a Type I error. We find that this is not the case. We simulate persistence tests based on artificially generated data, in which we preserve non-normality features and intentionally impose zero alpha. By doing so, we can determine the distributions of the tests statistics when persistence is predetermined to be a chance result. Even the most extreme values for the simulated test statistics are not in the order of the ones we obtain from the actual bond fund data.

The paper is organized as follows. Section 2 discusses our methods in the empirical analysis. Section 3 describes the bond fund sample. Section 4 presents the empirical results. Sections 5 Alternative model specifications, 6 Bootstrap analysis compare the robustness of our results to a wide range of model specifications and bootstrapped test statistics. Section 7 concludes.

Section snippets

Performance measurement

Consistent with most studies that hunt for new performance evaluation models for bonds, we measure bond fund performance relative to the return predicted by a multi-index model:Ri,t-Rf,t=αi+j=1Kβj,i(Ij,t-Rf,t)+ϵi,t,where Ri,t is the total return of fund i, Rf,t is the risk-free rate at time t, αi is the average risk-adjusted performance of fund i, Ij,t is the return on index j at time t, βj,i is the sensitivity of the excess return of fund i to index j, K is the number of indexes we use, and ϵi

Data

To our knowledge, ours is the first study on bond fund performance that fully exploits the information content of the CRSP mutual fund database. In our study we use the largest sample of bond funds investigated to date. It does not suffer from survivorship-bias of the kind described in Brown et al., 1992, Brown and Goetzmann, 1995.

Performance persistence

Our first series of results involve the nonoverlapping cross-sectional regressions of future alpha on past alpha, which we estimate at the beginning of each year over 12-month horizons. We regress bond fund alphas over 1991 on alphas over 1990, alphas over 1992 on those over 1991, and so forth. On average, 1128 funds are available. The number of available funds ranges from 420 over 1990–1991 to 1745 over 2002–2003.

Panel A in Table 1 reports the Fama and MacBeth (1973) averages of the

Alternative model specifications

In this section, we examine whether the persistence we found in previous sections of the paper can be explained by more complex models that account for managers’ timing ability, funds’ time-varying risk exposures, sensitivities to changes in fundamental economic variables, and changes in the term structure that are not fully picked up by duration. Here, we ask whether these extra performance attribution variables eliminate the persistence in the performance of the portfolios we formed based on

Bootstrap analysis

Although the evidence of a hot-hands phenomenon in the bond mutual fund industry is compelling, the distributional assumptions that typically underlie tests of performance persistence might be too stringent. Recent studies show that fund returns do not follow the normality assumption inherent in most popular research designs. Violation of the normality assumption could induce a Type I error, in the sense that empirical tests reject evidence of no persistence when persistence patterns are

Conclusion

Despite the massive size of the bond retail industry, empirical evidence on the performance persistence of actively managed bond mutual funds is surprisingly scarce. In this study, we provide a rigorous investigation into bond fund performance by using the entire fund population in the United States over the period 1990–2003. The central question that underlies our research asks if active bond fund performance in the past persists in the future.

We find strong evidence of performance persistence

Acknowledgements

We thank two anonymous referees, Nick Bollen, Frank de Jong, Jenke ter Horst, Kees Koedijk, Marno Verbeek, Bas Werker, seminar participants at RSM Erasmus University, Maastricht University, and Tilburg University, participants of the Robeco Asset Management roundtable session, and participants of the 2006 FMA Europe Conference for valuable comments on the paper. We thank Sandra Sizer for excellent editorial assistance.

References (32)

  • Chen, Y., Ferson, W., Peters, H., 2005. The timing ability of fixed income mutual funds. Working paper Boston...
  • E.J. Elton et al.

    Fundamental economic variables, expected returns, and bond fund performance

    Journal of Finance

    (1995)
  • E.J. Elton et al.

    The persistence of risk-adjusted mutual fund performance

    Journal of Business

    (1996)
  • E.J. Elton et al.

    A first look at the accuracy of the CRSP mutual fund database and a comparison of the CRSP and morningstar mutual fund databases

    Journal of Finance

    (2001)
  • E.F. Fama et al.

    Risk, return, and equilibrium: Empirical tests

    Journal of Political Economy

    (1973)
  • W.E. Ferson et al.

    Measuring fund strategy and performance in changing economic conditions

    Journal of Finance

    (1996)
  • Cited by (57)

    • How skilful are US fixed-income fund managers?

      2021, International Review of Financial Analysis
    • The impact of financial and economic factors on Islamic mutual fund performance: Evidence from multiple fund categories

      2019, Journal of Multinational Financial Management
      Citation Excerpt :

      The five Fama and French (1993, 2015) significantly affect balanced and equity funds, specially from July to December, with the risk premium effect appearing most frequently throughout the year. The three factors of Huij and Derwall (2008) are relevant for explaining the performance of cash and fixed income funds for almost over the whole year. The lagged spread between the 10-year and the three-month Treasury bill rate affects the performance of all funds, for all model specifications.

    View all citing articles on Scopus
    View full text