Mutual fund performance persistence: still true?

Fortin, Rich ; Michelson, Stuart

PURPOSE

The purpose of this paper is to examine the performance persistence of a large sample of mutual funds over time. Specifically do mutual fund managers show positive (negative) performance year after year? Alternatively, is mutual fund performance from one year to the next basically a random event? It's been many years since mutual fund performance persistence has been examined. This paper will examine whether persistence is still valid in mutual fund investing.

MOTIVATION

A number of researchers have examined mutual fund performance persistence, but the results are still inconclusive. Grinblatt and Titman (1992) find that there is positive persistence in mutual fund performance. They find that part of the persistence is due to differences in fees and transaction costs across funds. They conclude that past performance does provide useful information for investors.

Hendricks, Patel, and Zeckhauser (1993) find that the relative performance of growth, no-load mutual funds persists in the short-term, with the strongest results for the one-year horizon. Poor performing funds show significantly worse performance over time, although the better performing funds don't show significant results.

Carhart (1997) shows that common factors and investment expenses almost totally explain persistence in equity mutual funds. He indicates that "hot hands" is explained by the one-year momentum effect of Jegadeesh and Titman (1993). Carhart agrees that the only significant persistence not explained by his common factors is the underperformance of the lowest performing mutual funds. His results do not support the existence of skilled mutual fund managers.

Bollen and Busse (2005) show results that differ somewhat from Carhart. They demonstrate positive short-term performance persistence, from quarter to quarter. But, as with Carhart, the positive performance persistence disappears for longer investment horizons. They conclude that after considering transaction costs and taxes, investors may generate superior returns through a naive buyand-hold strategy over following a performance chasing strategy.

Brown and Goetzmann (1995) find that funds in the bottom octile show significant negative persistence, while funds in the top octile show non-significant positive performance persistence. They show that poor performance holds over time, although positive performance is dependent on the time period studied. They hypothesize that the positive performance is due to specific macroeconomic factors over time.

Eser (2008) examines shortcomings in the persistence literature. He finds that much of Carhart's (1997) persistence is due to calendar-related distortions and the use of a short-term momentum factor model. After using a longer-term momentum factor model and masking calendar year-end noise, Eser finds that performance persistence seems to disappear.

Malkiel (1996) notes that over the past 25 years, about 70% of active equity managers have been outperformed by the S&P 500 Stock Index. Gruber (1996) and Bogle (1995) also note similar results. They argue that index funds allow investors to buy securities of many different types with minimal expense and significant tax savings. Bogle (1996) states that "the case for selecting an index fund is compelling due to indexing's inherent cost advantage." Malkiel (1995) concludes by stating that "most investors would be considerably better off by purchasing a low expense index fund than by trying to select an active fund manager who appears to possess a hot hand".

While the literature appeared to support performance persistence in the past, it seems the results are mixed. Our study is intended to extend the previous research by examining a larger sample of mutual funds over a more recent and longer time period. Our sample includes nine mutual fund classification categories over a ten-year investment horizon.

HYPOTHESIS

This study will test the hypothesis that actively managed mutual funds show significant performance persistence over our study period, 1996 through 2005. This analysis includes nine classes of mutual fund categories, including five categories of equity funds, three categories of bond funds, and one category of balanced funds.

DATA

The mutual fund data used in this study is from the January 2006 Morningstar Principia Pro Plus for Mutual Funds (1). This database contains historical information on over 20,000 mutual funds through December 31, 2005 year-end. Data and information are provided on investment objective, total return, income and capital gain distributions, annual expense ratios, fund size, load, and turnover.

This study groups the funds into nine broad investment categories: Aggressive Growth and Growth (AGG), Growth/Income and Equity/Income (GIEI), International Stock (IS), Balanced Funds (AAB), Corporate Bond (CB), Government Bond (GB), Municipal bond (MB), Small Company Equity (SCE), and Specialty Equity (SP) categories. The final sample contains 44,560 funds in the categories described above.

METHODOLOGY

The methodology employed to test the hypothesis of significant performance persistence in mutual fund returns involves two methodologies. We first categorize funds as a "winner" or "loser" each year. Winner/Loser (W/L) is determined by comparing each fund's return to the median return for that funds Morningstar category. If a fund's return is greater than or equal to the median, it is classified as a Winner. Funds lower than the median are classified as a Loser. On an annual and overall basis we tabulate the number of funds that are Winner/Winner, Winner/Loser, Loser/Winner, and Loser/Loser. Using this data we compute the nonparametric Odds-Ratio to determine the performance persistence of our sample for each fund category (see Brown and Goetzmann (1995)). Using the Odds-Ratio we compute the Z-statistic and accompanying P-value. (2). Additionally we compute the nonparametric Chi-Square statistic to determine the P-value as well. The second methodology used categorizes all funds in performance quintiles from year to year. If a fund is a top performing quintile, it is categorized as a 5 and a bottom performing quintile is categorized as a 1. We then pair the prior year quintile rating with the current year quintile rating. We use this to determine those funds that maintained performance (55, 44, 33, 22, and 11) versus those that did not show performance persistence (51, 42, 13, etc.) from one year to the next. Graphs are presented to portray the performance persistence results. All returns are computed on a before-tax and after-tax basis and results are presented separately for each.

RESULTS

Table 1 presents the summary statistics for our sample, including: total return, after-tax return, net assets, turnover, and expense ratio (3). The total return for our full sample is 7.74% and the highest total return is in the specialty equity (SP) category at 13.34%. The lowest total return is in the government bond category (GB) at 3.49%. The largest funds by net assets are GIEI funds and the smallest are municipal bond funds. Turnover for the full sample is 82.895%, the largest turnover is in the GB category at 180.41%, and the smallest turnover is in the MB category at 37.18%. The mean expense ratio overall is 1.19%, the highest expense ratio is 1.536% in the SP category, and the lowest expense ratio is 0.479% in the GIEI category.

Table 2, Panels A and B present the number and percent of funds that are equal to or above the median return (and after-tax return) (W) and funds that are below the median (L). The columns labeled LL, LW, etc., indicate the fund's performance from the prior year to the current year. For example, LL (WW) indicates a fund's performance was below (equal to or above) the median for the prior year and the current year. As one scans across the rows for LL, LW, WL, and WW in each of the categories, it appears that there is persistence in the LL and WW categories (the number and percentage is higher for LL and WW than for LW and WL). The last two columns of Tables 2 present the Chi-Square statistic and the P-value for each of the categories to test for a significant difference in the four performance categories. All P-values, except one are significant at the 0.001 level, indicating a significant difference between groups (LL, WW, LW, WL). The one category that doesn't show significance is the Government Bond category. The results are similar for after-tax returns, although the non-significant category changes to Corporate Bonds and Government Bonds becomes significant.

Table 3 presents the results for the non-parametric Odds-Ratio statistic. A significant P-value indicates performance persistence for that fund category. Reviewing the P-values, all fund categories are significant at the 0.001 level, except for GB funds (before-tax) and CB funds (after-tax) which reinforces the results of the Chi-Square test.

[TABLE 3 OMITTED]

Figure 1 graphically illustrates these results. Note that for all fund categories, except GB, the percent of funds that are in the LL and WW categories are much higher than the LW and WL categories, which is a strong indicator of persistence in fund returns.

Table 4, Panels A and B present the funds sorted by performance quintiles. If a fund is in a top performing quintile, it is categorized as a 5 and a bottom performing quintile is categorized as a 1. We then pair the prior year quintile rating with the current year quintile rating to determine those funds that maintained performance (55, 44, 33, 22, and 11) versus those that did not show performance persistence (51, 42, 13, etc.) from one year to the next. Number and percent of fund pairs in each quintile are presented for before-tax (Panel A) and after-tax (Panel B) returns. As one scans the results for the pairs, it appears that more funds (number and percentage) are in the persistence categories (11, 22, 33, 44, 55) and fewer funds are in the other categories. Supplementing the data in Table 4--Panels A and B, we computed the Chi-Square statistic to test for a significant difference between groups (persistence quintile pair categories). All categories, except two, show significance at the 0.001 level. GB were not significant for before-tax returns (P-value of 0.6924) and CB were not significant for after-tax returns (P-value of 0.4358). Refer to Figures 2 and 3 for a graphical representation of these results. Figure 2 graphs all fund categories across the 25 fund pair quintiles. One can see that the persistence quintile pairs have many more funds than the non-persistent pairs. Figure 3 presents a bar graph that shows the 25 quintile pairs for all funds overall. Once again, this graph demonstrates that the persistence quintile pairs have many more funds than the non-persistent pairs.

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

[FIGURE 3 OMITTED]

[TABLE 4 OMITTED]

CONCLUSION

In this paper we examine the performance persistence of a large sample of mutual funds over time. We test 44,560 mutual funds in nine equity and bond fund categories over the time period 1996 through 2005. We utilize the non-parametric Odds-Ratio and Chi-Square tests to examine significance in performance persistence. We find that there is significant performance persistence in mutual fund returns. This outcome is true for both the lowest performing and highest performing mutual funds. The tests demonstrate this result for all fund categories, except government bond and corporate bond funds. These results are very important to individual investors when selecting mutual funds. Investors should be cognizant of previous returns for any funds under consideration. If a fund performed poorly during the past year, it is likely the fund will continue to perform poorly in the next year. Likewise if a fund performed well during the past year, it is likely the fund will perform well during the next year. Note that persistence appears to exist for the best and worst performing fund categories. Therefore, an investor selecting funds in the middle performance categories is not likely to see the same persistence in returns.

As a caveat we understand that there is survivorship bias when performing mutual fund research. A fund must have survived for the full ten-year period to be included in our study, so funds that under-performed and subsequently closed to investors would not be included in this study. This would actually bias against finding significant performance persistence for the worst performing quintile of funds. Additionally our sample period is a ten-year period from 1996 to 2005. We understand that this is a limited period and results could vary for other time periods.

REFERENCES

Bogle, J., April, 1995, "The Triumph of Indexing," The Vanguard Group, pp. 1-45.

Bogle, J., May 8, 1996, "Be Not the First ... Nor Yet the Last," The Vanguard Group, from a speech presented at the 1996 AIMR Annual Conference in Atlanta, Georgia.

Bollen, N. & J. Busse, 2005, Short-Term Persistence in Mutual Fund Performance, The Review of Financial Studies, 18(2), 569-597.

Brown, S., W. Goetzmann, 1995, "Performance Persistence," The Journal of Finance, 50(2), p. 679-698.

Carhart, M., 1997, "On Persistence in Mutual Fund Performance," The Journal of Finance" 52(1), p. 57-82.

Eser, Z., 2008, "Persistence in Mutual Fund Performance: 2.0," SSRN working paper.

Grinblatt, M. & S. Titman, 1992, The persistence of mutual fund performance, The Journal of Finance, 47(5), 19771984.

Gruber, M.J., 1996, Another Puzzle: The Growth in Actively Managed Mutual Funds, The Journal of Finance, 51(3),, 783-810.

Hendricks, D, J. Patel, and R. Zeckhauser, 1993, "Hot Hands in Mutual Funds: Short-Run Persistence of Relative Performance, 1974-1988," The Journal of Finance, 48(1), p. 93-130.

Malkiel, B., April 22, 1996, "Not So Random," Barron's, p. 55.

Malkiel, B., June, 1995, "Returns from Investing In Equity Mutual Funds 1971 to 1991," The Journal of Finance, Vol. 50, No. 2, pp. 549-572.

Mutual Funds OnDisk, Operations Manual, Morningstar Mutual Funds, Chicago, IL, January 2005.

Siegel, B. and D. Montgomery, 1995, "Stocks Bonds and Bills after Taxes and Inflation," Journal of Portfolio Management, Winter 1995, pp. 17-25.

ENDNOTES

(1) See References for version.

(2) The Odds-Ratio is computed using the number of funds in each category as follows: LN[(WW*LL) / (WL*LW)]. The Z-statistic is the Odds-Ratio divided by its standard error. The standard error is computed as follows: [sigma] = [square root of (1/WW) + (1/WL) + (1/LW) + (1/LL)].

(3) Since annual total returns (calculated assuming reinvestment of all dividends and capital-gain distributions) are provided by Morningstar, an important variable for individual investors is the after-tax total return. This calculation involved estimating the historical marginal tax rates on ordinary income and capital gains. This paper uses the marginal tax rates provided in Exhibit 1 of Siegel and Montgomery [Winter 1995]. Because tax rates are heterogeneous, they chose an arbitrary single taxpayer earning $75,000 in "earned" (noninvestment) income in 1989 dollars. This level of income was deflated (inflated) by the Consumer Price Index (CPI) for earlier (later) years. They argue that this investor would be typical of individuals with sizable investment portfolios subject to tax. Since our data starts in 1977, we use the Siegel and Montgomery marginal tax rates on ordinary income and capital gains from 1977 through the end of their study in 1993. For the years 1994 through 2005, we utilize tax code information on the ordinary income and capital gains rates and adjust earned income by the CPI for each year. After-tax returns for a given mutual fund in a given year are computed by adjusting the total return for the taxes that would have been paid on the dollar income and capital-gain distributions for that year. There is a slight upward bias in this after-tax return computation since Morningstar includes both short-term and long-term capital gains in its yearly dollar-per- share capital-gain figure. The short-term capital-gain distributions should be subject to the higher ordinary income tax rates, but it was not possible to make this adjustment. The differences between before- and after-tax returns presented in this article are thus slightly smaller than would actually be expected.

Rich Fortin, New Mexico State University

Stuart Michelson, Stetson University

Table 1
Summary Statistics

 After-Tax

 Total Total Net Expense
 Return Return Assets Turnover Ratio

Total N 44,560 44,560 35,212 41,197 32,859
Sample Mean 7.736 6.734 877.970 82.895 1.191
 Std 15.829 15.555 3454.020 115.551 0.704

AGG N 7,070 7,070 5,706 6,527 5,320
 Mean 10.290 8.963 1548.150 93.123 1.372
 Std 22.495 22.059 5021.700 103.639 0.865

GIB N 4,420 4,420 3,487 4,034 3,209
 Mean 9.942 8.560 2143.560 59.389 1.086
 Std 15.965 15.564 6867.510 47.650 0.479

IS N 3,530 3,530 2,927 3,319 2,710
 Mean 10.258 9.288 1088.170 74.838 1.628
 Std 24.544 24.414 3320.380 60.133 0.664

AAB N 3,120 3,120 2,508 2,845 2,259
 Mean 7.614 6.107 1062.500 89.298 1.294
 Std 12.303 12.037 3347.670 77.414 0.516

CB N 4,980 4,980 4,067 4,633 3,797
 Mean 5.688 3.849 705.227 144.384 0.968
 Std 7.058 6.987 2201.140 200.562 0.471

GB N 3,400 3,400 2,726 3,135 2,544
 Mean 5.006 3.493 409.166 180.410 0.999
 Std 4.051 3.875 1170.010 204.635 0.472

MB N 13,430 13,430 9,973 12,347 9,436
 Mean 4.784 4.738 243.241 37.181 1.009
 Std 3.979 3.979 707.623 42.703 0.416

SCE N 2,410 2,410 2,024 2,271 1,894
 Mean 12.076 10.686 655.515 87.244 1.415
 Std 23.160 22.861 1776.340 62.399 1.548

SCE N 2,200 2,200 1,794 2,086 1,690
 Mean 13.344 12.080 568.999 83.168 1.536
 Std 27.697 27.348 1349.960 88.773 0.570

Table 2--Panel A: Number and Percent of Funds Returns Equal to or
Above (W) and Below (L) the Median From Prior to Current Year for
Before Tax Returns

 Chi-Square
 LL LW WL WW Test P-Value

Total N 12466 9741 9745 12608
 Percent 27.98 21.86 21.87 28.29 701.663 0.0001

AGG N 2054 1472 1472 2072
 Percent 29.05 20.82 20.82 29.31 197.705 0.0001

GIEI N 1331 877 875 1337
 Percent 30.11 19.84 19.8 30.25 189.85 0.0001

IS N 1014 742 743 1031
 Percent 28.73 21.02 21.05 29.21 89.003 0.0001

AAB N 966 589 590 975
 Percent 30.96 18.88 18.91 31.25 186.156 0.0001

CB N 1315 1168 1168 1329
 Percent 26.41 23.45 23.45 26.69 19.128 0.0003

GB N 830 867 868 835
 Percent 24.41 25.5 25.53 24.56 10457 0.6924

MB N 3681 3004 3006 3739
 Percent 27.41 22.37 22.38 27.84 148.536 0.0001

SCE N 670 530 530 680
 Percent 27.8 21.99 21.99 28.22 34.979 0.0001

SP N 605 492 493 610
 Percent 27.5 22.36 22.41 27.73 24.0691 0.0001

Table 2--Panel B: Number and Percent of Funds Returns Equal to or
Above (W) and Below (L) the Median From Prior to Current Year for
After-Tax Returns

 Chi-Square
 LL LW WL WW Test P-Value

Total N 12341 9869 9866 12484
 Percent 27.7 22.15 22.14 28.02 582.389 0.0001

AGG N 2070 1456 1456 2088
 Percent 29.28 20.59 20.59 29.53 219.684 0.0001

GIEI N 1326 883 883 1328
 Percent 30 19.98 19.98 30.05 178.405 0.0001

IS N 1005 752 753 1020
 Percent 28.47 21.3 21.33 28.9 76.729 0.0001

AAB N 968 590 590 972
 Percent 31.03 18.91 18.91 31.15 185.139 0.0001

CB N 1265 1216 1217 1282
 Percent 25.4 24.42 24.44 25.74 2.726 0.4358

GB N 762 932 927 779
 Percent 22.41 27.41 27.26 22.91 19.927 0.0001

MB N 3692 2995 2996 3747
 Percent 27.49 22.3 22.31 27.9 156.572 0.0001

SCE N 653 547 547 663
 Percent 27.1 22.7 22.7 27.51 20.533 0.0001

SP N 600 498 497 605
 Percent 27.27 22.64 22.59 27.5 20.0691 0.0002