The persistence of unit trust performance for the period July 1985-June 1995

The purpose of this study was to determine whether any persistence of performance existed in the unit trust industry in South Africa over the ten-year period from July 1985 to June 1995. Calculations were done over different time periods (one-. twoand four-year periods) and using different definitions of superior performance (positive Jensen alphas or winner/loser phenomena). Results of nominal returns and risk-adjusted returns were also compared. Results obtained show that persistence in performance does exist, but that it is more of a 'loser' phenomenon than a 'winner' phenomenon.


Introduction
Investors usually judge and select unit trusts on the basis of the unit trusts' performance track record.Although the usefulness of the track-record approach seems obvious to participants, most academics do not believe in this approach.The efficient market hypothesis implies that past performance is no guide to future performance after adjustment for risk or other pricing factors.If the hypothesis is literally true, not only can the average manager not be expected to outperform passive management, but even managers with the best historical record can not be expected to keep up their performances in the future.This study focusses on the second part of the hypothesis, and asks whether persistence in performance does exist, and whether the best-performing funds of the past are likely to be the best-performing funds of the future.

Review of literature
Several mutual fund studies in the USA have looked at predictability of performance as part of a larger study of mutual fund performance.These include the studies by Lehmann & Modest ( 1987), Grinblatt & Titman ( l 989a) and Elton, Gruber, Das & Hlavka (1993).Recently, however, there have been several more focussed articles that directly examine persistence in mutual fund performance and claim to have isolated a 'hot hand' phenomenon.These include the articles by Grinblatt & Titman ( 1992), Shukla & Trzcinka (1992), Hendricks, Patel & Zeckhauser (1993), Goetzman & Ibbotson (l 994 ), Bauman & Miller (l 994 ), Brown & Goetzmann (1995), Malkiel (1995) and Elton, Gruber & Blake (1996).Grinblatt & Titman (1992), using an eight-portfolio benchmark.concluded that there is a positive persistence in mutual fund performance.Shukla & Trzcinka ( 1992) examined persistence in the performance of 1387 mutual funds listed on the NASDAQ (a computerized communications network that serves the over-the-counter [OTC] market in the USA) at the end of March 1989.They indicated two _problem a~eas associated with the fact that they found persistent superior performance: -The data is subject to survivorship bias since it consiSls only of funds that had survived until 1989; and -Most mutual funds are owned by investment advisers who manage groups of funds.These advisers may have the incentive to use superior information to build the reputation of the fund group rather than of the individual funds. 1   Their overall conclusion was that the persistence of per• formance in all funds arises from a persistence of inferior performance rather than from a persistence of superior per• formance.Hendricks, Patel & Zeckhauser ( 1993) examined the quarterly net returns of 96 no-load growth funds during the period from 1974 to 1987.They found that when perfonn• ance is measured using the Jensen alpha, mutual funds that have performed well in recent years tend to continue to be superior net performers in the next one to eight quarters.'Icy hands', the 'evil' counterpart of 'hot hands', also show up in their sample.Goetzmann & Ibbotson (1994) found that past returns and relative rankings are useful in predict• ing future returns and rankings.They reached this conclu• sion for nominal returns, Jensen risk-adjusted alpha measures and style-categorized subgroups.Brown & Goetzmann ( 1995) used a data set that they claim is free of most forms of survivorship bias and found that the persist• ence is strongest in losing mutual funds.Malkiel ( 1995) looked at mutual fund returns during the 1971 to 1991 period and confirmed the persistence phe• nomenon but noted two caveats.Firstly, the findings are likely to be influenced by survivorship bias.Secondly.the relationship may not be very robust since the strong persist• ence that characterized the 1970s failed to exist during the 1980s.Elton, Gruber & Blake ( 1986) also examined per• sistence in mutual fund performance and concluded that the past carries information about the future and that when per• formance is evaluated over a one-year evaluation period, the previous year's data conveys much more information about performance than data from the previous three years.Some degree of •persistence in performance has been established, hut three issues remain unresolved in the studies in the USA namely: ' -How can one truly eliminate survivorship bias?-What is the appropriate technique for risk adjustment?-Does the length of the period (selection period) influence the chance of correctly predicting the following period (evaluation period)?
This study attempts to find an answer for the initial issue of whether there is persistence in performance in the South African unit trust industry.The potential for survivorship bias is a real problem for studies in the USA because of the large number of funds that have closed down.In South Africa, this problem does not exist to a material extent, since only a few of the unit trusts have as yet closed down (in the sample period of July 1985 to June 1995, none closed down).The two remaining issues, namely whether risk-adjustment and time-periods influence results, are addressed in this study.Central to the issue of risk adjustment is how to define excess performance.A common yardstick of riskadjusted returns, the Jensen measure, was used in this article.Whether or not the security market line represents a legitimate and meaningful benchmark for managers' performance and whether the All Share index (ALSI) is the correct benchmark is not addressed in this study.To determine whether the length of the time periods is important to predict performance, the tests were done over four-year, two-year and one-year intervals.

Sample and measurement of performance
Unit trusts traded in South Africa for the period from July 1985 to June 1995 were used as the sample.By June 1995, there were 84 unit trusts (24 general funds, 38 specialist equity funds, 16 income funds and six gilt funds) in South Africa.However, only thirteen of these funds were in existence for the entire ten-year period.It was decided to use these thirteen funds as well as the 33 funds that were in existence for the five-year period from July 1990 to June 1995 as the sample for the investigation (sec Appendix I).
Repurchase prices and dividend information were obtained from the University of Pretoria.It was argued that for the purposes of this research, the repurchase price is more valid than the selling price, because repurchase returns are compared to indices that do not contain transaction charges.Monthly rates of return were calculated using equation I and crediting the dividends in the month of payment.where R111 = the return on the security in period t, P11, = the price of the security at the end of period t, and D"' = the total of all the dividends paid during period t.

(I)
To adjust returns for risk, the Jensen alpha was calculated for all the funds over the different periods.Using the CAPM as a theoretical framework.Jensen ( 1968) postulated that the overall performance of a portfolio could he assessed from a regression equation.The Jensen alpha (JA) IOI is defined as the difference between the actual average return by a portfolio and the equilibrium return that should have been earned by the portfolio, given market conditions and the risk of the portfolio.Suppose the portfolio actually earned an average return of Rr, then the Jensen alpha is defined as follows: (2) The logical choice for a share index for a South African study is the Johannesburg Stock Exchange All Share index (ALSI) which is the most comprehensive South African index.The ALSI is calculated using the market capitalization weighting method.Monthly rates of return were obtained from I-Net.~The methodology used is the same as that used to calculate the performance results published by the Johannesburg Stock Exchange (JSE).Data on the 3-month treasury bill is used to represent the risk-free rate of return in the study.
Initially, to answer the question whether there was persistence in performance on a risk-adjusted basis.data was divided into different overlapping periods.The ten years' worth of data was segmented into seven overlapping fouryear periods: July 1985-Junc I 989.July 1986-June 1990. July 1987-June 1991, July 1988-June 1992, July 1989-June IQ93, July 1990-June 1994and July 1991-June 1995.The five-year period data was divided into four overlapping two-year periods: July 1990-June 1992.July 1991-June 1993, July 1992-June 1994 and July 1993-June 1995.To avoid the problem that overlapping may lead to false conclusions about persistence, the alphas in adjacent period were also examined.The ten-year period was divided into three four-year adjacent periods: July 1985-June 1989and July 1989-June 1993. July 1986-June 1990and July 1990-June 1994and July 1987-June 1991and July 1991-June 1995.The performances in these three adjacent periods were further categorized according to winner/loser phenomena.Unit trusts were classified as winners if they had positive alphas and categorized as losers if they had negative alphas.Rank order correlations between the adjacent periods were calculated to further determine persistence in performance.
To investigate whether past performance can he used to predict future performance.data was categorized according to winner/loser phenomena based on the work done hy Goetzmann & lhhotson ( 1994).Unit trusts were classified as winners if their performance was better than the median performance of the funds in total and as losers if their performance was worse than the median performance of the funds in total.A measure of how funds rank ,•is a ,•is one another was preferred to one that measures performance against some benchmark because of the unresolved issues relating to benchmarks.The data was divided into successive two-year and one-year intervals to establish whether the length in the selection periods influences results in the evaluation period.This was done for nominal returns and risk-adjusted returns to see if there are differences in results.To establish significance in results, an alternative approach of running regressions of two-year alphas on subsequent two-year alphas was also done.A significant positive t-statistic for the slope coefficient in this regression would reject the null hypothesis that past performance is unrelated to future performance and support the alternative hypothesis that past performance is positively related to future performance.

Results
Tables I and 2 illustrate the results of the Jensen alphas in the overlapping periods for the two sample periods.
Of the 13 funds in the ten-year period, These results clearly indicate that some funds may have persistent superior performance and that some funds exhibit persistent negative performance.Similar results were obtained for adjacent periods.
Further investigations on the adjacent periods show that the chances of obtaining superior performance, based on the previous four years' results, was 100% for the first period, 77.8C'/c for the second period, 50% for the last period and 73% for the combined periods (see Table 3).The first period, as well as combined results, was significant at a 5% The results for the nominal returns and risk-adjusted returns over the successive two-year intervals for both sample periods are illustrated in Tables 5 and 6 (categorized according to whether or not the results were above or below the median performance).
The information in Table 5 shows the combined results of all the two-year periods for the 13 funds, indicating that the ratio associated with picking a winner, based upon past winning performance, is about 60/40 (58.3/41.6% ).In two periods there was a 66% chance, and in the remaining two periods, a 50% chance of picking a winner.If investors demand higher returns from riskier funds, one may argue that tests using returns uncorrected for risk merely document the differential expected return between high-risk versus low-risk funds.In order to investigate the question, the same test was done on average alphas over the two-year period.The repeat winner phenomenon remains visible (see Table 5-second part of table).Two periods still have a 66% chance and two a 50% chance, but in different time periods than for nominal returns.The chance of picking a loser, based on past losing performances, was even better (on nominal returns as well as on risk-adjusted returns), namely a 64% chance.Results from the Goetzman & Ibbotson study (1994) show that the chance of picking a winner improves from 59.1 to 62% with risk adjustment.The chance of picking a loser from losing performances ranges ~tween 59 and 63.6%.The results from this study differed m two respects from the Goetzmann & Ibbotson 's results, namely, in that similar results were obtained with nominal and risk-adjusted performance and that there was a bigger difference between repeat winner and repeat losers.• = significant at 0.05 confidence level For the 33 funds over the five-year period, the repeat winners and repeat losers phenomena only existed for risk-adjusted performances (56.3% and 58.8% ), see Table 6.To draw any conclusions from one period of observation may be insufficient.
From the ten-year sample it is clear that, over the twoyear period, the repeat loser phenomenon was more persistent than the repeat winner phenomenon.An alternative approach to run regressions to measure the magnitude of the two-year alpha on the subsequent two-year alpha is presented in Table 7.The results appear significant in only one of the four periods, assuming independence of observations.
Next, the results were analyzed assuming that winners and losers are ranked and determined over one-year peri- ods, and then ranked again over the subsequent one-year periods.The results were also categorised into high-variance funds and low-variance funds, to see whether the results are related to fund variance.The variance of the returns of all the funds was measured over the entire period and then ranked.The funds with variance above the median were categorised as high-variance, while median and below were categorised as low-variance funds (see Tables 8 to   11 ).
The repeat winner phenomenon existed only for low-variance funds, 53.3% on nominal returns and 54.5% on riskadjusted returns for the ten-year sample period.For the five-year sample period, the winner phenomenon existed for low-variance funds with nominal returns 52.8% and for high-variance funds on a risk-adjusted basis 55.6%.  .when compared lo the two-year perioru;.The 1:~ut hi:.t.>r phenomenon c.\.isted for lund~ m ltltal for •buth :.unwlc :pcno<i., 011 nommul return~ u~ well u., on II t1.'i,t,.-udJus11:dba.'11.., und t}:j~t.aall~for 'hiph-vurium:e fondh \tU11!Jtn.fhctw~un fi:2.Y und 5~.5~ {or lotul lund., und betwcun •Stund {( 7 tt,"li {or h1ph"vurmnc.:efund., 1.On 11Vfil1ll!e 11 :.\.'tlUi..~ thut both •the :nweat wmner .und:mpcat lo,,er :µhe-  these concerns, but some remain challenges for future research.Limitations of this research are the following: the small sample size for information currently available in South Africa and the absence of significance in the testing of the results.The Jensen alpha was used in all the tests to adjust for risk, however, not all the precautions for using a Jensen alpha were adhered to, for instance no adjustments were made for problems ansmg due to the non-stationarity of various distributions.A measure of how funds rank vis a vis one another (winner/loser phenomena) was preferred to one that measures performance against some benchmark because of the unresolved issues relating to benchmarks.Despite these limitations.few things can be learned from this study.The overall conclusion is that results in the South African market are more or less similar to those obtained in a much bigger market, for example, in the USA.Some persistence in performance of unit trusts in the South African environment does exist, although this persistence is not statistically significant.The repeat winner phenomenon exists over two-year periods for nominal returns as well as for risk-ad-    justed returns.These results are basically identical to results obtained in the USA.The repeat winner phenomenon is much lower over one-year intervals on both total and risk-adjusted returns.The best results for persistence were obtained over four-year periods, when winner performance is defined as positive alphas.This result is different from those of studies in the USA, where it seems that the repeat  winner phenomenon is stronger over shorter periods of evaluation.

Cmmdllllllimtn
The repeat loser phenomenon existed over the one-year, two-year and four-year time periods and at much higher percentages for the one-and two-year periods than for the four-year period.These results showing a stronger loser phenemenon are similar to results obtained in the USA.The results which show that the loser phenomenon is stronger over the shorter evaluation periods and that the winner phenomenon is stronger over the longer evaluation periods is an interesting feature that needs further investigation.
There is very little difference between the results obtained on nominal returns and those obtained on risk-adjusted returns.These results are also similar to results obtained in the USA.Persistence in performance seems to exist and it appears to be a guide to beat the pack in the long run.At this stage, it seems that the longer the evaluation period, the better the results for the winner phenomenon.

Notes
individual fund portfolio's return R, = the riskless rate (3-month treasury bill) R111 = the return on market portfolio (All Share index) ~r = the coefficient of OLS regression of fund return on the market portfolio over the period.

Table 2
Jensen's alphas of 33 unit trusts in four overlapping 1995 periods from July 1990 to June

Table 6
Tests of persistence of_ unit trust performance over successive two-year intervals from July 1990-June 1995 ranked on nominal returns/risk-adjusted performance

Table 1
Regression of last two years' alphas on the next two years' alphas

Table 8
Tests of persistence of unit trust performance over successive one-year intervals from July 1985 to

Table 9
Tests of persistence of unit trust performance over successive one-year intervals from July 1985 to June 1995 ranked on risk-adjusted returns

Table 10
Tests of persistence of unit trust performance over successive one-year intervals from July 1990 to June 1995 ranked on nominal returns

Table 11
Tests of persistence of unit trust performance over successive one-year intervals from July 1990 to June 1995 ranked on risk-adjusted returns I. The observer may miss this superior performance if only funds are examined.Hence, Shukla & Trzcinka also examined the best fund of the group and the average fund of the group.For both individual funds and fund groups.they found that inferior performance persists but superior performance does Elton, E.M .. Gruber.M.J. & Blake.C.R. 1996.The persistence of risk-adjusted mutual fund performance.Journal of Business.69(2): 133-157.Elton.E.M., Gruber, M.J., Das, S. & Hlavka.M. 1993.Efficiency with costly information: a re-interpretation of evidence for managed portfolios, The Review of Financial Studies.6( I): 1-22.Garvin.T. 1995.A study of the relative performance of South African unit trust fund managers utilizing the portfolio change measure technique.Unpublished MCom dissertation.Cape Town: University of Cape Town.