Evaluating Managed Fund Performance Using Conditional Measures: Australian Evidence. J. Sawicki* and F. Ong* The University of Western Australia

Transcription

1 Evaluating Managed Fund Performance Using Conditional Measures: Australian Evidence J. Sawicki* and F. Ong* The University of Western Australia Abstract Most studies of managed fund performance use measures that are susceptible to bias caused by common time variation in risks and risk premia. Ferson and Schadt (1996) propose a conditional performance measure that controls for the common variation. Their results suggest that incorporating lagged public information variables that have been shown to predict stock returns, such as interest rates and dividend yields, is an improvement over traditional measures. This study applies conditional performance models in analysing managed fund performance in Australia The results confirm the importance of using conditioning information, especially dividend yield, in performance evaluation. Consistent with Ferson and Schadt s (1996) findings, alphas are higher when estimated with the conditional model and the number of significant timing coefficients is greatly reduced. JEL Classification: G20; G23 Key words: Mutual funds; Performance evaluation; Conditional CAPM * The University of Western Australia, Department of Accounting and Finance, Nedlands WA 6907 Australia; ; jsawicki@ecel.uwa.edu.au I.

2 3 Introduction Fund manager performance is typically measured relative to some normal or expected performance. The predominant definition of normal applied in performance studies been based on the traditional Sharpe-Lintner-Black Capital Asset Pricing Model (CAPM) which assumes that fund risk and expected returns are stable over time. If fund risks and risk premia change over time, however, the traditional performance measures will confound time variation with abnormal performance. Evidence that stock returns are predictable using public information variables 1 may indicate that these variables proxy for a variation in the market risk premium. 2 If investors use these market indicators to update rationally their assessments of expected returns, measures of investment performance should incorporate the time variation by using variables that capture the state of the economy. Ferson and Schadt (1996) advocate using performance measures that are conditioned on public information variables in order to avoid the bias induced by using historical average returns to estimate expected performance. Their results demonstrate that using conditioning information is both statistically and economically significant. The average poor performance and perverse market timing ability indicated by unconditional performance measures is negated using the conditional model. This study examines the effect of incorporating lagged information variables in the analysis of investment performance using monthly data for 97 Australian Wholesale funds over the period In general, the results corroborate those of Ferson 1 For example: default risk spread, Keim and Stambauch (1986); dividend yield, Fama and French (1988); earnings yield, Campbell and Shiller (1988). 2 Ferson and Harvey (1991), Fama and French (1992), Evans (1994) provide evidence supporting this interpretation. 1

3 and Schadt (1996). Estimating conditional Jensen alphas causes the distribution of alphas to shift to the right towards positive values, making the sample of funds look better. The traditional Treynor-Mazuy (1966) measure of market timing produces evidence that an average fund takes on more market exposure when market returns are low. The conditional market-timing model reduces the number of significant negative timing coefficients, however the distribution of the timing coefficients remains similar with the traditional timing coefficients. Finally, the conditional performance measure produces no evidence of persistence in fund performance. The following section provides a discussion of the rationale for using conditional performance measures, followed by a description of the data and methodology used in this study in section three. Section four provides the results and interpretation. Section five concludes. 1 The Rationale for Using Conditional Measures The standard performance measures that have been applied extensively in evaluating managed fund performance were proposed more than three decades ago. Jensen s (1968) differential return measure is based on the CAPM where the risk premium on an asset, i (return on asset i less the risk free rate) is a linear function of the systematic risk (beta) of the asset and market risk premium, (r m r f ): r i,t r f,t = α i + β i (r m,t r f,t ) + ε i,t (1) where r i,t, r f,t, and r m,t are the return on asset i, the risk free asset and the market portfolio period t; β i and ε i,t are the beta and the random error of asset i. Manager 2

4 ability is measured by what has become known as the Jensen alpha. Superior (inferior) managers would have consistently positive (negative) random error terms which would be picked up the intercept, α i. In specifically considering managers ability to predict market moves, Treynor and Mazuy (1966) added a quadratic term in another model based on CAPM that has become a standard for measuring timing ability: r i,t r f,t = α i + β i (r m,t r f,t ) + γ i (r m,t r f,t ) 2 + ε i,t (2) By measuring whether the relationship between the portfolio return and the market return in non-linear, the coefficient γ indicates whether the manager is able to correctly predict market performance. Jensen (1968) applied his measure to examine annual rates of return net of management expenses on the portfolios of 115 open-end mutual funds for the period His findings of inferior performance by managed funds are typical of many subsequent studies. 3 Treynor and Mazuy (1966) report evidence of timing ability for only 1 of 57 funds. Other studies using their method produce similar evidence of no timing ability or perverse timing ability. 4 The single-index alpha model has been the predominant approach to performance evaluation until recently when researchers began employing a multi-index model to improve the accuracy performance measurement. 5 Both single- and multiindex models, however, may suffer from another problem: time-variation in risks and 3 On average, the managers were unable to predict security prices well enough to outperform a buyand-hold policy. Furthermore, there was very little evidence that any individual fund was able to perform significantly better than what is expected from pure random chance. Other studies providing corroborating results include Lehmann and Modest (1987) Cumby and Glen (1990) and Gruber (1996). 4 For example, Cumby and Glen (1990) and Coggin, Fabozzi and Rahman (1993). 5 If the portfolio comprises a diverse set of assets not contained in the index, manager ability can be confounded with the performance of the excluded category(ies) relative to the index. Multiple indices representing normal returns on various asset categories controls this. See for example, Gruber (1996). 3

5 expected returns that may be misinterpreted as superior selectivity or timing skills. 6 Performance measures which use unconditional expected returns calculate abnormal performance as the difference between the average portfolio excess return and a betaadjusted average market index excess returns. 7 If the market risk premium changes and the performance metric does not control for this, time variation in the market risk premium will be reflected in the estimate of abnormal performance and mistaken for manager under- or over-performance. Ferson and Schadt (1996) argue that evidence of return predictability using predetermined variables represents changing required returns. 8 They propose a modification to the Jensen alpha and market timing models to incorporate conditioning information that allows for the estimation of time-varying conditional betas. When incorporating lagged information variables on dividend yields and interest rates, they find that the conditional model improves measured performance. Unconditional alphas indicate average poor performance, however the distribution of conditional alphas is consistent with neutral performance. The evidence of perverse timing is removed when using the conditional market-timing model. Ferson and Schadt (1996) modify the traditional Jensen alpha model (Equation 1) by adding a vector of lagged public information variables to equation to estimate α cp the conditional measure of performance: R p,t = α cp + β p (R m,t ) +δ p (R m,t xz t 1 ) + ε p,t (3) where R pt and R mt are the excess return on portfolio p and on the market index over the risk-free rate at time t; B p is the parameter estimating the conditional beta of 6 For example, Jensen (1972) and Grant (1977). 7 Excess return for the market refers to the return on the market less the risk-free rate, also referred to the market risk premium. 8 See footnotes 1 and 2. 4

6 portfolio p; Z t-1 is the vector of public information variables at time t-1; δ p is the vector of parameters that measure how the conditional beta varies with respect to the vector of market indicators. The conditional measure of timing, γ cp, is estimated modifying Equation 2: 2 R p,t = α cp + β p (R m,t ) + δ p (R m,t xz t 1 ) + γ cp R m,t + ε i,t (4) The variables, Z, represent information available at t-1 for predicting returns. These interaction terms pick up movements of conditional betas as they relate to market indicators. The coefficients, δ p, estimate the response of conditional betas to lagged market indicators. By capturing information available to managers at t-1, the vector [R mt Z t-1 ] precludes strategies that can be replicated using public information from being ascribed with superior selectivity or timing ability on the basis of this information. The statistical reason for this is that the interaction terms measure the covariance between conditional beta and the expected value of the market return using lagged instruments. The difference between unconditional and conditional alpha is determined by the average values of the interaction terms. Positive (negative) covariance will result in lower (higher) abnormal performance. Ferson and Schadt (1996) and Ferson and Warther (1996) find that measured performance improves using conditional measures, indicating negative covariance between betas and expected market returns. They suggest two possibilities for this evidence that managers reduce (increase) betas when market returns are high (low). First is that the funds experience strong cash inflows when expected market return is high which are not immediately invested, causing beta to decline. Secondly, betas of underlying assets may change inversely with market performance. 5

7 The purpose of this study is to examine the performance of managed funds in Australia using conditional models. The investment performance of managed funds has been examined extensively in the United States and in United Kingdom, however, there has been little published research in Australia. Studies by Bird, Chin and McCrae (1993) and Robson (1986) employed the traditional Jensen measure and concluded that the overall fund performance was inferior to the market indices. Sinclair s (1990) unconditional timing measure produced evidence of perverse timing by pooled superannuation funds. A conditional performance evaluation measure has not been employed with Australian data. The tests performed here provide insight into the effect of using conditioning information in measuring the performance of Australian managed funds, and is a precursor to further work investigating the reasons for the differences. 9 In this study, performance of Australian Wholesale fund managers 10 is evaluated using traditional and conditional versions of the Jensen alpha (1968) and Treynor- Mazuy (1966) models. The conditional models are estimated using lagged public information on: (1) dividend yield (D/P), (2) short-term Treasury yield, (3) a measure of the interest rate term structure and (4) a dummy variable for January. Dividend yield and interest rates were chosen because previous studies have shown them to be useful market indicators. Fama and Schwert (1977), Ferson (1989), and Breen, Glosten and Jagannathan (1989) found that stock returns are expected to be low when interest rates are high, while Fama and French (1988) and Campbell and Shiller (1988) found stock 9 Preliminary analysis by Ferson and Schadt (1996) suggests that their evidence of a negative correlation between mutual fund betas and market returns may be related to net new flows of money. We anticipate investigating this relationship in an Australian context. 10 The funds management industry can be subdivided into two broad segments: retail, for example unit trusts or mutual funds managing portfolios of individual investors' funds; and wholesale, managing portfolios for large, primarily institutional, investors requiring a minimum investment of $500,000. 6

8 returns are expected to be high when dividend yields are high. The well-documented January seasonal 11 is controlled for using a lagged binary variable. 2 Data Managed Fund Returns The fund performance data were obtained from two commercial investment consultancies - William Mercer (Mercer) and FPG Research (FPG). The Mercer performance series covers one type of fund style, wholesale balanced pooled superannuation funds, which concentrate on portfolios comprising of a mix of asset classes. The performance series employs a monthly performance index with a base value of 1,000 that commenced on 31 December Data obtained from FPG Research includes Discretionary Growth funds (analogous to Balanced funds much of the data overlaps Mercer s), as well as funds that concentrate on a single asset class: domestic equities. The performance series utilises a performance index that has a base value of 10,000 and provided funds performance data up to the month of December The main requirement for inclusion in the sample was that a minimum of 24 consecutive months of performance be reported between January 1983 and December Funds following an index or specialist style 12 were excluded in order to focus on active funds of a reasonably homogeneous nature. Out of possible 190 funds, a total of 97 funds representing two active investment styles met these requirements: Balanced: investing primarily in Australian shares and fixed income securities; Australian equities: investing primarily in domestic equity. 11 See Fama (1991). 12 For example, Australian equity funds investing only in resource stocks. 7

9 Within each style group, both investment consultancies further classify the funds into either pooled superannuation trust (PST) or non-superannuation pooled trust (NPST). Funds classified into PST are funds that manage retirement assets; the PST performance series represent fund earnings after tax, where a 15% tax on earnings is paid at the fund level. NPST funds manage non-retirement assets; the NPST performance series represent fund earnings before tax. The sample for the balanced fund data was constructed from both the FPG and Mercer data set. Twenty-nine of the PST balanced funds used in this study are from Mercer s database. Thirty-seven funds from FPG s discretionary growth style category, including thirteen not reported in Mercer s database are included to make up the study s total of sixty-six funds in the PST balanced group. Sixty-six of the ninety-seven funds in the sample (68 percent) are PST balanced funds. In terms of the number of fund-year observations, the PST balanced funds dominate the other categories. 13 The total sample set contains 655 fund-years observations, with eighty percent (531 fund-years observations) comprised of PST balanced funds. In terms of performance history, the PST balanced funds have an average life of eight years per fund; the other style categories have an average life of four years per fund. Due to the difference in the data sets, separate tests were conducted. The models were estimated using the full sample comprising of the different active fund styles and its sub-samples, as well as using only the sub-samples within each fund style. Thus, the hypotheses are tested using five data set permutations: 13 The dominance of the PST balanced funds data in the sample reflects the history of managed funds in Australia. Tax incentives have encouraged savings in the form of retirement assets; the balanced investment style has traditionally been preferred to equity funds, which have only begun to proliferate in recent years. 8

10 3The PST balanced, 4The NPST balanced, 5The PST Australian equities, 6The NPST Australian equities, 7The all-style groups data set. Table 1 presents the descriptive statistics for the sample. Summary statistics for the individual funds are in Appendix A. [Insert Table 1 about here] As indicated by the mean relative to the median and the low degree of dispersion, it would seem that these measures of central tendency reliably describe the data. The mean monthly return for all the funds is (1.04%) with post-tax PST returns outperforming the pre-tax NPST returns by 0.1% to 0.2%, indicating that the PST funds would definitely outperform the NPST funds at a pre-tax level. The effect of the October 1987 stockmarket crash is captured by the minimum return observation and level of skewness of the PST fund groups. Eliminating the month of the stockmarket crash increased the minimum return by twenty to thirty percent, while the level of skewness changed from a negative skewness to a slightly positive level of skewness, indicating that the extremely low returns occurred during the stockmarket crash. Monthly returns were calculated from the consultancies month-end performance index. The returns include reinvestment of all distributions such as dividends and are net of all expenses except front-end or redemption load charges. It is assumed that investors evaluate performance on the basis of risk and return, and that fund managers 9

11 trade using a one-month horizon. Continuously compounded returns, R p,t, are calculated for each fund as follows: R p,t = ln P p,t P p,t 1 (5) where P pt is the month-end performance index of portfolio p at month t Potential Survival Bias An important threat to the validity of results of many studies is the effect of survivorship bias. If all members of the population being studied do not survive the entire study period, the data will include measures of the surviving members only. Test results will thus be biased to some degree, depending upon the attrition rate of the population, toward the survivors. This issue is one of particular concern to researchers studying the performance of professionally managed funds. Brown, Goetzman, Ibbotson and Ross (1992) demonstrate that sample truncation through attrition creates an appearance of performance consistency among mutual funds and argue that this effect can be severe enough to account for the strength of the evidence of return predictability. They do acknowledge, however, that whether survivorship is sufficient to explain the results of studies demonstrating performance predicability, such as Goetzman and Ibbotson (1991) and Patel, Hendricks and Zeckhauser (1991), is open to question. Interestingly, Patel, et al. (1991) take on this question in the published version of their working paper to which Brown, et. al. (1992) refer. They analyse sub-samples designed to induce survivorship bias and conclude that survivor bias appears unimportant for studying mutual fund performance. 10

12 Mercer s data includes results for funds that have ceased to exist as independent entities which reduces the potential for results biased by studying only surviving funds. It cannot be said, however, that the data are totally free of survival bias. For example, the entire group of 'starters' cannot be established with certainty and the performance of defunct funds 'had they survived' is unknown. FPG's database includes funds only, funds in operation as at May Predetermined Information Variables As market indicators that previous studies have identified as useful for predicting risks and security returns over time, this study employs: the the 30-day Treasury bill yield; the dividend yield of the All Ordinaries Accumulation Index; a measure of the slope of the term structure; and a dummy variable for the month of January. Two sources were used for the one-month Treasury bill yield: (1) RBA Bulletin Electronic Database and (2) RBA Monthly Bulletin. For ease of data collection, the RBA Bulletin Electronic Database was used. To improve data integrity, the Treasury bill yields were cross-checked with those provided in the monthly publication RBA Bulletin produced by the Reserve Bank of Australia. Any discrepancies were dealt with by using the monthly Treasury bill yield from the published RBA Bulletin. As monthly returns are used in this study, it is appropriate to use the yields on 30- day Treasury bills. The yields, stated in percent per annum are converted to continuous monthly rates of returns, R f,t, as follows: R f,t = ln[1+ R a, ft] 12 (6) where R a,ft is the annual yield on 30-day Treasury Bills at time t. The dividend yield represents the previous 12 months of dividend payments for the index divided by the price level at the end of the previous month on the All 11

13 Ordinaries Accumulation Index. The two data sources are: (1) RBA Bulletin Electronic Database and (2) ASX Monthly Index Analysis. The RBA Bulletin Electronic Database was chosen for ease of data collection. The dividend yields were crosschecked with those provided in the monthly publication ASX Monthly Index Analysis produced by the Australian Stock Exchange. Any discrepancies were dealt with by using the monthly dividend yield from the published ASX journals. The term structure represents the relationship between the interest rate and the term to maturity for securities with similar risk. It is a constant-maturity 10-year Treasury bond yield less the 3-month Treasury bill yield. Similar to the 30-day Treasury bill yield, the RBA Bulletin Electronic Database is used as the primary source. The yields were then crosschecked with the RBA Monthly Bulletin and any discrepancy would result in the use of the reported yield in the RBA Monthly Bulletin. The value-weighted All Ordinaries Accumulation Index for all stocks listed on the Australian Stock Exchange (ASX) was used as the market factor (or the benchmark portfolio). The market factor is a representative portfolio comprised of domestic equities. The performance of the market benchmark represents the return to a similar-risk buy-and-hold strategy. This return represents what the investor would have earned investing in the same class (classes) of security (securities), where the particular securities consist of randomly selected portfolios. The fund s performance is compared to the market benchmark and the difference between the two represents the over-or under-performance of the managed fund relative to what would have been earned, had there been no attempt at selectivity or market-timing. The All Ordinaries Accumulation Index represents the total pre-tax return from investments in listed shares after reinvesting 100 percent of dividends at the closing 12

14 market price on the ex-dividend date. The All Ordinaries Accumulation Index were obtained from the ASX Monthly Index Analysis. Because the ASX Monthly Index Analysis report the All Ordinaries Accumulation Index for the previous twelve months, earlier published journals were used to double check for data errors and misprints. The continuously compounded rate of return for the Australian All Ordinaries Accumulation Index over month t, R m,t are estimated as: R m,t = ln AOAI t AOAI t 1 (7) where AOAI t is the level of the Australian All Ordinaries Accumulation Index at month t. Appendix B depicts the behaviour of the market indicators for the period of January 1983 to December In general, the figures illustrate how the variables are related to the economic cycle over the period. Consistent with Ferson and Schadt s (1996) argument, if market indicators can be used to formulate expected returns, then measures of investment performance should accommodate them. 13

15 4 Evaluating Performance Using Traditional and Conditional Measures The main focus of this analysis is to determine whether conditioning on public information has an impact on performance evaluation. We test for differences between the unconditional and conditional approach to measuring performance with the Jensen (1968) and Treynor-Mazuy (1966) models. 4.1 The Jensen Model Data for an equal-weighted portfolio of the funds within (1) each fund style group and (2) the whole sample, are used in an ordinary least-squares (OLS) regression to estimate the unconditional Jensen model (Equation 1) and the conditional model (Equation 3). The results are reported in Table 2. [Insert Table 2 about here] There is some evidence to support proposition that the individual public information variables are related to excess fund return. In the all-funds sample tests, short-term interest rates and dividend yield are significant at the 1% level, while the slope of the term structure and January dummy are not significant. The results are somewhat different when tests are performed within style categories. For the PST Balanced fund groups, dividend yield is the only variable that is significant, while the January dummy is only significant at the 10% level for the two Australian Equities fund group. Finally, short-term interest rate is only significant for PST Australian Equities. This contrasts with Ferson and Warther (1996), who found that the coefficients on D/P and Treasury bill yield were significant at the 5% level or better for the four fund style groups in their study. 14

16 The results provide evidence of the marginal explanatory power of conditioning information in the performance measure. The adjusted R-squares are slightly higher for the conditional model and the partial F-test indicates that the additional variables are significant at the 1% level for the PST Balanced sub-set and the all funds portfolio. Regressions were also performed for each individual fund. Most of the tests indicate that only one of the four variables is significant, predominantly dividend yield (significant for 47 of the 97 funds). The other variables do not seem to be important in explaining returns with short term interest rates only being significant for 17 funds, slope of the term structure being significant for 10 funds and the dummy January variable being significant for 5 funds at the five percent level. Regarding marginal explanatory power, the partial F-test can reject the hypothesis that the additional public information variables do not explain returns at the five-percent level, for 49 of the 97 individual funds. Ferson and Schadt (1996) found that the variables are significant at the conventional level for 75% of their funds. Both measures provide little evidence of significantly abnormal return. Using the traditional measure, only 2 of 97 funds exhibit significant inferior performance and only 9 funds exhibit significant superior significant at the 5% level. Similarly, the conditional measure results in only 2 significantly inferior funds and 11 significantly superior funds. Table 3 provides evidence on whether the conditional alphas are significantly different from the traditional alphas. The traditional alphas are compared with the conditional alphas using the parametric t-test and the non-parametric Wilcoxon Matched-Pairs test. [Insert Table 3 about here] 15

17 The t-test suggests that for all funds the conditional alphas are not significantly different from the traditional alphas. In contrast, the non-parametric test indicates that the alphas estimated using the all-funds sample are significantly different from each other at the 1% level. However, when estimating alpha within style groups only the PST Balanced Funds indicate alphas that are significantly different at the 1% level. Given the small sample sizes for the other style categories, the insignificance of the results is not surprising. A binomial test is used to compare the distributions of conditional and traditional alphas. If there is a significant difference, the proportion of positive and negative alphas may be different between the two. The results are shown in Table 4. [Insert Table 4 about here] The number of negative alphas produced by the traditional measure is 43 (proportion = 44%). In the binomial test, this proportion is found not to be significantly different from a random proportion of 50%. On the other hand, the number of negative alpha estimates under the conditional model is 33 (proportion = 34%). This proportion is significantly different from a proportion of 50% at the 1% level. This implies that when public information is incorporated into the performance measure, the distribution of the alphas shifts to the right and consequently, makes the sample of funds look better. The results are consistent with Ferson and Schadt (1996) and Ferson and Warther (1996), that the conditional alphas are higher than the traditional alphas. 16

18 The results of a comparison of the performance of funds across the different fund style groups using both performance measures are summarised in Table 5. The performances of the fund groups are not significantly different from each other under both measures of performance. This is interesting because the PST funds performance is reported net of tax after These results suggest that on a pre-tax basis, PST funds would probably outperform the NPST funds. It is also evident that within the PST and NPST fund category, the Balanced and Australian equity groups do not perform differently. These results are not consistent with Shawky (1982) and Robson (1986), who found evidence that funds with different objectives or styles produce significantly different levels of risk-adjusted performance and may be due to the small sample size for the groups other than PST Balanced. [Insert Table 5about here] In order to compare evidence of consistency in fund performance, the Pearson Product Moment correlation and its non-parametric version, the Spearman Rank Order Correlations are performed for both performance measures. The correlations of the performance for successive three-year periods reported in Table 6 indicate that there is little consistency in performance from period to period. In addition, neither measures display greater consistency in successive period correlations indicating that evidence of persistence in abnormal performance is similar in both models. [Insert Table 6 about here] 17

19 4.2 The Treynor-Mazuy Model Timing coefficients in the traditional and conditional Treynor-Mazuy models (Equations (2) and (4)) are estimated using individual funds as well as fund groups. Results of for the individual funds, reported in Appendix D, indicate that 73 of the 97 timing coefficient estimates are negative. Of these 73 estimates, 42 are significant at the 5% level. This evidence is consistent with Grinblatt and Titman (1988) and Cumby and Glen (1990), who found negative timing coefficients in unconditional Treynor- Mazuy regressions. Negative timing coefficients indicate perverse timing where the manager lowers exposure to the market when the market performs well and increases exposure in poor markets. Under the conditional version, 71 of the 97 estimates of the conditional timing coefficients of the individual funds are negative. Of these 71 estimates, only 15 are significant at the 5% level. Consistent with Ferson and Schadt (1996) and Ferson and Warther (1996), the evidence of perverse timing ability is removed under the conditional model. Table 7 provides further evidence of this by examining whether the conditional timing coefficients are significantly different from the traditional timing coefficients. The traditional timing coefficients are compared with the conditional timing coefficients using the parametric t-test and the non-parametric Wilcoxon Matched- Pairs test. [Insert Table 7] 18

20 The t-test indicates that for the entire sample, conditional timing coefficients are not significantly different from the traditional timing coefficients. The non-parametric test provides evidence of a significant difference at the 10% level of significance. Within the fund style groups, both tests indicate that the traditional and conditional timing coefficients are not different from each other, which is not surprising given the low level of significance for the all funds group. The binomial test is used to further examine whether the distributions of traditional and conditional timing coefficients differ. The results in Table 8 indicate little difference between the distributions of the traditional and conditional timing coefficients. For both measures, the occurrence of negative timing coefficients is significantly different from chance at the 1% level and implies that both measures are biased towards negative values. This is inconsistent with Ferson and Schadt (1996) and Ferson and Warther (1996), who found that there is an equal chance of observing a positive and negative conditional timing coefficient. It should be noted, however, that even though the distribution of timing coefficients remain the same for both measures, the number of significant negative timing coefficients are greatly reduced when the conditional model is used. [Insert Table 8] 4 Conclusion This study examined the effect of incorporating lagged information variables into the evaluation of investment performance using Australian managed funds data. Although not as strong as the US findings, the use of conditioning information in performance is statistically significant, especially for dividend yield. Furthermore, using conditioning information improves the performance of the funds, causing the 19

21 distribution of the alphas to shift to the right and towards zero. We find that the number of significant negative timing coefficients is greatly reduced after incorporating public information variables into the model, however the distribution is almost the same for the traditional and conditional timing coefficients. The improvement in performance using the conditional approach is somewhat counter-intuitive. The lagged variables represent information available for predicting market returns. Assuming mangers rationally respond to this information and adjust their portfolios accordingly, they will increase (decrease) exposure to the market when an increase (decrease) in returns is predicted. Unconditional performance will confound these adjustments based on public information with abnormal performance. The conditional measures control for this and would be expected to produce more pessimistic results vis-à-vis traditional measures. The improvement in conditional results reported here and by Ferson and Schadt (1996) are attributable to negative covariance between fund betas and market return, which is controlled for by the conditioning information. An important part of the story is why? Negative covariance between beta and market returns suggest that managed funds reduce (increase) their exposure to the market when returns are high (low). This apparently irrational behaviour may be attributed to fund cash flows or changing asset betas. Further work should directed towards examining the reasons. Other areas further study is into performance persistence using the conditional models. For example, previous studies have found that performance persistence may be concentrated in the extreme-performance funds. It would be interesting to whether the extreme performers may be more easily detected using conditional methods. Finally, 20

22 Australian retail funds should be examined in order to compare results to wholesale funds and gain more insight into the reasons for the results. 21

23 References Bird, R., H. Chin and M. McCrae, 1983, The performance of australian superannuation funds, Australian Journal of Management 8(1), Breen, W., L. Glosten and R. Jagannathan, 1989, Economic Significance of predictable variations in stock index returns, Journal of Finance 44(5), Brown, S.J., W.N. Goetzmann, R.G. Ibbotson and S.A. Ross, 1992, Survivorship bias in performance studies, Review of Financial Studies 5, Campbell, J.Y. and R.J. Shiller, 1988, The dividend price ratio and expectations of future dividends and discount factors, Review of Financial Studies 1(1), Coggin, D.T., F.J. Fabozzi and S. Rahman, 1993, The investment performance of US equity pension fund managers: an empirical investigation, Journal of Finance 48, Cumby, R. and J. Glen, 1990, Evaluating the performance of international mutual funds, Journal of Finance 45, Evans, M., 1994, Expected returns, time-varying risk, and risk premia, Journal of Finance 40, Fama, E.F., 1970, Efficient capital markets: a review of theory and empirical work, Journal of Finance 25, Fama, E.F. and K.R. French, 1988, Dividend yields and expected stock returns, Journal of Financial Economics 22(1), Fama, E.F. and K.R. French, 1989, Business conditions and expected returns on stocks and bonds, Journal of Financial Economics 25, Fama, E.F. and G.W. Schwert, 1977, Asset returns and inflation, Journal of Financial Economics 5, Ferson, W. and C.R. Harvey, 1991, The variation of economic risk premiums, Journal of Political Economy 99, Ferson, W. and R. Schadt, 1996, Measuring fund strategy and performance in changing economic conditions, Journal of Finance 51, Ferson, W. and V.A. Warther, 1996, Evaluating fund performance in a dynamic market, Financial Analysts Journal 52(6), Goetzman, W. and R. Ibbotson, 1994, Do winners repeat? Patterns in mutual fund behaviour, Journal of Portfolio Management (Winter),

24 Grant, D., 1977, Portfolio performance and the cost of timing decisions, Journal of Finance 32, Grinblatt, M. and S. Titman, 1994, A study of monthly mutual fund returns and performance evaluation techniques, Journal of Financial and Quantitative Analysis 29, Gruber, M., 1996, Another puzzle: the growth in actively managed mutual funds, Journal of Finance 51, Jensen, M.C., 1968, The performance of mutual funds in the period , Journal of Finance 23, Jensen, M.C., 1969, Risk, The pricing of capital assets, and the evaluation of investment portfolios, Journal of Business 42, Lehman, B.N. and D.M. Modest, 1987, Mutual fund performance evaluation: a comparison of benchmarks and benchmark comparisons, Journal of Finance 42, Patel, J., R. Zeckhauser and D. Hendricks, 1991, the rationality struggle: illustrations from financial markets, Behavioural Finance, 81, Robson, G.N., 1986, The investment performance of unit trusts and mutual funds in Australia for the period 1969 to 1978, Accounting and Finance (November), Sharpe, W.F., 1966, Mutual fund performance, Journal of Business 39, Shawky, H.A., 1982, An update on mutual funds: better grades, Journal of Portfolio Management (Winter), Sinclair, N.A., 1990, Market timing ability of pooled superannuation funds january 1981 to december 1987, Accounting and Finance (May), Treynor, J. and K. Mazuy, 1966, Can mutual funds outguess the market? Harvard Business Review 44, Warther, V.A., 1995, Aggregate mutual fund flows and security returns, Journal of Financial Economics 39(2),

25 APPENDIX A Funds in Sample: Performance Summary Statistics No. Fund Name Period Mean Monthly Return Standard Deviation Min Max PST Balanced Funds 1 Aetna AMP Pooled Super Balanced Fund 3 AMP Pooled Super Managed Equity Fund 4 ANZ Life ANZ Super Pool Fund Growth 6 ANZCAP Australian Eagle Accent BNP - Balanced Pooled Super Trust 9 BT - Managed Fund PST BT - Retirement Fund BT Stable Growth Fund CBA Managed CFM PST - Managed Growth Fund 14 Citicorp Pooled Super'n Balanced Trust 15 Colonial Financial Management 16 Colonial Pooled Super'n Fund - Balanced Fund 17 Commonwealth Life Corporate Super - Managed 18 Credit Suisse PST - Capital Growth 19 Delfin Equitilink SuperTrust Equity Life Managed First National First State Fund Mgrs GIO Wholesale - Super Balanced Growth 25 Guardian Hambros HSBC - Growth Pooled Super Trust 28 Inlife J.P. Morgan - Balanced Trust 30 L & G Supermanagement Balanced Fund 31 L & G Supermanagement - Managed Fund

26 32 Macquarie W'sale Super Balanced Inv 33 Maple-Brown Abbott Pooled Super Trust 34 McIntosh APPENDIX A Funds in Sample: Performance Summary Statistics No. Fund Name Period Mean Standard Deviation Min Max 35 Merc Mutual Super Inv Managed Growth 36 Mercury Balanced Pooled Super'n 37 MLC - Balanced Fund MLC - Growth Fund Morgan Grenfell NAAM Balanced Nat Aust Pooled Super Balanced Pfolio 42 Nat Mutual Wholesale Balanced Fund 43 Nat Mutual Wholesale Superannuation Fd 44 National Australia Bank NM Equity Linked NMFM Superannuation NML Balanced Norwich Balanced Norwich Super & Investment - Managed Fnd 50 NZI Life Occidental Life Potter Warburg Balanced Prudential Wholesale Inv Balanced 54 Prudential Wholesale Inv Growth 55 Regal Life Rothschild - 5 Arrows Sup Investment Tr 57 Schroders Super Fund SMF Sector Leaders Capital Growth Fund 59 SPAL State Bank Victoria Sun Alliance - Premier Growth Fund 62 Suncorp - Balanced Super No 2 Fund 63 Tower Pooled Super Balanced Growth Fund 64 Tyndall Wholesale Managed Pooled Fund 65 Westpac Corporate Super - Capital Growth

27 66 Zurich - Managed Growth Fund APPENDIX A Funds in Sample: Performance Summary Statistics No. Fund Name Period Mean Standard Deviation NPST Balanced Min Max 67 Advance - Managed Pension Fund 68 AJ Wholesale - Balanced Fund 69 BNP - Managed Fund BZW Managed Inv Growth Fund 71 County NatWest Investment Growth Fund 72 L & G Supermgmt A Pens Managed 73 L & G Supermgmt A Pension - Balanced 74 Maple-Brown Abbott Diversified Inv Trs 75 Prudential Investment Trust Balanced 76 Prudential Wholesale Inv Balanced NTP 77 Westpac Pooled Inv Trs - Capital Growth PST Australian Equities 78 AMP No 2 Pooled Super - Australian Share 79 BZW Superannuation - Australian Equity 80 L & G Supermanagement - Aust Equities 81 MLC Corporate Pooled - Aust Share Fund 82 Nat Aust Pooled Super - Equity Portfolio 83 Norwich Super & Inv - Australian Eqtys 84 SMF Sector Leaders - Aust Equities Fund NPST Australian Equities 85 AJ Wholesale - Australian Share Fund

28 86 BZW Managed Inv - Australian Share Fund 87 CFM GIT - Australian Equities Fund 88 County NatWest - Australian Equity Trust 89 Credit Suisse - Australian Shares Fund 90 First State Wholesale - Equities Fund APPENDIX A Funds in Sample: Performance Summary Statistics No. Fund Name Period Mean Standard Deviation Min Max 91 Invia - High Asset Trust L & G Supermgmt A Pension - Aust Equity 93 Macquarie - Australian Enhanced Equities 94 Maple-Brown Abbott Australian Equity 95 Mercury - Master Australian Eqty 96 Prudential Investment Tr Aust Equities 97 SBC Key Investment - Australian Equities

29 APPENDIX B Figure 1 Cyclical Movements of Market Indicators One Month Treasury Bill Dividend Yield Term Structure Date 1

30 Figure 1 illustrates the behaviour of the market indicators for the period January 1983 December It can be seen that from , dividend yield was declining. This period was characterised as high economic growth. A stable level of dividend payments combined with rising share prices resulted in decreasing dividend yields. The nominal short-term interest rate significantly increased (by as much as 10 percent) due to rising inflation rates, which also may accompany high economic growth. The period was characterised by a small downturn in 1988 and the recession. The figure shows that dividend yield increased due to falling share prices. In 1990, interest rates started to fall as the government tried to stimulate growth by lowering interest rates. After the recession, dividend yield began to fall as share prices started to rise as the economy began to recover. The term structure has a counter-cyclical pattern similar to the pattern of the onemonth Treasury bill with short-term interest rates rising (falling) relative to long-term interest rates, and thus the premium of the long-term interest rates over the short-term interest rates declining (increasing). However, the counter-cyclical pattern did not exist for the year Both short-term interest rates and long-term interest rates declined because during this period, investors expected future inflation to fall. In 1994, longterm interest rates quickly rose as investors expected inflation to increase. In general, Figure 1 illustrates that market indicators can be used to show the economic cycle. If market indicators can be used to formulate expected returns, then measures of investment performance should accommodate them. 1

31 APPENDIX B Figure 2 Expected Compensation for Market Risk a b Risk Premium Jan-83 Jan-84 Jan-85 Jan-86 Jan-87 Jan-88 Jan-89 Jan-90 Jan-91 Jan-92 Jan-93 Jan-94 Jan-95 Date 1

32 Figure 2 plots estimates of the risk premium per unit of market beta. The vertical line a represents the beginning of an economic downturn that led to the recession, while vertical line b represents the start of the economic recovery after the recession. The market factor is used to obtain the risk premium by finding the difference between the return on the market factor and the risk-free interest rate. The risk premium represents the economy-wide expected compensation for exposure to economic risk. The figure shows that the risk premium for a unit of market beta increased during the recession and peaks at the start of the economic recovery. Furthermore, the stock market premium tended to decline during the economic expansion after the recession. This concurs with previous studies such as Fama and French (1989), who found that risk premium rises (falls) during economic contractions (expansions). The economic rationale for this pattern is that risk capital is relatively cheap during an economic boom period and is relatively expensive during recessions. During an economic boom period, low expected returns are good enough to induce investors to invest. On the other hand, during a recession, high expected returns are required to persuade investors to forgo current consumption in favour of investment. 1