On timing ability in Australian managed funds

Abstract

This paper investigates the timing abilities of Australian managed fund managers. To examine timing abilities, the cross-sectional bootstrap approach is adopted to determine whether timing ability is due to skill or luck. Based on three alternative timing measures, we find that top-ranked Australian fund managers have genuine timing abilities. In addition, the poor timing ability with bottom-ranked funds is not likely to be due to luck either, implying that the market exposure of some Australian managed funds is mistakenly adjusted when the stock market improves.

Keywords

Australian managed fund bootstrap approach joint timing return timing volatility timing

1. Introduction

Are fund managers able to time the market? Since this is one of the central questions in the finance literature, investors and researchers have devoted much attention to the performance evaluation of portfolio managers since Jensen (1968). Models that estimate both market timing and stock selection ability have been proposed by Treynor and Mazuy (1966), hereafter TM, and Henriksson and Merton (1981), hereafter HM. Market return timing¹ is a strategy designed to enhance performance, changing portfolio risk over time according to expectations about future returns in certain markets (Heaney et al., 2007). In addition to market return timing, Busse (1999) has recently proposed a model to measure market volatility timing. This author’s basic intuition is that a market timer should reduce market exposure when his forecast indicates an increase in market volatility, assuming all else is equal. In other words, market volatility timing shows how a fund manager reacts to changes in market volatility through the adjustment of fund beta.

The main objective of this paper is to investigate the timing ability of Australian managed fund managers. There are plentiful facts that warrant the importance of a market timing study of Australian funds. While Gharghori et al. (2007) quote from an industry survey that Australia ranked the highest in managed fund investments per capita in 2005, more recent figures from Lee (2012) confirm that Australia presently maintains its place in the ladder. In fact, Australia’s managed fund industry has more than tripled in the past 15 years (Yap and Pierce, 2008). Moreover, given that Australia fared well through the subsequent global financial crisis compared with other industrialized economies, its segment of the global funds market remains significant for research purposes. As of the first quarter of 2012, Australia’s share of total consolidated assets under managed funds has recovered and surpassed pre-crisis levels (Australian Bureau of Statistics, 2012). In addition, a compulsory superannuation scheme and tax incentives for additional voluntary contributions are likely some of the driving factors for the strong standing of Australian managed funds (Erskine, 2012). However, as Gharghori et al. (2007) note in motivating their study of Australian managed funds, consistently high investment earnings on Australian managed funds have been observed since the 1990s, which has perhaps offered investors an advantage with managed funds over conventional investment alternatives, such as cash and bonds. Moreover, the authors allude to the increased sophistication of the Australian financial market as a factor for the strong growth. In addition, the retail market for managed funds is dominated by domestic institutions, representing 85% of the entire market, suggesting the strength of the industry (Australian Trade Commission, 2011). An important question then emerges: Do Australian managers hold timing abilities that may explain the praised high earnings and sophistication?

The present article uses similar methodologies to Chen and Liang (2007), but the research subject is different. While Chen and Liang (2007) use hedge funds data, we focus on Australian managed funds. No prior studies have examined the timing ability of managed fund managers in the UK or the US, which avails us no comparable evidence.

Notwithstanding that the literature on Australian managed fund managers’ timing ability is sparse, overall evidence from many studies (e.g., Gallagher, 2001; Hallahan and Faff, 1999; Prather et al., 2001) suggests that there is little or no market return timing ability, while Sinclair (1990) and Sawicki and Ong (2000) document perverse market return timing ability. Sinclair (1990) examines the market timing ability of 16 pooled superannuation funds during the 1980s and finds that most had perverse timing ability. Holmes and Faff (2004) even discover strong evidence of negative market timing by including market volatility timing in their analysis. The majority of past studies on Australian funds adopted the TM quadratic market model and/or HM dual beta models in their measurement of market timing ability. In a recent study, Do et al. (2009) examine the market timing and volatility timing skills of Australian hedge fund managers and find that they possess neither market timing nor volatility timing skills. However, none of these studies examine timing ability by accounting for all timing measures and differentiating “skill” from “luck” in a bootstrap simulation framework.

Several statistical issues arise from timing the ability testing of mutual fund managers. When there is only one fund, a timing ability test can be based on the t-statistic of the timing measure. However, because neither the timing benchmark nor the strategy is observable, specification errors are likely. For instance, if a portfolio manager is evaluated using either the HM or Jensen (1968) model but acts according to the TM model, statistical inference about fund manager performance, timing aptitude, and selectivity ability can be erroneous. Not only sensitivity from strategies, as noted by Jiang et al. (2007), but apparent evidence of timing ability can result from just luck. In other words, even though none of the funds under examination exhibit true timing ability, some funds may have significant timing measures (based on t-statistics), due to pure chance, in an analysis involving a large number of funds (Jiang et al., 2007).

In examining whether or not market timing exists for Australian funds, we adopt a cross-sectional bootstrap approach² in an attempt to separate skill from luck. The cross-sectional bootstrap approach is adopted for at least two reasons. Firstly, the t-statistics of the timing measures are asymptotically normal under quite general conditions; however, the timing betas of funds may be correlated due to similar stock holdings and the t-statistics are not independent and identically distributed across funds. This issue is complicated further by the fact that funds can exist for only a short period and some funds may not overlap with each other at all. The finite sample distributions for the cross-sectional statistics, particularly for the extreme percentiles, can differ from their asymptotic counterparts (Chen and Liang, 2007).

Secondly, the cross-sectional bootstrap approach has the advantage of allowing researchers to obtain a timing distribution for all funds. More specifically, the cross-sectional bootstrap approach does not consider the luck distribution of individual funds but, rather, considers that of all funds, which enables us to draw statistical inferences of funds in the extreme tails of the cross-sectional distribution (i.e., extreme winner or loser). For this reason, the approach³ has been recently employed by Kosowski et al. (2006, 2007), Cuthbertson et al. (2008), and Fama and French (2010).⁴

This paper contributes to the market timing literature by examining Australian managed funds in the following two aspects. Firstly, both the ability of market return timing and market volatility timing are examined. Further, we adopt the test proposed by Chen and Liang (2007), which permits us to test market return timing and market volatility timing jointly. This newly proposed test relates fund returns to the squared Sharpe ratio of the market portfolio. Secondly, and more importantly, by adopting the cross-sectional bootstrap analysis, we do not examine the timing ability for individual funds but, instead, for the Australian managed fund sector as a whole. In addition, the cross-sectional bootstrap approach distinguishes timing skills from luck and compares the best and worst market timers with the luck distribution, derived from all funds. This is because individual investors are more concerned about extreme market timers.

Our core finding, based on three timing measures, is that the top-ranked Australian managed funds have genuine timing abilities, while bottom-ranked funds have poor timing ability, which is not likely to be associated with luck either.

The rest of this paper is organized as follows. Section 2 presents the research framework for examining timing ability and the bootstrap method. Data and empirical results are discussed in Section 3. Section 4 concludes the paper with a summary of our results.

2. Econometric models

Market timing refers to the dynamic allocation of capital among broad classes of investments depending on market conditions. When fund managers believe they possess timing ability, they increase the portfolio weight on equities prior to a forecast rise in the stock market and decrease it prior to a forecast fall. In other words, timing ability depends on a fund manager’s ability to change the beta of portfolios.

2.1. Market return timing

One possible explanation for superior risk-adjusted returns is that managers possess and utilize superior market return timing and security selection abilities. The test for market return timing is based on the assumption that fund managers who time the market adjust fund beta over time, with high-beta (low-beta) portfolios more likely during periods when equities are performing better (worse) (Heaney et al., 2007). That is, fund managers can adjust market factor loading according to future market conditions (i.e., $β_{i} = β_{i}^{M K T} + γ_{i} (R_{m, t})$ . This adjustment yields a non-linear characteristic line. Therefore, TM use the following quadratic equation:

R_{i, t} = α_{i} + β_{i}^{M K T} R_{m, t} + γ_{i} R_{m, t}^{2} + ε_{i, t}

(1)

where α_i is the intercept term of fund i, $β_{i}^{M K T}$ is the loading to market excess returns R_m,t, and the coefficient γ_i measures market return timing. In Equation (1), the intercept term captures abnormal excess returns attributable to stock selection ability (risk-adjusted returns) only and γ_i > 0 indicates successful market return timing (Gallagher and Jarnecic, 2002).

Our analysis extends the TM model by including the Fama–French factors,⁵ based on Goetzmann et al. (2000), who find this setup improves market return timing specifications by reducing measurement bias. Therefore, the extended model is

R_{i, t} = α_{i} + β_{i}^{M K T} R_{m, t} + γ_{i} R_{m, t}^{2} + β_{i}^{S M B} S M B_{t} + β_{i}^{H M L} H M L_{t} + ε_{i, t}

(2)

Similar to Equation (1), market return timing ability is measured by the coefficient γ_i. The magnitude of γ_i in Equation (2) measures the difference between the target betas and is positive for a manager who successfully times the market.

2.2. Volatility timing

Market volatility timing shows how a fund manager reacts to changes in market volatility by adjusting a fund’s beta. The intuition underlying this view was first examined by Busse (1999), who finds that fund beta can be expressed as a linear function of the demeaned market volatility from a Taylor’s approximation. Thus, market volatility timing can be captured as follows:

R_{i, t} = α_{i} + β_{i}^{M K T} R_{m, t} + γ_{i} R_{m, t} (σ_{m, t} - {\bar{σ}}_{m}) + ε_{i, t}

(3)

R_{i, t} = α_{i} + β_{i}^{M K T} R_{m, t} + γ_{i} R_{m, t} (σ_{m, t} - {\bar{σ}}_{m}) + β_{i}^{S M B} S M B_{t} + β_{i}^{H M L} H M L_{t} + ε_{i, t}

(4)

where σ_m,t is market volatility, which is measured by realized volatility. This paper uses the realized volatility estimated using the exponential general autoregressive conditional heteroskedastic (EGARCH) model.⁶ In this specification, a negative value for coefficient γ_i indicates successful market volatility timing because it indicates decreasing fund beta when the market becomes more volatile.

2.3. Joint timing

A mutual fund manager can change market exposure based on the forecast of both market returns and volatility. This means that even though fund managers may expect a high market return in the near future, they may not increase the portfolio weight on equities without also considering market volatility. This is because they need to behave conservatively when high market volatility is expected. Another observation of industry practice is that managers may be restricted by fund mandates, which may prevent them from being able to increase portfolio weights on equities.⁷ Based on this, Chen and Liang (2007) propose the following joint timing measure:

R_{i, t} = α_{i} + β_{i}^{M K T} R_{m, t} + γ_{i} {(\frac{R_{m, t}}{σ_{m, t}})}^{2} + ε_{i, t}

(5)

R_{i, t} = α_{i} + β_{i}^{M K T} R_{m, t} + γ_{i} {(\frac{R_{m, t}}{σ_{m, t}})}^{2} + β_{i}^{S M B} S M B_{t} + β_{i}^{H M L} H M L_{t} + ε_{i, t}

(6)

where the coefficient γi measures the timing ability of a manager who can forecast both the level and volatility of the market portfolio. As mentioned by Chen and Liang (2007), this joint measure relates fund returns to the Sharpe ratio of the market portfolio. As also noted by these authors, if a fund adopts buy-and-hold strategies, the beta ( $β_{i}^{M K T}$ ) can capture the fund’s market exposure and the timing coefficient (γ_i) should be zero. However, the performance of a timing fund can be enhanced if the market Sharpe ratio is non-zero. Therefore, a timing fund can benefit by increasing its market exposure with the expected Sharpe ratio of the market portfolio. In other words, if the coefficient (γ_i) is positive, the fund can be considered to have successful joint timing ability.

2.4. Cross-sectional bootstrap approach

This section presents the cross-sectional bootstrap approach to differentiating skill from luck and derives the luck distribution. To do so, it adopts the same cross-sectional bootstrap approach as used by Kosowski et al. (2006, 2007), Cuthbertson et al. (2008), and Fama and French (2010). A primary focus of these bootstrap studies has been to examine whether or not mutual funds consistently obtain abnormal returns (measured by alpha). These studies find marked evidence of fund manager skill from their bootstrap simulations.

The bootstrap method of Kosowski et al. (2006) and Cuthbertson et al. (2008) uses residual-only resampling as follows. First, estimate a chosen factor model for each fund and save the vectors of ( ${\hat{β}}_{i}$ , ${\hat{ε}}_{i}$ ). Next, resample the estimated residuals, ${\hat{ε}}_{i}$ , with replacement during the fund’s existing period and use these bootstrapped residuals to generate a simulated excess return series under the null hypothesis of no outperformance (α_i = 0). Using this simulated return series, re-estimate the chosen factor model. An advantage of this approach is that the simulated number of fund returns is always the same as the original number of fund returns, while a possible cross-sectional dependence of fund returns can be neglected.

Fama and French (2010) take a somewhat different approach. They jointly sample all fund returns to account for the correlation of α estimates for different funds. They then generate the simulated returns by subtracting α_i from the bootstrapped return series of each fund to ensure that the simulated return series have no outperformance. However, as caveated by the authors themselves, a drawback of this sampling method is that if a fund does not exist for the entire sample period, it is likely to show up in a simulation run for more or less than the number of months it is in the sample. Fama and French (2010) conclude that their evidence from the bootstrap method attests to manager skill.

Our study adopts the stationary bootstrap method of Politis and Romano (1994) and the algorithm of Politis and White (2004). The advantage of the stationary bootstrap is in mimicking the original model by retaining the stationarity property of the original series in the resampled pseudo-time series (Politis and Romano, 1994) and allowing time-series dependence (Kosowski et al., 2007) in managed fund returns. Basically, the purpose of using the bootstrap is simply to find out whether or not the timing ability of each managed fund hinges on sample variability. We resample the vector of each fund return and other explanatory variables (market returns, small minus big (SMB), and high minus low (HML)) simultaneously, using the stationary bootstrap to consider the dependence structure of managed fund returns to other explanatory variables.

To identify whether or not timing ability is skill or luck, we derive the cross-sectional luck distribution under the assumption that fund managers do not have timing abilities. Following Kosowski et al. (2006, 2007), Cuthbertson et al. (2008), and Fama and French (2010), we use the t-value of γ_i instead of the timing coefficient, where γ_i controls for differences in precision due to differences in residual variance and in the number of days funds are in the bootstrap simulation (Fama and French, 2010). More precisely, our bootstrap simulation consists of the following three steps.

Step 1. Assume there are N managed funds. Estimate the chosen model for each fund over the fund’s existing period ( $T_{i}$ ) and save the timing coefficient, ${\hat{γ}}_{i}$ , using the following equation:

R_{i, t} = α_{i} + β_{i}^{M K T} R_{m, t} + {\hat{γ}}_{i} R_{m, t}^{2} + e_{i, t}

(7)

Step 2. Simulate returns for each managed fund for the period T_i so that they do not have timing ability, as follows:

{\tilde{R}}_{i, t} = R_{i, t} - γ_{i} R_{m, t}^{2}

(8)

This procedure is similar to Fama and French’s (2010). As can be seen in Equation (8), subtracting the timing term from the original managed fund returns implies that fund i has no timing ability. This step is then repeated for all funds. For each fund i, we resample the generated returns of each mutual fund ( ${\tilde{R}}_{i, t}$ ) and explanatory variables (i.e., in this case, only market returns) simultaneously, replacing the length T_i.

Step 3. Using simulated returns ${\tilde{R}}_{i, t}$ on each fund by the bootstrap simulation, we re-estimate Equation (7) to derive ${\tilde{γ}}_{i}$ (i = 1, 2, …, n), which represents sampling variation around a true value of zero by construction, due entirely to luck.

This three-step procedure is repeated S = 3000 times for each of N funds. This gives us a separate luck distribution ( $f ({\tilde{γ}}_{i})$ ) in timing ability, from the extreme best market timer to the extreme worst market timer. Suppose we are interested in determining whether the timing ability of each fund is due to a fund manager’s skill or luck. If the estimated ${\hat{γ}}_{i}$ is greater than the 5% upper tail cutoff point from the estimated distribution of simulated ${\tilde{γ}}_{i}$ , then we can reject the null hypothesis that the fund’s timing ability is due to luck at the 95% confidence level. However, our bootstrap analysis focuses on the luck distribution for the t-statistic of ${\hat{γ}}_{i}$ , $t_{{\hat{γ}}_{i}}$ because it has better statistical properties (for further discussion, see Cuthbertson et al., 2008; Hall, 1992; Kosowski et al., 2007).⁸ The cross-sectional bootstrap approach differs from that of earlier studies of the timing ability of managed funds in a crucial manner: As mentioned by Cuthbertson et al. (2008), we can compare the best timer with the distribution ( $f ({\tilde{γ}}_{i})$ ), which provides information about luck represented by all funds, and not just the luck distribution of the best market timer.

To determine the statistical significance of whether timing ability is due to luck or skill, we use p-values, as do Jiang et al. (2007). To estimate the p-value, let $f ({\hat{t}}_{γ}^{i})$ be the cross-sectional statistic (say, the 95% percentile) of the timing ability using the original time series and let $f ({\hat{t}}_{γ}^{i})$ be a particular cross-sectional statistic under the null hypothesis that there is no timing ability, which has been estimated by the bootstrap simulation. Therefore, the p-value can be estimated by comparing the two cross-sectional distributions as follows:

p = \frac{1}{N} \sum_{i = 1}^{N} I (f ({\hat{t}}_{γ}^{i}) > f ({\tilde{t}}_{γ}^{i}))

(9)

where $I (f ({\hat{t}}_{γ}^{i}) > f ({\tilde{t}}_{γ}^{i}))$ is an indicator that takes the value of one if $f ({\hat{t}}_{γ}^{i}) > f ({\tilde{t}}_{γ}^{i})$ , and zero otherwise, and N is set to 3000 in our study. In the case of market return timing, a high value of p (close to 1) implies that the estimated timing measure is consistently higher than its bootstrapped value, which is evidence of positive market return timing ability. However, for market volatility timing, a low value of p (close to zero) is evidence of negative timing ability.

3. Data and empirical results

3.1. Data and descriptive statistics

We consider 1751 Australian managed funds (1268 live funds and 483 defunct funds) obtained from the Morningstar open-end database.⁹ We also use fund names and investment objectives from the Morningstar database. Our sample covers the period January 1995 to December 2009.

To obtain our final sample, we apply several filters on the initial set of 5425 funds. Since we are interested in fund managers’ market volatility timing abilities, we restrict our attention to domestic managed funds, which are categorized by Morningstar as either Equity, Allocation, Fixed Income, or Alternative. From these categories, we remove all Fixed Income and Alternative funds.¹⁰ However, the Morningstar categories still include funds that are not invested in domestic equity. We eliminate these funds by searching for keywords, such as Asian, European, Japan, and global, and we also drop real estate funds. In addition, we require the funds to have a minimum of 24 monthly returns.

Table 1 reports the number of funds in live and defunct funds, their mean returns, and average life. As can be seen in Table 1, the mean return of live funds (0.695%) is slightly higher than that of defunct funds (0.634%). The average life of live funds is 96.9 months, while that of defunct funds is 68 months.

Table 1.

Summary statistics for Australian managed funds This table shows the basic statistics for Australian managed funds. The sample period ranges from January 1995 to December 2009. The data are from Morningstar.

	Number of funds	Mean return	Average life (months)
Live	1268	0.695	96.885
Defunct	483	0.634	68.468
Total	1751	0.665	82.677

3.2. Empirical results for timing ability

As noted in Section 2, we use three different measures to examine whether or not the timing ability of Australian managed funds is genuine. To reduce measurement bias and improve the timing specification, we adopt the three-factor model by including the Fama–French factors SMB and HML.¹¹

First, we examine the results of the one-factor model as in Equations (1), (3), and (5). The estimation results for the one-factor models are reported in Table 2, which shows the number of funds partitioned by the significance of their estimated timing coefficients, at the 95% level of significance. The table suggests that a segment of Australian managed funds have timing abilities. More specifically, of the full sample of 1751 funds, 562 have market return timing ability, 401 have market volatility timing ability, and 383 have joint timing ability.

Table 2.

Estimation results for timing coefficients using the one-factor model This table shows the number of funds divided by the significance of their estimated timing coefficients using the one-factor model. The sample period ranges from January 1995 to December 2009. The data are from Morningstar.

	Return timing	Volatility timing	Joint timing
Significantly positive	562	262	383
Positive	634	525	787
Negative	467	563	540
Significantly negative	88	401	41

Our purpose is to assess whether or not these timing abilities are due to fund manager skill or just luck. To this end, we apply the cross-sectional bootstrap approach to derive the luck distribution. To interpret the bootstrap results for the three timing measures, Figure 1 presents the kernel density estimates of the distributions of the t-values of γ in the original data (indicated by a dashed line) and the bootstrap simulated distributions under the null that there is no timing ability (indicated by a solid line). In Figure 1, the bootstrap simulated distribution shows how many funds are expected to achieve a given level of timing ability due to random chance (i.e., luck distribution). Comparing this simulated distribution with the distribution of the original data, we can judge whether a fund’s timing ability is real.

Figure 1.

Kernel density estimates of the original and simulated t-values using the one-factor model.

Figure 1(a) compares the distribution of the original data and the bootstrap simulated distribution of market return timing. Overall, the distribution of the original data lies largely to the right of the simulated distribution. For example, as shown in Table 3, the top 10th percentile of the actual t-value, 4.242, for market return timing is much more extreme than the average estimate from the simulation, 1.601. This result suggests there are some genuine market timers. Joint timing, presented in Figure 1(c), can be similarly interpreted. Figure 1(b), however, shows that the distribution of the original data lies partly to the left of the luck distribution, implying that some funds possess negative market volatility timing, consistent with Busse (1999) and Chen and Liang (2007). This result may suggest that Australian managed funds provide investors with a type of volatility hedge.

Table 3.

Simulation results using the one-factor modelThe reported t-values are the Newey–West adjusted t-values. To determine the significance of the timing coefficients, the 95% significance level is adopted throughout the study. The p-values are in parentheses and are calculated using Equation (9).

	Bottom t-statistics of timing coefficient			Top t-statistics of timingcoefficient
	1%	5%	10%	10%	5%	1%
Return
Actual	−3.125	−1.955	−1.320	4.242	5.140	7.069
Simulated	−3.580	−2.194	−1.608	1.601	2.256	3.906
	(0.991)	(0.998)	(1.000)	(1.000)	(1.000)	(1.000)
Volatility
Actual	5.325	3.147	2.446	−3.066	−3.803	−5.421
Simulated	4.247	2.398	1.695	−1.529	−2.135	−3.639
	(0.999)	(1.000)	(1.000)	(0.000)	(0.000)	(0.000)
Joint
Actual	−2.308	−1.495	−1.099	2.816	3.631	5.035
Simulated	−3.394	−2.104	−1.556	1.438	1.964	3.248
	(1.000)	(1.000)	(1.000)	(1.000)	(1.000)	(1.000)

Table 3 tabulates the bootstrap results (actual t-statistics, average simulated t-statistics, and their corresponding p-values) using the full set of Australian managed funds during the sample period with the one-factor model. Comparing the actual t-statistics of the timing coefficients with their luck distributions in Table 3, we find that top-ranked Australian managed funds’ timing ability is genuine and not due to luck. For example, top-ranked Australian managed funds have significantly positive market return timing coefficients and their p-values, calculated by Equation (9), are close to one, implying that the estimated timing measure is consistently higher than its bootstrapped value. From the high p-values of market return timing for top-ranked Australian managed funds, we conclude that the market return timing ability is genuine. Moreover, the poor market return timing ability of bottom-ranked funds is not likely due to luck either, which contrasts with what Chen and Liang (2007) document for hedge funds. These authors provide evidence that the negative timing of the bottom-ranked (hedge) funds may be attributed to randomness. This suggests that the market exposure of some Australian managed funds mistakenly decreases when the stock market improves. For joint timing, a similar pattern is observed in Table 3.

However, the result for market volatility timing should be interpreted with caution. Note that a negative value for the coefficient γ_i indicates successful market volatility timing, because it indicates decreasing fund beta when the market becomes more volatile. Therefore, a low value of p (close to zero) implies that the estimated volatility timing measure is consistently lower than its bootstrapped value, which is evidence of negative timing ability. The results for volatility timing in Table 3 show that the p-values of top-ranked Australian managed funds are close to zero, indicating that the market volatility timing ability is genuine and not from luck. Based on the results of the three timing measures, we conclude that the top-ranked Australian managed funds have genuine timing abilities.

Next, with regard to the three-factor model, we draw a plot for the kernel density estimates of the original and simulated results in Figure 2 that is very similar to Figure 1. That is, the distribution of the original data lies largely to the right of the simulated distribution for market return timing and joint timing, while the distribution of the original data for market volatility timing lies to the left of the luck distribution, indicating that some funds possess negative market volatility timing.

Figure 2.

Kernel density estimates of the original and simulated t-values using the three-factor model.

As for the one-factor model, Table 4 reports the number of funds with timing ability relative to the three-factor model. The results are similar to those of Table 2, which shows that a group of Australian managed funds have timing ability.

Table 4.

Estimation results for timing coefficients using the three-factor model This table shows the number of funds divided by the significance of their estimated timing coefficients using the three-factor model. The sample period ranges from January 1995 to December 2009. The data are from Morningstar.

	Return timing	Volatility timing	Joint timing
Significantly positive	560	243	435
Positive	603	530	742
Negative	498	561	521
Significantly negative	90	417	53

The bootstrap results for the three-factor model are presented in Table 5. As in Table 3, it reports the actual t-statistics, the average simulated t-statistics, and their corresponding p-values. By and large, the results are similar to those for the one-factor model; regarding market return timing, the results are the same as those for the one-factor model. The top-ranked Australian managed funds show significant timing ability (return, volatility, and joint timing).

Table 5.

Simulation results using the three-factor model The reported t-values are the Newey–West adjusted t-values. To determine the significance of the timing coefficients, the 95% significance level is adopted throughout the paper. The p-values are in parentheses and are calculated using Equation (9).

	Bottom t-statistics of timing coefficient			Top t-statistics of timing coefficient
	1%	5%	10%	10%	5%	1%
Return
Actual	−3.494	−1.978	−1.342	4.190	5.079	6.772
Simulated	−3.598	−2.197	−1.617	1.643	2.330	4.037
	(0.676)	(0.995)	(1.000)	(1.000)	(1.000)	(1.000)
Volatility
Actual	5.446	3.173	2.399	−3.144	−3.811	−5.617
Simulated	4.063	2.370	1.700	−1.551	−2.179	−3.748
	(1.000)	(1.000)	(1.000)	(0.000)	(0.000)	(0.000)
Joint
Actual	−2.755	−1.559	−1.119	2.952	3.716	5.189
Simulated	−3.445	−2.113	−1.561	1.500	2.059	3.419
	(1.000)	(1.000)	(1.000)	(1.000)	(1.000)	(1.000)

Overall, the present article offers new findings for the Australian context. As stated in the introduction, there is little evidence of market return timing for Australian managed funds according to past studies (e.g., Gallagher, 2001). Sawicki and Ong (2000) suggest perverse market return timing for Australian data, while Holmes and Faff (2004) find negative marketing timing for the majority of their sample funds. We contend that our evidence could be profound, because it indicates that some Australian funds do possess timing ability. However, caution is warranted to avoid overstating the evidence. The methodology of the present analysis differs fundamentally from that of past work. Existing studies predominantly ask how many funds show negative market timing coefficients and whether these change signs when volatility timing is accounted for. On the other hand, our principal interest is in examining whether individual timing coefficients are statistically significant to discern skill from luck by applying a bootstrap procedure. In short, readers are warned that the present evidence is not directly comparable to past documentation on Australian managed funds.

4. Concluding remarks

This paper studies the timing abilities of Australian managed funds using monthly data for the period from January 1995 to December 2009. We include live and defunct funds in our sample to limit survivorship bias, which arises if only the returns of live funds are adopted (see Brown et al., 1992; Elton et al., 1996). This study differs from previous studies, firstly, because we compare the luck distributions generated from the bootstrap approach with the original distributions and, secondly, because we do not examine the timing ability of individual funds but, rather, for the Australian managed fund sector as a whole. In addition, market return timing, market volatility timing, and joint timing abilities are all examined.

Our empirical results can be summarized as follows. Firstly, from the kernel density estimates for the one-factor and three-factor models, the distribution of the original data lies largely to the right of the simulated distribution for market return timing and joint timing, while the distribution of the original data for market volatility timing lies to the left of the luck distribution, implying that some funds possess negative market volatility timing. The results for market volatility timing are consistent with Busse (1999) and Chen and Liang (2007), and may suggest that Australian managed funds provide investors with a type of volatility hedge. Secondly, when we focus on the top-ranked Australian managed funds for three timing measures, we can conclude that top-ranked funds have genuine timing abilities. The poor timing abilities of bottom-ranked funds are not likely to be due to luck either.

Footnotes

Date of acceptance of final transcript: 18 October 2012.

Accepted by Associate Editor, Tom Smith (Finance).

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Notes

References

Australian Bureau of Statistics (2012) Time series workbook: 5655.0 Managed Funds, Australia.

Australian Trade Commission (2011) Australia’s managed funds assets now at A$1.8 trillion. Available at: http://www.austrade.gov.au/Invest/Investor-Updates/Data-Alert-110311-Australias-Managed-Funds-Assets-now-at-1-8-trillion (accessed 10 March 2011).

Brown

Goetzmann

Ibbotson

. (1992) Survivorship bias in performance studies. Review of Financial Studies 5: 553–580.

Busse

(1999) Volatility timing in mutual funds: Evidence from daily returns. Review of Financial Studies 12: 1009–1041.

Chen

Liang

(2007) Do market timing hedge funds time the market? Journal of Financial and Quantitative Analysis 42: 827–856.

Cuthbertson

Nitzsche

O’Sullivan

(2008) UK mutual fund performance: Skill or luck? Journal of Empirical Finance 15: 613–634.

Faff

Veeraraghavan

(2009) Do Australian hedge fund managers possess timing abilities? Applied Financial Economics 19: 27–38.

Elton

Gruber

Blake

(1996) Survivorship bias and mutual fund performance. Review of Financial Studies 9: 1097–10120.

Erskine

(2012) Hedge funds, private equity and alternative investment in Australia. In: Athanassiou

(ed.) Research Handbook on Hedge Funds, Private Equity and Alternative Investments. Cheltenham: Edward Elgar, pp. 406–407.

10.

Fama

French

(2010) Luck versus skill in the cross section of mutual funds. Journal of Finance 65: 1915–1947.

11.

Gallagher

(2001) Attribution of investment performance: An analysis of Australian pooled superannuation funds. Accounting and Finance 41: 41–62.

12.

Gallagher

Jarnecic

(2002) The performance of active Australian bond funds. Australian Journal of Management 27: 163–185.

13.

Gharghori

Mudumba

Veeraraghavan

(2007) How smart is money? An investigation into investor behaviour in the Australian managed fund industry. Pacific-Basin Finance Journal 15: 494–513.

14.

Goetzmann

Ingersoll

Ivkovic

(2000) Monthly measurement of daily timers. Journal of Financial and Quantitative Analysis 35: 257–290.

15.

Hall

(1992) The Bootstrap and Edgeworth Expansion. New York: Springer.

16.

Hallahan

Faff

(1999) An examination of Australian equity trusts for selectivity and market timing. Journal of Multinational Financial Management 9: 387–402.

17.

Heaney

Hallahan

Josev

. (2007) Time-changing alpha? The case of Australian international mutual funds. Australian Journal of Management 32: 95–112.

18.

Henriksson

Merton

(1981) On market timing and investment performance. II. Statistical procedures for evaluating forecasting skills. Journal of Business 54: 513–533.

19.

Holmes

Faff

(2004) Stability, asymmetry and seasonality of fund performance: An analysis of Australian multi-sector managed funds. Journal of Business Finance & Accounting 31: 539–578.

20.

Jensen

(1968) The performance of mutual funds in the period 1945–1964. Journal of Finance 23: 389–416.

21.

Jiang

Yao

(2007) Do mutual funds time the market? Evidence from portfolio holdings. Journal of Financial Economics 66: 724–758.

22.

Kosowski

Naik

Melvyn

(2007) Do hedge funds deliver alpha? A Bayesian and bootstrap analysis. Journal of Financial Economics 84: 229–264.

23.

Kosowski

Timmermann

Wermers

. (2006) Can mutual fund “stars” really picks stocks? New evidence from a bootstrap analysis. Journal of Finance 61: 2551–2595.

24.

Lee

(2012) Funds management (except superannuation funds) in Australia. IBIS World Industry Report K7514, Oct 2012, IBIS World.

25.

Politis

Romano

(1994) The stationary bootstrap. Journal of the American Statistical Association 89: 1303–1313.

26.

Politis

White

(2004) Automatic block-length selection for the dependent bootstrap. Econometric Reviews 23: 53–70.

27.

Prather

Middleton

Cusack

(2001) Are N+1 heads better than one? The timing and selectivity of Australian managed investment funds. Pacific-Basin Finance Journal 9: 379–400.

28.

Sawicki

Ong

(2000) Evaluating managed fund performance using conditional measures: Australian evidence. Pacific-Basin Finance Journal 8: 505–528.

29.

Sinclair

(1990) Market timing ability of pooled superannuation funds January 1981 to December 1987. Accounting and Finance 30: 51–65.

30.

Treynor

Mazuy

(1966) Can mutual funds outguess the market? Harvard Business Review 44: 131–136.

31.

Yap

Pierce

(2008) Managed fund attributes and performance: Australian evidence. In: PBFEAM 2008: Conference Proceedings and Abstracts, 16th Annual Conference on Pacific Basin Finance, Economics, Accounting and Management Conference, Brisbane, Australia.