Market Timing, Selectivity and Alpha Generation: Evidence from Australian Equity Superannuation Funds

Transcription

1 Market Tmng, Selectvty and Alpha Generaton: Evdence from Australan Equty Superannuaton Funds AUTHORS ARTICLE INFO JOURNAL FOUNDER Mchael E. Drew Madhu Veeraraghavan Vanessa Wlson Mchael E. Drew, Madhu Veeraraghavan and Vanessa Wlson (2005). Market Tmng, Selectvty and Alpha Generaton: Evdence from Australan Equty Superannuaton Funds. Investment Management and Fnancal Innovatons, 2(2) "Investment Management and Fnancal Innovatons" LLC Consultng Publshng Company Busness Perspectves NUMBER OF REFERENCES 0 NUMBER OF FIGURES 0 NUMBER OF TABLES 0 The author(s) Ths publcaton s an open access artcle. busnessperspectves.org

2 Investment Management and Fnancal Innovatons, 2/ Market Tmng, Selectvty and Alpha Generaton: Evdence from Australan Equty Superannuaton Funds Mchael E. Drew, Madhu Veeraraghavan, Vanessa Wlson Abstract In ths performance evaluaton study, two questons are addressed. Frst, do actve fund managers possess macro and mcro forecastng sklls that delver superor rsk-adjusted returns? Second, what s the nature of market tmng/stock selectvty trade off n the generaton of alpha? The answers from ths study are as follows: as an ndustry, managers delvered nferor returns for superannuaton nvestors for the perod 1991 through The study provdes lttle evdence that the Australan funds management ndustry holds suffcent macro and/or mcro forecastng abltes to generate postve alpha. Whle prevous research has found that nferor market tmng decsons are compensated for by superor stock selecton sklls, ths study fnds no substantve nverse relatonshp between tmng and selectvty. Key words: Superannuaton funds, Australa. JEL Classfcaton: G23; G15. Introducton The economc functon of penson funds s to facltate the transformaton of retrement savngs nto retrement ncome. The effcency wth whch fund managers execute ths transformatve functon has receved much attenton from both academcs and practtoners alke. The fnancal economcs lterature suggests that the performance of actvely managed funds, on average, has been nferor to that of a passvely managed alternatve 1. Ths has led researchers, such as Gruber (1996), to ask why do nvestors buy actvely managed mutual funds (p. 783)? Ths paper nvestgates ths queston by examnng the role of market tmng and stock selectvty n the generaton of alpha ( ) 2. In order for actve managers to generate alpha, fnancal markets (at least n the short-run) must be predctable. Two possble methods used by managers to create value for nvestors are: superor market tmng abltes (macro-forecastng); and/or, superor stock selecton (mcroforecastng). Treynor and Mazuy (1966) argued that f managers could tme the market they would hold a larger share of volatle (less volatle) securtes n a bull (bear) market. Moreover, the documentaton of anomales relatng to the sze (Banz, 1981) and value (Rosenberg, Red and Lansten, 1985) premum may provde an opportunty for actve managers to garner superor rskadjusted returns from mcro forecastng. The breakdown of manager performance nto macro and mcro-forecastng decsons, formalsed by Fama (1972), and the role of anomales n the prcng of rsk (Fama and French, 1996) have two mportant mplcatons for the evaluaton of manager performance. Frst, studes that fal to consder tmng and selectvty smultaneously could lead to erroneous conclusons beng made about the sources of alpha generaton 3. Second, multple factors (apart from the overall 1 See, for example, Gruber (1996), Sawck and Ong (2000) and Wermers (2000). 2 Followng Warwck (2000), an nvestment manager s sad to generate alpha ( ) under the followng crcumstances: alpha s generated f nvestment returns exceed an approprate benchmark, f the rsk taken to acheve the return s smlar to that of the benchmark; or, alpha s generated f managers returns are equvalent to an approprate benchmark, f the rsk taken to acheve the return s less than that of the benchmark. 3 Chen and Stockum (1986) note that the use of the tradtonal Sharpe-Lnter-Mossn (SLM) model s lkely to generate based results as t treats the systematc rsk level of a fund as a fxed coeffcent rather than as a decson varable. The nonstatonarty of a fund s systematc rsk volates the basc assumpton of an OLS regresson model. Grant (1977) contends that ths wll result n a downwardly based alpha coeffcent. Treynor (1966) also argues that such measures do not capture the share of fund varablty due to a lack of dversfcaton, resultng n the possblty that managers could mprove ther ratngs wthout mprovng the qualty of ther sklls va securty selecton by gvng up more dversfcaton benefts.

3 112 Investment Management and Fnancal Innovatons, 2/2005 market factor) such as frm sze and the rato of book-equty to market-equty are requred to explan the cross-secton of returns n an economcally meanngful manner. The poneerng contrbuton of Treynor and Mazuy (henceforth TM) (1966) assumed that portfolo returns would be a non-lnear functon of market returns. The TM quadratc-regresson model permtted researchers to nvestgate the behavour of systematc rsk decsons made by managers. A second parametrc test of selectvty and tmng, developed by Henrksson and Merton (henceforth HM) (1981), used a dfferent nterpretaton of market tmng ablty. In the sprt of TM, funds may alter portfolo composton subject to market movements, but HM also ncorporated the dea that managers can elect the level of market rsk. The up and down-market beta model of HM provdes a useful confrmatory measure to test tmng performance 1. Internatonal evdence reports that an nverse relatonshp exsts between market tmng and stock selecton. Fund managers typcally have negatve or perverse market tmng sklls when evaluated by the TM and HM models. However, the poor market tmng decsons of managers appear to be somewhat offset by superor stock selecton skll. Ths negatve relatonshp has been consstently demonstrated n US (Coggn, Fabozz and Rahman, 1993; and Bollen and Busse, 2001), UK (Fletcher, 1995) and nternatonal (Cumby and Glen, 1990) mutual and penson fund returns. However, for the Australan settng, the relatonshp between market tmng and stock selectvty sklls s less clear. The prelmnary fndngs of Snclar (1990) usng a sample of 16 pooled superannuaton funds over the perod 1981 through 1987 found that 5 out of the 16 funds exhbted sgnfcant rsk level changes, consstent wth the proposton that managers were attemptng to tme the market. Moreover, all funds showed sgnfcant negatve market tmng ablty and sgnfcant postve securty selecton ablty, wth tmng domnatng overall performance. The small sample sze nvestgated by Snclar (1990) and use of a fve-year sample perod (whch ncluded the 1987 stock market crash) motvated Hallahan and Faff (1999) to undertake a detaled nvestgaton of the tmng and selectvty sklls of 65 Australan equty trusts 2. Hallahan and Faff (1999) selected a post-crash observaton perod from 1988 through 1997 usng the TM and HM models 3. Unlke prevous research, the contrbuton of Hallahan and Faff (1999) provded two nterestng results: frst, some evdence of postve tmng coeffcents was evdent; and, second, negatve selectvty coeffcents (although not always sgnfcant) were obtaned by managers. Whle there remaned a negatve correlaton between the tmng and selectvty varables, Hallahan and Faff (1999) found that where there was evdence of market tmng ablty, ths was beng offset by poor stock selecton ablty. Sawck and Ong (2000) undertook a thrd major study consderng the behavour of marketng tmng and stock selectvty n Australa. Usng Ferson and Schadt s (1996) lagged nformaton framework, the study examned the performance of 97 domestc managed funds over the perod 1983 through Sawck and Ong (2000) reported that condtonal alphas were hgher and postve (wth the number of sgnfcant negatve market tmng coeffcents greatly reduced) under condtoned nformaton. At frst glance, the results appeared to provde support to the status quo of negatve market tmng and postve selectvty. However, Sawck and Ong (2000) noted that the mprovement n performance usng the condtonal model was counter-ntutve. The negatve covarance found between fund beta and market return (resultng n mproved performance) should have been controlled for by the condtonng nformaton. However, a negatve correlaton would mply fund managers reduced ther exposure to market movements when market returns were hgh and vce versa. Ferson and Warther (1996) suggest that the negatve correlaton may be partly the result of mutual fund cash flows. Internatonal evdence suggests that actve fund managers face a trade-off between tmng and selectvty n the generaton of alpha. Specfcally, ths nverse relatonshp has predomnantly taken the form of negatve tmng decsons beng somewhat compensated for by postve selectv- 1 Dybvg and Ross (1985) argue that the HM model only tests f the fund manager had access to specal nformaton. Moreover, the HM model s not as sophstcated as the model of Jensen (1968) as t does not forecast the magntude of the superor performance, only the drecton of the performance. 2 The sample nvestgated by Hallahan and Faff (1999) ncluded dversfed growth trusts (37), dversfed ncome equty trusts (8), property equty trusts (9), dversfed resources equty trusts (6), and other equty trusts (5). 3 Hallahan and Faff (1999) also employed the specfcaton tests of Jagannathan and Korajczyk (1986).

4 Investment Management and Fnancal Innovatons, 2/ ty sklls. However, the Australan settng provdes an nterestng case study, wth Hallahan and Faff (1999) reportng contrary fndngs. The controversy surroundng the relatonshp between tmng and selectvty n Australa has mportant mplcatons for the performance of Australan equty superannuaton funds. Superannuaton s currently the second most mportant asset for Australans (after the home), wth assets totallng AUD bllon n 2001 (APRA, 2001). The retal funds management ndustry, the focus of ths study, managed a total of AUD 205 bllon of these assets as at the end of fscal 2000 (APRA, 2001) 1. The prmary concern for the fund manager s to maxmse returns for ther consttuents, n ths case, the superannuaton fund member. In generatng returns on retrement savngs, t s mportant to ascertan the role of market tmng and stock selectvty. Moreover, gven the controversy surroundng the nature of the trade-off between tmng and selectvty (and the ssue of sample perod selecton) t s tme to undertake a further nvestgaton of the Australan settng. The rest of the paper s organsed as follows. Secton I outlnes the performance evaluaton models and tests of tmng and selectvty used n the study. Secton II descrbes the data. Secton III presents the emprcal analyss, wth Secton IV provdng concludng remarks. I. Tests of Tmng, Selectvty and Performance Fund manager skll s evaluated through an analyss of the alpha generated over a gven perod. Ths study s concerned ntally wth manager skll aganst benchmark performance and, prmarly, wth the ablty for actve managers to forecast market movements and to select undervalued stocks to create value for superannuaton nvestors. In achevng the research objectve, the analyss commences wth the excess return from a sngle ndex model. R t R ( R R ), (1) ft where I = rsk adjusted abnormal return from the sngle ndex model; R ft = return on the Reserve Bank of Australa 13 week T-note n month t; R mt = return on the Australan Stock Exchange Top 100 accumulaton ndex n month t; = factor senstvty of dfference n fund return and the rsk free rate; and, = random error term. Jensen s (1968) sngle ndex model, posts that the securty s return should be lnearly related to ts rsk, as measured by beta. The ntercept term detects whether managers have superor forecastng abltes, wth alpha generated by selectng securtes resultng n > 0 2. As Equaton (1) s a sngle-perod model, estmatng the regresson over tme should allow nvestors to have heterogeneous nvestment horzons. Furthermore, returns are assumed to be ndependently and dentcally dstrbuted (IID) through tme and jontly mult-varate normal. Recent advances n the asset prcng lterature suggest that sngle ndex models are unable to capture the cross-secton of expected stock returns, especally those anomales relatng to the sze and value premum, n an economcally meanngful manner. The selecton of factors to be ncluded n any mult ndex model has been a controversal subject n the evaluaton lterature. Gruber (1996) suggests that the selecton of factors should nclude ndces that span the major types of securtes held by funds, that falure to do so wll make performance estmates more a matter of how the excluded categores of stocks dd than how well management could select securtes (p.787). Ths study therefore employs a four-factor model, ncludng factors relatng to the market, sze, style and bond ndces of funds. Agan, t s assumed returns condtonal on factor realsatons are IID through tme and jontly mult-varate normal. mt ft 1 Superannuaton s the Commonwealth Government s preferred system for the provson of retrement savngs for Australans. The mportance of superannuaton for the real sector cannot be underestmated. Superannuaton s now the second most mportant asset (after the home) for Australans, wth an average aggregated superannuaton membershp balance of AUD 60, Alpha generaton wll be sgnfcantly postve f the fund manager has the ablty to forecast future securty prces. Alpha wll be zero f the manager mmcs the composton of a reference benchmark. Fnally, alpha wll be sgnfcantly negatve f the fnd manager performs worse than a nave strategy of random selecton.

5 114 Investment Management and Fnancal Innovatons, 2/2005 R t R ( R R ) ( R R ) ( R R ) ( R R ), (2) ft mt mt ft s where: R st R lt = the sze effect captured by the dfference n return between a small market captalsaton portfolo and a large market captalsaton portfolo based on Australan Stock Exchange Frank Russell Company ndces n month t; R vt Rg t = the value premum effect captured by the dfference n return between a value portfolo and a growth portfolo 1 based on the Australan Stock Exchange Frank Russell Company ndces n month t; R dt R ft = the bond ndces effect captured by the dfference n return on a bond ndex representng the Commonwealth, sem-government and corporate bonds wth all maturtes; k = factor senstvty of dfference n return on fund to portfolo j (whch represents the market, sze, or value premum effect); and = random error term. As wth the sngle ndex model, the manager s ablty to generate alpha s ndcated by a sgnfcant postve result. Whle general crtcsms can be made of sngle and mult ndex models, one specfc crtcsm s that both Equatons (1) and (2) assume that a fund s systematc rsk s statonary over tme. The nablty of such tests to ncorporate dynamc rsk strateges by managers may result n a regresson estmate of that may be sgnfcantly based downward (Grant, 1977; and Lee and Rahman, 1990). TM (1966) addressed ths concern wth the development of a quadratc market model. Through the addton of a quadratc term to Equaton (1), portfolo returns are a non-lnear functon of the market return. Ths provdes a measure of the tmng abltes of fund managers. st lt g gt vt d dt ft R t R mt R 2 mt, (3) where = rsk adjusted measure of market tmng ablty of fund. The followng two hypotheses are tested wth ths model: Hypothess I: H 0 : = 0 H a : 0 Hypothess II: H 0 : = 0 H a : 0 Hypothess I s concerned wth testng for the presence of abnormal performance as mentoned n the models specfed prevously, but s measured net of the manager s tmng ablty. Hypothess II s concerned wth measurng the market tmng ablty of fund managers. Market tmng ablty wll be reflected by greater market exposure when the excess market returns are hgher and vce versa. A sgnfcantly postve value of gamma would ndcate superor market tmng ablty. If gamma does not devate sgnfcantly from zero, the manager cannot outguess the market. If gamma s sgnfcantly negatve, there has been perverse market tmng undertaken by the manager. HM (1981) took an alternatve approach to the ncorporaton of tmng ablty nto the tradtonal sngle ndex model. They assumed managers could elect the level of market rsk they wshed to encounter, ncorporatng an up-market beta ( 1 ) and a down-market beta ( 1-2 ) n the analyss. R t 1 Rmt 2 DRmt, (4) 1 Ths s the same as a portfolo of hgh book-to-market equty frms mnus a portfolo of low book-to-market portfolo equty frms.

6 Investment Management and Fnancal Innovatons, 2/ where: I = rsk adjusted abnormal return from dual-beta model or the fund s selectvty ablty; D = dummy varable that takes on a value of 1 for months when R mt s negatve, and zero otherwse; and, 2 = rsk adjusted measure of market tmng ablty of fund. Lke the TM s quadratc market model, two hypotheses are tested wth ths model. Hypothess III: H 0 : = 0 H a : 0 Hypothess IV: H 0 : 2 = 1 H a : 2 0 Agan, Hypothess III consders superor selectvty sklls net of tmng ablty. Hypothess IV s concerned wth the market tmng abltes of fund managers, determnng whether a manager s down-market beta s sgnfcantly dfferent from the up-market beta. A successful market tmer wll have a down-market beta greater than the up-market beta, ( 1-2 ) > 1, therefore the resultant estmate for 2 s sgnfcantly postve. If the estmate s sgnfcantly negatve, perverse market tmng decsons have been undertaken. If the manager s actual 2 s not zero, deductons made from Jensen s (1968) basc model may be rendered nvald. Snclar (1990) explans that alpha from a sngle ndex model would be overstated when 2 s greater than zero (a superor market tmer) and understated when 2 s less than zero (an nferor market tmer). Pror to descrbng the sample nvestgated n ths study, two ssues regardng the estmaton technque undertaken n ths study are noteworthy. Frst, a correcton for heteroskedastcty was necessary for both the TM and HM models. Ths s due to the error term demonstratng condtonal heteroskedastcty as managers attempted to tme market movements 1. Ths occurs despte the assumpton of securty returns beng ndependent and dentcally dstrbuted through tme. To correct for ths, Breen, Jagannathan and Ofer (1986), and Lehman and Modest (1987) suggest the use of heteroskedastcty-consstent standard errors employed by Whte (1980), Hansen (1982), and Hseh (1983). All tests for sgnfcance n ths study wll be based on heteroskedastctyadjusted t-statstcs. Second, dagnostc tests (reported n Appendx I) reveal the problem of multcollnearty for both market tmng models. Collnearty of the regressors yelds mprecse parameter estmates, weakenng hypothess testng. Auxlary regressons demonstrated that the squared excess market return varable and the excess market return varable wth a dummy for the TM and HM models, respectvely, s an approxmate lnear combnaton of the excess market return varable. The output from auxlary regressons s provded n Appendx II. The F-statstcs have p-values of zero under both market proxes used n ths study and are less than the sgnfcance values of 5% and 10%. Therefore the null hypothess of no multcollnearty s rejected, wth the varables mentoned beng collnear wth the other explanatory varable 2. Therefore, the emprcal nvestgaton has taken steps to correct for multcollnearty, as estmates wthout ths correcton wll yeld spurous results. 1 Prevous studes have demonstrated that gnorng heteroskedastcty often leads to the rejecton of the null hypothess of no market tmng ablty too often when the null s n fact true, and vce versa. Although Henrksson (1981) found that adjustng for heteroskedastc error terms dd not alter ther results, studes by Breen, Jagannathan, and Ofer (1986), and Lee and Rahman (1990) suggest the exstence of non-homoskedastc resduals can sgnfcantly affect the power of tests for market tmng. Ths s the result of OLS estmates beng neffcent, as systematc rsk s non-statonary. 2 Appendx III shows that for the frst market proxy, the ASX Top 100 Accumulaton ndex, the squared excess market return varable and the excess market return varable wth a dummy change from beng sgnfcant varables when regressed aganst the excess fund return varable to nsgnfcant varables when regressed wth all explanatory varables. Ths s

7 116 Investment Management and Fnancal Innovatons, 2/2005 Chapman and Pearson (2000) demonstrated that the problem of multcollnearty resultng from a model takng a non-lnear functonal form s resolved by the technque of orthogonalsed polynomals 1. Ths s acheved by transformng the squared excess market return varable and the excess market return varable wth a dummy under both models, resultng n: The transformed TM model: R t Wth the transformed HM model takng the form: R p ( R ). (5) R t mt mt 1 Rmt 2 p ( Rmt. (6) ) The new regressor, p (R mt ), s formed as the regresson resdual of R mt 2 and DR mt (under separate equatons) onto a constant. Ths s then scaled to have a standard devaton equal to the standard devaton of the dependent varable, R t. These orthogonalsed and scaled monomals have the ncremental effect of addng the orgnal terms n Equatons (3) and (4). The parameter estmates from Equatons (5) and (6) wll be reported throughout the study. II. Data Mornngstar Research Pty Ltd provded return data on retal Superannuaton Funds Australan Equty General for the perod of January 1991 through December Fund returns were net of management expenses but excluded entry and ext loads. To mnmse the problem of survvorshp bas, all funds n exstence over the observaton perod were ntally consdered, ncludng all termnated funds. The only excluson from the sample were funds that dd not have at least 30 months of data avalable. The populaton conssted of 142 funds, wth 8 funds beng excluded as the result of n 30, resultng n a sample of 134 funds. The funds are separated nto three categores by Mornngstar: open-end; closed-end; and, non-survvng or termnated funds. Open-end funds, commonly referred to as unt trusts, may ssue or redeem addtonal unts of the fund at net asset value. The retal funds consdered n ths study requre a mnmum ntal nvestment of AUD 2,000 wth mnmum contrbutons of AUD 100. A total of 68 open-end funds are examned n ths study. Closed-end funds sell unts to nvestors only once, at the tme of offer. These funds do not ssue addtonal unts and may not redeem unts on demand. A lack of lqudty may prevent an nvestor from extng ths fund. However, the effect of large captal nflows and outflows from contrbutors s mnmal, gvng managers some control over the assets under management. A total of 55 closed-end funds s examned n ths study. The non-survvng cohort ncludes funds that ceased operatons over the observaton perod. Elton, Gruber and Blake (1996) suggest that fund attrton s the result of ether poor fund performance over a perod of tme or because the total market value of the fund s suffcently small that the management judges that t no longer pays to mantan the fund. The latter reason for closng a fund s assocated wth the former reason: poor performance. The excluson of non-survvng funds therefore results n an overestmaton of hstorcal returns 2. Over the observaton perod 13 retal funds were termnated, and are ncluded n the pool of funds consdered n the analyss. Admat and Pflederer (1997) advocate the use of a benchmark proxy that reflects each fund s nvestment strategy to augment the precson of performance evaluatons. One of the advantages of the sample nvestgated n ths study s that the asset allocaton parameters are known. To have membershp n the category, funds are requred to hold at least 80% of assets n a general portfolo of Australan equtes, wth a maxmum of 20% n domestc fxed nterest securtes. After an nvestgaton of the fund mandates, the Australan Stock Exchange (ASX) Top 100 Accumulaton usually seen to be evdence of multcollnearty. Although the confrmatory proxy, the ASX Top 20 Accumulaton ndex stll shows some sgnfcant results, ts sgnfcance has been reduced dramatcally. 1 See Draper and Smth (1998) for a detaled dscusson of orthogonalsed polynomals. 2 For estmates of survvorshp bas, see Grnblatt and Ttman (1989), Brown, Goetzmann, Ibbotson and Ross (1992), and Brown and Goetzmann (1994).

8 Investment Management and Fnancal Innovatons, 2/ ndex has been selected as the benchmark market proxy. Gven the large-captalsaton bas of managers, the ASX Top 20 Accumulaton ndex s used as a confrmatory proxy. If managers have engaged n some strategc behavour over the observaton perod (for nstance, attemptng to explot the sze and/or value premum), the mult ndex model s desgned to capture these affects. The ablty of actve managers to generate alpha through superor macro forecastng abltes (for example, swtchng between equtes and fxed nterest), and/or the ablty to explot any defcences n the arbtrage functon of markets through stock selecton, wll be captured by Equatons (5) and (6). III. Emprcal Analyss Actve fund managers engage n macro and/or mcro forecastng to generate alpha. Alpha generaton results n value beng created for clents, wth the manager garnerng returns that exceed the cost of becomng nformed. Hence, the economcs of actve management requres that the margnal beneft of actve management exceeded ts margnal costs. Intal studes of manager performance focussed on the ablty of managers to generate superor rsk-adjusted returns through actve stock selecton. The analyss of selectvty ablty commences wth the sngle ndex model of Equaton (1). Cohort Sngle Index Performance Estmates Table 1 Lsted are the average realsed monthly percentage returns and coeffcents, net of management expenses, from a pooled regresson of excess returns aganst the sngle ndex model of the form: R t R ft = + (R mt R ft ) + The adjusted R-squared and Durbn-Watson statstcs are also reported. The ASX Top 100 (Panel A) and ASX Top 20 (Panel B) Accumulaton ndces provde a proxy for benchmark returns. The sample conssts of 68 open-end funds, 55 closed-end funds, and 13 fnalsed funds (136 retal funds n total). Cohort Stand. Error Stand. Error R 2 adj DW Retal open-end Retal closed-end Retal non-survvng All retal funds Bass Ponts (per annum) Retal open-end Retal closed-end Retal non-survvng All retal funds Bass Ponts (per annum) Panel A: R mt = ASX Top 100 accumulaton ndex (t = -0.95) (t = -1.38) (t = -0.99) (t = -1.13) (t = -1.08) (t = -1.39) (t = -0.85) (t = -1.19) (t = 19.73) (t = 22.77) (t = 8.35) (t =19.87) Panel B: R mt = ASX Top 20 accumulaton ndex (t = 13.24) (t = 15.78) (t = 7.18) (t = 13.69) Table 1 shows that the average manager underperformed by % per month, equvalent to 210 (-283) bass ponts p.a., usng the broad ASX Top 100 (Top 20) Accumulaton as the

9 118 Investment Management and Fnancal Innovatons, 2/2005 reference rate. The negatve result was evdent across all cohorts, wth the retal non-survvng category (as expected) provdng the greatest level of alpha destructon. However, the cohort alphas are all nsgnfcant from zero at both the 5% and 10% levels. The funds management ndustry was unable to predct securty prces wth suffcent accuracy to outperform a nave strategy per unt of systematc rsk. Therefore, alpha generaton was not suffcent to recover management expenses. Table 2 provdes estmates of the sngle-factor model on an ndvdual fund bass. Of the 136 funds, 109 (94) funds were charactersed by alphas that were nsgnfcant from zero aganst the Top 100 ndex at the 5% (10%) sgnfcance level. In only 27 (42) cases there was evdence of abnormal performance at the 5% (10%) level. Out of the sgnfcant cases, only 1 (1) fund had a sgnfcantly postve alpha whle 26 (41) funds had sgnfcant negatve alphas 1. Indvdual Sngle Index Performance Estmates Table 2 Lsted are the number of alphas that are nsgnfcant (Zero), and sgnfcantly dfferent from zero (Postve/Negatve) from a sngle ndex model. Also reported are the number of betas nsgnfcant (Unty), sgnfcantly dfferent from unty (Greater than unty/less than unty). The results are consdered at the 5% and 10% sgnfcance levels for both market proxes. Fnally, the mean alpha and beta terms for the sample are also shown. Top 100 Top 20 Top 100 Top 20 Sgnfcance = Zero Zero Zero Total Mean Panel A: Sngle Index Alphas for Indvdual Funds 5 % % % % Sgnfcance = Unty Unty Unty Total Mean Panel B: Sngle Index Betas for Indvdual Funds 5 % % % % The evdence provded n Table 2 rases some mportant concerns regardng the overall level of systematc rsk adopted by managers. On an ndvdual fund bass, only 3 funds reveal a beta estmate that s nsgnfcant from unty at the 5% level. Only 8 funds exhbted beta estmates that were sgnfcantly greater than unty usng the Top 100 ndex, wth 6 funds demonstratng ths characterstc aganst the Top 20 ndex. Pror to drawng strong conclusons from the estmaton of Equaton (1), t s mportant to acknowledge that the model has a number of lmtatons. These nclude a heavy relance on the ablty for equty rsk premum to capture fund return and Roll s (1977, 1978) concerns regardng the observablty of the market portfolo. Recent advances n the asset prcng lterature, led by Fama and French (1996), have allowed researchers to resolve some of these concerns. Table 3 provdes estmates of the mult ndex model of Equaton (2). 1 However, the hypothess that fund alphas are jontly equal to zero cannot be rejected at the 95% and 90% confdence levels as more than half of the sample had alphas that were not sgnfcantly dfferent from zero.

10 Investment Management and Fnancal Innovatons, 2/ Cohort Mult Index Performance Estmates Table 3 Lsted are the average realsed monthly percentage returns, from a pooled regresson of excess percentage returns aganst the mult ndex model of the form: Rt Rft = + mt (Rmt Rft) + s (Rst Rlt) + g (Rgt Rvt) + d (Rdt Rft) + The ntercept,, s a measure of selectvty ablty and the coeffcents s, g, and d represent the sze, style and bond portfolos respectvely. Cohort s g d R 2 adj DW Retal open-end Retal closed-end Retal nonsurvvng All retal funds Bass Ponts (per annum) Retal open-end Retal closed-end Retal nonsurvvng All retal funds Bass Ponts (per annum) (t = 0.11) (t = -0.63) (t = -0.71) (t = -0.29) (t = -0.11) (t = -0.68) (t = -0.58) (t = -0.39) - 77 Panel A: R mt = ASX Top 100 accumulaton ndex (t = 19.62) (t = 23.43) (t = 7.99) (t = 20.06) (t = 2.57) (t = 2.68) (t = 2.47) (t = 2.61) (t = 1.04) (t = -0.02) (t = 0.29) (t = 0.54) Panel B: R mt = ASX Top 20 accumulaton ndex (t = 15.68) (t = 19.56) (t = 7.68) (t = 16.49) (t = 3.76) (t = 4.38) (t = 3.15) (t = 3.96) (t = 0.55) (t = -0.19) (t = 0.40) (t = 0.24) (t = 2.63) (t = 3.65) (t = 3.66) (t = 3.14) (t = 2.08) (t = 2.88) (t = 3.62) (t = 2.55) The analyss presented n Table 3 provdes corroboratng evdence of the results from the sngle ndex model, wth one notable excepton. As an ndustry, funds underperformed the market by a range of 46 ( 77) bass ponts per annum, usng the ASX Top 100 (Top 20) accumulaton ndex as a proxy for passve returns. However, unlke the prevous analyss, all cohorts do not destroy value for nvestors. Managers n the retal open-end category generated a postve alpha of around 34 bass ponts per annum (net of the cost of becomng nformed) aganst the Top 100 ndex 1. Whle ths result s encouragng for those nvestors sklled enough (or lucky enough) to hold only survvng funds over the observaton perod, a complete dscusson of performance requres acknowledgment of any survvorshp bas. Incluson of termnated funds reverses ths result, confrmng that any analyss gnorng fund termnaton wll dramatcally overstate the ndustry s stock selectvty sklls. Fnally, all reported alpha estmates are nsgnfcant from zero at both 5% and 10% levels, so the hypothess that alpha s equal to zero cannot be rejected. 1 However, ths result reverses when the ASX Top 20 Accumulaton ndex s used as the market proxy, wth the ndustry destroyng around 30 bass ponts n value per annum. Ths hghlghts, out of sample, the concerns of Lehmann and Modest (1987) regardng the ssue of benchmark selecton n performance evaluaton.

11 120 Investment Management and Fnancal Innovatons, 2/2005 Indvdual Mult Index Performance Estmates Table 4 Lsted are the number of alphas that are nsgnfcant (Zero), and sgnfcantly dfferent from zero (Postve/Negatve) from a mult ndex model. Also reported are the number of betas nsgnfcant (Unty), sgnfcantly dfferent from unty (Greater than unty/less than unty). The results are consdered at the 5% and 10% sgnfcance levels for both market proxes. Fnally, the mean alpha and beta terms for the sample are also shown. Top 100 Top 20 Top 100 Top 20 Sgnfcance = Zero Zero Zero Total Mean Panel A: Mult Index Alphas for Indvdual Funds 5 % % % % Sgnfcance = Unty Unty Unty Total Mean Panel B: Mult Index Betas for Indvdual Funds 5 % % % % Turnng to the role of the explanatory varables n Equaton (2), the average fund beta ranged from 0.78 to 0.80, mplyng that managers adopted a portfolo composton less rsky than the general market. These values are sgnfcant at both the 5% and 10% levels under both market proxes. The ndvdual fund analyss ndcated that 133 (134) betas were sgnfcant at the 5% (10%) level aganst the Top 100 ndex. To provde some gude to the range of systematc rsk decsons made by managers, the hghest beta recorded beta by any fund was 1.32 (Top 100 ndex), wth the lowest estmated at The sze factor, s, s postve and hghly sgnfcant under both market proxes, rangng from 0.17 to The results provde some support to the noton that managers are attemptng to explot the sze anomaly. Moreover, the evdence hghlghts the lmtatons of usng sngle ndex models of Equaton (1) to analyse the cross-secton of fund returns. The regresson coeffcent relatng to sze was postve and sgnfcant for 88 (94) of the funds at the 5% (10%) level under the Top 100 ndex. The style factor, g, s postve but nsgnfcant under both market proxes sgnfcance levels. Ths s n contrast to the fndngs of Gruber (1996) who reported a sgnfcant negatve coeffcent for the style varable. Only 39 (48) funds were statstcally sgnfcant at the 5% (10%) level under the Top 100 ndex. In further contrast to Gruber s (1996) results from US mutual funds, 32 (33) were postve estmates. Ths rases some mportant ssues for superannuaton nvestors regardng the consstency of the manager s asset selecton style (value or growth). Moreover, research across a varety of nternatonal markets reports the exstence of a sweet spot for nvestors who hold small captalsaton stocks wth a value bas 1. Such opportuntes appear to reman largely unexploted by the majorty of managers nvestgated n ths study. Turnng to the fnal explanatory varable, the bond factor ( d ), plays an mportant role n capturng the return behavour of managers. Managers appear to be nvestng n a portfolo that has a sgnfcant holdng of returns provded by less volatle, fxed nterest securtes. Movng to the ndvdual fund level, 98 (113) funds had sgnfcant coeffcents at the 5% (10%) level under the Top 100 proxy. Moreover, these estmates were all postve. We now turn to another mportant 1 Ths fndng s confrmed by Fama and French (1996) for the US stock market, Hallwell, Heaney, and Sawck (1999) for Australa, and Drew and Veeraraghavan (2001) for the emergng markets n Asa.

12 Investment Management and Fnancal Innovatons, 2/ aspect of fund performance market tmng. The ssue often examned s how successful have funds been n tmng market movements and more mportantly how s tmng measured? Underlyng the estmates from Equatons (1) and (2) s the assumpton that a funds systematc rsk remans constant over tme. Relaxng ths assumpton, thereby decomposng the role of market tmng from stock selecton, requres regresson models to take a non-lnear form. The evaluaton of market tmng commences wth the augmented TM model of Equaton (5). The TM approach adds a quadratc term to Equaton (1) n an attempt to ncorporate the dynamsm of a fund s systematc rsk. TM Performance Estmates Table 5 Lsted are the average realsed monthly percentage returns, net of expenses, from a pooled regresson of excess percentage returns aganst TM s quadratc market model of the form: R t R ft = + (R mt R ft ) + (R mt R ft ) 2 + e I The ASX Top 100 (Panel A) and ASX Top 20 (Panel B) accumulaton ndces are the proxy for benchmark returns. The ntercept term, measures selectvty skll wth the coeffcent measurng market tmng ablty. Cohort I I R 2 DW Retal open-end Retal closed-end Retal non-survvng All retal funds Bass Ponts (per annum) Retal open-end Retal closed-end Retal non-survvng All retal funds Bass Ponts (per annum) Panel A: R mt = ASX Top 100 accumulaton ndex (t = -0.95) (t = -1.40) (t = -1.16) (t = -1.15) (t = 19.97) (t = 24.62) (t = 8.84) (t = 20.79) (t = -1.12) (t = -1.36) (t = -1.16) (t = -1.22) Panel B: R mt = ASX Top 20 accumulaton ndex (t = -1.11) (t = -1.48) (t = -1.05) (t = -1.26) (t = 14.36) (t = 18.17) (t = 7.78) (t = 15.27) (t = -2.30) (t = -2.59) (t = -1.41) (t = -2.33) The results reported n Table 5 further support the concluson that on average, actve managers have lmted stock selecton abltes, consstent wth the fndngs of Hallahan and Faff (1999) for Australan unt trusts. However, ths s contrary to the fndngs of accretve selecton sklls by Coggn, Fabozz and Rahman (1993), Fletcher (1995) and Bello and Janjgan (1997) amongst others, and Sawck and Ong (2000) for Australa. The alpha coeffcents reported n Table 5 usng both market proxes are slghtly greater compared to the coeffcents obtaned from the sngle-factor model shown n Table 1, a result consstent wth Grant s (1977) contenton that sngle ndex performance measures wll have a downward bas f market tmng effects are gnored.

13 122 Investment Management and Fnancal Innovatons, 2/2005 A total of 28 (45) funds exhbted sgnfcant abnormal performance aganst the Top 100 ndex at the 5% (10%) level. Of the funds wth sgnfcant alphas, just one generated a postve alpha, wth the remanng 27 managers recordng negatve performance. Wth approxmately threequarters of the sample portrayng no sgnfcant evdence of selecton ablty (and those wth a sgnfcant result demonstratng poor selectvty sklls), the null hypothess of no stock selecton skll across the ndustry s not rejected. The beta estmates provded n Table 5 renforce the fndng that Australan superannuaton funds took relatvely low rsks over the sample perod. However, the lnear combnaton of an ndustry beta less than unty combned wth negatve selectvty estmates volates the receved fndng of a perverse relatonshp between tmng and selectvty. Specfcally, ths volates the wdely cted fndngs of Black, Jensen, and Scholes (1972), and Kon and Jen (1978) who provde evdence that low-rsk portfolo compostons tend to result n postve selectvty performance estmates and vce versa 1. Interestngly, over 90% of the funds that exhbted sgnfcant negatve selectvty ablty had systematc rsk levels less than unty. Ths rases the ssue of whether the ndustry s nferor performance was a result of generally poor mcro forecastng ablty by managers or some, as yet undentfed, features of portfolo composton. Equaton (5) provdes gamma estmates as a measure of market tmng ablty. Aganst both market proxes, all cohorts showed negatve macro forecastng skll. The problem of benchmark selecton s agan hghlghted, wth all results beng nsgnfcant at the 5% and 10% levels aganst the Top 100, but sgnfcant at the both levels for the Top 20 ndex. Hence, t agan appears that selecton of the market proxy may sgnfcantly alter performance evaluaton results. However, some consensus s reached wth the estmated regresson coeffcents from both proxes beng negatve, leadng to a broad non-rejecton of the null hypothess 2. The evdence lsted n Table 5 fnds both support and controversy wth prevous studes examnng the Australan market. The fndngs are consstent wth prelmnary fndngs of perverse market tmng by Snclar (1990) and the recent contrbuton of Sawck and Ong (2000). However, the results do not reflect Hallahan and Faff s results, of lmted, manly postve market tmng sklls of the managers of Australan unt trusts. Indvdual fund gammas were sgnfcant n 31 (51) cases at the 5% (10%) level under the Top 100 proxy. All of the estmated coeffcents were sgnfcantly negatve under both market proxes, equatng to approxmately 23% (38%) of the sample havng sgnfcant perverse market tmng abltes aganst Top 100 (Top 20). The mplcaton of ths result s that around a quarter of managers tend to ncrease (decrease) ther equty exposure as the market falls (rses). To sum t up, we do not reject the null of hypothess of no tmng ablty for the ndustry at both sgnfcance levels. Fnally, t s mportant to comment on the nature of the relatonshp between tmng and selectvty. Aganst the Top 100 ndex, there s a negatve, but very low correlaton found between the two coeffcents, wth a reported value of The Top 20 ndex corroborates ths result, wth an estmated value of These low values would support Lehman and Modest s (1987) fndng of no substantve correlaton. Ths s n contrast to a large body of research reportng a large negatve correlaton between tmng and selectvty (Henrksson, 1984; Chang and Lewellen, 1984; Connor and Korajcyzk, 1991; Fletcher, 1995; and Bello and Janjgan, 1997). These results are also n contrast to prevous Australan studes of Snclar (1990), Hallahan and Faff (1999) and Sawck and Ong (2000). Therefore, gven these controversal fndngs, we turn to an alternatve market-tmng model, the HM model to further nvestgate the valdty of the results. 1 Moreover, Kon and Jen (1978) note that engagng n portfolo re-balancng (to mantan a target beta) wll result n a transacton cost bas n favour of low beta portfolos. 2 Ths poor performance s also evdent where funds havng the same selectvty and tmng coeffcents under TM generate an alpha of (-0.247) at the 5% (10%) level under the Top 100 ndex. 3 Ths s sgnfcant at the 5% and 10% levels wth a t-statstcs of

14 Investment Management and Fnancal Innovatons, 2/ HM Performance Estmates Table 6 Lsted are the average realsed monthly percentage returns, net of expenses, from a pooled regresson of excess percentage returns aganst HM s dual-beta model of the form: R t R ft = + 1 (R mt R ft ) + 2 D (R mt R ft ) + e I The ASX Top 100 (Panel A) and ASX Top 20 (Panel B) accumulaton ndces are the proxy for benchmark returns. The ntercept term, s a measure of selectvty ablty, and the coeffcents 1, and ( 1-2 ) represent the up and down-market betas respectvely. A successful market tmer wll have a down-market beta greater than the up-market beta. Therefore, 2 should be sgnfcantly postve for a successful market tmer. Cohort 1 2I R 2 DW Retal open-end (t = -0.98) Retal closed-end (t = -1.40) Retal non-survvng (t = -1.01) All retal funds Bass Ponts (per annum) Panel A: R mt = ASX Top 100 accumulaton ndex (t = -1.51) -217 Retal open-end (t = -1.18) Retal closed-end (t = -1.49) Retal non-survvng (t = -0.94) All retal funds Bass Ponts (per annum) (t = 19.85) (t = 23.97) (t = 8.94) (t = 20.48) (t = -0.72) (t = -0.94) (t = -0.88) (t = -0.82) Panel B: R mt = ASX Top 20 accumulaton ndex (t = -1.28) (t = 14.28) (t = 17.67) (t = 7.79) (t = 15.03) (t = -2.05) (t = -2.20) (t = -1.25) (t = -2.04) Smlar to the results reported earler n ths study, the returns are largely negatve but nsgnfcant from zero at the 5% and 10% levels. Agan, one fund generated a sgnfcant postve alpha for nvestors (the same fund found usng the TM technque) wth 28 funds havng sgnfcant estmates under the Top 100 ndex. As wth all of the tests undertaken n ths study, we do not reject the null hypothess of no selectvty ablty. The trend of managers havng a relatvely low rsk appette s confrmed by usng Equaton (6). A total of 7 (6) funds had beta coeffcents sgnfcantly greater than unty aganst the Top 100 (Top 20), at both the 5% and 10% levels. Ths small group of relatvely hgh-rsk funds has remaned the same for all of the models estmated. The beta 2 results from Table 6 provde a measure of market tmng ablty. Successful market tmng by the manager results n postve beta 2 coeffcent beng sgnfcantly dfferent from zero. As an ndustry, the estmates corroborate the result of no tmng ablty usng both market proxes. Movng to an ndvdual fund analyss, 8 (26) funds showed sgnfcant tmng skll at the 5% (10%) level usng the Top 100 ndex 1. Fnally, the HM measure confrms a margnal nega- 1 Ths poor performance s also evdent where the number of funds havng the same selectvty and tmng coeffcents under HM generate an alpha of (-0.227) at the 5% (10%) level under the Top 100 proxy.

15 124 Investment Management and Fnancal Innovatons, 2/2005 tve relatonshp between macro and mcro forecastng ablty, wth a correlaton of usng the Top 100, and aganst the Top 20 ndex. The nternatonal and domestc phenomenon of sgnfcant but perverse market tmng decsons, typcally offset by sgnfcant accretve stock selecton skll, s not supported n ths study. Ths study provdes lttle evdence that the Australan funds management ndustry has suffcent macro and/or mcro forecastng abltes to generate postve alpha for superannuaton nvestors. IV. Concludng Remarks The performance evaluaton lterature has long debated the relatonshp between macro and mcro forecastng skll. Elton and Gruber (1995) pont out that the acceptance of modern portfolo theory has changed the evaluaton process from crude calculatons to detaled explanatons of rsk and return. Whle the evdence has largely found that nferor market tmng decsons are compensated for by superor stock selecton sklls, the recent contrbuton by Hallahan and Faff (1999) reported contrary behavour for Australa. However, regardless of whether tmng sklls are postve or negatve, prevous studes have reported the exstence of a sgnfcant negatve relatonshp wth a manager s selecton ablty. The consensus vew posts a stuaton where managers face a trade-off between tmng and selectvty when generatng alpha. As Australa moves towards a full choce regme n superannuaton, permttng nvestors to move superannuaton savngs wth freedom among dfferent nvestment alternatves, the fndngs of ths study queston the ablty for the funds management ndustry to create value for consttuents through alpha generaton. Moreover, the fndngs of lmted macro or mcro forecastng skll queston the ongong role of dedcated, sngle-sector funds (such as the actvely managed Australan equty funds nvestgated n ths study) n the transformaton of retrement savngs nto retrement ncome. One legtmate economc functon for such ntermedares would be to provde a dversfcaton faclty for unt-holders, n ths case, superannuaton contrbutors. The study has been unsuccessful n unfyng the lterature regardng the tmng and selectvty performance of the Australan funds management ndustry by Snclar (1990), Hallahan and Faff (1999) and Sawck and Ong (2000). Prevous research shows that the relatonshp between tmng/selectvty trade-off, s a negatve one. The fndngs of ths study are consstent wth the work of Lehman and Modest (1987), n that no substantve nverse relatonshp exsts between tmng and selectvty. The evdence suggests that managers dd not try to sgnfcantly alter ther systematc rsk levels, resultng n largely fxed coeffcents over the sample perod. Our fndngs are also consstent wth Fabozz and Francs s (1979) observaton that managers n the US do not try to alter ther systematc rsk to explot market movements. Possble reasons offered as to why ths phenomenon occurs nclude the manager s nablty to tme market movements and/or costs of such tmng beng prohbtve. It s our conjecture that two mportant areas lead the agenda for future research on the role of tmng and selectvty n the generaton of alpha. The frst area relates to the defnton of style for Australan superannuaton fund managers. An nvestgaton nto the style of asset selecton decsons made by the Australan funds management ndustry s requred to permt a deeper understandng of alpha that s consumed by poor selectvty. A frutful area of research may relate to the ssue of style drft over the observaton perod. Second, future research may consder the ncorporaton of mult ndex models wth tradtonal selectvty and tmng technology, n a condtonal settng, to further decompose alpha generaton by managers. These ssues are left to future work. References 1. Admat, A.R., and S.A. Pflederer, 1997, Does It All Add Up? Benchmarks and the Compensaton of Actve Portfolo Managers, Journal of Busness, Vol. 70, pp Banz, R.W., 1981, The Relaton between Return and Market Value of Common Stocks, Journal of Fnancal Economcs, Vol. 9, pp Ths s sgnfcant at the 5% and 10% levels wth a t-statstcs of