Attribution of investment performance: An analysis of Australian pooled superannuation funds *

Transcription

1 Publshed n Accountng and Fnance, Vol. 41(1&2): pp41-62, 2001 Attrbuton of nvestment performance: An analyss of Australan pooled superannuaton funds * Davd R. Gallagher School of Busness, The Unversty of Sydney, Sydney, NSW 2006, Australa Abstract Ths paper evaluates the market tmng and securty selecton capabltes of Australan pooled superannuaton funds over the eght-year perod from January 1991 to December Evaluaton of both components of nvestment performance s surprsngly scarce n the Australan lterature despte actve nvestment managers engagng n both market tmng and securty selecton. The paper also evaluates performance for the three largest asset classes wthn dversfed superannuaton funds and ther contrbuton to overall portfolo return. The mportance of an accurately specfed market portfolo proxy n the measurement of nvestment performance s demonstrated. Consstent wth pror U.S. lterature, the emprcal results ndcate that funds n general do not exhbt securty selecton or market tmng skll. Keywords: Performance attrbuton; Market tmng; Securty selecton; Benchmarks JEL classfcaton: G2; G23 * The author s grateful to Dave Allen, Ron Brd, Bruce Grundy (Assocate Edtor), Amela Hll, Garry Hobbes, Crag Mellare, Vctor Souck and two anonymous referees for helpful comments and suggestons on earler drafts of ths paper. The author thanks Towers Perrn Australa for the provson of the unque fund performance data and nformaton used n the paper. Towers Perrn provded the data on the understandng that both the funds and nvestment managers would reman anonymous. 1

2 1. INTRODUCTION The performance of nvestment managers has long been of nterest to practtoners and nvestors, and n academa the performance evaluaton lterature spans at least four decades. Indeed the debate wthn ndustry between actve and passve nvestment management contnues, despte the overwhelmng emprcal evdence that actve funds, on average, do not earn superor rsk-adjusted returns. Ths can perhaps be consdered as paradoxcal when consderaton s gven to the relatve magntude of assets actvely managed n Australa. Ranmaker Informaton reports the sze of the nvestment ndustry at December 1999 was around $A632 bllon, of whch the overwhelmng majorty (approxmately 88.9 percent) of funds were actvely nvested. 1 In lght of the actve versus passve debate, ths paper evaluates the market tmng and securty selecton components of abnormal returns earned by actve Australan pooled superannuaton funds n the perod Most performance evaluaton studes have employed the Jensen (1968) approach where rsk-adjusted performance measures the ablty of funds to outperform the market (Jensen, 1972; Lee and Rahman, 1990). However, the Jensen Measure gnores the potental market tmng strateges employed by actve portfolo managers as the model does not partton the qualty of nformaton a manager holds from the aggressveness of the nvestment strategy. Indeed, actve nvestment managers commonly dstngush between both market tmng and stock selecton performance n the context of ther nvestment objectves. Therefore, performance evaluaton models gnorng market tmng strateges assume that rsk levels for managed funds reman statonary through tme, causng the estmate of abnormal return to be downward based where market tmng ablty s present (Dybvg and Ross 1985 and Grnblatt and Ttman 1989a). As a result, models that fal to measure market tmng and securty selecton smultaneously could lead to naccurate nferences beng made concernng the source of portfolo performance. Accordngly, ths paper evaluates both components of nvestment performance tmng and selectvty. Emprcal evdence n the U.S. wdely documents that actve funds do not outperform the market (for example Jensen, 1968; Grnblatt and Ttman 1989b; Elton, Gruber, Das and Hlavka 1993; Malkel 1995; and Gruber 1996). The lterature also confrms that funds do 1 Ranmaker Informaton Roundup, December Quarter

3 not successfully tme the market (Treynor and Mazuy 1966; Kon 1983; Chang and Lewellen 1984; Henrksson 1984; Lee and Rahman 1990; Coggn, Fabozz and Rahman 1993; Ferson and Schadt 1996; Danel, Grnblatt, Ttman and Wermers 1997; and Becker, Ferson, Myers and Schll 1999). Australan research supports the U.S. evdence that funds do not earn sgnfcantly postve rsk-adjusted returns attrbutable to securty selecton (Brd, Chn and McCrae 1983; Robson 1986 and Gallagher 1999). Snclar (1990) was the frst Australan paper to evaluate both market tmng and securty selecton performance, fndng that adverse market tmng by funds eroded the gans attrbutable to stock selecton. More recently, Hallahan and Faff (1999) examned selectvty and tmng ablty of Australan equty trusts, documentng that lttle evdence exsted to support the vew that such funds were successful market tmers. Sawck and Ong (2000) also document the nablty of funds to outperform market ndces where a condtonal performance evaluaton methodology was adopted. The paper makes the followng contrbutons to the Australan performance evaluaton lterature. Frst, the market tmng and securty selecton abltes of actve pooled superannuaton funds are evaluated at both the total portfolo level and across the three largest asset classes that comprse dversfed superannuaton portfolos; namely Australan equtes, nternatonal equtes and Australan fxed nterest. Second, the paper demonstrates the mportance of usng correctly specfed benchmarks n the measurement of performance where funds also hold non-australan equty assets n ther portfolos. Snclar s (1990) fndng that pooled superannuaton funds exhbt both postve and sgnfcant selectvty skll coupled wth sgnfcantly poor tmng s shown to arse when the market portfolo proxy s msspecfed. The potental bas n performance measurement where neffcent benchmark proxes are used s also evaluated. Fnally, the study utlses a unque data set comprsng pooled superannuaton fund asset allocatons relatve to strategc benchmark weghts and the performance of funds across ndvdual asset classes. Ths detaled level of nformaton provdes nsght nto the tactcal nvestment strateges that fund managers have used n ther quest for actve returns. The remander of ths paper s structured as follows. Secton 2 outlnes the emprcal tests for market tmng and securty selecton. Secton 3 descrbes the data and ths s followed by the emprcal results. The fnal secton concludes and provdes suggestons for future research. 3

4 2. EMPIRICAL FRAMEWORK 2.1 Rsk-Adjusted Performance Evaluaton Models Securty selecton represents the ablty of an nvestment manager to dentfy and explot msprced securtes (mcro forecastng). On the other hand, market tmng represents the ablty of portfolo managers to poston ther portfolos to take advantage of predcted market movements (macro forecastng). Successful market tmng occurs when portfolo rsk s ncreased n antcpaton of market rses. Extendng Jensen s (1968) model (based on the CAPM framework), Henrksson and Merton (1981) decompose performance nto selectvty and tmng as follows: R pt = α p + βp 1 xt + βp 2yt + εpt (1) Where: R pt = the portfolo return n perod t n excess of the rsk free return; α p = the abnormal return attrbutable to securty selecton; β p2 = the coeffcent estmatng tmng ablty; x t = the market return n excess of the rsk free rate n perod t; y t = max[ 0, x t ] ε pt = the random error term wth expected mean of zero. The term (β p2 ) s used by Henrksson and Merton (1981) to capture the market tmng component of nvestment performance followng Jensen (1972), Grant (1977), Dybvg and Ross (1985) and Grnblatt and Ttman s (1989a) demonstraton of potental bas n the estmates. These authors suggest that funds attemptng to market tme wll bas (β p1 ) upward and the abnormal return (α p ) wll be based downward f market tmng (β p2 ) s gnored. The Henrksson-Merton model assumes fund managers target two systematc rsk levels; one where the manager forecasts the rskless asset to outperform the market portfolo (β p1 ) and the other where the market return s expected to outperform the rsk-free 4

5 rate (β p2 ). 2 Successful market tmng exsts where the estmate of β p2 n (1) s sgnfcantly postve. The model does not predct the magntude of the return dfferental between rsky assets and the rskless asset, but rather consders the drecton of the forecast that a portfolo manager uses to re-weght the portfolo between rsky assets and the rskless asset. 3 An alternatve test for market tmng ablty s the Treynor-Mazuy model. Treynor and Mazuy (1966) propose the use of a quadratc term to capture market tmng ablty (compared wth Henrksson-Merton s β p2 measure), argung that funds wth market tmng ablty wll hold a greater (smaller) proporton of ther portfolos n the market portfolo of rsky assets when they expect the market to rse (fall). The Treynor-Mazuy approach ndcates successful market tmng where the coeffcent γ s sgnfcantly postve. R pt = α + β x + γ x + ε p p t p 2 t pt (2) Gven the Henrksson-Merton and Treynor-Mazuy models both rely on the CAPM framework, emprcal tests usng these models assume the market portfolo proxy s meanvarance effcent. Roll s (1977, 1978) crtcsms of the CAPM are well documented n the lterature. Dybvg and Ross (1985) also warn of the potental dangers of an neffcent market portfolo proxy, where abnormal returns reflect these neffcences rather than beng derved usng superor nvestment skll. For example, Grnold (1992) found n tests of benchmark effcency that the Australan All Ordnares Index s ex-ante neffcent. Fnn and Kovurnne (2000) also fnd evdence of benchmark neffcency for Australan stock market ndces. Measurng actve performance relatve to a passve benchmark ndex that s ndependent of prvate nformaton and mean-varance neffcent can overstate performance. Admat and Pflederer (1997) suggest an alternatve benchmark proxy that 2 Henrksson and Merton s (1981) β p2 accounts for market tmng on the bass of a fund manager engagng n a protectve put opton nvestment strategy. See Henrksson and Merton (1981) for a detaled descrpton. 3 The Henrksson-Merton model requres correctons for heteroskedastcty and ths paper employs Whte s (1980) method of adjustment. The ordnary least squares estmates n the model are neffcent gven systematc rsk s not statonary. Henrksson and Merton (1981) show that the standard devaton of the error term s an ncreasng functon of the absolute value of x t. Whle Henrksson (1984) found that adjustments for heteroskedastcty dd not affect the general conclusons made, other studes ncludng Breen, Jagannathan and Ofer (1986) and Lee and Rahman (1990) suggest that the presence of non-homoskedastc resduals sgnfcantly affects the power of tests for market tmng. Breen, Jagannathan and Ofer (1986) fnd that gnorng heteroskedastcty often leads to rejecton of the null hypothess for no market tmng too often when n fact the null s true. The converse s also the case. The Treynor-Mazuy model also requres correctons for heteroskedastcty (Coggn, Fabozz and Rahman, 1993). 5

6 employs the average return earned by managed funds as a group may allevate some of the problems of benchmark neffcency n performance evaluaton studes, and ths approach s consdered n the emprcal results secton. 2.2 Performance Attrbuton Framework Performance attrbuton measures the effect of the portfolo manager s actve nvestment decsons across asset sectors and ther respectve contrbuton to portfolo performance (Burne, Knowles and Teder, 1998). The monthly average asset allocatons for each fund across each asset class wthn the portfolo are used, where the attrbuton framework decomposes the raw actve return (fund return less return of the benchmark) nto securty selecton and market tmng components. 4 Attrbuton of nvestment performance can be performed usng ether an arthmetc approach (Karnosky and Snger (1995) and Snger, Gonzalo and Lederman (1998)) or geometrc approach (Burne, Knowles and Teder (1998)). In terms of the arthmetc approach, the methodology assumes the fund manager s portfolo management objectve s to outperform usng both top-down and bottom-up nvestment strateges. Whle ths assumpton has mert, gven managers are lkely to use elements of both styles, the attrbuton framework above leads to the necessty of a resdual term whch s potentally ambguous (see Karnosky and Snger (1995) and Snger, Gonzalo and Lederman (1998)). In order to elmnate ths resdual or nteracton term, Burne, Knowles and Teder (1998) develop a geometrc approach to decompose the actve return nto securty selecton and market tmng components only Top-Down Portfolo Management Ths geometrc methodology assumes fund managers prortse ther portfolo management strateges between top-down and bottom-up styles, thereby renderng the resdual term obsolete. 5 Top-down portfolo management assumes that nvestment managers prmary emphass s asset allocaton whereas the bottom-up strategy dentfes securty selecton as takng precedence. The top-down asset allocaton component (4) measures the portfolo 4 Ths study evaluates the components of performance n sngle currency terms. Where the portfolo manager makes actve decsons wth respect to currences, addtonal terms must be added to the attrbuton framework. 6

7 manager s ablty to underweght or overweght the asset classes wthn the portfolo relatve to each fund s unque strategc benchmark. The securty selecton component (5) for a top-down portfolo manager measures the stock selecton effect usng the portfolo s actual asset class weghts. The total portfolo s actve return (Tot), and the two components of total performance for a top-down nvestment strategy, asset allocaton (R a ) and securty selecton (R s ), are represented geometrcally: Tot [( 1+ R )(1 + R )] 1 (3) t = at st R at = (1 + (1 + r ω r ) b ) 1 (4) R st = Where: (1 + (1 + rp ) 1 ω r ) ω = average actual weght n asset class ; (5) ω = benchmark weght n asset class ; r = return earned by the fund n asset class ; r p = fund return for the total portfolo; r = benchmark return representng a passve nvestment strategy n asset class ; r b = benchmark return for the total portfolo. The ndvdual asset class contrbutons for a top-down portfolo manager can be expressed geometrcally as: R at (1 + r ) = ( ω )[ 1] (1 + r ) ω (6) b 5 The goal of parttonng managers on the bass of predomnant style used s amed at elmnatng the nteracton effect or resdual term. However, the dchotomy may appear overly smplstc, as some managers may not see themselves as clearly belongng to a sngle group, but a mxture of the two. 7

8 R st = ω ( r (1 + r ) ω r ) (7) Bottom-Up Portfolo Management Portfolo management decsons that are predomnantly bottom-up assume stock pckng s of hgher prorty than asset allocaton. Gven that managers select securtes across asset classes on the bass of fundamental value, bottom-up strateges are not lmted by asset allocaton weghts n the portfolo. Accordngly, the securty selecton component for a bottom-up portfolo manager reles on a fund s benchmark weght n each of the asset classes. The bottom-up asset allocaton component measures the mpact of the portfolo s actual asset allocaton dvergence from the strategc benchmark based on the fund s portfolo returns rather than the performance of the benchmark. The bottom-up attrbuton framework at the total portfolo level, geometrcally, can be represented as: Tot [( 1+ R )(1 + R )] 1 (8) t = at st R at = (1 + (1 + r p ) 1 r ) ω (9) R st = (1 + (1 + r ω r ) b ) 1 (10) The ndvdual asset class contrbutons for a bottom-up portfolo manager can be expressed geometrcally as: R at (1 + r ) = ( ω )[ 1] (1 + ω r ) ω (11) 8

9 R st ( r r ) = ω (12) (1 + r ) b The performance methodology outlned above s used to evaluate the extent to whch fund managers exhbt superor market tmng and securty selecton sklls wth reference to ther predomnant portfolo management strategy (top-down versus bottom-up), ndvdual asset allocaton decsons, strategc benchmarks and portfolo returns. 3. DATA Ths study uses monthly Australan pooled superannuaton fund returns for 16 average and above average volatlty funds usng a unque data set provded by Towers Perrn Australa. The Towers Perrn Pooled Superannuaton Funds database montors fund performance across the entre Australan market and therefore a representaton of fund manager performance. 6 Funds comprsng the sample were ncluded where Towers Perrn had complete hstorcal nformaton concernng performance, asset allocatons and strategc benchmark weghts provded by the nvestment managers over the entre 96-month perod. Towers Perrn classfes pooled superannuaton funds on the bass of hstorc volatlty n fund returns as well as fund nvestment style. Two of the funds n the sample (denoted fund A and B) are managed by the same nvestment organsaton. Fund B has therefore been removed from Table 4 n the results secton reportng the sector performances. 7 The perod of evaluaton s the 8-year perod January 1991-December The total assets under management for these 16 funds at December 1998 was around $A29.9 bllon and nvestment performance s reported before management fees and tax. The market ndces, 6 Whle the sample sze s relatvely small compared wth U.S. studes, the Australan market s consderably smaller coupled wth these types of funds (superannuaton) beng relatvely unque. Gven the crtera for ncludng funds, a number of funds were not ncluded as they were ether (1) not n exstence at January 1991 and/or (2) dd not have suffcent data (returns and asset allocatons) to perform the analyss over the entre 8- year perod. Therefore 10 funds (accountng for $A5.7 bllon at December 1998 or 16 percent of the total elgble market sze) could not be ncluded, prmarly on the bass of not havng exsted for the entre 8-year perod beng evaluated (.e. they were younger funds). Another vald pont concernng the sample sze s due to Australan fund managers (generally) not offerng multple pooled superannuaton vehcles to nvestors (whch may be contrary to sector specalst funds). Overall, these factors contrbute to the relatvely small number of funds ncluded n the study. 7 Whle funds A and B have dentcal sector performances n Australan equtes, nternatonal equtes and Australan fxed nterest, these funds have dfferent nvestment objectves. These nclude dfferent weghts across nvestment sectors and dfferent total fund returns. 9

10 outlned n Table 1, represent passve nvestment strateges across each asset sector. 8 rsk free rate used n the study s the 13-week Treasury note converted to a monthly rate. The <<INSERT TABLE 1>> The Towers Perrn Pooled Superannuaton Funds database ncludes monthly fund performance across ndvdual sectors and the total portfolo. 9 Average asset allocatons of each fund and across each month are also recorded, whch allows nferences to be made concernng the asset allocaton postons of nvestment managers relatve to each fund s unque strategc benchmark. The nvestment managers provde these strategc benchmark weghts for each of ther pooled funds to asset consultng frms such as Towers Perrn n order to better understand the nvestment strategy. 10 Strategc benchmarks are generally fxed across tme and represent a fund s long term nvestment objectve. Over the shortterm, managers may adopt strateges of under or overweghtng fund asset allocatons relatve to ther own strategc benchmark n an attempt to enhance portfolo performance. The funds ncluded n the sample are also classfed, where possble, accordng to the two dstnct nvestment management styles top-down and bottom-up. The parttonng of funds was performed based on nformaton provded to Towers Perrn by the fund managers. Half of the funds n the sample predomnantly used top-down strateges, 6 funds managed ther portfolos usng a bottom-up approach. 4. EMPIRICAL RESULTS 4.1 Overall Portfolo Performance 8 These market proxes are the most commonly used/cted ndexes n the Australan nvestment ndustry. 9 The sample group of superannuaton funds n the study contans the standard survvorshp bas problems faced by most performance evaluaton studes n the lterature, where funds ncluded n the sample reman n exstence at the end-date of the performance evaluaton perod. Studes ncludng Brown, Goetzmann, Ibbotson and Ross (1992) and Elton, Gruber and Blake (1996) hghlght the problems performance evaluaton studes face where survvorshp bas exsts. The extent to whch the results n the paper are based s not known, however, analyss of Towers Perrn s hstorcal performance surveys ndcate that t s lkely to be small. Gven the major source of bas generally arses due to poor performng funds havng hgher attrton probabltes, survvor based studes are lkely to postvely overstate performance than may otherwse be the case. 10 These ndependent strategc benchmark weghts provded by the nvestment managers have been used n the attrbuton analyss performed below. 10

11 The emprcal results derved from both the Henrksson-Merton and Treynor-Mazuy models and presented n Table 2 do not support the hypothess that funds collectvely have securty selecton or market tmng skll at the total fund level. Panel A of Table 2 (employng the Henrksson-Merton approach) reveals that a majorty of funds exhbt securty selecton and market tmng coeffcents nsgnfcant from zero. Three funds have selectvty estmates sgnfcantly dfferent from zero, where two funds are sgnfcantly postve. Approxmately half of the funds record negatve stock selecton estmates. The market tmng performance of funds provdes even greater evdence of an nablty by fund managers to outperform. The results show that whle a sgnfcant majorty of funds (15 out of 16) have nsgnfcant tmng coeffcents, the majorty of funds (11 out of 16) have negatve β p2 estmates. Further, the soltary fund exhbtng sgnfcantly postve market tmng underperforms n securty selecton. Panel B of Table 2, reports the securty selecton and market tmng estmates usng the Treynor-Mazuy approach, and the fndngs are largely consstent wth those n Panel A. <<INSERT TABLE 2>> An nterestng fndng documented n Table 2 s exstence of strong negatve correlaton (cross-sectonal) between selectvty and tmng estmates. 11 Around two thrds of funds exhbt ether postve selectvty coupled wth negatve tmng or postve tmng and negatve securty selecton coeffcents. Both the Pearson (-0.635) and Spearman (-0.435) correlaton coeffcents are sgnfcant at the 0.01 and 0.10 levels respectvely. Other studes, ncludng Henrksson (1984) and Coggn, Fabozz and Rahman (1993) fnd evdence of a strong negatve relatonshp between tmng and selectvty, ndcatng that perceved skll n one component of portfolo management actvty does not necessarly mply skll n the other. Henrksson (1984) hypothesses the exstence of a negatve relatonshp due to the market proxy beng msspecfed or the model omttng relevant factors explanng the dervaton of fund returns. Whle the former argument may appear to have lttle mert n ths study, due to the tests for tmng and selectvty relyng on the use of a more approprate benchmark, the ssue remans an emprcal queston. An alternatve possblty drvng the phenomena may be due to omtted rsk factors. Jagannathan and Korajczyk (1986) suggest the negatve correlaton between tmng and selectvty may 11

12 occur as a result of portfolo managers holdng optons or opton-lke securtes such as lsted securtes wth hgh leverage. Coggn, Fabozz and Rahman (1993) ndcate that the phenomenon of a negatve relatonshp between tmng and securty selecton s derved due to samplng errors of the two estmates beng negatvely correlated. However, whle not reported drectly n ths paper, evdence of negatve correlaton (tme seres) between tmng and selectvty s statstcally weak when consderaton s gven to the geometrc performance attrbuton approach. One problem of testng ths phenomenon n ths paper wth sgnfcant rgor s the lmtaton of only havng a small number of funds (where the sample sze s restrcted due to such funds beng unque). Indeed, future research s warranted concernng the contradcton s results concernng negatve correlaton between tmng and selectvty that encompasses a much larger sample of funds and across multple asset classes. 4.2 Senstvty of Performance to Choce of Benchmark Prevous performance evaluaton studes n both Australa and the U.S. have reled on the use of an equty market proxy as the benchmark, even where funds have non-equty assets as some proporton of the total portfolo. Henrksson (1984) states the use of such a benchmark s a suffcent market proxy where fund performance s hghly correlated wth the true market proxy. However, n response to Ippolto s (1989) concluson that U.S. mutual funds earned suffcent rsk-adjusted returns to recover expenses, Elton, Gruber, Das and Hlavka (1993) demonstrate that performance can be senstve to the choce of benchmark used. These authors show that Ippolto s (1989) results were due to the benchmark proxy excludng the performance of non-s&p 500 securtes. In vew of Elton et al. s fndng (1993), performance n ths study s also analysed usng the All Ordnares Accumulaton Index as the market proxy (followng Snclar s (1990) method) to evaluate the extent of possble bas generated for pooled superannuaton funds. 12 As outlned n Table 1, pooled superannuaton funds, on average, have less than 40 percent of ther strategc benchmark allocatons to the Australan equtes asset class. 11 Smlar to the results n Table 2, Table 4 also shows a strong negatve relatonshp (cross-sectonal) between tmng and selectvty estmates across Australan equtes, nternatonal equtes and Australan fxed nterest. 12 Fund returns n the sample, on average, had a correlaton coeffcent of 0.92 wth the All Ordnares Accumulaton Index. Ths compares wth a correlaton coeffcent of nvestment performance relatve to each fund s specfc strategc benchmark asset allocaton of approxmately

13 Snclar (1990) reports that 15 of the 16 funds examned n the perod exhbted sgnfcantly postve securty selecton estmates at 0.05 level under the Henrksson- Merton model. In contrast to the results presented n Table 2, Table 3 clearly demonstrates the problems that arse where a benchmark s used for dversfed funds that gnores other asset class exposures beyond Australan equtes provdes. Funds n the sample exhbt sgnfcantly hgher securty selecton estmates whle smultaneously recordng sgnfcantly worse market tmng. Whle the results n both Table 2 (Panel A) and Table 3 provde consstent evdence that funds do not exhbt superor tmng ablty under the Henrksson-Merton approach, the use of an equty market proxy overstates both pooled superannuaton funds poor tmng ablty and successful securty selecton. These fndngs support Elton, Gruber, Das and Hlavka s (1993) correcton of Ippolto s (1989) fndng that mutual funds outperform. <<INSERT TABLE 3>> 4.3 Performance of Indvdual Sectors Analyss of the performance of funds n Australan equtes, nternatonal equtes and Australan fxed nterest sectors was evaluated over the 8-year perod to December 1998 usng both the Henrksson-Merton and Treynor-Mazuy models. 13 Table 4 presents the results usng the Henrksson-Merton model, agan documentng the nablty of funds to outperform the relevant market ndces. 14 The overwhelmng majorty of funds exhbt postve securty selecton estmates n Australan equtes (6 funds sgnfcant) and Australan fxed nterest sectors. However around three-quarters of funds n Australan equtes (2 funds sgnfcant) have negatve tmng coeffcents. In Australan fxed nterest 14 of 15 funds (1 fund sgnfcant) record negatve tmng estmates. Internatonal equtes performance on the bass of securty selecton s the worst across all sectors, however only 13 The results derved usng the Treynor-Mazuy model were consstent wth the Henrksson-Merton and consequently are not drectly reported. 14 Tests for market tmng and selectvty were also performed to assess the potental bas n results arsng from benchmark neffcency followng the approach outlned by Admat and Pflederer (1997). These alternatve market proxes are more dffcult yardstcks for funds to outperform as they represent the average performance of potentally nformed nvestment managers. The securty selecton estmates were generally lower across all sectors for all funds and ndependent of the model used. Overall, the results ndcated that funds do not exhbt superor selectvty or tmng skll. 13

14 one manager recorded sgnfcantly negatve selectvty. Market tmng ablty n the nternatonal shares sector s shown to be non-exstent. <<INSERT TABLE 4>> 4.4 Geometrc Performance Attrbuton An alternatve test for securty selecton and market tmng ablty used n ths paper reles on a performance attrbuton methodology decomposng the actve raw return (not adjusted for rsk) n the perod across asset sectors gven the actve decsons employed by nvestment managers. The results n Table 5 ndcate 2 funds have postve and sgnfcant actve returns at the total fund level, and only one fund s successful n both tmng and stock selecton. The emprcal results across the ndvdual asset classes also ndcate the majorty of funds dd not exhbt superor performance. 15 Stock selecton n Australan equtes was generally the most successful asset class for the funds n the sample, however no evdence exsts of superor market tmng ablty. Four of the fve funds wth sgnfcantly postve selecton record postve tmng however none are statstcally sgnfcant. Fund performance n nternatonal equtes and Australan fxed nterest also supports the general fndng that funds overall do not outperform and therefore tmng or selecton skll beng absent. In nternatonal equtes, 12 funds have negatve mean securty selecton values (4 sgnfcant) and 8 of the 14 funds exhbt negatve tmng. Lttle evdence supports collectve tmng and selecton skll by managers n the Australan fxed nterest sector. Analyss of the performance of funds predomnantly top-down or bottomup does not ndcate that funds exhbt superor skll n asset allocaton or stock selecton respectvely. <<INSERT TABLE 5>> Further tests of performance are contaned n Table 6 evaluatng the consstency of tmng and selecton skll for pooled superannuaton funds. Analyss of the number of perods 15 Performance attrbuton was also performed usng an arthmetc approach whch assumes nvestment managers emphasse both securty selecton and market tmng. The results were consstent wth the evdence presented usng the geometrc performance attrbuton approach. Further, only 14 funds are evaluated as a result of 2 fund managers (C and M) not beng easly parttoned nto top-down or bottom-up styles. 14

15 (months) where nvestment managers make correct forecasts, rather than the magntude of the forecasts, provdes nformaton regardng the relatve success of the portfolo management process over tme. Hypothess tests are conducted over the 96-month perod to dentfy the ablty of nvestment managers to make correct forecasts. The null hypothess assumes the proporton of successful forecasts made by portfolo managers equates to 50 percent (H 0 : p=0.5). Rejecton of the null hypothess concludes the portfolo manager exhbts evdence of postve skll where the proporton exceeds 0.5 for both market tmng and stock selecton (H 1 : p 0.5). In Australan equtes, 5 funds record postve securty selecton sgnfcantly greater than 50 percent of months and 5 funds show sgnfcant consstency n market tmng forecasts n the Australan fxed nterest sector. However, the results provde further evdence that funds collectvely dd not exhbt successful securty selecton or tmng sklls. <<INSERT TABLE 6>> 5. SUMMARY AND SUGGESTIONS FOR FUTURE RESEARCH Ths paper evaluates the market tmng and securty selecton capabltes of Australan pooled superannuaton funds. The emprcal evdence ndcates that funds dd not exhbt superor selectvty or tmng skll at the total portfolo level, Australan equtes, nternatonal equtes and Australan fxed nterest and confrms the fndngs of prevous studes that funds do not outperform approprate market ndces. Whle funds are generally more successful n ther securty selecton strateges than market tmng, both components of performance do not provde nvestors wth both postve and statstcally sgnfcant rskadjusted performance. An nterestng fndng s the strong negatve cross-sectonal correlaton between selectvty and tmng usng both the Henrksson-Merton and Treynor- Mazuy models, supportng pror U.S. studes, however the phenomena s not supported usng the geometrc performance attrbuton methodology. The negatve correlaton phenomena requres further research, usng an expanded data set and alternatve evaluaton models. The paper also demonstrates the mportance of usng approprate benchmarks that are consstent wth the nvestment strateges and assets held n dversfed portfolos such as pooled superannuaton funds. Snclar s (1990) fndng that funds exhbt superor securty 15

16 selecton skll and sgnfcantly perverse tmng s shown to arse through the use of a msspecfed market proxy that excludes assets other than Australan equtes. Alternatve benchmarks reflectng each fund s unque nvestment strategy leads to more accurate nferences concernng portfolo performance. An extenson of ths research should nclude an nvestgaton of the market tmng and stock selecton capabltes of funds usng a condtonal performance evaluaton framework that accounts for publc nformaton and tme varaton n rsk. REFERENCES Admat, A., Pflederer, P. (1997), Does t all add up? Benchmarks and the compensaton of actve portfolo managers, Journal of Busness 70(3), Becker, C., Ferson, W., Myers, D., Schll, M. (1999), Condtonal market tmng wth benchmark nvestors, Journal of Fnancal Economcs 52, Brd, R., Chn, H., McCrae, M. (1983), The performance of Australan superannuaton funds, Australan Journal of Management 8, Breen, W., Jagannathan, R., Ofer, A. (1986), Correctng for heteroskedastcty n tests for market tmng ablty, Journal of Busness 59(4), Brown, S., Goetzmann, W., Ibbotson, R., Ross, S. (1992), Survvorshp bas n performance studes, Revew of Fnancal Studes 2(4), Burne, J., Knowles, J., Teder, T. (1998), Arthmetc and geometrc attrbuton, Journal of Performance Measurement, Fall, Chang, E., Llewellen, W. (1984), Market tmng and mutual fund nvestment performance, Journal of Busness 57(1), Coggn, T., Fabozz, F., Rahman, S. (1993), The nvestment performance of U.S.. equty penson fund managers: An emprcal nvestgaton, Journal of Fnance 48(3), Danel, K., Grnblatt, M., Ttman, S., Wermers, R. (1997), Measurng mutual fund performance wth characterstcbased benchmarks, Journal of Fnance 52(3), Dybvg, P., Ross, S. (1985), Dfferental nformaton and performance measurement usng a securty market lne, Journal of Fnance 40(2), Elton, E., Gruber, M., Das, S., Hlavka, M. (1993), Effcency wth costly nformaton: A renterpretaton of evdence from managed portfolos, Revew of Fnancal Studes 6(1), 1-22 Elton, E., Gruber, M., Blake, C. (1996b), Survvorshp bas and mutual fund performance, Revew of Fnancal Studes 9(4), Ferson, W., Schadt, R. (1996), Measurng fund strategy and performance n changng economc condtons, Journal of Fnance 51(2), Fnn, F., Kovurnne, T. (2000), The ex-ante effcency of Australan stock market benchmarks, Australan Journal of Management 25(1), 1-15 Gallagher, D. (1999), The performance of actvely managed Australan equty funds, Unpublshed paper, Unversty of Sydney, NSW 2006 Grant, D. (1977), Portfolo performance and the cost of tmng decsons, Journal of Fnance 32(2), Grnblatt, M., Ttman, S. (1989a), Portfolo performance evaluaton: old ssues and new nsghts, Revew of Fnancal Studes 2(3), Grnblatt, M., Ttman, S. (1989b), Mutual fund performance: an analyss of quarterly portfolo holdngs, Journal of Busness 62(3), Grnold, R. (1992), Are benchmark portfolos effcent?, Journal of Portfolo Management, Fall, Hallahan, T., Faff, R. (1999), An examnaton of Australan equty trusts for selectvty and market tmng performance, Journal of Multnatonal Fnancal Management 9, Henrksson, R., Merton, R. (1981), On market tmng and nvestment performance. II. Statstcal procedures for evaluatng forecastng sklls, Journal of Busness 54(4), Henrksson, R. (1984), Market tmng and nvestment performance: An emprcal nvestgaton, Journal of Busness 57(1),

17 Ippolto, R. (1989), Effcency wth costly nformaton: a study of mutual fund performance, Quarterly Journal of Economcs 104, 1-23 Jagannathan, R., Korajczyk, R. (1986), Assessng the market tmng performance of managed portfolos, Journal of Busness 59(2), Jensen, M. (1968), The performance of mutual funds n the perod , Journal of Fnance 23, Jensen, M. (1972), Optmal utlzaton of market forecasts and the evaluaton of nvestment performance, publshed n Szego, G. & Shell, K., Mathematcal Methods n Investment and Fnance, Amsterdam, North-Holland Karnosky, D., Snger, B. (1995), The general framework for global nvestment management and performance attrbuton, Journal of Portfolo Management, Wnter Kon, S. (1983), The market-tmng performance of mutual fund managers, Journal of Busness 56(3), Lee, C., Rahman, S. (1990), Market tmng, selectvty, and mutual fund performance: An emprcal nvestgaton, Journal of Busness 63(2), Merton, R. (1981), On market tmng and nvestment performance. I: An equlbrum theory of value for market forecasts, Journal of Busness 54(3), Robson, G. (1986), The nvestment performance of unt trusts and mutual funds n Australa for the perod 1969 to 1978, Accountng & Fnance 26(2), Roll, R. (1977), A crtque of the asset prcng theory s tests, Journal of Fnancal Economcs, Roll, R. (1978), Ambguty when performance s measured by the securtes market lne, Journal of Fnance 33(4), Sawck, J., Ong, F. (2000), Evaluatng managed fund performance usng condtonal measures: Australan evdence, Pacfc-Basn Fnance Journal 8(3-4), Sharpe, W. (1964), Captal asset prces: A theory of market equlbrum under condtons of rsks, Journal of Fnance 19, Snger, B., Gonzalo, M., Lederman, M. (1998), Multple-perod attrbuton: Resduals and compoundng, Journal of Performance Measurement, Fall, Snclar, N. (1990), Market tmng ablty of pooled superannuaton funds January 1981 to December 1987, Accountng & Fnance 30, Towers Perrn Pooled Funds Survey, 30 March 1999 Treynor, J., Mazuy, K. (1966), Can mutual funds outguess the market?, Harvard Busness Revew 44, Whte, H. (1980), Heteroskedastcty-consstent covarance matrx estmators and a drect test for heteroskedastcty, Econometrca, Vol. 48(4),

18 Table 1 Market ndces by asset class. The mean strategc benchmark weghts are at 31 December Average weghts are calculated by dvdng the sum of weghts to the respectve sectors by the number of funds that have benchmark exposures to those specfc asset classes. For ths reason, the sum of the weghts exceeds 100 percent. The Morgan Stanley Captal Internatonal Index ncludes gross dvdends renvested and s converted back nto Australan dollars. The Salomon Brothers Index s hedged back nto Australan dollars. Asset Class Market Index Benchmark Weght (%) Australan Equtes ASX All Ordnares Accumulaton Index 36.5 Internatonal Equtes MSCI World (ex-australa) Accumulaton Index 20.9 Australan Drect Property Towers Perrn Drect Property Index 8.2 Australan Lsted Property ASX Lsted Property Accumulaton Index 7.1 Australan Fxed Interest Warburg Dllon Read Composte Bond Index 20.1 Internatonal Fxed Interest Salomon Bros. World Bond Index 6.6 Australan Inflaton-Lnked Warburg Dllon Read Inflaton-Lnked Bond 5.7 Bonds Index Cash Warburg Dllon Read Bank Bll Index

19 Table 2 Pooled superannuaton fund performance at the total portfolo level before expenses usng the Henrksson-Merton model (Panel A) and Treynor-Mazuy model (Panel B) over the perod January 1991 to December Rsk-adjusted performance due to securty selecton (α p ) s expressed n percentage terms per month. Panel A: Henrksson-Merton Model Fund α p t(α p ) β p1 β p2 t(β p2 ) R 2 A B C D E F G H ** I ** J K L M N ** ** O P α β α, β α, β 2 +/- 10 Panel B: Treynor-Mazuy Model Fund α p t(α p ) β p γ p t(γ p ) R 2 A B C D E F G * H *** ** I *** * J ** K L M N ** * O P α γ α, γ α, γ +/- 7 *Sgnfcant at 0.10 level **Sgnfcant at 0.05 level ***Sgnfcant at 0.01 level The t-statstcs are calculated usng Whte (1980) heteroskedastc consstent standard errors. The coeffcent of determnaton s the adjusted R 2. 19

20 Table 3 Pooled superannuaton fund performance at the total portfolo level before expenses usng the Henrksson-Merton model over the perod January 1991 to December Performance s measured usng the All Ordnares Accumulaton Index s used as the market proxy consstent wth Snclar s (1990) study. Rsk-adjusted performance due to securty selecton (α p ) s expressed n percentage terms per month and market tmng estmates are represented n β p2. Fund α p t(α p ) β p1 β p2 t(β p2 ) R 2 A * * B C ** *** D ** ** E ** F ** G *** H *** ** I *** *** J *** K ** * L M * * N O * P *** *** *Sgnfcant at 0.10 level **Sgnfcant at 0.05 level ***Sgnfcant at 0.01 level The t-statstcs are calculated usng Whte (1980) heteroskedastc consstent standard errors. The coeffcent of determnaton s the adjusted R 2. 20

21 Table 4 The performance of pooled superannuaton funds before expenses across the three major nvestment sectors usng the Henrksson-Merton model. The perod of evaluaton s January 1991 to December 1998 where rsk-adjusted performance due to securty selecton (α p ) s expressed n percentage terms per month and market tmng estmates are represented n β p2. Australan Equtes Internatonal Equtes Australan Fxed Interest Fund α p t(α p ) β 2 t(β 2 ) α p t(α p ) β 2 t(β 2 ) α p t(α p ) β 2 t(β 2 ) A C D E ** *** F G ** H ** I ** J *** K ** ** L M N ** ** O P * α α α β β β α, β α, β α, β α, β 2 +/- 11 α, β 2 +/- 9 α, β 2 +/- 13 *Sgnfcant at 0.10 level **Sgnfcant at 0.05 level ***Sgnfcant at 0.01 level The t-statstcs are calculated usng Whte (1980) heteroskedastc consstent standard errors. 21

22 Table 5 Pooled superannuaton fund performance at the total portfolo level and sector level before expenses usng the geometrc attrbuton framework. Performance s expressed as the mean actve performance per month n percentage terms over the perod January 1991 to December Funds are parttoned on the bass of ther predomnant portfolo management style. Total Portfolo Australan Equtes Internatonal Equtes Australan Fxed Interest Fund Tot R s R a R s R a R s R a R s R a Panel A: Top-Down Portfolo Management A * * B ** D E ** *** J 0.167*** ** 0.059** 0.159*** K ** ** 0.011*** L P * Panel B: Bottom-Up Portfolo Management F * G 0.128* *** ** 0.180*** ** H ** ** I * ** ** N O * * SS MT SS, MT SS, MT +/ *Sgnfcant at 0.10 level **Sgnfcant at 0.05 level ***Sgnfcant at 0.01 level 22

23 Table 6 Evaluaton of the ablty of portfolo managers to make correct forecasts n ther nvestment decson-makng over the perod January 1991 to December Total Portfolo (%) Australan Equtes (%) Internatonal Equtes (%) Australan Fxed Interest (%) Fund R s R a R s R a R s R a R s R a Panel A: Top-Down Portfolo Management A 40.6* ** 61.5** 55.2 B ** 61.5** 58.3 D 37.5** ** ** E * *** ** J *** * 53.1 K * L P ** ** Panel B: Bottom-Up Portfolo Management F G 67.7*** 36.5*** 68.8*** 37.5** ** H * 61.5** *** I * 59.4* *** * N O 40.6* 57.3* * * *Sgnfcant at 0.10 level **Sgnfcant at 0.05 level ***Sgnfcant at 0.01 level 23