Momentum Investing and the Asset Allocation Decision

Transcription

1 Momentum Investing and the Asset Allocation Decision Karen L. Benson UQ Business School, The Universy of Queensland, Brisbane, AUSTRALIA David R. Gallagher School of Banking and Finance, The Universy of New South Wales, Sydney, AUSTRALIA Patrick Teodorowski BT Financial Group, Sydney, AUSTRALIA Abstract This paper examines the asset allocation decisions of Australian multi-sector fund managers. Specifically, we determine if active fund managers engage in momentum investing when reallocating funds between asset classes. We find evidence which supports the existence of momentum investing in the active asset allocation to the Australian Equies, Australian Fixed Interest and Listed Property asset classes. Interestingly, balanced funds adopt contrarian strategies in International Equies. We also consider momentum investing during periods of large cash inflows and outflows. Investment allocations are different between periods; however, this difference is consistent wh rebalancing to benchmark weight rather than momentum investing. Finally we test if managers wh different market timing abilies adopt momentum strategies. Our results show those funds wh no market timing skill are momentum investors.

2 . Introduction The asset allocation decision is fundamental in the determination of the portfolio composion for a fund manager. Once an active investment manager has set their strategic or long-term benchmark weight across multiple asset classes, the tactical allocation decision requires the manager to vary the overall exposure to individual assets classes (e.g. Australian and International Equies, Cash, Australian and International Fixed Interest, Property) as a means of enhancing the overall performance of the fund. Aggregate fund performance is both a function of passive and active asset allocation inputs. Passive changes in asset allocation occur when high (low) returns on an individual asset class results in an increase (decrease) in the weight of that class as a proportion of total fund assets. Given the importance of tracking error constraints, managers will be required to rebalance their portfolios in a manner which ensures any differential performance between asset classes does not compromise the overall investment strategy adopted by the fund. Active changes occur when the manager moves funds from one asset class to another wh respect to expectations about future returns. These active changes are referred to as tactical asset allocation. The importance of the asset allocation decision has been examined by Brinson, Hood and Beebower (986), Brinson, Singer and Beebower (99) and Ibbotson and Kaplan (2000). These studies show that more than 90 per cent of the variation in fund performance is explained by a fund s asset allocation decision. Blake, Lehmann and Timmerman (999) and Faff, Gallagher and Wu (2004) examine the dynamics of the tactical asset allocation decision in multiple asset funds in the UK and Australia, respectively. These studies find that, even though active fund managers do engage in tactical asset allocation strategies, they are unable to provide superior returns for investors. Diversified funds would 2

3 generate higher returns by following a passive asset allocation policy rather than employing an active asset allocation strategy. The skill and motivation of the fund manager in implementing active strategies have also been examined in the lerature. There is a vast body of research examining the selectivy and timing abilies of funds managers, (see for example: Treynor and Mazuy, 966; Kon, 98; Henriksson, 984; Chang and Lewellen, 984; Lee and Rahman, 990; and in the Australian context Sinclair, 990; Gallagher, 200; and Hallahan and Faff, 200). More recently, behavioural theories have provided insight into fund manager activies. Brown, Harlow and Starks (996) hypothesise that managers seek to increase the risk and returns of their fund portfolio after a poor performing period. They view the industry as a tournament between the participants, where interim losers increase their portfolio risk as a means of enhancing performance. Wermers (999) examines the herding behaviour of fund managers and shows that managers do follow one another into and out of the same securies. Grinblatt et al. (995) and Burch and Swaminathan (200) show instutional investors engage in buying stocks wh high past returns. This behaviour is consistent wh momentum investing. Bange and Miller (2004) examine momentum investing in the context of global asset allocation recommendations of investment houses, and find evidence to support the application of momentum investing in equy and cash allocation decisions. The concept of momentum investing wh respect to stock selection has had considerable attention in the lerature. However, research on the application of momentum investing to asset classes and the asset allocation decision is limed. In this paper we assess how momentum investment strategies are employed by fund managers in their tactical asset

4 allocation decisions. Specifically, we examine if historical returns of asset classes are a determinant of the amount invested in the key asset classes that comprise a multiple asset class managed fund. We focus on momentum investing as one possible explanation of the fund manager s asset allocation decision, and hence contribute to the behavioural research of assessing fund managers strategies. The examination is undertaken using a proprietary dataset of the strategic and tactical asset allocation components of fund managers asset allocation activies in Australia. Further, we extend the concept of momentum investing from stock selection to asset class allocation. The remainder of the paper is structured as follows: Section 2 provides a review of the lerature, Section outlines the data methodology and the results are discussed in Section 4. A conclusion is provided in section Lerature Review 2. Asset Allocation The asset allocation decision affects the performance of a fund in two main ways: ) long-term investment policy and 2) short-term (active) asset allocation, usually characterized as market timing abily. Brinson et al. (986) and Brinson et al. (99) find that investment policy explains the maory of return variation; however, active asset allocation fails to enhance the performance of pension plans. Conversely, Ibbotson and Kaplan (2000) find that 60% of return variation is unexplained, indicating a large degree of active management. Asset-class timing, asset style, secury selection and fees are suggested to account for this unexplained variation. Blake et al. (999) provides a comprehensive study that examines the abily of U.K. fund managers to generate posive returns through tactical asset allocation across multiple asset classes. The results show 4

5 that long-run asset allocations of pension funds account for a maory of the variation in returns. Further, fund managers show poor market-timing performance in line wh Kon (98) and Henriksson (984). Similarly, Gallagher (200) and Faff et al. (2004) find active asset allocation is unable to provide superior returns to investors. Faff et al. (2004) document that tactical asset allocation (hereafter TAA) whin international shares and domestic fixed interest asset classes detract from performance. 2.2 Momentum investing The existence of momentum in asset markets and stock returns may be tested by reference to the short term autocorrelation of returns. The obective is to determine if past performance is a predictor of future performance, and if economically profable trading strategies can be executed using historical information. Momentum studies consider both individual stock returns and market returns. Jegadeesh and Tman (99) and Chan et al. (996) find evidence to support return momentum in the short run and demonstrate that substantial returns can be made by buying past winners and selling past losers. Rouwenhorst (999) finds momentum in returns across 2 European countries. Hurn and Pavlov (200) and Demir, Muthuswamy and Walter (2004) find momentum profs are achievable whin the Australian equy market. More recently Chordia and Shivakumar (2004) and Cooper et al. (2004) find that the existence of momentum profs is evident during expansionary periods. Further, by controlling for macroeconomic variables, that is the Treasury bill yield, dividend yield, default spread and term structure of interest rates, the existence of momentum profs disappears. 5

6 The implementation of momentum strategies whin equy markets by instutional investors is examined in Grinblatt and Tman (989) and Grinblatt et al. (995). The latter study finds 77% of the funds examined were momentum investors. This percentage was higher for growth fund managers compared to balanced and income fund managers. Further, momentum investing realised higher performance outcomes. Burch and Swaminathan (200) find instutional investors, classified as insurance companies, banks, investment advisors and fund managers, adopt momentum trading strategies when selecting equy investments. Bange and Miller (2004) focus on momentum investing in the global asset allocation decision. This study examines the relationship between historical asset class returns and changes in asset allocation recommendations given by 6 investment houses, from 982 to 999. The results show mixed support for momentum investing as a determinant in the asset allocation decision.. Data and Methodology. Data The data is sourced from the Manager Performance Analytics database provided by Mercer Investment Consulting (MIC). From the database we access monthly data of actual asset allocations, strategic benchmark asset allocations, returns and size of the funds. The MIC Performance Analytics database comprises 44 capal stable funds and 88 growth/balanced funds. The sample period is January 99 to December 200 for Capal Stable funds and January 990 to December 200 for balanced funds. Funds are excluded where there is less than 24 months of data available, if there is only one asset class, or if 6

7 there is missing data in the key variables. After applying these filters, the database of 2 managed funds was reduced to a sample of 84; 5 growth/balanced and capal stable. Total funds under management for these two groups, as at December 200, were 0.2 billion and.8 billion respectively. The balanced and capal stable funds included in the sample consist of 490 and 274 fund years, respectively. The average life of the funds whin the sample is 09 months. The 75 th percentile statistic shows approximately 25% of the sample is alive for the entire sample period. The benchmark asset classes examined in the study are Australian Equies (AEQ), International Equies (IEQ) Listed Property (LP), and Australian Bonds (AFI). The indexes adopted for each of these asset classes are: S&P/ASX 00 Accumulation Index, MSCI world (ex-australia) Index, S&P/ASX 00 Listed Property Accumulation Index, UBS Warburg Compose Bond Index and Solomon Smh Barney World (ex-australia) Government Bond Index. The benchmark data is sourced from DataStream, as is the macroeconomic data on term structure, Treasury bill rates and dividend yield..2 Methods The empirical analysis comprises two sections. First we test for the existence of profable momentum investing opportunies whin four key and liquid asset classes. Second we assess if fund managers engage in momentum investing strategies wh respect to their tactical asset allocation decisions. ASX All Ordinaries Accumulation Index (equies) and ASX Listed Property Accumulation Index (listed property) was used prior to April For autocorrelation tests ASX All Ordinaries Index (equies), ASX Listed Property Accumulation Index (listed property) and UBS Semi-Government All Maturies Bond Index (Australian Fixed Interest) were used as this provided data back to January

8 A finding of profable momentum opportunies whin asset classes would provide support for fund managers who engage in momentum strategies. However, a lack of consistent momentum prof opportunies does not preclude managers from adopting a momentum-based strategy. Indeed there is evidence to suggest that fund managers are momentum investors in their stock selection (Grinblatt et al. 995). It is therefore interesting to consider if they are momentum investors in terms of asset class allocation.. Momentum in asset class returns To test for the relationships between past and future returns we employ a methodology similar to that used by Jegadeesh (990) and Gaunt and Gray (200) to test for autocorrelation in a return series: R, t 2 = α + α R ε () 0, t m= m, t, t m +, t where: R, t = return of asset class at time t. We also examine the autocorrelation of relative asset class returns. This relative measure becomes important in the context of funds. Whin a diversified fund, any active allocation of funds to one asset class must be offset by a corresponding net decrease in funds invested whin another asset class. By examining each asset class individually, momentum investing may be observed whin one class and contrarian investing 2 whin another class. If we consider the relative performance between two asset classes then, if 2 Contrarian investing is seen as investment behaviour whereby one invests money into investments that have performed poorly in the past. 8

9 one class outperforms the other then momentum investing will be observed in both asset classes. Further, if one asset class, say equies, outperforms all other asset classes; fund managers would allocate more of their funds into equies over the following period. To test the autocorrelation in relative asset returns, model () was extended to: where: R J, t Rk, t = 0, t + θ m, t R, t m Rk, t m ) + m= R t Rk, t θ ( ε (2) / k, t, = is the relative return between asset classes and k at time t. Equations () and (2) are estimated using both monthly and weekly returns across five asset classes: Australian Equies, International Equies, Australian Fixed Interest and Listed Property for the period January 980 to June Momentum based strategies in active asset allocation. To determine if past asset class performance is used in future asset allocation decisions, relative net cash flow (RNCF) is modelled as a function of six monthly lagged returns and five control variables. The measure of RNCF identifies only those cash flows due to active allocation changes. Passive effects, due to differential returns between the asset classes in the portfolio, are eliminated. RNCF represents the relative net cash flow for fund manager i to asset class at time t and is defined in Blake et al. (999). 4 To test if fund managers engage momentum-based strategies we estimate equation () for each of OLS is used in the estimation and all tests are corrected for heteroscedasticy. 4 This method assumes that cash flow to the asset class is made at the end of each month. 9

10 four asset classes; Australian equies (AEQ), international equies (IEQ), Australian fixed interest (AFI) and listed property (LP). 5 RNCF = β 0 + β R6 t + γ Benchi,t + γ 2 LogTNAi,t + γ k+ 2 X k,t + ε () k= where: { F ( F ω r ))} F RNCF = i, t ( + / ω i, t ω (4) F = net asset value of fund i at time t, ω = weight managed fund i holds in asset class at time t, r = return on asset class for fund i at time t. R6 t- = the 6 month cumulative benchmark return for asset class lagged by one month, Bench = the lagged measure of the fund s deviation from their strategic i,t benchmark allocation to asset class. Bench = ω ω, where ω = fund i s benchmark weight in asset class at the start of month t, LogTNA i,t- = the lagged measure of the log of the total net assets of fund i, X k,t- = represents the lag of three macroeconomic control variables, specifically, the -month Treasury bill yield, market dividend yield and the term structure premium. The six-month cumulative benchmark lagged return is used as the proxy for momentum whin an asset class, in line wh previous empirical lerature. 6 5 The maory of funds invested in the funds included in the sample are represented by these four asset classes plus cash. 0

11 The measure of the funds deviation from the strategic benchmark asset allocation (Bench) controls for the reallocation of funds back to the strategic posion. We expect a negative relationship between RNCF and strategic benchmark deviation since managers will make active asset allocation decisions to rebalance their portfolio wh respect to the fund s strategic benchmark. Size of the fund may impact on the cash flows allocated to asset classes. We control for this variable by computing the log of total net assets (LogTNA) consistent wh Sirri and Tufano (998), Del Guercio and Tkac (2002) and Sawicki (200). We include three macroeconomic variables: the -month Treasury bill yield (Znote), market dividend yield (Zdivy) and the term structure premium (Zterm). Ferson and Schadt (996) propose that managers use public information to aid their investment decisions. The importance of these variables is demonstrated in Chordia and Skivakumar (2004), where profs generated by momentum trading strategies dissipate when controlling for macroeconomic factors. Similarly Faff et al. (2004) show that active asset allocation may be explained by using a set of lagged macroeconomic indicators. We expect a negative coefficient on Znote (Fama 98), a negative coefficient on Zdivy (Keim and Stambaugh, 986 and Fama and French, 988) and a posive coefficient on Zterm (Fama and French, 998) 6 The maory of the equy momentum lerature adopts six-month lagged return as the period for forming portfolios of stocks when testing for momentum. However Bange and Miller (2004), in examining the recommendation of global asset allocations of investment houses, use a -month lagged return. Furthermore Grinblatt et al. (995) in testing for momentum investing in U.S. equy funds use lagged stock returns of one to four months. To check the robustness of our measure we also estimate equation () using three month lagged cumulative returns similar to that employed by Bange and Miller (2004).

12 .5 Relative return, active asset allocation and momentum investing Fund managers of diversified funds have the abily to choose between various asset classes. We examine the effect of past relative performance between four economically significant and liquid asset classes: Australian Equies (AEQ), International Equies (IEQ), Listed Property (LP) and Australian Fixed Interest (AFI). Cash is excluded from this analysis, largely because changes in cash may be biased by large inflows and cash is usually held as the residual investment. Equation (5) is estimated. 7 RNCF = β + β Rel 0 6,t 2 62,t 6,t 4 64,t γ Bench i,t + γ LogTNA 2 + β Rel i,t + + β Rel γ k+ 2 X k,t k= + ε + β Rel + (5) Where: RNCF Bench LogTNA i,t- and X k,t- are previously defined. i,t Rel6,t- = the difference between the cumulative benchmark asset class return over the prior 6 months for asset class and Australian equies; Rel6 2,t- = the difference between the cumulative benchmark asset class return over the prior 6 months for asset class and international eques; Rel6,t- = the difference between the cumulative benchmark asset class over the prior 6 months for asset class and Australian fixed interest; Rel6 4,t- = the difference between the cumulative benchmark asset class return over the prior 6 months for asset class and listed property;.6 Model extensions 7 Multicollineary is tested by reference to the variance inflation factors (VIF) in Kennedy (998). No VIF values are greater than 0, hence is concluded that multicollineary is not a problem. 2

13 We extend our inial model to consider momentum investing activies of managers when their funds are attracting time varying fund flows. We also classify managers on their timing abilies and examine momentum investing for different fund types. In periods of high inflow, managers may be required to assign new funds to asset classes that will provide attractive returns in the near term. Outflows may lead to the forced selling of (future) profable asset classes due to liquidy requirements. Edelen (999) finds that the market timing abily of fund managers is negatively impacted by increased fund flow volatily. 8 Similarly, Rakowski (200) finds supporting evidence that high fund flow volatily negatively influences fund performance. Liquidy-based trading activy also forces fund managers to operate below their optimal investment strategy, leading to adverse trading activy and performance (Edelen, 999). We examine three states of flows: periods of fund inflow, periods of relatively no fund flow and periods of fund outflows. Our measure of fund flow is expressed in equation (6) and is consistent wh Sirri and Tufano (998). TNA TNA ( + r ) FLOW = (6) TNA where; FLOW = net growth in fund assets beyond reinvested earnings, TNA = total net assets of fund i in month t, r = return of fund i for month t. 8 Gallagher and Jarnecic (2002) find that this negative relationship between fund flow and fund performance exists whin the Australian bond fund market.

14 Periods of inflows (outflows) for individual funds are determined when the monthly fund flow is greater (less) than 2.5%. The number of inflow and outflow months falling whin the top and bottom classifications is between 0% and 20% of the entire sample. 9 To determine if momentum investing activies vary between periods of inflow, balanced inflow or outflow dummy variables are incorporated in the model as in equation (7). RNCF where: = β + β R6 + γ Bench + γ D + k =,in k+ 9 Bench D i,t,out + β D + γ LogTNA i,t X k,t + γ D k = R6,out i,t k β D 0 t 2,out t,in 6t 7 γ γ + LogTNA D k=,in X γ k + 2 i,t k,t R X k,t + γ D + ε 9 + β D + γ D 6,in 4,out,out + β D Bench γ LogTNA 9 5 i,t i,t,in (7) D in D out = for periods of fund inflow for fund i, 0 otherwise, = for periods of fund outflow for fund i, 0 otherwise. The existence of momentum investing may also vary between good and poor market timers. Empirical research has shown that fund managers, as a group, are poor market timers (Kon, 98; Henriksson, 984; Chan et al., 984; Lee and Rahman, 990; Gallagher, 200). However, there is ltle evidence to suggest what drives good or poor market timing. We expect different strategies to be employed by both good versus poor market timers, and therefore we propose that the application of momentum investing will be different between these funds. 9 Previous studies by Sawicki (200), Del Guercio and Tkac (2002) use the top (bottom) 20% and 0% respectively in examining the impact of flow. This study is more interested in examining the impact of significant flow and for this reason a 2.5% inflow (outflow) threshold was set. 4

15 To ascertain which funds display good and poor market timing we implement the method in Burnie et al. (988). The returns generated by diversified funds are decomposed into three categories: secury selection, active asset allocation and residual performance. 0 Our focus is on the active asset allocation component of the funds which Burnie et al. (988) measure using the model expressed in equation (8). R i,at = ( ω )( rt ri,bt ) i ω (8) where; R i,at = return generated due to the active asset allocation strategy for fund i, ω = average actual weight in asset class for fund i, ω = benchmark weight in asset class for fund i, r = benchmark return in asset class in month t, r i, bt = benchmark return for the total portfolio at time t for fund i. Using each individual fund s measure of R i,at we classify the funds into three groups: funds that exhib statistically significant 2 posive return due to superior market timing abily, funds that exhib statistically significant negative return due to inferior market timing abily and funds that show statistically insignificant market timing abily. We then estimate equation (9). 0 Faff et al. (2004) state that the residual cannot strictly pertain to eher stock selection or asset allocation. It represents the interaction between both these facets of active management. The benchmark return represents the return on a passive investment strategy whin that asset class, therefore investing the benchmark. This is chosen to separate active asset allocation abily from secury selection abily. 2 Statistical significance was determined at the 5% level. 5

16 RNCF = β 0 + β R6 + β 2 D R6 + β D R6 + γ Bench + γ D + 7 k = γ,poor k+ 9 D t i,t + γ LogTNA Bench,good X 2 i,t k,t,good + γ D + 8 k = t i,t,good γ k LogTNA D k= γ,poor k + 2 X,poor X i,t k,t k,t t + γ D + γ D + ε 9 + β D 6 4,good,poor,good Bench LogTNA + β D 5 i,t i,t,poor (9) where; D,good D,poor = for good market timers, 0 otherwise. = for poor market timers, 0 otherwise. 4. Results 4. Descriptive Statistics Descriptive statistics are presented in Table. The summaries show that balanced funds hold a greater proportion of growth assets (equies) to defensive assets (cash and fixed interest) than Capal Stable funds. On average, as at December 200, the balanced funds weight in Australian and International Equies is 64% compared to 25% for Capal Stable funds. Capal Stable funds hold 70% of their funds in Cash, Australian and International fixed interest, whereas Balanced funds only hold 26% of their portfolios in these asset classes. Table 2 shows the variation in the strategic benchmark asset allocation for all funds in the sample over the 4 year period. There is ltle variation whin the benchmark asset allocation as represents a funds long term asset allocation. Balanced funds have decreased their benchmark allocation to Australian fixed interest (AFI) while increasing The statistics presented in Table 2 may overstate the yearly variation in the benchmark asset allocation, as individual managed funds enter and ex the sample each year. 6

17 their benchmark allocation to international equies (IEQ) over the sample period. Conversely, capal stable funds have increased their benchmark allocation to international fixed interest (IFI) and reduced the benchmark allocation to AFI. 4.2 Asset Class Momentum Table shows the results from equation () and (2) testing the existence of autocorrelation in monthly asset class returns and relative asset returns. We see only isolated incidences of autocorrelation in the asset class returns indicating a lack of opportuny to prof from momentum investing. Australian equies and fixed interest display no autocorrelation. Interestingly, there is a ltle more evidence of autocorrelation in the relative asset class returns wh five of the six relative classes showing autocorrelation in at least one of the 2 lags examined. Further, for relative returns between AEQ-IEQ there is persistent negative autocorrelation, although not significant in many cases. We also test autocorrelation using weekly return data and found more incidences of autocorrelation in the asset class returns but less in the relative returns. 4 Based on the results presented, asset class returns at monthly frequency are not predictable. This would suggest that fund managers who do engage in momentum investing in asset allocation would be unlikely to posively impact fund performance. Further, funds exhibing no or poor market timing would be more likely to engage in momentum investing. 4. Momentum Investing Tables 4 and 5 document the results for equations () and (5). 5 In Panel A we report the results for the full sample for each asset class. Panels B and C show the results for the two 4 Results are available from the authors on request. 5 The equations are also estimated using a three month cumulative return. The results using this variable were similar although less significant coefficients were found. 7

18 sub-samples: balanced funds and capal stable funds. These results show that momentum investing behaviour exists whin the funds. Significant posive relationships exist between the active asset allocation in the current period to AEQ, AFI and LP and the historical (six-monthly) asset class returns in benchmarks asset class returns. Active asset allocation whin IEQ is the only asset class that does not exhib momentum investing. Diversified funds do engage in momentum investing behaviour. These results are consistent wh Bange and Miller (2004) who find that investment houses engage momentum investing in determining their asset allocation. Panel B of Table 4 shows that the balanced funds display some similaries to the full sample. However, a significant negative relationship exists between the RNCF to the international equies sector and the historical returns of international equies. This result indicates that balanced fund managers engage in a contrarian investment strategy in the IEQ class. In the other asset classes, the relationship between historical asset class returns and RNCF is posive. Significant relationships exist in AEQ and AFI. These two asset classes represent more than half of the funds invested (5% as at December 200). There is persuasive evidence that momentum investing is evident whin the largest invested asset classes of Balanced funds. Results for the capal stable sub-sample are consistent wh the full sample. Momentum investing exists whin AEQ, AFI and LP. It is notable that the coefficient on R6 t- for the AEQ and AFI asset classes are twice the magnude of those for balanced funds. Almost one half of the strategic asset allocation for capal stable funds is whin these two classes. This result shows that for a given return whin AEQ and AFI, capal stable funds will reallocate greater amounts of invested funds to (from) these asset classes than 8

19 balanced funds. The result is consistent also wh Bange and Miller s (2004) finding that high bond houses engage to a greater extent in momentum investing than low bond houses. 6 Table 4 also shows significant coefficients on the deviation from the strategic benchmark asset allocation (Bench), the treasury bill rate, term structure premium and fund size. The significant negative coefficient on Bench - is consistent wh expectations. Performance differentials between asset classes will result in the fund s asset allocation deviating from s benchmark asset allocation. As this deviation occurs, fund managers will reallocate funds to move the asset allocation back to the long term benchmark asset allocation. The statically posive coefficients on Znote t- and Zterm t- indicates that as the economy is performing well, or is perceived to be performing well, more funds are invested whin Australian equies, listed property and Australian fixed interest. The posive coefficient on Znote t- is inconsistent wh expectations. Generally, if interest rates increase we would expect to see a move away from other sectors. Size is also significant whin the maory of all the panels of results in Table 4. The negative coefficient shows that smaller funds, on average, engage in larger monthly active movements of invested funds between asset classes. Consequently, larger funds have a more passive-like asset allocation strategy. Alternatively, large funds may be unable to shift large relative cash flows between asset classes easily in a month due to market liquidy and transaction cost constraints. 6 Bange and Miller (2004) spl their sample of investment houses into high bond houses those that invest large proportion in bonds (similar to capal stable funds) and low bond house (similar to Balanced funds) 9

20 Table 5 reports the results from equation (5). Posive and significant coefficients are found for AEQ relative to IEQ and LP (Rel6 AEQ 2t- and Rel6 AEQ 4t-), and for AFI relative to LP (Rel6 LP 4t-). The results whin the sub-samples provide for more interesting results for discussion. Panel B shows that for balanced funds, there are negative coefficients on the relative return in AEQ AFI and IEQ AFI (Rel6 t-). This finding shows that as Australian and International equies outperform Australian fixed interest in the prior period, there is a tendency for fund managers to reallocate investments from equies to the more defensive asset classes of fixed interest. Conversely, where equies underperform in the previous period, then managers allocate cash flows to equies in the current period. This approach is consistent wh a contrarian investment strategy. We also observe significant negative coefficients, and hence contrarian strategies, in the allocation to international equies when Australian equies under or over perform (Rel IEQ t-), and the allocation to Australian fixed interest when international equies under or over perform (Rel AFI 2t-). There are significant posive coefficients on each of the Rel6 4t- variables showing an active allocation to each of the equies and fixed interest sectors in the current period when these sectors outperform listed property in the prior period. This relationship is strongest for Australian and international equies. These asset classes comprise up to 64% of the assets of balanced funds as at December 200. Similarly there is a posive coefficient, indicating an increasing (decreasing) allocation to listed property in the current period, when this category outperforms (underperforms) fixed interest in the previous period (Rel LP t-). 20

21 The results for the capal stable funds are reported in Table 5, Panel C. There is a significant posive coefficient on the AEQ-IEQ (Rel6 AEQ 2t-) variable showing that, as the domestic equy market performs well relative to the international market, fund managers will increase their holding in Australian equies. There is a significant negative coefficient on AFI IEQ (Rel AFI 2t-), showing that where the fixed interest asset class outperforms international equies in the previous period, managers decrease their allocation to fixed interest in the next period. As wh Balanced funds, Capal Stable funds show significant results whin the asset classes they are most invested in, namely Australian equies and Australian fixed interest. These two asset classes represent 4% of invested funds as at December Reallocation of funds to these asset classes would directly affect allocation of funds from the remaining asset classes. 4.4 Impact of fund flows Table 6 presents the results from equation (7) estimated on the full sample of funds. There is evidence of contrarian investing during fund outflow periods whin the AEQ sector During periods of balanced flows there is momentum investing in AEQ but contrarian in IEQ. We expected more significant coefficients on the momentum variables. If money is flowing into the industry we would expect managers to invest the extra funds into sectors that have done well in the past. Conversely, in an outflow period, the managers may change their investment strategies and not allocate on the basis of past performance. We do find significant coefficients for all asset classes, on the dummy variables, for both inflow and outflows. This result shows that during periods of inflows, RNCF s are increasing to each sector and during periods of outflow the allocation to each sector 7 After adusting for the large cash holding (27.6%), Australian equies and Australian fixed interest makes up 60% of the balance of investments. 2

22 decreases. However, the allocations are not necessarily consistent wh a momentum strategy. It is interesting to note that the largest allocations go to AEQ and IEQ in inflow periods. Prior studies have addressed the implication of fund inflows on performance. Edelen (999) argues that the expected return would decrease due to the parking of funds in cash. Our results show increases to all sectors; however, the greater increase to equies indicates a timely allocation of funds to this category. During states of fund outflow the coefficient on AFI is lower than for the other asset classes. This result is consistent wh fund managers decreasing the relative holding in AFI, to a greater extent than other asset classes, to meet investor redemptions. We find significant negative coefficients on DBench -*D,in and DBench -. Managers are more likely to invest in asset classes that have deviated from their benchmark asset allocation during periods of fund inflow and balanced flows. We also find significant coefficients on the size control variable in states of fund inflow and outflow. The significant posive coefficients on LogTNA - * D,out indicate that during periods of outflow, larger funds actively decrease investment in the asset classes tested (AEQ, IEQ, AFI and LP) to a lesser extent than the smaller funds. This result is misleading because a dummy variable is used and does not adust for the level of relative cash outflow. 8 The posive coefficient indicates that large funds are less susceptible to large relative outflows compared to smaller funds. The results in Table 6 do not support the hypothesis that fund managers have a higher (or lower) propensy to engage in momentum investing during differing states of flow. We 8 Further investigation was carried out on the sample and was found that the observations included whin the outflow and inflow states are biased towards smaller funds. Therefore smaller funds are more likely to experience higher flow volatily compared to relatively larger funds. This would be expected as large dollar receipts are required to impact on the size of large funds (therefore also fund volatily of these funds). 22

23 do find that investment allocations change in periods of inflows and outflows, but this is more in line wh the realigning funds toward their benchmark asset allocations rather than as a result of momentum investing Impact of market timing The results for equation (9) are presented in Table 7. These findings indicate that funds wh no market timing abily adopt a momentum investing strategy in the AEQ, AFI and LP sectors. Good market timers do not use historical returns as a guide to future investment allocation, whereas poor market timers follow a contrarian strategy in the LP asset class and a momentum strategy in AEQ. These results are somewhat consistent wh expectations, in that there is some difference in the strategies employed by good market timers and others. There is ltle difference between good and poor timers on the macroeconomic and benchmark allocation variables. There are two significant coefficients, out of a possible 6, for good timers. The same two variables are also significant for poor timers wh and addional four. The three groups of market timers each have a significant coefficient on Bench IEQ - indicating a return to the benchmark allocation whin this asset class. Poor market timers appear to use the macroeconomic information a ltle differently to those good timers in the allocation to the AEQ and AFI asset classes. Both good and poor market timers show some posive signs on the coefficients for size (LogTNA - *D,good and LogTNA - *D,poor ). A Wald test was performed to test if LogTNA + LogTNA - *D,good/poor was significantly different to zero. For poor market timers, active asset allocation was not found to be related to the size of the fund. This result is different to the full sample, which shows larger funds engage to a lesser extent in active asset allocation. For good market timers there was no relationship between the size of the fund for AEQ 9 The equations are also estimated using 0% and 20% cut-off for the size measure. The results were very similar. 2

24 and IEQ, while AFI and LP display a posive relationship between fund size and active asset allocation to this asset class. 5. Conclusion The purpose of this study is to analyse the use of momentum strategies by fund managers in their asset allocation decisions, using a unique database of monthly asset allocation and strategic benchmark weights of instutional funds. We examine a sample of multi-sector funds that are classified as eher balanced or capal stable funds. Momentum strategies are assessed on the basis of historical returns of four of the most economically significant and liquid asset classes: Australian equies, Australian fixed interest, listed property and international equies. The relative returns of each asset class are also examined. As a preliminary, we tested the autocorrelation in past monthly returns whin various asset classes. This analysis provides ltle evidence to support profable trading opportunies on the basis of past asset class returns. However, we find that managers indeed implement momentum strategies. For the full sample we find the allocation decision is a posive function of past returns for AEQ, AFI and LP. Results do vary between the two groups of funds. Capal stable funds are consistent wh the full sample. For balanced funds, momentum investing is evident in the AEQ and AFI sectors. A contrarian strategy is evident in the IEQ sectors. When asset class returns are examined on a relative basis we find that momentum strategies are evident in the asset classes that the funds most heavily invest in. Balanced funds follow a momentum strategy in IEQ, AEQ and AFI relative to LP but a contrarian strategy in AEQ and IEQ relative to AFI. Capal Stable funds adopt a momentum strategy when AEQ performs well relative to IEQ but a contrarian strategy wh respect to AFI when outperforms IEQ. We also test momentum strategies controlling for cash inflows and outflows and market timing abilies. Although allocations do change in inflow and outflow periods there is no difference in the momentum investment behaviour. We find poor market timers are more 24

25 likely to adopt a contrarian strategy. Funds wh no market timing abily are more likely to be momentum investors. References Bange, M.M. and Miller Jr., T.W. (2004). Return Momentum and Global Portfolio Allocations, Journal of Empirical Finance, (4), Blake, D., Lehmann, B. and Timmerman, A. (999). Asset Allocation Dynamics and Pension Fund Performance, Journal of Business, 72, Brinson, G.P., Hood, L.R. and Beebower, G.L. (986). Determinants of Portfolio Performance, Financial Analysts Journal, 42, Brinson, G.P., Singer, B.D. and Beebower, G.L. (99). Determinants of Portfolio Performance II: An Update. Financial Analysts Journal, 47, Brown, K.C., Harlow, W.V. and Starks, L.T. (996). Of Tournaments and Temptations: An analysis of Managerial Incentives in the Mutual Fund Industry, The Journal of Finance, 5, Burch, T.R. and Swaminathan, B. (200). Are Instutions Momentum Traders? Working paper series, Graduate School of Business, Universy of Chicago - Graduate School of Business. Burnie, J., Knowles, J. and Teder, T. (998). Arhmetic and Geometric Attribution, Journal of Performance Measurement,, Chan, L.K.C, Jegadeesh, N. and Lakonishok, J. (996). Momentum Strategies. Journal of Finance, 5, Chang, E. and Lewellen, W. (984). Market Timing and Mutual Fund Investment Performance, Journal of Business, 57, Chordia, T. and Shivakumar, L., (2004). Momentum, Business Cycle and Time Varying Expected Returns, Journal of Finance, 57(2),

26 Cooper, M.J., Gutierrez Jr., R.C. and Hameed, A., (2004). Market States and Momentum, Journal of Finance, 59(), Del Guercio, D., and Tkac, P. (2002). The Determinants of the Flow of Funds of Managed Portfolios: Mutual Funds vs. Pension funds, Journal of Financial and Quantative Analysis, 7, Demir, I., Muthuswamy, J., and Walter, T. (2004). Momentum returns in Australian equies: The influence of size, risk, liquidy and return computation, Pacific-Basin Finance Journal, 2, Edelen, R. (999). Investor Flows and the Assessed Performance of Open-end Mutual Funds, Journal of Financial Economics, 5, Faff, R. Gallagher, D.R., and Wu, E. (2004). Tactical Asset Allocation: Australian Evidence, Australian Journal of Management, Forthcoming. Fama, E. (98). Stock returns, real activy, inflation and money, American Economic Review, 7, Fama, E. and French, K. (988). Dividend Yields and Expected Returns on Stocks and Bonds, Journal of Financial Economics, 25, Ferson, W. and Schadt, R. (996). Measuring Fund Strategy and Performance in Changing Economic Condions, Journal of Finance, 5, Gallagher, D.R. (200), Attribution of Investment Performance: An Analysis of Pooled Australian Superannuation Funds, Accounting and Finance, 4, Gallagher, D.R., Jarnecic, E. (2002), The Performance of Active Australian Bond Funds, Australian Journal of Management, 27, -2. Gaunt, C. and Gray, P. (200). Short-term Autocorrelation in Australian Equies, Australian Journal of Management, 28, Grinblatt, M. and Tman, S. (989a). Mutual Fund Performance: An Analysis of Quarterly Portfolio Holdings, Journal of Business, 62,

27 Grinblatt, M., Tman, S. and Wermers, R. (995). Momentum Investment Strategies, Portfolio Performance and Herding: A Study of Mutual Fund Behavior, American Economic Review, 85, Hallahan, T. and Faff, R. (999). An Examination of Australian Equy Trusts for Selectivy and Market Timing Performance, Journal of Multinational Financial Management, 9, Henriksson, R. (984). Market Timing and Investment Performance: An Empirical Investigation, Journal of Business, 57, Hurn, S. and Pavlov, V. (200). Momentum in Australian Stock Returns, Australian Journal of Management, 28, Ibbotson, R.G. and Kaplan, P.D. (2000). Does Asset Allocation Explain 40%, 90%, or 00% or Performance? Financial Analyst Journal, January, 26-. Jegadeesh, N. (990). Evidence of the Predictable Behavior of Secury Returns, Journal of Finance, 45, Jegadeesh, N. and Tman, S. (99). Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency, Journal of Finance, 48, Keim, D. and Stambaugh, R. (986). Predicting returns in the stock and bond market, Journal of Financial Economics, 7, Kennedy, P. (998). A Guide to Econometrics, Fourth edion, Massachusetts, USA, Blackwell Publishers Inc. Kon, S. (98). The Market-Timing Performance of Mutual Fund Managers, Journal of Business, 56, Lee, C. and Rahman, S. (990). Market Timing, Selectivy, and Mutual Fund Performance: An Empirical Investigation, Journal of Business, 6, Rakowski, D. (200). Fund Flow and Performance, Working Paper, Georgia State Universy. 27

28 Rouwenhorst, K.G. (999). International Momentum Strategies, Journal of Finance, 5, Sawicki, J. (200). Investors Differential Response to Managed fund Performance, Journal of Financial Research, 24-, Sinclair, N. (990). Market Timing Abily of Pooled Superannuation Funds January 98 to December 987, Accounting and Finance, 0, Sirri, E.R. and Tufano, P. (998). Costly Search and Mutual Fund Flows, Journal of Finance, 5, Treynor, J. and Mazuy, K., (966). Can Mutual Funds Outguess the Market. Harvard Business Review, 44, -6. Wermers, R. (999). Mutual Fund Herding and the Impact on Stock Prices, Journal of Finance, 54,

29 Table Descriptive Statistics and Equally-Weighted Active Asset Allocation for the Sample Panel A: Balanced Funds Average Funds / Year Total Fund Months / Year Total Year-end Size ($M) Australian Equies (%) International Equies (%) Direct Property (%) Listed Property (%) Aust Fixed Interest (%) Int l Fixed Interest (%) International Bonds (%) Domestic Cash (%) Other (%) Panel B: Capal Stable Funds Average Funds / Year Total Fund Months / Year Total Year-end Size ($M) Australian Equies (%) International Equies (%) Direct Property (%) Listed Property (%) Aust Fixed Interest (%) Int l Fixed Interest (%) International Bonds (%) Domestic Cash (%) Other (%)

30 Table 2 Equally-Weighted Benchmark Asset Allocation for the Sample Panel A: Balanced Funds Australian Equies (%) International Equies (%) Direct Property (%) Listed Property (%) Aust Fixed Interest (%) Int l Fixed Interest (%) International Bonds (%) Domestic Cash (%) Other (%) Panel B: Capal Stable Funds Australian Equies (%) International Equies (%) Direct Property (%) Listed Property (%) Aust Fixed Interest (%) Int l Fixed Interest (%) International Bonds (%) Domestic Cash (%) Other (%)