Active investment manager portfolios and preferences for stock characteristics

Transcription

1 Accounting and Finance 46 (2006) Active investment manager portfolios and preferences for stock characteristics Simone Brands, David R. Gallagher, Adrian Looi School of Banking and Finance, The University of New South Wales, Sydney, 2052, Australia Abstract The present study investigates the stock characteristic preferences of institutional Australian equity managers. In aggregate we find that active managers exhibit preferences for stocks exhibiting high-price variance, large market capitalization, low transaction costs, value-oriented characteristics, greater levels of analyst coverage and lower variability in analyst earnings forecasts. We observe stronger preferences for higher volatility, value stocks and wider analyst coverage among smaller stocks. We also find that smaller investment managers prefer securities with higher market capitalization and analyst coverage (including low variation in the forecasts of these analysts). We also document that industry effects play an important role in portfolio construction. Key words: Stock preferences; Managed fund performance; Equity portfolio characteristics; Market efficiency; Tracking error JEL classification: G12, G14, G23 doi: /j X x 1. Introduction The present study examines the portfolio preferences for stock characteristics of active institutional Australian equity managers. Empirical studies for US and Australian markets suggest that stock characteristics, beyond traditional proxies for The authors are grateful to the Securities Industry Research Centre of Asia-Pacific Stock Exchange Automated Trading System and Portfolio Analytics for the provision of data and the portfolio holdings data (respectively). Thanks are also extended to I/B/E/S for the provision of analyst coverage information that was provided for academic research purposes. We also gratefully acknowledge financial support from Mercer Investment Consulting, and helpful comments from Robert Faff (the Editor), James Jackson, Greg Liddell, Matt Pinnuck and two anonymous referees. Received 18 February 2004; accepted 6 May 2005 by Robert Faff (Editor).

2 170 S. Brands et al. / Accounting and Finance 46 (2006) risk, capture a wider spectrum of factors that also explain the inclusion of stocks in fund portfolios (Badrinath et al., 1989; Del Guercio, 1996; Falkenstein, 1996; Covrig et al., 2001; Gompers and Metrick, 2001; Chan et al., 2002; Bennett et al., 2003; Pinnuck, 2004). Specifically, the present study documents the preferences of active Australian institutional equity managers with respect to transaction costs, stock size, return variance, momentum, investment style and the degree of analyst coverage. The inclusion of a security in an active fund manager s portfolio is a function of three important decisions: (i) the types of stocks the portfolio manager should have exposure to in achieving the fund s investment objectives (i.e. stock characteristic preferences); (ii) prudential management in stock selection decisions; and (iii) the suitability of holding a stock given the manager s private information set. Our research considers the first two components of the decision-making process as a means of understanding the composition of active portfolios and the preferences for stock characteristics by institutional investors. The present study extends the published literature in several ways. First, our study considers an active manager s preferences for securities according to stock size, because small stocks might be included in a portfolio for different reasons than large stocks. This reflects the different level of tracking error risk associated with large and small firms in a portfolio. Second, unique preferences might also arise given the investment style of an active manager, or the size of the institution (in terms of their funds under management), because differences in investment processes and organizational features might affect investment decision-making. Third, our study also examines the stock-selection process on an individual manager basis, in addition to the approach adopted by Falkenstein (1996), Badrinath et al. (1989), Del Guercio (1996), Covrig et al. (2001), Gompers and Metrick (2001) and Pinnuck (2004), who evaluate preferences for stock characteristics at an aggregate (or market consensus) level. Accordingly, individual manager data result in improved granularity in understanding the diversity of stock characteristic preferences among individual portfolio managers. Finally, we consider the role of global industry classification system (GICS) industry classifications in the design of active portfolios, and whether Australian investment managers exhibit preferences for certain industries. Utilizing a unique database of equity fund portfolio holdings over a more recent period to Pinnuck (2004), the present study finds that active managers exhibit preferences for stocks with higher stock price variance, lower transaction costs, larger market capitalization, preferences towards value stocks, securities with higher analyst following and stocks with a lower standard deviation in analyst earnings forecasts. We also find that stock volatility and analyst coverage are of greater importance for small stock holdings. Smaller managers (by assets) exhibit preferences for stocks with even greater market capitalization and higher analyst coverage (including low variation in the forecasts of these analysts). Evidence is also presented by differences in stockholdings on the basis of industry exposures, demonstrating the existence of managerial preferences according to industry dimensions. The present study is structured as follows. Section 2 presents a theoretical discussion motivating research into the stock characteristic preferences of portfolio

3 S. Brands et al. / Accounting and Finance 46 (2006) managers. The next section describes the data and Section 4 outlines the research design. Section 5 presents the empirical results, and the final section concludes. 2. Theory This section outlines a theoretical discussion of the expected preferences for stock characteristics of active equity managers. In particular, this section develops the hypotheses by considering portfolio design that relates to investment performance and prudential requirements Investment performance In a world where capital market frictions exist, stock characteristic preferences might be affected by the following considerations: transaction costs, information asymmetries and historical returns. These are discussed below Transaction costs Keim and Madhavan (1998) provide a comprehensive review of the published trading costs literature. Studies show that transaction costs are related to both traderelated factors, such as skill and reputation, and stock-related factors, such as price level and liquidity (e.g. Aitken and Frino, 1996). Because these costs erode fund performance, a manager wishing to maximize after-cost performance will, ceteris paribus, prefer stocks with lower transaction costs. Given the empirical relationship between stock liquidity and transaction costs, one would expect managers to exhibit a preference for highly liquid stocks as a means of minimizing implicit transaction costs. A number of studies also find evidence that managers prefer liquid stocks with a high volume of trading. Falkenstein (1996) examines US mutual fund portfolio holdings and finds that managers prefer stocks with high liquidity (i.e. trading volume). Covrig et al., (2001) examine manager preferences across 11 developed countries and find both foreign and domestic institutions exhibit preferences for stocks with high daily turnover (a proxy for transaction costs). Therefore, we hypothesise: H 1 : Performance-maximizing managers, ceteris paribus, invest in stocks with low transaction costs, that is, those stocks which have low bid/ask spreads Historical returns and book-to-market equity Jegadeesh and Titman (1993) document the existence of positive serial correlation among short-horizon US security returns, suggesting the potential of exploiting

4 172 S. Brands et al. / Accounting and Finance 46 (2006) momentum strategies. Australian evidence also supports the US findings (see Hurn and Pavlov, 2003 and Demir et al., 2003). To date, studies also support investment managers using a momentum strategy, tending to purchase stocks with a track record of good past performance (e.g. Jegadeesh and Titman, 1993, 2001; Grinblatt et al., 1995; Chen et al., 2000 and Chan et al., 2002). Grinblatt et al. (1995) show that US mutual fund managers tend to purchase past winners, however they note that the momentum effect did not extend to the selling of past losers. This momentum strategy produced significant abnormal returns, whereas contrarian strategies produced little or no outperformance. Chan et al. (2002) investigate momentum within an investment style context, and find that managers generally hold stocks close to benchmark (S&P 500) weights, however, they are more likely to over-or-underweight the stock if it is a past winner or is growth-oriented. Chen et al. (2000) also document similar results. H 2 : There is a positive relation between stockholdings and past price momentum. If markets are not perceived to be fully efficient, active portfolio managers would be expected to strategically hold a subset of the market portfolio which biases the portfolio s constituents in favour of either value or growth stocks. Existing research documents superior one-year-ahead performance for value stocks (Halliwell et al., 1999), suggesting that managers would exhibit a preference towards stocks with high book-to-market ratios. Gompers and Metrick (2001) find supporting evidence for this among US institutional investors. Pinnuck (2004) finds no evidence of preferences for either growth or value stocks among Australian equity managers, and Chan et al. (2002) find evidence of US mutual funds exhibiting preferences for growth stocks. Stickel (1997) suggests that this might be a function of the increased analyst coverage of glamour stocks. Chan et al. (2002) also confirm these findings, as well as identifying a tendency for value strategies needing a longer time interval to yield superior profits. H 3 : Earnings per share yield is positively related to holdings Prudency constraints Investment managers have a fiduciary responsibility to clients. Fiduciaries have an incentive to ensure each component of the portfolio is considered as a prudent and defendable investment decision in the instance of extreme poor performance (Del Guercio, 1996). Badrinath et al. (1989) make reference to the managerial safetynet hypothesis, claiming that in such times of inferior performance a safety net is provided to managers provided they can show soundness of judgment with respect to their investment decision-making process. Del Guercio (1996) and Badrinath et al. (1989) document such evidence for US institutional investors, finding a tendency for this group of investors to tilt their portfolios towards higher quality or prudent stocks. H 4 : The volatility of stock returns is negatively related to portfolio holdings.

5 S. Brands et al. / Accounting and Finance 46 (2006) Institutional investors have been observed to exhibit higher levels of ownership in lower risk securities in the USA (Badrinath et al., 1989) as well as in Australia (Pinnuck, 2004). Interestingly, however, Falkenstein (1996) documents a preference by US mutual funds for high volatility stocks that might be a consequence of an agency problem existing between these funds and their investors. Recent evidence by Bennett et al. (2003) suggests that institutional investors have increased their preference towards stocks with higher volatility. Alternative motivations for the high stock volatility hypothesis might arise where there exists a divergence in opinion concerning the true price of a security among market participants. In short, there appears to be some debate about what preferences institutional investors might have in terms of stock return volatility; however the existing empirical evidence suggests institutional investors should prefer lower volatility stocks. Stock size has been previously identified as a prudent management variable by Del Guercio (1996). This is motivated by the observations of Shefrin and Statman (1995) and Lakonishok et al. (1994), where large stocks are considered to be good by investment managers. A preference for large stocks has been documented in the majority of markets and fund types (Badrinath et al., 1989; Del Guercio, 1996; Falkenstein, 1996; Gompers and Metrick, 2001 and Pinnuck, 2004). H 5 : Stock holdings of active portfolios are positively related to stock size. The degree of analyst coverage can also be related to prudent investment decisionmaking. A stock with a high degree of analyst coverage, as well as consensus among the earnings forecasts of these analysts (measured by the standard deviation of analyst forecasts), provides the manager with external validation for selection of the security. Given that a prudent investment is an investment that is deemed appropriate by other investment professionals, a security with high levels of positive analyst coverage should satisfy such a condition. Additionally, a momentum variable might also be influenced by prudential management concerns as stocks with good (poor) performance track records can be more easily justified for inclusion (exclusion) in an active portfolio. H 6 : Active managers prefer stocks with a higher degree of analyst coverage, and stocks with lower standard deviations of analyst forecasts. 3. Data The sample consists of the monthly portfolio holdings of a sample of 36 active Australian institutional equity managers over the period from 30 September 2000 to 30 September The data are sourced from the Portfolio Analytics Database. The fund holdings information includes all stocks, option securities, futures contracts, and cash positions. More than 96 per cent of holdings by funds are in equity securities. The average size of the funds in the sample by funds under management was $544 ($645) million in 2000 (2001). Ranked by funds under management, the sample

6 174 S. Brands et al. / Accounting and Finance 46 (2006) represents five of the top 10 managers from the Australian institutional fund manager population, four ranked 11 20, five ranked and the remainder outside the largest 30 managers. Given that the manager holdings data were collected from the institutions at a common time period, a degree of survivorship bias is present. However, after comparing the performance of the funds in the Portfolio Analytics Database to the performance of the survivorship-free population of funds in the Mercer Investment Consulting Manager Performance Analytics database, it appears that this bias is limited. An important feature of the Portfolio Analytics Database is the inclusion of options positions in a manager s holding data. Although US and Australian fund managers can hold exchange-traded options, US mutual fund data only report the physical stock holdings. Consequently, by ignoring options positions US studies have not captured the entire exposure to stocks. In the present study, options are accounted for by determining the equivalent number of ordinary shares (using the option s delta) and adding this to the manager s stock holding, consistent with the method adopted by Pinnuck (2003, 2004). The present study uses stock price information sourced from the Australian Stock Exchange (ASX) Stock Exchange Automated Trading System (SEATS). This database was provided by the Securities Industry Research Centre of Asia-Pacific (SIRCA). The SEATS data contain stock characteristics data including stock codes, bid-ask information, stock prices and other data permitting measurement of the variables empirically considered in the analysis of stock characteristic preferences. In terms of the sufficiency of historical data to be used in the analysis, securities are required to have one full-year of price history as at each of the dates examined. The analyst coverage data are provided by the Thomson Financial subsidiary company Institutional Brokers Estimate System (I/B/E/S) as part of their academic research program. Our data also include the ASX and S&P/ASX benchmark index weights for stocks that are constituents of the ASX All Ordinaries Accumulation Index (pre April 2000) and the S&P/ASX 100, 200 and 300 Accumulation Indices (post April 2000). This information is provided by SIRCA. In our sample, two managers are benchmarked to ASX 100, six to the ASX 200 and 28 the ASX Empirical design 4.1. Aggregate holdings This research uses a cross-sectional regression model to evaluate active manager preferences for stock characteristics. Observations are aggregated across the 37 managers in the sample following the methodology adopted by Falkenstein (1996), Chen et al. (2000), Covrig et al. (2001) and Pinnuck (2004). The dependent variable consists of the aggregate holdings in each stock and is defined as: PIH it = AGGNUM it TOTNUM it, (1)

7 S. Brands et al. / Accounting and Finance 46 (2006) where PIH i measures the fractional ownership of individual stocks at time t, AGGNUM it is the aggregate number of shares of stock i owned by all funds at the given date. TOTNUM it is the total number of outstanding shares of stock i at that date. Given the censored nature of the dependent variable a Tobit regression model is used in the estimation process. Use of ordinary least squares would give rise to biased and inconsistent coefficient estimates. The regression is specified as follows: PIH it = β 0 + β 1 MomentumPos it + β 2 MomentumNeg i,t + β 3 Variance it + β 4 EPSY it + β 5 Spread it + β 6 Size it + β 7 IBES it + β 8 IBESSD it + e it, (2) where MomentumPos denotes any positive percentage change in stock price over the prior 12 months in deciles and MomentumNeg denotes any negative percentage change in stock price over the prior 12 months in deciles. Var represents the log of the variance of daily stock returns for the month. EPSY equates to the earnings per share as a percentage of share price (or earnings yield) in deciles. Spread provides the stock s monthly time-weighted relative bid-ask spread, calculated using the methodology of McInish and Wood (1992). Size is the log of market capitalization of stocks expressed as a percentage of the S&P/ASX All Ordinaries Index. IBES represents the number of analysts covering the stock and IBES SD is the standard deviation of analyst forecasts. The IBES variable serves as a proxy for stock visibility in the market Disaggregate holdings In contrast to the above method, which examines aggregate manager holdings data in the evaluation of manager preferences, the individual stock selection decisions of active managers are examined. This improves granularity in the analysis and avoids distortions in the aggregation process. For example, distortions might arise in the aggregation of manager holdings where a single manager with a high allocation to a small stock is indistinguishable from a population-wide overweight allocation to the same small stock. The following cross-sectional regression is performed at each of the four points of time relevant to the study: Holding it = β 0 + β 1 MomentumPos it + β 2 MomentumNeg i,t + β 3 Variance it + β 4 EPSY it + β 5 Spread it + β 6 Size it + β 7 IBES it + β 8 IBESSD it J + D 8+i M it + e it. (3) i=1 Holding represents the actual portfolio holdings provided by each manager in the sample (in terms of the percentage weight that individual stocks constituted in

8 176 S. Brands et al. / Accounting and Finance 46 (2006) the aggregate portfolio). The actual holdings are used in preference over a relative measure based on a benchmark index because the inclusion of the intercept and the Size variable on the right-hand side of the equation is a less restrictive definition. 1 Zero holdings are included in the regression. The explanatory variables are identical to those observed in the earlier aggregate regression, with the addition of a set of j manager dummy variables, where j refers to the number of managers in the sample at the analysis point. 2 Given the censored nature of the dependent variable, a Tobit regression model is used in the estimation procedure Individual manager regressions The regression reported in equation (3) is also performed for the holdings of each individual manager. This is to ensure that evidence suggested by the disaggregated results is reliable and not a function of a high degree of cross-sectional dependence in the error term in the disaggregated regressions. 5. Empirical results 5.1. Regression results Our analysis commences with an examination of the stock characteristic preferences of all active managers in our sample. Table 1 reports both the individual manager (Panel A), disaggregate (Panel B) and aggregate (Panel C) results of the Tobit regression for all active managers and all stocks as at 30 September for each of the years 2000 and The Wald tests for each of the regressions are highly significant, showing the explanatory variables are jointly significantly different from zero. The coefficients on the Momentum variables are of mixed signs for each of the regressions, suggesting that the active managers in our sample do not consistently follow either a momentum or contrarian strategy. The results are also consistent with previous Australian findings, where Pinnuck (2004) finds Australian equity managers do not rely on momentum strategies. Gompers and Metrick (2001) examine US institutional funds and also find that managers are not momentum investors, in fact the evidence suggests that they use contrarian strategies. The coefficient Variance is positive and highly significant across all years, as well as for the individual manager, disaggregate and aggregate regressions, reported in Table 1. This indicates that active managers prefer to hold more volatile stocks, which is inconsistent with Pinnuck (2004), however given the evidence of Bennett 1 Using (holdings benchmark) assumes the coefficient of size is one. 2 Although manager dummy variables are included in the regression model, their coefficients are not reported in the results section.

9 S. Brands et al. / Accounting and Finance 46 (2006) Table 1 Regressions of active equity manager portfolio holdings The regression model is as follows: Holding it = β0 + β1momentumposit + β2momentumneg i,t + β3varianceit + β4epsyit + β5spread it + β6sizeit + β7ibesit J + β8ibessdit + D8+iMit + eit. i=1 MomentumPos is the positive percentage change in the price of stock i over the previous 12 months in deciles and MomentumNeg is the negative percentage change in the price of stock i over the previous 12 months in deciles. Var is the natural log of the variance of daily stock price returns of stock i in the previous month. EPSY is the natural log of the earnings per share for stock i as a percentage of share price in deciles. Spread is the time weighted relative bid/ask spread for the month of September. Size is the log of the market capitalization of stocks expressed as a percentage of the S&P/ASX All Ordinaries Index in Panels A and B and the log of market capitalization in Panel C. IBES is the number of analysts covering the stock and IBES SD the standard deviation of the forecasts of these analysts. NObsis the number of observations in the sample. t-statistics are calculated based on White s heteroskedastic-consistent standard errors. Panel A refers to individual manager regressions, Panel B to the disaggregate regressions where Holding is the percentage weight of stock i in each of the manager s portfolios. Panel C refers to aggregate regressions where the dependent variable (Holding) consists of the aggregate number of shares of stock i owned by all funds at t divided by the total number of outstanding shares of stock i at that date. Year (number Positive Negative Variance EPSY Spread Size IBES IBES SD in sample) momentum momentum Panel A: Individual Average coefficient 2000 (35) (38) Median coefficient 2000 (35) (38) Number of positive coefficients 2000 (35) (38) Number of negative coefficients 2000 (35) (38)

10 178 S. Brands et al. / Accounting and Finance 46 (2006) Table 1 (continued) Year (number Positive Negative Variance EPSY Spread Size IBES IBES SD in sample) momentum momentum Number of significant coefficients (5%) 2000 (35) (38) Number of significant coefficients (10%) 2000 (35) (38) Panel B: Disaggregate Coefficient 2000 (35) t-statistic Coefficient 2001 (38) t-statistic Panel C: Aggregate Coefficient 2000 (35) t-statistic Coefficient 2001 (38) t-statistic ,, indicates statistical significance at the 1, 5 and 10 per cent levels, respectively.

11 S. Brands et al. / Accounting and Finance 46 (2006) et al. (2003) that institutional preferences have been increasing towards more volatile stocks in recent years; our analysis covers a more recent period compared to Pinnuck (2004) and might also be consistent with changes in investor preferences across time. A positive preference for volatility is also consistent with managers exploiting mispricings for stocks for which there is little consensus in valuation. Overall, our findings are not consistent with H 4 and suggest that prudency constraints are not driving preferences concerning stock return variance. The EPSY coefficient is positive and significant for all of the aggregate regressions and disaggregate regressions. The evidence provided by the individual manager regressions is consistent although weaker. The average and median coefficients for the individual manager regressions are positive, although the majority of the coefficients are insignificant. Therefore, we find weak evidence that active managers prefer value stocks relative to growth. This finding is consistent with H 3 and indicates that managers have taken advantage of the historically superior returns on value stocks in Australia (Halliwell et al., 1999). Previous Australian evidence does not find evidence of preference for either growth or value stocks (see Pinnuck, 2004). Table 1 also shows that managers have a preference for stocks with low spreads, indicated by the negative and significant coefficients across all regressions. This is consistent with H 1 that states that performance-maximizing managers are expected to invest in stocks with low transaction costs. This finding is confirmed by Gompers and Metrick (2001), Falkenstein (1996), Del Guercio (1996), Badrinath et al. (1989) in the USA, and Pinnuck (2004) in an Australian context. Managers in the sample exhibit preferences for stocks with larger market capitalizations, as evidenced by the coefficients on the Size variable being greater than one, which are highly significant across all years in the disaggregate regressions and for the majority of the individual manager regressions. 3 The coefficients for the aggregate regressions are consistent in sign although they are not statistically significant. 4 The results are consistent with H 5, suggesting that managers are indeed concerned with prudential considerations. A preference for large stocks has also been documented in the majority of markets and fund types (Badrinath et al., 1989; Del Guercio, 1996; Falkenstein, 1996; Gompers and Metrick, 2001 and Pinnuck, 2004). The coefficients on the IBES variable are positive and significant across all years, and all regression forms finding evidence in favour of H 6 that managers prefer stocks with analyst coverage and, therefore, lower information asymmetries. Similar results are documented by Covrig et al. (2001) who also find a positive relationship between aggregate fund holdings and the number of analysts following the security. Falkenstein (1996) proxies information asymmetries with the number of major news articles and the number of months since listing on the exchange, finding that funds tend to avoid 3 As seen in Panel A of Table 3, 32 of the 35 size coefficients are significant at 5% in 2000 and 38 of 38 in For a sample of passive managers (i.e. with no size preference) we would expect a coefficient of zero; therefore, this variable is interpreted in terms of deviations from zero in the aggregate regressions.

12 180 S. Brands et al. / Accounting and Finance 46 (2006) stocks for which there is little information. This is also consistent with prudency issues identified previously, where the inclusion of a security in a portfolio that is followed by an analyst is likely to be much easier to justify. IBES SD measures the standard deviation of analyst earnings forecasts over the prior year. This coefficient is negative and highly significant for all of the disaggregate regressions. The average and median coefficients are also negative for the individual manager regressions and the majority of which are significant at 10 per cent. These results suggest that active managers prefer stocks for which there is a higher degree of consensus among the earnings forecasts of analysts. These results are also consistent with H Stock size To capture the importance of tracking error risk in portfolio construction, an individual stock s contribution to portfolio tracking error should be taken into consideration in the examination of a manager s preferences. Because of the nature of the value-weighted index to which managers in the sample are benchmarked, tracking error contribution is a function of stock size. As a consequence, small stocks might be included in a portfolio for different reasons than large stocks, and preferences are likely to depend on stock size. To capture these tracking error effects, we also partition the sample based on stock size. Table 2 reports the regression results for the top 25 stocks (i.e. largest) on the Australian Stock Exchange (on the basis of market capitalization) for which we had relevant data and Table 3 reports results for those stocks ranked outside the top 25. Variance is significant and positive for ex-top 25 stocks; however, there is no significant preference for volatility among top 25 stocks, as the coefficients are largely insignificant for these regressions. This suggests that managers view historical return variance as an indicator of mispricing for smaller stocks and that this represents an opportunity for them to exploit. The Size variables are significant and greater than one for both stock size partitions in the disaggregate regressions, indicating a preference for larger stocks within each subsample of securities (i.e. top 25 and ex-top 25). This result can be reconciled given the inclusion of zero-positions in our analysis of manager holdings. The Size coefficient is particularly large for the ex-top 25 sample, signifying that when active managers include small stocks in their portfolios, they are heavily overweight relative to benchmark weight. This is not surprising given that these stocks each comprise relatively small parts of the index, and managers might also need to trade these smaller stocks in round-lot quantities. The findings relating to analyst coverage for the entire sample are consistent with those relating to small stocks. The coefficients on the IBES variable are positive and significant across all years, suggesting that managers prefer stocks with analyst coverage and, therefore, lower information asymmetries. Perhaps of greater interest is the prudence issue. The ease of justification for a security that is covered by an analyst is likely to be far more relevant for smaller stocks, especially given the

13 S. Brands et al. / Accounting and Finance 46 (2006) Table 2 Regressions of active equity manager portfolio holdings of large stocks The regression model is as follows: Holding it = β0 + β1momentumposit + β2momentumneg i,t + β3varianceit + β4epsyit + β5spread it + β6sizeit + β7ibesit + β8ibessdit + J i=1 D8+iMit + eit. MomentumPos is the positive percentage change in the price of stock i over the previous 12 months in deciles and MomentumNeg is the negative percentage change in the price of stock i over the previous 12 months in deciles. Var is the natural log of the variance of daily stock price returns of stock i in the previous month. EPSY is the natural log of the earnings per share for stock i as a percentage of share price in deciles. Spread is the time weighted relative bid/ask spread for the month of September. Size is the log of the market capitalization of stocks expressed as a percentage of the S&P/ASX All Ordinaries Index in Panels A and B and the log of market capitalization in Panel C. IBES is the number of analysts covering the stock and IBES SD the standard deviation of the forecasts of these analysts. NObsis the number of observations in the sample. t-statistics are calculated based on White s heteroskedastic-consistent standard errors. Panel A refers to individual manager regressions, Panel B to the disaggregate regressions where Holding is the percentage weight of stock i in each of the manager s portfolios. Panel C refers to aggregate regressions where the dependent variable (Holding) consists of the aggregate number of shares of stock i owned by all funds at the given date divided by the total number of outstanding shares of stock i at that date. Year (number Positive Negative Variance EPS Spread Size IBES IBES SD in sample) momentum momentum Panel A: Individual Average coefficient 2000 (35) (38) Median coefficient 2000 (35) (38) Number of positive coefficients 2000 (35) (38) Number of negative coefficients 2000 (35) (38) Number of significant coefficients (5%) 2000 (35) (38)

14 182 S. Brands et al. / Accounting and Finance 46 (2006) Table 2 (continued) Year (number Positive Negative Variance EPS Spread Size IBES IBES SD in sample) momentum momentum Number of significant coefficients (10%) 2000 (35) (38) Panel B: Disaggregate Coefficient 2000 (35) t-statistic Coefficient 2001 (38) t-statistic Panel C: Aggregate Coefficient 2000 (35) t-statistic Coefficient 2001 (38) t-statistic ,, indicates statistical significance at the 1, 5 and 10 percent levels, respectively.

15 S. Brands et al. / Accounting and Finance 46 (2006) Table 3 Regressions of active equity manager portfolio holdings of small stocks The regression model is as follows: Holding it = β0 + β1momentumposit + β2momentumneg i,t + β3varianceit + β4epsyit + β5spread it + β6sizeit + β7ibesit + β8ibessdit + J i=1 D8+iMit + eit MomentumPos is the positive percentage change in the price of stock i over the previous 12 months in deciles and MomentumNeg is the negative percentage change in the price of stock i over the previous 12 months in deciles. Var is the natural log of the variance of daily stock price returns of stock i in the previous month. EPSY is the natural log of the earnings per share for stock i as a percentage of share price in deciles. Spread is the time weighted relative bid/ask spread for the month of September. Size is the log of the market capitalization of stocks expressed as a percentage of the S&P/ASX All Ordinaries Index in Panels A and B and the log of market capitalization in Panel C. IBES is the number of analysts covering the stock and IBES SD the standard deviation of the forecasts of these analysts. NObsis the number of observations in the sample. t-statistics are calculated based on White s heteroskedastic-consistent standard errors. Panel A refers to individual manager regressions, Panel B to the disaggregate regressions where Holding is the percentage weight of stock i in each of the manager s portfolios. Panel C refers to aggregate regressions where the dependent variable (Holding) consists of the aggregate number of shares of stock i owned by all funds at the given date divided by the total number of outstanding shares of stock i at that date. Year (number Positive Negative Variance EPS Spread Size IBES IBES SD in sample) momentum momentum Panel A: Individual Average coefficient 2000 (35) (38) Median coefficient 2000 (35) (38) Number of positive coefficients 2000 (35) (38) Number of negative coefficients 2000 (35) (38) Number of significant coefficients (5%) 2000 (35) (38)

16 184 S. Brands et al. / Accounting and Finance 46 (2006) Table 3 (continued) Year (no. Positive Negative Variance EPS Spread Size IBES IBES SD in sample) momentum momentum Number of significant coefficients (10%) 2000 (35) (38) Panel B: Disaggregate Coefficient 2000 (35) t-statistic Coefficient 2001 (38) t-statistic Panel C: Aggregate Coefficient 2000 (35) t-statistic Coefficient 2001 (38) t-statistic ,, indicates statistical significance at the 1, 5 and 10 percent levels, respectively.

17 S. Brands et al. / Accounting and Finance 46 (2006) previous observation that investors show a preference for volatile small stocks. The IBES SD variable is predominately negative for the ex-top 25 individual manager and disaggregate regressions, suggesting there is weak evidence that active managers prefer stocks for which there is a higher degree of consensus among analysts. Again, this is understandable from a prudency perspective, where active managers are concerned with the need to justify the inclusion of a stock in the portfolio. The results for large stocks suggest that analyst coverage and consensus in forecasts are not as important. This is not surprising given the belief that large stocks represent a more prudent investment Manager size Size partitions are also performed on the basis of funds under management as at the point of analysis. The active manager sample is divided into two equal size partitions, with one group containing the top half of the funds ranked by the fund size, whereas the other partition contains the remaining smaller funds. It is conceivable that small and large managers exhibit different preferences for stock characteristics on the basis of their organizational structure. Fund size also impacts on the number of unique stocks held in the portfolio (see Chan et al., 2004). We present the results only for the small manager size partition in Table 4, however we provide analytical comparison between the two groups in the text. 5 The important differences between large and small managers relate to the Variance and IBES variables. The coefficients of Variance, in both the aggregated and disaggregated regressions for small managers are of a greater magnitude than that for large managers, suggesting that smaller fund management organizations exhibit a greater preference for more volatile securities than large managers. This is likely to be as a result of smaller managers being less concerned, relatively speaking, with prudential constraints, and taking opportunities to profit from mispricing opportunities. Smaller managers also have a greater preference for stocks with high levels of analyst coverage and low standard deviation of forecasts (i.e. consensus among analysts) than large managers. This might be related to the stock volatility finding for small managers (i.e. a preference for more volatile stocks). Consistent with prudency concerns, if smaller managers are including more volatile securities in their portfolios, they might require stocks that are well followed by analysts to justify the inclusion of such securities. There are no obvious differences in the preferences of large and small managers for momentum, transaction costs, stock size and earnings yield. Falkenstein (1996) also partitions the sample on the basis of fund size, however no variation in the preferences of the two subsamples is reported. This might be a reflection of the difference in the nature of funds examined; 5 Given the constraints on journal space, the large manager results are omitted from the paper. These results are available on request from the authors.

18 186 S. Brands et al. / Accounting and Finance 46 (2006) Table 4 Regressions of small active equity manager portfolio holdings The regression model is as follows: Holding it = β0 + β1momentumposit + β2momentumneg i,t + β3varianceit + β4epsyit + β5spread it + β6sizeit + β7ibesit + β8ibessdit + j i=1 D8+iMit + eit. MomentumPos is the positive percentage change in the price of stock i over the previous 12 months in deciles and MomentumNeg is the negative percentage change in the price of stock i over the previous 12 months in deciles. Var is the natural log of the variance of daily stock price returns of stock i in the previous month. EPSY is the natural log of the earnings per share for stock i as a percentage of share price in deciles. Spread is the time weighted relative bid/ask spread for the month of September. Size is the log of the market capitalization of stocks expressed as a percentage of the S&P/ASX All Ordinaries Index in Panels A and B and the log of market capitalization in Panel C. IBES is the number of analysts covering the stock and IBES SD the standard deviation of the forecasts of these analysts. NObsis the number of observations in the sample. t-statistics are calculated based on White s heteroskedastic-consistent standard errors. Panel A refers to individual manager regressions, Panel B to the disaggregate regressions where Holding is the percentage weight of stock i in each of the manager s portfolios. Panel C refers to aggregate regressions where the dependent variable (Holding) consists of the aggregate number of shares of stock i owned by all funds at t divided by the total number of outstanding shares of stock i at that date. Year (number Positive Negative Variance EPS Spread Size IBES IBES SD in sample) momentum momentum Panel A: Individual Average coefficient 2000 (17) (19) Median coefficient 2000 (17) (19) Number of positive coefficients 2000 (17) (19) Number of negative coefficients 2000 (17) (19) Number of significant coefficients (5%) 2000 (17) (19)

19 S. Brands et al. / Accounting and Finance 46 (2006) Table 4 (continued) Year (number Positive Negative Variance EPS Spread Size IBES IBES SD in sample) momentum momentum Number of significant coefficients (10%) 2000 (17) (19) Panel B: Disaggregate Coefficient 2000 (17) t-statistic Coefficient 2001 (19) t-statistic Panel C: Aggregate Coefficient 2000 (17) t-statistic Coefficient 2001 (19) t-statistic ,, indicates statistical significance at the 1, 5 and 10 percent levels, respectively.

20 188 S. Brands et al. / Accounting and Finance 46 (2006) Falkenstein (1996) uses a sample of US mutual (i.e. non-pension) funds whereas the present study describes the preferences of Australian institutional (i.e. pensionoriented) funds. It is conceivable that our sample of institutional funds are more susceptible to prudency concerns, as US evidence suggests that mutual funds exhibit lower concern with prudent portfolio management (see Del Guercio, 1996) Industry classification To determine whether an industry effect exists in the stock characteristic preferences for our sample of managers, we also performed tests using eight sector dummy variables in the regressions (see equations (2) and (3)). Firms were classified into one of nine GICS sectors (energy, materials, industrials, consumer discretionary, consumer staples, health care, financials, telecommunication services, utilities). In unreported results, all coefficients on the sector dummies were significant for the disaggregate regressions, which provides evidence that industry effects on the stock holdings of Australian institutional investors are indeed present and important in the portfolio construction process Conclusion The present study examines the portfolio preferences for stock characteristics of actively managed Australian equity managers in the institutional market. The literature suggests that stock characteristics, beyond traditional proxies for risk, capture a wider spectrum of factors that explain the inclusion of stocks in active fund portfolios (Falkenstein, 1996; Gompers and Metrick, 2001 and Pinnuck, 2004). Our study examines the decision to include a security in a fund manager s portfolio on the basis of performance and prudential factors. Performance-maximizing managers are expected, ceteris paribus, to invest in stocks with lower transaction costs. Consistent with theory, we find evidence that active managers indeed exhibit a significant preference for stocks with small relative bid/ask spreads. It is apparent that managers prefer stocks with high levels of analyst coverage and, therefore, lower information asymmetries. This is also consistent with prudent management concerns, where the inclusion of a security also having analyst coverage is likely to be much easier to justify. Furthermore, we find the existence of preferences for stocks where a higher degree of consensus (earnings) exists among analysts. Despite the findings of Jegadeesh and Titman (1993) relating to the existence of positive serial correlation among short-horizon US security returns, the active funds in our sample do use contrarian strategies in large stocks (but not across 6 Again, because of journal space constraints, these results are removed from the paper, but are available on request from the authors.

21 S. Brands et al. / Accounting and Finance 46 (2006) their entire portfolios). Consistent with Falkenstein (1996), we also document preferences for higher volatility stocks. Our results concerning volatility are somewhat consistent with the changing preferences of institutional investors documented in the USA (see Bennett et al., 2003), but are different to those reported by Pinnuck (2004). Active managers are also found to prefer stocks with larger market capitalizations, suggesting that managers are concerned with both prudence considerations and tracking error considerations. Variance is of greater importance among smaller stocks, suggesting managers view historical return variance as an indicator of mispricing for smaller stocks. A higher degree of analyst coverage, as well as consensus in earnings forecasts, appears to be of greater importance for smaller stocks, perhaps reflecting the need to justify the inclusion of such securities in the portfolio. Small managers have a preference for more volatile securities, which is likely to be because of fewer concerns regarding prudential portfolio management. However, active managers also have a preference for stocks with high levels of analyst coverage and consensus of forecasts, which is likely to be linked to our findings concerning stock volatility. We also confirm the existence of an industry effect across Australian institutional investors, whereby active managers exhibit preferences for stocks classified according to the GICS industry classification system. References Aitken, M., and A. Frino, 1996, The determinants of market bid ask spreads on the Australian stock exchange: cross-sectional analysis, Accounting & Finance 36, Badrinath, S., G. Gay, and J. Kale, 1989, Patterns of institutional investment, prudence and the managerial safety net hypotheses, Journal of Risk and Insurance 56, Bennett, J., R. Sias, and L. Starks, 2003, Greener pastures and the impact of dynamic institutional preferences, Review of Financial Studies 16, Chan, L., H. Chen, and J. Lakonishok, 2002, On mutual fund investment styles, Review of Financial Studies 15, Chan, H., R. Faff, D. Gallagher, and A. Looi, 2004, Fund size, fund flow, transaction costs and performance: size matters, working paper (The University of New South Wales, Sydney). Chen, H., N. Jegadeesh, and R. Wermers, 2000, The value of active mutual fund management: an examination of the stockholdings and trades of fund managers, Journal of Financial and Quantitative Analysis 35, Covrig, V., S. T. Lau, and L. K. Ng, 2001, Do domestic and foreign fund managers have similar preferences for stock characteristics? A cross-country analysis, working paper (Nanyang Technological University and University of Wisconsin at Milwaukee). Del Guercio, D., 1996, The distorting effect of the prudent-man laws on institutional equity investments, Journal of Financial Economics 40, Demir, I., J. Muthuswamy, and T. Walter, 2003, Momentum returns in Australian equities: the influences of size, risk, liquidity, and return computation, Pacific-Basin Finance Journal 12, Falkenstein, E., 1996, Preferences for stock characteristics as revealed by mutual fund portfolio holdings, Journal of Finance 51, Gompers, P., and A. Metrick, 2001, How are large institutions different from other investors?, Quarterly Journal of Economics, 16,