Scholarly article on topic 'Mutual fund performance: A synthesis of taxonomic and methodological issues'

Mutual fund performance: A synthesis of taxonomic and methodological issues Academic research paper on "Economics and business"

Share paper
Academic journal
IIMB Management Review
OECD Field of science

Abstract of research paper on Economics and business, author of scientific article — S.G. Badrinath, Stefano Gubellini

Abstract This paper provides a comprehensive taxonomy of mutual funds and discusses the relative importance of these fund types. While most academic research focuses on US equity funds, we provide results for many more asset classes with this taxonomy—fixed income, balanced, global, International, sector, market-neutral and long-short funds. For each, we start by reporting statistics on the number of funds and their total net asset values at different intervals over the last four decades. We then identify short and long-term patterns in annual returns to mutual funds. We study the cross-sectional and time-series properties of the distribution of investor flows into different types of mutual funds, describe the relationship between flows and performance and discuss its implications for the strategic behaviour of managers and investors. We estimate and interpret fund performance alphas using both the single-factor and four-factor Fama-French models for each taxonomy type. Finally we describe the state of academic research on portfolio performance evaluation tilted towards an applied audience.

Academic research paper on topic "Mutual fund performance: A synthesis of taxonomic and methodological issues"

IIMB Management Review (2010) 22, 147-164

available at


journal homepage:


Mutual fund performance: A synthesis of taxonomic and methodological issues

S.G. Badrinath*, Stefano Gubellini

San Diego State University, 5500 Campanile Drive, San Diego, CA 92182-8236, United States Available online 23 October 2010

Abstract This paper provides a comprehensive taxonomy of mutual funds and discusses the relative importance of these fund types. While most academic research focuses on US equity funds, we provide results for many more asset classes with this taxonomy—fixed income, balanced, global, International, sector, market-neutral and long-short funds. For each, we start by reporting statistics on the number of funds and their total net asset values at different intervals over the last four decades. We then identify short and long-term patterns in annual returns to mutual funds. We study the cross-sectional and time-series properties of the distribution of investor flows into different types of mutual funds, describe the relationship between flows and performance and discuss its implications for the strategic behaviour of managers and investors. We estimate and interpret fund performance alphas using both the single-factor and four-factor Fama-French models for each taxonomy type. Finally we describe the state of academic research on portfolio performance evaluation tilted towards an applied audience.

© 2010 Indian Institute of Management Bangalore. All rights reserved.

* Corresponding author.

E-mail addresses: (S.G. Badrinath), (S. Gubellini).

1 Various versions of the Mutual Fund Fact Book, 1996-2007, report a range of $47.6 B to $12 trillion over this period. The numbers we report are taken from our sample from the CRSP -Survivor Bias-Free Mutual Fund database and confirm the general trend. Full details appear in Table 1, Panel B. 0970-3896 © 2010 Indian Institute of Management Bangalore. All rights reserved. Peer-review under responsibility of Indian Institute of Management Bangalore. doi:10.1016/j.iimb.2010.08.002


From 1970 to 2007, US mutual fund assets have grown at an annual rate of 16% from $43 billion in 1970 to nearly $11 trillion at the end of 2007.1 Nearly 44% of US households own mutual funds to participate in the purported benefits of portfolio diversification and manager skills as they attempt to secure a comfortable retirement. Globally, the US and Europe account for 80% of the assets under mutual fund management, with Asia and the Pacific bringing up the remainder. In India, assets under management have grown from Rs 24.67 crore (246.7 million) in 1965 to Rs. 341,378 crore (3413.78 billion) at the end of 2006, which is roughly a 26% growth rate. Higher growth rates in India are to be expected in a region where industry assets are still growing and only 2.2% of the population invests in mutual funds.2 In

2 See Sankaran (2007), Annexure 4.1, p. 57.

the US the rate of growth in the number of mutual funds appears to have slowed from about 15% per year in the 1980s to 10% annually in the 1990s to barely 0.01% annually in the 2000s. Most of the growth in the last decade is in exchange-traded funds because of their purported cost efficiency, from about 100 funds in 2000 to about 700 in 2007. The financial crisis of the last two years, the frequency with which such 'black swan' events appear to occur, the implications for the future regulatory landscape and the damage done to retirement portfolios has caused many to revisit the role and relevance of these investment vehicles in financial economics.

In response to this growth in net mutual fund assets (and perhaps because of it), the academic finance community now devotes a separate electronic journal to disseminate research on mutual funds, hedge funds and the investment industry.3 Mutual fund research uses and contributes to many of the central questions in our discipline, from multi-factor asset pricing models to behavioural finance. Scholars in the area continually develop, assess and improve models to measure mutual fund performance and examine how the behaviour of portfolio managers and investors impacts (and is impacted by) that performance.

To this growing literature, we believe that this paper makes several unifying contributions. First, we provide a more comprehensive taxonomy of the mutual fund universe — using (and improving upon) various versions of mutual fund objective codes made available in the CRSP — Survivor-Bias-Free Mutual Fund database.4 Our taxonomy enables us to categorise a whole range of asset classes — equity funds, fixed income funds, balanced funds, income funds, sector funds, value funds and long-short funds — although equity funds have attracted the largest academic interest. As a maturing market, a US based taxonomy has the merit of being exhaustive and can be easily adapted to Indian and emerging market settings. A scan of offer documents from Indian mutual fund schemes shows some similarity with the investment objectives that we classify in our taxonomy and we are confident that the methodological issues we discuss in this paper will transfer directly to an Indian context.5

Second, for each taxonomy group, we start by providing snapshots of total net assets at different points over the last four decades as well as snapshots of raw return performance at different investment horizons. The latter enables us to describe both cross-sectional and time-series patterns in simple ways. Short-term momentum in returns and long-run mean reversion are visible in our data. Third, we estimate net cash flows to different categories of mutual funds, discuss the distributional characteristics of

3 A search on the Financial Economics Network for'mutual fund performance' results in 673 hits.

4 Carhart, Carpenter, Lynch, and Musto (2002) document that the average annual attrition rate for mutual funds is about 3.6% per year over the period 1962—1995. For samples chosen with only surviving funds, the bias in annual performance estimates varies with sample length from 0.07% for 1-year samples to about 1% for 15-year samples.

5 In India, equity funds label themselves as growth or aggressive

growth, with goals that include long-term capital appreciation. Sector funds and balanced funds in India also appear to model

themselves along the lines of their Western counterparts.

cross-sectional flows and provide snapshots of the time-series properties of these flows.6 Some broad investment trends are evident and we are able to identify periodic asset-reallocations consistent with turning points in the economic climate. Fourth, we present portfolio performance results for different (and important) subsets of the mutual fund universe, with both single-factor and four-factor Fama—French models. Fifth, we provide an evolutionary and critical appraisal of the state of the mutual fund performance literature. In it we describe the rich set of hypotheses that arises from an examination of the strategic behaviour of both managers and investors along with directions for further research.

Broadly, our results on the return performance and flow behaviour give us some confidence in the discriminatory power of our taxonomy structure. While these employ US data, the methods we use to develop portfolio performance and fund flow estimates are easily replicable and applicable to mutual fund regimes in India and other markets. After all, the patterns of variation in flows and returns that we document arise from ostensibly value-maximising behaviour on the part of market participants — a motivation that is global even if developments and regulations in the mutual fund industry are local. Therefore, we expect that they should hold across regions.

We believe that our results and the accompanying survey of the mutual fund literature are of significant value to several clienteles. Scholars beginning mutual fund research may benefit from our taxonomy and descriptions of the database in their own work. Investment professionals should find our discussion of flows and the patterns visible in them to be of value in making their investment and security selection decisions. A possible use for our taxonomy is to assist portfolio managers in tailoring the construction of performance benchmarks to more closely match their investment styles. The comprehensive results and summaries that we provide can benefit teachers promoting financial literacy in the classroom or similar forums.

The rest of this paper is organised into four sections. The second section describes our efforts toward creating a unified taxonomy for mutual funds. The third section describes the returns to different fund types, discusses the flows of funds and documents single-factor and four-factor performance. The fourth section provides a discussion of the evolution of portfolio performance measurement in the academic finance literature and the fifth section concludes the paper.


This section develops our mutual fund taxonomy. Our primary database is the CRSP Survivor-Bias-Free Mutual Fund database 2007.7 On this database, available time-

6 These flows are inferred from monthly changes in fund net asset values and are the net of inflows over outflows.

7 Elton et al. (2001) examine the accuracy and comparability of the information provided on this database. Many of their concerns are being addressed in subsequent versions of this database and CRSP has done an excellent job of disseminating their'fixes' through quarterly updates. CRSP also updates the database on a regular basis, adding new funds to the universe and back-filling historical data on existing funds.

series of mutual fund records are indexed by a number which is unique for each share class. Share classes represent a myriad array of expense structures, characterised by front-end and back-end loads, and specific portfolios tailored towards financial advisors as well as institutional and retail clients.8 Our database, extracted from the 2007 version of the CRSP database contains 32798 unique share class records.9

Issues in fund style classification and aggregation

A survey of the academic finance literature on mutual funds reveals some variation in the samples of mutual funds used by different scholars. This is partly because sample sizes have been increasing with time. They also differ according to the quality of the fund classification codes available at the time each such study was conducted. Both the quality of fund classification code providers and their number have increased along with investor interest. CRSP initially provided fund-level S&P and Morningstar objective codes, then included codes from Wiesenberger, from Strategic Insight and most recently those provided by Lipper Associates. Sequential versions of the CRSP mutual fund database provide more detail on fund objectives. Clienteles served by these providers also differ, with Lip-per's services being largely institutional while Morningstar categories are the most popularly used by the retail investor. There is considerable variation in both the depth and the breadth of information available from different providers. Wiesenberger provides 27 codes that enable a classification of only 37% of the share classes in the database. Of these 27 codes, 15 are related to fixed-income securities and industry sectors. Strategic Insight provides 193 classification codes, of which 136 pertain to the fixed income group.10 These codes are available for only about 10% of the share classes in the database. Lipper Associates provides 166 different objective codes and is perhaps the most complete of the available sources with about 70% of the share classes carrying an objective code.11

We additionally examine the name of the individual share class to assist in making a classification assignment. Share class names carry a wealth of information. They identify whether the underlying fund is purely international, purely

8 Nanda, Wang, and Zheng (2005) find that the introduction of new share classes results in new investor dollars being allocated to those share classes for two years after which the expectation of poor performance causes those investors to exit.

9 By comparison, the 2008 Mutual Fund Fact Book published by the Investment Company Institute reports 21631 share classes which compares closely with the share classes in our database which have survived.

10 Some of the older share classes that did not survive till the end of our sample period may reflect a'coarser'than desirable taxonomy classification and/or a classification from only one provider who followed that share class.

11 Our assignment of names and objective codes is carried out using

the most recent value of that variable in the time-series for each

share class. Inspection of the data suggests that name changes and objective code changes within that time series are rare.

domestic or global (a mix of the two). Names can identify an investment style for the fund, indexed, income (dividend, bond, or high-yield) or total return, aggressive, moderate or conservative, long-short, neutral, bearorenhanced, growth, value or both. Names can also describe the class of investment being undertaken—fixed income (government, municipal or corporate), equity, or balanced. Names often indicate the type of securities comprising the portfolio — these can be by equity market capitalisation—micro-, small-, mid-, large- or multi-cap, by region, country or sector, by maturity or duration—short-, intermediate- or long-term. Various combinations of the above styles, classes and security types also appear for different share classes on the database. A computerised text search through the name records enables us to capture combinations of alphanumeric characters that are embedded in the name string and point to the same type of investment objective.12 Some names such as the Weingarten fund are not informative, while ambiguity in classification remains with other names such as the MFS New Discovery fund.13

We also conducted a random check on the accuracy of our name classification by examining the prospectus of the underlying mutual fund and found no significant errors. Wherever possible, we use this process to fill in the classification code information for those cases when it could not be inferred from an ambiguous name and when there was no classification code from the providers listed above.

In sum, an inspection of the name strings for individual funds in conjunction with other commonly used classification schemes permits a better assignment of mutual funds into their 'correct' categories. This enables a more comprehensive taxonomy, including debt funds, sector funds, enhanced or 'leveraged' funds. Our taxonomy is far more specific for debt funds than for equity funds reflecting different levels of product differentiation in those two asset classes. It is also more accurate in identifying sectors. The latter contribution is particularly noteworthy in comparative performance studies of passively constructed exchange-traded funds (ETFs) against their actively managed mutual fund peers—a study that is ongoing.

We recognise that even this level of detail does not enable a perfect classification, that names and styles can change over time and that some managers may obfuscate or even misrepresent their investment styles. The potential unreliability of stated style objectives causes Wei, Wermers, and Yao (2009) to identify funds as 'contrarian' by constructing 'anti-herding' measures from the characteristics of the component stocks in a portfolio. There does

12 In the database field where the fund name appears, the data record often contains words that are abbreviated according to the whim of the data provider. For instance, the term'small capitalisation' is described variously in the records as: smcp, sm cp, sm-cp, SmCap, SmCap, Small-Cap, Small Company. These abbreviations also appear at various positions in the records, requiring us to develop a computer programme to exhaustively search through the name records for all such possible combinations.

13 Even when the names are informative, some decisions regarding the hierarchy of different name strings are required to be made as for instance, the Dreyfus premier small cap growth equity fund.

seem to be some convergence in these reporting practices and we are heartened by Lipper's efforts to identify the characteristics of the common stocks in a fund's portfolio while making a classification. As an example, in describing their small-cap value group, they specify broad ranges for equity market capitalisation, for the price-earnings ratio, for the price-to-book ratio and for three-year per share sales growth. While an examination of the holdings of the fund can be used for style attribution, some amount of subjectivity will always remain in any broad taxonomy. For instance, Standard Industrial Classification (SIC) codes to classify firms into industries remain popular among scholars despite misgivings about their accuracy.

Despite these considerations, an accurate identification of share class is merely a first step. In most studies of performance, the common unit of interest is typically the mutual fund portfolio itself and this requires us to aggregate the different share classes in that fund. Our version of the database provides a portfolio number for about 58% of the available share classes. For the remainder, we match the names of the share classes and identify 13424 such unique mutual funds associated with the universe of share classes.14 About 47% of these funds have just one share class associated with them.15

Once the mutual fund with its various share classes is identified, there is a second level of aggregation by investment style which is common in samples of funds chosen for academic study. We find that these procedures are also not easily comparable. Some studies include global funds, many exclude sector funds, international funds and hybrid funds, and most exclude fixed-income funds. Moreover, after funds are selected by these criteria, they are further aggregated into broad classes such as growth, aggressive growth, income, and balanced funds. While these 'super categories' are in accord with common perceptions, empiricists can only hope that aggregation will not cancel out any systematic variations in the data.

Still, the broad criterion for including funds into any sample appears to be that of a significant domestic equity exposure. Several considerations are relevant in making this choice. First, global funds comprising both US and foreign firms appear in most samples studied. Since the pricing benchmarks used to measure portfolio performance are obtained from US data, the inclusion of global funds can bias the resulting estimates since some portion of their underlying portfolio is invested in firms with less sensitivity to these benchmarks. On the other hand it can also be argued that companies like Coca Cola domiciled in

14 Mutual fund mergers are fairly few with 6000 share class months out of 2.6 million in the mutual fund database being potentially affected. Multiple mergers in the same date characterise about 10% of the cases and less than 5% of overall mergers take place over the period 1970-1990. To treat mergers, we use the'follow-the-money' approach of Gruber (1996) by assuming that investors continue to invest with the surviving fund.

15 Multiple share classes appear to be more common for debt and

balanced funds which find homes in the retirement portfolios of many investors. Still about 86% of the funds in our sample have fewer than five share classes. The maximum number of share classes for a fund in our universe is 14.

the US have a substantial global presence as do companies like Roche that are domiciled overseas. While the decision to include or exclude global funds is researcher and project specific, it is important to classify such firms accurately in developing a meaningful and replicable taxonomy. Second, the exclusion of fixed income funds is common in the literature. By this criterion, many exclusively bond funds that investors would choose are absent from chosen samples, while balanced funds with some portion of their assets in debt securities are included. In addition to being somewhat arbitrary, this also restricts the sample size of funds in that subgroup. Third, value funds do not appear to merit special consideration, although the stated goal of value managers is often very different from that for growth funds and these goals are widely understood. Fourth, short, long-short and market-neutral funds with a significant portion of their holdings making downside bets, often with derivative products, are largely ignored.

Our taxonomy

The typical process that we follow is to infer an objective code from the name for a fund and to cross-check it with the Lipper code, the Strategic Insights code or the Wie-senberger code depending on which are available for the fund in question. While objective code descriptions from the latter two providers are not particularly informative, we find the supplementary information in the Lipper codes to be occasionally meaningful.16 However, even Lipper classifies funds that track equity sub-indices as growth funds while a name search correctly identifies them as passively managed. With sector funds and some fixed-income funds, our name matching rules also permit finer partitions of the universe of funds than those available with these objective code providers. In generating a taxonomy code, we provide detail up to four digits similar to the practices of the North American Industry Classification System (NAICS) for industry groups.

At the top level, funds are identified as domestic, global or international. At the next level, domestic funds are differentiated along 10 groups (small-cap, mid-cap and large-cap growth equity, growth and income, income, balanced, value, fixed income, sector funds and short-biased funds). The range of product offerings from fund families permits still finer levels of detail. Aggressive growth funds are separated into small-cap core funds, small-cap growth, micro-cap funds or other. Fixed-income funds are classified as government, municipal, corporate, money market, and within each, further distinguished by term to maturity or ratings level. Both international and global funds can be fixed income, equity or balanced.

We recognise that the taxonomy has widened as the mutual fund industry has grown. An examination of our

16 We individually examine cases where different procedures provide conflicting taxonomy assignments and classify funds based on our judgment of the relative values of these schemes. We defer to the Lipper classification if the supplementary information in their code description has a quantitative component.

database reveals that fixed-income mutual funds grew in number and type during the 1980s. The 1990s bull market created further product differentiation in the equity space with funds separated by investment style—value funds, growth funds and sector funds. In the latter half of this period, international, global, and emerging market funds became available to investors. As the fund industry matured in the 2000s, mutual funds that employed hedge fund like long-short, short and ultra-short strategies began to proliferate.

We view this taxonomy as a basis for evaluating dynamic changes to investor portfolios and offer it as a road-map for scholars considering research in this area in both developed and emerging markets. Since the US mutual fund market is the largest in the world and perhaps the most mature, we believe that a taxonomy structure using this universe would provide a good template to encompass the different types of mutual fund offerings in other locations. It partitions the mutual fund universe qualitatively by market capitalisation, by investment style, and by asset class, which are crude proxies for the type of risk in the underlying portfolios. As available, it aggregates and re-classifies objective codes assigned by information service providers in the US. One benefit of using this taxonomy is that its coarse descriptions of fund objectives are less likely to be influenced by changes in the manager's investment style and consequent exposure to portfolio risk. Despite being exhaustive, it does not however suffice to determine if the mutual fund product is 'suitable' for the risk tolerances of individual fund investors. For that purpose, a ratings box along the lines of Morningstar in the US or a Composite Performance Rating (CPR) from Crisil in India would be more appropriate.

The results below give us some confidence that our taxonomy generates reasonable (and expected) cross-sectional variation in terms of returns, flows and performance. (A complete listing of our taxonomy codes appears in the appendix.)

Results from our analysis Data

Armed with this taxonomy, we then report characteristics of the component fund categories. First we report basic statistics on the number of funds and assets under management by taxonomy type in Table 1. For Table 1, the fund categories are chosen in accordance with common investor perceptions of fund types. In later tables, we report information on the 12 categories described in the section 'our taxonomy'. In Table 2, we report past portfolio return performance over different horizons. In Table 3, we document patterns in net cash flows to these funds. Finally we provide performance results in Table 4 using both the single-factor CAPM and four-factor Fama—French models.

Estimation of rates of return and the treatment of missing values is important. Our version of the database has 2.97 million observations over the period 1961—2007. Available time-series of mutual fund observations can be annual, quarterly or monthly depending upon the reporting

requirements prevalent at the time of the inclusion of a share class into the CRSP database. For our purposes, the variables of interest are Net asset value (NAV), returns, total net asset (TNA) values, and shares outstanding. Holding period returns are typically obtained from changes in NAV with adjustments made for periodic redemptions and distributions. We explore patterns of discontinuities in the available time-series of these variables and impute missing observations with a view to preserving as much of the monthly-level data as possible. Over the period 1970—1990 the CRSP database reports NAV and returns monthly, while TNA is reported mostly at quarterly frequency. Share class months are not included in our sample when both return and TNA value for that month are unavailable. When the monthly return and NAV are available, we compute the corresponding TNA assuming that the number of shares (for that particular class) remains constant in the quarter of interest.17 Sample sizes before 1970 are extremely small and objective codes when available are not very unreliable for funds in that period. Therefore, we restrict our final sample to 2.67 million monthly observations for the period 1970—2007.

Results for the mutual fund sample

Generally, the data we report in Table 1, both in terms of the number of funds and TNA value is similar to that reported by the Investment Company Institute in various editions of their Mutual Fund Fact Books.18 Sample sizes are also in accordance with those reported in prior studies at the time they were conducted. Several observations are of interest. First, the increase in differentiated product offerings from the mutual fund industry is visible, with sector and short-biased funds as more recent entrants into the mutual fund stable. Growth funds and fixed income funds are the largest in number although several of the other taxonomy groups are well populated. The amount of funds committed to these two groups over time also supports the creation and examination of finer taxonomies. As an illustration, in later tables, we report results separately for small-cap, mid-cap and large-cap equity


Second, nearly 45—50% of the assets under management in the last few decades are in the fixed income, balanced and money-market space, and clearly reflect the conservative investment philosophy and retirement focus that motivates investment in these asset classes.20 Third,

17 In the period 1970—1990, our approach preserves between 80% and 90% of the monthly candidates for calculating FLOWS. We test the assumption that shares outstanding are constant within a quarter with other cases where data is available and find it to be reasonable.

18 Despite controlling for survivor-bias, our numbers are slightly smaller than those in the Mutual Fund Fact Books, since we exclude index funds and funds with missing returns data.

19 For brevity we do not report other partitions of the data. They are available upon request.

20 Estimates for the fixed income component are obtained by cumulating net assets invested in balanced, fixed-income, and money market funds in 2000 and 2007.

Table 1 Number of funds and total net asset value.

Panel A: number of mutual fundsa

Year 1970 1980 1990 2000 2007

N Prop N Prop N Prop N Prop N Prop

Domestic small-cap growth equity 20 (7.5) 21 (5.7) 85 (3.9) 392 (6.6) 437 (7.7)

Domestic mid-cap growth equity 19 (7.1) 21 (5.7) 65 (2.9) 262 (4.4) 289 (5.1)

Domestic large-cap growth equity 65 (24.3) 76 (20.7) 158 (7.2) 650 (10.9) 448 (7.9)

Domestic all-cap growth equity 41 (15.4) 47 (12.8) 170 (7.7) 413 (6.9) 462 (8.2)

Domestic growth and income 9 (3.4) 10 (2.7) 82 (3.7) 77 (1.3) 63 (1.1)

Domestic income 4 (1.5) 5 (1.4) 39 (1.8) 77 (1.3) 86 (1.5)

Domestic balanced 33 (12.4) 37 (10.1) 127 (5.8) 374 (6.3) 631 (11.1)

Domestic value 33 (12.4) 38 (10.4) 131 (5.9) 438 (7.3) 471 (8.3)

Domestic short-biased (0.0) (0.0) 3 (0.1) 33 (0.6) 70 (1.2)

Domestic fixed income 25 (9.4) 89 (24.3) 1015 (46.0) 1906 (31.9) 1528 (27.0)

Domestic- all sectors 9 (3.4) 11 (3.0) 133 (6.0) 451 (7.5) 380 (6.7)

International equity 5 (1.9) 6 (1.6) 97 (4.4) 643 (10.8) 552 (9.8)

International debt (0.0) (0.0) 5 (0.2) 54 (0.9) 51 (0.9)

Global equity 4 (1.5) 6 (1.6) 42 (1.9) 127 (2.1) 127 (2.2)

Global growth, income, balanced (0.0) (0.0) 17 (0.8) 28 (0.5) 31 (0.5)

Global debt (0.0) (0.0) 36 (1.6) 51 (0.9) 34 (0.6)

Global sectors (0.0) (0.0) (0.0) 2 (0.0) 1 (0.0)

All actively managed funds 267 (100.0) 367 (100.0) 2205 (100.0) 5978 (100.0) 5661 (100.0)

Money market funds 16 625 910 707

Index funds 1 28 221 209

All funds 267 384 2858 7109 6577

Panel B: Average TNA in billionsb

Avg Prop Avg Prop Avg Prop Avg Prop Avg Prop

Domestic small-cap growth equity 0.6 (1.5) 1.6 (2.8) 6.0 (1.1) 116.5 (2.9) 237.6 (3.3)

Domestic mid-cap growth equity 3.1 (7.2) 3.6 (6.3) 10.6 (2.0) 190.8 (4.8) 252.1 (3.5)

Domestic large-cap growth equity 17.4 (40.6) 18.9 (33.0) 69.8 (12.9) 1036.8 (25.9) 687.0 (9.6)

Domestic all-cap growth equity 3.6 (8.4) 3.7 (6.5) 26.7 (4.9) 469.5 (11.7) 868.1 (12.1)

Domestic growth and income 0.4 (0.8) 0.5 (0.9) 4.3 (0.8) 39.0 (1.0) 70.0 (1.0)

Domestic Income 0.1 (0.2) 0.2 (0.4) 8.7 (1.6) 83.6 (2.1) 159.2 (2.2)

Domestic balanced 7.6 (17.8) 5.3 (9.3) 32.0 (5.9) 274.9 (6.9) 949.4 (13.2)

Domestic value 7.8 (18.2) 9.6 (16.8) 53.4 (9.8) 381.6 (9.5) 862.4 (12.0)

Domestic short-biased 0.0 (0.0) 0.0 (0.0) 0.1 (0.0) 1.6 (0.0) 21.1 (0.3)

Domestic fixed income 0.9 (2.1) 9.5 (16.7) 279.5 (51.5) 701.7 (17.5) 1263.0 (17.6)

Domestic- all sectors 1.2 (2.8) 2.0 (3.5) 15.2 (2.8) 231.8 (5.8) 263.6 (3.7)

International equity 0.1 (0.3) 0.2 (0.4) 13.3 (2.5) 273.4 (6.8) 1022.4 (14.3)

International debt 0.0 (0.0) 0.0 (0.0) 0.8 (0.1) 8.6 (0.2) 43.6 (0.6)

Global equity 0.1 (0.3) 2.0 (3.4) 12.9 (2.4) 169.8 (4.2) 358.0 (5.0)

Global growth, income, balanced 0.0 (0.0) 0.0 (0.0) 1.7 (0.3) 15.0 (0.4) 78.8 (1.1)

Global debt 0.0 (0.0) 0.0 (0.0) 7.8 (1.4) 10.2 (0.3) 35.2 (0.5)

Global sectors 0.0 (0.0) 0.0 (0.0) 0.0 (0.0) 0.2 (0.0) 1.4 (0.0)

All actively managed funds 42.9 (100) 57.2 (100) 542.9 (100) 4005.0 (100) 7172.9 (100)

Money market funds 0.0 13.8 407.5 1646.4 2686.7

Index funds 0.0 0.1 4.6 331.4 732.7

All funds 42.9 71.1 955.0 5982.8 10592.3

Figures in parentheses represent the proportion ('Prop') of all active funds in that taxonomy group. a Aggregate statistics on the number of funds ('N') and the total net asset value of these funds at 10-year intervals through the period 1970-2007 for our broad taxonomy groups. b Aggregate statistics on the average Total Net Asset value ('Avg') of these funds at 10-year intervals through the period 1970-2007 for our broad taxonomy groups.

academic finance research concentrates primarily on diversified equity mutual funds and typically excludes debt funds, sector funds and international funds. This would imply that manager performance for a substantial portion

of assets under management has not attracted much research interest. Fourth, the proportion of funds invested in the different fund groups permits some preliminary inferences regarding investor choices. Assets in equity

Table 2 Average annual returns and future values.


Average annual returns for 1 year ending ina 1980 1990 2000 2007

Domestic small-cap growth equity 0.51 -0.11 -0.01 0.04

Domestic mid-cap growth equity 0.41 -0.07 -0.07 0.12

Domestic large-cap growth equity 0.33 -0.04 -0.10 0.10

Domestic growth and income 0.33 -0.03 -0.08 0.07

Domestic income 0.25 -0.10 0.10 0.03

Domestic balanced 0.20 -0.02 0.05 0.07

Domestic value 0.27 -0.07 0.10 0.02

Domestic short-biased funds 0.01 0.13 0.04

Domestic fixed income 0.00 0.05 0.08 0.04

Domestic all sectors 0.48 -0.05 -0.04 0.11

International all 0.28 -0.08 -0.14 0.18

Global all 0.30 -0.05 -0.07 0.12

NYSE composite index 0.33 -0.04 0.04 0.07

Average annual returns for 3 years ending in 1980 1990 2000 2007

Domestic small-cap growth equity 0.33 0.08 0.11 0.08

Domestic mid-cap growth equity 0.29 0.10 0.17 0.12

Domestic large-cap growth equity 0.22 0.13 0.14 0.09

Domestic growth and income 0.20 0.11 0.12 0.09

Domestic income 0.19 0.08 0.09 0.09

Domestic balanced 0.13 0.10 0.08 0.09

Domestic value 0.20 0.11 0.09 0.09

Domestic short-biased funds 0.00 0.13 -0.03 0.04

Domestic fixed income 0.02 0.08 0.04 0.04

Domestic all sectors 0.38 0.09 0.20 0.15

International all 0.20 0.08 0.10 0.20

Global all 0.26 0.09 0.13 0.14

NYSE composite index 0.20 0.13 0.11 0.11

Average annual returns for 5 years ending in 1980 1990 2000 2007

Domestic small-cap growth equity 0.25 0.05 0.14 0.16

Domestic mid-cap growth equity 0.20 0.10 0.16 0.17

Domestic large-cap growth equity 0.16 0.12 0.17 0.12

Domestic growth and income 0.17 0.10 0.18 0.13

Domestic income 0.20 0.07 0.15 0.13

Domestic balanced 0.12 0.10 0.12 0.11

Domestic value 0.17 0.11 0.15 0.14

Domestic short-biased funds 0.00 0.10 0.00 -0.01

Domestic fixed income 0.06 0.08 0.05 0.05

Domestic all sectors 0.25 0.13 0.20 0.20

International all 0.15 0.18 0.10 0.24

Global all 0.20 0.11 0.14 0.18

NYSE composite index 0.16 0.12 0.17 0.15

Average annual returns for 10 years ending in 1980 1990 2000 2007

Domestic small-cap growth equity 0.11 0.07 0.17 0.08

Domestic mid-cap growth equity 0.11 0.10 0.18 0.08

Domestic large-cap growth equity 0.09 0.12 0.17 0.05

Domestic growth and income 0.11 0.12 0.17 0.06

Domestic income 0.12 0.13 0.16 0.07

Domestic balanced 0.08 0.13 0.13 0.07

Domestic value 0.10 0.14 0.16 0.08

Domestic short-biased funds 0.00 0.00 0.02 0.02

Domestic fixed income 0.06 0.11 0.07 0.05

Domestic all sectors 0.13 0.10 0.18 0.11

International all 0.07 0.17 0.10 0.11

Global all 0.12 0.13 0.12 0.09

NYSE composite index 0.09 0.13 0.17 0.07

(continued on next page)

Table 2 (continued).


FV of $10000 invested for 3 years ending in

Domestic small-cap growth equity

Domestic mid-cap growth equity

Domestic large-cap growth equity

Domestic growth and income

Domestic income

Domestic balanced

Domestic value

Domestic short-biased funds

Domestic fixed income

Domestic all sectors

International all

Global all

NYSE Composite Index

FV of $10000 invested for 5 years ending inb

Domestic small-cap growth equity

Domestic mid-cap growth equity

Domestic large-cap growth equity

Domestic growth and income

Domestic income

Domestic balanced

Domestic value

Domestic short-biased funds

Domestic fixed income

Domestic all sectors

International all

Global all

NYSE composite index

FV of $10000 invested for 10 years ending in

Domestic small-cap growth equity

Domestic mid-cap growth equity

Domestic large-cap growth equity

Domestic growth and income

Domestic income

Domestic balanced

Domestic value

Domestic short-biased funds

Domestic fixed income

Domestic all sectors

International all

Global all

NYSE composite index

1980 23683.9

21377.1 18070.0

17387.5 16894.4 14279.4


10494.4 26201.4 17419.9 20186.9









13556.6 30955.4




28717.9 29521.8






18136.0 33475.7


30612.5 23777.2

1990 12723.3






13595.7 14261.3








17464.7 16440.5 14254.3 15769.5

16619.9 16266.3 14536.1 18336.9 23192.1 17001.9 17609.1

20448.3 26785.9



33975.3 35115.1 36924.5


26609.8 47873.3

35042.5 34941.3

13764.8 15839.7


13963.7 13070.6


13106.9 9092.4

11292.6 17396.5








19777.2 17323.0



12964.6 25370.0

16111.3 18942.5 21894.5

48445.7 52934.0

48454.4 46743.3 43181.3




19920.5 51890.0 26513.0



2007 12661.0 13926.1 12812.0


12854.8 12778.3


11359.3 11301.5



14806.6 13606.0

2007 21107.0


17530.8 18231.5


17094.9 19197.9





22810.1 20011.9



16425.4 17735.4 19466.8 19544.3


11745.1 16279.6 28228.0

28509.8 24748.1


Corresponding returns for the value-weighted NYSE/AMEX composite index are also reported for comparison.

a Panel A reports geometric average annual returns for 1, 3, 5 and 10-year periods ending in each of the last three decades and at the end of our sample period.

b Panel B reports 3, 5 and 10-year future values for a hypothetical investment in each of the fund categories at the end of each of the last 3 decades.

growth funds were near 40% in 1980, declining to 12% around the 1990s recession, peaking again to 35% at the tail end of the Internet 'bubble' and declining again in 2007 around the current financial crisis. Likewise, the size of assets managed by fixed-income funds increased to 29% in 1990 with an additional 42% in money market funds. They again drop thereafter. At the end of 2007, money market funds show the highest proportion of assets. While the size of assets under management has grown steadily over the

past four decades, transfers within mutual funds in response to economic events appear to be significant and are discussed later in terms of estimated flows.

Table 2, Panel A reports 3, 5 and 10-year annualised returns over 1, 3, 5 and 10-year holding periods in each of the fund categories at the end of each of the last three decades and for the final year of our sample period. For comparison, the corresponding returns for the NYSE composite index are listed. Inspection of the mean returns

across taxonomies over different time horizons provides some useful insights. Typically, the cross-sectional return variation is inversely related to the length of the holding period in several of our snapshot windows. In the short-run this is suggestive of momentum in some taxonomy groups playing an important role. In the long-run, mean reversions cause returns to be roughly comparable across taxonomy groups. This tendency is quite remarkable, noticed by the media and the last 10 years are frequently referred to by them as the 'lost decade'.

In Panel B of this table we report 1, 3, 5 and 10-year future values for a hypothetical $10,000 investment in each of the fund categories. Our intent in this table is to follow the manner in which mutual funds report performance in their prospectuses. For equities as a whole, the decades culminating in 1980 and in 2000 appear to have provided the best 10-year performance with mid-cap domestic funds and sector funds leading the way. Global and international funds (particularly emerging market funds) have been the most superior in the period ending in 2007.

Interpreting flows into mutual funds

Formal investigations of many of these patterns are pursued in the flow-performance literature that we turn to next. We generate flow estimates in the usual way, recognising that assets under management can either grow internally or by the flow of new cash to these funds. We define FLOW(t) as:21

FLOW (t) = [TNA(t)-TNA(t - 1) x 1 + R(t)} - MGTNA(t)]/TNA(t - 1) (1)

where: TNA represents the total net asset value of the mutual fund at times t - 1 and t, and R(t) represents the return earned by the fund over the period (t - 1,t), and MGTNA(t) represents the increase in assets due to mergers.

Examining flows into mutual funds has implications both for the timing ability of some investors and the consequent investment decisions of portfolio managers. Investors chasing performance may direct their dollars towards superior performing funds while others may choose to take their profits. Likewise, a reluctance to realise losses may cause some investors to leave their assets with under-performing funds. In turn, portfolio managers can use fund flows to ascertain investor appetites for risk and consider tailoring their product offerings accordingly.22 Studying flows enables the impact of the many effects to be disentangled.

Accordingly, Table 3 summarises the flows into our broad categories of mutual funds. In Panel A, we first provide descriptive statistics on the distribution of FLOW for the year 1990, roughly the mid-point of our sample period.

21 The flow estimates obtained from this equation represent both inflows and outflows. Additionally, it assumes that all these flows occur at the end of the period.

22 Fund flow data attracts frequent mention in the media. Trim Tabs Investment Research specialises in delivering current flow data to their institutional subscribers. Conversations with their customer relations department suggest that their method for estimating flows is identical to the procedure we follow.

These details are presented for selected domestic taxonomy cross-sections with relatively large sample sizes.23 The flow estimates are TNA-weighted flows aggregated within each taxonomy group with means, medians and standard errors alongside. Averages are reported in percentage terms. While the flow distributions display some central tendency with median flows close to zero, there are large values at both tails, with some skewness. To provide a feel for the distribution, we also report the inter-quartile range, the range and the range winsorised at 5%.24 Values for these variables suggest that larger flows are concentrated at both extremes of the distribution. These large flows are a reflection of both investor interest in that sector and their response to the performance of funds in it. Within each cross-section we also report the proportion of funds receiving positive flows. Extreme values for this variable would indicate that investors view the funds in the group homogenously and direct their flows towards or away from the group without discriminating between the performance of the individual funds. The panel reports this proportion to be between 37% (for global debt) to 77.1% for (international equity). Taken together the results in this panel support the notion that the FLOW variable is characterised by many extreme observations, both inflows and outflows and that investors are sensitive to performance.25

In Panel B, we report mean FLOWS at 10-year reporting periods to showcase the results from a comprehensive sample and to provide a feel for the time-series behaviour of mutual fund flows. Again, our pattern of annual flows very closely resembles those reported in Fig. 2.3 of the 2008 Mutual Fund Fact Book.26,27 A cell entry of 5.0 represents an annual flow of 5% of prior-year assets into that fund category. The reported numbers reflect the net effects of several competing reasons behind investors directing their flows into mutual funds. Nevertheless, several features are evident from this table. First, the flows reveal the rush towards growth in 2000 with positive fund flow into small and mid-cap stocks and a negative flow away from income funds, balanced funds, value funds and fixed income funds. Especially noteworthy is the flow of funds into sector funds of 28.41% at that time. Third, is the flow away from equities in particular and growth funds in general and into balanced and fixed income funds during 2007. Aversion to equities is also visible in the flow into

23 Despite the potential of FLOW estimates to shed light on a variety of research questions, we are surprised to find little documentation of the distributional properties of this variable.

24 While it is not customary to 'winsorise' cross-sectional data, we use this method to highlight the size of the extreme values in the flow distribution.

25 We observe similar patterns in flows at other points in our sample period. We chose 1990 simply for exposition.

26 Differences between the two samples are generally larger in the earlier years. We believe that this reflects differences in sample sizes, in the treatment of surviving funds and in the methods for aggregating and normalising flows into and out of mutual funds.

27 Despite the popularity of flow estimates in research, we are surprised to find only minimal reporting of basic flow statistics in the literature.

Table 3 Flow statistics.

Panel A: Descriptive statistics on FLOW for the calendar year 1990

N Mean flow Median flow Prop > 0 Range

Interq. Wins 5% Full

Domestic small-cap growth equity 80 2.361 -0.473 43.8 11.4 118.5 439.4

Domestic mid-cap growth equity 61 -0.496 -0.340 45.9 12.2 71.4 182.0

Domestic large-cap growth equity 145 1.738 0.180 52.4 5.3 62.1 173.0

Domestic growth and income 62 3.897 0.090 51.6 12.1 99.9 264.0

Domestic income 37 -8.968 -0.355 40.5 5.1 120.4 182.5

Domestic balanced 117 0.830 -0.250 42.7 6.3 89.9 392.3

Domestic value 117 -0.019 -0.144 41.9 4.2 51.4 383.5

Domestic short-biased funds 2 -1.925 -1.925 0.0 1.0 1.0 1.0

Domestic fixed income 913 -4.697 0.255 55.1 7.4 81.7 1076.4

Domestic all sectors 129 6.419 -0.203 45.0 9.4 103.9 333.0

International all 76 44.952 8.581 76.3 31.3 241.7 929.8

Global all 77 3.156 -0.557 49.3 17.4 104.3 147.6

Panel B: Mean FLOW at 10-year intervals over the sample period

1971 1980 1990 2000 2007

Domestic small-cap growth equity 7.275 -1.041 2.361 15.805 -8.311

Domestic mid-cap growth equity -1.837 -10.204 -0.496 29.367 0.289

Domestic large-cap growth equity -4.276 -8.261 1.738 3.495 -9.564

Domestic growth and income 2.015 -1.775 3.897 -1.168 4.95

Domestic income -9.684 5.904 -8.968 17.119 -1.506

Domestic balanced -4.723 -18.01 0.83 -9.806 9.667

Domestic value -2.321 -7.287 -0.019 10.384 -2.874

Domestic short-biased funds -1.925 1.281 5.012

Domestic fixed income 12.629 10.132 -4.697 -9.156 4.327

Domestic all sectors -6.355 2.742 6.419 28.407 -6.357

International all 33.354 -0.041 44.953 4.651 6.137

Global all 29.602 32.146 3.156 -3.802 19.632

This table reports descriptive statistics on net cash flow estimates. We define normalised flow as: FLOW(t) = [TNA(t) - {1 + R(t)} x TNA (t - 1) - MGTNA(t)]/TNA (t - 1) and report these in percentage terms. In Panel A, the mean and median flows and the proportion of positive flows ("Prop>0") for the 1990 cross-section are reported. In addition to the inter-quartile range ('Interq'), the range winsorised at 5% ('Wins 5%') and the range ('Full')_are also reported to characterise extreme values in the distribution. Panel B reports mean normalised flows for each taxonomy type at 10-year intervals.

short-biased funds in 2007. In sum, even at these aggregated levels, flows appear related to economic conditions.

Performance evaluation

In addition to economic conditions, flows into mutual funds are obviously related to the performance of the manager and a closer examination of the flow-performance relationship is the focus of one of our empirical investigations. In Table 4, we provide some basic results on mutual fund performance. Results in Panel A are based on monthly times-series regressions of the returns to portfolios of mutual funds for each taxonomy type. They are reported for both the single-factor CAPM and the four-factor Fama-French models.28

28 We thank Ken French for making the data on CRSP value-weighted market portfolio, HML, SMB and Momentum factors available at french/data_library.html.

Single-factor fund portfolio alphas are generally not significant. CAPM beta estimates are generally consistent with what one would expect, with aggressive equity portfolios, small-cap and sector funds showing the highest betas, fixed income funds and short-biased funds at the other extreme and balanced funds somewhere in the middle.

It is well known that the risk factors in the four-factor model are differences in average portfolios constructed according to various specifications, and that care must be taken in interpreting the resulting coefficients. To facilitate this interpretation, we break the universe of stocks into extreme deciles of value and growth based upon the book-to market ratio and, consistent with the literature, we find the spread between the corresponding loadings on HML to be about 1.45. This is the largest spread we can expect to find. We also break the universe of stocks into extreme deciles of market capitalisation and find the maximum possible spread between the loadings on SMB to be 1.51 (small-big). These estimates serve as measures against which we can assess the various mutual fund sensitivities to HML and SMB.

Table 4 Performance evaluation.

Panel A: CAPM and four-factor model estimates

CAPM Four-factor model

Alpha Beta R2 Alpha R(m)-Rf HML SMB UMD R2

Domestic Small-Cap Growth -0.0007 1.152 0.82 -0.0006 0.980 -0.200 0.591 0.090 0.95

Equity (-0.57) (39.58) (-0.84) (44.98) (-4.35) (19.55) (3.29)

Domestic mid-cap growth -0.0004 1.157 0.85 0.0002 1.004 -0.314 0.345 0.129 0.94

equity (-0.370 (42.26) (0.30) (44.01) (-8.94) (10.75) (4.45)

Domestic large-cap growth -0.0006 0.988 0.98 -0.0003 0.969 -0.101 -0.045 0.032 0.98

equity (-1.81) (133.56) (-0.91) (114.26) (-6.56) (-2.72) (2.89)

Domestic growth and income 0.0001 0.808 0.94 0.0000 0.845 0.142 0.016 -0.017 0.96

(0.23) (45.54) (0.05) (52.43) (5.45) (0.72) (-1.04)

Domestic income 0.0014 0.752 0.84 0.0000 0.846 0.347 0.009 -0.087 0.92

(1.71) (28.45) (0.06) (47.15) (9.10) (0.23) (-3.11)

Domestic balanced 0.0006 0.613 0.91 -0.0002 0.668 0.161 -0.038 -0.010 0.94

(1.27) (38.94) (-0.60) (58.61) (7.74) (-3.08) (-0.77)

Dopmestic value 0.0011 0.845 0.92 0.0000 0.928 0.259 -0.049 -0.056 0.96

(1.62) (36.58) (0.08) (85.31) (6.20) (-1.64) (-2.23)

Domestic short-biased funds 0.0005 0.095 0.02 -0.0024 0.195 0.248 0.004 0.159 0.12

(0.30) (0.75) (-1.35) (1.65) (3.33) (0.06) (2.15)

Domestic fixed income 0.0003 0.210 0.29 -0.0006 0.247 0.145 0.030 0.000 0.34

(0.32) (6.90) (-0.72) (7.82) (6.24) (1.25) (0.01)

Domestic all sectors 0.0011 0.956 0.79 0.0013 0.862 -0.204 0.209 0.105 0.84

(1.02) (26.86) (1.53) (26.64) (-3.87) (4.46) (2.73)

International all 0.0009 0.800 0.64 0.0000 0.784 0.064 0.192 0.052 0.66

(0.59) (24.65) (-0.05) (25.44) (1.48) (5.20) (1.76)

Global all 0.0008 0.903 0.87 0.0002 0.874 0.014 0.175 0.033 0.89

(0.96) (40.85) (0.39) (40.46) (0.36) (5.42) (1.57)

Panel B: Four-factor model alphas

Negative & Negative & Positive & Positive &

significant not significant not significant significant

Domestic small-cap growth 0.18 0.52 0.27 0.03


Domestic mid-cap growth 0.07 0.43 0.45 0.05


Domestic large-cap growth 0.15 0.57 0.27 0.01


Domestic growth and income 0.21 0.53 0.23 0.03

Domestic income 0.14 0.54 0.29 0.03

Domestic balanced 0.13 0.49 0.35 0.03

Domestic value 0.12 0.57 0.27 0.04

Domestic short-biased funds 0.37 0.41 0.18 0.04

Domestic fixed income 0.06 0.38 0.49 0.07

Domestic all sectors 0.03 0.45 0.48 0.04

International all 0.09 0.53 0.33 0.05

Global all 0.13 0.54 0.28 0.05

Panel A reports portfolio performance using the single-factor CAPM and four-factor Fama-French models. These regressions are estimated using monthly returns over the period 1970-2007 on a portfolio of funds for each taxonomy type. Monthly data for the Fama-French factors are obtained from his website (see footnote 28). t-statistics are reported in parenthesis and are based on standard errors computed using the Newey and West (1987) correction for heteroskedasticity and autocorrelation for up to six lags. Panel B reports the proportion of significant (at 5%) and insignificant four-factor alpha estimates at the individual fund level within each taxonomy type.

The loadings on the value-premium related factor (that is commonly attributed to the HML term) are consistent with what one would expect - namely that growth equity stocks load negatively while value stocks load positively. For our mutual fund sample, this spread is substantially smaller at about 0.4. The manner in which fund classifications recognise value and growth and the extent of aggregation in the data are responsible for this difference. For the SMB term, the spread is 0.64 again for the same reasons above. Compared with 0.46 for value-growth this could argue for a better size-based taxonomy classification rather than a book-to-market one. Finally, MOM is positive for equities, negative for value and balanced funds and insignificant for fixed income.

Panel B of Table 4 provides a picture of performance at the individual fund level rather than at the portfolio level. Here we report results on the proportion of individual mutual funds that exhibit inferior, superior or insignificant performance as represented by four-factor alphas. Significance tests are carried out using Newey and West (1987) corrections for hetero-skedasticity and autocorrelations up to six lags. As the table indicates, superior performance obtains for fewer than 7% of mutual funds across all taxonomy types—a proportion that is reasonably constant. One interpretation of this result is to view it as confirming general perceptions that most fund managers do not outperform. However, one should recognise that these proportions may simply represent Type-I errors or false positives and that a proper assessment of performance in the aggregate requires closer examination of the alpha distribution.29

The proportion of 'inferior performers' is however much larger for most of our taxonomy groups than the chance of a false negative. The attendant notion that more managers are significantly inferior performers reinforces the idea of index investing. However, one should recognise that these models are not well-specified for certain taxonomy types - for instance, fixed income portfolio measurement must be controlled for changes in default premiums and the term structure of interest rates. Indeed, one could argue that financial companies (that appear in most well diversified equity funds) are perhaps more sensitive to these factors rather than to the typical four factors that our benchmark models specify. Moreover, our performance results for short-biased funds are more likely to be a reflection of nonnormal return distributions and model mis-specification rather than inferior performance.30 Nevertheless, we offer these results to provide a picture of fund performance across different asset classes as we believe that,

despite the lack of precision, they highlight the criti-cality of benchmarking efforts.

The state of mutual fund portfolio performance evaluation

With the steady flow of investor dollars into the mutual fund space over the last four decades, the evolution in the number, the size, the range of strategies and investment styles, and the expense structure of fund offerings has been the primary success story of professional money management. Understandably, academic literature has been largely focused on whether the return performance of funds in this industry is adequate enough to compensate for the fees and expenses they generate. It also examines patterns in the flow of funds from investors. We summarise the state of this body of knowledge below.31

Performance studies

Portfolio performance evaluation studies in the finance literature are continually evolving - in the sample sizes of mutual funds available for study, the quality of the return series that are being made available and in the nature of the factor models and benchmarking techniques against which portfolio manager performance is measured. Early studies, Jensen (1968) and Gruber (1996) among others, used the well-worn single-factor CAPM and concluded that mutual fund managers did not exhibit much ability to select stocks. Later studies, beginning with Henriksson and Merton (1981), decomposed the performance of the portfolio manager into timing and selectivity and found no evidence of manager ability to time the market. In contrast, Goetzmann and Ibbotson (1994), and Hendricks, Patel, and Zeckhauser (1993) found that some portfolio managers have 'hot' hands, are able to consistently select superior stocks and that this performance has a tendency to persist. A large body of literature subsequently addresses this debate.

Brown, Goetzmann, Ibbotson, and Ross (1992) argue that this superior performance arises due to survivor bias. Using Morningstar's database with a well-known survivor bias, Elton, Gruber, and Blake (2001, p. 2417) document that overall performance measures are inflated by 0.4-1% depending upon the sample period studied. Therefore a sample of funds with survivor-bias can be shown to generate positive alphas when the true average for the population is negative.32 These concerns should at least be partly mitigated by the CRSP survivor-bias-free mutual fund

29 Barras, Scaillet, and Wermers (2010) and Fama and French (2009) propose methods for measuring the impact of false positives in performance evaluation.

30 Lo (2007) proposes a new measure of alpha that takes into account a manager's dynamic investment choices. He argues that this is a better measure of performance when long-short strategies, typically employed by hedge funds, are being evaluated.

31 We confine ourselves to describing issues surrounding portfolio performance evaluation. The literature on mutual funds has several papers studying related topics such as hedge funds, industry concentration and manager attributes.

32 Elton et al. (2001) go on to argue that, while a significant improvement, the CRSP database still suffers from a form of omission bias because of missing data particularly in the early years of the time series of mutual fund returns. Indeed, it is this consideration that gives rise to our discussion on missing data treatment in Section 3.1.

database that we presently use. Brown and Goetzmann (1995) claim that persistence results from 'inferior' managers consistently earning negative risk-adjusted returns. Carhart (1997) and Chen, Jegadeesh, and Wermers (2000) document that this observed performance persistence occurs from momentum. Since a disproportionally large portion of the portfolios of past winning funds will consist of stocks with high past returns, superior manager performance may reflect managers simply chasing momentum in the underlying stocks and not of any persistent security selection skills. Carhart's paper used a four-factor model in making this assessment.

Concerns regarding the benchmarks used to evaluate that performance prompt Daniel, Grinblatt, Titman, and Wermers (1997) to examine portfolio holdings. They argue that performance should really be measured in terms of whether the stocks chosen by a manager perform better than average stocks with the same characteristics. These characteristics are inferred from the portfolio holdings directly and their CS measure controls for size, book-to-market and momentum characteristics. In essence, both the characteristic benchmark approach and the factor model approach share a common theme -namely, that mechanistic, replicable adherence to a particular 'fad' that works should not qualify managers as 'superior' performers even if the chosen fad enabled them to perform well on an absolute basis.33 Jagannathan, Da, and Gao (2008) further decompose these measures into an informed trading and liquidity component.

Improving performance evaluation benchmarks is also the focus of Ferson and Schadt (1996) who caution that there is time variation in risks and risk-premiums and that performance evaluation should account for this condition-ality. Analogous to Daniel et al. (1997), they also explicitly recognise that manager performance stemming from publicly available information should not be treated as 'superior'. In their setting, conditional performance evaluation provides a nuanced re-interpretation of prior empirical results on two fronts. First, that negative unconditional alphas in traditional models are interpreted as inferior performance, but conditioning makes performance more neutral. Second, that conditioning the beta estimates impacts the negative timing results documented in previous studies. Changes in conditional betas are negatively related to changes in net new money flows into the funds. Efforts to fine-tune existing benchmarks also motivate (Cremers, Petajisto, & Zitzewitz, 2008). In a recent working paper they show that the Carhart and Fama-French four-factor models can result in significant alphas for passively managed equity index mutual funds. They explore alternative methods for constructing the Fama-French factors.

33 Kothari and Warner (2001) show that while regression-based multi-factor performance benchmarks should have lower statistical power than characteristics-based benchmarks, the magnitude of the differences is not large and is clouded by mis-specifications in manager style attribution.

Flow-related studies

In his presidential address Gruber (1996) addresses performance evaluation by comparing open-end and closed-end funds. He hypothesises that manager ability may not be reflected in actively managed open-end mutual funds since they are priced at net-asset value. Further, if such superior performance obtains, then it should be visible in investor flow of funds to those managers. He then examines 'new' money flows and finds that not only do investors chase superior performance, but that these new cash flows also earn positive risk-adjusted returns. He invokes a class of sophisticated investors ('smart-money') who are able to identify superior managers, send new funds to them and reap the benefits of that fund transfer. He finds this flow-performance relationship to be asymmetric in that while some investors chase superior performers, those in inferior funds appear hesitant to withdraw their money. He invokes a second, 'disadvantaged' clientele of investors and hypothesises that this reluctance may stem from institutional or tax reasons.

Zheng (1999) also documents evidence in support of this 'smart money' hypothesis by studying fund flows into a much larger sample of mutual funds and finds that it persists for up to 30 subsequent months. Wermers (2004) finds that fund performance is related to past and contemporaneous fund flows, and that the new funds appear to be invested by managers in increasing the positions in some of the stocks that they already own. As interest in studying the flows from investors and their deployment by fund managers grew, some scholars obtained data on 'gross' flows at the fund level rather than infer them from changes in net asset value. Cashman, Deli, Nardari, & Villupuram (2008) identify both inflows and outflows at the fund level and conclude that current investors punish poor performance by increasing (decreasing) their outflows (inflows). Ivkovich and Weisbenner (2008) use mutual fund trading data and find that inflows are related to relative performance while outflows are related to absolute performance. These results call into question the asymmetry of investor response to good and bad performance by focussing on heterogeneity in the behaviour of mutual fund investors.

Investor heterogeneity can also be argued in a more intuitive manner. The aggregate flows to mutual funds that are commonly estimated in the literature can mask several, often competing considerations. At the fund level, a portion of the flows could constitute a long-term asset allocation strategy by investors. Such flows will be regular and may exhibit positive autocorrelation. Another portion may be a direct response to fund performance, but these should largely cancel out at the taxonomy level. Flows may also be directed to certain types of funds as hedges against long-side portfolios in response to (or anticipation of) adverse market conditions and may be short-term. These considerations are consistent with Berk and Green (2004) who argue that the reaction of flows to manager performance is illustrative of an allocation of capital to where it is most productively utilised. In their framework, managers may not generate superior performance and investors may

not earn excess returns, but investors are rational in chasing performance. In contrast, Frazzini and Lamont (2008) treat flows as a 'sentiment' variable and find that flows are skewed towards funds that report subsequent lower returns and exhibit a tendency to follow behavioural prescriptions.

A parallel strand of research explores similar strategic and heterogenous behaviour by portfolio managers as they attempt to generate or preserve portfolio performance. These studies look at herding behaviour Wermers (1999) and risk-shifting Huang, Sialm, and Zhang (2009) by portfolio managers. Time-variation in portfolio risk exposure can be a response to dynamically available investment opportunities or as a result of tournament behaviour Brown, Van Harlow, and Starks (1996). Tournament behaviour refers to managers changing their portfolio's risk profile in response to performance in the middle of a year. Managers with superior (inferior) mid-year performance reduce (increase) their risk exposure. A related strategic aspect of portfolio manager behaviour relates to window-dressing, which is a cosmetic attempt to sell the 'losers' and hold/increase exposure to 'winners' in their portfolio. This tends to be a very short-term activity prevalent at the end of reporting quarters or fiscal years and does not necessarily result in dramatic changes to portfolio risk. Schwartz (2005) reports evidence on these two hypotheses using portfolio holdings and finds weaker evidence of tournament behaviour and stronger evidence in support of portfolio manager efforts to window-dress their portfolios.

Recent studies look at the behaviour of investors and managers according to the phase of the business cycle-dan investigation that is especially relevant in the current economic climate. Kosowski (2006) shows that mutual funds generate a greater positive alpha in recession periods than during expansions. Cederburg (2008) documents that investors chase performance and earn some positive alpha during expansions but not during recessions. These conflicting results are driven in part by the choice of conditional or unconditional benchmarking.

Summary and directions for further research

Early empirical evidence focused exclusively on the performance of the mutual fund portfolio and the manager's timing and/or security selection ability. Investigations then broadened into evaluating portfolio performance by examining the performance of the component stocks themselves and building a set of common, easily replicable benchmark characteristics. From portfolio performance evaluation, the literature has evolved towards studying patterns in the investment dollar flows that investors direct to the portfolio managers. The primary concerns are whether investors are rational in chasing performance and whether they are rewarded for it. Investors direct their dollars to funds as they anticipate and/or react to past performance while portfolio managers deploy these funds in winning stocks (herding), adjusting their risk exposure (performance anxiety and tournament behaviour) in order to maintain or improve upon that performance.

Both investor and manager actions are likely to have rational as well as behavioural implications. The view that we advocate is to first recognise that performance is jointly influenced by the heterogeneous actions of both portfolio managers and investors. One aspect of flow-manager interactions that has not been fully investigated is the extent to which investors shift flows between funds based on their expectation of future performance. Even if the flow is not a transfer from another fund, flows may be directed to specific funds because of a market expectation that might differ from that of the manager. The charter of most mutual funds requires them to be usually fully invested and differing expectations could mean that flows may increase (decrease) at times when the manager feels it is unwise (wise) to deploy them. How the manager reacts to the flows is subsequently likely to affect the fund's performance. Examining such dynamic interactions could shed further light on the efficacy of both investors' and managers' decisions.

Pedagogical considerations

The evolution of the literature follows a pattern common to most academic, empirical modes of inquiry. First, hypotheses are developed, then data is gathered and methodologies devised to test that hypothesis. The claim is accepted, rejected or deferred, with or without qualification depending on the strength of the evidence. Subsequent refinement to the data and the methodology results in rebuttals and refutations, often causing improvements to the original hypotheses. This process has pedagogical implications which we address below.

Frankfurter (2008) urges a less dogmatic approach to the pedagogy in financial economics based on the Toul-minian mode of argument—a process that closely resembles the empirical research mode of inquiry we describe above. This process is also visible in the social sciences where grey areas abound on many issues. A useful pedagogical tool in that context is 'teach the conflict' itself rather than sparing students from controversy. This approach has the advantage of enabling students to participate in an intellectual conversation and appreciate the complexities involved. They can only benefit from a nuanced understanding of how investors and portfolio managers might behave under various conditions. However, finance academics, who are usually trained as positivists argue for not questioning assumptions—a message that does not often translate well to the intended audience, many of whom continue to remain sceptical of the underlying theory. We feel that one aspect of the maturity of our discipline should be its willingness to expose its debates to public scrutiny. In this context, the different benchmarks that are brought to bear in measuring performance easily segue into a larger discussion of how risks are (and should be) measured and modelled.

Another illustration also highlights the pedagogic benefits of discussing conflicting evidence. The central theme underlying mutual fund performance measurement has always been that portfolio managers be held

to a higher benchmark than one achieved by merely mimicking public information better than their peers. Risk-adjusted performance has always been the goal behind most academic exercises—the plethora of models and methods simply reflects the discipline's technological advance towards that goal. This is clearly the highest standard and from a market efficiency standpoint, it should not be surprising that few managers are consistently able to outperform these ever-improving yardsticks. In contrast, portfolio managers prefer to showcase performance relative to a broad-based market index. Beginning students in finance classes tend to think of performance in terms of absolute returns—simply ending with more money than they started with.

At the forefront of financial literacy, it is incumbent upon us as teachers to improve the ways in which we can inform our audience of these developments. To us, the treatment of this material in our textbooks is not commensurate with the importance of mutual funds to that audience. Undergraduate textbooks typically discuss mutual funds by focussing primarily on load, no-load and index funds, with an occasional attempt at a broader taxonomy of funds by capitalisation and by security-type. Sources for these taxonomies are however varied, with classifications from Morningstar, the Wall Street Journal and Lipper Associates making an appearance. The presentation of this material is very qualitative with no supporting statistics to give the student a sense of the relative size and importance of fund categories. There is considerable discussion of fund expense structure, but very little on performance reporting beyond Sharpe, Treynor and Jensen ratios. Many have a disproportionate coverage of closed-end funds (with 2.4% of total fund assets as of December 2007), but relatively few focus in depth on funds with a significant fixed income component which have 45%—50% of total fund assets under management.34 Graduate textbooks discuss the taxonomies and investment styles for different funds in a little more detail. A few include the results of studies from the mutual fund performance literature, but the treatment is often selective. The reported studies are dated and mix results from a period of low sample sizes (pre-1990) with one from a large sample size (late 2000s). Only one (Bodie, Kane, & Marcus, 2008) offers some version of multi-factor performance and reports a distribution of monthly alphas for 1993—2007 for domestic equity funds. We recognise that there is some lead-time before research results appear in textbooks but our quibble is more with the arbitrariness of the results that are chosen for such presentation. Overall we believe that situating discussions of mutual fund performance and the behaviour of market participants to promote a view of the whole is preferable to a piece-meal approach where the intended audience takes away disconnected results with a potentially misleading message.

34 Data on closed-end funds is taken from the 2008 Mutual Fund Fact Book.


In this paper, we develop and describe the construction of a taxonomy of US mutual funds using the CRSP Survivor-Bias-Free Mutual Fund database. We believe that this taxonomy provides a basis for studying mutual fund subsets beyond that of equities. We provide snapshots of the number of funds and the total net asset values in each taxonomy group over time. We document estimates of the returns to different mutual fund portfolios for 3-year, 5-year and 10-year holding period intervals. We examine and interpret the flow of investor dollars into each taxonomy category. In these efforts, we describe patterns in the returns and flow data that form the basis of numerous research efforts. Additionally, patterns of investor fund flows appear broadly related to different stages of the business cycle—an area of our ongoing research. We report the results of single-factor and four-factor Fama—French performance measures for each taxonomy type and find little consistent evidence of superior performance. Finally, we discuss the evolutionary nature of academic research on mutual fund performance and describe the interactions between the strategic behaviour of investors and fund managers.

Throughout the paper, our intent is to present our research results in a simple way to make it more accessible and meaningful to the reader. Scholars beginning mutual fund research can use our taxonomy and descriptions of the database in their own work. Investment professionals can potentially construct more appropriate performance benchmarks and use the results on the strategic behaviour of market participants in their investment decisions. The comprehensive summary statistics, results and literature survey that we provide should be valuable to teachers promoting financial literacy as we emerge from turbulent economic times.

Appendix Taxonomy groups

This section specifies all the taxonomy codes into which we group our sample of mutual fund share classes. Below, we list the individual codes for domestic share classes. The same structure is maintained when we classify global funds and international funds. Some sub-categories are presently populated by relatively few share classes, which we expect will eventually fill over time.

We have chosen to preserve the essence of existing classification schemes rather than radically alter them. In the body of the paper, we generally report results for selected top-level taxonomy groups in the interest of brevity.

Total share classes are those available over the period 1970-2007. The length of the time-series observations of returns, net asset values, expense ratios will vary depending upon how long each share class remained in existence on the database and the extent of missing data.

Number of share classes

Aggressive growth funds (Code 11)

1110 Small capitalisation equities (core) 830

1120 Small capitalisation equities (growth) 1006

1130 Micro-capitalisation equities (core) 5

1140 Micro-capitalisation equities (growth) 1

1150 Other aggressive growth equity funds 212

Growth funds (code 12)

1220 Large capitalisation equities (core) (3055) 1208

1230 Large capitalisation equities (Growth) (4837) 1399

1240 Mid-capitalisation equities (Core) 855

1250 Mid-capitalisation equities (Growth) 473

1260 Multi-cap equities (Core) 814

1270 Multi-cap equities (Growth) 1108

1210 Index funds 192

1290 Other growth funds 363

Growth and income funds (Code 13)

1310 Growth and income funds 267

1320 Total return funds 262

Income funds (Code 14) 10

1410 Mixed income funds 52

1420 Equity income funds 380

1490 Other income funds (includes option income) 17

Balanced funds (Code 15) 336

1510 Target date funds 920

1520 Mixed-asset funds 1762

1530 Asset allocation funds 84

1590 Other balanced funds 12

Value funds (code 16)

1610 Large-cap value funds 709

1620 Mid-cap value funds 417

1630 Small-cap value funds 401

1640 Other value funds 693

Sector funds (codes 18 and 19)

1810 Bio-technology 255

1820 Consumer 36

1840 Financial services 149

1850 Health-care 3

1860 Leisure 11

1870 Precious metals 94

1880 Real estate 428

1890 Natural resources 157

1910 Technology 511

1920 Telecommunications 58

1930 Utilities 151

1940 Commodity funds 42

1950 Environmental funds 6

1960 Industrial Funds 13

1970 Transportation 7

1980 Merger funds 5

1990 Other sectors 11

Specialty funds

1690 Short-biased funds 70

1691 Long-short funds 139

1692 Market-neutral funds 74

1693 Bear and ultra-short funds 91

Fixed income funds (Code 17) 6

1710 Government/treasury funds 511

1711 Short-term government funds 214

1712 Short-intermediate term government funds 162

1713 Intermediate-long term government funds 151

1720 TIPS 156

Municipal bond funds

1730 General municipal bond funds 391

1731 State specific municipal bond funds 1803

1732 Insured state specific municipal bond funds 116

1733 Short-term municipal bond funds 103

1734 Short-intermediate municipal bond funds 106

1735 Intermediate municipal bond funds 588

1736 High-yield municipal bond funds 115

1740 Money market funds 2227

1741 Insured money market funds 791

1742 Loan participation funds 68

1750 Mortgage funds 283

1751 Adjustable rate mortgage funds 32

1760 Corporate bond funds 154

1761 Short-maturity investment grade bond funds 338

1762 Short/intermediate investment grade bond funds 208

1763 Intermediate investment grade bond funds 824

1764 A-rated corporate bond funds 299

1765 BBB-rated corporate bond funds 239

1766 High current yield corporate bond funds 664

1770 Convertible bond funds 121

1780 High-yield bond funds 54

1790 Other fixed-income funds 306

2000 Global funds 1191

3000 International funds 3195

9999 Unclassified funds 282

Total share classes 32798


Barras, L., Scaillet, O., & Wermers, R. (2010). False discoveries in mutual fund performance: measuring luck in estimated alphas. Journal of Finance, 65(1), 179-216. Berk, J., & Green, R. (2004). Mutual fund flows and performance in rational markets. The Journal of Political Economy, 112(6), 1269-1295.

Bodie, Z., Kane, A., & Marcus, A. (2008). Investments (8th ed.).

New York: McGraw Hill/Irwin Companies. Brown, K. C., Van Harlow, W., & Starks, L. T. (1996). Of tournaments and temptations: an analysis of managerial incentives in the mutual fund industry. Journal of Finance, 51(1), 85-110. Brown, S. J., & Goetzmann, W. N. (1995). Performance persistence. Journal of Finance, 50, 679-698. Brown, S. J., Goetzmann, W. N., Ibbotson, R. G., & Ross, S. A. (1992). Survivorship bias in performance studies. Review of Financial Studies, 5, 553-580. Carhart, M. M. (1997). "On persistence in mutual fund performance. Journal of Finance, 52, 57-82. Carhart, M. M., Carpenter, J. N., Lynch, A. W., & Musto, D. K. (2002). Mutual fund survivorship. Review of Financial Studies, 15(5), 1439-1463. Cashman, G., Deli, D. N., Nardari, S., & Villupuram, S. V. (2008). Investors do respond to poor mutual fund performance:

Evidence from inflows and outflows. Working Paper. W.P. Carey School of Business, Arizona State University.

Cederburg, S. (2008). Mutual fund investor behavior across the business cycle. University of Iowa Working Paper.

Chen, H., Jegadeesh, N., & Wermers, R. (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(3), 343-368.

Cremers, M. A., Petajisto, & Zitzewitz, E. (2008). Should benchmark indices have alpha? Revisiting performance evaluation. Yale University. Working paper.

Daniel, K. D., Grinblatt, M., Titman, S., & Wermers, R. (1997). Measuring mutual fund performance with characteristic-based benchmarks. Journal of Finance, 52(3), 1035-1058.

Elton, E. J., Gruber, M. J., & Blake, C. R. (2001). A first look at the accuracy of the CRSP mutual fund database and a comparison of the CRSP and Morningstar mutual fund databases. Journal of Finance, 56, 2415-2430.

Fama, E. F., & French, K. R. (2009). Luck versus skill in the cross-section of mutual fund alpha estimates. Working Paper. Center for Research in Security Prices, University of Chicago.

Ferson, W. E., & Schadt, R. W. (1996). Measuring fund strategy and performance in changing economic conditions. The Journal of Finance, 51(2), 425-461.

Frankfurter, G. M. (2008). On the teaching of financial economics: a pedagogic note. Journal of Investing, 17(3), 105-110.

Frazzini, A., & Lamont, O. (2008). Dumb money: mutual fund flows and the cross-section of stock returns. Journal of Financial Economics, 88(2), 299-322.

Goetzmann, W. N., & Ibbotson, R. G. (1994). Do winners repeat: predicting mutual fund performance. The Journal of Portfolio Management, 20, 9-18.

Gruber, M. (1996). Another puzzle: the growth in actively managed mutual funds. Journal of Finance, 51, 783-810.

Hendricks, D., Patel, J., & Zeckhauser, R. (1993). Hot hands in mutual funds: short-run persistence of relative performance, 1974-1988. Journal of Finance, 48, 93-130.

Henriksson, R. D., &Merton, R. C. (1981). On the market timing and investment performance of managed portfolios II - statistical procedures for evaluating forecasting skills. Journal of Business, 54, p513-533.

Huang, J., Sialm, C., & Zhang, H. (2009). Risk shifting and mutual fund performance. Working paper. Mc Combs School of Business, University of Texas at Austin.

Ivkovich, Z., & Weisbenner, S. (2008). Individual investor mutual fund flows. NBER. Working Paper 14583.

Jagannathan, R., Da, Z., & Gao, P. (2008). Informed trading, liquidity provision, and stock selection by mutual funds. NBER. Working Paper No. 14609.

Jensen, M. C. (1968). The performance of mutual funds in the period 1945-1964. Journal of Finance, 23, 389-416.

Kosowski, (2006), "Do Mutual Funds Perform When It Matters Most to Investors? US Mutual Fund Performance and Risk in Recessions and Expansions," Working Paper, Insead.

Kothari, S., & Warner, J. (2001). Evaluating mutual fund performance. Journal of Finance, 56, 1985—2010.

Lo, A., 2007, "Where do alphas come from?" "A new measure of the value of active investment management," Working paper.

Mutual Fund Fact Book. 2008. The investment Company Institute

Nanda, V., Wang, Z. J., & Zheng, L. (2005). The ABCs of mutual funds: On the introduction of multiple share classes. Working paper. University of Michigan.

Newey, W. K., & West, K. D. (1987). A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covari-ance matrix. Econometrica, 55, 703—708.

Sankaran, S. (2007). Indian mutual funds handbook: A guide for industry professionals and intelligent investors. Vision Books.

Schwartz, C. G. (2005). Mutual fund tournaments: The sorting bias and new evidence. University of California at Irvine. Working paper.

Wei, K., Wermers, R., & Yao, T. (2009). Uncommon value: The investment performance of contrarian funds. Working Paper. University of Texas at Dallas.

Wermers, R. (1999). Mutual fund herding and the impact on stock prices. The Journal of Finance, 54(2), 581—622.

Wermers, R. (2004). Is money really smart? New evidence on the relation between mutual fund flows, manager behavior and performance persistence. Working Paper. University of Maryland.

Zheng, L. (1999). Is money smart? A study of mutual fund investor's fund selection ability. Journal of Finance, 54, 901 —933.