Is Timing Everything? The Value of Mutual Fund Manager Trades

Authors

  • Jon A. Fulkerson


  • I am grateful for comments from Bradford D. Jordan, Jason Smith, Yuehua Tang, seminar participants at the 2011 EFA and FMA annual meetings, and an anonymous referee.

  • Jon A. Fulkerson is an Assistant Professor of Finance at Loyola University Maryland in Baltimore, MD.

Abstract

I develop new measures of the value of active mutual fund management using portfolio holdings. These measures simultaneously test for trading and selection skill within stocks, industries, and characteristics. I demonstrate that most of the skill documented in prior studies comes from correctly trading stocks within industries, though funds also have some skill in timing industries. However, prior research focuses on the period 1980-1994. I also test the hold out sample 1995-2007. Contrary to prior results, the latter period (and the full sample) demonstrates that mutual funds generate no excess returns from any category of skill.

The mutual fund literature examining the trading behavior of mutual funds documents skill in gross returns (Chen, Jegadeesh, and Wermers, 2000). Mutual fund buys outperform sells, and gross returns demonstrate positive excess returns. Therefore, fund managers are viewed as having some stock trading ability prior to expenses. What is less clear is the source of manager skill.

I contribute to the literature on mutual fund performance by simultaneously testing for skill in a united framework of performance measures that control for both selection and trading ability. My measures decompose returns into two broad components. The trading component measures the additional return gained or lost through changing the portfolio and captures the value created by the short run anticipation of returns. The selection component measures how long-term holdings would have created value. A manager with selection ability creates value by tending to hold stocks that outperform over a longer period. The selection and trading components are not mutually exclusive and a fund may benefit from both.

My measures also test for multiple sources of skill within the trading and selection components. Prior research confirms skill in areas as diverse as market timing, anticipating individual stock news, and industry selection, but few studies control for multiple sources simultaneously. For example, Daniel et al. (1997) only test for trading skill in broad characteristics, while Kacperczyk, Sialm, and Zheng (2005) only allow for timing in industries. Chen et al. (2000) test for individual stock timing for the whole industry, but do not control for other forms of timing. In contrast, my measures allow researchers to isolate selection and trading skill in all three areas: 1) individual stocks, 2) industries, and 3) characteristic styles.

I apply these measures to mutual fund holdings from 1980 to 2007. I confirm that mutual funds demonstrate little trading ability with respect to individual stocks, industries, or characteristics. The funds do show some economically large stock selection ability, though most specifications do not reach conventional levels of statistical significance. Stock selection skill within industries is present from 1980 to 2007, but the measures are noisy and sensitive to timeframe. I conclude that industry expertise provides, at best, only modest benefits to the average mutual fund over the full sample.

My second contribution to the literature stems from the prior finding that excess returns can be attributed, in part, to timeframe. Daniel et al. (1997) find economically and statistically large excess returns for the shorter period of 1975-1994. I confirm that mutual funds generated excess returns over a similar period and that these excess returns come from stock trading (as suggested in Chen et al., 2000). The following thirteen years (1995-2007), however, indicate small losses from trading. Looking at the overall timeframe (1980-2007), the gains from trading are inconsistent through time and close to zero, on average.

My finding of small gains from industry expertise differs from Kacperczyk, Sialm, and Zheng (2005). Kacperczyk et al. (2005) find that funds with higher industry concentrations have higher excess returns and suggest that industry expertise may generate value. I find that trading within an industry generated large gains over the period analyzed by Kacperczyk et al., (2005), but the same trading generated small losses in more recent years and almost zero return for the full sample. Industry expertise over the full sample provided only minor benefits.

For robustness, I test the sensitivity of my results to portfolio construction. Mutual funds only report holdings quarterly and I may be using stale positions to calculate trading skill. I recalculate each measure assuming managers acquire the reported positions one to three months prior to the report date. Like Nicolosi (2009), I find that excess returns increase when the manager is assumed to have traded sooner. These higher excess returns are reflected in higher trading skill, while selection skill stays mostly constant. Though suggestive, further research with higher frequency data would be necessary to confirm that the current methods underestimate manager skill.

The remainder of this study proceeds as follows. The next section reviews previous studies of fund manager performance and timing. Section II describes the new measures implemented in this study and the sample on which the measures are applied. Section III presents the empirical results, while Section IV provides some tests of robustness. Section V provides my conclusions.

I. Background

Mutual fund holdings may be studied for insight into how mutual fund managers add (or subtract) value through active management. Skilled active managers should buy stocks that appreciate and sell stocks before they perform poorly. Consistent with this, some studies find that stocks sold by mutual funds do poorly in the period after selling and those purchased provide moderate positive returns (Chen et al., 2000; Pinnuck, 2003; Nicolosi, 2009). This evidence suggests that fund managers create value through trading.

Similarly, a fund manager may have the ability to select industries. Funds concentrating in fewer industries perform better than diversified funds, perhaps because the portfolio manager has expertise in a set of industries (Kacperczyk et al., 2005; Avramov and Wermers, 2006; Busse and Tong, 2012). Busse and Tong (2012) attribute around one-third of observed excess returns to industry selection by fund managers.

Holdings are also used to study fund manager timing abilities. Studies generally focus on one of three different aspects of timing. The first approach recognizes that a mutual fund's risk level may be time varying if the manager has an ability to anticipate market movements. Such a manager would increase beta exposure in an up market and decrease exposure in a down market. The evidence suggests small, but statistically significant differences in beta and, as such, at least some degree of market timing ability (Bollen and Busse, 2001; Chance and Hemler, 2001; Jiang, Yao, and Yu, 2007). Looking only at industry timing, Jiang et al. (2007) and Busse and Tong (2012) find similar results.

In contrast, Elton, Gruber, and Blake (2009) obtain different results using a slightly different approach. They estimate a fund's beta every month from portfolio holdings. Using the prior periods’ beta as an estimate of the fund's beta without timing, the timing ability is the difference in expected return earned using the actual beta and the prior beta. They find consistent losses due to timing on a monthly basis, suggesting poor market timing ability.

Studies examine the shift between equity and cash as a second method of measuring market timing. Hybrid (balanced) mutual funds, barring a very few periods, rarely correctly time the market (Comer, 2006). Similar results indicate that nonequity holdings have very little ability to predict general market movements (Sensoy and Kaplan, 2005; Yan, 2006). Thus, attempts to time the market with changes in cash holdings add a small amount of value at best.

The final method is the one most related to this study. Daniel et al. (1997) develop the characteristic timing (CT) measure. Looking at portfolio weight changes over time, Daniel et al. (1997) and Wermers (2000) find no ability for mutual fund managers to time broad stock characteristics (based on market capitalization, market-to-book ratios, and momentum). Since I build on the Daniel et al. (1997) approach to measuring the value of trades and timing in this study, their methods are discussed in greater depth in the next section.

Overall, previous research using holdings data produces multiple indications as to the source of skill in managers. Some, but not all, studies examining mutual fund holdings find that at least some actions taken by mutual fund managers suggest skill in stock selection, but there is conflicting evidence regarding the nature of this skill (if it exists at all). The evidence on trading ability also seems thoroughly mixed. Using new return decomposition measures and a much larger sample of fund holdings than most of the previous studies, I contribute to the debate regarding manager ability by examining trading and selection ability jointly. By design, my approach allows a more finely grained analysis of the various components of fund manager activities than those used in previous research with the hope of identifying how trading adds value.

II. Methods

I begin by examining alternate ways to measure mutual fund manager skill with holdings data. The measures developed here allow for a separation of the impact of active and passive management by managers, as well as measuring trading in multiple contexts. I first review the measures and assumptions used in previous research and then introduce new measures.

A. Characteristic Methods for Measuring Skill

Grinblatt and Titman (1993) were the first to use holdings to evaluate skill. They were followed by Daniel et al. (1997), who propose a more sophisticated decomposition for mutual fund returns that suggests three sources of returns. These three components describe the ability of a mutual fund manager to select specific stocks and to time the categories of stocks. The original three Daniel et al. (1997) measures, characteristic selectivity (CS), characteristic timing (CT), and average selectivity (AS), are as follows:

image
image
image

where wi,t is the weight of stock i at the end of month t in the mutual fund portfolio; ri,t is the return on stock i for the month ending at t; inline image is the return of the benchmark portfolio b for stock i for the month ending at t; and N is the number of stocks that exist at the beginning of month t.

The three measures in the decomposition add up to the gross portfolio return and have intuitive interpretations.1 The CS measure captures the additional return earned by selecting the best performing stocks within a benchmark portfolio.2Daniel et al. (1997) find an average CS measure of 0.77% return per year, suggesting that the typical fund manager is skilled at stock selection.

The CS measure ignores trading. In contrast, CT measures how well a mutual fund manager changes portfolio weights in anticipation of benchmark portfolio returns. A skilled manager would anticipate high (low) returns by increasing (decreasing) weights in benchmarks that subsequently do well (poorly). Daniel et al. (1997) find no significant evidence of value added through characteristic timing.

Finally, AS measures the returns earned by a fund that may be solely attributed to the characteristics of the stocks in the portfolio. Prior period weights and benchmark portfolio returns eliminate the impact of stock selection and benchmark timing. Thus, AS is the return a fund would have earned if it did no trading and only held broad characteristic portfolios. This measure typically comprises most of a fund's return (the other two measures are return differentials, not return levels), but this return component does not represent any value added by the fund manager over the period studied.

Results of Daniel et al. (1997) are confirmed in Wermers (2000), but neither study distinguishes between stock and industry skill. Also, there is a distinction that can be made with respect to the meaning of weight changes by a mutual fund. The next section demonstrates alternate approaches to measure trade value that lead to more detailed performance information. These alternate measures are one of the core contributions of this study.

B. Alternative Decompositions—Active versus Passive Returns

To begin, I isolate the additional return gained by changing the portfolio weights from the prior period. I define the gain from trading (GT) as:

image

where wi,t−13 is the weight one year ago. GT is the difference between the gross return assuming the most recent report date weights and the gross return assuming one year lagged weights.3 Thus, GT measures the additional return the portfolio earned as a result of active management over the past year.

GT provides a very basic measure of the benefits of active management. However, both the passive and active returns can be examined in more detail. To do so, I decompose GT into the performance due to trading on individual stock performance (stock selection trading, SST) and a characteristic timing similar to CT measure of Daniel et al. (1997). Specifically:

image

where

image(1)

and

image(2)

SST measures the ability of fund managers to trade in anticipation of individual stock excess returns. It will be positive when a fund increases the weight on stocks that outperform the benchmark portfolio and/or when a fund decreases the weight of a stock prior to a negative excess return. Thus, SST measures the trading ability of a manager for specific stocks separately from the manager's ability to trade broad characteristics. CT in Equation (2) has the same interpretation as the CT measure of Daniel et al. (1997). Managers who anticipate the time-varying premium on the various characteristic portfolios can add value. Therefore, SST and CT simultaneously control for two distinct types of trading ability.

SST and CT capture the active component to returns. A separate question concerns the return a fund would have experienced if its manager did no trading. I decompose this passive return to explore the excess return that would have occurred with no active management. To do so, I define stock selectivity (SS) as follows:

image(3)

SS measures the excess return the portfolio would have earned without active management. Whereas SST is the excess returns added by trading specific stock purchases, SS is the excess return the fund would have realized with no trading and is, therefore, a passive measure of performance. I interpret this performance as resulting from a manager's general style and not from their ability to anticipate returns.

To complete the decomposition, the characteristic selectivity (CS) measure below is the return the portfolio would have earned if it had only invested in broad benchmark portfolios. This is the mutual fund's return considering only stock characteristics and is similar to the Average Style (AS) measure of Daniel et al. (1997):

image(4)

CS does not represent skill since, by construction, the benchmark portfolios have no excess return. Instead, it measures the return on a passive portfolio that weighted the characteristic benchmark portfolios in the same way as the manager one year ago.

The top three levels of Figure 1 illustrate the relationship between the active and passive measures of performance. This alternative decomposition of returns provides two additional insights over Daniel et al.'s (1997) decomposition. First, by examining SST and CT, the measures isolate actual active management by separating out the excess returns that directly result from changes by management. Additionally, a fund's own past weights serve as a passive benchmark of performance. I use past weights to capture the value added over the past year compared to the low cost strategy of maintaining the same portfolio composition for this time period. In essence, the manager's past holdings serve as their own benchmark.

Figure 1.

Active and Passive Returns

The figure below illustrates how the various measures relate to the gross return. Gross return has two parts: 1) passive returns and 2) active returns. The passive returns involve selectivity of stocks, industries, and characteristics in prior periods. The active returns measure the impact on gross return for position changes by the mutual fund for individual stocks, industries, and characteristics.

C. Industry Trading

Many previous return decomposition studies (Daniel et al., 1997) do not explicitly consider industry effects. Kacperczyk et al. (2005), however, determine that funds with greater industry concentration demonstrate better industry selection and timing (and also use their own decomposition). Busse and Tong (2012) attribute one-third of a fund's gross excess return to industry selection. Therefore, industry skill affects a mutual fund's return and may be an important source of skill.

Based on these studies, I introduce two additional measures to account for industry skill. My previously defined Stock Selection Trading (SST) can be decomposed into two parts. The first is Industry-Adjusted Stock Selection Trading (ISST):

image(5)

where inline imageis the return on stock i's industry for period t. ISST measures the ability to trade a particular stock's performance relative to its industry. ISST will be positive when the fund increases (decreases) the weight on stocks that outperform (underperform) the industry portfolio. I claim that this is a direct measure of a manager's industry expertise.

The second part of SST is Industry Trading (IT):

image(6)

IT is positive when a mutual fund successfully anticipates an industry's performance relative to characteristic benchmarks. These characteristic-Adjusted returns are most analogous to the decomposed alpha of Busse and Tong (2012) and can be interpreted as the part of SST that comes from average industry excess return. A manager that can select, among industries, those that will typically beat their characteristic benchmarks will have a high IT return measurement.

Passive stock selection can also be decomposed into Industry-Adjusted Stock Selection (ISS) and Industry Selection (IS) as follows:

image(7)

and

image(8)

Industry-Adjusted Stock Selection is the passive excess return earned by a mutual fund over the industry portfolio of the stock. ISS will be positive when a manager consistently holds stocks that subsequently outperform their industry competitors. Industry Selection will be positive when a mutual fund tends to hold stocks whose industry returns beat the benchmark for that stock. Like the SS measure, IS measures the excess return a manager would have received with passive management with no subsequent weight changes.

The gross portfolio return on a mutual fund is now decomposed into six parts as illustrated in the bottom two tiers of Figure 1. Active management is measured with the three trading measures, Industry-Adjusted Stock Selection Trading (ISST), Industry Timing (IT), and Characteristic Trading (CT). Passive performance is measured with the three passive measures, Industry-Adjusted Stock Selection (ISS), Industry Selection (IS), and Characteristic Selection (CS).

These decompositions are most similar to the industry timing and selection measures of Kacperczyk et al. (2005)[Equations (7) and (8)]. However, like Daniel et al. (1997), their measures do not isolate the value of active management. Specifically, their IS measure captures the return in excess of the industry, but does not measure whether the fund actively pursued this excess return or if it is the passive result of their general investment style. Their IT measure does capture active trading, but there is no measure of industry selection. So, while similar, the two industry measures of Kacperczyk et al. (2005) do not fully capture the industry skill. Similarly, most prior research has ignored the impact of explicit and implicit trading by the fund, as discussed in the next section.

D. Explicit and Implicit Trading

One additional concern when evaluating fund manager trading is the difference between explicit changes in portfolio weights and implicit changes. If managers trade on momentum and the benchmark portfolios do not adequately control for momentum, “winner” (“loser”) stocks will mechanically get a higher (lower) weight in the portfolio. A positive weight change may result either from an uncontrolled momentum effect or from actual purchases of the stock by the fund manager. I refer to changes in weight resulting only from price effects as implicit changes. In contrast, an explicit weight change occurs when the manager increases or decreases shares owned (i.e., actual transactions by the manager).

To study this issue, I define the implicit weight as:

image(9)

where n is the number of shares of stock i held at time t-13, and Pi,t−1 is the price of stock i at time t-1. This weight would have occurred had there been no trading (but a potential change in price). With this definition, the change in weights used for the trading measures can be divided into explicit and implicit measures:

image

where inline image is the explicit weight change and inline image is the implicit weight change. By substituting the explicit or implicit weight change for the difference in weights, I can separate the trading performance measures into explicit and implicit management of stocks, industries, and characteristics.

Naturally, implicit changes may or may not be active management. Cohen, Polk, and Silli (2010) argue that managers include many stocks in the portfolio simply to track the market and only a few stocks represent the manager's “best” investments. In that case, some weight changes may be effectively ignored by the manager. I assume that since a manager has total control over a portfolio's weights, the manager (implicitly or explicitly) condones any changes in weight. Even if this is not the case, the explicit trading measures have a neat interpretation as being the value added directly from the manager's trades.

E. Implementation

I calculate my alternate measures from the previous subsections using holdings data on all equity mutual funds from 1980 to 2007 in the Thomson/Reuters Financial Network database. Following Daniel et al. (1997), I only update holdings after the fund reports their current position. For subsequent months, I assume the number of shares in each position stays the same as the prior month until a new report is found in the database. The weight of the asset in the portfolio is calculated each month using the share price from the beginning of the month. Assets dropped from the portfolio have a prior period weight greater than zero, but a current period weight equal to zero. Stocks added to the portfolio have zero prior period weight.

I calculate each measure of portfolio performance for every month for every fund in the database, thereby minimizing the impact of any survivorship bias. This process yields a time series of performance measures for each fund. Since fund holdings are usually updated quarterly, I compound monthly measures to form quarterly measures.4 Each quarter, the measures are averaged, weighting each fund equally or by total net assets (TNA) to get a time series of quarterly rebalanced performance measures for the entire population of mutual funds. The time series for each performance measure is used to test whether funds had significant performance over the sample period.

The sample includes all mutual funds with TNA data available and an investment objective that usually holds the majority of assets in equity. Mutual fund objectives and monthly TNA are taken from the Center for Research in Security Prices (CRSP) Survivor-Bias-Free Mutual Fund database using the March 2008 update. The CRSP and Thomson databases are matched by ticker where available, and by name for all other cases. The final sample includes 4,144 unique funds.5

Stock returns for gross and benchmark return calculations come from the CRSP NYSE/AMEX/Nasdaq Monthly Stock Database. I only include US common shares and exchanges. I use the 49-industry definition from Fama and French (1997) for industry benchmarking, and I obtain the industry returns from Ken French's website.6 The construction of benchmark returns also requires book equity information from Compustat.

III. Empirical Results

A. Active and Passive Returns

In this section, I present tests of trading ability and industry skill using the new measures developed in Section II. Daniel et al. (1997) and Wermers (2000) find positive stock selectivity and zero characteristic timing for mutual funds. If positive stock selectivity results from actively increasing (decreasing) asset weights prior to a positive (negative) abnormal return for a stock, I expect a positive SST measure. If there is no trading in anticipation of performance, the SS measure using prior period weights should be positive to be consistent with the prior literature. The values for the characteristic-adjusted measures are reported in Table I.

Table I.  Active and Passive Returns—Characteristic-Adjusted Mutual Fund Return Decomposition, 1980-2007
Period Equal-Weighted Returns TNA-Weighted Returns
SST CT SS CS SST CT SS CS
  1. a

    A decomposition of mutual fund returns is provided below. The returns are based on the Thomson/Reuters Financial Network mutual fund holdings and the CRSP stock return databases. The Equal-Weighted Returns section of this table provides equal-weighted averages for subperiods from 1980 to 2007. The TNA-Weighted Returns section uses TNA-weighted averages. The statistics include the following measures: industry-adjusted stock selectivity trading [SST, Equation (1)], characteristic trading [CT, Equation (2)], stock selectivity [SS, Equation (3)], and characteristic selectivity [CS, Equation (4)]. Portfolio measures are calculated monthly and compounded to get quarterly values for each measure. Annualized returns are given below, but t-statistics are calculated on the quarterly time series. In all measures in this table, I limit the sample to those funds with a self-reported investment objective of “aggressive growth,”“growth,”“growth and income,” and “balanced” in the most recent annual report in the CRSP mutual fund database. For the periods 1980-1994, 1995-2007, and 1980-2007, the t­-statistics for significance are presented in parentheses. The number of distinct mutual funds equals 4,144.

  2. ***Significant at the 0.01 level.

  3. **Significant at the 0.05 level.

1980-19840.0082−0.00100.00720.10620.0078−0.00090.00200.0971
1985-19890.00490.00200.00820.17940.00000.00190.00810.1782
1990-19940.01400.00430.00350.10190.01450.00460.00050.0943
1995-19990.00300.0064−0.00090.25330.00150.0054−0.00040.2522
2000-2007−0.0054−0.00310.01860.0587−0.0054−0.00380.01220.0566
1980-19940.0091***0.00200.00620.1308***0.0074**0.00200.00360.1251***
 (3.02)(1.24)(1.17)(3.51)(2.18)(1.54)(0.54)(3.65)
1995-2007−0.00220.00060.01110.1335***−0.0028−0.00030.00740.1318***
 (−0.59)(0.21)(1.00)(2.83)(−1.00)(−0.11)(0.91)(2.90)
1980-20070.00370.00130.00860.1321***0.00250.00090.00540.1283***
 (1.44)(0.85)(1.45)(4.52)(1.06)(0.65)(1.05)(4.65)

For both equal-weighting and TNA-weighting, SST is positive and significant for the period 1980-1994 and generated equal-weighted (TNA-weighted) excess returns of 0.91% (0.74%) per year. CT is much smaller and statistically indistinguishable from zero. The subperiod 1995-2007 and the full sample from 1980 to 2007 display different results. Trading on stocks and characteristics generated losses for the later part of the sample and trading generates effectively zero excess returns over the entire time frame.

Stock Selection (SS) does demonstrate economically important excess returns. The excess returns from selection were almost double that of timing for the full sample and were consistently large for both halves of the time period. Low statistical significance for SS suggests selection ability generated large, but noisy excess returns.

My findings differ from earlier research and suggest that if there was skill in the earlier years of the sample, it has diminished (or disappeared) in more recent years. Over the period studied by Daniel et al. (1997), the excess returns are similar in magnitude. My results suggest that their CS measure is primarily derived from active stock trading. Recent years indicate fewer total excess returns for the mutual fund industry and poorer trading ability.

When comparing studies, it is important to note that the cross-sectional aggregation of returns gives the first half of the sample the same weight as the second half even though, in terms of observations, 1995-2007 represents over 70% of the holdings database. The decrease in trading ability has been accompanied by an increase in the number of funds.

Next, I test the industry expertise of mutual fund managers using the industry-adjusted measures. The industry-adjusted measures allow for skill in both trading and selection in the areas of individual stocks, industries, and characteristics. If the trading skill from Table I actually results from industry trading, IT should be positive and ISST should be zero. Likewise, the stock selection measures may be, in part, due to industry selection. Table II displays the empirical results for industry-adjusted returns and addresses these two issues.

Table II.  Active and Passive Returns—Industry-Adjusted Mutual Fund Return Decomposition, 1980-2007
Period ISST IT CT ISS IS CS
  1. a

    A decomposition of mutual fund returns is provided below. The returns are based on the Thomson/Reuters Financial Network mutual fund holdings and the CRSP stock return databases. Panel A of this table provides equal-weighted averages for subperiods from 1980 to 2007. Panel B includes TNA-weighted averages. The statistics include the following measures: industry-adjusted stock selectivity trading [ISST, Equation (5)], industry trading [IT, Equation (6)], characteristic trading [CT, Equation (2)], industry-adjusted stock selectivity [ISS, Equation (7)], industry selectivity [IS, Equation (8)], and characteristic selectivity [CS, Equation (4)]. Portfolio measures are calculated monthly and compounded to get quarterly values for each measure. Annualized returns are given below, but t-statistics are calculated on the quarterly time series. In all measures in this table, I limit the sample to those funds with a self-reported investment objective of “aggressive growth,”“growth,”“growth and income,” and “balanced” in the most recent annual report in the CRSP mutual fund database. For the periods 1980-1994, 1995-2007, and 1980-2007, the t­-statistics for significance are presented in parentheses. The number of distinct mutual funds equals 4,144.

  2. ***Significant at the 0.01 level.

  3.  **Significant at the 0.05 level.

  4.   *Significant at the 0.10 level.

Panel A. Equal-Weighted
1980-19840.00370.0043−0.00100.00510.00190.1062
1985-19890.0049−0.00010.00200.00580.00300.1794
1990-19940.01110.00300.00430.00180.00200.1019
1995-19990.0045−0.00160.0064−0.01210.01100.2533
2000-2007−0.00550.0001−0.00310.0266−0.00790.0587
1980-19940.0068**0.0023*0.00200.0042*0.00230.1308***
 (2.45)(1.89)(1.24)(1.82)(0.40)(3.59)
1995-2007−0.0017−0.00050.00060.0117−0.00060.1335***
 (−0.41)(−0.93)(0.21)(1.15)(−0.09)(2.83)
1980-20070.00270.00090.00130.00780.00090.1321***
 (1.08)(1.27)(0.85)(1.56)(0.20)(4.52)
Panel B. TNA-Weighted
1980-19840.00410.0036−0.00090.0020−0.00020.0971
1985-19890.0010−0.00110.00190.0117−0.00320.1782
1990-19940.01190.00250.0046−0.00070.00140.0943
1995-19990.0018−0.00010.0054−0.00400.00340.2522
2000-2007−0.00670.0010−0.00380.0134−0.00100.0566
1980-19940.0058**0.00150.00200.0045−0.00070.1251***
 (2.14)(1.21)(1.54)(1.47)(−0.13)(3.65)
1995-2007−0.00340.0006−0.00030.00670.00070.1318***
 (−1.05)(0.81)(−0.11)(1.14)(0.15)(2.90)
1980-20070.00140.00100.00090.0056*0.00000.1283***
 (0.61)(1.46)(0.65)(1.74)(−0.01)(4.65)

The equal-weighted (TNA-weighted) ISST for 1980-1994 is 0.68% (0.58%) per year and is highly significant; however, it is negative and insignificant in the second half of the sample. Thus, the significant SST found in the first half of the sample in Table I appears to be due, in large part, to trading stocks within industries. However, IT is only marginally significant for equal-weighting (at the 10% confidence level) and is not significant at all for TNA-weighting for the 1980-1994 period. The periods 1995-2007 and 1980-2007 demonstrate no contribution of excess returns from trading within stocks, industries, or characteristics.

These results demonstrate that, in the first half of the sample, managers had limited ability related to selecting industries, but they did have skill in identifying winners (and/or avoiding losers) within industries. As such, there is evidence of industry expertise. None of the other performance measures are significant. Thus, there is no other source of skill identifiable in Table II.

B. Explicit and Implicit Trading

In this section, I decompose the trading measures from Tables I and II to test for explicit and implicit trading. If the implicit trading measures are positive, mutual funds would earn excess returns with a buy-and-hold strategy (and might be better off by not trading to maintain the same weights). The explicit trading measure will be positive when a mutual fund actively changes the number of shares held so as to benefit from abnormal returns. A positive measure indicates an increase (decrease) in shares held for stocks with positive (negative) excess returns. Table III displays the explicit and implicit measures.

Table III.  Explicit and Implicit Trading—Industry-Adjusted Mutual Fund Return Decomposition, 1980-2007
Period ISST ISST –Explicit ISST –Implicit IT IT –Explicit IT –Implicit CT CT –Explicit CT –Implicit
  1. a

    A decomposition of mutual fund returns is provided below. The returns are based on the Thomson/Reuters Financial Network mutual fund holdings and the CRSP stock return databases. Panel A of this table provides equal-weighted averages for subperiods from 1980 to 2007. Panel B includes TNA-weighted averages. The statistics include the following measures and their explicit and implicit counterparts: industry-adjusted stock selectivity trading [ISST, Equation. (5)], industry trading [IT, Equation (6)], characteristic trading [CT, Equation (2)], industry-adjusted stock selectivity [ISS, Equation (7)], industry selectivity [IS, Equation (8)], and characteristic selectivity [CS, Equation (4)]. The explicit measures only include the impact of actual position changes, while the implicit measures only include the impact of price changes over the prior year. Portfolio measures are calculated monthly and compounded to obtain quarterly values for each measure. Annualized returns are given below, but t-statistics are calculated on the quarterly time series. In all measures in this table, I limit the sample to those funds with a self-reported investment objective of “aggressive growth,”“growth,”“growth and income,” and “balanced” in the most recent annual report in the CRSP mutual fund database. The t­-statistics for significance are presented in parentheses.

  2. ***Significant at the 0.01 level.

  3.  **Significant at the 0.05 level.

  4.   *Significant at the 0.10 level.

Panel A. Equal-Weighted
1980-19840.00370.0086−0.00490.00430.00340.0009−0.0010−0.00140.0004
1985-19890.00490.0078−0.0030−0.0001−0.00090.00080.00200.0023−0.0002
1990-19940.01110.0121−0.00090.00300.00150.00150.00430.00390.0004
1995-19990.00450.0082−0.0037−0.0016−0.0013−0.00030.00640.00530.0011
2000-2007−0.0055−0.0014−0.00420.00010.0006−0.0005−0.0031−0.0025−0.0005
1980-19940.0068**0.0096***−0.0028***0.0023*0.00120.0011***0.00200.00180.0002
 (2.45)(3.48)(−4.54)(1.89)(1.11)(3.55)(1.24)(1.21)(0.93)
1995-2007−0.00170.0023−0.0040***−0.0005−0.0001−0.00040.00060.00050.0001
 (−0.41)(0.74)(−2.85)(−0.93)(−0.21)(−1.28)(0.21)(0.20)(0.13)
1980-20070.00270.0061***−0.0034***0.00090.00060.00040.00130.00120.0001
 (1.08)(2.82)(−4.54)(1.27)(0.86)(1.37)(0.85)(0.87)(0.41)
Panel B. TNA-Weighted
1980-19840.00410.0080−0.00390.00360.00270.0009−0.0009−0.00140.0004
1985-19890.00100.0038−0.0028−0.0011−0.00220.00110.00190.0023−0.0004
1990-19940.01190.01190.00000.00250.00100.00140.00460.00400.0005
1995-19990.00180.0046−0.0027−0.00010.0003−0.00040.00540.00420.0011
2000-2007−0.0067−0.0034−0.00330.00100.0017−0.0008−0.0038−0.0035−0.0004
1980-19940.0058**0.0079***−0.0021***0.00150.00030.0012***0.00200.00190.0002
 (2.14)(3.30)(−3.27)(1.21)(0.32)(3.79)(1.54)(1.50)(0.75)
1995-2007−0.0034−0.0004−0.0031***0.00060.0012−0.0006−0.0003−0.00050.0002
 (−1.05)(−0.14)(−2.97)(0.81)(1.31)(−1.58)(−0.11)(−0.22)(0.28)
1980-20070.00140.0039**−0.0026***0.00100.00070.00030.00090.00070.0002
 (0.61)(2.06)(−4.32)(1.46)(1.05)(1.03)(0.65)(0.58)(0.52)

The equal-weighted (TNA-weighted) implicit ISST indicates that mutual funds would have lost a statistically significant 0.34% (0.26%) in excess return if their holdings had not changed over the year. The implicit ISST is significant in both halves of the sample and larger (in absolute value) in the second half. For both equal- and TNA-weighting, the full-sample explicit ISST measure is a significant and positive 0.61% and 0.39% per year, respectively, though explicit ISST is not significant in the latter half of the sample.

A positive explicit trading measure and a negative implicit trading measure provide greater insight into manager activity. These results only occur under two conditions. First, mutual funds increased the number of shares owned, the price of the stock decreased over the prior period, and the current quarter excess returns are positive. Second, mutual funds decreased the number of shares owned, the price of the stock increased over the prior period, and the current quarter excess returns are negative. In the first case, the mutual funds follow price drops by buying up more shares in stocks that subsequently outperform their benchmark. In the second case, the mutual funds sell off stocks following price increases and avoid the subsequent price correction.

The explicit and implicit returns could be possibly explained assuming managers are contrarian investors. Specifically, managers sell winners and buy losers. While a possibility, this specification does not allow us to distinguish between idiosyncratic price changes or broad market movements. Regardless, the strategy appears to add value since the trading allows the fund to avoid the negative excess returns associated with no portfolio changes at all.

Though the IT measures are not significant for the full sample, the implicit IT measure is significant for the period 1980-1994. An implicit buy-and-hold strategy for industries would have yielded a significant and positive 0.11% (0.12%) for equal-weighting (TNA-weighting). However, the inclusion of explicit trading appears to eliminate the small returns from industry selection. As in Daniel et al. (1997) and Wermers (2000), characteristic trading does not appear to be an important part of fund performance.

Taken together, the empirical results of Tables I-III provide a more nuanced picture of fund manager skill than that found in previous research. Table I provides some evidence of stock selection, at least in the first half of the sample period. The second half and full sample demonstrate no skill at all. Table II indicates that the trading ability from 1980 to 1994 is largely due to selecting stocks relative to their industry. Finally, Table III finds that managers are, on average, able to earn excess returns (before expenses) with a specific strategy of actively buying (selling) stocks that outperform (underperform) their industries. However, this trading only offsets losses the fund would have realized had the portfolio's holdings never changed.

IV. Alternate Estimation Methods

In the previous section, I explore new performance measures using gross returns and demonstrate some skill in choosing stocks within industries. One problem with any research utilizing holdings data is that the holdings are only allowed to change quarterly. If a manager earns excess returns on a short term (e.g., monthly) basis, the absence of monthly trading would unfairly penalize good funds and unfairly favor bad funds by ignoring the value added. The impact of this problem is demonstrated in Puckett and Yan (2010), who find that the absence of intra-quarter transactions leads to an underestimation of excess returns by 20 to 26 basis points per year.7 In this section, I propose alternate methods of constructing gross returns that allow portfolio trades to occur prior to the report date.

A. Alternate Gross Return Construction Methods

Mutual fund holdings are reported quarterly or semiannually. Daniel et al. (1997) (and essentially all other studies that use 13-f holdings data) approximate gross returns by assuming positions only change on the day the holdings are reported (at the end of a quarter). In reality, of course, a mutual fund changes positions at various times during the quarter and certainly not all at once. Thus, the gross returns used in most research incorrectly estimate the performance of managers whose trades during the quarter may have generated positive or negative excess returns. Put another way, the standard approach in the literature penalizes managers whose trades added value over the short run and rewards those managers who destroyed value in the short run. Nicolosi (2009) discusses this issue and demonstrates that changing the portfolio construction method changes the economic magnitude of the trading results documented in Chen et al. (2000).

A simple alternative to the Daniel et al. (1997) method is to assume holdings change some time during the quarter. To explore the impact of this issue on my evaluation of manager skill, I compare results from assuming that trading takes place at four alternative points in time: 1) at the end of quarter, 2) two months into the quarter, 3) one month into the quarter, and 4) at the beginning of the quarter.8 As seen in Figure 2, Daniel et al. (1997) only changes on the report date. The other measures assume that the end of the quarter holdings actually trade sooner and sooner into the quarter. A priori, there is no reason to prefer any particular assumption since there is little data on the intra-quarter trading behavior of mutual funds.9

Figure 2.

Gross Return Estimation Methods

The figure demonstrates the different ways holdings may be used to estimate gross returns for a quarter beginning at Time 0 and ending at Time 1. The “Holdings Reported” line illustrates what holdings the mutual fund reports, where R0 is the holdings at the beginning of the quarter and R1 is the holdings at the end of the quarter. The “Daniel et al.” line demonstrates the assumptions made in Daniel et al. (1997) regarding portfolio holdings. The “Two Month Switching,”“One Month Switching,” and “Immediate Switching” lines report three alternative portfolio construction methods. Two Month Switching assumes that a mutual fund does not change holdings until the end of the second month of the quarter, changing from R0 to R1 at the beginning of the third month. One Month Switching assumes holdings do not change from R0 to R1 until one month has passed. Immediate Switching assumes that the mutual fund immediately acquires R1 and holds it until the end of the quarter.

B. Performance and Alternate Approximations

Table IV replicates Table II under the alternative assumptions regarding the timing of trades. To save space, only the results for the full sample and the two half samples are shown. For ease of comparison, Table II results are repeated in Table IV[labeled End of Quarter Switching (Daniel et al., 1997)]. Broadly speaking, Table IV reports that the earlier the trading is assumed to take place, the larger the trading measures become (and the greater the estimated gross returns become). In contrast, the selectivity measures (and the CS measure) are relatively unaffected. For the trading measures, the results for the one- and two-month switching assumptions essentially look like monotonic interpolations of the immediate and the end-of-quarter numbers, declining relatively smoothly as trading is pushed later into the quarter.

Table IV.  Active and Passive Returns—Mutual Fund Return Decompositions Assuming Alternate Estimation Methods, 1980-2007
Estimation Method Period ISST IT CT ISS IS CS
  1. a

    A decomposition of mutual fund returns is provided below. The returns are based on the Thomson/Reuters Financial Network mutual fund holdings and the CRSP stock return databases. Panel A of this table provides equal-weighted averages for subperiods from 1980 to 2007. Panel B includes TNA-weighted averages. The column “Estimation Method” indicates when holdings are assumed to change. “Immediate Switching” assumes holdings reported at the end of the quarter were acquired at the beginning of the quarter. “One Month Switching” means the holdings were acquired one month into the quarter and “Two Month Switching” indicates the holdings were acquired two months into the quarter. “End of Quarter Switching (Daniel et al., 1997)” assumes holdings only change on the date holdings are reported. (See Figure II for a graphical example.) The results for “End of Quarter Switching (Daniel et al., 1997)” are the same as Table II and are reproduced for convenience. The statistics include the following measures: industry-adjusted stock selectivity trading [ISST, Equation (5)], industry trading [IT, Equation (6)], characteristic trading [CT, Equation (2)], industry-adjusted stock selectivity [ISS, Equation (7)], industry selectivity [IS, Equation (8)], and characteristic selectivity [CS, Equation (4)]. Portfolio measures are calculated monthly and compounded to get quarterly values for each measure. Annualized returns are given below, but t-statistics are calculated on the quarterly time series. In all measures in this table, I limit the sample to those funds with a self-reported investment objective of “aggressive growth,”“growth,”“growth and income,” and “balanced” in the most recent annual report in the CRSP mutual fund database. The t­-statistics for significance are presented in parentheses.

  2. ***Significant at the 0.01 level.

  3.  **Significant at the 0.05 level.

  4.   *Significant at the 0.10 level.

Panel A. Equal-Weighted
Immediate1980-19940.0248***0.0034***0.0043***0.00340.00300.1335***
 Switching (6.21)(2.68)(2.99)(1.38)(0.66)(3.62)
 1995-20070.0232***0.00080.00380.0079−0.00200.1332***
  (5.34)(0.68)(1.46)(0.87)(−0.30)(2.78)
 1980-20070.0240***0.0021**0.0041***0.00560.00060.1333***
  (8.31)(2.41)(2.84)(1.25)(0.14)(4.54)
One Month1980-19940.0170***0.0032***0.0029*0.00200.00170.1358***
 Switching (4.77)(2.91)(1.91)(0.89)(0.38)(3.80)
 1995-20070.0139***−0.00050.00340.0074−0.00330.1319***
  (3.35)(−0.42)(1.19)(0.82)(−0.47)(2.84)
 1980-20070.0155***0.00140.0031**0.0046−0.00070.1339***
  (5.77)(1.64)(2.01)(1.03)(−0.17)(4.71)
Two Month1980-19940.0108***0.0025**0.00250.00190.00160.1340***
 Switching (3.69)(2.07)(1.62)(0.89)(0.34)(3.72)
 1995-20070.0071−0.00070.00190.0078−0.00320.1312***
  (1.62)(−0.67)(0.60)(0.86)(−0.44)(2.85)
 1980-20070.0091***0.00100.00220.0048−0.00070.1327***
  (3.50)(1.11)(1.32)(1.06)(−0.16)(4.67)
End of Quarter1980-19940.0068**0.0023*0.00200.00420.00230.1308***
 Switching (2.45)(1.89)(1.24)(1.82)(0.40)(3.51)
 (Daniel1995-2007−0.0017−0.00050.00060.0117−0.00060.1335***
 et al., 1997) (−0.41)(−0.93)(0.21)(1.15)(−0.09)(2.83)
 1980-20070.00270.00090.00130.00780.00090.1321***
  (1.08)(1.27)(0.85)(1.56)(0.20)(4.52)
Panel B. TNA-Weighted
Immediate1980-19940.0151***0.0041***0.0030***0.0045*0.00000.1244***
 Switching (4.31)(3.15)(2.70)(1.88)(0.01)(3.67)
 1995-20070.0134***0.00090.00310.0038−0.00100.1300***
  (3.33)(0.87)(1.14)(0.76)(−0.19)(2.73)
 1980-20070.0143***0.0026***0.0030**0.0041−0.00040.1271***
  (5.48)(2.90)(2.19)(1.56)(−0.15)(4.48)
One Month1980-19940.0105***0.0043***0.00190.0033−0.00160.1268***
 Switching (3.42)(3.58)(1.56)(1.44)(−0.43)(3.78)
 1995-20070.0086**0.00000.00300.0034−0.00260.1316***
  (2.26)(−0.02)(1.10)(0.65)(−0.53)(2.88)
 1980-20070.0096***0.0022**0.0024*0.0033−0.00210.1291***
  (4.02)(2.46)(1.70)(1.23)(−0.70)(4.70)
Two Month1980-19940.0068**0.0035**0.00190.0030−0.00180.1257***
 Switching (2.54)(2.50)(1.50)(1.41)(−0.49)(3.73)
 1995-20070.0037−0.00030.00160.0040−0.00240.1313***
  (1.01)(−0.31)(0.55)(0.76)(−0.49)(2.89)
 1980-20070.0053**0.0017*0.00180.0035−0.00210.1284***
  (2.40)(1.77)(1.14)(1.28)(−0.70)(4.68)
End of Quarter1980-19940.0058**0.00150.00200.0045−0.00070.1251***
 Switching (2.14)(1.21)(1.54)(1.47)(−0.13)(3.65)
 (Daniel1995-2007−0.00340.0006−0.00030.00670.00070.1318***
 et al., 1997 (−1.05)(0.81)(−0.11)(1.14)(0.15)(2.90)
 1980-20070.00140.00100.00090.0056*0.00000.1283***
  (0.61)(1.46)(0.65)(1.74)(−0.01)(4.65)

Table IV also indicates that regardless of the time at which trading is assumed to take place, ISST is much more important economically than the other two trading measures, reinforcing my main conclusion that choosing stocks within industry is the primary source of manager skill. Whether managers possess skill in trading industries or characteristics depends critically upon the assumption of when the trades occur. However, even with immediate switching, neither IT nor CT is significant for the second half of the sample.

As previously noted, it is not possible to determine (using available data) which assumption regarding trade timing yields the most accurate gross return measures.10 The issue is very important, however. Assuming mutual fund managers change holdings at (or near) the beginning of a quarter implies much greater levels of performance (and greater total returns) than the Daniel et al. (1997) method. If the intra-quarter gross returns are closer to the true return, the prior literature on mutual funds has been underestimating both gross return and manager skill.

Whether the results assuming immediate switching are more accurate depends upon whether managers genuinely possess trading skill and how managers’ trades are influenced by prior returns. Under the assumption that managers are typically adept at trading on short run returns, then end-of-quarter switching will tend to mask any gains. However, if managers are skilled in this manner, then we need to determine why these excess returns are not present in the net returns. Specifically, the net returns historically demonstrate zero or negative excess returns for the typical fund manager. The difference between gross and net returns arises from fees, transaction costs, and nonequity holdings. If gross returns are larger than previously measured, the prior literature must be underestimating the impact of management costs on fund performance.

An alternative explanation is much simpler. The results are also consistent with mutual fund managers acting like momentum traders, buying stocks after excess returns are realized. If managers purchase a stock after it experiences higher returns during the quarter, the trading measures would be higher when trades are assumed to happen earlier. While there is evidence in the literature that mutual fund managers are momentum traders (Chen et al., 2000), very little research has addressed whether mutual funds buy (sell) stocks soon after those stocks realize positive (negative) excess returns.

The difference in performance between estimation methods also addresses a policy concern for mutual funds. The Securities and Exchange Commission requires reporting of mutual fund holdings quarterly. A common complaint of investors is that infrequent reporting hampers oversight and gives mutual fund managers too much latitude. The counter-argument by the mutual fund industry is that more frequent reporting encourages freeloading. Other funds or private investors can invest the same way as a successful trader without paying the successful trader's overhead. Since excess returns are higher when it is assumed that holdings are acquired before being made public, freeloading becomes less effective. Ge and Zheng (2006) analyze fund reporting frequency and find support for both transparency and freeloading effects. My results imply that freeloading does not yield the full benefit of having a successful fund manager.

Unfortunately, all models that rely on holdings data assume static trading in an environment that is well known to have frequent trading. Further clouding the issue is the holdings manipulation (or “window dressing”) that may occur around report dates such that the reported holdings do not represent a fund's actual holdings during the reporting period. Finally, and unavoidably, intra-period in-and-out trading will be missed completely. Until higher frequency holdings data becomes more widely available, mutual fund researchers inevitably must work with approximations to gross returns that can make measuring manager skill difficult.

V. Conclusion

When it comes to performance in gross returns, timing appears to be everything. I create a framework that measures the effect of mutual fund trading in several dimensions and find that the excess gross returns previously documented from 1980 to 1994 result entirely from trading stocks within industries. This value is highest when a fund explicitly trades stocks and does not necessarily result from medium-term momentum.

Using a holdout sample from 1995 to 2007, I find that mutual funds have performed much worse. Neither the trading nor selection measures demonstrate significant excess returns. My results suggest that manager skill has diminished or that 1980-1994 was not a representative period for mutual funds. I also find no industry expertise in the more recent sample period.

I do determine that assumptions regarding the timing of mutual fund trades have an impact on my conclusions. Allowing a trade to occur prior to the report may increase excess returns by as much as 2% per year. As in Nicolosi (2009), I demonstrate that the current methods of using holdings biases against finding large magnitude excess returns.

Footnotes

  • 1

    More specifically, the three measures add up to the gross return on the equity portion of the portfolio (cash and nonequity positions are ignored). Also, the return calculations assume that the portfolio is purchased using the weights from the end of the previous period and then held for one period. This assumption is necessary due to the “snapshot” nature of the holdings data and is discussed in detail in Section IV.

  • 2

    Characteristic benchmark portfolios are constructed by sorting stocks into 125 portfolios based on size, book-to-market, and prior year returns. For a complete discussion of characteristic portfolios see Daniel et al. (1997). For a discussion of the relative merits of characteristic portfolios, see Daniel and Titman (1997) and Kothari and Warner (2001).

  • 3

    Shorter lags yield similar results. I stay with one year lags to be consistent with prior research.

  • 4

    Holdings are sometimes (though rarely) reported in the middle of a quarter or there are multiple reports in a quarter. Following Daniel et al. (1997), only those holdings reported closest to the end of the quarter are included and are treated as having been reported at the end of the quarter like the majority of mutual funds.

  • 5

    I include balanced funds to be consistent with Daniel et al. (1997). This implicitly assumes that the equity portion of a mixed portfolio would be managed similarly to an equity-only fund. I tested for systematic differences between balanced funds and equity-only funds and found no economically important differences in any performance measure.

  • 6

    Industry returns, definitions, and construction are provided by Ken French at http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/.

  • 7

    Most of the holdings data is based on quarterly 13-f filings to the Securities and Exchange Commission (SEC). Puckett and Yan (2010) use a sample including intra-quarter trades and document that quarterly holdings will underestimate the trading skill of fund managers. I come to a similar conclusion in this section in that allowing trades to occur sooner than the end of the quarter demonstrates greater stock picking ability.

  • 8

    Nicolosi (2009) also investigates mid-quarter switching.

  • 9

    Both Elton et al. (2009) and Puckett and Yan (2010) find that some trades during the quarter are missed, but not necessarily that funds tend to buy or sell during certain parts of the quarter. Only persistent changes during certain times of the quarter would make the choice of approximation more or less correct.

  • 10

    The correlation coefficient of each gross return method with net returns exceeds 0.85 for each measure. Two month switching correlates the highest, but the difference is not economically significant relative to the other measures.

Ancillary