In search for managerial skills beyond common performance measures

Journal of Banking & Finance xxx (2016) xxx–xxx

Contents lists available at ScienceDirect

Journal of Banking & Finance journal homepage: www.elsevier.com/locate/jbf

In search for managerial skills beyond common performance measures Fan Chen a,⇑, Meifen Qian b,1, Ping-Wen Sun c,2, Bin Yu d a

Department of Finance, Merrimack College, United States School of Management, Zhejiang University, China c International Institute for Financial Studies and RCFMRP, Jiangxi University of Finance and Economics, China d School of Economics, Zhejiang University, China b

a r t i c l e

i n f o

Article history: Received 10 April 2015 Accepted 23 December 2015 Available online xxxx JEL classification: G11 G12 G14 G23 Keywords: Manipulation-proof performance measure (MPPM) Stock picking Managerial skills

a b s t r a c t One caveat of current literature on the value of active management is the lack of treatment for the performance measures that can be gamed. We propose to use the performance measure that can’t be manipulated with respect to the underlying distribution, time variation, nor estimation error, (the manipulation-proof performance measure (MPPM, Goetzmann et al. (2007)), to rank all active U.S. domestic equity mutual funds from 1980 to 2013 on a quarterly basis to analyze managerial skills. We find fund managers in the higher ranked persistently outperform lower ranked managers by posting higher gross and net fund returns, higher holding-based returns, and generating positive return gap. Analyzing the holdings of the portfolios indicates higher ranked managers hold stocks with higher information asymmetry, especially the growth companies that are younger, smaller, and with lower liquidity. Our results show that the spread on gross and net fund returns between highest ranked and the lowest ranked fund managers is between 49 and 52 basis points per month. The holding returns are statistically significant for up to six months indicating the stock picking skills exist for those higher ranked managers. Even though MPPM identifies managerial skills, the positive alphas may not be warranted due to their operating expenses. Ó 2015 Elsevier B.V. All rights reserved.

1. Introduction Anecdotal evidence suggests mutual fund managers have stock picking and/or market timing skill to outperform their peers on a pre-expenses and risk adjusted basis. Kacperczyk et al. (2014) document an existence of value creation mutual fund managers who have both market timing (in recessions) and stock picking (in booms) skills to generate persistence performance up to one year. Chung and Kim (2014) find that high consistency funds generate more than 2% additional risk adjusted returns in the subsequent year after accounting for fund size, past performances, sample period, and expenses. Petajisto (2013), and Cremers and Petajisto (2009) find fund managers who hold different holdings than their benchmark index could outperform these benchmarks on fee

⇑ Corresponding author at: 315 Turnpike St, North Andover, MA 01845, United States. Tel.: +1 225 288 7880. E-mail address: [email protected] (F. Chen). 1 Meifen Qian gratefully acknowledges the financial support from the National Natural Science Foundation of China (Grant No. 71303098). 2 Ping-Wen Sun gratefully acknowledges the financial support from the National Natural Science Foundation of China (Grant No. 71463018).

adjusted bases. Without using the holding data, Amihud and Goyenko (2013) regress fund returns on multifactor benchmark models to estimate the R2 and document managerial stock selection skills would predict their future performance. On the other hand, a vast amount of literature find little evidence that fund managers generate positive abnormal returns over long horizons. French (2008), and Fama and French (2010) argue that actively managed mutual funds cannot outperform passively managed funds to conclude on average those fund managers do not have stock picking skills. Similarly, Bollen and Busse (2004) find superior performance from mutual fund managers are short-lived.3 Given the controversial on the duration and the existence of managerial skills, perhaps a more important issue is whether we are using the appropriate performance measure that would be less likely for managerial manipulation. Recent mutual fund scandals

3 As compared to Jensen (1969) for stock selection over periods of 10–20 years, and Treynor and Mazuy (1966) and Henriksson (1984) for market timing over periods of 6–10 years. Others such as Brown and Coetzmann (1995), Carhart (1997), and Porter and Trifts (1998) find the persistence performance is either time sensitive/sample specific or cannot provide additional risk-adjusted returns beyond common risk factors.

http://dx.doi.org/10.1016/j.jbankfin.2015.12.008 0378-4266/Ó 2015 Elsevier B.V. All rights reserved.

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

2

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx

such as late trading and market timing has caused regulators emphasize on extensive disclosures and regulation reform.4 These issues are equally important if not more in hedge fund industry as many hedge funds in U.S. registered with the Securities and Exchange Commission (SEC) on a voluntary basis prior to the Dodd–Frank Act of 2010. Cici and Palacios (2015) find some mutual funds effectively use options to generate income even though the trades could lead to underperformance. If managers could generate high investment returns with little risk or report overly consistent returns over a longer time frame, it is important to know the persistent performance on managerial skills would not be another Bernie Madoff or Ponzi Schemes. Consequently, we demand a manipulation-proof performance measure that can really separate a skillful manager from a manipulated one. Goetzmann et al. (2007) argue fund managers could use informationless manipulations such as writing at-the-money call options and/or increasing their leverages to alter/improve fund performance.5 Since the performance measure in the current literature either assumes the underlying distribution (regardless of including the estimation error), or assumes stationary in the estimation of return distribution, or induces estimation error (for example, induces positive biases), managers could manipulate their fund returns since the current compensation structure to mutual fund managers rewards mainly on fund size.6 Consequently, the (shortlived) persistence on performance could be achieved without the contribution from their informed managers and their intellectual buy side analysts. This makes the findings on managerial skills less robust. In this paper, we use the manipulation-proof performance measure (MPPM, Goetzmann et al. (2007))7 to rank all active domestic equity fund managers. We also analyze the quarterly holdings to examine the stock picking skills of fund managers. Using the MPPM to rank fund managers and to examine their quarterly holdings has major advantages. First, by using the MPPM to rank fund managers on their ex ante performance allows us to filter out those uninformed trades from fund managers who achieve superior fund performance through writing calls and puts options or simply altering leverages. The measures allow us to precisely identify the percentage of funds that are beating the benchmark index. Second, by looking into their quarterly holdings and analyzing the changes through a time series trend allows us to establish linkages on their fund performance to stock selection and market timing skills. If managers achieve superior performance through luck or manipulation, the results from analyzing their quarterly holdings would serve this purpose to filter those managers who game through the performance measures. Our findings not only draw policy implications to the regulators but also shed lights to fund complexes that are in search for better compensation mechanisms to reward truly skillful fund managers from their manipulated counterparts. By combining the MPPM and quarterly holdings from fund managers, we could alleviate labeling

4 Since September 2003, the U.S. mutual fund industry has been mired in the worst scandal in its 65-year history. The scandal has produced in excess of 40 civil and criminal prosecutions, more than $2 billion in monetary sanctions, numerous Congressional hearings and bills, and a bevy of new regulations. For details, please refer to Bullard (2006). 5 Similarly, Weisman (2002), and Brown et al. (2004) also show that the traditional mutual fund performance measures can be gamed by fund managers. 6 For example, Carhart (1997) ranks by prior year return and by prior three-year abnormal return. 7 The MPPM has been gradually adopted in recent mutual fund performance studies. Titman and Tiu (2011) apply MPPM to examine hedge funds. Huang et al. (2011) apply MPPM to evaluate actively-managed mutual funds for robustness check on their empirical results. Bhattachara et al. (2012) find using MPPM can distinguish sophisticated investors from retail investors while Sharpe ratio cannot. Even though mutual funds managers are less found to engage on return smoothing or investing on illiquid assets, mutual fund are still the target since they are widely documented using derivatives (Lynch-Koski and Potiff (1999), and Cao et al. (2010)).

fraud-like outperformance or informationless trades from skillful managers. Our results show that managers in the higher ranked MPPM deciles persistently outperform lower ranked managers by posting higher gross and net fund returns, higher holding-based returns, and generating positive return gap. Those higher ranked fund managers tend to hold stocks with higher information asymmetry. Their holdings tend to be clustering in younger, smaller, growth, and lower liquidity stocks. To quantify the managerial skills through the Fama and French (2015) five-factor model analysis, our results show that higher ranked managers could generate 15–29 basis points while lower ranked managers would show a loss of 20–26 basis points on the subsequent month based on the quarter-end holdings. The differences of the monthly gross and net fund returns of the highest and the lowest rank can account for 49–52 basis points. Our results are also important to investors and fund holders. Investors could earn trading profits if they follow the disclosed quarterly holdings from the highest ranked managers to establish long positions of the stocks that are added to the portfolios and short positions of the stocks that are removed from their portfolios. The differences on long and short positions following the highest ranked managers based on their quarter end holding would generate 31 basis points based on the Fama and French (2015) fivefactor model. The persistent of the stock picking skills could live up to two consecutive quarters. A further analysis on the return gap indicates a positive and significant relationship between MPPM rank and return gap. However, it is important to note that even though the highest ranked managers have better stock picking skills and are significant different than the lower ranked managers, their fund returns are not robust enough among other asset pricing models to warrant positive alphas due to their frequent transactions and related operating expenses Our findings are consistent with Daniel et al. (1997) who show that stocks that are picked by mutual funds outperform a characteristic-based benchmark with the gain being approximately equal in magnitude to the funds’ management fee. Our findings are also consistent with Fama and French (2010) who find mutual funds in aggregate realize net returns that underperform four-factor benchmark by about the costs in expense ratios and most mutual funds do not have the skill to produce benchmark adjusted expected returns that cover costs. On the other hand, our results do show that highest ranked managers trade more often (have higher quarterly churn rate) and hold smaller and growth firms to outperform their counterparts, a result that is consistent with Yan and Zhang (2009) who argue those short-term institutional traders are more informed. However, our findings are contrast to Wermers (2000) who finds fund managers could earn more from picking stocks to offset their trading costs. Our results indicate even though top decile fund managers have stock picking skills, those skills do not warrant risk adjusted performance considering the trading and managerial expenses. Overall, our findings show that MPPM is a more reliable performance measure for investors to select equity mutual funds. We claim the MPPM to be a more effective measure to rank managers and to predict fund performance than other performance measures which are not isolated from manipulation and less robust in measuring managerial skills. Section 2 describes the research design, our hypothesis, and data construction. Section 3 reports our empirical findings. Section 4 concludes the paper. 2. Research design, hypothesis, and data construction Our research questions are (1) Whether MPPM can truly differentiate skillful and informed managers from their manipulated counterparts and be able to predict future fund returns, holding returns, and return gap without bias? (2) Whether the

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

3

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx

highly-ranked managers would outperform their lower-ranked managers or generate excess returns above the benchmark on an after-cost basis? Based on our research questions, we hypothesize funds with higher MPPM rank will be more informed so they will demonstrate better stock picking (Kacperczyk et al. (2014)) and add value from generating positive return gap (Kacperczyk et al., 2008). We hypothesize sample funds with higher MPPM ranked managers could outperform the lower ranked managers on a preexpense basis. Goetzmann et al. (2007) argue a manipulation-proof performance measure should meet the following four criteria: (1) the measure should produce a single score to rank each fund, (2) the score’s value is irrelevant to fund size, (3) only informed investors are able to produce higher scoring portfolios by taking advantage of arbitrage opportunities, and (4) the measure should be consistent with standard financial market equilibrium conditions. Our first step on the research design is to verify that MPPM is a more efficient performance measure than the common performance measures that do not accommodate for manipulation. We follow Qian and Yu (2015) to establish the comparison8 and to conclude MPPM indeed could mitigate the manipulation due to the lower (3% to 10% lower) frequency to beat the market when compared with the other common performance measures. Based on the results from Qian and Yu (2015), younger, larger, retail-based funds, and funds suffered from bad performance are more likely to manipulate. In addition, we need to test informed and uninformed market timers and be able to show that MPPM could more correctly recognize both informed and uninformed mutual fund managers. We will specify those hypotheses and how we design and execute our empirical tests in the following section.

The measure of MPPM is defined as shown in Eq. (1).

!

‘n½Eð1 þ r b Þ  ‘nð1 þ r f Þ Var½‘nð1 þ rb Þ

j

Timing t ¼

Nj X m ðxi;tj  xm i;t Þðbi;t Rtþ1 Þ

ð3Þ

i¼1

For fund j at the end of quarter t, we calculate each stock i’s value weight wi;tj in the fund and each stock i’s value weight wm i;t in the market portfolio of common stocks. bi;t is each stock i’s factor loading on market returns from past 12 months from the market model. Rm tþ1 is the quarterly returns of the market portfolio of common stocks in the following quarter t þ 1. When skillful fund managers expect the market returns are positive in the next quarter, they will put more weights on stocks with higher betas in their funds than those in the market portfolio. Similarly, when skillful managers expect the market returns to be negative in the next quarter, they will put less weights on stocks with higher betas in their funds than those in the market portfolio. From the above portfolio management, skillful fund managers will receive higher j

scores in market timing, Timing t . Nj   X i m ðxi;tj  xm i;t Þ Rtþ1  bi;t Rtþ1

j

Picking t ¼

ð4Þ

i¼1 j

For the measure of stock picking, Picking t of fund j at the end of Ritþ1

is the quarterly return of stock i in the following quarter t, quarter t þ 1. When fund managers are better at stocks picking

j

ð1Þ

ð2Þ

We choose our benchmark portfolio rb as the monthly market value-weighted returns of common stocks (with share code 10 or 11) from the CRSP database. We take average monthly r b and r f from January 1980 to December 2013 and substitute the average values into Eq. (2) to get our q estimation to be around 2.9 in our MPPM calculation. It is important to know why MPPM is a good measure because not only the measure traces the fund performance over an extended period above the risk-free rate, it also traces the fund performance relatively to the market index. An informed manager would earn a higher MPPM score through generating positive and persistent fund performance above the risk-free rate and benchmark. Goetzmann et al. (2007) compare the MPPM to the Sharpe, Sortino, SVP ratios, the alphas, and the Henriksson–Merton and Treynor–Mazuy timing measures in simulations and document only MPPM correctly shows that the manipulated portfolios are not as good as the market. 8

j

j

stock picking, Picking t .

We use Dt as one month and T as twelve months in our study. Hence, r t is a fund’s monthly returns and rft is monthly risk-free rate that can be pulled from the CRSP database. The parameter q is defined as shown in Eq. (2)

q¼

We follow Kacperczyk et al. (2014) to construct market timing Timing t and stock picking Picking t as shown in Eqs. (3) and (4).

where future idiosyncratic component Ritþ1  bi;t Rm tþ1 will outperform its systematic component bi;t Rm tþ1 , they will put more weights on those stocks in their funds than the weights in the market portfolio. Consequently, fund managers will receive higher scores in

2.1. MPPM

T 1 1X ‘n MPPM ¼ ½ð1 þ rt Þ=ð1 þ r ft Þ1q ð1  qÞDt T t¼1

2.2. Market timing and stock picking

We compare the Sharpe ratio, Jensen’s alpha, Treynor ratio, Sortino’s downsiderisk, Sortino measure, van der Meer and Plantinga’s upside-potential measures, Henriksson-Merton and Treynor-Mazuy timing measures, and few others (Lhabitant (2000), Ferson and Siegel (2001), Richard (2001), Bollen and Pool (2008)).

j

j

After calculating Timing t and Picking t , Skill Index could be calculated as follows:

Skill Indextþ1 ¼ xt Timing t þ ð1  xt ÞPicking t j

j

j

ð5Þ

It is defined that for fund j in quarter t þ 1 as a weighted averj

j

age of Timing t and Picking t , in which the weights we place on each measure are based on the state of the business cycle, where weight on Timing equal to 0 6 xt 6 1, where xt is the recession probability from the survey from the Federal Reserve Bank of Philadelphia website. Picking always gets the complementary weight 1  xt . 2.3. Average quarterly churn rate CRKT Even though we only have the end of the quarter holding data, we follow Yan and Zhang (2009) to construct a fund’s turnover estimate as shown in Eqs (6)–(9).

CR buyk;t ðSk;i;t > Sk;i;t1 Þ ¼

Nk X jSk;i;t Pi;t  Sk;i;t1 Pi;t1  Sk;i;t1 DPi;t j i¼1

ð6Þ CR sellk;t ðSk;i;t 6 Sk;i;t1 Þ ¼

Nk X jSk;i;t P i;t  Sk;i;t1 Pi;t1  Sk;i;t1 DPi;t j i¼1

ð7Þ We first summarize each fund k’s stock i holding cash inflow/ outflow in quarter t and aggregate those stocks’ cash inflow/outflow to be fund k’s aggregate purchase and sale at the end of quarter t as shown in Eqs. (6) and (7). Sk,i,t1 and Sk,i,t are number of

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

4

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx

shares held by fund k at the end of quarter t  1 and t. Pi,t1 and Pi,t are listed share price of stock i at the end of quarter t  1 and t. We use CRSP’s price adjustment factor to adjust stock splits and stock dividends to calculate stock i’s price change DPi,t at the end of quarter t.

CRk;t 

minðCR buyk;t ; CR sellk;t Þ PNk Sk;i;t Pi;t þSk;i;t1 Pi;t1 i¼1

ð8Þ

2

We then calculate each fund k’s churn rate as shown in Eq. (8). We use the minimum of aggregate purchase and sale price divided by the fund k’s average portfolio holding value during quarter t to be its churn rate at the end of quarter t.

AVG CRk;t ¼

3 1X CRk;tj 4 j¼0

ð9Þ

Finally, we calculate each fund’s average churn rate over past four quarters as shown in Eq. (9).

servative total asset investment and aggressive total asset investment firms.

rpt  r ft ¼ apt þ bMTKRF;pt  MKTRF t þ bSMB;pt  SMBt þ bHML;pt  HMLt þ bRMW;pt  RMW t þ bCMA;pt  CMAt þ ept

ð10Þ

For robustness, we propose to use the q-factor model that is proposed by Hou et al. (2015), hereafter HXZ q-factor, which consists of the market factor, a size factor, an investment factor, and a profitability factor that largely summarizes the cross section of average stock returns. It is worth to note the investment factor has a high correlation (0.69) with HML and the ROE (profitability) factor has a high correlation (0.50) with the momentum factor, UMD, by Carhart (1997). Since the HXZ q-factor model could explain Carhart (1997) momentum factor and represent investment-based asset pricing, we will use the HXZ q-factor model as our robustness measure. Due to the economy of the presentation, we will report our main results based on Fama and French (2015) five-factor model and the selective robustness check with the HXZ q-factor result in the appendix (Appendix Table 1 and Table 2).9

2.4. Choosing the Fama and French (2015) five factor model 2.5. Other stock characteristic variables Since our hypothesis is to examine the gross and net fund returns, holding returns, and return gap after we carefully mitigate the possibility from manipulation their reported fund performance, we need to be carefully select the performance measures that can capture whether those managers would create value pre- and posttransaction and account for the related operation expenses. The common performance measure in active equity funds then to fall into the Fama and French (1993) three-factor model and the Carhart (1997) four-factor model. Fama and French (2015) argue that book to market is a noisy proxy for the expected returns because the market cap also responds to forecasts of earnings and investment. Therefore, firm’s size, book to market ratio, profitability, and investment are all correlated with its expected stock returns. Alti et al. (2012) show institutional investors’ trend chasing does not rely on the momentum anomalies. They do not trade stocks based on past returns per se but on news that affect investors’ confidence which changes the demand for stocks. Fund managers appear to believe that they possess superior skills in analyzing their past profit makers. Such beliefs may represent rational expectations or wishful thinking. Given Novy-Marx (2013) documents that the gross profitability (difference between total revenue and total cost divided by total asset) predicts future stock returns and suggests gross profitability could be a good proxy for expected earnings. Besides, Titman et al. (2004) find firms that substantially increase capital investments subsequently achieve negative benchmark-adjusted returns. On the other hand, Cooper et al. (2008) also find asset growth rates are strong predictors of future abnormal returns. Their findings suggest asset growth could proxy for expected growth in book equity. We believe choosing an asset pricing model that factors the asset growth and profitability is important. We choose the Fama and French (2015) five-factor model that captures size, value, profitability, and investment patterns to apply to measure the fund and holding returns since those factors thoroughly factoring the cross section of stock returns. The model is shown in Eq. (10) to calculate the risk adjusted returns ap of different MPPM ranked fund managers and their portfolio returns rp, where MKTRF is the market risk premium (market return in excess of 3 month Treasury bill rate), SMB is the return difference between small size and big size firms, HML is the return difference between high book to market ratio to low book to market ratio firms, RMW is the return difference between robust operating profitability and weak operating profitability firms, and CMA is the return difference between con-

In order to find out the information content for the stock holdings among different portfolio managers, we have chosen several stock characteristic variables in our analysis. Specifically, we include four risk related characteristic variables. They are size, book to market, past performance, and gross profits to total assets (Novy-Marx (2013)). We also include three variables to measure the liquidity characteristics of their holdings. They are monthly turnover, illiquidity ratio (Amihud (2002)), the monthly trading volume divided by its share outstanding in the most recent 3 months), and the daily (high-low) spread (Corwin and Schultz (2012)) numerical solution of high-low estimator for the most recent quarter). Following Falkenstein (1996), Gompers and Metrick (2001), and Yan and Zhang (2009), we have also include age, dividend yield, and return volatility in our analysis. 2.6. Data construction For our empirical analysis, we use the CRSP Survivorship Bias Free Mutual Fund Database for mutual fund return and fund characteristics. We use quarterly holdings for all mutual funds from the Morningstar Direct Database. We merge the two databases only if the observations of the monthly return data on both databases exist and to merge the CRSP dataset and Morningstar dataset by share class, year, and month. We also use the CRSP U.S. Stock Database to extract stock price to compute holding returns. The holdings database includes only common stock positions and excludes none equity holdings. To adjust holding returns for various asset classes, we proxy these assets’ returns using published indices. We use the total return of the Barclay Aggregate Bond Index for bonds. We use the Treasury bill rate for cash holdings. Since there is no reliable index available for preferred stocks and other assets, we assume that the return on preferred stocks equals the return of the Barclay Aggregate Bond Index. Although bonds are not traded as often as stocks and nonsynchronous trading may be an issue to calculate funds’ portfolio value, the concern about wrongly-calculated value of funds due to nonsynchronous trading for holdings in bonds could be alleviated since we only 9 We have also performed Fama and French (1993) three-factor and Carhart (1997) four-factor analysis for the performance measures after we rank managers based on MPPM. Those results are also available based upon request.

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

5

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx Table 1 Summary statistics of sample funds. Panel A: Fund characteristics in sample period and sub-periods Variables Overall sample

TNA (millions) Age (years) Expense ratio (%) Portfolio turnover (%) Common stock proportion (%) Bond proportion (%) Cash proportion (%) Fund returns (%) MPPM (%) Sharpe ratio Number of unique funds Total number of observations (fund-quarters)

1980–1989

1990–1999

Mean

Std. Dev.

Median

Mean

Std. Dev. Median

Mean

Std. Dev. Median

Mean

Std. Dev. Median

1278.04 14.20 1.18 86.42 90.76 0.63 4.15 2.68 1.97 0.67 3903 109,517

5763.98 14.21 0.51 126.21 17.20 3.35 7.66 10.54 24.01 1.24

213.80 10.00 1.15 59.00 95.87 0.00 1.92 3.43 6.71 0.72

309.56 20.51 1.06 75.97 67.76 2.30 9.67 4.32 3.26 0.66 473

633.70 15.50 0.46 67.80 36.24 5.63 11.83 10.13 20.50 1.13

867.59 15.25 1.24 79.17 81.93 0.89 8.46 4.79 7.92 0.94 1702

2803.62 15.51 0.53 84.92 24.23 4.69 9.44 10.87 16.32 1.06

181.58 9.00 1.17 57.00 90.30 0.00 5.90 4.16 8.70 0.86

1444.56 13.42 1.18 88.47 94.46 0.41 2.95 2.14 0.75 0.62 3541

6388.61 12.70 0.51 135.19 8.00 2.67 6.19 10.45 25.30 1.27

109.40 17.00 1.00 58.00 85.00 0.00 6.00 4.79 6.77 0.68

2000–2013

238.50 10.00 1.17 59.00 96.67 0.00 1.43 3.11 6.21 0.69

Panel B: Fund characteristics in different MPPM rank groups Fund characteristics

TNA (millions) Age (years) Expense ratio (%) Portfolio turnover (%) Common stock proportion (%) Bond proportion (%) Cash proportion (%) Fund returns (%) MPPM (%) Sharpe ratio Number of unique funds Total number of observations (fund-quarters)

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

739.03 13.34 1.29 104.49 90.00 0.50 4.22 0.67 16.18 0.09 2409 10,667

1114.51 14.44 1.21 89.61 90.53 0.51 3.86 0.75 6.57 0.23 2980 10,964

1233.79 14.54 1.15 82.51 90.84 0.57 3.90 1.36 3.27 0.37 3029 11,049

1426.29 14.69 1.13 79.12 91.30 0.54 3.75 1.89 0.79 0.49 2975 10,986

1577.92 14.66 1.11 75.36 91.26 0.60 3.81 2.27 1.25 0.60 2934 10,916

1546.14 14.93 1.13 76.16 91.23 0.63 3.97 2.77 3.22 0.70 2918 11,134

1501.05 14.76 1.15 76.31 91.09 0.64 4.20 3.25 5.37 0.82 2929 11,064

1393.74 14.16 1.16 77.61 91.01 0.68 4.32 3.83 7.67 0.96 2911 10,971

1277.34 13.80 1.19 80.80 90.66 0.77 4.39 4.72 10.81 1.12 2788 11,042

943.79 12.59 1.26 90.72 89.59 0.82 5.10 6.58 17.95 1.47 2312 10,724

Panel A in this table reports mean, median and standard deviation on fund characteristics of all domestic mutual funds in our sample. Data is from CRSP Survivorship Bias Free Mutual Fund Database. The sample period is from January 1980 to December 2013. Fund characteristic variables include total net assets (TNA) which measures total net assets under management (in million dollars). Age (in years) is measured since fund’s inception. Expenses ratios (in percentage) is the annual operating expenses divided by the Asset under management. Portfolio turnover (in percentage) is calculated by taking either the total amount of securities purchased or the total amount of securities sold over a year divided by the total net assets of the fund. Common stock proportion (in percentage) is the holding that is composed in common stocks. Bond proportion (in percentage) is the holding that is composed in all kinds of bond. Cash proportion (in percentage) is the holding that is composed in cash. Fund returns (in percentage) are quarterly raw returns calculated from CRSP monthly fund returns. Sharpe ratio is calculated by subtracting the risk-free rate from the fund returns and divided by the standard deviation of the fund returns. MPPM (in percentage) is calculated from the past 12-month raw fund returns. We report fund characteristics for the all sample funds and split the sample funds into sub-period from 1980–1989, 1990–1999, and 2000–2013. In Panel B, we report the mean of the fund characteristics after we rank all domestic equity mutual funds by manipulation-proof performance measure (MPPM (Goetzmann et al. (2007)) from their past 12-month returns to sort them into deciles at the end of each quarter. Rank 10 indicates the highest score and Rank 1 for the lowest among the deciles.

choose equity funds in our analysis and most of the sample funds hold more than 95% in equity positions during the sample period. Our sample period is from January 1980 to December 2013. We restrict our data on actively managed domestic equity mutual funds (with the objective of aggressive growth, growth, and growth and income) for which the holdings data are mostly complete and reliable. Consequently, we eliminate bond, balanced, money market, international, and index funds.10 We further exclude funds less than $5 million and funds that hold less than 10 stocks per quarter in the previous month. We also exclude funds without total net assets. In order to calculate the annualized MPPM score, funds in the final sample should have at least 12 months of consecutive returns data. Funds that end up in the graveyard are included in the sample. For funds with multiple share classes, we compute fund-level variables by aggregating across the multiple share classes. Since the holding data is generally available on quarterly except few funds voluntarily report their holdings on monthly frequency, we restrict our analysis on quarterly basis. The final sample includes 3903 distinct funds and a total of 109,517 fund-quarter observations.

10 For a complete data process on how to select the active domestic equity funds can refer to Huang et al. (2011).

3. Empirical results 3.1. Fund characteristics We report summary statistics of our sample funds in Table 1. In Panel A, we split the sample across three different time period to see how sensitive the data of each sub-sample to the whole sample period. On average, our sample funds are 14.20 years from their inceptions and carrying expense ratios of 1.18% annually. The quarterly raw returns are on average of 2.68%, or about 10.72% annually. The average fund size has grown rapidly across sub-sample, from 310 million dollar in total net assets in the 1980s to grow to more than 1 billion dollar in the 2000s. In order to examine the different skillsets of fund managers, at the end of each quarter from our entire sample period from January 1980 to December 2013, we rank each fund by its MPPM from its past 12-month returns to sort them into deciles. In Panel B, the average total net assets for rank 1 (lowest MPPM rank) group is 739 million dollars and 944 million dollars for the rank 10 (highest MPPM rank). All other ranks are all above 1 billion dollars. The relationship between fund size and performance is consistent with the diseconomies of scales from the mutual fund literature. Fund age, expense ratios, and portfolio turnover show u-curve patterns. Highest ranked fund managers trade more often and charge higher expenses when they

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

6

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx

Table 2 Subsequent managerial skills of sample funds based on MPPM rankings. Managerial skills measure

Equally-weighted fund returns Equally-weighted portfolio holding returns Equally-weighted return gap Value-weighted fund returns Value-weighted holding returns Value-weighted Return gap Value-weighted timing Value-weighted stock picking Manager-skill Value-weighted average churn rate

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

Diff (Rank 10–Rank 1)

2.764 2.778 0.015 2.641 2.780 0.136 3.252 0.979 1.734 5.418

3.111 3.053 0.058 2.755 2.779 0.032 2.888 0.583 0.837 5.542

3.080 2.987 0.094 2.994 3.021 0.025 2.824 0.394 0.383 5.103

3.218 3.147 0.074 3.069 3.066 0.009 2.638 0.321 0.118 4.708

3.283 3.104 0.176 3.223 3.180 0.040 2.623 0.186 0.04 4.8

3.470 3.296 0.174 3.261 3.193 0.068 2.547 0.101 0.526 4.857

3.557 3.314 0.244 3.346 3.191 0.154 2.606 0.019 0.765 4.942

3.700 3.547 0.153 3.513 3.460 0.051 2.503 0.218 1.196 4.833

3.884 3.752 0.134 3.587 3.554 0.034 2.661 0.234 1.699 5.259

4.100 3.796 0.309 3.893 3.768 0.122 2.758 0.438 2.742 5.395

1.340⁄⁄ (2.55) 1.036⁄ (1.70) 0.314⁄⁄ (2.05) 1.252⁄⁄ (2.14) 0.988⁄ (1.69) 0.258⁄⁄ (2.11) 0.494 (1.34) 1.417⁄⁄⁄ (3.37) 4.476⁄⁄⁄ (12.30) 0.023 (0.09)

This table reports subsequent managerial skills measure of the sample funds. We rank all domestic equity mutual funds by manipulation-proof performance measure (MPPM (Goetzmann et al. (2007)) from their past 12-month returns to sort them into deciles at the end of each quarter. Rank 10 indicates the highest score and Rank 1 for the lowest among the deciles. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stocks returns are from CRSP U.S. Stock Database. The sample period is from January 1980 to September 2013. Quarterly fund returns are calculated from CRSP monthly fund returns. Quarterly portfolio holding returns are based on the most recently quarterly-end holdings which include stocks, bonds and cash. Return gaps are calculated based on Kacperczyk et al. (2008) by using the difference between fund returns and holding returns. Market timing, stock picking, and manager-skill variables are based on Kacperczyk et al. (2014). Average churn rate (per annual) is based on Yan and Zhang (2009). Average number of fund quarters in each group is the average funds multiply by the 135 quarters of the entire sample period. T-statistics are reported in the parentheses only for the last column. We test the differences (Diff (Rank 10–Rank 1)) of the highest rank (rank 10) and the lowest rank (rank 1). The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels.

are compared to the managers ranked in the middle deciles. However, the lowest ranked managers charge the highest expenses and have the highest portfolio turnover among the 10 groups. We do not find major differences on common stock holding proportion among these 10 groups. For the quarterly value-weighted fund returns, the lowest ranked funds carry 0.67% while the highest ranked funds carry 6.58%. After using the past 12 months fund returns to calculate the quarterly value-weighted MPPM, the average quarterly MPPMs across the 10 groups ranges from the lowest of 16.18% to the highest of 17.95%. 3.2. Can MPPM predict future fund and holding returns? Since there are some variation among fund characteristics among different ranked fund managers, we are interested to perform a variety measures of managerial skills to see how those measures hold up through different MPPM ranked managers. We rank our sample funds from their past MPPM ranking to sort them into decile at the end of each quart to use the measure to test the subsequent fund and holding returns. Since Kacperczyk et al. (2008) use the return gap as the measure for a fund’s short-term performance due to unobserved actions to capture the manager’s value-added relative to the previously disclosed holdings, we adopt the return gap as one proxy for the managerial skills. We extract weights of stock holdings in the current quarter to calculate their value-weighted holding returns and use the fund size to calculate value-weighted fund returns for the subsequent quarter to provide insights on whether higher ranked managers produce higher fund returns, holing returns, and positive return gap. We also include the measure of market timing, stock picking, and manager-skill measured used in Kacperczyk et al. (2014) to see how robust the MPPM ranked managers are. We report our findings in Table 2. Sample funds with higher ranked MPPM this quarter generate both higher future fund returns and holding returns in the subsequent quarter. The lowest ranked MPPM funds generate value-weighted fund returns of 2.64% in the next quarter. Their holding returns are 2.78% per quarter. On the other hand, the highest ranked MPPM funds achieve value-weighted fund returns of 3.89% in the subsequent quarter. Their holding returns are 3.77% per quarter. The differences between the top and the bottom decile are 125 basis points on quarterly fund returns and 99 basis points on quarterly holding

returns. The return gap between the top to the bottom ranked managers indicates 26 basis points. All those three measures are statistically and economically significant. The monotonically increase on those measures are suggesting higher ranked MPPM managers persistently create value, while the lower ranked counterparts destroy value. Moreover, the top ranked managers exhibit the best stock picking skills with a score of 0.438 while lower ranked managers produce a negative score. Moreover, the difference in picking and overall skill between two extreme deciles are also significant. Those measures confirm the findings with Kacperczyk et al. (2014) that there are group of fund managers that are consistent demonstrating their managerial skills. On the other measures, fund managers with highest MPPM rank also exhibit higher average quarterly churn rate of 5.40%, a pattern of the U-curve that is similar to the lowest ranked managers (with the churn rate of 5.40%). High churn rate funds are possibly more informed (Yan and Zhang (2009)). Since the lowest ranked managers also carry high churn rate, it is reasonable to suspect those managers are more opt to satisfy the redemptions from their existing fund holders due to sub-par performance. On the other hand, highest ranked managers are more likely to have rapid fund inflows and thus have to trade more often to offset the growing size that might deteriorate the fund performance. 3.3. Stock holding characteristics From the first two Tables, it is reasonable to conclude the highest ranked fund managers have better managerial skills to generate higher fund and holding returns. We need to further investigate whether their stock holdings contain valuable information to separate the skillful managers from those that take higher risks but do not pay-off. We report stock holding characteristics of sample funds from those ten different ranked MPPM groups in Table 3. Our results show that fund managers with highest rank tend to hold smaller stocks (with average size of 18.752 billion dollars). They hold stocks with the lowest book-to-market ratio (0.370). They appear to use momentum strategies as they tend to hold the stocks with the highest past year performance with the annual rate of returns of 46.73%. There is no statistically different on the gross profits (6.44%). From the findings, it appears that highest ranked managers prefer small, growth, and momentum stocks that have better past performance.

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

This table reports value-weighted stock holding characteristics for the sample funds. We rank all domestic equity mutual funds by manipulation-proof performance measure (MPPM (Goetzmann et al. (2007)) from their past 12month returns to sort them into deciles at the end of each quarter. Rank 10 indicates the highest score and Rank 1 for the lowest among the deciles. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stock characteristics are from CRSP U.S. Stock Database. The sample period is from January 1980 to December 2013. Size (in millions) is the quarterly-end market capitalization of the stock. Book to market ratio is the most recent quarterly-end book equity value to divide by fiscal quarterly-end market equity value. Past year performance is stock’s cumulative past 12-month returns. Gross profits to total asset is taking gross profits (revenue minus cost of goods sold) to divide by total assets. Monthly turnover is taking average trading volume to divide by shares outstanding in most recent three months. Amihud (2002) illiquidity ratio of stock is taking the average daily absolute return to divide by dollar trading volume in millions at the current quarter. Daily (high-low) spread is the numerical solution defined in Corwin and Schultz (2012). Age is number of months since recorded in CRSP Database. Dividend yield is the cumulative cash dividends of the recent 12 months to divide by stock price at the quarter end. Monthly volatility is the monthly standard deviation of recent 24 months stock returns. T-statistics are reported in the parentheses only for the last column. We test the differences (Diff (Rank 10 – Rank 1)) of the highest rank (rank 10) and the lowest rank (rank 1). The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels.

Diff (Rank 10–Rank 1) Rank 10 (highest)

18,752.93 0.370 46.73% 6.44% 14.78% 0.067 1.45% 296 0.53% 9.62% 22,018.07 0.377 36.36% 6.50% 13.65% 0.051 1.38% 323 0.55% 9.33%

Rank 9 Rank 8

24,005.47 0.376 34.23% 6.58% 13.59% 0.049 1.37% 327 0.54% 9.31% 24,867.55 0.381 30.24% 6.52% 13.25% 0.042 1.36% 336 0.55% 9.21% 24,757.88 0.379 28.44% 6.59% 13.23% 0.047 1.36% 333 0.54% 9.25% 25,838.73 0.381 26.69% 6.59% 13.18% 0.047 1.36% 336 0.56% 9.19% 26,182.63 0.381 25.13% 6.52% 13.28% 0.045 1.38% 335 0.56% 9.28%

Rank 7 Rank 6 Rank 5 Rank 4

26,642.04 0.387 22.67% 6.47% 13.13% 0.051 1.38% 339 0.55% 9.21%

Rank 3 Rank 2

25,580.99 0.386 21.94% 6.56% 13.49% 0.057 1.40% 334 0.55% 9.30% 21,637.29 0.401 17.68% 6.38% 15.00% 0.092 1.52% 305 0.51% 9.97% Size (in millions) Book to market ratio Past year performance (month t  11 to t) Gross profits to total asset Monthly turnover (month t  2 to t) Amihud (2002) illiquidity ratio Daily spread Age (in months) Dividend yield (month t  11 to t) Monthly volatility (month t  23 to t)

Rank 1 (lowest)

MPPM rank Stock holding characteristics

Table 3 Stock holding characteristics of sample funds based on MPPM rankings.

2884.35 (1.42) 0.031⁄⁄ (2.55) 29.05%⁄⁄⁄ (9.73) 0.06% (0.50) 0.22% (0.35) 0.025⁄⁄ (2.42) 0.07%⁄⁄ (2.02) 8 (1.01) 0.02% (0.74) 0.37%⁄ (1.95)

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx

7

We further look into stock liquidity characteristics to see whether different ranked managers have any preference on liquidity. Our results show managers with highest ranked hold stocks with average monthly turnover rate of 14.78% (second highest from the ten groups), with Amihud (2002) illiquidity ratio of 0.067 (second highest), and with daily spread of 1.45% (second highest among the deciles). The findings suggest highest ranked managers prefer stocks with higher turnover rate, above average price impact and bid-ask spread, indication of stocks with relative higher information asymmetry. Even though managers in the lowest ranked display similar liquidity preferences on stocks, they apparently do not earn illiquidity premiums to cover their risks. Highest ranked managers, on the other hand, are better informed to profit from the illiquidity. Further we examine three additional stock characteristics from their holdings for robustness check. Among 10 groups of ranked managers, highest ranked managers hold stocks with the youngest companies (average age of 296 months)), lower dividend yield (second lowest annual dividend yield (0.53%)), and higher return volatility stocks (second highest monthly return volatility (9.88%) among the 10 groups). Not surprisingly, lowest ranked managers have similar holding characteristics with highest ranked managers. However, lowest ranked managers take additional idiosyncratic risks that do not pay off. Overall, our results show that even though lower ranked managers take additional risks on stocks with low liquidity with higher information asymmetry. However, the risks do not pay-off. Lower ranked managers end up with the lowest subsequent fund and holding returns. Highest ranked managers, on the other hand, have informed information to earn additional returns based on the risks they take. 3.4. Common risk factors analysis (Fama and French (2015) fivefactor) for subsequent holding returns through reported quarterly holdings Up to this point, we are convinced that lower and higher ranked managers prefer taking additional risk but not necessarily generating better risk-adjusted fund and holding returns on the subsequent quarter. We further apply common risk factors to examine the stock holdings to see whether their managerial skills in stock picking can be explained by the common risk factors. In Table 4, we report Fama and French (2015) five-factor model through examine the quarterly holdings. Our results show the higher ranked managers would be able to pick stocks that could earn positive abnormal returns (alphas), at about 15–29 basis points, in the subsequent month. On the other hand, the lower ranked managers would earn negative returns (alphas), at 20–26 basis points, in the subsequent month. The middle ranked managers do not generate significant risk-adjusted returns from the holdings for the subsequent month. The test for the differences from the highest ranked and lowest ranked indicates the managerial skills on the quarterly holdings could account for 49 basis points. From the common risk factor analysis, higher ranked managers have the lowest factorloading on market risk premium (1.03), the highest factorloading on SMB (0.26), the most negative factor-loading on HML (0.24), the most negative factor-loading on RMW (0.12), and the highest factor-loading on CMA (0.08). Analysis from the fivefactor loading concludes managers of highest ranked could earn additional risk premiums on loading small, growth, stocks with higher information asymmetry but be able to predict their future performance from the existing lower profitability and conservative on the capital expenditure investment. Before account for the transaction cost, the holdings for higher ranked managers do contain informed information and their risk-taking strategies in identify stocks with higher information asymmetric payoff through earning positive risk-adjusted performance of the subsequent

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

8

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx

Table 4 Monthly five-factor model analysis of portfolio holding returns of sample funds based on MPPM rankings. 5-Factor model

Alpha t-stat Mktrf_beta t-stat SMB_beta t-stat HML_beta t-stat RMW_beta t-stat CMA_beta t-stat

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

Diff (Rank 10–Rank 1)

0.0020 1.39 1.0995 21.00 0.1179 1.33 0.1725 1.89 0.0378 0.25 0.3027 2.01

0.0026⁄⁄⁄ 2.99 1.0934 53.45 0.0580 1.04 0.1261 1.91 0.1550 1.75 0.1203 1.48

0.0013⁄ 1.73 1.0821 54.43 0.0390 1.05 0.0768 1.48 0.1207 1.92 0.0495 0.81

0.0009⁄ 1.65 1.0490 96.28 0.0399 2.00 0.0100 0.36 0.0797 2.70 0.0070 0.18

0.0006 1.64 1.0776 89.41 0.0205 0.76 0.0351 1.14 0.1007 2.85 0.0518 1.63

0.0002 0.57 1.0742 104.32 0.0566 3.84 0.0029 0.15 0.0604 2.48 0.0100 0.28

0.0005 1.24 1.0646 83.97 0.0913 3.72 0.0114 0.41 0.1344 3.98 0.0662 1.68

0.0007 1.22 1.0712 83.47 0.1245 4.14 0.0776 2.39 0.0431 0.80 0.0993 1.77

0.0015⁄⁄ 2.03 1.0360 61.66 0.1756 3.88 0.1037 2.07 0.0171 0.22 0.0749 0.94

0.0029⁄⁄ 2.42 1.0317 35.97 0.2559 3.60 0.2380 2.71 0.1197 0.91 0.0793 0.58

0.0049⁄⁄ 2.08 0.0678 0.97 0.1380 0.91 0.4105⁄⁄ 2.37 0.1574 0.57 0.3820 1.37

This table reports subsequent monthly value-weighted alphas and betas from the portfolio holdings of sample funds by using five-factor model from Fama and French (2015). We rank all domestic equity mutual funds by manipulation-proof performance measure (MPPM (Goetzmann et al. (2007)) from their past 12-month returns to sort them into deciles at the end of each quarter. Rank 10 indicates the highest score and Rank 1 for the lowest among the deciles. Value weights of each stock of a fund in the subsequent quarter is based on the value weights of each stock of a fund at the end of current quarter. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stock characteristics are from CRSP U.S. Stock Database. The sample period is from April 1980 to December 2013. Alpha is the intercept of the five-factor model. Mktrf_beta is the factor loading on the market risk premium. SMB_beta is the factor loading on the size factor. HML_beta is the factor loading on the value factor. RMW_beta is the factor loading on the profitability factor. CMA_beta is the factor loading on the investment factor. The Newey–West corrected t-statistics (Bartlett kernel with a lag length of 3) are reported below the coefficient estimates. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels (for the alpha and the differences for alpha and other loading coefficients).

quarter. It is worth noting that all t-stats reported for the coefficients of alphas and betas are robust for standard errors from the analysis.

3.5. Common risk factors analysis (Fama and French (2015) fivefactor) for monthly fund returns Applying the same common risk factors, we next to examine the gross fund returns and net fund returns to see whether funds returns can be explained by the common risk factors prior and post to the fees. In Table 5, we report Fama and French (2015) fivefactor model through examine the subsequent monthly fund returns. Our results show the higher ranked managers would be able to generate positive abnormal returns (alphas), at about 20–34 basis points in the subsequent month on the equallyweighted gross fund returns. The value-weighted gross fund returns of the higher ranked managers generate positive abnormal returns (alphas), at about 14–31 basis points in the subsequent month. On the other hand, the lower ranked managers would earn negative returns (alphas), at 7–18 basis points, in the subsequent month on the equally-weighted fund of the same decile. An analysis of the net returns generating the same patterns while higher ranked managers generate 11–24 basis points in the subsequent month on the equally-weighted net fund returns. The results for the value-weighted net returns indicate 7–23 basis points for the higher ranked managers. The test for the differences from the highest ranked and lowest ranked indicates the abnormal returns for five-factor alphas can be as much as 52 basis points of the equally-weighted results and 49 basis points for the valueweighted results on the gross returns and net returns. From the common risk factor analysis, higher ranked managers have the lowest factor-loading on market risk premium, the highest factor-loading on SMB, the most negative factor-loading on HML, the most negative factor-loading on RMW, and the highest factor-loading on CMA, consistent with the holding level analysis. Analysis from the five-factor loading at the gross and net fund returns concludes managers of highest ranked could earn additional risk premiums on loading small, growth, stocks with higher information asymmetry but be able to predict their future perfor-

mance from the existing lower profitability and conservative on the capital expenditure investment. Before account for the fund expenses, higher ranked managers outperform lower ranked managers both economically and statistically significant. However, after account for the higher expenses, higher ranked managers are not necessarily earning additional returns due to their higher transaction and operating expenses. The differences on performance of the higher ranked and lower ranked managers are consistently statistically and economically significant. For robustness, we also perform HXZ q-factor (four-factor) alphas analysis for the gross and net fund returns and report our results in the Appendix Table 1. It is worth noting that all t-stats reported for the coefficients of alphas and betas are robust for standard errors from the analysis. 3.6. Are funds with highest MPPM more informed? Next we perform multivariate regression analysis to control for holding characteristics and to see whether the prior univariate tests of the higher ranked managers indeed show that they are better informed. We perform the Fama and MacBeth (1973) regression to regress the subsequent quarterly stock returns on total equity mutual fund ownership, stocks owned by the highest MPPM ranked funds, and stocks owned by the lowest MPPM ranked funds for the current quarter, after controlling for other stock characteristics. The results in Model 1 of Table 6 show that aggregate holdings by all sample funds negatively predicts future holding returns. For common stock characteristics, stock size, book-to-market ratio, past year performance, and gross profits to total asset do have some explanation power on future stock returns. As expected, higher book-to-market ratio stocks tend to have higher future returns. Stocks that have better past year performance tend to have higher future returns. Stocks with higher gross profits to total asset ratio tend to earn higher returns in the future. However, larger size stocks tend to have lower future return but not statistically significant. Models 2 and 3 indicate lowest MPPM ranked funds and highest MPPM ranked funds have prediction power to the future stock returns. Highest MPPM ranked funds positively predict the stock returns of the subsequent quarter, and the lowest MPPM

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

9

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx Table 5 Monthly five-factor model analysis of fund returns of sample funds based on MPPM rankings. Panel A: Equally-weighted five-factor model of MPPM ranked funds (dependent variable: gross fund returns) 5-Factor model MPPM rank

Alpha t-stat Mktrf_beta t-stat SMB_beta t-stat HML_beta t-stat RMW_beta t-stat CMA_beta t-stat

Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

Diff (Rank 10–Rank 1)

0.0018 1.54 1.0288 27.73 0.1823 2.20 0.1526 1.59 0.0735 0.51 0.2324 1.53

0.0007 0.94 0.9805 45.03 0.1498 2.93 0.1048 1.76 0.0160 0.19 0.1081 1.35

0.0010⁄ 1.77 0.9703 56.90 0.1686 4.23 0.0814 1.85 0.0607 1.02 0.0694 1.25

0.0008⁄ 1.74 0.9852 87.40 0.1575 5.90 0.0342 1.11 0.0712 1.77 0.0065 0.16

0.0004 1.03 0.9782 78.74 0.1359 5.24 0.0325 1.11 0.0738 2.06 0.0304 0.97

0.0002 0.45 0.9744 95.25 0.1592 10.00 0.0003 0.01 0.0690 3.25 0.0216 0.72

0.0005 1.11 0.9784 81.13 0.1892 10.01 0.0381 1.53 0.0427 1.45 0.0769 2.15

0.0011⁄⁄ 2.19 0.9616 77.74 0.2206 8.75 0.0529 1.73 0.0146 0.37 0.0686 1.37

0.0020⁄⁄⁄ 2.85 0.9434 50.10 0.3074 7.17 0.1360 2.63 0.0613 0.85 0.1327 1.65

0.0034⁄⁄⁄ 2.84 0.9116 30.66 0.3769 5.85 0.2121 2.61 0.1191 1.09 0.0970 0.78

0.0052⁄⁄ 2.36 0.1172⁄ 1.88 0.1946 1.37 0.3646⁄⁄ 2.13 0.0456 0.18 0.3294 1.24

Panel B: Equally-weighted five-factor model of MPPM ranked funds (Dependent variable: net fund returns) 5-Factor model

Alpha t-stat Mktrf_beta t-stat SMB_beta t-stat HML_beta tstat RMW_beta t-stat CMA_beta t-stat

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

Diff (Rank 10–Rank 1)

0.0028⁄⁄ 2.42 1.0289 26.84 0.1821 2.15 0.1526 1.52 0.0738 0.49 0.2327 1.49

0.0017⁄⁄ 2.26 0.9807 44.53 0.1496 2.84 0.1050 1.70 0.0158 0.18 0.1083 1.31

0.0020⁄⁄⁄ 3.33 0.9705 56.92 0.1686 4.07 0.0814 1.80 0.0610 0.97 0.0699 1.24

0.0017⁄⁄⁄ 3.86 0.9853 89.08 0.1577 5.63 0.0344 1.09 0.0713 1.68 0.0066 0.16

0.0013⁄⁄⁄ 3.61 0.9782 78.56 0.1361 5.05 0.0326 1.08 0.0739 1.98 0.0307 -0.96

0.0008⁄⁄ 2.00 0.9746 96.28 0.1591 9.96 0.0006 0.03 0.0690 3.27 0.0219 0.72

0.0005 1.18 0.9785 82.16 0.1890 10.08 0.0380 1.51 0.0426 1.49 0.0765 2.12

0.0002 0.35 0.9617 79.28 0.2202 8.94 0.0530 1.70 0.0144 0.37 0.0681 1.34

0.0011 1.46 0.9435 50.49 0.3071 7.17 0.1361 2.52 0.0616 0.84 0.1325 1.62

0.0024⁄ 1.91 0.9117 30.59 0.3770 5.83 0.2120 2.48 0.1191 1.06 0.0965 0.76

0.0052⁄⁄ 2.38 0.1172⁄ 1.88 0.1949 1.38 0.3645⁄⁄ 2.13 0.0453 0.18 0.3291 1.24

Panel C: Value-weighted five-factor model of MPPM ranked funds (Dependent variable: Gross fund returns) 5-Factor model

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Alpha 0.0018 0.0021⁄⁄ 0.0011 0.0007 0.0003 t-stat 1.32 2.27 1.45 1.30 0.84 Mktrf_beta 0.9871 0.9858 0.9749 0.9494 0.9787 t-stat 20.83 41.14 47.76 79.10 86.89 SMB_beta 0.0987 0.0430 0.0164 0.0314 0.0168 t-stat 1.21 0.76 0.42 1.49 0.66 HML_beta 0.1947 0.1482 0.0931 0.0188 0.0514 t-stat 2.16 2.16 1.87 0.70 1.86 RMW_beta 0.0728 0.0703 0.0331 0.0023 0.0373 t-stat 0.50 0.75 0.50 0.07 1.15 CMA_beta 0.2866 0.1201 0.0518 0.0166 0.0391 t-stat 1.99 1.53 0.85 0.43 1.47 Panel D: valued-weighted five-factor model of MPPM ranked funds (Dependent variable: 5-Factor model

Alpha t-stat Mktrf_beta t-stat SMB_beta t-stat HML_beta t-stat RMW_beta t-stat

Rank 6

Rank 7

0.0002 0.0002 0.43 0.56 0.9642 0.9700 89.89 74.40 0.0459 0.0845 3.16 3.94 0.0248 0.0079 1.45 0.32 0.0176 0.0670 0.82 2.17 0.0010 0.0742 0.03 2.09 Net fund returns)

Rank 8

Rank 9

Rank 10 (highest)

Diff (Rank 10–Rank 1)

0.0007 1.35 0.9610 72.62 0.1176 4.57 0.0354 1.17 0.0275 0.65 0.0805 1.63

0.0014⁄⁄ 2.08 0.9263 57.09 0.1724 4.34 0.0645 1.34 0.0873 1.27 0.0777 1.10

0.0031⁄⁄⁄ 2.78 0.9047 31.21 0.2374 3.64 0.2069 2.43 0.1731 1.46 0.1039 0.82

0.0049⁄⁄ 2.24 0.0824 1.22 0.1387 1.00 0.4016⁄⁄ 2.50 0.1003 0.40 0.3906 1.51

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

Diff (Rank 10–Rank 1)

0.0026⁄ 1.93 0.9870 20.79 0.0981 1.20 0.1949 2.16 0.0731 0.50

0.0028⁄⁄⁄ 3.12 0.9860 41.18 0.0423 0.74 0.1489 2.17 0.0695 0.74

0.0018⁄⁄⁄ 2.38 0.9749 47.74 0.0160 0.41 0.0938 1.89 0.0325 0.49

0.0013⁄⁄⁄ 2.60 0.9494 78.87 0.0309 1.46 0.0194 0.72 0.0029 0.09

0.0009⁄⁄ 2.81 0.9788 87.04 0.0167 0.66 0.0516 1.87 0.0375 1.16

0.0005 1.23 0.9644 89.83 0.0454 3.10 0.0250 1.46 0.0182 0.84

0.0009⁄⁄ 2.23 0.9701 74.20 0.0842 3.92 0.0079 0.32 0.0670 2.16

0.0000 0.01 0.9611 72.82 0.1169 4.54 0.0351 1.16 0.0283 0.67

0.0007 1.04 0.9266 57.17 0.1716 4.33 0.0645 1.34 0.0880 1.28

0.0023⁄⁄ 2.06 0.9048 31.27 0.2373 3.64 0.2070 2.43 0.1730 1.46

0.0049⁄⁄ 2.25 0.0822 1.22 0.1392 1.00 0.4019⁄⁄ 2.50 0.0999 0.39 (continued on next page)

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

10

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx

Table 5 (continued) Panel D: valued-weighted five-factor model of MPPM ranked funds (Dependent variable: Net fund returns) 5-Factor model

CMA_beta t-stat

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

Diff (Rank 10–Rank 1)

0.2882 2.00

0.1217 1.55

0.0534 0.88

0.0152 0.39

0.0403 1.51

0.0014 0.04

0.0737 2.08

0.0799 1.62

0.0772 1.09

0.1033 0.81

0.3915 1.51

Panel A of this table reports subsequent monthly equally-weighted alphas and betas from the gross fund returns of sample funds by using Fama and French (2015) five-factor model. We rank each fund by its MPPM from its past 12 months returns to sort them into deciles at the end of each quarter. Rank 10 indicates for best decile and Rank 1 for the bottom decile. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stock characteristics are from CRSP U.S. Stock Database. The sample period is from April 1980 to December 2013. Alpha is the intercept of the five-factor model. Mktrf_beta is the factor loading on the market risk premium. SMB_beta is the factor loading on the size factor. HML_beta is the factor loading on the value factor. RMW_beta is the factor loading on the profitability factor. CMA_beta is the factor loading on the investment factor. The Newey–West corrected t-statistics (Bartlett kernel with a lag length of 3) are reported below the coefficient estimates. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels (for the alpha and the differences for alpha and other loading coefficients). Panel B of this table reports subsequent monthly equally-weighted alphas and betas from the net fund returns of sample funds by using Fama and French (2015) five-factor model. We rank each fund by its MPPM from its past 12 months returns to sort them into deciles at the end of each quarter. Rank 10 indicates for best decile and Rank 1 for the bottom decile. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stock characteristics are from CRSP U.S. Stock Database. The sample period is from April 1980 to December 2013. Alpha is the intercept of the five-factor model. Mktrf_beta is the factor loading on the market risk premium. SMB_beta is the factor loading on the size factor. HML_beta is the factor loading on the value factor. RMW_beta is the factor loading on the profitability factor. CMA_beta is the factor loading on the investment factor. The Newey–West corrected t-statistics (Bartlett kernel with a lag length of 3) are reported below the coefficient estimates. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels (for the alpha and the differences for alpha and other loading coefficients). Panel C of this table reports subsequent monthly valued-weighted alphas and betas from the gross fund returns of sample funds by using Fama and French (2015) five-factor model. We rank each fund by its MPPM from its past 12 months returns to sort them into deciles at the end of each quarter. Rank 10 indicates for best decile and Rank 1 for the bottom decile. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stock characteristics are from CRSP U.S. Stock Database. The sample period is from April 1980 to December 2013. Alpha is the intercept of the five-factor model. Mktrf_beta is the factor loading on the market risk premium. SMB_beta is the factor loading on the size factor. HML_beta is the factor loading on the value factor. RMW_beta is the factor loading on the profitability factor. CMA_beta is the factor loading on the investment factor. The Newey–West corrected t-statistics (Bartlett kernel with a lag length of 3) are reported below the coefficient estimates. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels (for the alpha and the differences for alpha and other loading coefficients). Panel D of this table reports subsequent monthly valued-weighted alphas and betas from the fund returns of sample funds by using Fama and French (2015) five-factor model. We rank each fund by its MPPM from its past 12 months returns to sort them into deciles at the end of each quarter. Rank 10 indicates for best decile and Rank 1 for the bottom decile. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stock characteristics are from CRSP U.S. Stock Database. The sample period is from April 1980 to December 2013. Alpha is the intercept of the fivefactor model. Mktrf_beta is the factor loading on the market risk premium. SMB_beta is the factor loading on the size factor. HML_beta is the factor loading on the value factor. RMW_beta is the factor loading on the profitability factor. CMA_beta is the factor loading on the investment factor. The Newey–West corrected t-statistics (Bartlett kernel with a lag length of 3) are reported below the coefficient estimates. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels (for the alpha and the differences for alpha and other loading coefficients).

ranked funds negatively predict the stock returns of the subsequent quarter while the total fund holding is no longer significant negative after controlling for the impact of lowest ranked MPPM group. This findings confirm that managers with the highest ranked have better stock picking skills that positively predict future stock returns. 3.7. Quarterly buying and selling portfolio returns of MPPM ranked funds Our results also have implication to the general investors. Investors can learn from the higher ranked managers from their quarterly reported holdings. In this analysis, we construct a copycat portfolio that tracks manager’s buying and selling activities from their two consecutive quarterly holdings. Our results in Table 7 indicate in general stocks bought by managers will outgain stocks sold by the managers in the subsequent quarter. However, Stocks bought by highest ranked managers significantly outperform stocks sold by those managers at 0.67% in the following quarter. This finding shows that highest ranked managers are more informed. If investors take long position on the stocks purchased by the highest ranked managers and short on the stocks sold by those managers could earn trading profits prior to their transaction costs. We use Fama and French (2015) five-factor models to reexamine the copycat portfolios and report the results in Table 8. If investors follow the highest ranked managers to buy stocks based on their quarterly disclosed stocks that newly added and sell stocks that managers newly sold, could they earn risk-adjusted returns for the subsequent quarter? Our result shows that the

copycat portfolios would earn additional 31-basis-point per quarter if investors construct the portfolios exactly the ways of the highest ranked managers and hold onto the next quarter. For robustness, we perform the same tests through adopting the HXZ q-factor (four-factor) to re-examine the copycat portfolios. The results of the 32-basis-point per quarter from the copycat portfolio (reported in Appendix Table 2) are consistent with the five-factor analysis, suggesting the stock picking skills are statistically and economically significant and are not sensitive with respect to the different asset pricing models we select. All t-stats reported for the coefficients of alphas and betas are robust for standard errors from the analysis. 3.8. The quarterly holding returns based on the portfolios constructed from the previous quarters For the robustness for the managerial stock picking skills, we extend the analysis to analyze the holdings that managers construct from the previous two quarters. We report only the equity holdings and to calculate the differences from the stock holding returns between the previous quarterly holdings and holdings from two quarters prior to the rank, the calculation is as follows.

Difft ¼ Rett ðHolding t1 Þ  Ret t ðHolding t2 Þ Our results in Table 9 indicate the quarterly stock returns in time period t  1 outperforms the returns in two quarters prior for the sample funds from the highest ranked managers. The value-weighted returns are 5.90%, significantly greater than the returns of two quarters prior (3.8%). The equally-weighted returns are 4.9% on one quarter prior, significantly greater than the returns

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

11

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx Table 6 MPPM ranked fund ownership and subsequent stock returns.

Intercept t-stat Total fund holding t-stat Stocks held by highest MPPM ranked funds t-stat Stocks held by lowest MPPM ranked funds t-stat Ln (Size) t-stat Ln (B/M) t-stat Past year performance t-stat Ln (Gross Profits to Total Asset) t-stat Ln (Monthly Turnover) t-stat Ln (Age) t-stat Ln (Dividend Yield + 1) t-stat Ln (Monthly Volatility) t-stat Adjusted R-squared Number of observations

Model 1

Model 2

Model 3

⁄⁄⁄

⁄⁄⁄

0.102⁄⁄⁄ 3.57 0.026 (1.20)

0.102 3.54 0.044⁄⁄ (2.19)

0.101 3.51 0.055⁄⁄ (2.57) 0.101⁄⁄ 2.12

0.001 (0.75) 0.019⁄⁄⁄ 7.60 0.014⁄⁄ 2.24 0.017⁄⁄⁄ 13.91 0.000 (0.06) 0.001 (0.97) 0.042 0.45 0.001 (0.15) 0.074

0.113⁄⁄ (2.30) 0.001 (0.79) 0.019⁄⁄⁄ 7.69 0.014⁄⁄ 2.23 0.017⁄⁄⁄ 14.00 0.000 (0.08) 0.001 (0.93) 0.042 0.47 0.001 (0.14) 0.074

0.001 (0.73) 0.019⁄⁄⁄ 7.68 0.014⁄⁄ 2.23 0.017⁄⁄⁄ 14.02 0.000 (0.09) 0.001 (0.83) 0.04 0.44 0.001 (0.16) 0.074 401,363

This table reports Fama and Macbeth (1973) regression results for subsequent quarterly stock returns on being held among highest and lowest MPPM (Goetzmann et al. (2007)) ranked funds. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stock characteristics are from CRSP U.S. Stock Database. The sample period is from April 1980 to September 2013. Total fund holding of a stock is shares held by the domestic equity mutual funds in our sample divided by shares outstanding of the stock. Stocks held by highest (lowest) MPPM ranked funds is shares held by the funds in the highest (lowest) MPPM ranked group divided by shares outstanding of the stock. Ln (Size) is the logarithm of a stock’s market capitalization in millions. Ln (B/M) is the logarithm of a stock’s book to market ratio. Past year performance is the cumulative stock returns of a stock from past 12 months. Ln (Gross profits to total asset) is the logarithm of a stock’s gross profits (revenue minus costs of goods sold) divided by total assets. Ln (Monthly turnover) is the logarithm of a stock’s recent three months’ monthly turnover. Ln (Age) is the logarithm of a stock’s number of months since recorded in the CRSP Database. Ln (Dividend yield + 1) is the logarithm of one plus a stock’s past 12-month cash dividends divided by quarter end stock price. Ln (Monthly volatility) is the logarithm of the monthly standard deviation of recent 24 months of stock returns. Newey and West (1987) corrected t-statistics are reported under each coefficient. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels.

Table 7 Subsequent quarterly returns from copycat portfolios following buying and selling among different MPPM ranked managers. Returns from copycat portfolio

Subsequent portfolio Subsequent portfolio Subsequent portfolio

quarterly returns of copycat from buying quarterly returns of copycat from selling quarterly returns of copycat (buying–selling)

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

2.75%

2.83%

3.26%

3.24%

3.49%

3.50%

3.46%

3.90%

4.09%

4.37%

2.65%

2.98%

3.12%

3.17%

3.26%

3.25%

3.38%

3.73%

3.57%

3.70%

0.11% (0.30)

0.14% (0.56)

0.14% (0.52)

0.06% (0.25)

0.23% (0.80)

0.26% (1.03)

0.07% (0.30)

0.18% (0.57)

0.53%⁄⁄ (2.14)

0.67%⁄⁄ (1.98)

This table reports the value weights subsequent quarterly performance from the copycat portfolios who buy stocks after the stocks purchased by the managers and sell stocks after the stocks sold by the portfolio managers of sample funds. We rank all domestic equity mutual funds by manipulation-proof performance measure (MPPM (Goetzmann et al. (2007)) from their past 12-month returns to sort them into deciles at the end of each quarter. Rank 10 indicates the highest score and Rank 1 for the lowest among the deciles. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stocks returns are from CRSP U.S. Stock Database. The sample period is from April 1980 to September 2013. A fund’s buying portfolio is categorized in the top decile group of stocks that gained the highest market value among the holdings in the current quarter. A fund’s selling portfolio is categorized in the top decile group of stocks that lost the highest market value among the holding in the current quarter. T-statistics are reported among different MPPM group of the copycat portfolio combining Buying minus Selling of the holdings. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels.

formed at two quarters prior (3.9%) to the rank. The quarterly analysis shows that higher ranked managers do have stock picking skills and are better informed. 3.9. Are high MPPM funds generating higher return gap? Next, we want to test whether the MPPM can predict the return gap where Kacperczyk et al. (2008) conclude the return gap can be

proxy for the unobserved actions. We design the empirical test to see if the MPPM can predict the return gap. We also want to control other fund characteristic variables that the existing literature adopt that has shown to affect fund returns. We first run a correlation of the return gap to the major fund characteristics. We find an insignificant negative relation between expenses and the return gap, which indicates that funds do not compensate investors for their expenses. The proxy for trading activities is to look at the

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

12

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx

Table 8 Subsequent quarterly alphas and betas from copycat portfolios that added or sold by the current quarter of different MPPM ranked managers. Diff (added–sold)

Alpha t-stat Mktrf_beta t-stat SMB_beta t-stat HML_beta t-stat RMW_beta t-stat CMA_beta t-stat

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

0.0009 0.69 0.0235 0.63 0.0401 0.71 0.1830⁄⁄ 2.12 0.1266⁄ 1.76 0.1808⁄⁄ 2.03

0.0010 0.97 0.0846⁄⁄⁄ 3.25 0.0589 1.30 0.0097 0.18 0.0243 0.48 0.0053 0.09

0.0011 1.19 0.0213 0.85 0.0874⁄ 1.86 0.0921⁄ 1.69 0.0462 0.80 0.1679⁄⁄⁄ 2.70

0.0001 0.09 0.0473⁄ 1.78 0.0736⁄ 1.93 0.0248 0.42 0.0721 1.01 0.0638 0.74

0.0005 0.48 0.0591⁄⁄ 2.18 0.0544 1.29 0.0489 0.81 0.0417 0.73 0.0985 1.26

0.0004 0.33 0.0416 1.57 0.0990⁄ 1.77 0.0494 0.72 0.0732 0.84 0.0720 0.72

0.0001 0.09 0.0583⁄⁄ 2.19 0.0515 1.16 0.0363 0.75 0.0044 0.07 0.0334 0.35

0.0007 0.63 0.0334 1.27 0.0302 0.64 0.0808 1.31 0.0495 0.60 0.0056 0.05

0.0018⁄ 1.87 0.0291 0.97 0.1003⁄⁄⁄ 2.56 0.0204 0.32 0.0312 0.47 0.0822 0.81

0.0031⁄⁄⁄ 2.73 0.0262 1.09 0.1051⁄⁄ 2.24 0.0871⁄ 1.71 0.0722 1.13 0.0655 0.82

This table reports the subsequent quarterly five-factor (Fama and French (2015)) alphas and betas by taking the differences of using the portfolios added to the holdings to subtract the portfolios sold by the managers of the sample funds. We rank all domestic equity mutual funds by manipulation-proof performance measure (MPPM (Goetzmann et al. (2007)) from their past 12-month returns to sort them into decile at the end of each quarter. Rank 10 indicates the highest score and Rank 1 for the lowest among the deciles. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stocks returns are from CRSP U.S. Stock Database. The sample period is from April 1980 to September 2013. A fund’s buying portfolio is categorized in the top decile group of stocks that gained the highest market value among the holdings in the current quarter. A fund’s selling portfolio is categorized in the top decile group of stocks that lost the highest market value among the holding in the current quarter. The Newey–West corrected t-statistics (Bartlett kernel with a lag length of 3) are reported below the coefficient estimates. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels.

Table 9 Quarterly holding returns (time t) among different MPPM ranked managers (who form the portfolios at previous quarter (t  1) and two quarters prior (t  2)). Holding returns

EW returns on time t based on time t  1 holding EW returns on time t based on t  2 holding Diff EW holding returns (t  1 minus t  2) VW returns on time t based on time t  1 holding VW returns on time t based on time t  2 holding Diff VW holding returns (t  1 minus t  2)

MPPM rank Rank 1 (lowest MPPM)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest MPPM)

2.653 3.006 0.375 2.943 2.881 0.060

3.072 3.179 0.126 2.494 2.881 0.407

2.883 2.987 0.126 1.980 2.960 0.975

3.302 3.219 0.068 3.123 3.037 0.096

3.259 3.310 0.058 3.346 3.216 0.145

3.921 3.404 0.496 4.301 3.319 0.958

4.189 3.544 0.625 3.951 3.429 0.543

4.625 3.592 1.020 4.731 3.539 1.192

5.226 3.666 1.539⁄⁄⁄ 4.984 3.519 1.457 ⁄

4.931 3.909 0.997⁄⁄⁄ 5.895 3.771 2.114⁄⁄⁄

⁄

This table reports the equally-weighted (EW) and value-weighted (VW) subsequent quarterly equity holding returns of different MPPM ranked fund managers who form the portfolios at Time t  1 (the previous quarter) and Time t  2 (two quarters prior). We rank all domestic equity mutual funds by manipulation-proof performance measure (MPPM (Goetzmann et al. (2007)) from their past 12-month returns to sort them into deciles at the end of each quarter. Rank 10 indicates the highest score and Rank 1 for the lowest among the deciles. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stocks returns are from CRSP U.S. Stock Database. The sample period is from April 1980 to September 2013. For the economy of the presentation, the coefficients of T-statistics are not reported. The significance level among different MPPM group for the differences on Time t  1 (the previous quarter) and Time t  2 (two quarters prior) of the holding returns (equally- and value-weighted) are marked. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels.

relation between turnover and the return gap. We do not find a significant relation between turnover and the return gap. Kacperczyk et al. (2008) document that return gap decreases with fund size. Consistent with their findings, we find that smaller funds tend to exhibit more favorable return gaps. Furthermore, we find that a fund’s age is negatively related to its return gap. Our results from the Fama and Macbeth (1973) regression in predicting the return gap is reported in Table 10. In Model 1, our results are consistent with the ‘‘smart money’’ effect in Zheng (1999) and Kacperczyk et al. (2008). We find a significant and positive relation between the mean lagged money flow and the return gap. In Model 2, we find that MPPM has strong predictive power on return gap, which indicates that highest MPPM manager has skills in engaging in unobserved actions through stock picking to provide additional value. Finally, in Model 3, we combine the holding proportion with MPPM, the regression results indicate that funds with higher proportions in cash and less proportions in equity and higher MPPM score predict higher return gaps after controlling for the other variables. Since skillful managers would hold less liquid stocks, the higher proportions in cash in comparison to the lower ranked managers is probably to satisfy the investors’ redemption.

3.10. The persistence of fund returns by MPPM Lastly, we want to test whether the managerial skills are persistent. We sort all funds in our sample into deciles based on their MPPM scores during the previous 12 months and compute the average fund returns during the subsequent quarter by weighting all funds in each decile. Table 11 reports accumulative gross fund returns for the first, first to second, first to third, and first to fourth quarters. We find managers of the highest ranked funds persistently outperform managers of the lowest ranked funds up to 3 quarters at the equally-weight (each fund) and 2 quarters for our value-weighted (each fund) in that decile. The results are consistent with Kacperczyk et al. (2014), who document that the persistence of skill index is up to 6 months.

4. Conclusion With the controversial findings on the existence of managerial skills in the asset management industry and the battle between active management versus passive (index) strategies, we propose

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

13

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx Table 10 Predictive power of MPPM on return gap. Model 1 Intercept t-stat MPPM t-stat Correlation between holding and fund ret t-stat Fund expense t-stat Turnover t-stat Ln (fund Size) t-stat Ln (age) t-stat Stock proportion t-stat Bond proportion t-stat Cash proportion t-stat Quarterly flow t-stat Quarterly flow squared t-stat Adjusted R-squared Number of observations

0.536 3.13

Model 2

⁄⁄⁄

⁄⁄

6.628⁄ 1.79 5.173 (0.50) 0.008 0.27 0.059⁄⁄ (2.26) 0.057⁄ (1.94)

0.375⁄ 1.69 0.116 (0.43) 0.088 94,215

Model 3

0.468 2.28 1.661⁄⁄⁄ 3.22 6.968⁄⁄ 2.34 3.586 (0.35) 0.014 0.50 0.058⁄⁄ (2.30) 0.057⁄ (1.85)

1.590⁄⁄⁄ 2.77 1.895⁄⁄⁄ 4.09 9.486⁄⁄⁄ 2.98 9.084 (1.58) 0.004 0.15 0.027⁄⁄ (2.53) 0.040⁄ (1.87) 0.015⁄⁄ (2.40) 0.005 (0.65) 0.000⁄⁄⁄ (5.99)

0.255

0.073

This table reports Fama and Macbeth (1973) regression results by regressing quarterly return gap on MPPM (Goetzmann et al. (2007)) and various characteristics. Return gaps are calculated based on Kacperczyk et al. (2008) by using the difference between fund returns and holding returns. MPPM (in percentage) is calculated from the past 12-month raw fund returns. The correlation variable we consider measures the transparency of a fund’s investment strategy and is defined as the correlation coefficient between monthly holdings and investor returns during the previous quarter. Turnover is the fund turnover ratio recorded in CRSP Database. Fund expenses (in percentage) is the annual expense ratio. Ln (Fund Size) is the logarithm of Fund TNA in millions. Ln (Age) is the logarithm of a fund age recorded in the CRSP Database. Common stock proportion (in percentage) is the holding that is composed in common stocks. Bond proportion (in percentage) is the holding that is composed in all kinds of bond. Cash proportion (in percentage) is the holding that is composed in cash. Fund flow is calculated by fund returns and TNA that recorded in the CRSP Database. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10-, 5- and 1-percent significance levels.

Table 11 Persistence of performance of best and worst sample funds based on MPPM rankings. MPPM rank based on current quarter

Equally-weighted fund returns (MPPM ranked 1 funds) Equally-weighted fund returns (MPPM ranked 10 funds) Equally-weighted fund returns Diff (Rank 10–Rank 1) Value-weighted fund returns (MPPM ranked 1 funds) Value-weighted Fund Returns (MPPM ranked 10 funds) Value-weighted Fund Returns Diff (Rank 10–Rank 1)

Subsequent quarters t+1 (first quarter)

t + 1 to t + 2 (first quarter to second quarter)

t + 1 to t + 3 (first quarter to third quarter)

t + 1 to t + 4 (first quarter to fourth quarter)

0.0277 3.11 0.0410 4.88 0.0133 2.25⁄⁄ 0.0264 3.08 0.0389 4.68 0.0125 2.15⁄⁄

0.0601 4.76 0.0775 6.05 0.0174 2.04⁄⁄ 0.0580 4.75 0.0725 5.73 0.0145 1.74⁄

0.0941 5.94 0.1117 7.22 0.0176 1.77⁄ 0.0891 5.89 0.1051 6.79 0.0160 1.57

0.1305 7.34 0.1427 7.89 0.0123 1.08 0.1222 7.32 0.1341 7.38 0.0118 1.00

This table reports equally-weighted and value-weighted accumulative gross returns for the subsequent first quarter, first to second, first to third, and first to fourth quarter for all domestic equity mutual funds ranked by manipulation-proof performance measure (MPPM (Goetzmann et al. (2007)) from their past 12-month returns. We sort fund returns into deciles at the end of each quarter. Rank 10 indicates the highest score and Rank 1 for the lowest among the deciles. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stocks returns are from CRSP U.S. Stock Database. The sample period is from April 1980 to September 2013. T-statistics are reported in the parentheses for the last column of the differences (Diff (Rank 10–Rank 1)) of the highest rank (rank 10) and the lowest rank (rank 1). The significance levels are denoted by ⁄ ⁄⁄ ⁄⁄⁄ , , and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels.

to use a better and efficient performance measure that can’t be manipulated to re-examine the managerial skills of the active domestic equity fund managers. By analyzing fund returns, holding returns, and the return gap, we find a subgroup of fund managers consistently generate higher gross, fee-adjusted fund returns, and demonstrate better stock picking skills. Analyzing the holding of

the higher ranked managers indicates those managers hold stocks with higher information asymmetry and can positively predict the subsequent quarterly holding returns. They earn risk-adjusted returns (positive alphas) on both their fund and holding returns when they are compared to their lower ranked counterparts. They also generate positive return gap which adds value to their

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

14

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx

fundholders on pre-expense basis. Those managers are able to persistently outperform their lower ranked counterparts up to six months. The findings are useful for the regulators and investors. With regulators focus on higher frequency of disclosure, investors can avoid selecting fund managers who achieve superior performance through engaging derivative contracts and using leverage. Even though it is encouraging to know a subgroup of fund managers are better informed, it is also important to know those managers charge higher expenses and trade more often so the returns are not necessarily large enough to offset their expenses to the passive (index) strategies. This paper also calls for new compensation mechanisms to reward managers who have skillsets. The current managerial compensation on mutual funds reward primarily on fund size thus induces managers to manipulate their fund returns to attract additional fund flows. Through our analysis on the MPPM, one can distinguish skillful managers from their manipulated counterparts

thus would mitigate the agency conflicts in the asset management industry. Acknowledgements The authors would like to thank Carol Alexander (the managing editor), two anonymous referees, and the seminar participants at 2014 Financial Management Association conference. The data was extracted while Fan Chen at Quinnipiac University. An earlier draft of the paper was circulated under the title ‘‘On market timing, stock picking, and managerial skills of mutual fund managers with manipulation-proof performance measure”. Inaccuracies that may have been introduced in subsequent revisions are the sole responsibility of the authors. Appendix A See appendix Tables A.1 and A.2.

Table A.1 Monthly HXZ q-factor (four-factor) model analysis of fund returns of sample funds based on MPPM rankings. Panel A: Equally-weighted HXZ q-factor (four-factor) model of MPPM ranked funds (Dependent variable: Gross fund returns) 4-Factor model MPPM rank

Alpha t-stat Mktrf_beta t-stat ME_beta t-stat I/A_beta t-stat ROA_beta t-stat

Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

0.0007 0.62 1.0206 29.15 0.1024 1.02 0.0647 0.48 0.1790 1.93

0.0002 0.29 0.9744 46.23 0.1019 1.53 0.0092 0.12 0.0445 0.80

0.0008 1.26 0.9654 55.84 0.1263 2.33 0.0238 0.42 0.0107 0.26

0.0007 1.46 0.9822 94.59 0.1298 3.66 0.0302 0.80 0.0456 1.61

0.0004 1.22 0.9810 78.06 0.1087 2.65 0.0235 0.74 0.0619 2.71

0.0001 0.19 0.9785 114.71 0.1491 7.54 0.0350 1.41 0.0807 4.59

0.0000 0.13 0.9855 95.18 0.1934 12.11 0.0446 1.61 0.0856 4.24

0.0007 1.30 0.9699 99.41 0.2350 10.34 0.0090 0.20 0.0800 3.41

0.0013 1.60 0.9579 50.64 0.3547 6.00 0.0354 0.49 0.0743 1.64

0.0023 1.63 0.9386 29.22 0.4491 4.84 0.1685 1.46 0.0994 1.36

Panel B: value-weighted four-factor model of MPPM ranked funds (Dependent variable: Gross fund returns) 4-Factor model

Alpha t-stat Mktrf_beta t-stat ME_beta t-stat I/A_beta t-stat ROA_beta t-stat

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

0.0008 0.62 0.9782 21.94 0.0273 0.28 0.0822 0.57 0.1578 1.73

0.0014 1.46 0.9762 40.07 0.0252 0.31 0.0547 0.61 0.0353 0.65

0.0008 0.93 0.9707 48.66 0.0234 0.44 0.0618 0.93 0.0145 0.35

0.0005 0.89 0.9465 84.51 0.0245 1.30 0.0308 0.87 0.0052 0.25

0.0003 0.96 0.9824 82.35 0.0047 0.13 0.0321 1.13 0.0424 1.98

0.0001 0.17 0.9670 101.68 0.0501 4.18 0.0258 1.15 0.0188 1.33

0.0004 1.15 0.9715 73.07 0.0776 3.60 0.0967 3.02 0.0738 3.63

0.0003 0.65 0.9669 79.90 0.1443 4.96 0.0309 0.87 0.0475 1.92

0.0008 1.07 0.9396 56.35 0.2158 4.21 0.0112 0.16 0.0482 1.11

0.0019 1.44 0.9371 28.16 0.3212 3.36 0.1373 1.17 0.0697 0.82

Panel C: Equally-weighted four-factor model of MPPM ranked funds (Dependent variable: net fund returns) 4-Factor model

MPPM Rank Rank 1 (lowest)

Rank 2

Alpha 0.0018 0.0012 t-stat 1.53 1.56 Mktrf_beta 1.0208 0.9747 t-stat 29.12 46.23 ME_beta 0.1024 0.1017 t-stat 1.02 1.52 I/A_beta 0.0646 0.0095 t-stat 0.48 0.12 ROA_beta 0.1786 0.0442 t-stat 1.93 0.80 Panel D: value-weighted four-factor model of MPPM 4-Factor model

Alpha t-stat Mktrf_beta

Rank 3 0.0017⁄⁄ 2.81 0.9657 55.78 0.1264 2.32 0.0236 0.41 0.0111 0.27 ranked funds

Rank 4

Rank 5

0.0016⁄⁄⁄ 0.0013⁄⁄⁄ 3.54 3.79 0.9823 0.9811 94.45 78.07 0.1299 0.1088 3.66 2.65 0.0304 0.0236 0.80 0.75 0.0457 0.0624 1.61 2.74 (Dependent variable: Net fund

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

0.0010⁄⁄⁄ 3.07 0.9788 114.77 0.1489 7.53 0.0351 1.41 0.0806 4.58 returns)

0.0009⁄⁄⁄ 2.45 0.9857 95.19 0.1932 12.12 0.0445 1.61 0.0857 4.25

0.0003 0.56 0.9701 99.46 0.2347 10.33 0.0086 0.19 0.0802 3.40

0.0003 0.40 0.9582 50.70 0.3545 5.99 0.0355 0.49 0.0744 1.64

0.0013 0.90 0.9388 29.26 0.4492 4.84 0.1686 1.46 0.0997 1.36

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

0.0017 1.23 0.9783

0.0022⁄⁄ 2.27 0.9765

0.0015⁄ 1.79 0.9708

0.0011⁄⁄⁄ 2.13 0.9465

0.0010⁄⁄⁄ 2.97 0.9826

0.0006 1.63 0.9673

0.0011⁄⁄⁄ 3.07 0.9717

0.0004 0.74 0.9672

0.0000 0.05 0.9400

0.0011 0.82 0.9373

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

15

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx Table A.1 (continued) Panel D: value-weighted four-factor model of MPPM ranked funds (Dependent variable: Net fund returns) 4-Factor model

t-stat ME_beta t-stat I/A_beta t-stat ROA_beta t-stat

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

21.91 0.0270 0.28 0.0830 0.58 0.1572 1.73

40.10 0.0257 0.31 0.0542 0.60 0.0353 0.65

48.61 0.0237 0.45 0.0610 0.92 0.0144 0.35

84.23 0.0240 1.27 0.0301 0.84 0.0054 0.26

82.24 0.0048 0.13 0.0317 1.11 0.0428 2.00

101.27 0.0498 4.14 0.0260 1.15 0.0188 1.33

72.89 0.0774 3.59 0.0961 3.00 0.0742 3.65

79.94 0.1437 4.93 0.0309 0.87 0.0473 1.90

56.39 0.2153 4.20 0.0114 0.16 0.0484 1.11

28.20 0.3213 3.36 0.1375 1.17 0.0703 0.82

Panel A of this table reports subsequent monthly equally-weighted alphas and betas from the fund returns of sample funds by using Hou et al. (2015) four-factor model. We rank each fund by its MPPM from its past 12-month returns to sort them into deciles at the end of each quarter. Rank 10 indicates for best decile and Rank 1 for the bottom decile. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stock characteristics are from CRSP U.S. Stock Database. The sample period is from April 1980 to December 2013. Alpha is the intercept of the four-factor model. Mktrf_beta is the factor loading on the market risk premium. ME_beta is the factor loading on the size factor. I/A_beta is the factor loading on the investment factor. ROA_beta is the factor loading on the profitability factor. The Newey–West corrected t-statistics (Bartlett kernel with a lag length of 3) are reported below the coefficient estimates. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels (for the alpha coefficients). Panel B of this table reports subsequent monthly value-weighted alphas and betas from the fund returns of sample funds by using Hou et al. (2015) four-factor model. We rank each fund by its MPPM from its past 12-month returns to sort them into deciles at the end of each quarter. Rank 10 indicates for best decile and Rank 1 for the bottom decile. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stock characteristics are from CRSP U.S. Stock Database. The sample period is from April 1980 to December 2013. Alpha is the intercept of the four-factor model. Mktrf_beta is the factor loading on the market risk premium. ME_beta is the factor loading on the size factor. I/A_beta is the factor loading on the investment factor. ROA_beta is the factor loading on the profitability factor. The Newey–West corrected t-statistics (Bartlett kernel with a lag length of 3) are reported below the coefficient estimates. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels (for the alpha coefficients). Panel C of this table reports subsequent monthly equally-weighted alphas and betas from the fund returns of sample funds by using Hou et al. (2015) four-factor model. We rank each fund by its MPPM from its past 12-month returns to sort them into deciles at the end of each quarter. Rank 10 indicates for best decile and Rank 1 for the bottom decile. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stock characteristics are from CRSP U.S. Stock Database. The sample period is from April 1980 to December 2013. Alpha is the intercept of the four-factor model. Mktrf_beta is the factor loading on the market risk premium. ME_beta is the factor loading on the size factor. I/A_beta is the factor loading on the investment factor. ROA_beta is the factor loading on the profitability factor. The Newey–West corrected t-statistics (Bartlett kernel with a lag length of 3) are reported below the coefficient estimates. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels (for the alpha coefficients). Panel D of this table reports subsequent monthly valued-weighted alphas and betas from the fund returns of sample funds by using Hou et al. (2015) four-factor model. We rank each fund by its MPPM from its past 12-month returns to sort them into deciles at the end of each quarter. Rank 10 indicates for best decile and Rank 1 for the bottom decile. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stock characteristics are from CRSP U.S. Stock Database. The sample period is from April 1980 to December 2013. Alpha is the intercept of the four-factor model. Mktrf_beta is the factor loading on the market risk premium. ME_beta is the factor loading on the size factor. I/A_beta is the factor loading on the investment factor. ROA_beta is the factor loading on the profitability factor. The Newey–West corrected t-statistics (Bartlett kernel with a lag length of 3) are reported below the coefficient estimates. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ and indicated whether the results are statistically different form zero at the 10%, 5% and 1% significance levels (for the alpha coefficients).

Table A.2 Subsequent quarterly HXZ q-factor (four-factor) alphas and betas from copycat portfolios. Diff (added–sold)

Alpha t-stat Mktrf_beta t-stat ME_beta t-stat I/A_beta t-stat ROA_beta t-stat

MPPM rank Rank 1 (lowest)

Rank 2

Rank 3

Rank 4

Rank 5

Rank 6

Rank 7

Rank 8

Rank 9

Rank 10 (highest)

0.0009 0.66 0.0250 0.62 0.0553 1.15 0.0710 0.77 0.1354 2.16

0.0010 0.94 0.0871 3.24 0.0560 1.42 0.0329 0.54 0.0236 0.46

0.0007 0.68 0.0391 1.34 0.0949 1.98 0.0260 0.42 0.0263 0.53

0.0000 0.01 0.0491 1.80 0.0955 1.68 0.0173 0.25 0.0106 0.25

0.0008 0.75 0.0625 2.17 0.1009 1.97 0.0017 0.03 0.0245 0.49

0.0005 0.51 0.0436 1.52 0.1315 1.53 0.0421 0.52 0.0048 0.10

0.0007 0.64 0.0519 2.04 0.0793 1.95 0.0214 0.21 0.0699 1.21

0.0016 1.36 0.0290 1.28 0.0616 1.21 0.0906 1.00 0.1289 2.28

0.0018⁄ 1.77 0.0351 1.23 0.1112 3.11 0.0372 0.54 0.0270 0.49

0.0032⁄⁄⁄ 2.80 0.0403⁄ 1.80 0.1051⁄⁄ 2.57 0.1374⁄⁄ 2.21 0.0551 1.05

This table reports the subsequent quarterly four-factor (Hou et al. (2015)) alphas and betas by taking the differences of using the portfolios added to the holdings to subtract the portfolios sold by the managers of the sample funds. We rank all domestic equity mutual funds by manipulation-proof performance measure (MPPM (Goetzmann et al. (2007)) from their past 12-month returns to sort them into decile at the end of each quarter. Rank 10 indicates the highest score and Rank 1 for the lowest among the deciles. Mutual fund data is from CRSP Survivorship Bias Free Mutual Fund Database and stocks returns are from CRSP U.S. Stock Database. The sample period is from April 1980 to September 2013. A fund’s buying portfolio is categorized in the top decile group of stocks that gained the highest market value among the holdings in the current quarter. A fund’s selling portfolio is categorized in the top decile group of stocks that lost the highest market value among the holding in the current quarter. The Newey–West corrected t-statistics (Bartlett kernel with a lag length of 3) are reported below the coefficient estimates. The significance levels are denoted by ⁄, ⁄⁄, ⁄⁄⁄ that are different form zero at the 10%, 5% and 1% significance levels (for the alpha coefficients).

References Alti, Aydogan, Kaniel, R., Yoeli, U., 2012. Why do institutional investors chase return trends? Journal of Financial Intermediation 21, 694–721. Amihud, Y., 2002. Illiquidity and stock returns: cross-section and time-series effects. Journal of Financial Markets 5, 31–56. Amihud, Y., Goyenko, R., 2013. Mutual fund’s R2 as predictor of performance. Review of Financial Studies 26, 667–695.

Bhattacharya, U., Hackethal, A., Kaesler, S., Loos, B., Meyer, S., 2012. Is unbiased financial advice to retail investors sufficient? Answers from a large field study. Review of Financial Studies 25, 975–1032. Bollen, N., Busse, J., 2004. Short-term persistence in mutual fund performance. Review of Financial Studies 18, 569–597. Bollen, N., Pool, V., 2008. Conditional return smoothing in the hedge. Journal of Financial and Quantitative Analysis 43, 267–298. Brown, S., Gallagher, D., Steenbeek, O., Swan, P., 2004. Double or Nothing Patterns of Equity Fund Holdings and Transactions, Working Paper, New York University.

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008

16

F. Chen et al. / Journal of Banking & Finance xxx (2016) xxx–xxx

Brown, S., Goetzmann, W., 1995. Performance persistence. Journal of Finance 50, 679–698. Bullard, M., 2006. The mutual fund as a firm: frequent trading, fund arbitrage, and the SEC’s response to the mutual fund scandal. Houston Law Review 42, 1271– 1330. Cao, C., Ghysels, E., Hatheway, F., 2010. Derivatives do affect mutual fund returns: Evidence from the financial crisis of 1998. Journal of Futures Market 31, 629– 658. Carhart, M., 1997. On persistence in mutual fund performance. Journal of Finance 52, 57–82. Chung, Y., Kim, T., 2014. Law of large numbers in mutual funds: A simple but effective way to identify persistent performances among actively-managed mutual funds, Working paper, University of California, Riverside. CiCi, G., Palacios, L., 2015. On the use of options by mutual funds: do they know what they are doing? Journal of Banking and Finance 50, 157–168. Cooper, M., Gulen, H., Schill, M., 2008. Asset growth and the cross-section of stock returns. Journal of Finance 63, 1609–1651. Corwin, S., Schultz, P., 2012. A simple way to estimate bid-ask spreads from daily high and low prices. Journal of Finance 67, 719–760. Cremers, M., Petajisto, A., 2009. How active is your fund manager? A new measure that predicts performance. Review of Financial Studies 22, 3329–3365. Daniel, K., Grinblatt, M., Titman, S., Wermers, R., 1997. Measuring mutual fund performance with characteristic-based benchmarks. Journal of Finance 52, 1035–1058. Falkenstein, E., 1996. Preferences for stock characteristics as revealed by mutual fund portfolio holdings. Journal of Finance 51, 111–135. Fama, E., French, K., 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33, 3–56. Fama, E., French, K., 2010. Luck versus skill in the cross-section of mutual fund returns. Journal of Finance 65, 1915–1947. Fama, E., French, K., 2015. A five-factor asset pricing model. Journal of Financial Economics 116, 1–22. Fama, E., Macbeth, J., 1973. Risk, return, and equilibrium: empirical tests. Journal of Political Economy 81, 607–636. Ferson, W., Siegel, A., 2001. The efficient use of conditioning information in portfolios. Journal of Finance 3, 967–982. French, K., 2008. The cost of active investing. Journal of Finance 63, 1537–1573. Goetzmann, W., Ingersoll, J., Spiegel, M., Welch, I., 2007. Portfolio performance manipulation and manipulation-proof performance measures. Review of Financial Studies 20, 1503–1546. Gompers, P., Metrick, A., 2001. Institutional investors and equity prices. Quarterly Journal of Economics 116, 229–259. Henriksson, R., 1984. Market timing and mutual fund performance: An empirical investigation. The Journal of Business 57, 73–96.

Hou, K., Xue, C., Zhang, L., 2015. Digesting anomalies: an investment approach. Review of Financial Studies 28, 650–705. Huang, J., Sialm, C., Zhang, H., 2011. Risk shifting and mutual fund performance. Review of Financial Studies 24, 2575–2616. Jensen, M., 1969. Risk, the pricing of capital assets, and the evaluation of investment portfolios. Journal of Business 42, 167–247. Kacperczyk, M., Sialm, C., Zheng, L., 2008. Unobserved actions of mutual funds. Review of Financial Studies 21, 2379–2416. Kacperczyk, M., Van Nieuwerburgh, S., Veldkamp, L., 2014. Time-varying fund manager skill. Journal of Finance 69, 1455–1484. Lhabitant, F.S., 2000. Hedge funds and investment partnerships: a primer. Discussion paper, Thunderbird, the American Graduate School of International Management. . Lynch-Koski, J., Pontiff, J., 1999. How are derivatives used? Evidence from the mutual fund industry. Journal of finance 54, 791–816. Newey, W., West, K., 1987. A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica 55, 703–708. Novy-Marx, R., 2013. The other side of value: the gross profitability premium. Journal of Financial Economics 108, 1–28. Petajisto, A., 2013. Active share and mutual fund performance, Yale University, working paper. Porter, G., Trifts, J., 1998. Performance persistence of experienced mutual fund managers. Financial Service Review 7, 57–68. Qian, M., Yu, B., 2015. Do mutual fund managers manipulate? Applied Economics Letters 22, 967–971. Richard, B., 2001. How to game our sharpe ratio. Journal of Alternative Investments 4, 38–46. Titman, S., Wei, K.C., Xie, F., 2004. Capital investments and stock returns. Journal of Financial and Quantitative Analysis 39, 677–700. Titman, S., Tiu, C., 2011. Do the best hedge funds hedge? Review of Financial Studies. 24, 123–168. Treynor, J., Mazuy, K., 1966. Can mutual funds outguess the market? Harvard Business Review 44, 131–136. Wermers, R., 2000. Mutual fund performance: an empirical decomposition into stock-picking talent, style, transaction costs, and expenses. Journal of Finance 55, 1655–1703. Weisman, A., 2002. Informationless investing and hedge fund performance measurement bias. Journal of Portfolio Management 28, 80–91. Yan, X., Zhang, Z., 2009. Institutional investors and equity returns: are short-term institutions better informed? Review of Financial Studies 22, 893–924. Zheng, L., 1999. Is money smart? A study of mutual fund investors’ fund selection ability. Journal of Finance 54, 901–933.

Please cite this article in press as: Chen, F., et al. In search for managerial skills beyond common performance measures. J. Bank Finance (2016), http://dx. doi.org/10.1016/j.jbankfin.2015.12.008