Brothers from different mothers how distribution fees change investment behavior

Journal of Banking & Finance 51 (2015) 12–25

Contents lists available at ScienceDirect

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

Brothers from different mothers how distribution fees change investment behavior Marco Navone a,⇑, Marco Pagani b a b

Finance Discipline Group, UTS Business School, University of Technology, Sydney San Jose State University, San Jose, CA, United States

a r t i c l e

i n f o

Article history: Received 5 December 2013 Accepted 29 October 2014 Available online 6 November 2014 JEL classification: G11 G14 G23 Keywords: Mutual fund flows Brokerage services Loads Irreversible investments

a b s t r a c t We ask whether loads affect investment flows in the US mutual fund industry. We argue that sales fees make the investment decision partially irreversible. Under these circumstances investors await for a stronger signal of managerial ability before committing to a new fund. This stronger signal can take the form of a particularly strong performance or a particularly long series of positive performance realizations. Looking at pairs of fund shares with the same portfolio but different sales fee arrangements we show that investment flows in share classes with front loads react disproportionally to good performances (higher convexity in the flow-performance relationship) and react to performance realizations further back in time (longer memory). A counterfactual example of fund shares with back-end loads allows us to rule out the hypothesis that this behavior is due to the incentive structure of brokers. Finally we show that these behavioral modifications induced by front loads have a negative and significant effect on investors’ timing ability. Ó 2014 Elsevier B.V. All rights reserved.

1. Introduction Investors pay for asset management services via annual management and performance fees. When investment flows are intermediated by a third party fund management companies can charge a one-off sales fee to be paid at the subscription or at the redemption of the fund shares. While the relevance of sales fee has steadily decreased in the last ten years it still remains substantial1 and a significant number of contributions have addressed the effect of loads on investment decisions. The effect of sales fees can be analyzed at two different levels: on the one side they usually indicate the presence of a broker within the investment decision process2 and many authors have argued that broker incentives may pay a key role in the choice of

⇑ Corresponding author at: UTS Business School, PO Box 123, Broadway, NSW 2007, Australia E-mail address: [email protected] (M. Navone). 1 The 2013 Investment Company Fact Book reports a weight of load fund shares in terms of asset under management declining from 36.5% in 2003 to 25.4% in 2012. 2 See Pozen and Hamacher (2011) for examples of revenues sharing agreements between brokers and investment companies. http://dx.doi.org/10.1016/j.jbankfin.2014.10.013 0378-4266/Ó 2014 Elsevier B.V. All rights reserved.

investment product (see for example O’Neal, 1999; Zhao, 2008; Del Guercio et al., 2010; Lemeunier, 2011). On the other side investment cost is one of the key determinants, together with past performance, of investment decisions. While some authors have argued that investors only care about a comprehensive ‘‘total investment cost’’ measure where all the cost components are lumped together (see for example Sirri and Tufano, 1998; Khorana et al., 2009) other authors have demonstrated that the ‘‘form’’ of the charge (one-off vs. recurring annual fee) is relevant as it may affect investors’ behavior (see for example Barber et al., 2005). Overall, our understanding of the investment effect of distribution fees is so far limited. A major problem in the previous literature is that load and no-load funds investment policies may differ significantly. If for example, as postulated by Houge and Wellman (2007), load funds cater to less sophisticated investors, then we should expect them to invest more conservatively. Under these conditions is very hard to disentangle the effect of the load from the effect of the different investment policy when looking at investors’ behavior: failure to properly control for portfolio differences between load and no load fund could lead to erroneously attribute to loads what is in reality caused by differences in investment strategies.

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25

In this paper we address this issue looking at mutual funds with multiple share classes and considering couples of fund shares (with and without loads) with the same portfolio. This allows us to isolate the effect of the load on investors’ behavior. We call these pairs ‘‘brothers from different mothers’’ because we document striking dissimilarities in investors’ behavior in spite of the identical portfolio composition. In this paper we hypothesize that when investors infer managerial ability from funds’ past performance, their investment behavior will be affected by the presence of loads. Specifically we will argue that investors become ‘‘more cautious’’ and wait for a stronger signal of managerial ability before committing to a new fund. This effect derives from the fact that while dissatisfied investors in no-load funds can easily switch to another product, the front load acts as a non-recoverable upfront cost and makes the investment decision partially irreversible.3 A ‘‘stronger’’ signal of managerial ability can derive from a particularly good past performance or from a long series of positive performances. We empirically show that investment flows in fund shares with front load are characterized by: (a) a more convex relationship with past performance, they react disproportionally to strong returns; (b) a longer memory, they react to performance realizations further back in time. A counterfactual example looking at fund shares with back-end loads confirms that these effects are generated by the salience and the upfront nature of front loads rather than by the involvement of a broker in the investment decision process. Finally we look at how these behavioral modifications induced by front loads affect investors’ timing ability measured as the difference between dollar-weighted and time-weighted rates of returns of investment flows (as in Friesen and Sapp, 2007). We show that investment flows into share classes with front loads show worse timing ability than flows into the no-load class of the same fund with the difference both statistically and economically significant. The rest of the paper is organized as follows: in Section 2 we summarize the relevant literature and develop our hypotheses; This paper is divided into four sections. Section 2 covers the relevant literature and develops the main hypotheses; Section 3 describes the data, Section 4 analyses the effect of loads on investment decisions; Section 5 analyses the effect of loads on timing ability; Section 6 summarizes and concludes. 2. Relevant literature and hypotheses In this paper we address three strands of financial literature: (a) The effect of sales fees on investment flows in the mutual fund industry; (b) The role of brokers as advisors for mutual fund investors; (c) The role of (partial) irreversibility on investment decisions. 2.1. Loads and investment flows The effect of loads on investment flows has been the object of a number of relevant contributions. Sirri and Tufano (1998) and, more recently, Khorana et al. (2009) show that investment flows respond negatively to a total investment cost measure built as the expense ratio plus the amortized value of loads. Barber et al. (2005) separate the expense ratio from loads and show that the latters are more salient for investors and have a stronger (negative) impact on flows than the former. This result has been later contradicted by Zhao (2008) and Bergstresser et al. (2009) who, 3 The reader should note that investment in a portfolio of liquid financial asset is totally reversible in absence of transaction costs. The load here acts as a form of transaction cost that is paid at the moment of the purchase (front load) or at moment of the sale (back-end load).

13

looking at a sample of funds distributed through the broker channel, find a positive relationship between investment flows and loads. More recently Christoffersen et al., 2013 find a positive relationship between investment flows and the portion of loads paid to brokers. Different hypothesis have been advanced to justify this positive effect: Nanda et al. (2009) argue that offering multiple share classes mutual funds are able to attract investors with a range of different preferences.4 Iannotta and Navone (2012) argue that loads increase fund visibility and lower search costs through the fund selection activity of the broker. A common characteristic of these papers is that they do not account for portfolio differences between load and no-load funds. While every analysis includes some controls such as portfolio performance, size, turnover, volatility etc. it is impossible to completely rule out the possibility that the samples of load and no-load funds differ along some unobserved dimension. 2.2. Brokers and investment decisions Investors can buy mutual fund shares through a variety of channels. Brokers can offer both transactional and advisory services. In the latter case they may advise the client in asset allocation and or fund selection. While the obvious rationale for this service is that brokers should have better knowledge and higher skill Bergstresser et al. (2009), and Friesen and Sapp (2007) show that investment flows intermediate through the broker channel either do not earn any extra performance or they outright underperform. While Bergstresser et al. (2009) suggest that investors may derive some unobserved non-portfolio benefit from their relationship with brokers, other authors have pointed out that brokers advice may be affected by conflict of interest. O’Neal (1999) shows that brokers have the incentive to steer clients towards fund shares with load structure that is not optimal for their planned investment horizon. Houge and Wellman (2007) and Zhao (2008) argue that brokers target clients with lower sophistication and empirically show that brokers distribute funds with higher fees. Since these studies do not properly account for portfolio differences between fund sold through direct and brokered channel is not possible to attribute the lack of performance of brokered flows to bad timing of investment decisions or bad fund selection. 2.3. Irreversibility and investment decisions The effect of sunk costs and irreversibility on investment decisions has been thoroughly investigated both in economics and corporate finance literature. While basing investment decisions on past financial outlays is considered irrational (the so-called sunk cost fallacy documented, among many others, by Kahneman and Tversky, 1979 and Thaler, 1980) firms rationally anticipate the unrecoverable nature of their future investments when they evaluate prospective projects. Pindyck (1991) summarizes the main findings of economic literature on irreversible investment decisions under uncertainty. These contributions mainly deal with investments in physical assets with limited secondary market value such as industrial plants or infrastructures such as pipelines and oil wells. On the other side financial assets are inherently more liquid and thus portfolio investments are eminently reversible, with transaction costs being the only unrecoverable component. When transaction costs are low irreversibility should not play a 4 An exploratory analysis of our database suggests that out of 7636 unique funds with a single share class active between 1999 and 2011 5570 where no-load. The predominance of no-load share classes among single-class mutual fund could account for the positive effect between loads and flows under the Nanda et al. (2009) argument that the introduction of multiple share classes increases the total flows to the fund.

14

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25

role in investment decisions. On the other side if market frictions are substantial (for example acquisition of a majority stake in a company) we should expect some of the intuitions developed in this literature to carry over. In this paper we look at loads as an unrecoverable component of mutual fund investment: while the investment in a no-load fund share is completely reversible investors committing to a load share class will not be able to recover a portion of their investment if they decide to switch in a new product. The relevance of loads as a source of irreversibility is somewhat an empirical question: while the maximum load advertised by mutual funds is sizeable (5.3% on average in the last ten years according to Investment Company Institute (2013)) the actual average load paid by clients is significantly lower and has decreased from 3.9% in 1990 to 1.0% in 2012. In this paper we postulate that the effect is strong enough to significantly affect investment flows. Pindyck (1991) shows that a company evaluating an irreversible investment in the production of a good with a variable market price will commit to the project only when the current market price of the output is above a certain threshold (i.e. the expected return on the investment is above a certain level). The reservation price increases with the volatility of output price and with the degree of irreversibility. This example can easily be translated to a partially irreversible investment in a mutual fund with an uncertain return. The investor will commit only when the expected performance is above a certain reservation level. If the fund is actively managed the expected performance is a function (also) of the expected unobservable managerial ability. If investors infer managerial ability from past performance they will commit to the fund only if past performance signal a sufficiently strong managerial ability. Based on this intuition we develop two testable hypotheses:

of mind’’ argument specifically consider front-end load, and argue that ‘‘front-end loads are more salient than operating expenses. Frontend-load fees are paid when a fund is purchased and generally obvious in nominal terms on the first statement following the transaction’’. They purposefully avoid back-end load because they ‘‘are often waived if an investor holds a fund for a specified period of time’’ and thus is difficult for the investor to correctly anticipate their value at the moment of the investment decision. Both Hortaçsu and Syverson (2004) and Iannotta and Navone (2012) show that front-end and back-end loads have remarkably different effects on investment flows. Specifically they show that the main effect of back-end loads is to act as ‘‘switching costs’’: they affect the decision to abandon a fund rather than the decision of which fund to select. Iannotta and Navone (2012) empirically show that, while for funds with front load investment flows are better explained by past performance (and other explanatory variables), the opposite is true for fund shares with back-end loads: investors may be unwilling to disinvest from a poorly performing fund due to the exit cost represented by the deferred load. If investors were able to fully discount the expected value of the deferred load at the moment of the investment decision we should not observe any differential effect on investment flows (with respect to front loads). The opposite empirical evidence leads us to conclude that investors do not fully anticipate the value of deferred loads so we expect their effects on investment flows to be somewhat weaker than front loads. Finally we do not have any specific hypothesis on the effect of loads on investors’ timing ability. To the best of our knowledge there is no reason to assume this effect to be positive or negative so we consider this to be an empirical question.

3. Database H1. Investment flows into load fund shares will exhibit a more convex relationship with past performance. The intuition is that a stronger signal of managerial ability can derive from a particularly good performance realization. Huang et al. (2007) develop a similar intuition looking at investors who have to pay a cost to collect information prior to committing to a fund. They show that when these search costs are high the relationship between investment flows and performance becomes indeed more convex because investors will be willing to commit to a fund only after a strong past performance (when their assessment of managerial ability is above the reservation level). H2. Investment flows into load fund shares will exhibit a longer memory responding to performance realization further back in time. In a multiperiod setting the subjective assessment of managerial ability can improve after a strong performance realization or after a long series of positive outcomes. Persistence of over-performance has been considered the hallmark of managerial ability (see, for example Hendricks et al., 1993 and Carhart, 1997). A longer series of positive performance realizations reduces the uncertainty about the ability of the fund manager. In the context of Pindyck (1991) this is equivalent to reduction in the risk of the investment. The same result can be observed in a recursive application of the mechanism with which investors infer managerial ability modelled by Huang et al. (2007). Both of these hypotheses can be applied in line of principle to front loads and back-end loads alike. Both forms of sales fee generate partial irreversibility and if properly anticipated should affect investors’ behavior. The extent to which both forms of loads are ‘‘within the sight’’ of investors is ultimately an empirical issue. Barber et al. (2005) in developing the original ‘‘out of sight, out

We use data from the Center for Research in Security Prices (CRSP) Survivorship Bias Free Mutual Fund Database, from which we obtain information about funds’ net asset values, returns and characteristics. Data are collected from 1999 to 2011 on all US mutual funds with assets under management, at the beginning of the year, above 10 million dollars. Following Evans (2010) we only consider funds with age non shorter than 36 months in order to avoid investment vehicles in their incubation period. Since our main experiment is centered around couples of share classes from the same portfolio but with different fee structure we need a portfolio identifier that is provided by CRSP only from 1999. Our experiment also provides a natural control for the specificities of different investment objectives so we do not impose any restriction on fund strategy. One concern with our sample selection is that while our experiment is designed to avoid any bias due to different portfolio composition of load and no-load funds (we confront couples of share classes with the same portfolio but different fee structure) our results could still be biased by unobservable differences between investors of load and no-load funds. Specifically different authors have argued that no-load funds cater to more sophisticated investors unwilling to pay for the advice of a broker (Bergstresser et al., 2009). If this is true investment flows in the two groups of funds could reflect this different level of sophistication more than the effect that we are trying to measure here. We try to minimize this problem in two ways: first of all we drop all share classes targeted at institutional investors. Goetzmann and Peles (1997) and Bailey et al. (2011) show that retail investors are prone to a number of behavioral biases. If our subsample of load and no-load shares reflect a different composition of institutional and retail shares our results could be biased. Restricting the analysis to fund shares targeted at retail investors should reduce the difference in sophistication levels between the two subsamples of fund shares.

15

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25

Table 1 Descriptive statistics (complete sample). The table reports descriptive statistics for a sample of US mutual funds from 1999 to 2011. Each observation represents a quarter/fund share. Flows are calculated monthly and aggregated on each quarter. Fund Size (in m$), Age (in months) and Expense Ratio (in excess with respect to the average of each investment objective) are measured at the beginning of each quarter. Performance is the fractional ranking of the fund within the investment objective in the previous year. Mean values (and standard deviations in parenthesis) are reported for the total sample and for three subsamples: No load funds, funds that charge Front loads and funds that charge back-end loads. The last two columns report T-tests for the significance of the differences between the two subsamples with load and the no-load subsample (standard errors in parenthesis). ⁄⁄⁄, ⁄⁄ and ⁄ represent significance at the 1%, 5% and 10% respectively. Total

No load

Front load

Back-end load

Front load – no load

Back-end load – no load

N. of observations N. of unique fund shares N. of unique portfolios Size

568,427 25,554 14,113 733.3 (3152.1)

260,720 16,072 10,515 1023.0 (3860.3)

126,184 5189 5899 670.6 (3312.0)

221,986 9820 6148 410.8 (1465.0)

310.9⁄⁄⁄ (15.323)

650.8⁄⁄⁄ (8.409)

Performance

0.500 (0.286)

0.543 (0.282)

0.528 (0.282)

0.486 (0.290)

0.017⁄⁄⁄ (0.001)

0.067⁄⁄⁄ (0.001)

Flows

0.777% (0.130)

1.738% (0.146)

0.663% (0.111)

0.245% (0.117)

1.143%⁄⁄⁄ (0.000)

2.219%⁄⁄⁄ (0.000)

Age

131.0 (98.8)

128.9 (93.0)

158.0 (131.7)

122.1 (86.8)

31.5⁄⁄⁄ (0.491)

13.7⁄⁄⁄ (0.250)

Expense Ratio

0.000% (0.005)

0.264% (0.004)

0.129% (0.003)

0.181% (0.005)

0.119%⁄⁄⁄ (0.000)

0.506%⁄⁄⁄ (0.000)

As a second robustness check we try to identify, among fund shares with front loads, the investment vehicles targeted at large (and possibly more sophisticated investors). We do this by looking at the fee discount granted to investors willing to commit at least one million of dollars. This ‘‘discount’’ is a measure of how interested the fund is in attracting large investors. The median value of this discount in our sample is 3.5% while the 5th percentile is 2% and the 95th is 3.75%. We can also note that 1% of the funds do not offer any discount to large investors. A large drop indicates willingness to offer strong discounts to attract large investors. We use this variable to flag as ‘‘Large Oriented’’ all the funds and with a ‘‘drop’’ above the median. All our results are robust to the inclusion of this additional measure of investors’ sophistication.5 It is important to notice that while these robustness checks suggest that our results are not driven by different sophistication among investors in load- and no-load fund shares we are not in any way testing whether difference in sophistication actually exists or whether it affects in any way the demand function of mutual fund shares. We calculate net investment flows defined as the percentage growth of total net assets (TNA) adjusted for the fund return net of expenses (rit):

Flowit ¼

TNAit  TNAit1 ð1 þ r it Þ TNAit1

ð1Þ

Flows are calculated on a monthly basis and aggregated over the quarter in order to minimize the approximation errors due to the timing of investment decisions. The quarterly variable is then winsorized at the 5% level to eliminate outliers. As the performance measure we consider the fund’s fractional rank in the previous period (Rank). This represents its percentile performance relative to other funds with the same investment objective, and ranges from 0 to 1. This measure captures the tournament nature of the mutual fund industry (Brown et al., 1996) and has been proven to be highly relevant in terms of its ability to capture investors’ behavior.6 Performance is calculated over the previous year, the previous two semesters or the previous four quarters. Basing the performance variable on a risk-adjusted return measure such as the Sharpe Ratio does not change the results of our analysis.7

As a measure of fund cost we consider the total expense ratio (Exp Ratio) in excess over the average expense ratio for funds with the same investment objective. Descriptive statistics for our complete sample are in Table 1. The sample contains data on 25,554 unique fund shares belonging to 14,113 different portfolios. Subsamples of no load and load funds differ along multiple dimensions, with no-load funds being on average larger and experiencing higher performance and flows, while at the same time charging lower expense ratios. The average no-load fund is also 13 months older than the average fund that charges back-end loads, and 31 months younger than the average fund share with front loads. Our main experiment will compare couples of fund shares from the same portfolio but with different fee structure. We look at mutual funds with multiple share classes that have at least one share class with front load and one no-load share class. These two fund shares have, by definition, the same portfolio and the only difference between them is the fee structure. When multiple no-load fund shares are available we consider the largest one in terms of asset under management. We repeat the same procedure for our second subsample where we pair no-load share classes with share classes that charge back-end loads. Table 2 reports descriptive statistics for these two subsamples. When we compare load and no-load funds removing all the portfolio differences we get a different picture. For example we see that now front-load fund shares have lower expense ratios (and consequently higher net performance) than the respective no-load share. They also experience significantly higher investment flows and consequently have larger size. The difference between this comparison and the one in the complete sample of Table 1 highlights the importance of removing portfolio differences when comparing load and no-load funds. Interestingly the back-end load sample behaves differently, with load funds charging higher expense ratios and experiencing lower performance and flows. The fact that funds trade front-loads for a lower expense ratio whereas no such trade is in place for back-end loads seems to support the intuition of Barber et al. (2005) that front-load are particularly salient for investors and command some sort of compensation in terms of lower annual fee. In this sense back-end loads are clearly less relevant for the initial investment decision, but may create a barrier to exit, allowing for an increase in expense ratios.8

5

Results for this robustness check are available from the authors upon request. See, for example, Sirri and Tufano (1998), Huang et al. (2007), Kempf and Ruenzi (2008) and Kempf et al. (2009). 7 Results of this robustness check are available from the authors upon request. 6

8 The fact that back-end loads can affect exit decisions and pricing of mutual fund has been documented by Hortaçsu and Syverson (2004) and Iannotta and Navone (2012).

16

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25

Table 2 Descriptive statistics (matched sample). The table reports descriptive statistics for two samples of matched fund shares. The front load sample considers only funds with at least one no-load share and one share that charges front loads. Where multiple shares in each category are available only the share with the largest AUM in each group is chosen. The back-end load sample considers only funds with at least one no-load share and one share that charges back-end loads. Where multiple shares in each category are available only the share with the largest AUM in each group is chosen. Each observation represents a quarter/fund share. Flows are calculated monthly and aggregated on each quarter. Fund Size (in m$), Age (in months) and Expense Ratio (in excess with respect to the average of each investment objective) are measured at the beginning of each quarter. Performance is the fractional ranking of the fund within the investment objective in the previous year. Mean values (and standard deviations in parenthesis) are reported for the different subsamples. Columns number three and six report T-tests for the significance of the differences between the mean values of the two subsamples in each group (standard errors in parenthesis). ⁄⁄⁄, ⁄⁄ and ⁄ represent significance at the 1%, 5% and 10% respectively. Front load sample

N. of observations N. of unique portfolios Size Performance Flows Age expense ratio

Back-end load sample

Load

No load

12,243 1526 798.1 (2068.4) 0.537 (0.275) 0.937% (0.119) 171.2 (132.6) 0.126% (0.003)

12,243

Difference 726 397.5⁄⁄⁄ (21.337) 0.035⁄⁄⁄ (0.004) 0.636%⁄⁄⁄ (0.002) 34.1⁄⁄⁄ (1.479) 0.298%⁄⁄⁄ (0.000)

400.7 (1138.3) 0.502 (0.280) 0.301% (0.122) 137.1 (96.0) 0.172% (0.005)

4. Loads and investment behavior We want to test whether the (partial) irreversibility of investment decisions induced by an up-front one-off sales fee that investors have to pay when they buy mutual fund shares affects their investment behavior. Our hypothesis is that as investors infer managerial ability from fund past performance, the presence of an unrecoverable investment cost will force them to wait for ‘‘stronger evidence’’ of managerial ability before investing in the fund. This may take two forms: an increase in the convexity of the flow-performance relationship where flows respond disproportionately to particularly strong performances (similar to what Huang et al., 2007 postulate when investors face initial search and participation costs) or an increase of the relationship between flows and older performance, when investors wait for a longer series of positive outcomes before investing in the fund. To start we measure the ‘‘baseline’’ investors’ behavior by running a standard flow-performance regression where the dependent variables are quarterly net investment flows and the independent variables are past performance and other fund characteristics that from previous research we know to be relevant in determining fund flows.

Flowit ¼ a þ bC þ e

ð2Þ

In model (1) C is a vector of explanatory variables that contains relevant fund share characteristics: Past Performance as the performance measure we consider the fractional rank (Rank) defined in the previous paragraph. In different models we will consider the performance in the past year or indifferent time subsamples (past two semesters or past four quarters). In order to allow for non-linearity in the flow-performance relationship we use a piecewise specification similar to Sirri and Tufano (1998) where we define Rank High and Rank Low considering 0.5 as the breakpoint. Expenses I consider the total expense ratio (Exp Ratio) in excess of the category average in each given quarter. The normalization avoids biases due to differences in expense ratios between investment objectives and time trend in expense ratios. We also include a group of variables to control for relevant fund characteristics: the (natural log of) fund asset under management at the end of the previous quarter (Size), the (natural log of) fund age in months (Age), a dummy variable equal to 1 if the fund share charges loads (Load).

Load

No load

5085

5085

395.5 (1106.1) 0.457 (0.283) 1.176% (0.116) 126.5 (88.7) 0.442% (0.004)

475.7 (1143.8) 0.516 (0.278) 0.612% (0.125) 124.7 (95.5) 0.002% (0.005)

Difference

80.2⁄⁄⁄ (22.313) 0.059⁄⁄⁄ (0.006) 1.788%⁄⁄⁄ (0.002) 1.8 (1.828) 0.444%⁄⁄⁄ (0.000)

Previous contributions also consider other characteristics such as turnover and flows into other funds managed by the same investment company or other funds with the same investment objective. These factors, albeit important, are all defined at the fund and not at the fund share level. As such they will not affect the difference in investment flows for fund shares with the same portfolio but different fee structure. To simplify our analysis, and focus our attention on the variables that differentiate fund shares within the same fund. We can capture all these other fund-level effects including, as an additional independent variable, the average net investment flows to ‘‘other’’ fund shares of the same fund (PortFlow). This variable should capture all the effects hinted above: for example if flows are affected by turnover, since all fund shares have the same portfolio (hence the same turnover) the variable will capture this effect. Similarly if all the funds in a certain investment objective experience high flows (we can think for example at technology funds during the so-called ‘‘internet bubble’’), since all the fund shares of the same fund have the same investment objective the effect will be captured. Finally if we have a ‘‘star’’ effect and all the funds managed by a certain investment company experience high flows, again this phenomenon will be captured by PortFlow. We also include the lagged value of net investment flows (Lag Flows) to capture the stickiness of investment flows. In all our models standard errors are clustered for portfolio identifier and quarter. Results in Table 3 show that investors’ baseline behavior in both our subsamples follows the regularities documented in previous literature. Specifically in models (1a) and (1b) we see that investment flows are positively correlated with past performance and negatively correlated with expense ratios, while the presence of loads does not seem to affect flows in any significant way. From models (2a) and (2b) we learn that the relationship between flows and past performance is indeed convex with the sensitivity of flows to performance of funds in the top half of the ranking significantly stronger than the sensitivity for funds in the lower half. In the last three models we split the performance of the last year into two (models 3a, 3b, 4a and 4b) or four (models 5a and 5b) sub periods and we can see that more recent performances are clearly more salient for investors and have a significantly stronger effect on investment flows. In models 4a and 4b we split the two six-months performance measures in High and Low performance to test whether the time frame effect is a by-product of the flow performance convexity. Results show that the two effects coexist and are indeed independent.

17

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25

Table 3 Basic investment flow regression. The table reports the results of a linear model where the dependent variable is the estimate of net quarterly investment flows and the independent variable fund performance and other characteristics. The analysis is repeated for two subsamples of matched fund shares. The front load sample considers only funds with at least one no-load share and one share that charges front loads. Where multiple shares in each category are available only the share with the largest AUM in each group is chosen. The back-end load sample considers only funds with at least one no-load share and one share that charges back-end loads. Where multiple shares in each category are available only the share with the largest AUM in each group is chosen. The performance measure is the fractional ranking of the fund within the investment objective in the previous year (Y1 Rank), the previous two semesters (S1 and S2 Rank) or the previous four quarters (Q1,. . .,Q4 Rank). In order to test for non-linearity in the flow-performance relationship we apply a piecewise specification to some of these performance variables using the median ranking (0.5) as the breaking point between High and Low. PortFlow is the average investment flow in the other fund shares of the same fund. T-Statistics in parenthesis. ⁄⁄⁄, ⁄⁄ and ⁄ represent significance at the 1%, 5% and 10% respectively. Front load sample

Constant Lag (Flows) Load Y1 Rank

Back-end load sample

(1a)

(2a)

(3a)

(4a)

(5a)

(1b)

(2b)

(3b)

(4b)

(5b)

0.047⁄⁄⁄ (6.748) 0.312⁄⁄⁄ (12.735) 0.004 (1.561) 0.026⁄⁄⁄ (7.227)

0.050⁄⁄⁄ (7.008) 0.311⁄⁄⁄ (12.733) 0.004 (1.522)

0.043⁄⁄⁄ (6.050) 0.312⁄⁄⁄ (12.698) 0.004 (1.568)

0.047⁄⁄⁄ (6.311) 0.312⁄⁄⁄ (12.670) 0.004 (1.509)

0.038⁄⁄⁄ (5.374) 0.313⁄⁄⁄ (12.758) 0.004 (1.555)

0.092⁄⁄⁄ (7.273) 0.201⁄⁄⁄ (8.006) 0.004 (1.191) 0.019⁄⁄⁄ (2.648)

0.094⁄⁄⁄ (7.166) 0.200⁄⁄⁄ (7.991) 0.004 (1.201)

0.092⁄⁄⁄ (6.975) 0.202⁄⁄⁄ (8.021) 0.004 (1.220)

0.097⁄⁄⁄ (6.629) 0.201⁄⁄⁄ (7.982) 0.004 (1.228)

0.091⁄⁄⁄ (6.667) 0.202⁄⁄⁄ (8.067) 0.004 (1.208)

0.015⁄⁄⁄ (2.655) 0.037⁄⁄⁄ (5.273)

Y1 Rank Low Y1 Rank High

0.011 (0.894) 0.027⁄⁄⁄ (2.926) 0.024⁄⁄⁄ (7.307) 0.012⁄⁄⁄ (3.489)

S1 Rank S2 Rank

0.016⁄⁄⁄ (2.833) 0.004 (0.696) 0.017⁄⁄⁄ (4.010) 0.030⁄⁄⁄ (4.688) 0.001 (0.089) 0.022⁄⁄⁄ (3.982)

S1 Rank Low S1 Rank High S2 Rank Low S2 Rank High

0.003 (0.277) 0.029⁄⁄⁄ (3.281) 0.001 (0.114) 0.009 (1.312)

0.005⁄⁄⁄ (6.652) 0.008⁄⁄⁄ (4.600) 1.709⁄⁄⁄ (6.309) 0.413⁄⁄⁄ (16.462)

0.005⁄⁄⁄ (6.699) 0.008⁄⁄⁄ (4.571) 1.752⁄⁄⁄ (6.414) 0.412⁄⁄⁄ (16.429)

0.005⁄⁄⁄ (6.651) 0.008⁄⁄⁄ (4.656) 1.677⁄⁄⁄ (6.216) 0.412⁄⁄⁄ (16.464)

0.005⁄⁄⁄ (6.704) 0.008⁄⁄⁄ (4.605) 1.742⁄⁄⁄ (6.380) 0.412⁄⁄⁄ (16.486)

0.024⁄⁄⁄ (8.115) 0.007⁄⁄ (2.544) 0.011⁄⁄⁄ (3.793) 0.004 (1.256) 0.005⁄⁄⁄ (6.595) 0.008⁄⁄⁄ (4.710) 1.671⁄⁄⁄ (6.119) 0.413⁄⁄⁄ (16.578)

24,486 0.355

24,486 0.355

24,486 0.356

24,486 0.356

24,486 0.356

Q1 Rank Q2 Rank Q3 Rank Q4 Rank Log (Size) Log (Age) Exp Ratio PortFlow Observations Adj R2

Now that we have the baseline behavior we can test whether this is in any way affected by the presence of an unrecoverable investment cost in the form of front loads. We run the following augmented version of the basic regression model

Flowit ¼ dL þ kL ½L  C þ cN þ kN ½N  C þ e

ð3Þ

where L is a dummy variable equal to one for fund shares that charges load and N, defined as (1  L) is a dummy variable that identifies no-load fund shares. The vectors of the interaction coefficients kL and kN capture the different behavior of investors in load and no-load fund shares. The difference between the two vectors captures the changes in investors’ behavior induced by the presence of loads. 4.1. Convexity of the flow-performance relationship In this paper we postulate that when prospective acquirers have to pay an unrecoverable cost in order to buy fund shares they will

0.004⁄⁄⁄ (3.711) 0.017⁄⁄⁄ (6.339) 1.865⁄⁄⁄ (4.799) 0.503⁄⁄⁄ (18.590)

0.004⁄⁄⁄ (3.715) 0.017⁄⁄⁄ (6.333) 1.899⁄⁄⁄ (5.000) 0.503⁄⁄⁄ (18.621)

0.004⁄⁄⁄ (3.648) 0.017⁄⁄⁄ (6.379) 1.872⁄⁄⁄ (4.897) 0.505⁄⁄⁄ (18.618)

0.004⁄⁄⁄ (3.674) 0.017⁄⁄⁄ (6.360) 1.948⁄⁄⁄ (5.210) 0.504⁄⁄⁄ (18.689)

0.010⁄ (1.827) 0.008⁄ (1.683) 0.000 (0.032) 0.005 (0.863) 0.004⁄⁄⁄ (3.626) 0.018⁄⁄⁄ (6.405) 1.875⁄⁄⁄ (4.940) 0.505⁄⁄⁄ (18.672)

15,732 0.306

15,732 0.306

15,732 0.306

15,732 0.306

15,732 0.306

want stronger evidence of managerial ability before committing to the fund. As a result our first hypothesis on the effect of front loads on investors’ behavior is that we expect an increase in the convexity of the flow performance relationship. We estimate four models with different configurations of the past performance variable. Table 4 reports, for each model, coefficients (and t-statistics in parenthesis) for no load and load funds and the difference between the two (and Wald statistic in parenthesis). From models 1 and 3 we see that load funds exhibit higher performance sensitivity in general, both when we look at the ranking over the previous year or the previous six months. Is interesting to note that load funds exhibit, at the same time, lower price sensitivity with a coefficient on expense ratio five times lower than no-load funds. This evidence is in contrast with the argument of Christoffersen and Musto (2002) according to whom: ‘‘from its investors’ point of view, a fund’s fee is simply a direct reduction of performance, so performance insensitivity implies price insensitivity’’. In models 2 and 4 we break the performance variables

18

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25

Table 4 Loads and the convexity of the flows-performance relationship. The table reports the results of the OLS estimation of Eq. (3) on a sample of matched fund shares. In Panel A the sample considers only funds with at least one no-load share and one share that charges front loads. Where multiple shares in each category are available only the share with the largest AUM in each group is chosen. In Panel B the sample considers only funds with at least one no-load share and one share that charges back-end loads. For each regression the table reports the interaction between the each independent variable and the dummies front load and no load. The third column in each model reports v2-tests for the difference between the two interaction coefficients. The dependent variable is an estimate of net quarterly investment flows and the independent variables are fund performance and other characteristics. The performance measure is the fractional ranking of the fund within the investment objective in the previous year (Y1 Rank), or the previous semester (S1 Rank). In order to test for non-linearity in the flow-performance relationship we apply a piecewise specification to some of these variables using the median ranking (0.5) as the breaking point between High and Low. PortFlow is the average investment flow in the other fund shares of the same fund. T-statistics and Wald statistics in parenthesis. ⁄⁄⁄, ⁄⁄ and ⁄ represent significance at the 1%, 5% and 10% respectively. (1) No load Panel A: Front load sample Constant 0.066⁄⁄⁄ (6.694) Lag (Flows) 0.350⁄⁄⁄ (11.652) Y1 Rank 0.017⁄⁄⁄ (3.884) Y1 Rank Low

(2)

(3)

(4)

Load

Diff

No load

Load

Diff

No load

Load

Diff

No load

Load

Diff

0.031⁄⁄⁄ (3.826) 0.267⁄⁄⁄ (10.325) 0.037⁄⁄⁄ (8.228)

0.035⁄⁄⁄ (8.650) 0.082⁄⁄⁄ (7.467) 0.02⁄⁄⁄ (15.679)

0.068⁄⁄⁄ (6.886) 0.350⁄⁄⁄ (11.650)

0.034⁄⁄⁄ (4.085) 0.267⁄⁄⁄ (10.332)

0.034⁄⁄⁄ (7.718) 0.083⁄⁄⁄ (7.543)

0.065⁄⁄⁄ (6.746) 0.351⁄⁄⁄ (11.675)

0.032⁄⁄⁄ (4.069) 0.272⁄⁄⁄ (10.366)

0.033⁄⁄⁄ (7.983) 0.079⁄⁄⁄ (6.921)

0.066⁄⁄⁄ (6.648) 0.351⁄⁄⁄ (11.667)

0.035⁄⁄⁄ (4.203) 0.272⁄⁄⁄ (10.345)

0.031⁄⁄ (6.359) 0.08⁄⁄⁄ (6.993)

0.010 (1.427) 0.024⁄⁄⁄ (2.670)

0.023⁄⁄⁄ (3.124) 0.049⁄⁄⁄ (4.944)

0.013 (2.037) 0.025⁄ (3.775) 0.017⁄⁄⁄ (4.172)

0.033⁄⁄⁄ (6.728)

0.016⁄⁄⁄ (6.755) 0.019⁄⁄⁄ (2.576) 0.045⁄⁄⁄ (5.313) 0.005⁄⁄⁄⁄⁄⁄ (4.851) 0.005⁄⁄ (2.122) 0.533 (1.145) 0.442⁄⁄⁄ (15.293) 24,486 0.360

0.005 (0.228) 0.025⁄⁄ (4.481) 0 (0.019) 0.006⁄⁄ (6.568) 1.496⁄⁄ (5.814) 0.051 (2.441)

Y1 Rank High S1 Rank

0 (0.008) 0.006⁄⁄ (6.333) 1.625⁄⁄⁄ (6.885) 0.048 (2.256)

0.005⁄⁄⁄ (5.059) 0.011⁄⁄⁄ (5.176) 2.036⁄⁄⁄ (5.741) 0.390⁄⁄⁄ (12.770)

0.005⁄⁄⁄ (5.313) 0.005⁄⁄ (2.129) 0.487 (1.085) 0.438⁄⁄⁄ (15.088) 24,486 0.360

0 (0.004) 0.006⁄⁄⁄ (6.645) 1.55⁄⁄⁄ (6.730) 0.048 (2.244)

0.005⁄⁄⁄ (5.046) 0.011⁄⁄⁄ (5.194) 2.021⁄⁄⁄ (5.722) 0.392⁄⁄⁄ (13.027)

0.005⁄⁄⁄ (4.809) 0.005⁄⁄ (2.175) 0.439 (0.942) 0.443⁄⁄⁄ (15.321) 24,486 0.360

0 (0.026) 0.006⁄⁄ (6.313) 1.582⁄⁄ (6.457) 0.051 (2.470)

0.014⁄⁄ (2.260) 0.020⁄⁄ (2.117) 0.005⁄⁄⁄ (5.046) 0.011⁄⁄⁄ (5.192) 2.029⁄⁄⁄ (5.735) 0.392⁄⁄⁄ (13.013)

0.035 (1.794) 0.01 (0.079) 0.004 (0.124)

0.075⁄⁄⁄ (5.431) 0.202⁄⁄⁄ (6.785)

0.113⁄⁄⁄ (4.661) 0.192⁄⁄⁄ (6.280)

0.039 (2.002) 0.01 (0.087)

0.075⁄⁄⁄ (5.543) 0.203⁄⁄⁄ (6.819)

0.112⁄⁄⁄ (4.829) 0.194⁄⁄⁄ (6.341)

0.037 (2.104) 0.009 (0.065)

0.077⁄⁄⁄ (5.603) 0.203⁄⁄⁄ (6.798)

0.116⁄⁄⁄ (4.735) 0.194⁄⁄⁄ (6.299)

0.039 (2.145) 0.009 (0.069)

0.021 (1.619) 0.020 (1.626)

0.005 (0.284) 0.031⁄⁄ (2.543)

0.016 (0.423) 0.01 (0.362) 0.021⁄⁄⁄ (3.558)

0.012 (1.228)

0.009 (0.696) 0.012 (0.899) 0.030⁄⁄ (2.497) 0.007⁄⁄⁄ (4.806) 0.010⁄⁄⁄ (3.798) 2.523⁄⁄⁄ (5.191) 0.430⁄⁄⁄ (15.165) 15,372 0.313

0.001 (0.048) 0.028⁄ (1.810) 0.001 (1.015) 0.025⁄⁄⁄ (5.111) 1.381⁄⁄ (2.504) 0.595⁄⁄⁄ (14.325)

0.013 (0.301) 0.002 (0.014) 0.006⁄⁄⁄ (8.825) 0.015⁄⁄⁄ (7.603) 1.141 (2.310) 0.165⁄⁄⁄ (13.737)

S1 Rank Low S1 Rank High Log (Size) Log (Age) Exp Ratio PortFlow

0.005⁄⁄⁄ (5.050) 0.011⁄⁄⁄ (5.183) 2.017⁄⁄⁄ (5.678) 0.391⁄⁄⁄ (12.777)

Observations Adj R2

0.005⁄⁄⁄ (5.220) 0.005⁄⁄ (2.169) 0.392 (0.846) 0.439⁄⁄⁄ (15.144) 24,486 0.360

Panel B: Back-end load sample Constant 0.075⁄⁄⁄ 0.110⁄⁄⁄ (5.505) (4.694) Lag (Flows) 0.202⁄⁄⁄ 0.193⁄⁄⁄ (6.801) (6.302) Y1 Rank 0.021⁄⁄⁄ 0.017 (3.058) (1.592) Y1 Rank Low Y1 Rank High S1 Rank S1 Rank Low S1 Rank High Log (Size) Log (Age) Exp Ratio PortFlow Observations Adj R2

0.007⁄⁄⁄ (4.865) 0.010⁄⁄⁄ (3.753) 2.484⁄⁄⁄ (5.040) 0.429⁄⁄⁄ (14.963) 15,372 0.313

0.001 (1.016) 0.025⁄⁄⁄ (5.094) 1.233⁄⁄ (2.238) 0.594⁄⁄⁄ (14.065)

0.006⁄⁄⁄ (9.304) 0.015⁄⁄⁄ (7.607) 1.251⁄ (2.799) 0.165⁄⁄⁄ (13.531)

0.007⁄⁄⁄ (4.855) 0.010⁄⁄⁄ (3.755) 2.483⁄⁄⁄ (5.044) 0.429⁄⁄⁄ (14.962) 15,372 0.313

0.002 (1.103) 0.025⁄⁄⁄ (5.083) 1.313⁄⁄ (2.404) 0.593⁄⁄⁄ (14.089)

0.006⁄⁄⁄ (8.908) 0.015⁄⁄⁄ (7.581) 1.17 (2.375) -0.164⁄⁄⁄ (13.434)

around the median ranking and we can observe that while there is not statistical difference between load and no-load shares in flows reaction for low performing funds, investors’ reaction to past performance for funds in the top half of the ranking is significantly stronger for load with respect to no-load shares. When facing a front load investors are willing to commit only when their subjective assessment of managerial ability is sufficiently high. As a result they respond disproportionately to strong performances.

0.007⁄⁄⁄ (4.848) 0.010⁄⁄⁄ (3.826) 2.493⁄⁄⁄ (5.081) 0.430⁄⁄⁄ (15.172) 15,372 0.313

0.001 (0.917) 0.025⁄⁄⁄ (5.122) 1.297⁄⁄ (2.370) 0.596⁄⁄⁄ (14.289)

0.006⁄⁄⁄ (9.437) 0.015⁄⁄⁄ (7.660) 1.196 (2.618) 0.166⁄⁄⁄ (13.783)

How material is this effect? Some comparative static can help us answer this question. Let’s consider a fund that from the middle of the category ranking increases its performance until it reaches the top 10% of its peers (90th percentile of the ranking), what would be the additional annualized investment flow into this fund? The answer of course depends on whether we consider the load or the no-load fund share of this product. From model 2 of Table 4 Panel A we can estimate that the no-load share of this fund

19

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25

Table 5 Loads and the ‘‘memory’’ of investment flows. The table reports the results of the OLS estimation of Eq. (3) on a sample of matched fund shares. In Panel A the sample considers only funds with at least one no-load share and one share that charges front loads. Where multiple shares in each category are available only the share with the largest AUM in each group is chosen. In Panel B the sample considers only funds with at least one no-load share and one share that charges back-end loads. For each regression the table reports the interaction between the each independent variable and the dummies front load and no load. The third column in each model reports v2-tests for the difference between the two interaction coefficients. The dependent variable is an estimate of quarterly investment flows and the independent variables are fund performance and other characteristics. The performance measure is the fractional ranking of the fund within the investment objective in the previous two semesters (S1 and S2 Rank) or the previous four quarters (Q1,. . ., Q4 Rank). In order to test for non-linearity in the flow-performance relationship we apply a piecewise specification to some of these performance variables using the median ranking (0.5) as the breaking point between High and Low. PortFlow is the average investment flow in the other fund shares of the same fund. T-statistics and Wald statistics in parenthesis. ⁄⁄⁄, ⁄⁄ and ⁄ represent significance at the 1%, 5% and 10% respectively. No load

Load

Diff

(1) Panel A: Front load sample Constant 0.064⁄⁄⁄ (6.327) Lag (Flows) 0.351⁄⁄⁄ (11.674) S1 Rank 0.017⁄⁄⁄ (4.086) S2 Rank 0.004 (1.027) S1 Rank Low

No load

Load

Diff

(2) 0.024⁄⁄⁄ (2.973) 0.267⁄⁄⁄ (10.215) 0.032⁄⁄⁄ (6.801) 0.020⁄⁄⁄ (4.594)

0.04⁄⁄⁄ (10.435) 0.084⁄⁄⁄ (7.660) 0.015⁄⁄⁄ (6.852) 0.015⁄⁄⁄ (7.331)

S1 Rank High S2 Rank Low S2 Rank High

0.065⁄⁄⁄ (6.165) 0.351⁄⁄⁄ (11.672)

0.031⁄⁄⁄ (3.530) 0.266⁄⁄⁄ (10.198)

0.035⁄⁄⁄ (7.449) 0.084⁄⁄⁄ (7.825)

0.015⁄⁄ (2.434) 0.019⁄⁄ (2.072) 0.000 (0.012) 0.009 (1.080)

0.022⁄⁄⁄ (3.127) 0.041⁄⁄⁄ (5.079) 0.003 (0.455) 0.035⁄⁄⁄ (4.617)

0.007 (0.532) 0.023⁄⁄ (3.862) 0.003 (0.098) 0.026⁄⁄ (5.095)

Q1 Rank Q2 Rank Q3 Rank Q4 Rank Log (Size) Log (Age) Exp Ratio PortFlow

0.005⁄⁄⁄ (5.067) 0.011⁄⁄⁄ (5.212) 2.003⁄⁄⁄ (5.657) 0.391⁄⁄⁄ (12.758)

Observations Adj R2

0.005⁄⁄⁄ (5.127) 0.005⁄⁄ (2.251) 0.324 (0.685) 0.438⁄⁄⁄ (15.200) 24,486 0.361

0 (0.005) 0.006⁄⁄ (5.966) 1.68⁄⁄⁄ (7.091) 0.048 (2.236)

(3) Panel B: Back-end load sample Constant 0.074⁄⁄⁄ (5.267) Lag (Flows) 0.203⁄⁄⁄ (6.813) S1 Rank 0.021⁄⁄⁄ (3.619) S2 Rank 0.004 (0.691) S1 Rank Low

0.005⁄⁄⁄ (5.069) 0.011⁄⁄⁄ (5.208) 2.019⁄⁄⁄ (5.680) 0.390⁄⁄⁄ (12.796)

0.005⁄⁄⁄ (5.232) 0.005⁄⁄ (2.155) 0.502 (1.063) 0.438⁄⁄⁄ (15.197) 24,486 0.361

0 (0.000) 0.006⁄⁄ (6.341) 1.518⁄⁄ (5.927) 0.047 (2.211)

(4) 0.111⁄⁄⁄ (4.566) 0.194⁄⁄⁄ (6.315) 0.011 (1.317) 0.004 (0.513)

0.037 (1.881) 0.009 (0.067) 0.01 (0.981) 0.001 (0.006)

S1 Rank High S2 Rank Low S2 Rank High

0.074⁄⁄⁄ (5.110) 0.203⁄⁄⁄ (6.794)

0.118⁄⁄⁄ (4.530) 0.192⁄⁄⁄ (6.241)

0.043 (2.357) 0.01 (0.085)

0.009 (0.752) 0.032⁄⁄⁄ (2.964) 0.012 (1.048) 0.004 (0.443)

0.002 (0.147) 0.023 (1.501) 0.011 (0.747) 0.023⁄⁄ (2.028)

0.007 (0.106) 0.009 (0.244) 0.023 (1.427) 0.027⁄ (2.845)

Q2 Rank Q3 Rank Q4 Rank 0.007⁄⁄⁄ (4.854)

0.001 (0.938)

0.006⁄⁄⁄ (9.395)

0.007⁄⁄⁄ (4.821)

Load

Diff

0.061⁄⁄⁄ (6.053) 0.352⁄⁄⁄ (11.719)

0.018⁄⁄ (2.107) 0.268⁄⁄⁄ (10.242)

0.043⁄⁄⁄ (11.991) 0.084⁄⁄⁄ (7.568)

0.021⁄⁄⁄ (5.705) 0.002 (0.663) 0.005 (1.505) 0.002 (0.520) 0.005⁄⁄⁄ (5.053) 0.011⁄⁄⁄ (5.226) 1.996⁄⁄⁄ (5.572) 0.391⁄⁄⁄ (12.855)

0.027⁄⁄⁄ (6.057) 0.013⁄⁄⁄ (3.560) 0.017⁄⁄⁄ (4.179) 0.010⁄⁄⁄ (2.869) 0.005⁄⁄⁄ (5.058) 0.005⁄⁄ (2.363) 0.308 (0.644) 0.439⁄⁄⁄ (15.296) 24,486 0.361

0.007 (1.339) 0.01⁄⁄ (5.574) 0.011⁄⁄ (4.671) 0.012⁄⁄ (6.424) 0 (0.008) 0.006⁄⁄ (5.487) 1.688⁄⁄⁄ (7.031) 0.048 (2.305)

0.070⁄⁄⁄ (4.909) 0.203⁄⁄⁄ (6.802)

0.111⁄⁄⁄ (4.413) 0.194⁄⁄⁄ (6.373)

0.041 (2.142) 0.008 (0.060)

0.020⁄⁄⁄ (3.930) 0.011⁄ (1.898) 0.001 (0.156) 0.004 (0.757) 0.007⁄⁄⁄ (4.867)

0.000 (0.004) 0.007 (0.872) 0.002 (0.370) 0.006 (0.803) 0.001 (0.910)

0.02⁄⁄⁄ (7.543) 0.004 (0.198) 0.003 (0.118) 0.001 (0.028) 0.006⁄⁄⁄ (9.678)

(5)

Q1 Rank

Log (Size)

No load (3)

0.002 (1.116)

0.006⁄⁄⁄ (8.576)

(continued on next page)

20

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25

Table 5 (continued) (3) Log (Age) Exp Ratio PortFlow Observations Adj R2

0.010⁄⁄⁄ (3.817) 2.473⁄⁄⁄ (5.013) 0.430⁄⁄⁄ (14.968) 15,732 0.313

(4) 0.025⁄⁄⁄ (5.107) 1.263⁄⁄ (2.334) 0.595⁄⁄⁄ (14.064)

0.015⁄⁄⁄ (7.603) 1.21 (2.616) 0.166⁄⁄⁄ (13.584)

0.010⁄⁄⁄ (3.804) 2.488⁄⁄⁄ (5.092) 0.430⁄⁄⁄ (14.968) 15,732 0.313

would experience an additional annualized inflow equal to 3.91% of its asset under management, while the differential flow for the corresponding load share would be equal to 8.13%. As we can see the increase in convexity generates a much stronger response to this improved perceived managerial ability. So far we have demonstrated that investment flows into fund shares that charge front loads exhibit a stronger convexity in the flow-performance relationship. While this is coherent with our hypothesis that upfront unrecoverable costs generate a nonreversibility of investment choices and as a consequence investors want a stronger confirmation of managerial ability, we cannot completely rule out an alternative hypothesis that this effect is actually due to different characteristics of investors in load and no-load funds. Reid and Rea (2003) show how loads are used to pay for advice in the fund selection process, and in their survey on mutual fund shareholders Bogdan and Schrass (2013) show that investors who buy mutual funds through Sales Force Channel have, for example, a slightly lower education level than investors who buy through direct channels. Houge and Wellman (2007) also argue that load share classes cater to a less sophisticated clientele. One could thus argue that for load funds advisers and broker pay a key role in the decision process and the difference in the flow-performance relationship could simply reflect different incentives or a different level of sophistication rather than the irreversibility of the investment decision induced by the load. We try to disentangle these alternatives looking at a counterfactual example of fund shares with back-end loads. According to Reid and Rea (2003) investors can also pay for advice through a combination of 12b-1 fees and back-end loads (usually ‘‘B’’ shares have this arrangement). If the effect documented above for share classes with front loads is caused by the presence of a broker in the decision process we should observe a similar effect for share classes with back-end loads. On the other side if the effect is due to a behavioral modification induced by the unrecoverable nature of the load we would expected to be much weaker for back-end load due to the much lower salience (Barber et al., 2005). In Panel B of Table 4 we repeat our analysis for our back-end load subsamples that only consider funds with no-load and backend load shares. As we can see while we still observe a lower price-sensitivity for load funds, we no longer observe any difference in performance sensitivity. This counterfactual analysis show that the increase in the convexity of the flow-performance relationship is not a result of different incentives or sophistication of the decision maker (broker/adviser vs individual investor) but likely stems from the unrecoverable nature of front loads (coupled with their greater salience).

(5) 0.025⁄⁄⁄ (5.095) 1.418⁄⁄ (2.566) 0.594⁄⁄⁄ (14.108)

0.015⁄⁄⁄ (7.553) 1.07 (1.937) 0.164⁄⁄⁄ (13.375)

0.011⁄⁄⁄ (3.886) 2.449⁄⁄⁄ (4.964) 0.429⁄⁄⁄ (15.034) 15,732 0.313

0.025⁄⁄⁄ (5.106) 1.295⁄⁄ (2.378) 0.596⁄⁄⁄ (14.054)

0.015⁄⁄⁄ (7.537) 1.155 (2.323) 0.167⁄⁄⁄ (13.714)

performance measure into two semiannual or four quarterly measures. In Panel A we look at our front load sample and we can see in models 1 and 3 that while for investors in no-load funds only the performance in the most recent semester is relevant, investments in load funds also responds to older performance. In model 2 we split the two six-month performances into high and low using our piecewise specification and show that this ‘‘memory’’ effect is not byproduct of the increase in convexity previously analyzed, both effects coexist and are thus, at least partially, independent. When investment is easily reversible (no entry cost) flows respond quickly to short term performance, while when reversibility is hampered by non-zero entry costs in the form of front loads, flows have a longer memory and investors commit only when a fund manager can show a stable track record of good performance. Again we can use comparative static to gauge how material this behavioral modification is. This time we can consider two mutual funds that in a given semester are both in the top 10% of their category ranking. Where they differ is that while the first fund (that we will call ‘‘consistent’’) was in the top 10% also in the previous semester the second fund (that we will call ‘‘inconsistent’’) was, in the previous semester, in the bottom 10% of the ranking. The good performance of the consistent manager is less likely to be fruit of pure luck, we have stronger signal of managerial ability. What would be the expected investment flow in the two funds given what the investors know about the ability of the two fund managers? Investors in the no-load fund shares will not give much consideration to the older performance and would treat the two funds in very similar way: the investment flow would be 3.11% (of the asset under management) in the inconsistent fund and 4.42% in the consistent one. On the other side investors in the load fund shares face a non-reversible investment decision and are thus much more cautious: the investment flow would be 4.68% in the inconsistent fund but 11.46% in the consistent fund. As we can see investors who face upfront unrecoverable costs exhibit a longer memory. Again we could argue that what we are capturing are simply different characteristics of the different decision makers (brokers and advisors for the load fund shares versus individual investors for the no-load shares). As before we will use fund shares with back-end loads as a counterexample. In Panel B of Table 5 we can observe that while investors in load funds now exhibit generally weaker performance sensitivity there is no differential sensitivity to performance in periods preceding the last one.9 Concluding we can say that the presence of front loads significantly affect investors’ behavior. Investors who buy no-load fund shares can easily switch between investment products and thus may be optimal for them to buy shares in a fund where they have moderately good expectations on managerial ability, knowing that

4.2. Investment flows memory Our second hypothesis is that investors will wait for a longer series of positive returns before committing to a new fund when this commitment involves an upfront sunk cost in the form of a load. Table 5 explores this problem by decomposing our annual

9 With only one exception when we combine multiple periods with the piecewise specification to capture the non-linear relationship between flow and performance. In this case flows in load fund shares exhibit stronger sensitivity to older/High performance (S2 Rank High). The fact that this result is not confirmed when we do not consider the piecewise specification tells us that it could be a by-product of the convexity in the flow-performance relationship.

15.78 15.31 0.00 0.00 0.00 0.60 3.88 3.60 0.20 Panel D: No-load fund shares Time weighted return 5.31 Dollar weighted return 4.21 Timing ability 1.08

16.25 16.12 2.40

0.29 0.10 1.43

11.21 10.33 0.00

8.24 7.63 0.60

3.28 3.19 0.05

30.97 30.63 1.51

23.25 22.53 0.02 6.65 7.27 0.94 5.84 5.21 0.74 Panel C: Back-end load fund shares Time weighted return 6.29 Dollar weighted return 4.78 Timing ability 1.51

20.26 20.11 2.54

1.43 2.66 2.06

15.95 14.65 0.14

9.92 9.06 0.82

5.67 5.31 0.26

36.44 36.12 1.63

20.95 20.38 0.02 4.90 5.35 0.80 5.59 5.15 0.54 Panel B: Front load fund shares Time weighted return 6.29 Dollar weighted return 5.02 Timing ability 1.28

18.67 18.60 2.31

0.54 1.56 1.76

14.40 13.36 0.11

9.58 8.83 0.72

5.75 5.48 0.20

34.23 33.90 1.51

19.50 18.89 0.00 3.08 3.52 0.76 4.24 4.05 0.15 9.03 8.31 0.70 13.50 12.32 0.02 0.05 0.00 1.74 18.15 18.02 2.43 4.64 4.31 0.45 5.77 4.50 1.26 Panel A: Whole sample Time weighted return Dollar weighted return Timing ability

25th Percentile (%) Standard deviation (%) Median (%)

33.53 33.21 1.55

75th Percentile (%) 25th Percentile (%) Standard deviation (%) Mean (%)

Median (%)

Quarterly measures

Mean (%)

75th Percentile (%) Yearly measures

Table 6 Descriptive statistics of investors’ timing ability. The table reports descriptive statistics for time weighed and dollar weighted rates of returns and investors’ timing ability defined as the difference between dollar-weighted and timeweighted return. Measures are calculated from monthly performances and flows and aggregated at the quarterly or yearly level. All performances are annualized for sake of comparison. Descriptive statistics are provided for the whole sample and for subsamples of funds with different load arrangements.

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25

21

if at a later stage these expectations should prove unfounded they will leave the fund and look for new opportunities. Investors in fund shares that charge a front load have to pay an upfront price that will likely make this switching strategy unprofitable. As a result they will be willing to invest only in funds where performance conveys a stronger signal of managerial ability through a particularly high performance or a longer series of positive outcomes. In terms of flow-performance relationship this generates a higher degree of convexity and a longer memory.

5. Loads and timing ability So far we have shown that front loads affect investor’s behavior, now we will analyze whether these changes significantly affect investors’ timing ability. Gruber (1996) argues that investors’ timing ability in the mutual fund market could justify investment in underperforming actively managed mutual funds. Our goal here is to test whether investors in fund shares with front loads exhibit higher or lower timing ability than investors in no-load fund shares. Following Friesen and Sapp (2007) we measure investors’ timing ability as the difference between dollar-weighted return and the time-weighted return for a specific mutual fund. The dollarweighted return, derived as the internal rate of return of assets under management, measures the performance of investors’ flows in and out a specific mutual fund, while the time-weighted returns measures the performance of the fund itself and is not affected by investors’ timing decisions. The time-weighted return measures the performance of a dollar invested in the mutual fund for the entire period, a buy and hold strategy of an investor who refrains from any timing decision. On the other side, a dollar-weighted return explicitly accounts for investment flows over time, reflecting the average investor’s performance during the sample period. We estimate this measure using monthly investment flows and fund returns and calculate the internal rate of returns over non overlapping quarters or years. With respect to other methodologies used in literature (see for example Zheng, 1999 and Sapp and Tiwari, 2004) this solution has the distinctive advantage of allowing for different holding periods among funds. While this may not have been a key problem in previous research10 it is when it comes to confronting load and no-load fund shares. O’Neal (1999) shows that fund shares with different load arrangements cater to investors with different investment horizon. In this context assuming that ‘‘new money’’ is invested in the fund for a single period would be an unacceptable simplification. When we calculate the dollar-weighted rate of return of the fund funds invested in one period will remain invested (and will learn the respective return) until offsetted by a money outflow of the same size. Coherently with Friesen and Sapp (2007) we find that, on average, investors exhibit negative timing ability. Descriptive statistics in Table 6 show that the difference between a buy and hold strategy and the performance of actual investment flows is 1.26% when measured over yearly periods and 0.70% (annualized) when measured over quarterly periods. As we can see only around 25% of the observations (funds/period) show positive timing ability. These results are quite stable across subsamples of fund shares with different load arrangements. In Fig. 1 we plot the asset-weighted average timing ability for all the fund shares in our sample. As we can see there is a general market trend in this variable. 10 Both Zheng (1999) and Sapp and Tiwari (2004) look at different returns for funds that experience positive flows vs funds with negative investment flows. There is no reason to assume that investors in these two groups of financial products may have different investment horizons.

22

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25

Yearly Timing Ability

2011q1

2010q1

2009q1

2008q1

2007q1

2006q1

2005q1

2004q1

2003q1

2002q1

2001q1

2000q1

1999q1

0.0% -0.5% -1.0% -1.5% -2.0% -2.5% -3.0% -3.5%

Quarterly Timing Ability

Fig. 1. Investors’ timing ability. The figure reports the time series of the asset weighted timing ability for investors in the US mutual fund market.

Table 7 Timing ability and funds characteristics. The table reports the average value of time-weighted and dollar-weighted total returns for mutual funds in our sample sorted in quintile portfolios according to fund size, age, expense ratio (in excess over the category average) and past performance (category fractional ranking over the previous year). Returns are measured on quarterly (Panel A) or annual (Panel B) basis. For each characteristic the table reports results of a t-test on the significance of the difference between the first and the last quintile. Standard errors parenthesis. ⁄⁄⁄, ⁄⁄ and ⁄ represent significance at the 1%, 5% and 10% respectively. Size DWRR

Age

Expense ratio

Past Performance

TWRR

Diff

DWRR

TWRR

Diff

DWRR

TWRR

Diff

DWRR

TWRR

Diff

Panel A: Quarterly returns Q1 10.33% Q2 9.74% Q3 9.09% Q4 8.55% Q5 7.50% Q5  Q1 2.83%⁄⁄⁄ (0.002)

11.24% 10.53% 9.87% 9.29% 8.19% 3.05%⁄⁄⁄ (0.002)

0.82% 0.74% 0.71% 0.68% 0.64% 0.18%⁄⁄⁄ (0.000)

10.17% 9.08% 9.09% 8.66% 8.15% 2.03%⁄⁄⁄ (0.002)

11.20% 9.90% 9.80% 9.39% 8.77% 2.43%⁄⁄⁄ (0.002)

0.89% 0.75% 0.70% 0.69% 0.57% 0.32%⁄⁄⁄ (0.000)

11.08% 9.02% 7.98% 7.94% 9.12% 1.95%⁄⁄⁄ (0.002)

12.03% 9.71% 8.64% 8.68% 9.98% 2.05%⁄⁄⁄ (0.002)

0.87% 0.66% 0.61% 0.66% 0.78% 0.1%⁄⁄⁄ (0.000)

8.74% 8.41% 8.68% 9.30% 10.08% 1.33%⁄⁄⁄ (0.002)

9.63% 9.11% 9.39% 10.02% 10.97% 1.34%⁄⁄⁄ (0.002)

0.77% 0.68% 0.68% 0.69% 0.78% 0.01% (0.000)

Panel B: Yearly returns Q1 5.27% Q2 5.23% Q3 4.78% Q4 4.36% Q5 3.26% Q5  Q1 2.01%⁄⁄⁄ (0.002)

6.75% 6.54% 6.07% 5.56% 4.41% 2.34%⁄⁄⁄ (0.002)

1.43% 1.30% 1.25% 1.19% 1.12% 0.31%⁄⁄⁄ (0.000)

4.52% 4.08% 5.11% 4.48% 4.78% 0.27%⁄ (0.002)

6.04% 5.37% 6.37% 5.71% 5.79% 0.25% (0.002)

1.47% 1.27% 1.24% 1.21% 0.99% 0.48%⁄⁄⁄ (0.000)

5.72% 5.03% 4.22% 3.82% 4.08% 1.65%⁄⁄⁄ (0.002)

7.20% 6.16% 5.30% 5.05% 5.60% 1.6%⁄⁄⁄ (0.002)

1.44% 1.11% 1.04% 1.19% 1.50% 0.06%⁄⁄⁄ (0.000)

4.23% 4.23% 4.75% 4.93% 4.87% 0.64%⁄⁄⁄ (0.002)

5.71% 5.47% 5.93% 6.19% 6.24% 0.53%⁄⁄⁄ (0.002)

1.44% 1.21% 1.15% 1.22% 1.33% 0.1%⁄⁄⁄ (0.000)

Table 8 Loads and timing ability (quarterly returns). The table reports the average value of the difference between (annualized) quarterly dollar-weighted and time-weighted total returns for load and no-load funds in our sample sorted in two portfolios according to fund size, age, expense ratio (in excess over the category average) and past performance (category fractional ranking over the previous year). For each characteristic the breaking point between the two portfolios is the sample median value. The last two columns report t-tests on the difference between the mean value of each of the two load subsamples and the no-load subsample. ⁄⁄⁄, ⁄⁄ and ⁄ represent significance at the 1%, 5% and 10% respectively. No load (%)

Front load (%)

Back-end load (%)

Front load – No load (%)

Back-end load – No load (%)

Panel A: Fund size Small Large

0.77 0.54

0.67 0.68

0.79 0.87

0.11⁄⁄⁄ 0.14⁄⁄⁄

0.02⁄⁄ 0.33⁄⁄⁄

Panel B: Fund age Young Old

0.76 0.51

0.76 0.61

0.83 0.80

0 0.1⁄⁄⁄

0.07⁄⁄⁄ 0.29⁄⁄⁄

Panel C: Expense ratio Cheap Expensive

0.70 0.51

0.57 0.81

0.98 0.76

0.14⁄⁄⁄ 0.3⁄⁄⁄

0.28⁄⁄⁄ 0.25⁄⁄⁄

Panel D: Past performance Low Performance High Performance

0.67 0.61

0.70 0.65

0.74 0.92

0.03⁄⁄⁄ 0.04⁄⁄⁄

0.08⁄⁄⁄ 0.31⁄⁄⁄

23

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25

Table 9 Loads and timing ability (yearly returns). The table reports the average value of the difference between yearly dollar-weighted and time-weighted returns for load and no-load funds in our sample sorted in two portfolios according to fund size, age, expense ratio (in excess over the category average) and past performance (category fractional ranking over the previous year). For each characteristic the breaking point between the two portfolios is the sample median value. The last two columns report t-tests on the difference between the mean value of each of the two load subsamples and the no-load subsample. ⁄⁄⁄, ⁄⁄ and ⁄ represent significance at the 1%, 5% and 10% respectively. No load (%)

Front load (%)

Back-end load (%)

Front load – No load (%)

Back-end load – No load (%)

Panel A: Fund size Small Large

1.30 0.90

1.13 1.21

1.45 1.61

0.17⁄⁄⁄ 0.31⁄⁄⁄

0.15⁄⁄⁄ 0.71⁄⁄⁄

Panel B: Fund age Young Old

1.22 0.88

1.30 1.07

1.51 1.50

0.08⁄⁄⁄ 0.18⁄⁄⁄

0.3⁄⁄⁄ 0.62⁄⁄⁄

Panel C: Expense ratio Cheap Expensive

1.15 0.91

0.95 1.44

1.69 1.44

0.2⁄⁄⁄ 0.53⁄⁄⁄

0.54⁄⁄⁄ 0.54⁄⁄⁄

Panel D: Past performance Low Performance High Performance

1.14 1.00

1.25 1.09

1.41 1.64

0.11⁄⁄⁄ 0.1⁄⁄⁄

0.26⁄⁄⁄ 0.64⁄⁄⁄

Table 10 Loads and timing ability (multivariate analysis). The table reports the results of a linear model where the dependent variable is the difference between money-weighted and timeweighted total returns funds in our two matched subsamples. The front load sample considers only funds with at least one no-load share and one share that charges front loads. Where multiple shares in each category are available only the share with the largest AUM in each group is chosen. The back-end load sample considers only funds with at least one no-load share and one share that charges back-end loads. Where multiple shares in each category are available only the share with the largest AUM in each group is chosen. Load is a dummy variable equal to one if the fund share charges the relevant load. Flow is the quarterly (or annual) investment flow (in percentage over the AUM of the fund) and Flow2 is its squared value. PortDiff is the average difference between money-weighted and time-weighted total returns in the other fund shares of the same fund. T-statistics in parenthesis. ⁄⁄⁄, ⁄⁄ and ⁄ represent significance at the 1%, 5% and 10% respectively. Front load sample

Back-end load sample

Quarterly returns

Constant Load Flow Flow2 Log (Size) Log (Age) Exp Ratio

Quarterly returns

(2a)

(3a)

(4a)

(1b)

(2b)

(3b)

(4b)

0.005⁄⁄⁄ (2.759) 0.001⁄⁄ (2.455) 0.000 (0.016) 0.021 (1.174) 0.000 (0.001) 0.000 (0.722) 0.205⁄⁄ (2.557)

0.009⁄⁄⁄ (4.453) 0.001⁄⁄⁄ (3.146) 0.001 (0.112) 0.017 (1.039) 0.000 (1.532) 0.000 (0.624) 0.238⁄⁄⁄ (2.898) 0.062⁄⁄⁄ (4.535) Yes 24,469 0.098

0.010⁄ (1.863) 0.002⁄⁄ (2.141) 0.001 (0.053) 0.000 (0.072) 0.000 (0.627) 0.001 (1.014) 0.503⁄⁄⁄ (2.992)

0.018⁄⁄⁄ (2.701) 0.002⁄⁄⁄ (2.655) 0.000 (0.031) 0.000 (0.013) 0.001 (1.605) 0.001 (0.709) 0.515⁄⁄⁄ (2.898) 0.053⁄⁄⁄ (3.480) Yes 5868 0.097

0.013⁄⁄⁄ (5.092) 0.001⁄⁄⁄ (3.274) 0.002 (0.196) 0.017 (1.367) 0.001⁄⁄⁄ (3.496) 0.000 (0.639) 0.034 (0.493)

0.014⁄⁄⁄ (6.062) 0.001⁄⁄ (2.263) 0.003 (0.337) 0.014 (1.204) 0.001⁄⁄⁄ (4.202) 0.001 (1.144) 0.126⁄ (1.836) 0.079⁄⁄⁄ (4.182) Yes 15,715 0.148

0.021⁄⁄⁄ (4.059) 0.002⁄⁄ (2.021) 0.000 (0.072) 0.001 (0.731) 0.002⁄⁄ (2.325) 0.000 (0.266) 0.257 (1.524)

0.023⁄⁄⁄ (4.422) 0.002⁄⁄ (2.202) 0.000 (0.045) 0.001 (0.787) 0.002⁄⁄ (2.452) 0.000 (0.019) 0.332⁄ (1.930) 0.028 (1.328) Yes 3620 0.067

Yes 24,473 0.009

Yes 5984 0.008

Specifically we can observe sharp drops in timing ability during the burst of the tech bubble and during the recent financial crisis. In Table 7 we look at investors’ timing ability in funds with different characteristics. We see that investment flows in larger and older funds exhibit better timing ability. Size and age have often been used as proxies for information availability,11 and our results are coherent with this view: when information about a fund is more readily available investors can make more timely investment decisions. We do not find any strong and stable difference in timing ability in funds with different expense ratios and different past performance. As a first test of the effect on loads on timing ability we look at the difference between average time ability between load and no-

11

Annual returns

(1a)

PortDiff Time F.E. Observations Adj R2

Annual returns

See for example Huang et al. (2007), Iannotta and Navone (2012) and Navone (2012).

Yes 15,717 0.015

Yes 3693 0.024

load funds with different characteristics. Results for quarterly returns (Table 8) and yearly returns (Table 9) show that in general investors in load fund exhibit lower timing ability than investors in no-load funds, with the only exception being investment flows for front loads funds in the subsamples of small funds (size below that sample median), young funds (age below the sample median) and cheap funds (normalized expense ratio below the sample median). These results seem to point to a negative impact of loads on timing ability, with maybe the exception of funds with low visibility where information acquisition for retail investors may be more expensive.12 As we have seen before a major drawback is that it’s hard to account for portfolio differences among the subsample of load and no load funds: if load and no-load funds cater to investors 12 Huang et al. (2007), Iannotta and Navone (2012) and Navone (2012) use size and age as proxies or information availability. Sirri and Tufano (1998) consider expense ratio as a proxy for marketing expenses and thus information availability.

24

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25

with different sophistication there may be significant differences in portfolio composition and these, in turn, may affect returns predictability and timing. In order to properly address this issue we go back to our samples of matched load and no-load share and regress our timing ability variable over fund characteristics (Size, Age and Exp Ratio as defined in the previous paragraph) and a dummy variable for fund shares with loads (Load). Since our timing variable is mechanically related to the intensity of investment flows (when there are no inflows or outflow the dollar-weighted return is by construction equal to the time-weighted return) we also consider on the righthand side of our model investment flows and their squared value (Flow and Flow2). Finally as we did before in order to capture portfolio level effects we also include the weighted average timing ability for the other share classes of the same fund (PortTiming). Given the non-negligible time trends in timing ability as depicted in Fig. 1, we also include time fixed effects. In an unreported regression we show that their exclusion would not affect the results in any material way. Results in Table 10 paint a different picture, indicating again that portfolio differences can indeed affect our perception of the effect of load on investors’ behavior. Specifically what we observe is that front loads have a negative impact on investors timing ability. The increase in the convexity of the flow-performance relationship and the increase in ‘‘memory’’ seem to worsen the ability of investors to time investment flows. How relevant is this effect? From models 2 and 4 we see that investment flows into fund shares with front loads have timing ability that is 19 basis points lower on an annual scale (8 basis points on the quarterly measures) lower than flows into the no-load fund share of the same fund. The effect is clearly material if we consider that the sample average timing ability is 1.26% on yearly basis and 0.70% on a quarterly basis. Again one may argue that two different factors may generate this effect: on the one side our main hypothesis is that the upfront unrecoverable cost represented by the front load induces a change in the relationship between past performance and investment decision, on the other side if front load is an indication that a broker is intervening in the decision process the results could simply imply that brokers have worst timing ability than retail investors. As we did before we can address this issue by looking at funds with no-load and back-end load share classes. Here we have a broker involved in the decision process but a far less salient unrecoverable investment cost. From Table 10 we see that this time the load is associated with an increase in timing ability of 13 bps on a yearly scale (7 bps on a quarterly scale). It would be beyond the scope of this work to speculate on the relative timing ability of brokers vs individual investors, what is relevant here is that clearly we cannot attribute the negative timing effect of front loads to the simple presence of a broker: if this were true we would observe the same effect for share classes with back-end loads. The fact that for these shares the effect of loads on timing ability is actually positive lends credit to our main hypothesis that front loads directly affect investment behavior through their nature of salient upfront unrecoverable costs. 6. Conclusion In this paper we look at the effect of loads on investors’ behavior. We argue that loads make the investment in mutual fund shares partially irreversible. As a consequence investors will wait for a stronger signal of managerial ability before committing to specific mutual fund. This stronger signal may come in the form of a particularly good performance or of a long series of positive outcomes. In order to test this hypothesis we build a sample of matched share classes from the same fund but with different fee

arrangements. We empirically show that, coherently with our hypothesis, investment flows into share classes with front loads exhibit a more convex flow-performance relationship and a longer memory, with investment flows reacting also by performances in periods prior to the last one. The fact that we do not see the same effect for share classes with back-end loads confirm that these effects are due to the salient nature of front loads as unrecoverable investment costs rather than to the involvement of a broker in the decision process. Finally we show that these effects have a negative and significant impact on investor’s timing ability. While we acknowledge that load and no-load fund shares may cater to investors with different levels of sophistication, and that that this may lead to differences in the demand function, the structure of our experiment and the robustness checks performed seem to indicate that the empirical effects documented in this paper do not purely reflect heterogeneity among investors but a real, and material, effect of the fee structure on investors’ behavior. This evidence has important implications both for fund management companies and for regulators, as it highlights an additional dimension that has to be considered in the determination of the optimal fee structure. Acknowledgements Marco Navone would like to acknowledge financial support received from CAREFIN, the Center for Applied Research in Finance of Bocconi University. References Bailey, W., Kumar, A., Ng, D., 2011. Behavioral biases of mutual fund investors. Journal of Financial Economics 102 (1), 1–27. Barber, B., Odean, T., Zheng, L., 2005. Out of sight, out of mind: the effects of expenses on mutual fund flows. Journal of Business 78 (6), 2095–2120. Bergstresser, D., Chalmers, J., Tufano, P., 2009. Assessing the costs and benefits of brokers in the mutual fund industry. Review of Financial Studies 22 (10), 4129– 4156. Bogdan, M., Schrass, D., 2013. Profile of Mutual Fund Shareholders, 2012, ICI Research Report. Brown, K., Harlow, W., Starks, L., 1996. Of tournaments and temptations: an analysis of managerial incentives in the mutual fund industry. Journal of Finance 51 (1), 85–110. Carhart, M., 1997. On persistence in mutual fund performance. Journal of Finance 52 (1), 57–82. Christoffersen, S., Musto, D., 2002. Demand curves and the pricing of money management. Review of Financial Studies 15 (5), 1499–1524. Christoffersen, S., Evans, R., Musto, D., 2013. What do consumers’ fund flows maximize? Evidence from their brokers’ incentives. Journal of Finance 68 (1), 201–235. Del Guercio, D., Reuter, J., Tkac, P., 2010. Broker incentives and mutual fund market segmentation, Working paper, National Bureau of Economic Research. Evans, R., 2010. Mutual fund incubation. Journal of Finance 65 (4), 1581–1611. Friesen, G., Sapp, T., 2007. Mutual fund flows and investor returns: an empirical examination of fund investor timing ability. Journal of Banking & Finance 31 (9), 2796–2816. Goetzmann, W., Peles, N., 1997. Cognitive dissonance and mutual fund investors. Journal of Financial Research 20 (2), 145–158. Gruber, M., 1996. Another puzzle: the growth in actively managed mutual funds. Journal of Finance 51 (3), 783–810. Hendricks, D., Patel, J., Zeckhauser, R., 1993. Hot hands in mutual funds: short-run persistence of relative performance, 1974–1988. Journal of Finance 48 (1), 93–130. Hortaçsu, A., Syverson, C., 2004. Product differentiation, search costs, and competition in the mutual fund industry: a case study of S&P 500 index funds. Quarterly Journal of Economics 119 (2), 403–456. Houge, T., Wellman, J., 2007. The use and abuse of mutual fund expenses. Journal of Business Ethics 70 (1), 23–32. Huang, J., Wei, K., Yan, H., 2007. Participation costs and the sensitivity of fund flows to past performance. Journal of Finance 62 (3), 1273–1311. Kahneman, D., Tversky, A., 1979. Prospect theory: an analysis of decision under risk. Econometrica 47 (2), 263–291. Kempf, A., Ruenzi, S., 2008. Tournaments in mutual fund families. Review of Financial Studies 21 (2), 1013–1036. Kempf, A., Ruenzi, S., Thiele, T., 2009. Employment risk, compensation incentives, and managerial risk taking: evidence from the mutual fund industry. Journal of Financial Economics 92 (1), 92–108. Khorana, A., Servaes, H., Tufano, P., 2009. Mutual fund fees around the world. Review of Financial Studies 22 (3), 1279–1310.

M. Navone, M. Pagani / Journal of Banking & Finance 51 (2015) 12–25 Iannotta, G., Navone, M., 2012. The cross-section of mutual fund fees dispersion. Journal of Banking & Finance 36 (3), 846–856. Investment Company Institute, 2013, 2013 Investment Company Fact Book, Available online at . Lemeunier, S., 2011. On the origins of a Conflict of Interest in the Mutual Fund Industry, Working paper, ESSEC. Nanda, V., Wang, Z., Zheng, L., 2009. The ABCs of mutual funds: on the introduction of multiple share classes. Journal of Financial Intermediation 18 (3), 329–361. Navone, M., 2012. Investors’ distraction and strategic repricing decisions. Journal of Banking & Finance 36 (5), 1291–1303. O’Neal, E., 1999. Mutual fund share classes and broker incentives. Financial Analysts Journal 55 (5), 76–87. Pindyck, R., 1991. Irreversibility, uncertainty, and investment. Journal of Economic Literature 29 (3), 1110–1148.

25

Pozen, R., Hamacher, T., 2011. The Fund Industry: How Your Money Is Managed. John-Wiley and Sons Inc., Hoboken, NJ. Reid, B., Rea, J., 2003. Mutual fund distribution channels and distribution costs, Perspectives (Investment Company Institute), 9 (3). Sapp, T., Tiwari, A., 2004. Does stock return momentum explain the ‘‘smart money’’ effect? Journal of Finance 59 (6), 2605–2622. Sirri, E., Tufano, P., 1998. Costly search and mutual fund flows. Journal of Finance 53 (5), 1589–1622. Thaler, R., 1980. Toward a positive theory of consumer choice. Journal of Economic Behavior and Organization 1 (1), 39–60. Zhao, X., 2008, Conflict of interests between load fund investors and brokers and financial advisors, Working paper, China Europe International Business School. Zheng, L., 1999. A study of mutual fund investors’ fund selection ability. Journal of Finance 54 (3), 901–933.