Marco Navone 1 Università Bocconi CAREFIN. This Draft: September Abstract

Transcription

1 INVESTORS DISTRACTION AND STRATEGIC RE-PRICING DECISIONS Marco Navone Università Bocconi CAREFIN This Draft: September 009 Abstract In this paper, I analyze re-pricing decisions for mutual fund management services. I derive measures of performance and price sensitivity and show that investors do not consider expense ratios simply as a negative component of expected returns: while performance sensitivity monotonically increases with past performance, price-sensitivity does not. Investors that buy top pastperformers seems to be distracted by the fund previous return and pay relative little attention to expense ratios. Moreover price-sensitivity increases with fund visibility while performance sensitivity decreases, and while looking at data from 980 to 006 no discernible trend can be observed in the average performance sensitivity, price sensitivity strongly increases after 990 due to the dramatic increase in the availability of mutual funds information for retail investors. Finally I show that investment companies strategically time their re-pricing decisions in order to exploit time variations in price and performance sensitivities. Jel codes: G, G3 Key words: mutual funds, expense ratios, price sensitivity Introduction Pricing in the mutual fund industry is rapidly emerging as a new puzzle in finance. Several papers document a significant level of price dispersion for homogeneous groups of funds (see Christoffersen and Musto (00) on money market funds and Hortaçsu and Syverson (004) on S&P 500 index funds), while others document a negative relationship between fees and gross risk-adjusted performance (Gil-Bazo and Ruiz-Verdú 009). Recent contributions have shed some light on these anomalies: for example we now know that, like in any other industry, search costs play a significant role in generating non-optimal pricing poli- Contact information: Finance Department Università Bocconi, Via Roentgen, 036 Milano (MI) ITALY. Tel , fax , marco.navone@unibocconi.it.

2 cies, and we also know that governance quality 3 and industry concentration 4 matter in the determination of the price structure in the mutual fund industry. Nonetheless many empirical facts remain unexplained. One strand of literature deals with loads, whose role and market rationale seems to represent a puzzle per se 5, while other authors look at the long-term increase of expense ratios 6 and at the negative relationship between managerial ability and expense ratios 7. My contribution focuses on the dynamics of expense ratios looking at the determinants of the repricing decisions. Particularly I show that when deciding to change the expense ratio of a fund investment companies strategically exploit cross-sectional and time variations of investors pricesensitivity. The works more closely related to this research are Christoffersen and Musto (00) and Gil-Bazo and Ruiz-Verdú (009). The first contribution focuses on the negative relationship between expense ratios and past returns for money market funds. The authors conclude that funds that exhibited the worst past performance are more likely to be owned by investors with a low performance-sensitivity (the ones that didn t left the fund after the bad performance) and that investment companies exploit this situation by charging higher expense ratios. The second contribution applies the same intuition to equity mutual funds in order to justify the negative relationship between expense ratios and managerial ability (measured with a 4-factors alpha). Previous studies have focused on cross-sectional variation of expense ratios. Mutual fund fees also show significant time dynamics: from 980 to 006 US equity mutual funds exhibited a mean absolute change of expense ratios from year to year (in percentage terms) of 6.8%, that translates in around 8 basis points of increase or decrease every year, and since the standard deviation is.9% (or 3.7 basis points) changes of 0% or more in a single year are not uncommon. See for example Sirri and Tufano (998), Huang et al. (007) and Iannotta and Navone (009) for a discussion of the effect of search and participation costs on the investors behavior and fee dispersion. 3 The effect of governante on mutual fund fees is addressed, among others, by Tufano and Sevick (997), Meschke (007), Gil-Bazo and Ruiz-Verdú (009). On the contrary Ferris and Yan (007) conclude that board composition, and other governance quality measures, are irrelevant. Khorana et al. (008) address the related topic of investors protection quality and mutual fund fees. 4 Luo (00) 5 On this topic see Livingston and O Neal (998), O Neal (999), Christoffersen et al. (005), Zhao (005), Nanda et al. (005), Houge and Wellman (007), Bergstresser et al. (009) and Navone and Pagani (009). 6, Ferris and Chance (987), McLeod and Malhotra (994) and Duke and Davis (006) look at the introduction of b- fees as a determinant of the increase of expense ratios in the last decades. The same increase is attributed by Golec (003) to the limitations to performance-based compensation imposed by current regulation.

3 Investment companies can change the expense ratio by amending the fund prospectus or changing existing fee-waiving policies. For example, according to a recent article on the Wall Street Journal 8, one of the largest US investment companies decided, at the end of 008, to terminate a 0% feewaiving policy that had been in place since 004. The dynamics of these decisions are intriguing and have not, so far, been investigated. Practitioners usually justify re-pricing decisions considering that investment companies have fixed costs and so when assets under management go down (because the market goes down, for example) they are forced to increase expense ratios 9. This explanation seems to be rather unconvincing: increasing expense ratios improve profit margins also when the stock market is going up and assets under management are increasing, so why don t investment companies always increase expense ratios? And if fee increases are just a response to poor fund performance and assets shrinking why fee changes do show a coefficient of variation seven time larger than that of fund yearly returns and almost four time larger than that of assets under management percentage changes? Abundant anecdotical evidence could also be provided to show the insufficiency of the above mentioned explanation: in 996 the Prudential Utility Fund produced a performance of % ending in the top 0% of its category ranking. During the fiscal year assets under management grew from.7 to bn$, and, coherently with the fixed-cost explanation, in the next fiscal year the fund reduced the expense ratio from 88 to 57 bps (a 35% decrease). In 997 the fund repeated its good performance yielding a total return of 7.7% (top 5% of the ranking), and experienced an increase of asset under management from to.58 bn$. Surprisingly the investment company decided to increase, the following year, the expense ratio from 57 to 78 bps, with a percentage increase close to 37%. A common feature of past studies on mutual fund fees is to consider expense ratios as a negative component of net performance and thus they try to justify fund prices looking at investors performance-sensitivity. In this paper I investigate the determinants of changes in expense ratios modeling price-sensitivity as separate (and different) from performance-sensitivity. Although investors utility function is (should be) defined on the fund performance net of expenses, and thus expense ratios should only be considered as a negative component of the fund net performance, empirical evidence on investors behavior suggest that fund expenses may affect investors decisions in a different way. Prices and performances are considered separately in a number of contributions that investigate the determinants of investors fund choices: for example Capon et al. (996) using survey data 7 Christoffersen and Musto (00), Gil-Bazo and Ruiz-Verdú (008) and Gil-Bazo and Ruiz-Verdú (009). 8 Battered Mutual-Fund Firms To Raise Fees on Shareholders, by Sam Mamudi, 3 March 009, The Wall Street Journal. 9 See for example the Wall Street Journal quoted above. 3

4 cluster investors into four groups based on fund selection criteria. Among these groups two are characterized by high performance-sensitivity but while the first shows low price-sensitivity the second is characterized by a significant relevance of expense ratios in the decision process. More recently Wilcox (003) shows, in an experimental setting, that price and performance sensitivities are statistically different (the former being significantly smaller than the second) and that the relative importance of the two is correlated with demographic variables (with highly educated investors, for example, paying relatively more attention to past performance than to expense ratios and loads). More recently Choi et al. (009) in an experiment on Harvard staff members, MBA and College Students show that the demographics that affect past return sensitivity are not the same of those that drive the choice of funds with lower fees 0. The fact that investors do not consider expenses simply as a negative component of net performance can be inferred also by the results of Alexander et al. (998). In a survey on more than 000 investors the authors asked if a fund with higher than average expenses had to be expected to deliver, in the future, a performance above, about or below the average. The bulk of investors (64.4%) expected this fund to deliver a performance close to the average of the industry, 9.9% expected a performance higher than the average (implying a positive relationship between performance and fees) and only 5.7% answered that a fund with higher fees should deliver a lower than average performance. The fact that the role of expenses and past performance can be affected by behavioral bias has been investigated by Bailey et al. (009). The authors show that investors affected by narrow framing and overconfidence show a performance sensitivity higher than the average but they also choose funds with higher expense ratios. Judging the rationality of portfolio choice criteria is beyond the scope of this contribution, the fact that fees and performance are considered separately may be a signal of a naïve investment behavior or of the fact that they both affect in a non-trivial way the expectation of future performances. What is relevant here is that there is additional insight that can be gained by modeling separately the two sensitivities. In this paper I jointly estimate price and performance sensitivities from fund-level net investment flows and show that consistently with previous research performance-sensitivity is positively related to the fund past return. Christoffersen and Musto (00) argue that when the fund has a good performance attracts performance-chasers increasing the average performance-sensitivity of fund 0 Engström (007) reaches the same conclusion on a large sample of Swedes retail investors. 4

5 shareholders. The opposite is true when the fund experience bad performances: return-chasers leave and the pool of existing shareholders exhibit a relatively low performance-sensitivity. The analysis of price sensitivity is more interesting because it exhibit a non-monotonic relationship with past performance: in the lower part of the performance ranking we observe an increasing pricesensitivity (coherent with the idea that expenses are simply negative component of net performance), surprisingly in the top half of the performance ranking we observe a decreasing pricesensitivity. I call this a distraction effect: investors simply pay less attention to fees for funds that experienced, in the previous year, a noteworthy performance. Price-sensitivity differs from performance-sensitivity in two other key aspects: first of all I show that while the first increases with fund visibility, the second decreases. It seems that investors rely heavily on past performance as an investment criterion when fund visibility is low (the fund exhibits high search costs) and acquiring additional information would be too expensive. On the contrary, for high visibility funds investors tend to rely more heavily on additional pieces of information (for example about expense ratios) and less on past performance. Finally I also show that looking at the evolution of the two sensitivities from 980 to 006 no discernible trend can be observed for the sensitivity of investment flows to past performance while price-sensitivity exhibit a strong and stable increase after 990. It seems that the increase in mutual fund information availability for retail investors generated by dedicated providers such as Morningstar has changed the investment decision process increasing the relative importance of hard-to-find information such as expense ratio with respect to the widely available past-performance data. In the last part of the work I also show that investment companies strategically time their re-pricing decisions in order to exploit time variations in price and performance sensitivities: increases of expense ratios are positively correlated with decreases of the expected sensitivity of investment flows to both performance and prices. The rest of the paper is organized as follows: in Section, I present the dataset and some descriptive statics. Section 3 deals with the quantification of performance and price sensitivity. Section 4 analyzes re-pricing decisions. Section 5 concludes. Dataset 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. We collect data from 980 to 006 on all non-industry-specific US domestic equity funds 5

6 with assets under management, at the beginning of the year, not smaller than USD 0 million. Since CRSP does not provide consistent fund investment objectives and fund family names for the years prior to 99, we classify funds into different types and identify their family affiliation based upon the CDA-Spectrum mutual fund data from Thomson Financial, Inc. In order to have a sample of rather homogenous investment products we only consider funds the follows three investment objectives: aggressive growth, growth, and growth and income. Because we focus on flows into actively managed retail funds, we exclude index and institutional funds from our sample. Since CRSP does not identify index and institutional funds prior to 999 we follow, prior to this date, the identification methodology proposed by Gil-Bazo and Ruiz-Verdú (009) based on text recognition in the name of the fund. The number of funds in our sample grows from 00 in 993 to 34 in 006. As can be noted from Table sample size is smaller than what reported in comparable studies, as for example in Huang et al. (007). The difference is due to our minimum size requirement. Removing this constraint would bring sample size and other characteristics perfectly in line. [Insert Table about here] For the period considered in this study CRSP reports some unreasonably high values for expense ratios (for example there are 0 observations with values above 30%). In order to avoid bias in our estimates due to these outliers I winsorize the expense ratio distribution at the % level, leaving a maximum value of 3.0%. Following a common practice in mutual fund flows research I also winsorize net flows at the 5% level in order to avoid extreme values generated errors associated with mutual fund mergers and splits in the CRSP mutual fund database (for a detailed description of this problem see Elton, Gruber and Blake, 00). Since I focus my research on expense ratios and their dynamics the definition of the observation period is key element in the database building process: mutual funds report expense ratios on the base of a fiscal year that seldom coincides with the calendar year. In the database all the expense-related variables are defined over the fiscal year and performance-related variable are calculated accord- A fund is considered an index fund if the name contains any of the following strings: Index", Idx, Ix, Indx, NASDAQ, Nasdaq, Dow, Mkt, S&P 500, BARRA. IN the same way a fund is considered institutional if the name contains Inst or inst or if it belong to share classes Y or I. 6

7 ingly: the month rank of fund, for example, is the position of the fund in its category ranking in the months prior to the closing of the fiscal year. Calculating performance measures coherent with the fiscal year is important because we assume that investment companies decide, at the beginning of the new fiscal period, to increase or decrease the expense ratio on the base of the behavior of the fund in the previous period. A side effect of this methodology is that when we look at a given fiscal year, performance measures of the different funds are not calculated exactly over the same period because they refer to the exact fiscal year of each fund. Table reports summary statistics for expense ratio changes from 980 to 006. Although in the whole sample we have roughly the same frequency of increases and decrease we also observe a significant time variation with years where more than two thirds of the funds increased the expense ratios and years where the opposite is true. The spread between the 0 th and the 90 th percentile of the distribution of percentage changes in the expense ratios, ads well as the cross sectional standard deviation, testify to the existence also of a significant cross sectional variation: in the entire sample mean and median percentage changes are close to zero but with a standard deviation of 4.37%. [Insert Table about here] 3 Price and Performance Sensitivity In order to estimate price- and performance-sensitivity I follow the intuition of Gil-Bazo and Ruiz- Verdú (009) and model fund net flows as a function of past performance, expenses and other control variables that capture stylized facts reported in previous literature. Net flows are defined, as the percentage growth of total net assets (TNA) adjusted for fund return net of expenses (r it ): Flow it TNA it TNA TNA r it Flows are calculated on a monthly basis and aggregated over the fiscal year in order to minimize approximation error due to the timing of investment decisions. 7

8 As performance measure I consider the fund's fractional rank in the previous fiscal year (RANK it- ) represents its percentile performance relative to other funds (on the basis of a funds' one-year raw return) with the same investment objective in the same period, and ranges from 0 to. This measure captures the tournament-nature of the mutual fund industry (Brown, Harlow and Starks, 996) and has been proven highly relevant in terms of ability to capture investors behavior 3. Expenses are represented by two variables, the total expense ratio (EXP it- ) and a dummy variable for load funds (LOAD it- ). The results of Barber et al. (005) suggest to considerate the two cost component separately as they may affect investors behavior in different ways. Moreover Bergstresser et al. (009) document relevant differences between load and no-load funds and Navone and Pagani (009) show how loads can affect the flow-performance relationship. I will estimate three different models: in Model only linear relationship between flows and both performances and expense ratios will be considered. Model the performance measure will enter the equation with a quadratic term in order to capture the well known asymmetry in the relationship between investment flows and performance (Ippolito, 99). Finally Model 3 will include also a quadratic term for the expense ratio. In all models past performance and expense ratio will be interacted with each other and with the time fixed effects. The first interaction will allow us to analyze how fund past performance affects investors price sensitivity and the latter allows for time variation in price and performance sensitivities. Several recent contributions have also highlighted the stickiness of mutual fund flows, suggesting the introduction of the lagged value of flows (Flow t- ) as an explanatory variable in the model 4. I will also control for (the natural logarithm of) fund size (SIZE it- ) and age (AGE it- ) and investment company size (ICSIZE it- ) the fact that a fund belongs to a star family (STAR it- ) 5, the normalized standard deviation of the fund (STD it- ) and the asset weighted net flow into funds with the same investment objective (IO_Flow it- ). These variables have been shown relevant in terms of forecasting power on investment flow. Moreover Huang et al. (007) and Iannotta and Navone (008) consider the first four as proxies for information availability. All these control variables will be interacted with past performance and expense ratios in order to capture the effect of search costs on price and performance sensitivity. 3 See, for example, Sirri and Tufano (998), Huang et al. (007), Kempf and Ruenzi (008), Kempf et al. (009). 4 To the best of my knowledge the first paper to introduce this topic was Fant and O Neal (000). More recently Cashman et al. (007) have addressed the same problem looking at gross investment flows. Lagged flows explanatory variables are also used by Gil-Bazo and Ruiz-Verdú (009). 8

9 Finally the three models will also include time (TFE), investment objective (IOFE) and fund (FFE) fixed effects 6. In order to properly address the correlation between the individual fixed effect and the lagged dependent variable in the right-hand side of the equation the models will be estimated using the Arellano-Bond (99) robust estimation 7. Complete estimation results are reported in Appendix. Starting from estimated coefficients of Model 3 I develop measures of price and performance sensitivities as the first derivatives of fund flows with respect to expense ratio and past performance respectively. Considering the equation form of Model 3 Flow a b Flow c d e it EXP RANK c EXP RANK c EXP RANK c EXP RANK EXP SIZEit d EXP AGEit d3exp ICSIZEit d 4 EXP STDit d5exp LOADit d6exp STARit RANK e RANK e RANK ICSIZE e RANK STD e RANK LOAD e RANK STAR g SIZE 6 h k k 0 g AGE b RANK RANKit TFEk mk EXPit TFEk nktfek pk IOFEk it k g ICSIZE 3 6 b RANK g STD b EXP 3 k g LOAD 6 5 b EXP 4 k g STAR g IO _ Flow 7 it 5 6 Price and performance sensitivities can be derived as follows: PERF _ Sens e SIZE and d SIZE it PRICE _ Sens e AGE Et Flow RANK it d E AGE it e ICSIZE 3 Flow b b RANK b e STD 4 c EXP e LOAD 5 c EXP e STAR 6 c 3 h EXP RANK c EXP RANK t 4 EXP RANK c RANK c EXP RANK t it EXPit d ICSIZE 3 b EXP d STD 4 c RANK d LOAD 5 c d STAR 6 m t It is important to note that PRICE_Sens it can be seen as the expected value in t- of the sensitivity of flows to expense ratios in time t. 5 Consistently with Nanda et al. (004) I define as a star a fund that ranks in the top 5% of its own category. The variable STAR is a dummy set equal to one for all the funds belonging to a complex that has at least one star. 6 As a robustness check I ve also estimated the sensitivities using no fixed effects at all and investment company fixed effects (instead of fund FEs). All the results are robust and available from the author. 7 Alternatively the models have been estimated using pooled regressions and standard within estimation. Results are robust to all three methodologies. In the paper I prefer to present the results of the Arellano-Bond specifications because there is no valid a-priory reason to rule out the possibility of significant biases due to the correlation between the fixed effects coefficients and the lagged dependent variable. 9

10 The presence of numerous interaction terms makes the economic interpretations of the single coefficients difficult. Some comparative statics may help in understanding few interesting features of the estimated price- and performance-sensitivity coefficients. Figure shows that while performance sensitivity is monotonically increasing with fund past performance, thus generating the well documented asymmetry in the flow-performance relationship, price-sensitivity is not. Specifically for funds in the lower half of the performance ranking pricesensitivity increases with past performance (the reader should remember that fees enter with a positive sign in the equation, thus when the sensitivity becomes more negative we can say that pricesensitivity increases). This behavior mimics that of performance-sensitivity and is coherent with the finding that when return is very low the investors with the higher performance sensitivity will leave the fund. Since fees have a negative effect on performance it is reasonable to observe the same effect on the two sensitivities. If we observe the top half of the performance ranking we see a different picture: performance sensitivity is still increasing while price-sensitivity drops (it becomes less negative). I call this effect investors distraction : when a fund shows a very good performance investors pay less and less attention to fund fees. According to Christoffersen and Musto (00) the funds that had the best performance in the previous period should be populated by the investors with the highest performance-sensitivity (as confirmed in this paper) nonetheless they do not seem to react to the negative performance implied in higher expense ratios. This empirical result confirms the intuition of Capon et al. (996), Wilcox (003) and Choi et al. (009) according to which investors do not consider expense ratios simply as a negative element of fund performance. The consequence of this non-monotone relationship between past performance and price-sensitivity is that both funds that had a very bad performance or a very good performance in the previous period are more likely to increase their expense ratios: the formers are populated by investors that are not very performance-sensitive, and thus do not care much about the decrease in fund expected returns generated by the increased fees, while the latters are populated by investors distracted by the strong performance and thus not very concerned about the price they have to pay in order to buy the fund manger services. [Insert Figure about here] A second interesting result is the effect of fund visibility on performance and price sensitivities: according to Huang et al. (007) past performance is a relatively cheap information in the sense that investors can acquire it relatively easily, on the other side fund expenses are more costly to gather 0

11 and thus we should observe a different effect of fund visibility on the two types of sensitivity. Figure shows that price sensitivity is stronger for high visibility funds with respect to the median fund and low visibility funds. I define high visibility a fund that is large (size in the 75 th percentile), old (age in the 75 th percentile), managed by a large investment company (size of the investment company in the 75 th percentile) and charges front loads (to account for the increased visibility generated by the broker). In the same way a low visibility fund will be a no-load fund relatively small, young and managed by a small invest company (all the variables at the 5 th percentile). The empirical result is coherent with the idea that for investors its easier to acquire information on expense ratios for funds with low search costs and thus, for these funds, investment flows show higher sensitivity to fund expenses. This result is particularly interesting if confronted with what emerges from Figure 3 where performance sensitivity is plotted for funds with different visibility. Here the opposite empirical effect can be observed: sensitivity to past performance is stronger for funds with low visibility. It seems that investors rely heavily on past performance as an investment criterion when the fund exhibits high search costs and acquiring additional information would be too expensive. On the contrary, for high visibility funds investors tend to rely more heavily on additional information (for example about expense ratios) and less on past performance. [Insert Figure and Figure 3 about here] A final important result is related to the evolution of price and performance sensitivities trough time. In this work I use data from 980 to 006. During this period mutual funds information availability for retail investors has increased dramatically thanks to specialized information providers such as Morningstar and the diffusion of Internet and on-line brokerage. Figure 4 shows the evolution of price and performance sensitivities trough time for the median fund (with a median past performance). For performance sensitivity we cannot observe any discernible time trend while for price sensitivity we see a stable increase from 990 to 006. This evidence strongly supports the intuition that greater information availability has enabled investors to base their allocation decision on a wider information set, they are no longer constrained to rely on past performance, the most available piece of information. [Insert Figure 4 about here]

12 4 Price Sensitivity and Re-pricing Decisions Table shows that actively managed retail equity mutual every year re-price their services altering the expense ratio charged to investors. The aim of this analysis is to investigate if investment management companies time re-pricing decisions in order to take advantage of changes in investors price-sensitivity. Christoffersen and Musto (00) show that money market fund managers increase fees after a year of bad performance taking advantage of the reduce average performance-sensitivity of fund shareholders. In the previous session I ve demonstrated that price-sensitivity is different from and not always positively correlated with performance-sensitivity: investors in funds with high past performance may be distracted by the exceptional return and thus exhibit a decreasing pricesensitivity. In order to test for this assumption I regress changes in expense ratios over changes in pricesensitivity controlling for changes in performance-sensitivity and other fund-related control variables. Specifically Table 3 reports the results of five panel regression models where the dependent variable is the difference between the expense ratios charged by each mutual fund in years t and t-. [Insert Table 3 about here] The first three models only consider changes in expected price and performance sensitivities of fund shareholders between years t and t-, PRICE_Sens and PERF_Sens respectively, as well as time, investment objective and fund fixed effects. The two sensitivities are considered separately (Models and ) and together (Model 3). The results confirm that investment companies time both sensitivities with their re-pricing decisions: we see that expense ratio changes are positively correlated with decreases of expected price and performance sensitivities 8. Model 4 also consider, as possible explanatory variables, changes in the mutual fund size (SIZE) and relative risk (STDEV) as well as the asset weighted average change of expense ratios for other funds managed by the same investment company (IC_EXPCH) or with the same investment objective (IO_EXPCH). The first two variables should capture changes in the cost of managing the fund due to a different investment policy or to a change in the incidence of fixed costs, while the last two should take care of market- and investment company-level effects. For example we could observe an increase in demand of funds 8 The reader should remember that since expense ratios are expressed as a positive number a positive change in price sensitivity has to be interpreted as a reduction of price sensitivity: the relationship between flows and fees becomes les negative.

13 with a certain investment policy wit ha subsequent increase in the price not motivated by fund-level variables or we could have changes in a fund expense ratio due to bottom-line issues at the investment company level. Results of Model 4 confirm that the decision to change the expense ratio of a fund may be motivated by issues related to the economics of the whole investment company, nonetheless this effect does not reduce the significance of the main variables. Finally Model 5 also considers two additional explanatory variables related to changes in the performance of the fund from year t- to year t-: RANK is the change in the positions in the performance rankings of the fund and STAR is a variable with a value of if the fund became a star in t-, a value of - if the fund ceased to be a star in t- and a value of 0 otherwise 9. Of course these two variables play a major role in the determination of the changes in price and performance sensitivities so their inclusion will tell us if my expected sensitivity measures contain any additional information beyond, the simple change in past performance (as would be implied by Christoffersen and Musto, 00). The results show that while the inclusion of these two variables strongly reduces the significance of the performance-sensitivity measure it does not alter significantly the coefficient and the p-value of the price-sensitivity measure. This is a last confirmation of the fact that investors exhibit a price sensitivity that is not entirely captured by their sensitivity to fund returns: expense ratios and fund management costs in general are not considered simply as a negative portion of expected future returns. 5 Conclusions In this paper, I analyze effecter-pricing decisions for mutual fund management services. From the behavior of net investment flows for a large sample of US actively managed equity mutual funds from 980 to 006 I derive measures of performance and price sensitivity. Contrary to previous empirical research I show that investors do not consider expense ratio simply as a negative component of expected returns: their behavior show a price-sensitivity that is different and not always positively correlated with performance sensitivity. Specifically I show that while performance sensitivity monotonically increases with past performance, price-sensitivity does not: investors that buy top past-performers seems to be distracted by the fund previous return and pay relative little attention to the expense ratio. I also demonstrate that price-sensitivity increases with fund visibility 9 Consistently with the previous chapter a fund is labeled a star if it ranks in the top 5% of its category. 3

14 while performance sensitivity decreases: when search costs are high investors heavily rely, for their allocation decision, on the most widely available piece of information: the fund past performance. When the fund becomes more visible other variables enter in the decision process. A third important result is related to the evolution of price and performance sensitivities trough time. Looking at data from 980 to 006 no discernible trend can be observed in the average performance sensitivity while price sensitivity strongly increases after 990. This result shows that the dramatic increase in the availability of mutual funds information for retail investors due to specialized providers such as Morningstar has changed the investment decision process increasing the relative importance of less-visible pieces of information such as expense ratios. Finally I show that investment companies strategically time their re-pricing decisions in order to exploit time variations in price and performance sensitivities: increases of expense ratios are positively correlated with decreases of the expected sensitivity of investment flows to both performance and prices. 4

15 Appendix This appendix report complete estimation of the three regression models described in Section 3 of the paper. Table A reports estimated coefficients for the main and the control variables as well as the respective interactions. Table A and A3 reports coefficients for the interactions between time fixed effects and past performances and expense ratios respectively. Table A () () (3) Coeff. P-Value Coeff. P-Value Coeff. P-Value Constant 0.04 (0.967) (0.87) (0.855) FLOWS t (0.355) (0.300) (0.59) RANK.84*** (0.000).08*** (0.000).496*** (0.000) RANK (0.58) -0.4** (0.044) EXP *** (0.004) *** (0.003) 79.39*** (0.000) EXP (*) -'003*** (0.000) EXP_RANK 5.0** (0.04) -3.5 (0.646) ** (0.06) EXP_RANK 9.3 (0.7) 83.09*** (0.006) EXP _RANK (*) '3** (0.03) EXP _RANK (*) -'445** (0.06) EXP_SIZE *** (0.00) *** (0.00) *** (0.000) EXP_AGE -4.54** (0.034) -4.60** (0.03) -4.90** (0.07) EXP_ICSIZE 0.06 (0.984) (0.946) 0.95 (0.803) EXP_STD (0.56) 0.8 (0.648) (0.364) EXP_LOAD (0.609) (0.660).336 (0.459) EXP_STAR.54 (0.44) (0.806) 0.84 (0.579) RANK_SIZE (0.36) -0.0 (0.53) (0.9) RANK_AGE -0.0*** (0.000) -0.6*** (0.000) -0.6*** (0.000) RANK_ICSIZE (0.40) 0.00* (0.087) 0.00* (0.084) RANK_STD 0.05** (0.05) 0.06** (0.0) 0.07*** (0.008) RANK_LOAD 0.00 (0.674) (0.686) 0.00 (0.650) RANK_STAR 0.068** (0.0) (0.8) (0.3) SIZE *** (0.000) *** (0.000) -0.64*** (0.000) AGE 0.30*** (0.000) 0.6*** (0.000) 0.4*** (0.000) ICSIZE (0.49) (0.53) (0.637) STD -0.0 (0.45) -0.0 (0.5) -0.06* (0.090) LOAD (0.534) 0.07 (0.57) (0.59) STAR 0.00 (0.938) 0.06 (0.385) 0.00 (0.507) IO_FLOWS t 0.73*** (0.000) 0.75*** (0.000) 0.73*** (0.000) Time FE Yes Yes Yes Inv. Obj FE Yes Yes Yes Fund FE Yes Yes Yes N of Obs m (*) Estimated coefficient divided by 000. m is the p-value on a test for the second order serial correlation of the residuals of the first-difference equation (see Arellano and Bond, 99). 5

16 Table A () () (3) Coeff. P-Value Coeff. P-Value Coeff. P-Value RANK_ ** (0.04) 0.37** (0.039) 0.36** (0.044) RANK_ *** (0.008) -0.0*** (0.007) -0.07*** (0.006) RANK_98-0.0** (0.0) -0.09** (0.0) -0.8*** (0.008) RANK_ (0.07) 0.3 (0.) 0.7 (0.8) RANK_ (0.78) (0.745) (0.706) RANK_ (0.) -0.5 (0.09) -0.3 (0.7) RANK_ * (0.058) 0.8* (0.065) 0.8* (0.06) RANK_ (0.) (0.0) (0.96) RANK_ *** (0.003) -0.76*** (0.003) -0.76*** (0.004) RANK_ (0.65) (0.65) (0.3) RANK_ ** (0.03) -0.6** (0.07) -0.0** (0.04) RANK_ (0.68) 0.04 (0.63) 0.07 (0.5) RANK_ (0.93) (0.94) 0.00 (0.988) RANK_ (0.34) (0.333) (0.39) RANK_ *** (0.000) 0.9*** (0.000) 0.9*** (0.000) RANK_ (0.50) (0.4) (0.07) RANK_ (0.94) (0.39) (0.) RANK_ *** (0.000) 0.60*** (0.000) 0.59*** (0.000) RANK_ *** (0.000) -0.45*** (0.000) -0.44*** (0.000) RANK_ * (0.090) 0.06* (0.08) 0.06* (0.086) RANK_ *** (0.000) 0.30*** (0.000) 0.30*** (0.000) RANK_ *** (0.00) -0.35*** (0.00) -0.9*** (0.003) RANK_ (0.5) (0.09) 0.04 (0.88) RANK_ *** (0.000) 0.093*** (0.000) 0.093*** (0.000) RANK_ *** (0.000) 0.67*** (0.000) 0.69*** (0.000) RANK_ *** (0.000) 0.37*** (0.000) 0.40*** (0.000) RANK_ *** (0.000) 0.36*** (0.000) 0.36*** (0.000) 6

17 Table A3 () () (3) Coeff. P-Value Coeff. P-Value Coeff. P-Value EXP_ (0.458) (0.43) -3. (0.444) EXP_ ** (0.07) -3.** (0.08) ** (0.08) EXP_ (0.435) (0.448) (0.498) EXP_ (0.840) -.59 (0.838) (0.87) EXP_ (0.933) -.03 (0.933) (0.989) EXP_ (0.383) (0.366) (0.409) EXP_ (0.535) (0.556) (0.54) EXP_ (0.86) 8.6 (0.86) (0.99) EXP_ (0.699) (0.703) (0.589) EXP_989 -.** (0.00) -.48** (0.00) -.900** (0.03) EXP_ *** (0.000) *** (0.000) *** (0.000) EXP_ *** (0.000) -7.5*** (0.000) *** (0.000) EXP_ *** (0.004) -6.66*** (0.004) -7.74*** (0.003) EXP_ *** (0.000) *** (0.000) *** (0.000) EXP_ *** (0.00) -.899*** (0.00) -4.37*** (0.00) EXP_ *** (0.000) *** (0.000) -3.36*** (0.000) EXP_ *** (0.000) -3.59*** (0.000) *** (0.000) EXP_ *** (0.000) *** (0.000) *** (0.000) EXP_ *** (0.000) *** (0.000) *** (0.000) EXP_ *** (0.000) *** (0.000) *** (0.000) EXP_ *** (0.000) *** (0.000) *** (0.000) EXP_ *** (0.000) *** (0.000) *** (0.000) EXP_ *** (0.000) -6.07*** (0.000) *** (0.000) EXP_ *** (0.000) *** (0.000) *** (0.000) EXP_ *** (0.000) *** (0.000) *** (0.000) EXP_ *** (0.000) *** (0.000) *** (0.000) EXP_ *** (0.000) *** (0.000) -0.56*** (0.000) 7

18 Bibliography Alexander, G., J. Jones and P. Nigro, 998, Mutual fund shareholders: characteristics, investor knowledge, and sources of information, Financial Services Review, 7, pp Arellano, M. and S. Bond, 99, Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations, Review of Economic Studies, 58, pp Bailey, W., A. Kumar and D. Ng, 009, Behavioral Biases and Mutual Fund Clienteles, mimeo. Barber, B., T. Odean and L. Zheng, 005, Out of Sight, Out of Mind: The Effects of Expenses on Mutual Fund Flows, Journal of Business, 78(6), pp Bergstresser, D., J. Chalmers and P. Tufano, 009, Assessing the costs and benefits of brokers in the mutual fund industry, Review of Financial Studies, forthcoming. Brown, K., W. Harlow and L. Starks, 996, Of Tournaments and Temptations: An Analysis of Managerial Incentives in the Mutual Fund Industry, Journal of Finance, 5(), pp Capon, N., G. Fitzsimons and R. Prince, 996, An individual level analysis of the mutual fund investment decisions, Journal of Financial Services Research, 0, pp Cashman, G., D. Deli, F. Nardari, and S. Villupuram, 007, Investor Behavior in the Mutual Fund Industry: Evidence from Gross Flows, mimeo. Christoffersen, S. and D. Musto, 00, Demand curves and the pricing of money management, Review of Financial Studies, 5, pp Christoffersen, S., R. Evans and D. Musto, 005, The Economics of Mutual-Fund Brokerage: Evidence from the Cross Section of Investment Channels, mimeo. Choi, J., D. Laibson and B. Madrian, 009, Why Does the Law of One Price Fail?An Experiment on Index Mutual Funds, mimeo. Dukes, W., P. English and S. Davis, 006, Mutual fund mortality, b- fees, and the net expense ratio, Journal of Financial Research, 9(), pp Elton, E., M. Gruber and C. Blake, 00, A First Look at the Accuracy of the CRSP Mutual Fund Database and a Comparison of the CRSP and Morningstar Mutual Fund Databases, Journal of Finance, 56(6), pp Engström, S., 007, Preferences and Characteristics of Mutual Fund Investors, mimeo. Fant, F. and E. O Neal, 000, Temporal changes in the determinants of mutual fund flows, Journal of Financial Research, 3(3), pp Ferris, S. and D. Chance, 987, The effect of b- plans on mutual fund expense ratios: A note, The Journal of Finance, 4(4), pp Gil-Bazo, J. and P. Ruiz-Verdú, 009, Yet another puzzle? The relation between price and performance in the mutual fund industry, Journal of Finance, Forthcoming. 8

19 Golec, J., 003, Regulation and the Rise in Asset-Based Mutual Fund Management Fees, Journal of Financial Research, 6(), pp Hogue, T. and J. Wellman, 007, The use and abuse of mutual fund expenses, Journal of Business Ethics, 70(), pp Hortaçsu, A. and C. Syverson, 004, Product differentiation, search costs, and competition in the mutual fund industry: A case study of S&P 500 index funds, Quarterly Journal of Economics, 9, pp Huang, J., K.D. Wei and H. Yan, 007, Participation costs and the sensitivity of fund flows to past performance, Journal of Finance, 6(3), pp Iannotta, G. and M. Navone, 008, Search Costs and Mutual Fund Fees Dispersion, CAREFIN Research Paper, 3/08. Ippolito, R., 99, Consumer reaction to measures of poor quality: Evidence from the mutual fund industry, Journal of Law and Economics, 35, pp Kempf, A. and S. Ruenzi, 008, Tournaments in Mutual Fund Families, 008, Review of Financial Studies, (), pp Kempf, A., S. Ruenzi and T. Thiele, 009, Employment risk, compensation incentives, and managerial risk taking: Evidence from the mutual fund industry, Journal of Financial Economics, 9(), pp Khorana, A., H. Servaes and P. Tufano, 008, Mutual fund fees around the world, Review of Financial Studies, forthcoming. Latzko, D., 999, Economies of scale in mutual fund administration, Journal of Financial Research, (3), pp Livingston, M. and E. O Neal, 998, The Cost of Mutual Fund Distribution Fees, Journal of Financial Research, (), pp Luo, G., 00, Mutual fund fee-setting, market structure and mark-ups, Economica, 69(), pp McLeod, R. and D. Malhotra, 994, A Re-examination of the Effect of b- Plans on Mutual Fund Expense Ratios, Journal of Financial Research, 7(), pp Meschke, F., 007, An Empirical Examination of Mutual Fund Boards, Mimeo. Nanda, V., Z.J. Wang and L. Zheng, 004, Family values and the star phenomenon: Strategies of mutual-fund families, Review of Financial Studies, 7, pp Nanda, V., Z.J. Wang and L. Zheng, 005, The ABCs of Mutual Funds: On the Introduction of Multiple Share Classes, mimeo. Navone, M. and M. Pagani, 009, Loads and Investment Decisions, CAREFIN Research Paper, /09. 9

20 O Neal, E., 999, Mutual Fund Share Classes and Broker Incentives, Financial Analysts Journal, 55(5), pp Sirri, E.R. and P. Tufano, 998, Costly search and mutual fund flows, Journal of Finance, 53, pp Tufano, P. and M. Sevick, 997, Board Structure and and Fee-Setting in the U.S. Mutual Fund Industry, Journal of Financial Economics, 46(3), pp Wilcox, R., 003, Bargain Hunting or Star Gazing? Investors' Preferences for Stock Mutual Funds, The Journal of Business, 76(4), pp Zhao, X., 005, The role of brokers and financial advisors behind investments into load funds, mimeo. 0

21 Table Summary Statistics This table reports summary statistics of our sample from 980 to 006. At the end of each year, we calculate the crosssectional mean value of total net asset value, fund age, expense ratio, annualized mean monthly return, standard deviation and net investment flow. Year Number of Funds TNA (Millions) Age (Years) Expense Ratio (%) Annualized Return (%) Standard Deviation (%) Net Investment Flow (%)

22 Table Changes in Expense ratios This table reports summary statistics of changes in expense ratios in our sample from 980 to 006. Year % of Increases % of Decreases % Change in the Expense Ratio Mean Median 0th pctile 90th pctile Standard Deviation Total

23 Table 3 Changes in Expense ratios This table reports the results of five panel regressions where the dependent variable is the difference between the expense ratios charged by each mutual fund in year t and t-. PRICE_Sens and PERF_Sens are the changes in the expected price and performance sensitivities of fund shareholders between years t and t-. RANK is the change in the positions in the performance rankings of the fund in years t- and t-. STAR has a value of if the fund became a star in t-, - if the fund ceased to be a star in t- and 0 otherwise. A fund is labeled a star if it ranks in the top 5% of its category. SIZE is the change in the asset under management of the fund between years t- and t-. STDEV is the change in the normalized standard deviation of the fund between years t- and t-. IC_EXPCH is the asset weighted average change of expense ratios for funds managed by the same investment company and IO_EXPCH is the asset weighted average change of expense ratios for funds with the same investment objective. Each model includes time, investment objective and individual fund fixed effects. Standard Errors are clustered at the fund level. ***, **, * represent significance at the %, 5% and 0% levels respectively. () () (3) (4) (5) Constant *** ** 0.067*** (0.68) (0.000) (0.57) (0.09) (0.00) PRICE_Sens 0.003*** 0.004*** 0.004*** 0.005*** (0.000) (0.000) (0.000) (0.000) PERF_Sens *** *** *** -0.03* (0.000) (0.000) (0.000) (0.06) RANK *** (0.00) STAR ** (0.07) SIZE (0.00) (0.838) STDEV * (0.50) (0.09) IC_EXPCH 0.03*** 0.03*** (0.000) (0.000) IO_EXPCH (0.78) (0.63) Time FE Yes Yes Yes Yes Yes Inv Obj FE Yes Yes Yes Yes Yes Fund FE Yes Yes Yes Yes Yes N of Obs. 0'898 0'898 0'898 9'09 9'09 Adj R