«Edwin J. Elton* Martin J. Gruber* Jeffrey A. Busse* October 2002 Abstract Financial theory is often based on the belief that the actions of rational ...»
Since both 12b-1 fees and loads are primarily payments to support marketing and distribution, an investor might examine them separately from other expenses in deciding which fund to buy.
III. Cash Flows and Fund Characteristics In this section we discuss a set of variables, including performance characteristics, that might be associated with future fund cash flows and then examine their impact. As the prior section shows, the important performance characteristics of index funds are easily understood, and for most of these characteristics we can easily predict future
We measure each of these variables at the start of the period in which we measure cash flow. Thus, all of them are known to the investor before they decide on their investment. Several researchers have suggested that, in addition to performance characteristics, a number of other variables might affect cash flow.11 We can categorize these variables into two groups: marketing and spillover. Marketing includes two types of influences. The first is any action management takes to increase investor awareness of the fund. This includes advertising, promotional brochures, website development, etc. The second involves compensating brokers and investment advisors for selling the fund (rather than competing funds). Although we do not have direct measures for either of these influences, there are some good proxies. Funds primarily use a large portion of 12bfees and the majority of loads to reward salespersons for selling the fund, with the remainder paying for general marketing effort. Consequently, the size of 12b-1 fees and loads should be related to marketing effort. In addition, the management fee is the part of the expense ratio that is less likely to be used for marketing effort.12 Thus we use the difference between the expense ratio and management fee as another measure of marketing effort. Investor awareness of a fund can also be a function of its frequency of inclusion in financial publications as well as direct marketing effort. To proxy for general investor awareness we use both the size of the fund and the age of the fund.
Spillover is the name given to the idea that being a member of a fund family increases inflows because investors prefer to concentrate their investments within one family.13 For example, an investor attracted to a fund family because of a desirable small stock fund might invest in the fund’s index fund for diversification purposes and a desire to hold all mutual fund investments in one family. We use the following variables to capture spillover effects: 1) the natural log of the total net assets of the fund family (to measure family size), 2) the number of different Morningstar categories held by the See Sirri and Tufano (1998) for a discussion of marketing variables.
It might seem surprising that the part of expenses labeled management fee is not simply used to compensate managers. Conversations with fund managers indicate they can and do sometimes pay brokers part of the management fee.
This becomes less important with the advent of systems that allow an investor to transfer funds among many families, such as the systems marketed by Fidelity and Schwab.
family (to measure the investor’s chance to diversify within the family), 3) fund flows into a family excluding the flows into the index fund (to measure the popularity of the family), and 4) the presence of a fund within the family that received a Morningstar fivestar rating. We include this last variable because Khorana and Servaes (1999) show that a major determinant of flows into a family is the presence of a highly rated fund.
There is another variable that might impact cash flow: the impression investors and financial professionals have of the service provided by the fund family. Dalbar, Inc.
constructs the leading measure of service rankings. Dalbar surveys more than 2,600 investors and professionals and records impressions of fund family service quality.
Dalbar was generous in providing us with data over our sample period for the families it follows. However, Dalbar gathers data for less than half of the families represented in our sample (25 out of 52 in the year with the greatest coverage). Consequently, we examine the service variable separately rather than including it directly in our regression.
Like others, we use a percentage flow variable to measure the flow into a fund.14 However, our measure is slightly different than ones used in previous research. The index funds in our sample are of dramatically different size. When a fund gets extremely large it becomes more and more difficult to attract the same percentage of new capital. Thus, we first compute expected flow based on a regression of cash flow on size. We then define percentage surprise by taking the difference between the actual flow (the net growth rate in fund assets beyond that due to capital gains and reinvested dividends) and expected flow and dividing the difference by the fund’s total net assets.
Table 6 shows the results of our analysis.15 The odd-numbered columns show the coefficients of the regression including one variation of the variables we discuss above.
See, for example, Sirri and Tufano (1998) and Warther (1995).
We also run each of the regressions on panel data incorporating an intercept for each fund in the sample.
An F test indicates we cannot reject the hypothesis that the firm intercepts are jointly equal to zero. In addition, we also use the criterion proposed by Akaike to test for the inclusion of additional variables. This test indicates that the firm intercepts should not be included. Hence, we report the simpler results in Table 6.
The even numbered columns show the coefficients of the regressions that include only those variables that have a significant relationship to the surprise in cash flow. Columns also differ in the performance variable used; 1 and 2 use differential return, 3 and 4 use alpha, and 5 and 6 use the expense ratio. The results are very similar regardless of the predictive measure of performance.
Although columns 1, 3, and 5 use one iteration of the full range of variables that have been suggested in the literature, we examine variations of similar variables separately. Thus, we use log of total net assets and age, which are highly correlated, in separate regressions. Table 6 reports only the most relevant alternative. However, in the text we discuss variations of variables not shown in the table.
We first examine the regressions shown in columns 2, 4, and 6. First, note that the overall explanatory power is at least as high as it is in other studies of this type based on actively managed funds.16 This is understandable given the ease with which we can predict characteristics for index funds. Of the variables measuring fund characteristics, overall performance and one risk variable are significant. Whether we measure overall performance by differential return, alpha, or expense ratio, fund flow shows a highly significant coefficient on the performance measure. The coefficient is also economically significant. Taking the difference between the 75th and 25th percentile of any performance measure leads to a difference in flows of about 30 percent of total net assets.
Although the regression coefficient on R2 – 1 is very large and statistically significant, the differences in R2 – 1 across funds are very small, so that the net impact is small. The average difference in cash inflow associated with the 25th percentile of R2 – 1 rather than 75th percentile is about 0.2 percent. The other risk measure, the absolute value See Sirri and Tufano (1993 and 1998) and Chevalier and Ellison (1997), for example. In fact, we find higher explanatory power since before running this regression we remove the impact of size on cash flow.
The R 2 between cash flow and size is 0.21. See Bergstresser and Poterba (2001) for a study of cash flows and taxes. See DelGuercio and Tkac (2001) for a study of cash flows and Moningstar ratings. See Jain and Wu (2000) for a study of advertising and fund flows.
of beta minus one, is insignificant. This is expected given the small variation in the variable.17 Somewhat more surprising is the lack of significance for the tax efficiency variables. Although there is little variation in dividends, there is substantial variation in capital gains, and capital gains efficiency is predictable. Investors are either unconcerned with capital gains, or they are unaware of its predictability.
Load is the only marketing variable that is significant, and it enters with a positive sign.18 Load has two impacts. First, load is a major cost to investors and reduces substantially the return they receive. Second, load is a reward for brokers and financial planners for including the fund in their clients’ portfolios. On net, the incentive for brokers and financial planners to push the fund is more important for new flows than the effect of reduced flows associated with investors avoiding high cost and poorly performing funds. None of the other marketing variables—12b-1 fees, fund size, age (results not shown), or expenses less management fee (results not shown)—are significant.
The number of types of funds in the fund family is the only spillover variable that is important. An index fund that is part of a family that offers a variety of other types of funds attracts more cash flow. This is logical because many investors wish to stay with one fund family to facilitate inter-fund transfers and record keeping. Neither family size, flow of funds into family (results not shown), nor the presence of a star fund explains cash flows.
We separately run the analysis for institutional funds and funds for individuals.
The results are essentially the same, with the following caveats. First, we anticipated that capital gains would be important in the individual sample. Most institutions are taxexempt, but individuals are not. Furthermore, there is substantial variation in capital gains The extremely small variation in beta indicates that S&P index funds have developed a cash management policy (possibly involving futures) that allows minimization of the impact of cash flows on their performance.
Excluding load decreases the explanatory power to 19 percent.
payouts across funds. The relationship is not significant, which indicates that individuals do not pay attention to taxes or are unaware of the differences across funds and their predictability. Second, we had anticipated that performance would be more important in affecting cash flows for institutional funds. This does not occur. The principal reasons for this are likely to be the small size of the institutional sample and the lack of a substantial variation in the performance of funds primarily sold to institutions.
As a final step in this part of the analysis, we examine the relationship between unexplained cash flow (the residuals from each of the regressions described above) and service rankings. Dalbar ranks fund families from best to worst according to perceived service quality. To examine the relationship, we compute the Spearman rank correlation between the perceived service quality of quartiles formed according to unexplained cash flow. The rank correlation is insignificant each year and overall. In fact, it is very close to zero and often has the wrong sign. Dalbar exceedingly and carefully constructs its survey of perceived service. The results indicate that, at least for index funds, either the variables we incorporate in the regression already capture the important aspects of service or the quality of service that the fund family offers does not matter to index fund investors.19 IV. How Well Do Investors Do?
In this section we examine the performance of investors who purchase S&P 500 index funds. At any point in time the dollar flows into index funds purchase a portfolio of holdings for which we can easily measure future returns. We compare the actual returns earned by investors to returns of naive portfolios and returns earned by investors making informed choices.
We first describe our methodology in more detail and then present our results.
Each month we form a portfolio where the weight on each fund is the net cash flow into There is a third explanation. Given the small number of funds for which we have service measurements, the sample may be inadequate to detect a relationship.
that index fund divided by the aggregate cash flow into all index funds. This converts the dollar cash flow into each fund into the fraction of the total cash flow invested in all S&P index funds that went into each fund. We then measure the return on this portfolio over one- and three-year holding periods. More specifically,
where RA,t,t+j is the actual annualized return earned on the portfolio of index funds bought by investors at time t assuming they hold the portfolio until time t+j, j is 12 or 36 months, N is the number of funds in our sample, CFi,t is the net cash inflow to fund i at time t,20 and Ri,t,t+j is the actual annualized return of fund i between time t and t+j.21 Equation (2) measures the actual return earned by new purchases of index funds assuming either a one-year or a three-year holding period. We construct a new portfolio each month from January 1996 to January 2001 for the one-year holding period and from January 1996 to January 1999 for the three-year holding period.
For both the one- and three-year holding periods, we construct reference portfolios each month to determine how well investors do. The first two reference portfolios examine the return earned when we use naive rules to form portfolios. The first is an equally weighted portfolio. Each month we allocate all net new cash inflow equally to each fund in the sample. The second is a market-weighted reference portfolio. We allocate each month all net new cash inflow across index funds in proportion to the total net assets of each index fund at that point in time.
We also compare actual performance with the performance that an investor could earn by selecting index funds using past data. If an investor recognizes that index funds We exclude funds with net outflows in the month this occurs. Our intent is to see how new cash invested in mutual funds performs.
Since we form a new portfolio every month, the returns on portfolios formed in successive months, while based on the same return series for individual funds, are different because the weights on each fund’s return are different.