The Reaction of Mutual Fund Flow to Short-Term Returns

The Reaction of Mutual Fund Flow to Short-Term Returns
Whether investors take the short-run performance of mutual funds into account when making investment decisions has important ramifications for both investor education and the behaviour of fund managers.
by Steve Beveridge

A prevalent belief among mutual fund sponsors, managers and financial commentators is that investors place too much stress on recent fund performance. If investors do behave in this manner, their myopia leads to flitting from fund to fund chasing chance, instead of recognizing that such behaviour increases transaction costs and encourages the tendency to "buy high, sell low."

While there is still some dispute about whether longer-term fund performance acts as an indicator of future returns, the evidence seems overwhelmingly against using short-run returns to predict future performance.1 Nevertheless, casual empiricism suggests an undue emphasis on short-run returns: newsletters and newspaper articles often flag "hot funds" as those who have had stellar performances over the past few months, and the financial press and Internet sites report recent returns, thereby imbuing the numbers with more meaning and importance than they carry.

Several studies have documented a strong relationship between the inflow of new investment into a fund and the fund's performance over the past year or years.2 However, there is no corresponding evidence that investors consider a fund's short-run historical record when making buy and sell decisions. A short-run performance influence on investment patterns has a couple of important consequences.

First, such a relationship would indicate investor reaction to a noisy, inconsequential signal and demonstrate the amount of investor education that remains to be done. Second is the possible reaction by fund managers. When management remuneration is based on fund size, an element of "short-termism" may be introduced into fund management by trying to capture short-run related flows. Turnover costs would rise and the tax efficiency of the fund would fall as managers partially focus on short-run gains.

However, it has not been demonstrated that enough investors react to short-term performance that it is an issue to be concerned about. Does a detectable association exist between flow and short-run return? Additionally, is it return itself, or return relative to competitors that sways investors? This study examines these questions by evaluating the influence of recent return on Canadian equity funds' monthly sales and redemptions.

Investment Flows
Investment flowing in and out of a fund is represented by the proportionate change in the number of outstanding units. Two somewhat different derivations are used. One views reinvested distributions as new monies attracted to the fund. This quantity is referred to as "included flow," since reinvested distributions are included in flow. The other measure treats all distributions as automatically reinvested regardless of fund performance. Since distributions are excluded from flow, the series is consequently referred to as "excluded flow."

The period covered is February 1989 to November 1998 and uses around 12,000 fund-months of data. Complete details are provided in the appendix.

Table 1 displays the monthly behaviour of flow. Months without significant seasonality are not shown. Clearly, the treatment of distributions has an impact on measured seasonal fluctuation. Differences in monthly sales are most noticeable for January, June and December. Since the bulk of distributions occur in December, it is not surprising that sales take a jump up by about 2.3% when assuming reinvested distributions represent new investment. Notice that without this restriction, sales actually decline slightly in that month, probably due to a Christmas induced switch from savings to consumption by investors. The excluded flow series also seems to capture the RRSP season pattern better.

The Model
We want to gauge the impact of return measured over a short period ending this month on next month's flow. Three returns are used to calculate short-run return: the one-month, three-month and six-month returns. These returns are commonly reported in the financial press and other sources and are therefore likely candidates for influencing investor decisions. Table 2 (where R1, R3 and R6 denote one-month, three-month and six-month returns, respectively) shows that short-run returns are positively correlated with next month's sales. Although the correlation coefficients are not large, they are highly significant, suggesting that short-run return may be an important determinant of unit sales.

Correlation, however, does not adjust for the fact that returns themselves are correlated with each other and thus does not separate out their individual impacts. Moreover, since it is known that annual return influences sales and six-month return in particular is correlated with 12-month return, Table 2 may be showing the influence of annual return rather than short-horizon relationships. Consequently, 12-month return is included as a control variable. To put the returns on a similar basis, the three, six and 12-month returns are expressed as geometric monthly average returns.

Another factor that may be important is the relative performance of a fund. Perhaps it is not so much that a fund has a high or low return, but whether performance is high or low relative to competing funds. To judge the importance of this effect, two shift variables are related to each return series. When, for example, the three-month return is greater than the three-month return on an index of common stock funds, one of the variables allows for the possibility of the fund gaining extra flow because the manager has outperformed the other managers. The second variable allows the extra flow to increase with the magnitude of the superior performance.

Four non-performance control variables are included. First is lagged flow, which allows for stickiness in sales. Advertising campaigns, investor sluggishness when responding to performance and other effects can result in flows being related to the previous period's flows.

If performance attracts (or scares off) investment, does the same return bring in the same number of dollars regardless of fund size? In other words, do sales increase proportionately or absolutely? To control for the possibility that large funds cannot achieve the same proportionate growth in sales as small funds, the logarithm of total net assets is also added as a control variable.

Two other non-performance control variables are part of the model. One is a dummy variable to indicate if the fund is an index fund or not. At least with regard to return experience, it shows if investors treat index funds differently than those that are actively managed. The other is another dummy variable that flags funds offered by banks, trust companies and credit unions. This is an approximate separation between load and no load funds, as investors who are willing to pay for investment advice (load funds) may react differently than investors who do not seek advice (no load funds).

Fund risk is not included in the model, since monthly change in risk for these well-diversified portfolios should be negligible. In addition, including performance relative to the index of funds to some degree acts as a risk-adjusted benchmarking.

As a result, the initial model fit to the data is:
Flow i,t = (Flowi,t-1, Returnsi,t-1, Shifted Returnsi,t-1, Log(Assetsi,t-1), Indexi, Banki) where for fund i Flowi,t-1, Log(Assetsi,t-1), Indexi, and Banki are the four non-performance control variables, Returnsi,t-1 are the one, three, six and 12-month returns, and Shifted Returnsi,t-1 are their associated relative performance series.

Table 3 presents the final estimated model. Notice first that for most variables the results are not sensitive to how distributions are treated in the definition of flow. This may not be so if, say, bond funds are examined, but for stock funds distribution effects are small relative to all the other sources of variation.

Of all the performance variables, only three-month and 12-month returns are of any consequence. Although one-month and six-month returns were irrelevant, the presence of the three-month return shows that short-run return does sway some investors. It is not as important as annual return, either statistically or economically, but it does affect sales. A 1% increase in average monthly return over a three-month period enhances sales by about 0.3% - 0.4%, while the same average monthly increase over a year pushes sales up by roughly 1.4% - 1.5%.

None of the relative performance variables were noteworthy. As an example, included in the table is the 12-month relative return variable XR12. It potentially shows that flow rises with the degree of superior annual performance. However, although it is the most important of the relative performance variables, it is not statistically significant at conventional levels. Either the specification of relative performance is not a good representation of what investors evaluate, or relative performance over periods of up to one year is not as important to investors as commonly made out.

Clearly, the previous month's flow has the most statistically significant impact on current flow. There is inertia present in unit sales, with about 30% of a month's growth or decline being carried into the next month. This may strengthen mutual fund managers' incentives to pursue strategies that boost short-run return, since good short-run performance not only translates into higher flow next month, but via the autocorrelation into subsequent months as well.

Asset size is also important and has a negative impact. This indicates that large funds receive less proportionate flow than small funds, all else equal. One reasonable interpretation is that new flow comes primarily from investors who are not already in the fund. They can be thought of as a pool of investors whose money goes to the best performing funds and is withheld from poor performers. Their impact will, of course, be greater on smaller funds than on larger ones. If new flow derived mainly from existing unitholders, we would expect proportionate flow to rise and fall in the same manner across funds, because fund size depends directly on the number of unitholders, who would be the ones adding or withdrawing investment. One consequence is that excluding distributions from the flow variable is probably the more accurate of the two definitions, since including distributions as new flow relies on existing unitholder reaction.

Of the other two non-performance control variables, the one that flagged index funds was of no importance. In fact, dropping index funds out of the sample and re-estimating the model has no important effect. It appears that everything else held constant, index and actively managed funds have the same chance of attracting or losing sales. On the other hand, when distributions are not defined as new flow, the bank fund variable is significant at the 10% level. That is to say, there is a slight tendency for more flow for no load funds, all else the same. This is not surprising, given that load fees are a barrier to fund entry and exit.3

Short-run performance does have a detectable effect on mutual fund flow, and three-month return is found to be the best measure of short-term performance. Moreover, return itself is important, not return relative to returns on competing funds. Short-term performance does not influence unit sales and redemptions as much as annual performance, but itseffect is significant enough to warrant including short-run return in flow performance models. In other words, because short-term returns provide no useful information about future fund performance, this paper documents the presence of noise trading in mutual funds.

No load funds, or at least funds offered by financial institutions, appear to be slightly more sensitive to the performance-flow relationship than other funds. No difference is discovered, though, between index and actively managed funds. Finally, the effect of performance on proportionate flow declines with fund size.

Appendix Data
The individual fund information comes from Funddata Canada Inc. and Financial Results Ltd. and the second source provides the index of mid- to large-cap equity funds. Because during their early life most funds grow rapidly regardless of their return performance, the first 12 months of data are discarded for new funds. Mergers are handled by throwing out the two months of asset growth rates influenced by such events.

The analysis covers the period February 1989 to November 1998 and the sample consists of Canadian common stock funds that are not defined as small cap. Included in the sample are most of the funds that dropped out along the way, i.e., it is not a sample of survivors. Be that as it may, Sirri and Tufano (1998), Chevalier and Ellison (1997), and Goetzmann and Peles (1997) have shown that inferences on flow and performance are insensitive to survivorship bias. The end result is a data set of about 12,000 monthly observations.

Two potential biases remain. First, funds that are closed to new investors will distort the flow-performance relationship. As far as could be determined, though, this is a concern for less than 0.5 % of the sample. Another possible bias is funds with large minimum investments. For example, flows into a fund with a minimum investment of $5,000 or $10,000 or more may not be sensitive to short-run return. Again, this possible influence is present in only a small part of the sample. Notice that both effects bias against finding a flow and short-term performance association and are thus of concern mainly if no relationship is found.

1. See, for example, "Win Some, Lose Some," J. Ramseyer and L. Ackert, Canadian Investment Review, Fall 1996, pp. 9-11.

2. R. Ippolito, "Consumer Reactions to Measures of Poor Quality: Evidence from the Mutual Fund Industry," Journal of Law and Economics 35, 1992, pp. 45-70; M. Berkowitz and Y. Kotowitz, "Investment Management Services," Canadian Investment Review, Winter 1998, pp. 42-48; and E. Sirri and P. Tufano, "Costly Search and Mutual Fund Flows," Journal of Finance 53, 1998, pp. 1589-1622, provide evidence on U.S. funds and M. Berkowitz and Y. Kotowitz, "The Market for Investment Management Services: Is it Rational?" Canadian Investment Review, Spring 1991, pp. 69-75, show the same for Canadian mutual funds.

3. While it is true that most load-fund companies do not charge a switching fee, in most cases fund retailers are free to do so. In addition, more likely than not, the best fund to switch into is in another family.

Steve Beveridge is a professor at the Faculty of Business, University of Alberta in Edmonton. This article was made possible by a grant from the Canadian Investment Review's Adademic Sponsorship Program.

Transcontinental Media G.P.