Disposition effect and mutual fund performance

Abstract

This article finds strong evidence for the presence of the disposition effect among US mutual fund managers. The analysis can establish a link between the disposition effect and mutual fund characteristics as well as changes in the macroeconomic environment. Managers with a lower disposition effect are found to invest in larger equities with a higher trade volume, a higher past performance, lower idiosyncratic risk, and a higher risk-adjusted performance. However, fund characteristics and the economic environment can only explain a limited amount of the variation in the disposition effect across mutual funds. Using a new methodology to reduce the disposition effect exhibited by mutual fund investments, we find no increase in their profitability. Although statistically significant, the disposition effect has only a minor economic effect on fund performance.

Keywords:

• disposition effect
• mutual fund performance
• behavioural finance

JEL Classification:

• G11

Notes

¹ Odean (1998) shows that US investors realize tax loses in December, which is the end of the US fiscal year.

² Da Silva Rosa et al. (2006) use daily holding positions from September 2001 to September 2004 for all 4264 UK managed funds and find a disposition effect that is lower than the one reported by Odean (1998).

³ The analysis of mutual funds is particularly interesting because they have low transaction costs, thereby reducing the noise level in the data and making the measurement of the disposition effect easier and more reliable (see Shefrin and Statman, 1985).

⁴ However, mutual funds having a ticker for some but not for all records are kept.

⁵ This number is subject to sensitivity tests, but the funds deleted are comparably insensitive to this threshold.

⁶ The price of the stock is taken at the reporting date. This corresponds to the procedure followed by Frazzini (2006).

⁷ The average purchase price is calculated as the purchase price of the stock weighted by the amount of shares bought as given in Odean (1998), Grinblatt and Keloharju (2001) and Da Silva Rosa et al. (2006).

⁸ By our procedure of calculating realized gains and losses we do not include possible gains or losses that result by shares bought before 1993 so that we will have a survivorship bias.

⁹ In this approach we assume that the disposition effect is independent across mutual funds.

¹⁰ 23 mutual funds are excluded because of computational reasons such as not having a sale during their reporting period or selling shares for which the purchase price is not known.

¹¹ There exists a higher disposition effect from 1993 to 1999 than in the second subperiod.

¹² In this approach PGR, PLR and DEM are calculated by aggregating paper gains, paper losses, realized gains and realized losses across all dates and mutual funds on a number of shares basis. This corresponds to the basic approach for the calculation of the disposition effect chosen by Odean (1998). Using this approach, we obtain a relatively low disposition effect of 0.016, but we can still reject the null hypothesis at the 1% significance level (t-statistic of 41.79). This value is in line with results by Frazzini (2006) as well as Da Silva Rosa et al. (2006) (values of 0.031 and 0.0012 with t-statistics of 43.6 and 28.7, respectively) for professional investors, but is significantly lower than the values obtained by Odean (1998) for private investors (value of 0.050 with a t-statistic of 34.98). Thus, the disposition effect appears to be more dominant among private investors.

¹³ Since the correlation between mutual fund characteristics and the economic environment is low, we do not include these variables as control variables.

¹⁴ Correlations are between −0.29 for the volatility of the S&P500 and the GNP growth and 0.25 for the mean returns of the S&P500 and consumer confidence. Only for the correlation between the volatility and the mean return of the S&P500 we obtain a larger correlation value of −0.43. However, since we work with a very large dataset, multicollinearity does not pose a problem.

¹⁵ Results are not reported in tables for brevity.

¹⁶ However, repeating the analysis with the available daily data confirms our results.

¹⁷ The disposition effect reflects a behavioural pattern which is similar to momentum and reversal investment. Possibly, the four-factor model erroneously captures the impact of the disposition effect on alpha in the momentum component. Therefore, we repeat the analysis with the three-factor model by Fama and French (1993), which does not have a momentum component and obtain similar results. The disposition effect is the largest (value of 0.1477) for fund managers with the lowest alpha (value of −0.0167). Although in general an increase in alpha goes along with a lower disposition effect, the link between the disposition effect and the performance is not linear.

¹⁸ The intercept of the regression is −0.0050 and has a t-statistic of 36.863. Cici (2005) finds in his regression a lower value of −0.0110 (t-statistic: −2.60).

¹⁹ Regressing the Fama–French alpha of the fund on the disposition effect returns a significant parameter of −0.0029 (t-statistic: −5.145) for the disposition effect figure. Cici (2005) finds a value of −0.0219 (t-statistic: −4.15) for the three-factor model. With the average disposition effect measure of 0.082, our result amounts to an average performance impact of −0.02%. The impact of the disposition effect on the mutual fund performance is weak from an economic perspective. The low R² of 0.0110 lends further support to this claim. Therefore, our results are robust to the usage of momentum in our risk factor regressions.

²⁰ The sale proceedings of a stock that is held longer in our approach might have been originally used to buy another stock. By postponing the sale, this new stock might have to be financed by a credit, causing costs we do not consider. However, since this approach is solely taken to analyse the disposition effect, this simplification is reasonable.

²¹ Our results could be biased because we assume all transactions take place at the reporting period although in reality they take place between the two reporting dates. To control for this possible mismatch, we repeat the calculations with prices 1 month before the reporting date. However, the results are essentially unchanged.

²² Minor changes in PGR can be attributed to the fact that changes in sales of losing stocks have an impact on subsequent holdings and, therefore, on PGR if the price increases and the position becomes a gain.

²³ The reason for the decline of the realized gains is that the amount of shares that are sold with a gain sometimes is smaller than in the standard case due to the fact that there are less shares left in the portfolio that can be sold.