The return of a mutual fund can basically be divided into two main parts: the alpha, which is generated by the manager’s selectivity skill, and the beta, which is related to the exposure to the stock market. The fund can decide to play on the market exposure depending on its market direction forecasts: this is market timing behavior. On the basis of a more detailed view of the market exposure, the authors develop an extended model to decompose mutual fund returns. They add three non-skill components to the selectivity and market timing, which refer to manager skill: the long-term market exposure, the exposure to the current macroeconomic situation and the market exposure in the recent past.

          Mutual funds that take market timing decisions are called asset allocation mutual funds. The sample extracted from the database provided by Morningstar is composed of 78 asset allocation mutual funds. The data runs from June 1972 to May 2002, with an inception prior to March 1995 for the funds that do not exist over the entire period.

          Considering the estimation results for the alpha, 47 funds have a negative estimate. The median is –3 bp. The 80 percent interval around the median ranges from –18 bp and 16 bp per month. This dispersion shows that the managers are not all equal in terms of selectivity. A small share of the funds exhibit a statistically significant alpha.

          The market timing is expressed by the correlation of the exposure to the market with the excess return on the market in the same month. 47 funds have a positive estimate. The median is close to 0. The 80 percent interval around the median ranges from –8.7 bp and 12.1 bp per month. Six mutual funds have a statistically significant coefficient.

          According to the results, the selection of a fund with both top decile alpha and timing results in an expected return which exceeds that of a fund with both bottom alpha and timing by 55.5 bp per month. However, previous studies have highlighted the fact that timing ability and selectivity are negatively correlated. This is due to the fact that the manager artificially improves timing by buying put options. At the same time, this reduces the selectivity level. It is confirmed by the negative correlation between selectivity and timing of –0.71 found in the sample. Moreover, the average return of the sum of selectivity and timing is close to zero.

          The long-term market exposure exhibits a median of 0.61. The 80 percent interval around the median ranges from 0.26 bp and 0.88 bp per month. The expected return that can be attributed to this component (equal to the long-term exposure multiplied by the conditionally expected risk premium) is 34 bp per month.

          Focusing on the market exposure in the recent past (in other words a delayed reaction), it appears from the sample that there is a short delay to adapt the fund’s exposure. However, for eight funds the coefficient is more important, indicating that the exclusion of this term may lead to an underspecification of the model.

          The last term is related to the exposure to the current macroeconomic situation. The average contribution of this term is of 2.8 bp per month, and the 80 percent interval around the median ranges from –2.5 bp and 8.0 bp per month. Even if the return attributed to this factor is weak, its positive contribution for three quarters of the funds suggests that a large part of the funds may use the public forecast to increase their returns.

          To examine the variation in the market exposure over time, the authors use the minimum and maximum estimated exposure. At the 78 funds level, the median from the time-series minima (respectively maxima) is 0.21 (0.92), indicating a strong time-series variability in the market exposures. This leads the authors to allow market exposures to change over time in the model, through a random exposure shock. On the basis of an equally weighted fund of the 78 mutual funds of the sample, the average contribution of each of the terms is calculated. A cross-sectional regression exhibits selectivity and timing with a very low contribution, while the long-term market exposure explains the largest share of the return over the whole period.

          The authors present a comparison of the market-timing skills results obtained by their extended model with those obtained from previous models, namely Treynor and Mazuy's (1966) traditional timing model, Lockwood and Kadiyala's (1988) stochastic timing model and Ferson and Schadt's (1996) conditional timing model. For most of the funds the timing coefficients are similar, but several funds exhibit considerable coefficient differences, highlighting the importance in such cases of a more specified model than a general model.