|
Constructing Peer Benchmarks for Mutual Funds: A Style Analysis-Based Approach Authors: Arik Ben Dor, Vernon Budinger, Lev Dynkin, Kenneth Leech Source: Journal of Portfolio Management, vol. 34, n°2 Date: Winter 2008 |
In this article, the authors stress the importance of being able to accurately evaluate the performance of mutual funds. The performance of funds is generally evaluated relative to a benchmark. Most of the time, however, the methodology used to derive the benchmark does not make it possible to have a suitable representation of the fund to be evaluated. The authors mention the use of a single index as a representative benchmark, which does not allow representation of all the asset classes a fund may be invested in. Another more sophisticated approach involves building a benchmark based on mutual fund holdings. In this case, the problem will be collecting the data, as it is difficult to obtain accurate information on fund holdings on a regular basis.
In response to these shortcomings, the authors propose benchmarks based on Sharpe’s (1988, 1992) return-based style analysis. Using only historical fund returns, return-based style analysis makes it possible to construct a passively managed portfolio that replicates the style of the portfolio to be evaluated. The exposures of the fund to different style indices are evaluated using a constraint regression, so that these exposures represent portfolio holdings, i.e., so that they are positive and add up to one, and so that they minimise the unexplained variation in returns.
To go one step further, the authors explain that this model, suitable for a fund managed by a single manager, may be extended to an aggregation of funds managed by several managers, by adding up the exposures of all managers for each asset class weighted by the proportion of the whole portfolio managed by each manager.
To illustrate the application of the model for the construction of a benchmark, the authors consider 38 multi-sector bond funds with data covering the eight years from January 1999 to July 2006. These funds can include investment-grade debt securities, high-yield, emerging market, and international bonds. They can also be exposed to various currencies and have a limited exposure to equity. The authors select representative indices to cover the asset classes that were part of these funds, leading to a ten-factor model. Using the data for the whole period, they compute a benchmark both for each fund considered individually and for the equally-weighted aggregation of the 38 funds. The adjusted R² obtained for each fund individually is quite high, with an average of 84%, to be compared with a value of 32% obtained when evaluating funds with a single index to represent the category of funds. The authors also note that the explanatory power of the style benchmark derived for the aggregate fund, which is equal to 96%, is much higher than that obtained on average for the individual funds. They explain that the elimination, in the aggregate fund, of the bulk of the idiosyncratic risk of each fund accounts for this result.
In addition, they compute the same regression using rolling windows with 36-month data to evaluate the evolution of fund asset class allocation. Their results reveal that the change in exposure to style is sufficiently progressive to allow the style benchmark to adapt over time in response to these changes.
The authors conclude that the model they propose can be implemented with successful results both for performance and risk analysis.
References
Sharpe, W. F., “Determining a Fund’s Effective Asset Mix”, Investment Management Review, 1988, pp. 59-69.
—, “Asset Allocation: Management Style and Performance Measurement”, Journal of Portfolio Management, vol. 18, n°2, 1992, pp.7-19.
