home>/ Methodology

Style Drift

When building a multi-manager portfolio, funds are usually selected based on the strategies that they follow. Selecting funds that belong to different strategies provides diversification in the portfolio. The strategy that a fund follows may change significantly over time. This is known as style drift. Identifying style drifts is important because they can give a warning signal about diminishing diversification benefits in the portfolio. Changes in strategies over time may also bring about undesired tilts in the portfolio exposure to market factors. Detection of style drifts can be done either through rolling window stepwise regression or through the Kalman filter technique.

The advantage of the rolling window stepwise regression method is that it allows you to track both changes in the factors to which the fund is exposed as well as in the magnitude of these exposures. The disadvantage of rolling window regression is that it is relatively slow to reflect changes in exposures to market factors. The Kalman filter overcomes this problem. The disadvantage of the Kalman filter is that it requires a lot more data compared to rolling window regression to work reliably. The key disadvantage of the Kalman filter however, is that it cannot be used together with a stepwise method to identify exposures to new factors. This means that the Kalman filter is useful to detect changes in exposures to a known fixed set of factors but not to detect changes in the factors to which the fund is exposed. In cases of style drift it cannot help you identify the new factors to which the fund is exposed.

 

 

 

Another way to approach the problem of identifying style drifts is to fit a factor model on the most recent data and try to replicate past performance through it. In case the factor model is representative for the current style and fits well the fund returns, we can verify if it is representative for past data as well. If the style of a manager does not change through time you can expect that the model will be able to replicate out of sample manager returns sufficiently well.

If the style of a manager changes over time, then using a model fitted on current exposures will produce returns significantly different from the ones actually generated by the fund. When style drift is detected, then the older observations may not be useful since they are no longer representative after the style drift occurred. Therefore, one approach is to replace them taking advantage of a backfilling method.