A new paper from the smart researchers at Resolve Asset Management, "Global Equity Momentum - A Craftman's Perspective" shows the value of ensemble modeling for momentum and trend-following. Their careful research suggests that there is nothing inherently wrong with trying to find the best look-back specification for a trend and momentum model. Many specifications work and this is a testimony to the robust nature of trend and momentum styles. Nevertheless, the dispersion in performance relative to a median can be quite large. This style may work but if you happen to have picked the "wrong" look-back period you will be disappointed. This suggests that choosing the median look-back across a broad range of choices makes sense, but investor can do even better if they blend or diversify momentum/trend timing approaches.
Instead of just looking at the historical record, the folks at Resolve ran extensive bootstrap analysis coupled with hundreds of variations to provide distributional properties on the look-back choice set. Many specifications seem to work and generate positive returns with good information ratios, but a better approach is to blend timeframes. Timeframe diversification will reduce risk and dispersion relative to a median specification; consequently, blending look-back periods is an effective way of reducing underperformance or regret. There are well-defined approaches for this blending through the use of ensemble techniques.
Many modelers have intuitively known that time scale diversification works and have incorporated this knowledge in forming more robust trend-following and momentum models, yet few have done the exhaustive work to truly study the value of diversification. As important, the Resolve researchers have shown how to develop a robust approach to find the best ensemble over some ad hoc blend. Their application to a well-known data set and modeling approach for global equity momentum should satisfy even the most skeptical investor that there is a value with being a model craftsman.