Nonlinear momentum - Increase positions with signal strength

 

Many firms are using trend-following. Their distinctions are often based on the length of trend signals, the set of markets used, and the risk management techniques employed. A recent paper, Nonlinear Time Series Momentum, measures the nonlinear relationship between trend and risk-adjusted returns using machine learning techniques. Using techniques to exploit nonlinear relationships within momentum outperforms simple linear methods. This nonlinear value-added is observed across all asset classes, frequencies, and horizons and lookback periods. This is especially true during market downturns. 

For modelers, this research concludes that simple nonlinear transformation of momentum signals will improve strategy performance, and signal strength interacts with predictability. However, as signals move to extremes, the extra risk from adding to position sizing diminishes, and at some point is not worth taking the extra risk. Hence, there is a complex nonlinear function between signal strength, sizing, and optimal risk-adjusted returns. Increase position exposure when signals are stronger, but reduce the signal exposure as you move to extremes.