Decision Tree for Fast and Slow Momentum

Article BigQuant

Summary

This article presents a time-series momentum strategy that switches between a short lookback signal and a longer one according to market volatility. The short signal uses one month of returns and the slow signal uses twelve months; each takes a long or short position based on the sign of its past return. The motivation is that slower signals can capture persistent trends in calm markets, while faster signals may respond better to turning points in turbulent markets, though they can react to false turns.

A decision tree is trained on periods when the two signals disagree, using historical monthly S&P 500 volatility to select which signal to use next. The account reports a volatility threshold learned from an earlier training period and says the rule performed better out of sample than either signal alone or a fixed blend. It attributes returns mainly to market timing and also discusses volatility timing through alpha and beta decompositions. The cited comparisons and claims are reported without underlying figures here, and the article notes that differences in alpha are not statistically significant.

Key ideas

The strategy uses the sign of past returns to take long or short positions at two different lookback speeds.
It uses market volatility in a decision tree to choose between fast and slow momentum when their signals disagree.
The article reports that slower momentum performs better in calmer conditions and faster momentum in more volatile conditions.
Its out-of-sample comparison favors the volatility-based switching rule, though supporting charts are absent from the text.
The reported alpha differences between the fast and slow signals are not statistically significant.

Using a Decision Tree to Switch Between Fast and Slow Momentum

Summary

Key ideas

Tags