Trend-Following versus Cross-Sectional Momentum: A Data-Driven Statistical Factor Comparison

Abstract

The persistence of returns for an asset (momentum effect) is well known and robust: Practitioners use a variety of different approaches and methodologies to capture it. They often distinguish between trend-following (or time series) and cross-sectional momentum models. This article presents a comparison of the two methodologies based on statistical factors. For that purpose, we create trend-following and cross-sectional models in four asset classes (Equity, Forex, Interest Rate, and Commodity). We show that the first have relied on a dynamic (time varying) exposure to a few dominant statistical factors to generate performance. Interestingly, the predominant factor can easily be interpreted as the exposure to the asset class. On the other hand, cross-sectional models are more diversified but historically underperformed trend-following models. They, however, yield interesting results in more heterogeneous asset classes (for instance, Commodity and Interest Rate). Finally, we show that the categorization is of minor importance, as returns are driven by exposure to factors, which can be modified in multiple ways, such as, for example, by changes in the portfolio construction methodology.

TOPICS: Quantitative methods, portfolio construction, factor-based models

Key Findings

• • Trend-following models by asset class have relied on a dominant statistical factor to generate performance, which can be interpreted as exposure to the asset class.

• • Cross-sectional models are more diversified and appear to be more interesting in heterogeneous asset classes as they tend to hedge the dominant statistical factor.

• • The categorization in cross-sectional and trend-following models is of minor importance, as they might both load on similar performance drivers depending on portfolio constraints or portfolio construction methodologies.