Needles, Haystacks, and Hidden Factors

Abstract

Statistical factor modeling is often described as a way to identify commonality among returns in a financial market. Statistical models examine returns over many time periods, and from them identify relationships between and among the different assets, unlike fundamental factor models, which from the outset group assets that are likely to experience similar returns. Yet the statistical approach is better at finding some types of factors than others. Statistical factors generally work best with high-frequency return data, and even then may not pick up distinctions that apply to a relatively small subset of assets (such as distinctions associated with industry membership); they are most useful when they supplement fundamental factors. We see this in a case study that adds statistical factors to an MSCI Barra fundamental factor risk model.