Unique Factors

Abstract

It is well known that 70% of individual stocks' returns are classified as idiosyncratic returns under a conventional asset pricing model. In this study, we raise an important question as to whether majority return variations are truly influenced by idiosyncratic risks that at most affect several stocks from an empirical perspective. In other words, we explore a possible middle ground between common risk factors that influence almost all stocks and idiosyncratic risks by proposing unique factors that may only affect certain groups of stocks. In particular, we present a simple iterative approach both to extract unique factors from individual stock returns and to group stocks simultaneously. Comparing with industry groupings, this approach not only allows for a parsimonious group structure but also ensures that unique factors are indeed unique to individual groups. With the deterioration in the explanatory power of common factors such as the Fama and French factors in recent years, we find that unique factors not only are pervasive within their groups but also have played an increasingly important role in explaining individual stock returns over the past forty years. As an example, a multifactor model with two common factors and two unique factors not only provides a 10% extra explanatory power over other popular models but also outperforms the Fama and French model in out-of-sample tests. Therefore, the simple structure of unique factor model may hold the key to improve the performance of an asset pricing model in general.

Keywords: Explanatory Power, Fama and French Factor, Heterogeneity, Idiosyncratic Volatility, Industry factor, Multifactor Model, Principal Component, Unique Factor

JEL Classification: G10, G12