Factor Zoo and Multi-Factor Models [/ˈfæktər ˈzu ənd ˈmulti-faktor ˈmɑdəlz/] n - The factor zoo names the observation that hundreds of claimed factors have been published in academic finance. Most do not survive out of sample; a few only are robust, economically grounded, and widely traded. The labor lies in telling the durable from those mined upon the same datasets everyone uses.
The classic starting point is the Fama-French three-factor model:
Market, size, and value were later joined by momentum and quality. The Fama-French five-factor model adds profitability and investment; the q-factor model offers an alternative founded on investment theory. Each family keeps its defenders and its critics.
A factor earns its place only if it carries independent information. Two factors nearly redundant raise complexity without improving forecasts; VIFs and correlation matrices should be checked before they are combined. See Linear Factor Models in Trading for the regression mechanics.
Data mining is the central danger. A factor formed by sorting on a ratio that chanced to work in the past may have no economic mechanism behind it. The strongest factors come with a story: compensation for risk, behavioral bias, institutional frictions. Without a story, a factor is a pattern and not an edge.
Crowding makes even real factors perilous. When too many managers tilt toward value, momentum, or low volatility, the factor can suffer long drawdowns and sharp liquidations. See Alpha Decay and Model Lifecycle for the watching of a factor grown crowded.
The practical counsel is plain: begin with a small set of well-understood factors, and admit new ones only when they improve out-of-sample predictions or risk-adjusted returns. A model of twenty factors is not twenty times better than one of three; it is usually worse.