The factor zoo is the name John Cochrane (2011) gave to the hundreds of stock-return factors that researchers have published. Most are variations on a few ideas, and many fail when tested on new data, which raises the question of which are real.
How large is the zoo
Harvey, Liu and Zhu (2016) catalogued over 300 factors published in academic journals. The count has kept growing since, as researchers test new combinations of the same accounting and price data.
Why so many look good
- Multiple testing. If hundreds of researchers test many variables, some will look significant by chance. See data snooping.
- Publication bias. Strong results are published and weak ones are not.
- Overlap. Many factors are close to each other, such as different measures of profitability or value.
- Small, illiquid stocks drive many results and are costly to trade. See transaction costs.
What replication found
Hou, Xue and Zhang (2020) reproduced a large set of published anomalies and found that a majority failed to be significant once microcap stocks were handled properly and stricter statistical thresholds applied. McLean and Pontiff (2016) found that returns of published factors fell by about a third after publication, consistent with both data snooping and arbitrage.
Judging a new factor
| Question | Why it matters |
|---|---|
| Is there an economic reason? | Signals without a reason are likelier to be noise |
| Does it hold out of sample and in other markets? | See out-of-sample testing |
| Does it survive costs and exclude tiny stocks? | Net returns are what matter |
| Is it new, given known factors? | Test against standard factors |
| Is the statistical bar high enough? | Harvey, Liu and Zhu suggest a t-statistic above 3 |
The few factors with long evidence, such as value, momentum, quality and low volatility, are the core of factor investing.
Frequently asked questions
What is the factor zoo?
It is the large collection of published stock-return factors, hundreds in number, many of which overlap or fail in later tests.
Why do many published factors fail?
Because of data snooping, publication bias, overlap among factors and reliance on small illiquid stocks that are costly to trade.
How many factors are real?
There is no agreed number. A handful, such as value, momentum, quality and low risk, have the strongest and longest evidence.
How should a new factor be evaluated?
Look for an economic rationale, out-of-sample and cross-market results, net-of-cost returns and a stricter significance threshold.