A Practical Order for Factor Cleaning, Imputation, Scaling, and Neutralization
Summary
The document gives a suggested sequence for preparing quantitative factors: remove extreme values, handle missing observations, standardize the data, and then neutralize it. It describes neutralization as removing exposures that can obscure a factor’s relationship with stock returns, with market capitalization and industry given as common sources of unwanted influence.
Daily trading value illustrates the issue: larger companies tend to trade more, so the raw factor may partly reflect company size rather than an independent relationship between turnover and returns. Neutralizing for size can help isolate the factor’s contribution. The guidance is concise and general; it does not specify particular methods for outlier treatment, imputation, scaling, or regression, nor does it discuss cases where ordering choices may depend on the data or model. No empirical comparison or performance evidence is provided.
Key ideas
- The suggested preparation order is outlier removal, missing-value treatment, standardization, and neutralization.
- Neutralization can reduce the influence of market capitalization or industry on a factor.
- Daily trading value may partly proxy for company size, complicating interpretation of its relation to returns.
- The document gives general guidance without comparing procedures or measuring their effects.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.