Quantile Regression for Multi-Factor Stock Selection
Summary
The document explains quantile regression as an alternative to ordinary least squares for modeling stock returns. Because it estimates relationships at different points in the return distribution, it can reveal effects that an average-based model may hide, and it is presented as more robust to heavy tails and changing variance. The example studies China Securities 500 constituents using market capitalization, prior-month return, and average daily turnover as factors.
Prior-month return is associated with reversal near the middle of the distribution, while its estimated relationship is steeper at the 10th percentile and turns positive at the 90th percentile. For ranking stocks, the document recommends choosing a statistically significant quantile with a large absolute slope. It then compares a 10th-percentile model with OLS in benchmark-relative portfolios, reporting stronger excess returns and Sharpe ratios for the quantile approach, with similar volatility, drawdowns, and turnover. The source provides no underlying tables, sample dates, transaction-cost assumptions, or detailed testing design, so the reported comparison cannot establish broad or forward-looking performance.
Key ideas
- Quantile regression estimates factor relationships at selected points across the return distribution.
- It can be useful when returns have heavy tails or nonconstant variance.
- The prior-month return factor shows different estimated effects at different quantiles, including a reversal pattern near the center and a positive slope at the upper quantile.
- For stock ranking, the document favors a statistically significant quantile with a large absolute slope.
- Its reported portfolio comparison favors the 10th-quantile model, but methodological details and generalizability are limited.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.