Random Forest Stock Selection with Price and Volume Factors
Summary
This article explains random forests and decision trees, then applies a random forest regressor to short-horizon Chinese stock selection. It describes bootstrap sampling, random feature selection, averaging tree predictions, tree pruning, regression split criteria, model parameters, and feature-importance measures. The input features are derived from price and volume data, while the target is five-day stock return. The strategy ranks stocks by predicted return and forms a periodically rebalanced portfolio.
The study compares model settings and reports out-of-sample fit statistics and backtest results, including annual return, Sharpe ratio, and drawdown. Its experiments suggest that feature subsampling materially changes results, and it also describes annual rolling retraining to adapt to changing market conditions. The evidence is specific to the stated Chinese A-share sample and test periods; the authors note that performance varies across periods and that the parameter analysis, training-window choice, and factor set need further investigation. The reported backtests do not establish that the strategy will generalize to other markets or future data.
Key ideas
- Random forests combine bootstrap-trained decision trees and average their regression predictions.
- Random feature selection adds diversity between trees and can affect out-of-sample performance.
- The study builds price and volume factors and predicts five-day returns for Chinese stocks.
- Its portfolio ranks stocks by predicted score and rebalances on a five-trading-day schedule.
- Annual rolling retraining is presented as a way to respond to shifting market styles, though results vary by period.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.