Random Forests for Short-Horizon Equity Factor Research

Summary

The article introduces random forests as an ensemble of decision trees and explains how bootstrap sampling and random feature selection create diversity among trees. It reviews classification and regression trees, impurity measures, pruning, regression losses, and ways to combine learners. It also outlines common forest settings and model outputs, including tree count, depth, feature sampling, predictions, and fit scores.

For the empirical setup, it describes deriving short-horizon stock-selection factors from market data such as prices, volume, and turnover, using statistical aggregations including correlations, dispersion, extrema, sums, and weighted averages. The supplied text stops while introducing factor expressions, so it does not provide completed experiments, performance results, or enough detail to assess the proposed stock-selection model. The discussion is primarily a conceptual and implementation overview; its parameter recommendations and general statements should not be treated as validated trading guidance.

Key ideas

• Random forests average or vote across decision trees trained on bootstrapped samples and randomized feature subsets.
• Tree splits use different criteria for classification and regression, while pruning and tree limits can help control overfitting.
• Regression trees commonly minimize squared or absolute error, with squared error more sensitive to outliers.
• The proposed stock-selection inputs are short-horizon factors derived from price, volume, and turnover data.
• The provided article excerpt does not report completed model results or trading performance.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.