Machine-Learning Stock Selection with Rolling Factor Features

Summary

This overview describes a practical Python workflow spanning data collection, database access, and machine-learning, deep-learning, and natural-language-processing tools. Its investing example uses traditional factors measured over rolling 12-month windows as features. Stocks are ranked by subsequent returns to define relative strength groups, and a machine-learning model uses current factor values to predict relative performance for the following month.

For the portfolio study, the overview selects the top and bottom portions of predicted rankings as long and short groups, with equal weighting and industry neutrality. It reports an out-of-sample period ending in November 2017, along with annualized volatility and maximum drawdown, but the annualized long–short return difference is missing from the supplied text. The source is an overview rather than a complete methodological account: it does not identify the model algorithm, provide full data and portfolio construction details, or establish how robust the reported results are. The figures should therefore be read as a limited summary, not as sufficient evidence for replication or deployment.

Key ideas

• The example uses rolling 12-month traditional factor values as stock-level model features.
• Stocks are ranked by subsequent returns to define stronger and weaker groups for training.
• The model predicts relative stock performance over the following month from current factor values.
• The portfolio study takes long and short positions in the high and low ends of predicted rankings.
• The overview reports volatility and drawdown but omits the annualized long–short return difference.

Tags

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