Decision Trees for Classifying Daily Stock Moves
Summary
The document introduces decision trees as supervised models for classifying a stock’s next daily move as up or down. It outlines a workflow using historical OHLCV data, technical indicators such as RSI, moving averages and ADX, and a target class derived from daily returns. Indicator values are lagged to reduce look-ahead bias, and the dataset is divided into training and test sets before fitting the model.
It explains how a tree routes observations through a root split, internal tests and leaf predictions, then describes recursive tree construction. Split criteria such as information gain and the Gini index select attributes, while stopping rules and pruning limit further growth and classification error. The example illustrates how indicator thresholds can form trading rules, but the document supplies no measured strategy performance or detailed validation results. Its discussion is introductory: a learned tree can overfit, and a train-test split alone does not establish that a strategy will generalize or account for trading costs.
Key ideas
- Decision trees classify observations by routing feature values through sequential threshold tests.
- Lagged technical indicators can serve as predictors while reducing look-ahead bias.
- Training data determines the tree’s splits, and a separate test set is used to assess predictions.
- Information gain or Gini criteria can guide splits, while stopping rules and pruning constrain tree growth.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.