Training a GBDT Model for Multi-Factor Stock Selection

Summary

This brief tutorial describes replacing the default stock-ranking model in a visual AI strategy template with gradient boosted decision tree training and prediction modules. It gives an example configuration: a regression loss, root mean squared error as the evaluation metric, and a tree-based model. The intended workflow is to train the model on the strategy's factor data and use its predictions for stock selection.

The document is mainly a setup recipe rather than a full strategy explanation. It does not identify the factors, target construction, training and validation periods, parameter choices beyond the listed settings, or portfolio and execution rules. It also reports no test results or comparison with the default model, so it offers no evidence that the substitution improves selection quality. These omissions limit reproducibility and make the example a starting point for experimentation rather than a validated approach.

Key ideas

• The example replaces a default stock-ranking model with GBDT training and prediction modules.
• It configures the learner for regression and evaluates predictions using RMSE.
• The model uses a tree-based architecture.
• The note does not specify factor inputs, validation design, portfolio rules, or performance results.

Tags

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