This is the public version of a machine learning model that can very accurately predict movements in stock prices. Because much of that accuracy is attributed to the data, I have kept that for myself. However, the model architecture that I used is free for others to use.
Key Points:
- For CV, I employed combinatorial purged cross-validation, useful for financial applications
- I created a manual stacking ensemble of three models (more can be added):
- XGBoost: widely regarded as powerful for financial machine learning
- TabNet: useful for tabular datasets
- Logistic Regression: although basic on its own, good for ensembling
- I tuned the hyperparameters of each model in the ensemble, as well as the meta-learner
- I used XGBoost as both a base model and as the meta-learner
- The scoring function used in tuning must be implemented (I removed it because it gave too much about my strategy away)