---

Residual Stock Return Prediction: Data Challenge by QRT (see ENS website)

This project concerned a binary classification, predicting whether a stock will outperform the median over total market residual return in the next day. I implemented and fine-tuned a stacked model of random forest and gradient boosted tree ensemble in XGBoost and LightGBM.

My model attained test set accuracy of 52.18%, putting me at 62nd place (out of 950) in the public ranking (as of 5 May 2025).

To achieve this performance, I crafted features for various GICS groupings and well as stock-level features and carefully considered methods of model regularisation for the ensembles.

Using the Polars dataframe package, I engineered the project code to be extensible and conducive to quick research iteration via cross-validation experiments.