Trading Signal [COMPETITION-5]

This competition focuses on developing a cross-sectional trading signal using walk-forward evaluation and correlation clustering analysis. Empirical validity for the AlphaNova competition is achieved through a sequential walk-forward framework that implements leak-free feature engineering and strict cross-sectional de-meaning on the prediction outputs [1]. The use of XGBoost with high regularization (L1/L2) and feature engineering based on cross-sectional ranking is designed to generate unique signals with an angular distance of >60° from existing signal cities, aiming to maximize the Sharpe ratio [1]. To test the empirical validity locally, run the command python runner.py submission_xgb.py --gauge-fix to ensure the model does not overfit, complies with time constraints, and passes the novelty check.