This project develops a predictive model for sector rotation using high-frequency price data condensed into 30-minute bars to forecast the highest returning sector in the next interval. The sectors are defined by averaging the returns of DJIA stocks within each sector.

The model aims to predict one out of five sectors that will yield the highest return in the upcoming 30-minute interval.

Data is transformed from high-frequency price data to 30-minute bars, then to returns.

The project also considers alternate data such as GDP, Jobs, and SPDR ETF Daily Returns as part of the features (X).

The project addresses challenges such as insufficient granularity of external data and potential data leakage.

Random Forest LSTM XGBoost

The current models show poor performance (F1/AUC scores), with results not significantly better than chance. Future improvements may include careful data shifting and inclusion of alternate data to enhance predictions.

Data leakage is a concern; care is taken to not pair future X data with the wrong Y. Predictions are shifted accordingly to ensure we are forecasting a future winning sector (Y).

This project is part of a group effort, and contributions are welcome to improve the model's predictive capabilities.