Abstract:
This study examines the predictive capability of machine learning and hybrid Principal Component Analysis PCA based models to forecast stock market returns. This study considers key features of the macro economy such as money supply, oil prices, exchange rate, gold prices, inflation and Quantum Index Manufacturing. The analysis began with correlation analysis that showed significant positive relationships between money supply, exchange rate, inflation and QIM. Contrarily, oil prices had consistent negative relationships with all major variables, indicating the existence of nonlinear and interdependent economic dynamics. Baseline forecasting models showed different performances and the best one was LightGBM PCA in terms of accuracy. The integration process of PCA enhanced the performance of the model to a great extent, especially LightGBM PCA, which contained the lowest error values RMSE 0.216 and MAE 0.080. SVR PCA also showed moderate improvement, and the RandomForest PCA showed no change much. Comparative evaluation confirmed that hybrid PCA models are more effective than baseline models in terms of improving the reduction of noise and the multicollinearity and then can represent latent economic structures more accurately. The study comes to the conclusion that the use of hybrid PCA augmented machine learning models offers robust and reliable forecasting capabilities with evident benefits as compared to standalone approaches for capturing the complexities of the financial markets.
