금융변수들과 주가 수익률 간의 관계에 관한 연구: 통계모형과 기계학습 모형 간의 효율성 비교

To analyze the impact of macro-financial variables on stock returns, we employed both traditional regression models and machine learning methods. For this purpose, we estimated statistical regression models with time series data that account for weak stationarity and autocorrelation, specifically the multiple linear regression model and the ARDL model. In addition, using the same dataset, we trained representative machine learning models, including the SVR model, tree-based models such as XGBoost and Light GBM, and the deep learning-based LSTM model, to generate learning outcomes. The estimation and training results indicated that machine learning models generally outperformed statistical regression models. Among the machine learning models, the LSTM model demonstrated the highest efficiency, followed by the SVR model and the tree-based models such as Light GBM model and XGBoost model. These results imply that combining both methodologies can allow researchers to leverage the financial insights obtained from statistical regression models while simultaneously benefiting from the measurement and predictive efficiency of machine learning models.