「預測」之所以誘人，在於對未來有效的掌握度。本文使用多種降維技術，其目的是有效利用多維度數據，提升預測準確性和投資決策的可靠性。 ;This study examines the impact of various stock trading indicators on individual stock return prediction and explores supervised dimensionality reduction models. With growing data complexity, effectively utilizing trading behaviors, institutional trends, technical factors, market sentiment, and international indicators is crucial. This research applies Partial Least Squares (PLS) and Partial Quantile Regression (PQR) to extract key information from high-dimensional data while addressing collinearity among indicators.
Empirical analysis using multi-period data confirms the models’ stability and effectiveness. Results show that PLS and PQR outperform traditional models in capturing stock price volatility and nonlinear return characteristics, significantly enhancing prediction accuracy. PLS compresses data into key factors, reducing overfitting risks, while PQR offers flexibility under various quantiles, adapting to diverse market conditions and risk preferences.
This study highlights the potential of supervised dimensionality reduction in improving stock return predictions and supporting investment decisions, especially when faced with numerous correlated indicators. By leveraging dimensionality reduction, this approach enhances predictive accuracy and decision-making reliability.
