Declan P. Mallamo1, Helene P. Hoang2, Michael H. Azarian1, and Michael Pecht1
1Center for Advanced Life Cycle Engineering (CALCE), University of Maryland, College Park, MD
2Costello College of Business, George Mason University, Fairfax, VA
For more information about this article and related research, please contact Dr. Michael H. Azarian or Prof. Michael G. Pecht
Abstract:
Predicting stock movements remains a significant challenge due to the volatility, complexity, and nonlinear dynamics of financial markets. Accurate forecasting is essential for optimizing trading strategies and achieving financial gains. However, existing models often struggle to integrate multivariate functional patterns, effectively balance the bias-variance tradeoff, and maintain computational efficiency for real-time trading. This study proposes a novel feature extraction framework that combines functional Principal Component Analysis (fPCA) and functional Linear Discriminant Analysis (fLDA). The approach captures the intrinsic variability and continuous nature of financial time-series data, transforming it into structured and interpretable representations for decision-making. A Mahalanobis distance-based decision function is applied within the discriminant space to classify buy, hold, and sell signals, leveraging clusters formed by the functional discriminant directions. The framework's performance is validated through simulated trading strategies on historical stock data, with evaluations based on portfolio growth and predictive accuracy. Results demonstrate the method's effectiveness in capturing market dynamics and providing actionable insights for financial analytics.
This article is available online here and to CALCE Consortium Members for personal review.
.