AI-Driven Quantitative Stock Selection with Return–Risk Prediction and Agentic XAI

Quantitative investment increasingly relies on artificial intelligence to analyze complex financial data and support investment decisions. This study develops an AI-driven stock selection framework for the Taiwan equity market by integrating industry-specific modeling, multi-period return prediction, downside risk estimation, and explainable artificial intelligence.

The proposed framework groups stocks according to industry characteristics and compares several deep learning models, including LSTM, GRU, Transformer, Temporal Convolutional Network (TCN), and Temporal Fusion Transformer (TFT). The models are used to predict future stock returns and Value at Risk (VaR), allowing the stock selection strategy to consider both return potential and downside risk rather than relying on expected returns alone.

To improve the transparency of investment decisions, the study further introduces an Agentic Explainable AI (Agentic XAI) system that combines SHAP, Retrieval-Augmented Generation (RAG), and a ReAct Agent. The system automatically analyzes model predictions, identifies potential contradictions between quantitative signals, and generates readable explanations for individual stocks. By integrating predictive modeling, risk-aware decision-making, and AI-based explanations, this research provides a comprehensive framework for intelligent and interpretable quantitative investment.