Enhancing Financial Market Predictions in Taiwan: A Hybrid Approach of Traditional Analysis and Machine Learning Techniques

This research delves into the comparative effectiveness of traditional panel data analysis versus machine learning (ML) algorithms in forecasting stock returns, specifically within the context of the Taiwanese stock market. It aims to assess these methodologies against the backdrop of the semi-strong form of the Efficient Market Hypothesis (EMH) [1]. Employing a dataset spanning from 2015 to 2023, which encapsulates financial metrics from 54 prominent companies listed on the FTSE TWSE Taiwan 50 Index and TSEC Taiwan Mid-Cap 100 Index, the study juxtaposes traditional econometric methods with advanced ML techniques such as Support Vector Machines (SVM), Random Forests (RF), and Long Short-Term Memory (LSTM) networks [2], [3], [4]. The analysis incorporates a dynamic panel data regression model to scrutinize various financial ratios, while ML algorithms are evaluated for their precision in handling complex data structures and predicting market trends. The outcome of the study reveals that traditional panel data analysis not only holds its ground but also occasionally outperforms ML algorithms in ensuring stability and consistency in return predictions across both equally weighted and value-weighted portfolios. These findings highlight the enduring relevance of conventional financial analysis methods and suggest a symbiotic integration of traditional and modern approaches to enhance the robustness and accuracy of financial market predictions. The study advances the discourse on the utility of integrating computational technologies within traditional financial frameworks, providing insights that could influence future investment strategies and market analysis methodologies.