Comparison of two approaches for financial time series forecasting

Abstract

The problem of forecasting price movements in financial markets using historical data is a critical challenge in modern quantitative finance. This study focuses on comparing the effectiveness of machine learning (ML) methods [1], specifically XGBoost [2], with stochastic approaches based on Markov chains (MC) [3] and hidden Markov models (HMMs) [4] for predicting the direction of stock price changes in the S&P 500 index. Theoretical and empirical analyses are conducted, including data preprocessing, model implementation, and accuracy evaluation using classification metrics. The results provide insights into the strengths and limitations of each method, along with recommendations for future research