Machine Learning and Statistical Model Hybrid Approach to Optimizing Financial Data Prediction

Abstract

Accurate price volatility prediction is a cornerstone of sound investment decisions and effective dynamic risk management in financial markets. This study addresses a significant research gap: the limited number of studies exploring the systematic integration of traditional statistical models and artificial intelligence techniques within emerging financial markets, despite their high levels of instability and volatility. The research aims to develop a hybrid predictive framework that combines the flexibility of linear models, specifically ARIMA, with the ability of machine learning algorithms to grasp the complex, nonlinear patterns inherent in financial time series. Furthermore, the study highlights an application gap: the underutilization of advanced volatility estimators. The Garman-Class estimator was adopted as a more efficient and accurate alternative to traditional estimators for measuring daily volatility, due to its reliance on four-part price information (open, close, high, and low). The proposed framework was applied to data from Savola Group, listed on the TASI. The results demonstrated the superiority of the proposed hybrid model in improving forecast accuracy and reducing predictive error measures, particularly the MAE, and RMSE, compared to traditional single-model models.

The scientific value of this research lies in its contribution to bridging the knowledge gap related to the integration of statistical models and artificial intelligence techniques in the emerging markets environment. Furthermore, it provides an advanced analytical tool that can enhance asset allocation efficiency and support decision-makers and portfolio managers in navigating the dynamics of highly volatile markets.