Time Series Analysis: Hybrid Econometric - Machine Learning Model for Improved Financial Forecasting

Abstract

Financial forecasting in stock markets is a complex problem due to inherent volatility and non-linearity. This research proposes a hybrid Auto-Regressive Integrated Moving Average (ARIMA) - Convolutional Neural Network (CNN) - Support Vector Machine (SVM) model to enhance the accuracy of time-series predictions. The hybrid model integrates ARIMA for capturing linear trends, CNN for extracting non-linear features, and SVM for the final classification of trend directions. The study is conducted on an 8-year stock market dataset (2015-2023) from Euronext, with 21 features and 1787 observations. The Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and F1-score are used to evaluate performance. Results indicate that the hybrid model achieves a prediction accuracy of 59%, outperforming standalone ARIMA (48%) and CNN (54%). Comparative analysis with ARIMA-LSTM and ARIMA-RNN further validates the robustness of the proposed approach. This study contributes a novel econometric-machine learning hybrid framework for financial forecasting with superior predictive power.