Abstract: Forecasting financial time series is a fundamental challenge in finance and econometrics, largely due to the complexity of volatility dynamics and interdependencies among assets. This study evaluates and compares the forecasting performance of MGARCH models, BEKK GARCH and DCC GARCH, with two deep learning networks, Single LSTM and BiLSTM, across short, medium and long-term forecast horizons. Two datasets, comprising simulated data and bank stock data were used. Forecast accuracy was assessed using Root Mean Squared Error (RMSE) on both simulated data and real-world stock returns. The findings from simulated data reveals that deep learning models, particularly BiLSTM, consistently outperform traditional GARCH models, with performance gains increasing over longer horizons. Similar trends are observed in the real data, where LSTM networks maintain lower RMSE values, indicating greater robustness in capturing complex time series patterns.

Keywords: Multivariate Time Series, Forecasting, MGARCH, Deep Learning, LSTM, Rooted Mean Square Error (RMSE).

Title: Evaluating the Forecast Accuracy of MGARCH Models and LSTM Networks for Multivariate Financial Time Series

Author: Saminu Umar, Gafar M. Oyeyemi

International Journal of Novel Research in Physics Chemistry & Mathematics

ISSN 2394-9651

Vol. 12, Issue 2, May 2025 - August 2025

Page No: 1-8

Novelty Journals

Website: www.noveltyjournals.com

Published Date: 17-May-2025