A Hybrid LSTM-DCC Model for Multivariate Cryptocurrency Volatility Prediction
Saminu Umar *
Department of Mathematics and Statistics, Umaru Ali Shinkafi Polytechnic, Sokoto, Nigeria.
Gafar M. Oyeyemi
Department of Statistics, University of Ilorin, Ilorin, Nigeria.
*Author to whom correspondence should be addressed.
Abstract
Accurate volatility forecasting remains a central challenge in the analysis of cryptocurrency markets, where extreme price fluctuations, nonlinear dependencies and evolving cross-asset correlations complicate traditional modeling approaches. This study proposes a hybrid framework that integrates the Dynamic Conditional Correlation (DCC) Multivariate Generalized Autoregressive Conditional Heteroskedasticity (MGARCH) model with Long Short-Term Memory (LSTM) networks to enhance forecasting accuracy. The LSTM–DCC model improves the representation of volatility clustering, structural breaks and interdependencies among digital assets by feeding the time-varying covariance matrix from the DCC-MGARCH model into the LSTM networks as an additional input at each time step. Using daily return data for Bitcoin (BTC), Ethereum (ETH) and Binance Coin (BNB) from January 2018 to March 2025, the hybrid model was developed and evaluated across multiple forecast horizons and compared with standalone LSTM and DCC-MGARCH models. Forecast performance was measured using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The LSTM–DCC model achieved the lowest average errors across all horizons, with MAE values of 0.00165 (train), 0.00243 (30-step) and 0.00371 (60-step) and RMSE values of 0.00417 (train), 0.00445 (30-step) and 0.00572 (60-step). These results outperform both standalone LSTM and DCC-GARCH models. This confirms the hybrid model’s superiority in capturing nonlinear temporal dynamics and cross-asset interactions. The findings support the adoption of integrated deep learning and econometric models for robust and reliable multivariate cryptocurrency volatility forecasting.
Keywords: Deep learning, MGARCH, LSTM-DCC, hybrid models, cryptocurrency, volatility forecasting