Hybrid Error Correction Model and Long Short-Term Memory Approach for Predicting Ethereum Opening Prices

Abstract

The rapid growth of the cryptocurrency market has raised the need for an effective model to predict opening prices and assist investors and policymakers in decision-making. Traditional econometric models often struggle with the high volatility and nonlinear patterns inherent in digital asset prices. Long short-term memory networks are effective at recognizing complex patterns, yet they lack interpretability. This study bridges this gap by integrating the error correction model with long short-term memory to improve prediction of Ethereum’s opening price. Using daily price data from January 2018 to June 2024, the model captures both long-term equilibrium relationships and short-term fluctuations, resulting in more accurate forecasts. The findings confirm a significant long-run equilibrium relationship between Bitcoin and Ethereum prices. The integrated model outperforms standalone models, by achieving a mean absolute error of 46.76, a mean squared error of 5,544.05, and an R-squared of 88%. This study contributes to both econometric and deep learning literature, highlighting Bitcoin’s influence on Ethereum, and offering a practical framework for financial forecasting. Future research could expand this work by incorporating additional macroeconomic variables, exploring alternative deep learning architectures, and testing the robustness of the model across time and market conditions.