"Time Series Forecasting of Ethereum Prices: An ARIMA Model Approach for Predicting Cryptocurrency Volatility"

Abstract

This study investigates the forecasting of Ethereum prices through time series analysis using the Augmented Dickey-Fuller (ADF) test, ARIMA modeling, and subsequent model evaluation. The first step involved testing for stationarity in Ethereum’s price series, where the ADF test confirmed that the first differences of the series were stationary, thus establishing an Integrated of Order 1 (I(1)) process. Building on this result, multiple ARIMA models were estimated, and ARIMA(1,1,0) was identified as the most suitable model based on the lowest Bayesian Information Criterion (BIC) and favorable regression diagnostics. The ARIMA(1,1,0) model was further validated through Maximum Likelihood estimation, revealing that past price changes significantly affect current price movements. Model performance was evaluated using forecasting metrics, including Mean Absolute Percentage Error (MAPE) and Theil’s Inequality Coefficient, yielding MAPE of 1.43% and a Theil coefficient of 0.0072, indicating robust forecasting accuracy. However, the model also exhibited some bias, suggesting the need for refinement. The results underscore the potential of ARIMA models for short-term Ethereum price forecasting, but also highlight the increasing uncertainty in long-term forecasts, necessitating caution in investment and trading strategies. Future research could expand the model to include exogenous variables and alternative forecasting methods to address the observed biases and enhance prediction reliability in volatile markets.