Comparative Forecasting of Major Cryptocurrencies: An Empirical Study Using Four Timeseries Forecasting Models

This study analyzes the in-sample forecasting power of four univariate time-series models—Naive, Autoregressive Moving Average (ARMA), Exponential Smoothing, and Autoregressive Fractionally Integrated Moving Average (ARFIMA)—for predicting 5-day returns of Bitcoin, Ether, Tether tokens, and Binance Coin (BNB). Each model's suitability varies due to the unique consensus systems and market dynamics of each cryptocurrency, suggesting no single model is universally optimal. While the Naive model proved unsuitable for most cases, ARMA demonstrated better accuracy for several cryptocurrencies, highlighting the importance of tailored forecasting approaches for diverse cryptocurrencies.