Forecasting cryptocurrency prices using artificial neural networks

In an era of high volatility in cryptocurrency markets, this study provides evaluation of Artificial Neural Network (ANN) models for forecasting the prices of three major assets: Bitcoin (BTC), Ethereum (ETH), and Binance Coin (BNB). The research systematically investigates the impact of model architecture on predictive accuracy as measured by Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). The findings reveal that a moderately complex architecture with two hidden layers consistently yields the best forecasting performance across all three cryptocurrencies. However, the optimal node configuration varies by asset, indicating that model must be tailored to the unique characteristics of each coin. Furthermore, this study translates these forecasts into practical trading strategies based on signal thresholds and evaluates their performance against a traditional buy-and-hold strategy. Although the ETH-based strategy had the highest daily return (0.19%), the BNB-based strategy was clearly better when adjusting for risk, with a higher Sharpe Ratio of 2.82, showing strong returns for the amount of risk taken. Despite these promising results, a conclusive statistical validation using one-way ANOVA and a Tukey’s HSD post-hoc test revealed no statistically significant difference in performance between the model-driven strategies and the passive buy-and-hold strategies (p = 0.660).