Application of the arbitrage pricing based on Gaussian temporal factor analysis (TFA) for the prediction of cryptocurrency prices

Abstract

The study demonstrates how the Gaussian TFA model can be applied for cryptocurrency price forecasting. The study considers four approaches of the Extended Normalized Radial Basis Function (ENRBF), namely, the N-Adaptive ENRBF, S-Adaptive ENRBF, ICA-ENRBF and APT-based TFA-ENRBF to make one-time-step predictions of the closing price of four different cryptocurrency prices (BTC, ETH, XRP and SOL) in the financial markets. We’ve evaluated and compared the performance of the proposed models with a conventional Root Mean Squared Error (RMSE) using daily series data from January 01, 2020 to December 31, 2024. The result indicates that the APT-based TFA-ENRBF consistently outperformed the other models by achieving the highest predictive performance for look-back and look-ahead periods in terms the least RMSE. This shows that the APT-based TFA-ENRBF approach shows consistently superior performance over three other conventional approaches. Moreover, we’ve found that the ICA-ENRBF approach comes second, whilst the N-ENRBF approach appears to be the worst compared to others. Because the paper ignores the influence of other factors that can likely influence price predictions, we suggest that future research may involve a more in-depth exploration of hybrid methods. which can help examine how other factors predict the cryptocurrency prices.