Comparison Of Vector Autoregressive and Multiresponse Fourier Series for Cryptocurrency Prediction Post-Bitcoin Halving 2024

This study examines cryptocurrency price modeling after the 2024 Bitcoin halving by comparing two multivariate forecasting methods: the Vector Autoregressive (VAR) model and the cosine-based Multiresponse Fourier Series Estimator. The research aligns with SDGs 8 on Decent Work and Economic Growth, as accurate forecasting in digital asset markets can support financial stability and informed investment decisions. The novelty of this study lies in applying a Fourier-based multiresponse model for post-halving cryptocurrency prediction, an approach that remains limited in existing literature. The dataset consists of daily prices of Bitcoin, Ethereum, and Litecoin from April 2024 to August 2025 (t=480), obtained from investing.com, with 90% for data training and 10% for data testing. Prior to modeling, the Bartlett test produced p-value 0.000α, indicating significant correlations among cryptocurrencies, thereby validating the use of simultaneous multivariate analysis. The results show that the Fourier Series Estimator with five oscillation parameters (k=5) provides superior predictive accuracy, achieving a MAPE of 3.768%, compared to the VAR model’s MAPE of 8.503%. These findings demonstrate that the Fourier estimator more effectively captures cyclical and nonlinear patterns in digital assets and offers valuable contributions to financial statistics, providing practical insights for investors and policymakers in the highly volatile cryptocurrency market.