Universitas Airlangga Official Website
Digital currencies or cryptocurrencies, especially Bitcoin, have attracted significant attention in the financial world due to their volatile nature and growing role in the global digital economy. According to the TripleA report, global crypto holders reached approximately 562 million people, or 6.8% of the world’s population, representing a 34% increase from 2023. The 2024 Bitcoin halving represents a critical structural turning point in the cryptocurrency market. The reduction of block rewards from 6.25 to 3.125 Bitcoins constrained new supply, while demand continued to rise amid growing institutional participation and tighter global macroeconomic conditions. This imbalance contributed to price appreciation accompanied by persistent volatility. Despite the importance of this period, empirical studies focusing specifically on multivariate cryptocurrency forecasting in the immediate post-2024 halving phase remain limited. This study aims to bridge this gap by conducting a systematic comparison between two multivariate forecasting approaches: the Vector Autoregressive (VAR) model and a multivariate Fourier series estimator.
This study uses secondary data obtained from the website investing.com, which consists of daily closing price data covering the period from April 20, 2024, to August 12, 2025. The Variance Decomposition results indicate a clear dominance of Bitcoin in explaining forecast error variance across the system. At the beginning of the forecast horizon, Bitcoin price movements are entirely driven by its own innovations and remain overwhelmingly self-determined throughout the period, with a contribution of approximately 98.84%. In contrast, Ethereum and Litecoin display substantial exposure to cross-asset shocks. VAR models are powerful for multivariate time series analysis and forecasting, especially in economic contexts, but they can be limited by their linear assumptions and high dimensionality. On the other hand, nonparametric Fourier series offer flexibility and robustness for modeling periodic data, although they can be more complex to implement and interpret. Therefore, both approaches are considered in this study to compare their performance in modeling the observed time series data. Fourier Series Estimator method has a MAPE value of 3.768%, which is smaller than that of the VAR method, indicates that the Fourier Series Estimator model provides more accurate predictions.
This study compares the effectiveness of two multivariate time series analysis approaches, namely VAR and Fourier Series Estimator based on cosine functions. The analysis results show that the Fourier Series Estimator method is superior to VAR in predicting time series data with complex fluctuation patterns. Validation using Bitcoin, Ethereum, and Litecoin price data after the 2024 Bitcoin halving event shows that Fourier produces a MAPE value of 3.768%, which is lower than VAR. This finding confirms Fourier’s ability to stably model nonlinear and periodic data dynamics. Methodologically, this research positions the Fourier Series Estimator as an efficient alternative for multivariate time series modeling, particularly in contexts with high volatility such as financial analysis. These findings contribute to financial inclusion, digital economic participation, and inclusive growth, aligning with SDG 8 on decent work and economic growth.
Author: Dita Amelia
Details of the research can be viewed here: https://ejournal.uin-malang.ac.id/index.php/Math/article/view/37720
Copyright 2025 Universitas Airlangga. All Rights Reserved.
CENTER FOR COMMUNICATIONS AND PUBLIC INFORMATION (PKIP)
Management Office Building, MERR (C) Campus Mulyorejo – Surabaya
Tel. (031) 5914042, 5914043, 5915551 Fax. (031) 5915551
WhatsApp. +62 821-3004-0061
Email: adm@pkip.unair.ac.id
