Investigating and comparing the efficiency of econometric models (ARIMA and GARCH) and artificial intelligence (SVM and ANFIS) in predicting the price of cryptocurrencies (Bitcoin case study)

اقتصاد پولی مالی

Volume 31, Issue 2 - Serial Number 28, September 2025

Document Type: Original Article

DOI: 10.22067/mfe.2024.86389.1382

Authors

sadegh bafandeh imandoust^{* 1}; mehdi behname²; Mehdi Roshan³

¹Associate Professor, Department of Economics, Payame Noor University, Tehran, Iran.

²Associate Professor, Department of Economics, Ferdowsi University of Mashhad, Mashhad, Iran.

³Master of economics, Payame Noor University,Tehran, Iran.

Abstract

In this study, the comparative prediction of Bitcoin price was discussed using econometric models and artificial intelligence. For this purpose, daily Bitcoin price data for the period January 1, 2016 to January 1, 2023 was collected from the Bitstamp exchange website. In addition, in this study, in all horizons, 70% of the data is used to train artificial intelligence models and specify econometric models, and the remaining 30% is used to test the output of artificial intelligence models and predictions outside of Samples of econometric models were assigned. Also, in order to evaluate the efficiency of the models, R^2, MAD and RMSE criteria were used. Finally, in order to check the statistical difference in the effectiveness of the investigated models in predicting the daily price of Bitcoin, the restricted F test was used. The results showed that the average daily price of Bitcoin fluctuated from $449.71 on January 1, 2016 to $16,555.75 on January 1, 2023. Also, its minimum and maximum value was $370.21 (February 3, 2016) and $67,482.75 (November 9, 2021), respectively. In addition, in all models, by increasing the prediction horizon from 1 to 4 days ahead, the effectiveness of predicting the daily price of Bitcoin decreases. Finally, the results of comparing the effectiveness of the investigated models confirm that SVM, ANFIS, GARCH and ARIMA models have the most efficiency, respectively.

Keywords

Predicting methods; ARIMA model; autoregressive conditional heterogeneity variance model; SVM model; ANFIS model

References