نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشیار دانشگاه آیت الله بروجردی
2 گروه اقتصاد، دانشکده علوم انسانی، دانشگاه آیت الله بروجردی (ره)، بروجرد، ایران.
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
The present study aims to model and forecast Bitcoin prices using the Support Vector Regression (SVR) approach and daily Bitcoin price data from July 13, 2010, to January 29, 2026. To this end, first, seven components were extracted from the fundamental and technical variables affecting Bitcoin prices using Principal Component Analysis (PCA) and employed as input features for the model. Subsequently, the Support Vector Regression (SVR) model, along with eight competing models, including ARIMA and GARCH models, machine learning algorithms such as Random Forest (RF) and XGBoost, and deep learning models including LSTM, GRU, RNN, MLP, and CNN, was utilized to model the Bitcoin market price. The results obtained from comparisons of error metrics, the correlation coefficient, and Theil’s index indicated that, in forecasting the Bitcoin price index, the SVR model outperformed the other models under consideration. Furthermore, the findings of the Diebold–Mariano, Wilcoxon, Hansen–Lunde–Nason, Giacomini–White, and variance-corrected Diebold–Mariano tests also confirmed the statistically significant superiority of the proposed model over the competing models, thereby reinforcing the validity of the results derived from the error metrics. These findings indicate that the SVR model, due to its high capability in modeling the nonlinear and noisy behavior of the Bitcoin market, can provide more accurate and more stable forecasts of Bitcoin prices. The SVR model's forecasting capabilities for predicting Bitcoin prices indicate the transformative capacity of machine learning in investment management and in reducing market volatility, and can provide a scientific basis for the development of risk management tools and the regulation of digital asset markets.
کلیدواژهها [English]