Volatility Forecasting in Cryptocurrencies and Traditional Financial Assets: Comparative Analysis of ARIMA, GARCH (1,1), and XGBoost Models

<p dir="ltr">This paper presents a comparative analysis of volatility forecasting models applied to both cryptocurrency markets, specifically Bitcoin and Ethereum, and traditional financial assets including the S&P 500 and gold. The study evaluates the forecasting accuracy of two econometric models, ARIMA and GARCH(1,1), alongside a machine learning model, XGBoost. Forecast performance is assessed using Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). Results show that cryptocurrency markets display heavier-tailed return distributions, greater unconditional volatility, and more frequent price jumps than traditional assets, which presents challenges for models relying on stationarity assumptions. XGBoost outperforms traditional methods in capturing nonlinear patterns and regime shifts in crypto markets, while ARIMA-GARCH models remain competitive in more stable environments such as equities and commodities. The inclusion of exogenous factors, such as macroeconomic indicators and regulatory developments, improves forecasting accuracy, especially for digital assets. Furthermore, hybrid models that integrate outputs from econometric frameworks into machine learning architectures yield additional performance gains. These findings emphasize the importance of selecting models that align with asset-specific characteristics and market structures. The study contributes to financial forecasting literature by demonstrating the complementary strengths of statistical and machine learning methods across heterogeneous asset classes.</p>