Advanced CEEMD hybrid model for VIX forecasting: optimized decision trees and ARIMA integration

This study focuses on the time series analysis and forecasting of the Volatility Index (VIX) data in the United States. We employed the Complete Empirical Mode Decomposition (CEEMD) method to decompose the original VIX data into seven Intrinsic Mode Functions (IMFs). Subsequently, each IMF determined its stationarity using the Augmented Dickey-Fuller (ADF) test. For stationary IMFs, an AutoRegressive Integrated Moving Average (ARIMA) model was built, while non-stationary IMFs were fitted using machine learning regression algorithms. The sum of the predicted values of all IMFs was considered as the forecast VIX value. About the parameter selection in machine learning algorithms, we employed the Optuna library to optimize the model's parameters. We selected the parameters that resulted in the lowest Mean Squared Error (MSE) on a validation set to be the optimal model parameters. Applying this methodology, we found that our hybrid model has greater predictive capabilities compared to simple machine learning models when the market fluctuates greatly. Specifically, the AdaBoost model outperformed a basic decision tree model and a random forest model in terms of MSE and MAE (mean absolute error) accuracy.