Time Varying Multivariate Autoregressive Modeling

Matlab code for time-varying multivariate autoregressive (TV-MVAR) modeling

This toolbox contains Matlab codes for time-varying multivariate autoregressive (TV-MVAR) modeling. MVAR models are usually applied to investigate couplings between various time-series in frequency domain. Herein, changes in the model parameters are tracked using the conventional Kalman Filer (KF) and a proposed modified KF. Model order selection and hyperparameter optimization is realized using Genetic Algorithms, significantly improving accuracy and run-time. Residual heteroskedasticity is tackled by using Generalized Autoregressive Conditional Heteroskedasticity (GARCH) models leading to more accurate representations of the strength and directionality of the underlying couplings.

The code can be also found at: https://github.com/BioSigSystLab/TV-MVAR