Dynamic Mixture Vector Autoregressions With Score‐Driven Weights

We propose a novel dynamic mixture vector autoregressive (VAR) model where the time‐varying mixture weights are driven by the predictive likelihood score. Intuitively, the weight of a component VAR model is increased in the subsequent period if the current observation is more likely to be drawn from this state. The model is not limited to a specific distributional assumption and allows for straightforward likelihood‐based estimation and inference. In a Monte Carlo study, we document the model's ability to filter and predict mixture dynamics across different data‐generating processes. Moreover, we illustrate the model's empirical performance with the help of two applications.