Time-Varying Assets Clustering via Identity-Link Latent-Space Infinite Mixture: An Application on DAX Components

Finance literature suggests that cross-correlations among assets increase during periods of financial distress, and that cross-correlation's very own clustering structure varies over time. This work proposes an Identity-Link Latent-Space Infinite-Mixture model to analyze the clustering structure of cross-correlation over time. The model allows for the representation of stocks on a d-dimensional Euclidean space and the clustering of assets into groups. Model estimation is carried out within a Bayesian framework, which allows including prior extra-sample information in the inference and accounting for parameter uncertainty. We apply the model to time-varying correlations among the DAX components. We find evidence of clustering effects and positive dependence between the number of clusters and both annualized volatility and average cross-correlation.