DataScience@BI invites Yiru Wang
Assistant Professor Yiru Wang, Department of Economics at the University of Pittsburgh to give a research talk titled "Estimating Large Covariance Matrices with Unobserved Clustering".
- Starts:12:00, 22 October 2024
- Ends:13:00, 22 October 2024
- Location:BI - campus Oslo, B3 inner area - next to meeting room B3i-108 or Zoom
- Contact:Siri Johnsen (siri.johnsen@bi.no)
DataScience@BI invites Assistant Professor Yiru Wang, Department of Economics at the University of Pittsburgh to disuss the paper "Estimating Large Covariance Matrices with Unobserved Clustering".
Research fields: Econometrics, Time Series, Forecasting, Empirical Macroeconomics, Empirical Finance.
Abstract:
In this paper, we address the complexities of unobserved heterogeneity and the curse of dimensionality in the estimation of large covariance matrices. We introduce latent group structures in multivariate generalized autoregressive conditional heteroskedasticity (GARCH) models. Within this framework, groups are characterized based on the diagonal parameters of the MGARCH models and the unconditional covariance matrix which are the same within a group but differ among groups, with the group memberships unknown. Our methodology leverages a composite likelihood approach in conjunction with the Sequential Binary Segmentation Algorithm (SBSA) to effectively identify group memberships and estimate group-specific GARCH parameters. We demonstrate the practicality of our approach through application to optimal minimum-variance portfolio analysis, documenting significant time-varying clustering patterns within large covariance matrices.