A dynamic model for daily equity covariances based on multiple realised measures

OMI Seminar Series

We propose a new observation-driven model for high dimensional time-varying covariances of financial assets.  The key novelty is the dynamic model formulation for daily covariances based on model-implied weighted combination of multiple realised measures of different properties. We demonstrate that in our model specification each individual innovation driving the daily update of covariance entries exploits full local likelihood information. Consequently, even parsimonious or scalar model formulation adopts for cross-equity effects implying that a number of model static parameters can be kept low in high dimensions without much loss of flexibility yet resulting in ease of estimation. We illustrate the merits of the new model in the simulation study as well as in the application to a portfolio of NYSE equities.

Joint work with Peter R. Hansen and Siem Jan Koopman


Pawel Janus (UBS, Zurich)

Tuesday, March 5, 2013 - 12:30
to 13:30