A state space model approach to integrated covariance matrix estimation with high frequency data

We consider a state space model approach for high frequency financial data analysis. An expectation-maximization (EM) algorithm is developed for estimating the integrated covariance matrix of the assets. The state space model with the EM algorithm can handle noisy financial data with correlated microstructure noises. Difficulty due to asynchronous and irregularly spaced trading data of multiple assets can be naturally overcome by considering the problem in a scenario with missing data. Since the state space model approach requires no data synchronization, no record in the financial data is deleted so that it efficiently incorporates information from all observations. Empirical data analysis supports the general specification of the state space model, and simulations confirm the efficiency gain and the benefit of the state space model approach.