Title: Bayesian nonparametric covariance estimation with noisy and nonsynchronous asset prices
Authors: Jia Liu - Saint Mary's University (Canada) [presenting]
Abstract: A Bayesian nonparametric approach is proposed to estimate the ex-post covariance matrix of asset returns from high-frequency data in the presence of market micro-structure noise and non-synchronous trading. Several contributions are made. First, pooling is used to group returns with similar covariance matrices to improve estimation accuracy. Second, a new synchronization method of observations based on data augmentation is introduced. Third, the estimator is guaranteed to be positive definite. Finally, the new approach delivers exact finite sample inference without relying on asymptotic assumptions. All of those benefits lead to a more accurate estimator, which is confirmed by Monte Carlo simulation results. In real data applications, the proposed covariance estimator results in better portfolio choice outcomes.