Title: Singular conditional autoregressive Wishart model
Authors: Gustav Alfelt - Stockholm University (Sweden) [presenting]
Taras Bodnar - Stockholm University (Sweden)
Farrukh Javed - Orebro University (Sweden)
Joanna Tyrcha - Stockholm University (Sweden)
Abstract: A Singular Conditional Autoregressive Wishart model that aims to capture the dynamics of singular realized covariance matrices of asset returns is suggested. Such singularity arises in high-dimensional cases where the dimension of the return process exceeds the number of intraday returns sampled each day. The model assumes that the non-singular scale matrix of the underlying Singular Wishart process follows an autoregressive moving average structure with a BEKK specification, and can be estimated by the Maximum Likelihood method. In order to facilitate feasible estimation in high-dimensional cases, the model is fitted to a transformation of the data series together with the application of covariance targeting. Finally the model is applied to high-frequency data from AMEX, NASDAQ and NYSE, and is evaluated by out-of-sample forecast accuracy.