A Simple Boundary Model of Dynamic Covariance

One of the major goals of modern finance is to accurately model, test and forecast the conditional covariance of very large multivariate data-sets. Traditional methodologies have approached this problem by assuming a matrix-autoregressive framework, however for very large systems this requires the estimation of a large number of parameters and can result in a very flat objective function, when estimating via maximum likelihood. Our model strips down the modelling of the time evolution of the conditional covariance matrix into a simple two stage boundary framework.