DYNAMIC MODELING OF HIGH-DIMENSIONAL CORRELATION MATRICES IN FINANCE
Abstract
A class of dynamic factor and dynamic panel models is proposed for daily high dimensional correlation matrices of asset returns. These flexible semiparametric predictors process ultra high frequency information and allow to exploit both realized correlation matrices and exogenous factors for forecasting purposes. The Fisher-z transformation offers the transmission from (factor and panel) time series models operating on unrestricted random variables to bounded correlation forecasts. Our methodology is contrasted with prominent alternative correlation models. Based on economic performance criteria dynamic factor models turn out to carry the highest predictive content.
Remember to check out the Most Cited Articles! |
---|
Be inspired by these new titles
|