Do spatial structures yield better volatility forecasts? (in Russian)

We evaluate, using forecasting experiments with real stock return data, forecasting ability of spatially structured BEKK specifications relative to standard BEKK. We confirm that the class of spatial BEKK has a potential of improving a quality of multivariate volatility forecasts. However, there is a sharp disagreement among forecast performance criteria on which types of further restrictions on coefficient matrices are most promising, on which degree of homogeneity of matrix coefficients is most beneficial, and on which grouping criteria and their number deliver highest improvements in volatility forecasts. The numerosity and composition of the portfolio also have a big influence on how well volatility is forecast by spatially structured BEKK compared to its standard configuration.