Multivariate functional-coefficient regression models for nonlinear vector time series data

Vector time series data are widely met in practice. In this paper we propose a multivariate functional-coefficient regression model with heteroscedasticity for modelling such data. A local linear smoother is employed to estimate the unknown coefficient matrices. Asymptotic normality of the proposed estimators is established, and bandwidth selection is considered. To deal with the co-integration commonly observed in financial markets, we propose an error-corrected multivariate functional-coefficient model. Simulations show that our proposed estimation procedures capture nonlinear structures of coefficients well. Analysis of United States interest rates illustrates the proposed methods.