Multivariate Linear And Bilinear Time Series Models And Their Applications To Revenue Data

ABSTRACT

Multivariate time series modeling has been carried out in this work. We considered three vector series; a response vector X 11 and two predictor vectors X21 and X31. The distribution of autocorrelation and partial autocorrelation functions of the non-stationary series suggested a pure autoregressive process for the three vector series. These give basis for expression of order of relationship between the response and predictor vectors. In this work, generalized form of multivariate bilinear time series models have been established. The Autoregressive Vector (ARV) models were initially estimated with the three vector series. From the multivariate linear models. we made an extension to bilinear vector series. Bilinear Autoregressive Vector (BARV) models and Pure Diagonal Bilinear Autoregressive Vector (PDBARV) models have been established for vectors, which follow a pure autoregressive process. The estimates from both multivariate linear and bilinear models indicate that the I BARY and PDBARV models have higher precision than ARV models, especially j when estimating economic and financial data