A GQL-based inference in non-stationary BINMA(1) time series

This paper introduces a non-stationary bivariate integer-valued moving average of first-order (BINMA(1)) model with corresponding negative binomial innovations under different levels of over-dispersion that are pairwise unrelated. In the proposed BINMA(1), the interrelation between the series is induced by the relation of the current observation with the previous-lagged innovation of the other series, while the non-stationarity is captured through the time-variant covariate specification. Under such condition, the likelihood construction is cumbersome to formulate. Thus, a generalized quasi-likelihood equation based on an exact auto-covariance specification via multivariate thinning structures is proposed to estimate the regression, over-dispersion and dependence effects, and its performance and efficiency measures are compared with other common established techniques: generalized least squares and generalized method of moment based on simulated data from the proposed model under different scenarios of over-dispersion and serial coefficients. The model is further applied to analyze the intraday transactions of two major banks in Mauritius.