Simulation Study to Evaluate Full Information Maximum Likelihood as Parameter Estimation Methods for Spatial Vector Autoregressive Model with Calendar Variation

Abstract

Spatial Vector Autoregressive models with calendar variation can be used to analyze the interrelationships between variables, the relationship of variables with their past and the relationship of variables in a location with those variables in other locations. It also can accommodate the effects of calendar variations. The parameter estimation of this model can be done using Full Information Maximum Likelihood (FIML). The purpose of this study is to evaluate the performance of FIML and it is compared to Ordinary Least Square (OLS) through simulation. The simulation is done using generated data which is designed following Spatial Vector Autoregressive model with calendar variation. There are three aspects studied in this simulation, namely how the effect of error correlation between equations, the variance of error and length of period on the performance of FIML Method and OLS Method. The result of the simulation is the variance of FIML parameter estimator is smaller than OLS, especially when the error correlation between equation are high. While the variance of errors and length of periods have no effect on performance of the estimator. The simulations also show that the mean of parameter estimators both FIML and OLS are very close to the parameters specified.