Spatial extension of generalized autoregressive conditional heteroskedasticity models

This paper proposes an extension of generalized autoregressive conditional heteroskedasticity (GARCH) models for a time series to those for spatial data, which are called here spatial GARCH (S-GARCH) models. S-GARCH models are re-expressed as spatial autoregressive moving-average (SARMA) models and a two-step procedure based on quasi-likelihood functions is proposed to estimate the parameters. The consistency and asymptotic normality are proven for the two-step estimators. S-GARCH models are applied to simulated and land-price data in areas of Tokyo to demonstrate the empirical properties.