The SPATIALREG Procedure

Spatial Durbin Autoregressive Moving Average Models

You can also accommodate exogenous interaction effects in the SARMA model. The term spatial Durbin autoregressive moving average (SDARMA) model is used to refer to such an extension of the SARMA model. Let y Subscript i denote the observation associated with the spatial unit bold s Subscript i for i equals 1 comma 2 comma ellipsis comma n. For these spatial units, let bold upper W 1 and bold upper W 2 be two spatial weights matrices. Further, let bold x Subscript i be a p times 1 vector that denotes values of p regressors recorded at unit bold s Subscript i. Similarly, let bold z Subscript i be a q times 1 vector that denotes values of q regressors measured at unit bold s Subscript i.

The SDARMA model can be described in vector form by using the following two-stage formulation,

where bold y equals left-parenthesis y 1 comma y 2 comma ellipsis comma y Subscript n Baseline right-parenthesis prime and bold-italic epsilon equals left-parenthesis epsilon 1 comma epsilon 2 comma ellipsis comma epsilon Subscript n Baseline right-parenthesis prime, with epsilon Subscript i Baseline tilde Overscript normal i normal i normal d Endscripts upper N left-parenthesis 0 comma sigma squared right-parenthesis prime. Moreover, bold upper X is an n times p matrix that has bold x prime Subscript i in each row, bold upper Z is an n times q matrix that has bold z prime Subscript i in each row. In addition, bold-italic beta is a p times 1 parameter vector.

By letting upper X overTilde equals left-bracket bold upper X bold upper W 1 bold upper Z right-bracket and bold-italic beta overTilde equals left-parenthesis bold-italic beta prime bold-italic theta Superscript prime Baseline right-parenthesis prime, the SDARMA model can be written as

The log-likelihood function for the SDARMA model is

where bold upper A equals bold upper I Subscript n Baseline minus rho bold upper W 1 and bold upper B equals bold upper I Subscript n Baseline minus lamda bold upper W 2.

For the SDARMA model, the gradients are

Last updated: August 11, 2020