Simulating mixed regressive spatially autoregressive estimators
Document Type
Article
Publication Date
12-1-1998
Abstract
This paper discusses fast maximum likelihood computational considerations for the mixed regressive spatially autoregressive model which involves a regression of the dependent variable on the independent variables plus the spatial lags of all the variables. As discussed in the paper, estimation precision of the spatial autoregressive parameter does not depend upon the inherent variability of the true errors. This fact, along with the sparse spatial structure, and the quick maximum likelihood computational algorithm proposed, greatly facilitates simulations of this type of model. The paper demonstrates the utility of the techniques by simulating 4,500 spatially correlated vectors of 41,372 observations each and estimating these via maximum likelihood. Each of the simulated vectors requires the solution of 41,372 equations and the Jacobian required by maximum likelihood involves determinants of 41,372 by 41,372 matrices. With the proposed techniques, these computations took under 2.5 hours on a Pentium Pro 200 megahertz personal computer.
Publication Source (Journal or Book title)
Computational Statistics
First Page
397
Last Page
418
Recommended Citation
Pace, R., & Barry, R. (1998). Simulating mixed regressive spatially autoregressive estimators. Computational Statistics, 13 (3), 397-418. Retrieved from https://repository.lsu.edu/finance_pubs/113