Memory Properties and Aggregation of Spatial Autoregressive Models

T.T. Azomahou*

*Corresponding author for this work

Research output: Contribution to journalArticleAcademicpeer-review


A random field displays long (resp. Short) memory when its covariance function is absolutely non-summable (resp. Summable), or alternatively when its spectral density (spectrum) is unbounded (resp. Bounded) at some frequencies. Drawing on the spectrum approach, this paper characterizes both short and long memory features in the spatial autoregressive model. The data generating process is presented as a sequence of spatial autoregressive micro-relationships. The study elaborates the exact conditions under which short and long memories emerge for micro-relationships and for the aggregated field as well. To study the spectrum of the aggregated field, we develop a new general concept referred to as the ‘root order of a function’. This concept might be usefully applied in studying the convergence of some special integrals. We illustrate our findings with simulation experiments and an empirical application based on gross domestic product data for 100 countries spanning over 1960–2004.
Original languageEnglish
Pages (from-to)2581-2597
JournalJournal of Statistical Planning and Inference
Publication statusPublished - 1 Jan 2009


Dive into the research topics of 'Memory Properties and Aggregation of Spatial Autoregressive Models'. Together they form a unique fingerprint.

Cite this