ABSTRACT
Depending on the ordering, cross-sectional dependence changes. Except for some obvious cases, such as the location of a wireless communication center, however, the actual cross-sectional dependence may not follow the spatial autocorrelation model. In social science and economics, hence, the spatial autocorrelation has been used as a filtering rule. That is, after eliminating the spatial dependence, if there exists cross-sectional dependence, then this result becomes supportive evidence for the social interaction. When the time series information is not available, the notion of cross-sectional dependence is purely hypothetical. Not like the time series dependence, the cross-sectional dependence is hard to model econometrically because the ordering of the cross-sectional units is usually unknown. Historically, the gravity model is the first economic model to take the cross-sectional dependence seriously. Contrasted with the Heckscher-Ohlin model, the gravity model can be statistically interpreted as a factor model. Hence the statistical factors and factor loadings are not equivalent to the true factors and loadings.
