Testing for Separation in Agricultural Household Models and Unobservable Individual Effects: A Note

When market structure is complete, factor demands by households will be independent of their characteristics, and households will take their production decisions as if they were profit-maximizing firms. This observation constitutes the basis for one of the most popular empirical tests for complete markets, commonly known as the 'separation' hypothesis. In this paper, we show that all existing tests for separation using panel data are potentially biased towards rejecting the null-hypothesis of complete markets, because of the failure to adequately control for unobservable individual effects. Since the variable on which the test for separation is based cannot be identified in most panel datasets following the usual covariance transformations, and is likely to be correlated with the individual effect, neither the within nor the variance-components procedures are able to solve the problem. We show that the Hausman-Taylor (1981) estimator, in which the impact of covariates that are invariant along one dimension of a panel can be identified through the use of covariance transformations of other included variables that are orthogonal to the individual effects as instruments, provides a simple solution. We furnish an empirical illustration in which separation —and thus the null of complete markets— is strongly rejected using the standard approach, but is not rejected once correlated unobservable individual effects are controlled for using the Hausman-Taylor instrument set.