Estimation with inequality constraints on the parameters: dealing with truncation of the sampling distribution.

Theoretical constraints on economic-model parameters often are in the form of inequality restrictions. For example, many theoretical results are in the form of monotonicity or nonnegativity restrictions. Inequality constraints can truncate sampling distributions of parameter estimators, so that asymptotic normality no longer is possible. Sampling theoretic asymptotic inference is thereby greatly complicated or compromised. We use numerical methods to investigate the resulting sampling properties of inequality constrained estimators produced by popular methods of imposing inequality constraints. In particular, we investigate the possible bias in the asymptotic standard errors of estimators of inequality constrained estimators, when the constraint is imposed by the popular method of squaring. That approach is known to violate a regularity condition in the available asymptotic proofs regarding the unconstrained estimator, since the sign of the unconstrained estimator, prior to squaring, is nonidentified.