Finite sample performance of the robust Wald test in simultaneous equation systems

The estimator of the coefficient covariance matrix proposed in White (1982) can be used to robustify the classical Wald test. Sampling experiments recently performed on linear regressions and simultaneous equation models, however, suggest that such an estimator tends to underestimate the covariance matrix if the model is correctly specified. In the classical framework of simultaneous equation systems, this paper aims at investigating the consequences of the use of robust covariance matrix estimator in the Wald test, when there is no misspecification.