The Robustness of the RESET Test to Non-Normal Error Terms

In systems ranging from 1 to 10 equations, the size and power of various generalization of the Regression Specification Error Test (RESET) test for functional misspecification are investigated, using both the assymptotic and the bootsrap critical values. Furthermore, the robusteness of the RESET test to various numbers of non-normal error terms has been investigated. The properties of eight versions of the test are studied using Monte Carlo methods. Using the assyptotic critical values together with normally distributed error terms,we find theRao’smultivariate F-test to be best among all other alternative testmethods (i.e.Wald, Lagrange Multiplier and Likelihood Ratio). In the cases of heavy tailed error terms, short or long tailed errors, however, the properties of the bestRao test deteriorates especially in larg systems of equations.By using the bootstrap critical values, we find that the Rao test exhibits correct size but still slightlyunder reject the null hypothesis in cases when the error terms are short tailed. The powerof the test is low, however, in small samples and when the number of equations grows.