The family of Inverse Gaussian distributions has applications in finance, lifetime testing, etc. In this manuscript we consider the problem of testing the IG assumption due to the relevance of this model in statistical applications. For this problem there exist several modified versions of the classical empirical distribution function (EDF) tests and some tests based on the empirical Laplace transform of the observations. Koutrouvelis and Karagrigoriou (2012) provide a complete review on existing tests for this model.
Many tests for the IG distribution rely on the use of tables containing critical values for different values of the shape parameter ϕ=λ/μ since the null distributions of their test statistics depend on the unknown value of ϕ. Other tests rely on the use of parametric bootstrap for approximating null distributions, from which approximated p-values and/or critical values are obtained. Here we propose a variance ratio test and two additional tests of fit for the IG distribution based on property (1), which are asymptotically independent of parameter ϕ under the null hypothesis. Vexler et al. (2011) proposed a test based on an empirical likelihood ratio, which can be considered of the same type of the tests proposed here.
This manuscript is organized as follows. In Section 2 we present the proposed tests and the asymptotic null distribution of the variance ratio test is obtained. Section 3 contains the results of a simulation study conducted to compare the power functions of these tests under three classical alternative distributions for this model. Finally, some conclusions are included.