[Gretl-users] SFspec - function package for testing the Stochastic Frontier model

*To those gretl users that engage in stochastic frontier analysis (SFA).* (and also to those gretl users that intend to apply general regression analysis but using maximum likelihood and wanting to allow for a regression error with skewness). I have been notified that the function package "SFspec" has been approved and is now available in the gretl server. The package implements the specification test of Papadopoulos A. and Parmeter C. (2023), "A Specification Test for the Composed Error Term in the Stochastic Frontier Model". /Economics Letters/, 233, 111390. In Monte Carlo simulations, this test had best power among all the tests that have been proposed for the SF model. It has been already named "the PP specification test" and has been included in the "frontier" gretl package as an after-estimation test of the chosen error specification. But the SFspec package has a different purpose: It tests 12 different null hypotheses. The test depends on an OLS regression only, so the idea is to use it before maximum likelihood estimation, in order to get a sense of which distributional assumptions are a good match with the data, and which are not. Mostly, it tells you what /not/ to do, and then leaves you with the decision which specification to select among those that remain statistically admissible. This test is not available to any other statistical or econometric package. */HOW IT WORKS/* After starting a gretl session, uploading your data set, and installing the package, you only need to define a gretl List of regressors, and then in the GUI of the test, select the dependent variable and the regressor List. Hit "OK" and in the script output window you will get the results: the value of the test statistic and the associated p-value for the 12 distributional null hypotheses, that are the combinations of four zero-mean symmetric distributions for the noise component of the error term (Uniform, Normal, Logistic, Laplace) and three for the inefficiency component (Half Normal, Exponential, Generalized Exponential). The test works as-is for both production and cost SF models. Note: the test is applicable to inefficiency distributions that have constant skewness and excess kurtosis, which in practice means that they must have a single parameter. This is why it is not applicable to test for inefficiency that follows the Truncated Normal (that has two parameters and so varying skewness and excess kurtosis). Opportunity given, the Generalized Exponential gives an inefficiency component with its mode away from zero, see Papadopoulos, A. (2021). "Stochastic frontier models using the Generalized Exponential distribution./" Journal of Productivity Analysis/, /55/(1), 15-29. -- Alecos Papadopoulos PhD Affiliate Researcher Dpt of Economics, Athens University of Economics and Business Foundation for Economic and Industrial Research (IOBE) web: alecospapadopoulos.wordpress.com/ ORCID:0000-0003-2441-4550