Autore
Monfardini, Chiara

Titolo
Simulation-based encompassing for non-nested models: a Monte Carlo study of alternative simulated cox test statistics
Periodico
European University Institute of Badia Fiesolana (Fi). Department of Economics - Working papers
Anno: 1995 - Fascicolo: 37 - Pagina iniziale: 1 - Pagina finale: 49

We consider the possibility of using simulated methods in order to compute test statistics for non-nested models, in the general framework represented by the encompassing principle. Simulation techniques are useful whenever the evaluation of a test statistic involves a difficult integration problem, and represent a (simpler) alternative to numerical procedures. Focusing on the Cox test statistic, we extend a procedure proposed in the literature and calculate both its numerator and its denominator through simulation, getting a refinement of previous proposals which considered only simulation of the numerator. Some Monte Carlo experiments are then conducted taking as an example the test of linear versus loglinear models. Our results suggest that simulating both the quantities in the numerator and the denominator leads to a finite sample distribution of the test statistic closer to the asymptotic one.



Testo completo: http://hdl.handle.net/1814/562

Esportazione dati in Refworks (solo per utenti abilitati)