This paper uses a modiﬁed rank score test for non-nested linear regression models. The modiﬁed rank score test is robust with respect to models with nonnormal distributions and can be viewed as a robust version of the J test of Davidson and MacKinnon (Econometrica 49:781–793, 1981). Therefore, this test does not require a speciﬁcation of error density function and is easy to implement. Also, a modiﬁed rank score test for multiple non-nested models is provided. Monte Carlo simulation results show that the test has good ﬁnite sample performances. Financial applications for two competing theories, the capital asset pricing model and the arbitrage pricing theory, are considered herein. Empirical evidence from the modiﬁed rank score test shows that the former is a better model for asset pricing.
AStA Advances in Statistical Analysis , 15(1), 93-111