February 1, 2004
This paper proposes pure significance tests for the absence of nonlinearity in cointegrating relationships. No assumption of the functional form of the nonlinearity is made. It is envisaged that the application of such tests could form the first step towards specifying a nonlinear cointegrating relationship for empirical modelling. The asymptotic and small sample properties of our tests are investigated, where special attention is paid to the role of nuisance parameters and a potential resolution using the bootstrap.
J.E.L classification codes: C32, C45
Keywords:Cointegration, Nonlinearity, Neural networks, Bootstrap