No. 474: Testing for Neglected Nonlinearity in Long Memory Models
George Kapetanios ,
Queen Mary, University of London
November 1, 2002
Interest in the interface of nonstationarity and nonlinearity has been increasing in the econometric literature. The motivation for this development maybe be traced to the perceived possibility that processes following nonlinear models maybe mistakenly taken to be unit root or long-memory nonstationary. This paper considers the possibility that processes may exhibit both long memory and nonlinearity. We test against the possibility that the process ut in the model (1-L)dyt = ut is nonlinear. We do not assume a particular parametric form for the nonlinear process but construct a pure significance test. Clearly, such a test could be straightforwardly constructed if d were known. Unfortunately, if a linear model is assumed while estimating d the power of the test will be reduced. We propose new more powerful tests for this problem. We present Monte Carlo evidence on the performance of the new tests and apply them to Yen real exchange rates.
J.E.L classification codes: C22, C14, F31
Keywords:Long memory, Nonlinearity, Neural networks,