Specification testing of ARCH and nonlinear time-series models

Since the introduction of autoregressive conditional heteroscedasticity (ARCH) by Engle, there has been considerable interest in econometrics in models in which the variance of the current observation is a function of past observations. A difficulty in applying the ARCH model and other types of nonlinear time series models is lack of specification test for determining the appropriate model. This dissertation proposes several new tests to aid the selection of ARCH and nonlinear time series models.