Three essays on more powerful unit root tests with non-normal errors

Abstract

This dissertation is concerned with finding ways to improve the power of unit root tests. This dissertation consists of three essays. In the first essay, we extends the Lagrange Multiplier (LM) unit toot tests of Schmidt and Phillips (1992) to utilize information contained in non-normal errors. The new tests adopt the Residual Augmented Least Squares (RALS) estimation procedure of Im and Schmidt (2008). This essay complements the work of Im, Lee and Tieslau (2012) who adopt the RALS procedure for DF-based tests. This essay provides the relevant asymptotic distribution and the corresponding critical values of the new tests. The RALS-LM tests show improved power over the RALS-DF tests. Moreover, the main advantage of the RALS-LM tests lies in the invariance feature that the distribution does not depend on the nuisance parameter in the presence of level-breaks. The second essay tests the Prebisch-Singer hypothesis by examining paths of primary commodity prices which are known to exhibit multiple structural breaks. In order to examine the issue more properly, we first suggest new unit root tests that can allow for structural breaks in both the intercept and the slope. Then, we adopt the RALS procedure to gain much improved power when the error term follows a non-normal distribution. Since the suggested test is more powerful and free of nuisance parameters, rejection of the null can be considered as more accurate evidence of stationarity. We apply the new test on the recently extended Grilli and Yang index of 24 commodity series from 1900 to 2007. The empirical findings provide significant evidence to support that primary commodity prices are stationary with one or two trend breaks. However, compared with past studies, they provide even weaker evidence to support the Prebisch-Singer hypothesis. The third essay extends the Fourier Lagrange Multiplier (FLM) unit root tests of Enders and Lee (2012a) by using the RALS estimation procedure of Im and Schmidt (2008). While the F\LM type of tests can be used to control for smooth structural breaks of an unknown functional form, the RALS procedure can utilize additional higher-moment information contained in non-normal errors. For these new tests, knowledge of the underlying type of non-normal distribution of the error term or the precise functional form of the structure breaks is not required. Our simulation results demonstrate significant power gains over the FLM tests in the presence of non-normal errors.