Testing for purchasing power parity: a nonlinear approach

This dissertation focuses on testing purchasing power parity. The special feature of the paper is that we switch from conventional unit-root test and cointegration test to a category of asymmetric testing methodology, namely, TAR (Threshold AutoRegressive) model test and M-TAR (Momentum Threshold AutoRegressive) model test. Enders & Granger (1998) and Enders & Sikols (1998) recently made these approaches available;Conventional testing methodology rarely supports PPP strongly even though PPP has very sound theoretical foundation. The deviation could come from two perspectives: first, by claiming the national price index usually is downward stickey, we question the correctness of the symmetric adjustment mechanism which is implied by the alternative hypothesis in the conventional test method. Second, the data set people usually employed are not very appropriate for testing PPP in that CPI or WPI usually are not constructed by using the same mixture of goods, and same weights on each good across countries;Based on these two observations, we apply TAR model and M-TAR model to a very special data set which was kindly provided to us by John Pippenger. Because the data was published to adjust the salaries of German diplomats and foreign service people outside Germany, the price levels are constructed using weights that represent the typical spending pattern of a four-person family in Germany. Therefore, this data set does not have weight problem;As we expect, these two improvements together made we accept PPP in general;A fractional cointegration test is applied to the same data set for a comparison purpose. We found that the more powerful mean-reversion detecting method did not work better than Engle-Granger test significantly. Therefore we conclude that the incorporation of the asymmetric test methodology is very essential;We also tested PPP by using a very general Fourier convergence test.