Re-examining purchasing power parity for selected emerging markets and African countries

Abstract

The validity of Purchasing Power Parity (PPP) is re-examined using data for some emerging market economies and African countries, extending recent works of Doğanlar et al. (2009) and Chang et al. (2010), respectively. For this purpose, we apply new unit root tests that allow for nonlinearities and structural change in the data-generating process. The results of this study suggest that although linear unit root test provides evidence in favour of PPP only in a few cases, the new nonlinear unit root tests suggest that the PPP proposition holds in majority of the sample countries.

Keywords:

• purchasing power parity
• nonlinearity
• structural breaks
• emerging markets

JEL Classification:

• C12
• C32
• F31

Notes

¹Mexico and Peru are the only two countries in which the Purchasing Power Parity (PPP) hypothesis is supported.

²We choose the sample periods used in Doğanlar et al. (2009) and Chang et al. (2010), and extend the data until last available date. Although South Africa was included in emerging markets list in Doğanlar et al. (2009), we show this country only under the list of African countries.

³It is a well-known fact that financial time series may suffer from time-varying volatility. In order to see whether the residuals of the test regressions exhibit Autoregressive Conditional Heteroscedasticity (ARCH) effect or not, we applied Engle's (1982) ARCH Lagrange Multiplier (LM) test against conditional heteroscedasticity. The results of the tests that are available upon request suggest that ARCH effects are prominent in some series, as is often the case with financial variables. Kapetanios et al. (2003) argued that ‘… time varying volatility may raise two problems. First, heteroscedasticity will interfere with inference on the appropriate number of augmentations to be included for the test. Second, the power of both the nonlinear and ADF tests may be affected, though the asymptotic size of both tests is not affected by the existence of heteroscedasticity. We addressed the first issue by using heteroscedastic consistent SEs in regressions determining the augmentation terms, but we can do little about the second issue. However, there is no prior reason to suppose that the power of our test is less or more affected by heteroscedasticity than that of the ADF test, so in this sense comparison with the ADF test remains fair’. Therefore, following Kapetanios et al. (2003), we also selected the number of augmentation terms using heteroscedastic consistent SEs in the test regressions.