Purchasing power parity in Mexico: a historical note

Abstract

Purchasing power parity has been the subject of many empirical studies. Much of this work has focused on recent history in developed countries. This article reports results of tests for nonlinear, mean reversion of the real exchange rate for a less-developed country, Mexico, using a previously unexploited data set of monthly observations for 1930–1960. The test results provide limited support for Purchasing Power Parity (PPP).

Notes

¹See Sarno and Taylor (2002) for a comprehensive survey of the evidence.

²The upward bias can be compounded by the degree of temporal aggregation in the data, as demonstrated by Taylor.

³In studies of the law of one price Sarno et al. (2004) and Juvenal and Taylor (2008) found that transactions costs are sizeable and vary substantially across countries and sectors. See also, the special issue of Applied Economics Letters, January 2009, devoted to long-run PPP.

⁴The peso depreciated during this period so that the decline in per capita GDP is even more dramatic when expressed in dollars; from about $1239 to $562.

⁵Historical price indices are not available for the entire country. Mexico City was, and remains, the most important commercial and industrial location in the country.

⁶Stationarity implies a process that is mean reverting. In PPP studies it is assumed that the mean is the PPP value of the real exchange rate, an assumption maintained in this article. Conceptually, it is possible that the rate reverts to a mean different from its PPP value. See the discussion in Taylor and Taylor (2004).

⁷Even though a central bank might implement a discrete devaluation once the real exchange rate is outside its target range, the prices of individual, traded goods can also adjust to move the real exchange rate towards its PPP value. Because individual goods likely have different threshold values because of different transactions costs, a test based on a smooth transition, autoregressive model is more appropriate than a threshold autoregressive model.

⁸See SLN for additional discussion of the test equation. In particular, the symmetric version of the test is valid when the data are demeaned using the sample mean rather than the population mean.

⁹See Table 1 of the SLN paper. The critical value is scarcely affected by the number of observations; it is −2.83 for 500 observations for example.

¹⁰For example see the discussion on pages 49 and 101–2 of Cádenas.

¹¹Convergence is sensitive to the value of the delay parameter, but a number of the specifications converge and none produce results indicating rejection of the unit root null.