Monetary Policy and Real Currency Appreciation: A BEER Model for the Mexican Peso

Abstract

A notable feature of the Mexican economy since the late 1980s was the persistent real appreciation of the peso. The appreciation – a key development that helps to explain Mexico's slow rate of economic growth – took place despite changes in the exchange-rate regime, yet with an unchanging focus of monetary policy on gradually reducing the inflation rate. Thus, the frequent assumption that only real-side variables (as opposed to monetary ones) have a lasting or ‘long-run’ effect on the real exchange may not suit the recent Mexican case. The paper presents the results of an econometric study of exchange rate determination in Mexico for the period 1990Q1–2006Q4. The study is based on the so-called BEER (Behavioral Equilibrium Exchange Rate) model, which relies on Johansen's cointegration methodology and jointly considers real-side and monetary determinants. The estimation results – in the form of two- and three-equation cointegration models – show that, controlling for the influence of real-side determinants, the peso–dollar interest differential had a statistically and economically significant long-run effect on the peso's real exchange rate.

Keywords:

• Real currency appreciation
• real exchange rate
• monetary policy
• Mexican peso
• BEER model
• Johansen's cointegration methodology

JEL CLASSIFICATIONS :

• C32
• F31
• F35
• F41
• O54

Acknowledgements

Financial support from Consejo Nacional de Ciencia y Tecnología (CONACYT), project 47140-S, is gratefully acknowledged.

Notes

¹See Williamson (1994) for a discussion of the proposition, Kildegaard (2006) for an example applied to Mexico, and Frankel (2007) for a recent counter-example. The proposition does not imply, of course, that an appreciation caused by real-side variables, such as an oil windfall or a surge in foreign capital inflows, will not have negative growth effects, particularly in the tradables sector (see Krugman, 1987).

²See Chinn (2006) for a discussion of alternative measures of the real exchange rate.

³The real exchange rate indices in Table 1 are lagged one year to better visualize their effect on macroeconomic performance.

⁴The definition of ‘long-run’ can be a matter of debate (see Driver & Westaway, 2004). In what follows, a long-run relationship is understood as a relationship in levels (as opposed to first differences) between the real exchange rate and its determinants within the specific horizon of 17 years of quarterly observations spanning 1990 to 2006.

⁵BEER studies usually include the country's net foreign assets (NFA) position among the determinants of the real exchange rate. Since there are no readily available quarterly series for Mexico's NFA position, one was constructed using the current account balance as a proxy for the change in NFA. The constructed series resembled closely the annual series presented by Lane and Milesi-Ferretti (2006); compared with the rest of the variables in the models though, it had a higher degree of integration, making its inclusion problematic. Following Fazio et al. (2007) the current account balance itself was considered as a proxy for NFA, but its estimated coefficient tended to have the wrong sign in the real-exchange-rate equation (a result also obtained by those authors). See Note 14 for the inclusion of government debt as a possible determinant of the risk premium.

⁶To avoid confusion, keep in mind that the number of cointegration equations in Johansen's reduced-rank regression methodology is always smaller than the number of endogenous variables in the VAR.

⁷While the series for the multilateral rate are available from the Bank of Mexico, the bilateral US–Mexico ratio must be used to be consistent with the peso–dollar interest differential.

⁸The LR test statistic was 5.65, with a p-value of 0.0592. Note that, somewhat surprisingly, the test would reject the (weak) exogeneity of the oil price at 10% significance, despite the oil price being determined in the world market with no influence from the Mexican economy. The weak exogeneity tests were carried out within an otherwise unrestricted cointegrated VAR, estimated under the assumption of two cointegration relationships (see trace test results below).

⁹An F-test showed that no lag could be eliminated; in addition, removing the fourth lag introduced serial correlation in the residuals.

¹⁰The trace statistic was corrected for small-sample bias using the factor suggested by Johansen (2002) (see note in Table 4). Following standard practice, the cointegration relationships are reported in the form β′ x=0. The estimations were carried out in PcGive 10.

¹¹In most models the interest differential in the inflation equation has a non-significant coefficient. If in equation E.2 the interest differential – which shows a significant positive coefficient – is eliminated, then the real exchange rate coefficient rises strongly, from 45.9 to 88.3; this suggests that the positive coefficient on the interest differential is capturing some of the inflationary effect of the real exchange rate during periods of financial turbulence – when the interest differential and the real exchange rate move in the same direction. Eliminating the interest differential, however, is rejected at 10% by the LR test. In the inflation equation of models E and G, the manufacturing production ratio has a positive coefficient, presumably reflecting a cost-push factor.

¹²The share of oil in goods exports fell from more than 60% in the early 1980s to less than 10% twenty years later. The variable has been problematic in previous studies. Kildegaard (2006) found a significant but positively-signed oil price coefficient in a cointegration analysis for the period 1969–2000; Dabós and Juan-Ramón (2000) obtained a similar result in a regression analysis for the period 1982Q1–1998Q4 – using the international terms of trade rather than the oil price. Edwards (1994) and Elbadawi (1994) discuss theoretically why the terms of trade coefficient may have an ambiguous sign.

¹³Interestingly, our point estimates for the interest-differential coefficient in the long-run equation for the real exchange rate are very similar to those obtained by Frankel (2007) in his study of the South African rand (his point estimates range from 0.019 to 0.022 for the pre-liberalization period).

¹⁴A larger coefficient on the inflation differential was expected under the assumption that the inflation differential could be a significant determinant of the unobserved risk premium. Following the BEER literature, model A was re-estimated with the addition of the government's total debt (in percentage of GDP) as a possible determinant of the risk premium (results available upon request). As expected – given the stationarity of the new variable – the trace test indicated the existence of a new, third cointegration relationship, which captured a positive relationship between government debt and the interest differential. The estimated coefficient on the government's debt in the long-run equation for the real exchange rate showed the expected positive sign, but no economic or statistical significance. Compared with model A, the other coefficients in the real exchange rate equation remained unchanged, including those on the interest and the inflation differentials.

¹⁵The trend had to be removed from model G for the LR test to accept the symmetry restriction.