Real exchange rate and relative profit margins: Evidence from non-linear ARDL models for Mexico

ABSTRACT

While many studies have documented the positive effect the real exchange rate (RER) can have on the rate of economic growth in developing countries, the channels of this effect are not yet well understood. The paper contributes to this literature by studying the pass-through of the RER into aggregate relative profit margins in the manufacturing and whole tradables sectors with respect to non-tradables in Mexico during the period 1990–2017. Based on estimates from non-linear error-correction ARDL models, the paper shows the RER affects relative profit margins asymmetrically – with larger reductions after appreciations than increases after depreciations – and that this asymmetry comes from the asymmetric adjustment of both relative costs and to lower extent relative prices. The results support the hypothesis of an RER’s tradables-led economic growth channel and help explain why the RER may affect capital accumulation asymmetrically, with stronger negative effects from appreciations than the positive effects of depreciations, as reported in recent research on Mexico.

KEYWORDS:

• Real exchange rate
• asymmetric exchange rate pass-through
• relative profit margins
• tradables
• non-tradables
• non-linear ARDL model

JEL CODES:

• E31
• F31
• O11
• O14
• O15
• O54

Acknowledgements

The author would like to thank Jaime Ros (†) and three anonymous referees from this journal for their comments on a previous version of the paper. The usual caveat applies.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1 See Table 1 in Aron, MacDonald, and Muellbauer (2014) for a summary of studies specifying country and period coverage, methodology, and main results. As pointed out by the authors, asymmetric pass-through is a novel research topic in emerging market economies. In their large survey, they distinguish only three such studies (131).

2 Other authors have tested for potential asymmetries in the response of other macroeconomic variables in Mexico. See, for example, Bahmani-Oskooee, Halicioglu, and Hegerty (2016) for an analysis of trade flows and Ibarra (2018) for an application to capital accumulation.

3 Broadly speaking, in our dataset the RER and relative costs, prices, and profit margins are integrated of order one, while output gaps and production growth rates are integrated of order zero. While unit-root tests may fail to identify a variable as integrated of order zero (incorrectly indicating an integration of order one due to the presence of structural breaks), what matters for the proper use of the bounds-testing methodology is that the variables do not show an order of integration higher than one.

4 As indicated in equation (2), in the majority of the estimated equations it was also necessary to include intercept shift dummies to control for outliers. The dummies are identified in the notes to each of the tables presented.

5 Thus, following what has become the common practice (see, for example, Giles 2013), the t- and F-tests are applied after simplifying the lag structure of the model.

6 While the number of annual observations is relatively small, they cover a long historical period of a little more than a quarter century. As further evidence, Appendix 2 presents estimations based on a sample of nearly 100 quarterly observations that confirm the existence of a long-run relationship between relative prices and the appreciation and depreciation components of the RER. We cannot use these series in the main part of the paper, though, as there are no quarterly data that enable us to calculate unit costs and profit margins and to distinguish with enough precision between tradables and non-tradables.

7 Changing the base period (or initial value) of the RER in equation (3) would not affect in any way the econometric results reported below, except for a change in the estimated value of the equations’ intercept.

8 Following the literature, we use a threshold of zero-growth to distinguish between RER appreciations and depreciations. An alternative threshold could be the RER’s long-run rate of growth. Given that the growth rate is very low (0.3% annually), however, and that the actual positive changes in the RER in any given year are always above this threshold, the resulting series for RERA and RERD would be identical to the ones we use. In any event, to test for the possibility that the effect of changes in the RER depends on their size, we include quadratic terms for RERA and RERD in one of our robustness exercises.

9 Given the relatively small number of annual observations, and to enhance the confidence in our results, Appendix Table A2 presents equations for the relative price of total and manufacturing goods with respect to services using a sample of nearly 100 quarterly observations from the available period 1994–218 (see sources and definitions of the variables in Appendix 1). The estimations strongly support the existence of a long-run relationship between relative prices and the RER’s appreciation and depreciation components.

10 The null hypothesis takes the form d_i=–ρ, for i=1, 2, in equation (2). There is some ambiguity in the results of the Wald tests. They cannot reject the hypothesis that the appreciation and depreciation coefficients are each equal to one, but they reject the hypothesis that they are equal to each other. As shown by the respective p-values, the latter test has a much smaller statistical significance than the first one. Note, however, that the probability of the hypothesis that the RERA coefficient is equal to one is always visibly greater than the probability of a similar hypothesis for RERD (except in column 1.2, which excludes outlier dummies for illustrative purposes). This is consistent with the existence of an asymmetry.

11 The same results obtain if the equation is estimated without outlier dummies, in this case a single one for the year 2012 (not shown for brevity).

12 As was the case with relative prices, in the relative cost equations a Wald test cannot reject the hypotheses that the coefficients on RERD and RERA are equal to one; but, again, the probability of the hypothesis for the RERA coefficient is always the largest, which is consistent with the existence of an asymmetry.

13 These long-run coefficients are similar but not identical to those shown in column (3.4), which are somewhat smaller, because the latter come from an equation that controls for an outlier year dummy in 2012. The same observation applies to the long-run multipliers shown in the other panels of the figure.

14 Regarding the fall in the relative labor cost, Ibarra and Ros (2019b) propose the explanation that the gains in labor productivity in the tradables sector have accrued to profits rather than wages because of the stagnant or even falling productivity of labor in the informal non-tradable activities, which acts as a drag on wages in the tradables sector.

15 The dataset in Excel format is available from the author upon request.