The long-run effects of the real exchange rate on employment and wages in Canadian manufacturing

ABSTRACT

The literature on real exchange rate effects on the labour market is dominated by short-run analysis showing that there is heterogeneity in the responses of firms or industries to a real exchange rate shock. Analysing data on Canadian manufacturing industries, I conclude that there is a common long-run equilibrium across all manufacturing industries controlling for their openness to trade after varying adjustments to a real exchange rate shock have taken place. This conclusion is important from the perspective of policy making because it helps to form expectations about the effects of a real exchange rate movement on the labour market. The results suggest that real appreciation leads to economically significant reductions in employment in manufacturing in the long run. Real wages decrease in industries that are highly engaged in international trade and somewhat increase in industries that are relatively closed to international trade. Both employment and real wages converge quickly to the long-run equilibrium.

KEYWORDS:

• Real exchange rate
• labour markets
• employment
• error-correction models
• Canada

Acknowledgements

I would like to thank Danny Leung and Lynda Khalaf for their encouragement and advice and two anonymous referees of this journal for their very insightful and helpful comments.

Disclosure statement

No potential conflict of interest was reported by the author. The views expressed in the paper are solely the views of the author, and no responsibility for them should be attributed to any institution with which the author is affiliated.

Notes

1. Although the ARDL model allows for the estimation of short- and long-run coefficients, this is essentially a model of short-run dynamics because the long-run coefficients are a function of the short-run coefficients.

2. See Pesaran (2006), Chudik and Pesaran (2015) and Binder and Offermanns (2014) for the derivation of the approximation.

3. The PMG procedure includes running industry-specific regressions to calculate mean group estimates used as initial estimates of long-run parameters and to calculate industry-specific short-run effects and the speed of adjustment. There are only 41 observations in each industry-specific regression (43 years minus the maximum number of lags set at 2).

4. The average employment correlates with the domestic demand for manufacturing output at 0.95 and the average real wage at 0.86.

5. Pesaran, Shin, and Smith (1999) show expressions of coefficients in (2.2) in terms of coefficients in (2.1).

6. I omit the notation of logarithm in the equations, but for greater clarity I use it in the tables showing the results of the analysis.

7. Some authors include a measure of the alternative wage in their models (e.g. Chakrabarti 2003; Revenga 1992). I considered but passed on this control variable. I calculated it as the real wage in the Canadian business sector excluding the manufacturing sector, and I found it very strongly correlated, at 0.96, with the domestic demand for manufacturing output and that it is never statistically significant in regressions. A related issue is that I found a strong correlation between the domestic demand for manufacturing output and US GDP, 0.86. Nevertheless, I retain both variables in the model because it may be important to control for the Canadian-specific component of the aggregate demand even if it is small. Indeed, in some regressions, both variables appear to be statistically significant.

8. I use the GAUSS code available on the Journal of Applied Econometrics Data Archive webpage http://qed.econ.queensu.ca/jae/datasets/banerjee003/.

9. I use the code of PMG estimator written by Dr. Y. Shin in Gauss language, which is available on Dr. Pesaran’s webpage: http://www.econ.cam.ac.uk/people/emeritus/mhp1/published-articles. I made two modifications of the code. First, I modified the code so that no lags of the structural break dummy were included in the model; without the modification the code would determine the number of lags using an information criterion on par with other regressors. Second, I modified the code so that the interaction term with the net trade exposure was differenced in the short-run part of the model as $\mathrm{\Delta }ln(re{r}_{it})\cdot t{r}_{it}^F$; without the modification, it would be differenced as $\mathrm{\Delta }\left[(ln(re{r}_{it})\cdot t{r}_{it}^F\right]$.

10. Standard errors in parentheses, calculated using the formula shown in Greene (2000, 326). * p<0.01, ** p<0.05, *** p<0.1.

11. The authors point out that one of the reasons for the difference in the results between the United Kingdom and the United States could be that US employment is measured by hours of work (as in this paper) and UK employment is measured by the number of employees.

12. To perform this calculation, the data on the price of output of the US manufacturing sector were obtained from the Bureau of Labor Statistics, the data on the price of output of the Canadian manufacturing sector were obtained from CANSIM, table 3830022, and the data on the nominal exchange rate were obtained from CANSIM, table 1760064.

13. Standard errors in parentheses. They are calculated using the formula shown in Greene (2000, 326). * p<0.01, ** p<0.05, *** p<0.1.

14. For example, since the 1960s, the United States has run the Trade Adjustment Assistance programme aimed at helping workers adjust to import competition, and the EU recently introduced a programme, the European Globalisation Adjustment Fund, aimed at helping workers harmed by structural changes in world trade patterns.