Population age structure and real exchange rates in the OECD

Abstract

Macroeconomic theory predicts that variations in population cohort sizes will lead to demographically induced real exchange rate movements. While such effects have previously been established for individual countries, this paper exploits cross-sectional time series data to test the prediction for a larger number of economies. A reduced form model with population age shares as regressors is estimated using a panel of 25 OECD countries between 1971 and 2002. The results confirm that demographic structure has significant explanatory power for the real exchange rate and the estimated relationship supports age structure effects in accordance with the life cycle hypothesis.

Keywords:

• Demography
• real exchange rates
• panel data
• forecasts

Acknowledgments

We are grateful to Thomas Lindh, Johan Lyhagen, seminar participants at Uppsala University and three anonymous referees for valuable comments on this paper. Financial support from Sparbankernas forskningsstiftelse is gratefully acknowledged.

Notes

¹See for instance Solow (1956) and Romer (1990).

²Age structure effects of a slightly different kind can possibly be found on the supply side where the Balassa–Samuelson theorem states that countries experiencing a higher relative growth rate should have an appreciating real exchange rate. Demographic impact on growth has been shown in several articles, such as McMillan & Baesel (1990), Bloom & Sachs (1998) and Lindh & Malmberg (1999) and are expected given the relevance of the human capital theory of Schultz (1960) and Becker (1964).

³See for instance Meese & Rogoff (1988) and Mark & Choi (1997).

⁴See for instance for instance Clark & MacDonald (1998) and Alberola et al. (1999).

⁵Apart from aggregate saving, investment and the current account – which were mentioned above – demography has been shown to affect a number of other variables, such as growth, inflation, real interest rates and budget balance. See for instance McMillan & Baesel (1990), Bloom & Sachs (1998) and Lindh & Malmberg (2003).

⁶Moreover, Andersson & Österholm (2005) point out that when using a reduced form model with only age structure data as explanatory variables, the forecast errors of the explanatory variables would be extremely small. This should improve the quality of real exchange rate forecasts compared with the alternative of using other macro variables as explanatory variables, which themselves can be difficult to forecast.

⁷The TCW index is a trade weighted index reflecting bilateral import and export competitiveness as well as competitiveness in third country markets.

⁸For different ways of representing the demographic structure in empirical papers, see, for instance, Fair & Dominguez (1991), Higgins (1998), and Lindh & Malmberg (1999).

⁹Note that the scale is different on the y-axis in the different plots.

¹⁰Children have been excluded from the equations to avoid perfect collinearity since the age shares sum to one.

¹¹Phillips & Moon (1999) showed that even when regressing independent random walks on each other in a panel, the estimated parameter vector is consistent for the average long-run regression coefficient given a number of assumptions.

¹²See for instance Kao & Chiang (2000).

¹³See for instance Abuaf & Jorion (1990) and Taylor & Sarno (1998).

¹⁴The relative effects implied by the point estimates are also in line with the results found in Andersson & Österholm (2005).

¹⁵This could be interpreted as if the FE model should be preferred over the RE model; however, the rejection of the null hypothesis in the Hausman test could also be due to model misspecification. Such misspecification is not completely unlikely since the age model is a reduced form model and not a complete description of everything that affects the real exchange rate. In particular, short-run dynamics are by definition excluded from the model due to the slow movements in age structure.

¹⁶Usage of ex post real interest rates implies imposing rational expectations on the agents in the economy with respect to forecasting the real interest rate.

¹⁷We choose to conduct this exercise with the RE model since this has shown the most robust results across different specifications in the sensitivity analysis.