Growth and exchange rate volatility: a panel data analysis

Abstract

The aim of this article is to assess the role of Real Exchange Rate (RER) volatility on long-run economic growth for a set of 82 advanced and emerging economies, using a panel data set ranging from 1970 to 2009. With an accurate measure for exchange rate volatility, the results for the two-step system Generalized Method of Moments (GMM) panel growth models show that a more (less) volatile RER has a significant negative (positive) impact on economic growth. The results are also robust for different model specifications.

Keywords:

• exchange rate
• volatility
• economic growth
• panel data analysis

JEL Classification::

• F43
• C33
• F31

Notes

¹ In this study we actually use Real Effective Exchange Rate (REER) instead of RER for the reasons discussed ahead.

² See Balassa (1978), Chow (1987), Bahmani-Oskooee et al. (1991), Ahmad and Kwan (1991), Oxley (1993), Ahmad and Harnhirun (1995), Krueger (1998) and Alguacil et al. (2002).

³ See Edwards (1988), Rodrik (2008), Eichengreen (2008), Aghion et al. (2009) and Berg and Miao (2010).

⁴ We have addressed possible endogeneity of the explanatory variables and the correlation between the error term and the lagged-dependent variable, which is a common issue present in this sort of analysis, through the use of a system of regressions in differences and levels as suggested by Blundell and Bond (1998).

⁵ OxMetrics by Timberlake Consultants Ltd.

⁶ The construction of the REER index uses the nominal exchange rate as units of US dollar relative to domestic currency, meaning that a higher (lower) value is associated to REER appreciation (depreciation).

⁷ Source: IFS, Penn World Table, WDI (2010), Barro and Lee (2000).

⁸ See Romer (1986), Lucas (1988), Barro and Sala-i-Martin (1991, 1992, 1995).

⁹ DULatin: Argentina, Brazil, Chile, Mexico, Peru, Ecuador, Paraguay, Uruguay, Colombia, Bolivia, Nicaragua, Costa Rica, Panama, Dominican Republic, El Salvador, Guatemala, Honduras, Haiti, Trinidad and Tobago, Venezuela and Jamaica. DUG7: Canada, France, Germany, Italy, Japan, United States, United Kingdom. DUAsia: South Korea, China, India, Sri Lanka, Bangladesh, Malaysia, Pakistan, Philippines, Singapore, Thailand and Indonesia.

¹⁰ Fixed and random effects models are not reported here for convenience, but the results are available upon request.

¹¹ Table 4, for all estimated system GMM growth models, reports the overidentification tests (Hansen and Hansen-in-difference).

¹² We have set the Laglimits to (1 1). A more detailed presentation of both methods to reduce the number of instruments, including matrix notation, can be found in Roodman (2009b), pp. 148–149.

¹³ Measured as the log of secondary schooling years of the total population aged 15 and over in the first year of each 5-year period.

¹⁴ The first set of empirical results is for fixed and random effects (robust and bootstrap). Again, these results are not reported for convenience, but they are available upon request. The fixed and random effect estimations do not include lagged growth or initial GDP level (convergence) as explanatory variables. All estimated models include time dummy variables. The results show that all estimated coefficients for the conditional REER Volatility are negative and statistically significant, regardless of changes in model specification and of the correction (robust or bootstrap) in the standard error of the regression coefficient. Such outcome indicates that countries with lower (higher) REER volatility face higher (lower) long-run growth over time, which is in line with other works, such as Dollar (1992).

¹⁵ The GMM estimators have one- and two-step variants. The two-step variant is asymptotically more efficient, but the reported standard errors tend to be downward biased (Arellano and Bond, 1991; Blundell and Bond, 1998). To deal with this problem, our estimated models (Table 4) use a finite sample correction to the covariance matrix (Windmeijer, 2005) to make two-step robust estimations more efficient.

¹⁶ The Jackknife method with the cluster option in Stata is used by clustering on the panel identifier variable (countries) in order to drop each observational unit in turn.