Exchange rate volatility and export performance: a cointegrated VAR approach

Abstract

During the last decades Norwegian exporters have–despite various forms of exchange rate targeting–faced a rather volatile exchange rate which may have influenced their behaviour. Recently, the shift to inflation targeting and a freely floating exchange rate has brought about an even more volatile exchange rate. We examine the causal link between export performance and exchange rate volatility across different monetary policy regimes within the cointegrated Vector Autoregression (VAR) framework using the implied conditional variance from a Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model as a measure of volatility. Although treating the volatility measure as either a stationary or a nonstationary variable in the VAR, we are not able to find any evidence suggesting that export performance has been significantly affected by exchange rate uncertainty. We find, however, that volatility changes proxied by blip dummies related to the monetary policy change from a fixed to a managed floating exchange rate and the Asian financial crises during the 1990s enter significantly in a dynamic model for export growth–in which the level of relative prices and world market demand together with the level of exports constitute a significant cointegration relationship. A forecasting exercise on the dynamic model rejects the hypothesis that increased exchange rate volatility in the wake of inflation targeting in the monetary policy has had a significant impact on export performance.

Acknowledgements

The authors thank Ådne Cappelen, Roger Bjørnstad, Håvard Hungnes and Terje Skjerpen for helpful advice and discussions during the work with the project. Comments from Neil Ericsson at the 2008 Econometric Society European Meeting in Milan and Graham Mizon and Kevin Hoover during their visit at Statistics Norway in April and September 2008 are gratefully acknowledged. The econometric modelling and testing were performed using PcGets (Hendry and Krolzig, 2001) and PcGive 10.3 (Hendry and Doornik, 2001; Doornik and Hendry, 2001a, b). The usual disclaimer applies.

Notes

¹ Bahmani-Oskooee and Hegerty (2007) review recent articles and conclude in line with McKenzie (1999) that the relationship between exchange rate volatility and trade is still unresolved at the empirical level.

² The volume of total Norwegian exports amounted to 738 billions (at fixed 2004-prices) in 2005, which made up around 40% of total Gross Domestic Product (GDP); see Economic Survey (2007) available at http://www.ssb.no/english/subjects/08/05/10/es/.

³ Pagan (1984) does not consider the generated regressor problem in the case of GARCH-based measures of the kind used here. McKenzie (1999) points out, however, that the consistency property of estimated parameters in models with ARCH generated regressors extends to cases of more complicated conditional variance models.

⁴ In what follows, lower case letters indicate natural logarithms of a variable, unless otherwise stated.

⁵ That hedging in forward markets fails to completely eliminate exchange rate risk is discussed in e.g. Arize et al. (2000). The difficulty to provide perfect hedge in financial markets is inter alia related to the fact that forward rates are a poor predictor of future spot rates; see Choudhry (1999) and the references cited therein.

⁶ Bredin et al. (2003) provide a formal model and discuss in detail the conditions under which increased exchange rate variability would lead to increased exports.

⁷ The financial crises, which broke out in Thailand in July 1997, spread itself quickly to several countries in the southeast of Asia during the following autumn and successively to the rest of the world, mainly through lower domestic demand in the troubled Asian economies. Consequently, international trading partners faced reduced export possibilities, which were further amplified by stronger competition in the wake of falling exchange rates in the Asian economies, with downward pressure on prices for trading partners and thus lower earnings, see Economic Survey (1998) available at http://www.ssb.no/english/subjects/08/05/10/es/9801/

⁸ Excluding the smaller trading partners these weights are 23.4% for Sweden, 49.4% for the Euro area, 14.1% for the UK, 7.5% for the US and 5.6% for Japan. These weights are kept time independent in (3) as they have been quite stable throughout the sample period. See the Appendix for further details about the underlying series in (3) and their sources.

⁹ The time series are normalized to unity in 1985Q1.

¹⁰ These and the other test results not reported below are available upon request. Several previous studies also provide evidence to suggest that the distinction between nominal and real exchange rate volatility makes no difference to the results obtained; see, e.g. Thursby and Thursby (1987) and Qian and Varangis (1994).

¹¹ The time series are normalized to unity in 1985Q1.

¹² A battery of augmented Dickey–Fuller tests suggest that the time series for exports, foreign demand and relative prices are all I(1), whereas the volatility series is a borderline case when it comes to being integrated of order zero or one. We remark that the order of integration of the volatility series in principle is possible to deduce from the GARCH model. From Equation 5 we have that . Inserting this in the expression for h_t in (5) assuming a GARCH(1, 1), known α₀, α₁ and β₁ and neglectable estimation uncertainty with respect to these parameters (due to enough available data) gives . Now assume that this expression holds for t = 1, …, T and that h ₀ = 0, then it follows that h ₁ = α₀, , and so on. Since the distribution of ε is specified, it follows what the order of integration of the volatility series would be. We thank Terje Skjerpen for pointing out this to us.

¹³ The choice of lag length and the insignificance of the linear trend are unaltered when the cumulated volatility series is included restrictedly in addition to the volatility series itself (and its lags) entering unrestrictedly in the VAR.

¹⁴ The rank tests are virtually unchanged with different lag length of the volatility series in (8) and (9).

¹⁵ The issue of possible effects on exports of the move to inflation targeting in 2001 is pursued further in the next section.

¹⁶ We should mention that an additional blip dummy D892 _t –albeit with no particular economic rationale–enters the VAR unrestrictedly to mop up extraordinary large residuals in 1989Q2. Noticeably, including D892 _t , D931 _t and D972 _t does not change the estimates of Π₁, Π₂ and Π dramatically compared to the no-dummy VAR. We also notice that the blip dummies in Equation 10 appear less significant when modelled as transitory rather than permanent variables, see Juselius (2006, p. 106).

¹⁷ Although the sign and magnitude of the estimated coefficients are hardly affected by the presence of D892 _t , it is included in the specific model to remove borderline autocorrelation in the residuals at conventional levels.

¹⁸ The share of total exports to the troubled Asian countries was only between 5 and 10% for the European countries; see Economic Survey (1998).

¹⁹ See http://www.norgesbank.no/Pages/Article_55476.aspx