Exchange rate volatility and exports: a firm-level analysis

Abstract

The relationship between real exports and exchange rate volatility is investigated using panel data analysis at the firm level. Results indicate that there is no negative or positive relationship between volatility and real exports. In addition, firm size and level of international activity do not influence the size and significance of the volatility effect on exports. However, there is some evidence that firms use import revenue to lower their exchange rate exposure.

Notes

¹ The existence of a forward market to hedge currency risk separates this risk from the firm's trade decisions. When hedging is possible, all production and trade decisions are made on the basis of the forward exchange rate. This is known as the ‘Separation Theorem’. For more details, see Ethier (1973), Baron (1976) or Kawai and Zilcha (1986) for the separation theorem in trade literature, and Holthausen (1979) in the theory of the firm.

² By a lack of consensus, we mean that theoretical works do not clearly specify which variables should be included in the conditioning set, or the measure of volatility, or correct model specification. Solakoğlu (2000) investigates the robustness of this so-called relationship by using Extreme Bound Analysis of Leamer and Leonard (1983), Leamer (1985) and finds that it is not robust. In a simulation, Gagnon (1993) also shows that this effect is too small to be detected.

³ Instead of using official exchange rates, Bahmani-Oskooee (2002) study utilizes black market exchange rates, which may be significantly different from official rates for developing countries, and finds adverse effects of exchange rate volatility on trade flows.

⁴ Many of the Turkish firms are smaller compared to European countries. In addition, smaller-firms are mostly owner-managed and they do not have access to personnel with required knowledge to hedge. It is also known that pricing strategy of these firms are based on costs plus a profit margin.

⁵ www.ise.gov.tr

⁶ About 87% of the firms are in manufacturing, and about 10% in construction industries. The remaining firms are operating in other industries.

⁷ GDP value of industrialized countries is used as a measure of economic activity in the importing country.

⁸ In calculating the relative price measure, CPI for industrialized countries is used.

⁹ If the coverage ratio is above one, it will indicate that the value of imports is larger than the value of exports. Firms who do not have access to other hedging tools will try to protect themselves by exporting more and lowering their exposure in the importing side.

¹⁰ It can be argued that firms will first try to sell to the markets where they will maximize their profits. If they shift their export to another country, we may expect that the prices they receive for the goods sold should be either at most the same or lower.

¹¹ Given that relative price, income measure for importing countries and exchange rates do not change from firm-to-firm in our estimation period, it is impossible to include all in one equation. Therefore, our dependent variable is regressed individually on these variables and information that is not captured by them – the error term – is used as the dependent variable in the final equation. In a way, we are using a two-step estimation which leads to inefficiency, but not inconsistency.

¹² For the sake of brewity, we do not provide the coefficient estimates on constant(s) and on macro variables used in the first step of the estimation process.

¹³ High values of LM test favour GLS over OLS suggesting some exogenous factors, which may be correlated with the dependent variable and possibly omitted from the model, are not correlated with the right hand side variables which results inefficient OLS estimates whereas GLS gives efficient estimates.

¹⁴ The null hypothesis states no correlation, thus large values of the Hausman's X2 test suggest statistical preference for a fixed effects model specification. Fixed-effects estimation assumes that differences across firms can be captured by differences in the constant term. However, if the differences between firms are not just parametric shifts of the regression function, it may be more appropriate to view individual specific constant terms as randomly distributed across cross-sectional units with random-effects model.

¹⁵ Note that it would be better to use revenue to determine firm size. However, due to reporting problems in Turkey, these figures were not reliable. Each segment, in Table 3, approximately represents 25% of the firms.

¹⁶ The most comprehensive specification is used in Table 3. Additionally, results from Tables 1 and 2 show that we can focus only on this specification without losing any important findings. Yet, we tried other two specifications and were not able to find any different results.

¹⁷ On average, firms in segment 1 and 4 have the lowest coverage ratios indicating that firms in segment 4 might be using other hedging tools more often than others.