1 Introduction

In 1999, the Euro replaced the national currencies of 11 European Union member states. While this has had a significant impact on the volume of intra-EU trade, mostly due to reducing transactions costs, it may also have had an effect on EU nations’ trade with countries outside the region. The introduction of a common currency was just the final phase of a number of fixed- and floating-rate regimes that have been in place in Europe over the past half century. Regardless of the choice of regime, the main determinants of trade flows—income and exchange rates—can still influence trade flows outside the region. This study looks at the case of Spanish trade over the period from 1962 to 2009 to address the role of these determinants on the country’s industry-level bilateral trade flows with the United States. Controlling for the switchover to the Euro, we find that for the majority of the 74 industries examined, currency depreciations or devaluations do not significantly improve the trade balance.

Since the 1973 collapse of Bretton Woods, which specified a fixed exchange rate of 60 (and later 70) pesetas per dollar, Spain has remained tightly integrated into the European exchange-rate system. Nevertheless, there has been ample room for nominal as well as real currency fluctuations. Under the European Exchange Rate Mechanism, which was instituted in 1979, the peseta was allowed to fluctuate within a wider band than many other European nations. Vis-à-vis the dollar, of course, the European currencies floated—just as the Euro does today. The Spanish–US real exchange rate, which exhibits alternating periods of appreciation and depreciation, but no overall recent trend, is shown in Fig. 1.

Fig. 1
figure 1

Spanish–US real exchange rate, 1962–2009

Full size image

Previous studies tend not to focus on the specific case of Spanish trade, particularly with the United States. Perhaps this is because the countries are not considered to be “major” trade partners. Nevertheless, their combined exports and imports are generally around $1.5 billion per year. Certain studies that have examined the exchange-rate sensitivity of trade flows have included Spain in their list of countries, however. Some, which focus on the Marshall–Lerner condition, include Warner and Kreinin (1983), Andersen (1993), Bahmani-Oskooee and Niroomand (1998), and Bahmani-Oskooee and Kara (2005). Others, which take more of a “J-curve” approach, include Miles (1979), Himarios (1989), and Bahmani-Oskooee and Alse (1994). These tend to show that Spain’s trade balance responds little to exchange-rate changes. More recently, Langwasser (2009) confirms this, finding no long-run response to changes in the exchange rate for Spain.

Further discussion of these branches of the literature are provided by Bahmani-Oskooee and Ratha (2004), Bahmani-Oskooee and Hegerty (2010), and Bahmani-Oskooee et al. (2011). In general, because of the tendency of these studies not to find evidence of a link between currency changes and countries’ trade balances, the recent trend in the literature has been to disaggregate countries’ trade flows by industry, usually at the SITC three-digit level. This study does the same, examining the Spanish trade balance with the United States for 74 individual industries. Cointegration analysis using annual data show that only 40 industries exhibit a long-run relationship among the variables. For 26 of them, the trade balances improve after a currency depreciation.

This paper proceeds as follows: Sect. 2 outlines the empirical methodology. Section 3 provides the results, and Sect. 4 concludes. The data are described in detail in the Appendix.

2 Methodology

A country’s bilateral trade balance can most easily be analyzed with a reduced-form model in which both the exporting and the importing countries’ income, as well as a measure of the relative price, serve as the main determinants. Following Bahmani-Oskooee and Hegerty (2009), among others, we use the CPI-based real exchange rate as our price variable. We also include a dummy variable (which equals 1 beginning in 1999) to account for the Euro conversion.

Our choice of cointegration methodology is the Autoregressive Distributed Lag (ARDL) approach of Pesaran et al. (2001). This method has been shown to have good small-sample properties; works with stationary as well as I(1) variables; and is able to provide short-run estimates, long-run estimates, and a cointegration test within a single equation. We begin the procedure by identifying the long-run determinants of the trade balance of industry i outlined by Eq. (1):

$$ \ln \left( {\frac{{X_{it} }}{{M_{it} }}} \right) = \alpha + \beta \ln Y_{t}^{\text{US}} + \lambda \ln Y_{t}^{\text{Spain}} + \theta \ln {\text{REX}}_{t} + \varepsilon_{t} $$
(1)

where the trade balance is defined as ratio of Spain’s export of industry i to US over her imports of industry i from the US. As argued in the literature this measure is unit free and it allows us to express the model in natural log form. Three main long-run determinants are identified to be level of economic activity in the US denoted by Y US and in Spain denoted by Y Spain as well as the real exchange rate denoted by REX. We expect an estimate of β to be positive. As US economy grows, Spanish exports rise. By the same token an estimate of λ is expected to be negative since an increase in Spanish economic activity induces an increase in her imports. Finally, since an increase in REX reflects a real depreciation of pesota against US dollar, an estimate of θ is expected to be positive if depreciation is to improve industry i’s trade balance.

Estimate of Eq. (1) by any method does not allow us to infer the short-run effects or the J-Curve effect of a depreciation. The remedy is to express (1) in an error-correction format as in Eq. (2):

$$ \begin{aligned} \Updelta \ln \left( {\frac{{X_{it} }}{{M_{it} }}} \right) & = \alpha + \beta {\text{EURO}}_{t} + \sum\limits_{j = 1}^{n1} {\gamma_{j} \Updelta \ln \left( {\frac{{X_{i,t - j} }}{{M_{i,t - j} }}} \right)} + \sum\limits_{j = 0}^{n2} {\delta_{j} \Updelta \ln Y_{t - j}^{\text{US}} } + \sum\limits_{j = 0}^{n3} {\lambda_{j} \Updelta \ln Y_{t - j}^{\text{Spain}} } \\ & \quad + \sum\limits_{j = 0}^{n4} {\pi_{j} \Updelta \ln {\text{REX}}_{t - j} } + \theta_{1} \ln \left( {\frac{{X_{i,t - 1} }}{{M_{i,t - 1} }}} \right) + \theta_{2} \ln Y_{t - 1}^{\text{US}} + \theta_{3} \ln Y_{t - 1}^{\text{Spain}} + \theta_{4} \ln {\text{REX}}_{t - 1} + \varepsilon_{t} \\ \end{aligned} $$
(2)

Equation (2) is an error-correction specification in which lagged level variables are included as a substitute for lagged error term from (1) in an Engle-Granger sense. The main advantage of this specification is that the short-run effects and the long-run effects are estimated by a single specification such as (2). Since this specification is to be estimated only, we add the mentioned dummy (i.e., EURO) at this stage. Pesaran et al. (2001) propose applying the F test for joint significance of lagged level variable to justify cointegration. However, the F test has new critical values that they tabulate.

Following the literature, in estimating (1) we use a set criterion such as Akaike Information Criterion to select the lag length (out of a maximum of four). Our short-run estimates are then obtained via the difference terms in this OLS estimation, while the long-run estimates are obtained by the estimates of θ2–θ4 normalized on θ1.Footnote 1

3 Results

We now examine the bilateral trade flows for our 74 industries over the (annual) period from 1962 to 2009. Our first step is to test for cointegration. As mentioned above we do this with Pesaran et al.’s (2001) F test for joint significance of lagged level variables in (2). The authors’ nonstandard set of critical values is extended by Narayan (2005) for small samples such as ours. This test requires both a lower bound value, below which the variables are not cointegrated, and an upper bound value, above which the variables have a long-run relationship. If an equation’s F-statistic falls in between the bounds, we use an alternative test, grouping the fitted values of the lagged level terms into a single variable (i.e., error-correction term denoted by ECMt−1) and re-estimating the equation after imposing the optimum lags. If the coefficient of the lagged error-correction term is significantly negative, we can say that any shock is eventually “undone” as the variables move together back toward equilibrium.Footnote 2

Our cointegration test statistics are provided in Table 1. We find that out of the 74 industries, 30 have F-statistics below Narayan’s (2005) lower bound of 2.873, while 23 statistics are above the upper bound of 3.973. Most of the remaining 21 industries that lie in the intermediate range are shown to be cointegrated when the ECM test is applied; overall, 40 of the 74 industries are shown to be cointegrated. We thus continue our analysis with short-run coefficients for all 74 industries, but examine the long-run estimates only for the 40 cointegrated industries.

Table 1 Cointegration test results
Full size table

Table 2 shows that, even in the short run, relatively few industries are affected by currency depreciations or devaluations. Of the 74 industries, 47 have at least one significant REX coefficient. This is a relatively low percentage, compared with similar studies of this type that have been performed for other countries, as detailed by Bahmani-Oskooee and Hegerty (2010). The signs of these coefficients are both positive as well as negative, and a single industry might have both at different lags. As we will discuss below, there is little evidence of any “J-curve” effect here.

Table 2 Short-run coefficient estimates, all industries
Full size table

Our long-run coefficients, given in Table 3, show that the determinants have a stronger effect. Here, though, the income coefficients do not consistently have their expected signs. Only eight industries carry the expected signs for both Spanish and US income; 10 industries actually carry “perverse” coefficients that are negative for US income and positive for Spanish income. The EURO dummy is significant in only eight cases. It is clear that something else is driving these trade flows.

Table 3 Long-run coefficient estimates, cointegrated industries
Full size table

The real exchange rate is by far the strongest determinant of the trade flows of these 40 cointegrated industries. Twenty-six have significant coefficients, all of which carry the expected positive sign. These are presented separately in Table 3. This proportion—26 out of 74—is consistent with those found in other industry-level studies. In the detailed analysis by Bahmani-Oskooee and Hegerty (2010), a large number of country studies find that around one-third to one-half of industries respond positively to an exchange-rate depreciation.

One interesting result of our analysis is that we find almost no evidence of any “J-curve,” by any definition. As Bahmani-Oskooee and Hegerty (2010) note, the J-curve effect (where a currency depreciation initially worsens a country’s trade balance before short-run rigidities are worked through and the trade balance improves) can be isolated in two ways with short- and long-run coefficients. In one, a J-curve can be determined when short-run exchange-rate coefficients are negative at early lags and positive at later lags. In another, short-run coefficients are negative while the long-run coefficient is positive. We find possible evidence of the first definition of the J-curve in only one industry (numbered #612), but this industry does not carry a significant long-run exchange coefficient. Support for the second definition is somewhat stronger, as four industries (numbered 532, 642, 696, and 724) fulfill the criteria. But, since 90% of our cointegrated industries and 94% of all industries do not, we cannot claim much overall support for this effect.

We provide diagnostic statistics in Table 4. In general, our model appears to be free of serial autocorrelation (via a Lagrange multiplier test), functional misspecification (RESET test) and instability (CUSUM and CUSUM-squared tests of residuals). In addition, adjusted R 2 is relatively high.

Table 4 Diagnostic statistics, cointegrated industries
Full size table

It is clear from Table 3 that all classes of industry have significant, as well as insignificant REX, coefficients. Are there any characteristics that might help us describe which sectors or industries might be more likely to respond significantly to exchange-rate fluctuations? To answer this question, we first rank the industries according to certain classifications and then examine how these ranks differ for cointegrated industries and those that have significant REX coefficients. These results are given in Table 5.

Table 5 Comparison of industries with and without significant REX coefficients
Full size table

We rank the industries in three ways: By growth (in the trade balance over the 1962–2009 period), the size of each industry’s trade balance in 2009, and each industry’s 2009 trade share. These shares and their construction are given in Table 1; the others are available upon request. We then compare the average rank once we separate the industries by different types of statistical significance. The average rank of each group should be 37.5, since 74 industries are examined. We then look for a wide gap between the set that contains a significantly positive REX coefficient and the other groups. If this group has a noticeably lower average rank, we can say that faster-growing or large industries are more likely to benefit from a depreciation in the long run. Likewise, larger average ranks suggest that benefits would accrue to slower-growing or smaller industries.

We compare three sets of groups, but find that these characteristics explain relatively little. First, we compare cointegrated versus uncointegrated industries. The average rank of the two groups is very similar for our industry characteristics. This implies that industries with a large trade share are no more or less likely to be cointegrated. The one characteristic where the average trade shares diverge is growth, suggesting that growing industries (with a low rank) are more likely to be cointegrated.

These patterns persist in our other two choices of groups. When we compare three groups (industries with a significantly positive REX coefficient, those that are cointegrated but have an insignificant coefficient, and uncointegrated industries), we again find very little difference in average rankings for the trade balance and trade share, but a possible relationship where faster-growing industries are more likely to respond positively to a real depreciation. When we combine all industries that do not have a significant REX coefficient (regardless of cointegration) with those that do, we confirm our results above. We are not surprised that industries where the trade balance is growing might be responding to the exchange rate, as that is the point of our study. We can only say that large trading industries do not seem to respond any differently than small ones to currency fluctuations in the long run.

We also examine our industries by sector, assessing what fraction of the industries within each 1-digit SITC classification experience an increase in the trade balance after a depreciation. These fractions are given in Table 6. Within most sectors, between 20 and 40% of the three-digit industries have significantly positive long-run REX coefficients. The only sector with a relatively large share is category 7 (Machinery and Transport Equipment), with six of 13 industries responding. On the other hand, category 8 (Miscellaneous Manufactured Articles) has a share that is lower than 20 percent. Considering these results alongside those of category 6 (Manufactured Goods), for which only a quarter of industries were affected, we can say that there is no specific pattern regarding durable versus non-durable goods. Again, industry characteristics provide only limited explanation of which types of product might be more sensitive to exchange-rate fluctuations.

Table 6 Industry classifications and significant REX coefficients
Full size table

4 Conclusion

While Western Europe has often linked its countries’ currencies to one another in the post-Bretton Woods era—culminating in the Euro—exchange rates have nevertheless floated freely against the US dollar. As a result, real appreciations and depreciations have the capacity to hinder or help European trade with the largest economy in the world. At the same time, individual industries might respond differently from one another, due to specific characteristics that might influence their particular sensitivity to currency fluctuations. The recent trend in the literature has been to analyze the effects of currency movements on industry-level trade flows rather than aggregate or bilateral ones.

This study examines the specific case of Spain’s trade with the United States, particularly for 74 individual industries. We model each industry’s trade balance as a function of Spanish and US income, as well as the real exchange rate. Using cointegration analysis, we then test what effect real depreciations vis-à-vis the dollar have on trade in the short- and long run.

We find that only slightly more than half of the industries show a significant short-run relationship, which is fewer than similar studies of other countries’ industry trade have found. Our long-run evidence, however, finds some support for long-run improvements in the trade balance: While only 40 of the 74 industries show evidence of cointegration, the trade balances of 26 of these (about a third of the total) increase after a devaluation. This is a similar proportion to what has been found in previous studies.

We then rank the countries according to their growth and relative trade share and compare whether these characteristics are more likely to be found among the industries that respond to a devaluation. We find little relationship, except for some evidence that industries with faster-growing trade might also be sensitive to the real exchange rate. Another analysis, where we examine what proportion of trade flows in broader, SITC 1-digit sectors are sensitive to devaluations, find that category 7 (Machinery) has the largest fraction of industries that respond to changes in the real exchange rate.