Export Hysteresis, Capacity Constraints, and Uncertainty: A Smooth Transition Analysis for Euro Area Member Countries

Abstract

We argue that, under certain conditions described by a sunk cost hysteresis model, firms consider exports as a substitute for domestic demand. This is valid also on the macroeconomic level where the switch from the domestic market to the export market and vice versa takes place in a smooth manner. Areas of weak reaction of exports to changes in domestic demand are widened by uncertainty. Our econometric model for six euro area countries suggests domestic demand and capacity constraints as additional variables for export equations. We apply the exponential and logistic variant of a smooth transition regression model and find that domestic demand developments and uncertainty are relevant for short-run export dynamics particularly during more extreme stages of the business cycle. A substitutive relationship between domestic and foreign sales can most clearly be found for France, Greece, and Ireland (exponential smooth transition regression (ESTR) model) and France, Portugal, and Italy (logistic smooth transition regression (LSTR) model), providing evidence of the importance of sunk costs and hysteresis in international trade between the member states of the Economic and Monetary Union (EMU). What is more, our empirical results are robust to the inclusion of a variable measuring European policy uncertainty. In some cases (Italy, Greece, and Portugal) the results underscore the empirical validity of the export hysteresis under uncertainty model. (JEL codes: F14, C22, C50, C51, F10).

1. Introduction

A number of euro area countries which recorded large current account deficits in the period prior to the European debt and banking crisis starting in 2010 have seen a significant correction of their external imbalances, in particular the trade balance, over recent years. Falling imports have been an important part of this correction due to low domestic demand. However, at the same time, exports and export market shares have been continuously increasing in most of these countries since 2009. Shrinking unit labour costs and falling real effective exchange rates are able to explain only part of the gains in export market shares. Christodoulopoulou and Tkacevs (2014) find that only 60–70% of variation in exports can be explained by standard export equations. It thus seems likely that non-price-related factors have been important in explaining export performance. The residuals from a standard approach to model exports are potentially consistent with the parallel dramatic fall of domestic demand. A possible relationship between domestic demand and exports could be particularly important in the current economic situation of substantial macroeconomic adjustment needs and very low domestic demand.

The relation between domestic demand and exports is not straightforward and could be either negative (substitutive) or positive (complementary). A recent survey of literature on this topic is presented in Esteves and Rua (2013). Theoretical reasons for a positive link between domestic demand and exports may be due to increased efficiency from learning by doing effects (Belke et al. 2013) or due to liquidity generated by cash flow from exports which can help overcome liquidity constraints for domestic operations (Berman et al. 2011). Theory has identified a negative relationship between domestic demand and exports mostly at the firm level. Several studies have been concerned with the effects of domestic demand pressure on the inclination and capacity to export. These studies are not numerous, but go back several decades.¹

The main argument is that—in the short run—exporting firms face capacity constraints or increasing marginal costs and thus have to substitute sales between their domestic and foreign markets. An increase in demand for exports cannot be satisfied in the short run as long as capacity is highly utilized and most of production is sold on the domestic market. Conversely, with low domestic demand, for instance during a domestic recession, firms will be able to shift more resources to export activities; to compensate for the decline in domestic sales, firms will increase their efforts to export. Besides pull factors (e.g. foreign demand), export performance can thus also be determined by push factors (such as low capacity utilization). Besides the studies mentioned above, more recent empirical literature (Ilmakunnas and Nurmi 2007; Máñez et al. 2008; Berman et al. 2011; Blum et al. 2011; Vannoorenberghe 2012 or Ahn and McQuoid 2013) generally identifies a significant negative effect of domestic demand pressure on exports for several countries, among them the UK, the USA, Germany, Spain, Israel, Turkey, Morocco, and India.

The main lesson from the literature is that any exercise of modelling export performance should take into account not only the factors driving external demand (and thus impact export activity from the demand side) but also those influencing domestic demand (which affect export activity mostly through the supply side). Moreover, the studies underline the necessity of clearly differentiating between the short run and the long run. One potential limitation of the previous literature is that the complementarity versus substitutability property of domestic demand and exports has often been analysed in a linear framework. The relationship between domestic demand and export performance may however vary with economic conditions and thus be of a non-linear nature.

Assuming a substitutive relationship between domestic demand and exports, following a domestic demand shock, firms will try to shift sales between the two markets. However, entering the export market or shifting more sales towards it usually implies sunk costs. These are costs firm need to pay that are irreversible ex post (Baldwin and Krugman 1989), and the significance of this knowledge diminishes rapidly after leaving the export market (Belke et al. 2013).

In that respect, we can distinguish two cases. First, with a negative domestic demand shock and sunk costs for entering or shifting to the export market, firms will therefore be reluctant to pay these costs as long as capacity is still relatively highly utilized. Once capacity utilization falls below a certain threshold, firms might be more willing to pay sunk investment costs, as these costs and the effort of selling in the foreign market might be lower than the cost of running excess capacity.²

Exports in this case can be considered as ‘survival-driven’. Secondly, following a positive domestic demand shock, firms might not be able to serve both domestic and foreign markets due to highly utilized capacities. If they prefer producing for the domestic market, firms would consider shifting sales to that market once a certain high capacity utilization threshold has been crossed. With sunk costs, leaving the export market or shifting sales away from it implies that these costs would have to be paid again upon trying to re-enter the export market or reshifting sales towards it in the future.

Overall, these arguments suggest that only if certain low or high capacity utilization thresholds have been crossed, firms will change their export behaviour. Only if a domestic demand shock is accompanied by extreme changes in capacity utilization will firms shift their sales to another market. As long as capacity is utilized to a more normal degree and operates within these lower and upper thresholds, firms are working in a ‘band of inaction’ where sunk costs hinder firms from changing their export behaviour, even though capacities might exist for those firms that are not yet very active in foreign markets.³

This also implies that, once capacity utilization thresholds have been crossed on either end and firms have shifted sales among markets, they will be reluctant to shift again once capacity returns back to more normal levels. There is thus strong persistence in export behaviour which can be traced back to the theory of hysteresis (Baldwin and Krugman 1989). Export hysteresis is the tendency of a temporary change in export behaviour to become permanent. It is particularly important in the current weak economic situation of several euro area member states; firms increase efforts to shift sales to the export market, given weak domestic demand, and this might not be a cyclical change but rather a persistent improvement as firms will often decide to stay in the foreign market even once domestic demand picks up again as they are trying to avoid repaying sunk costs.

We thus essentially present a story of dynamic investment in the presence of high fixed cost and capacity constraint. This story is consistent with firms switching from selling in the domestic market to the foreign market as soon as the level of domestic demand falls short of a given trigger threshold. But what about the heterogeneity element that induces switching only by some firms, but not in all firms? Here we refer to Belke and Goecke (2005) who, starting from the idea of a ‘band of inaction’, focus on the issue of aggregation. They are able to derive an aggregation process, considering heterogeneity of sunk exit/entry costs and/or the extent of uncertainty of the future market situation and/or the elasticity of demand. This is resulting in different triggers for different firms. This (realistic) consideration of heterogeneity alters the hysteresis characteristics when aggregating from the microeconomic to the macroeconomic level. Due to heterogeneity in firm characteristics such as the magnitude of sunk costs or the productivity level, firms exit (and entry) sequentially and not all in a time from (into the) the market, and the resulting aggregated hysteresis loop thus shows no discontinuities. This is rather important in our context because, absent this feature, all firms would switch if there is a large negative domestic demand shock. This would contradict the abundant micro-evidence in the trade literature that actually the most engaged exporters are also faring best on the domestic market.⁴

Notably, in this model of export hysteresis, the band of inaction is widened by uncertainty (Belke and Goecke, 2005). This is because a forward-looking firm considers future effects of a present sunk cost ‘investment’. If the exogenous variable demand is stochastic, a real option approach applies (Dixit 1989, Pindyck 1998, 1991; Belke and Goecke 2001). An inactive firm deciding on a present entry or to stay passive will include the option to enter later as a potential alternative. Demand which is presently contributing to cover costs may in a stochastic situation decrease in the future. By staying passive the firm can avoid future losses if this situation will realize. Moreover, an instantaneous entry kills the option to enter later and to ‘wait-and-see’ if the future demand movement will turn out to be (un)favourable. Thus, in a stochastic situation, the sunk costs and, additionally, an option value of waiting have to be covered to trigger an entry. Therefore, uncertainty implies an upward shift of the entry trigger demand. The same is valid for the exit trigger demand which will shrink in a situation with uncertainty. Belke and Goecke (2005) show that this line of reasoning is valid also at the macroeconomic aggregated level. Thus, uncertainty leads to a widening of the band of inaction also at the macroeconomic level, aggravating the hysteresis property of the firm’s export behaviour.⁵

Our article builds on this sunk cost hysteresis model and explicitly tests for a short-run non-linear relationship between domestic demand and exports from a macroeconomic perspective. A particular asymmetric effect was already considered in Esteves and Rua (2013) for the case of Portugal. Belke, Oeking, and Setzer (2015) consider the relation of domestic demand and export of goods in several euro area countries. Our analysis goes beyond these studies by investigating six euro area countries with significant current account deficits in the pre-crisis period (Spain, Portugal, Italy, France, Ireland, and Greece) employing the export of both goods and services.

The focus upon these former current account deficit countries reflects our intent to analyse countries in which firms were experiencing a fall in domestic demand. After the start of EMU, a real appreciation sets in for peripheral Euro area member countries, especially in those experiencing a massive housing boom, namely, Ireland and Spain. As a consequence, their competitiveness went down further. At the same time nominal long-term interest rates converged among EMU member countries, inducing low real interest rates in the peripheral countries with higher inflation. This in turn stifled spending and inflation even further, leading to growing current account deficits (see, for instance, Krugman et al. 2015, pp. 687ff.). Since currency devaluation was no option for these countries, it became clear that the necessary real exchange rate adjustment implied a period of low inflation or even deflation in combination with significant unemployment and protracted recession including weakness of domestic demand.⁶

The ‘doom loop’ among banks and governments contributed significantly to this development.

Moreover, we go beyond the papers mentioned above by thoroughly conducting tests for structural breaks common to the countries under investigation and integrating an uncertainty variable in our estimations.

Following Belke et al. (2015), we implement a smooth transition regression (STR) model such that we can specify aggregated non-linearities with a high degree of flexibility. We argue that the strength of the relation between domestic demand and exports depends on capacity constraints and more generally the business cycle. Besides the possibility that substitutability will increase after reaching either the upper or lower threshold (i.e. giving rise to symmetry around the band of inaction), we also allow for the possibility that exports react sharper in a recession than during an economic expansion (giving rise to asymmetry around the band of inaction). This is achieved by relying on either an exponential or logistic variant of smooth transition specification. The aggregation at the macro level allows us to draw results on net effects of capacity utilization on the economies as a whole. This is of special importance in the discussion of macroeconomic adjustment and the reduction of current account imbalances in the euro area.

The article proceeds as follows. Taking the simple sunk cost-based hysteresis model as a starting point, we carry out some pretesting in terms of unit roots and cointegration in Section 2. Based on the cointegration results, we set up an error-correction export equation and incorporate non-linearities as suggested by our theoretical considerations. These STR models, including several robustness tests among them the incorporation of an uncertainty variable, are estimated in Section 3. Section 4 finally concludes and conveys an outlook on further research avenues.

2. Empirical Strategy

2.1 Data

Our data stem from different sources (cf. Table A1): data on real exports (⁠$xt$⁠) and real domestic demand (⁠$ddt$⁠) come from the national statistical offices (either obtained from Eurostat or Oxford Economics). Value-added exports (⁠$xtva$⁠) have been constructed by data from the World Input-Output Database (wiod.org); the annual data were converted to quarterly data by applying cubic spline interpolation. The real effective exchange rate has been obtained from Eurostat and is an index deflated by consumer price indices with a country’s 15 main trading partners (⁠$rt$⁠). Alternatively, the same source provides an index deflated by unit labour costs with a country’s 24 main trading partners (⁠$rtULC$⁠). Data on foreign demand (⁠$yt*$⁠) from the ECB are based on trade-weighted imports for a country’s 15 main trading partners. Capacity utilization data in the manufacturing industry (⁠$zt$⁠) come from the Business and Consumer Surveys by the European Commission, available from Eurostat or Insee in the case of France. For Ireland, data on capacity utilization are not available. Instead, we use the output gap (interpolated data from AMECO). As an uncertainty variable for our robustness checks, we employ the economic policy uncertainty index relevant for the European Union as a whole because the respective index was not available for the individual Euro area member countries for such a long sample period like ours (http://www.policyuncertainty.com/europe_monthly.html). The final data set is quarterly and mostly available from 1980:Q1 to 2012:Q4.

2.2 Non-stationarity and cointegration tests

We adopt a methodology developed by Bai and Perron (1998, 2003) to account for possible instabilities in the long-run coefficients of Equation (1). Their basic idea is to choose break points such that the sum of squared residuals for all observations is minimized. The estimated break points by definition represent the linear combination of these segments which achieve a minimum of the sum of squared residuals (Bai and Perron, 2003). Table 1 shows the two most important break points for the six countries analysed, accompanied by the 95% upper and lower confidence intervals. In case of Spain the first break point occurs in 1993Q4 with 1993Q3 and 1994Q1 providing the 95% confidence intervals.

The results of Table 1 are also useful in the context of unifying our testing and estimation approach. One may ask, for instance, to what extent the differences in the cointegrating test and cointegrating equation estimation results across countries we usually gain for the export equations of the six EMU member countries analysed here (available on request) are driven by the fact that the specific break points are different across countries?⁷

To check this, we follow the option to see what happens when imposing a common break point for all countries. Do the data strongly reject such an assumption? To account for this issue, we have implemented one common break point for all countries in all estimations contained in this article with an eye on the results displayed in Table 1, that is by a (permanent) dummy denoting the most common break in 1993:04 which may proxy the fallout of the 1992/1993 crisis of the European Monetary System (EMS).⁸

The respective findings (long-term relation and Engle -Granger test for cointegration) based on a common break in the fourth quarter of 1993 are provided in Table 2, alternative specifications with country-specific break points were contained in the previous version of this article and are available upon request. To be more concrete, we focus on estimating the long-run equilibrium of Equation (1) by FMOLS (fully modified ordinary least squares). To test for cointegration, we apply the Engle–Granger test for cointegration. The results are displayed in the last column with the respective critical values from MacKinnon (1991). Because $êt∼I0$⁠, we can conclude that for each country the error-terms are stationary and a cointegration relationship between the variables is thus present. It should be mentioned that the findings are not greatly affected by the specification of the break points. The error terms from estimations based on common and individual break points turn out to be highly correlated and the short-term findings provided in the following are not affected by these findings.

The sign of the estimated coefficients (negative for our exchange rate variable and positive for the foreign demand variable) overall corresponds with our priors from theory. The effects are in line with theory. As an example, the effect of y (exp) is always positive. Our estimation results for the long-run relations largely match those of other studies, both in terms of sign and size of the coefficients (see, for instance, European Commission 2011). We do not come up with a more detailed analysis here, as our main focus is on the short-run relation, and slightly different long-run specifications did not change the following results in a noteworthy way.⁹

2.3 Empirical model

As explained above, we apply a non-linear framework to capture any short-run non-linear impact in the relation between domestic demand and exports regarding the state of the economy. We consider each country’s economic condition by looking at deviations of its capacity utilization from its mean. Looking at short-run adjustments and in particular at the short-run relation between exports and domestic demand, we take into account the long-run equilibrium estimated above. For this purpose, we apply an error-correction model. As already mentioned in the introduction, we take into account the possibility of a non-linear adjustment process to a linear long-run equilibrium relationship depending on the state of the economy. Based on an economy’s export performance where individual firm-level decisions are aggregated, it may not seem adequate to assume that this threshold is a sudden and abrupt change which is identical for all firms and which is commonly known; the smooth transition regression (STR) model thus allows for gradual regime change when the timing of the regime switch varies on an aggregated level.

The main difference between our short- and long-run specifications is the inclusion of the domestic demand variable. Based on the theoretical arguments given in the introduction above, domestic demand should enter our estimations in the short run only.¹⁰

In contrast to the long-run estimation, we do not include a structural break in the short-run estimations of Equation (2) because this specification already includes the smooth transition non-linearities. Furthermore, a break in the long-run relation does not imply that short-run dynamics change as well; by excluding these breaks we are also able to reduce our model’s complexity.

The logistic transition function increases monotonically from 0 to 1, as the value of transition variable $z$ increases. We can therefore distinguish two different regimes in the extreme and a gradual transition between these two: (i) negative deviations of the transition variable from its threshold: $limzt-j→-∞⁡ Fzt-j,γ,c=0$⁠, when the model collapses to just the first set of brackets in Equation (2), that is the linear part, and (ii) positive deviations of the transition variable from its threshold: $limzt-j→+∞⁡ Fzt-j,γ,c=1$⁠. The coefficients $α, β, θ, μ, η, δ$ gradually change between these two extreme values with changing $zt-j$⁠.

In our setting, this implies testing the hypothesis that domestic sales are substituted by foreign sales once capacity utilization falls below a certain threshold. Further reductions in capacity reduction strengthen the substitution of domestic demand by exports. Note that there is no threshold for the opposite case of high capacity utilization. In other words, the band of inaction is only constrained on one side (for negative but not for positive deviations of capacity utilization from its mean).

This transition function is symmetric (U-shaped) around $zt - j=c$⁠, so that the two different regimes to distinguish between are: (i) large deviations of the transition variable from its threshold: $limzt - j→±∞⁡ Fzt - jt - j,γ,c=1$ and (ii) small deviations of the transition variable from its threshold: $limzt-j→c⁡Fzt - jt - j,γ,c=0$⁠, that is the linear part.

In our case, the ESTR model represents the hypothesis of symmetric hysteresis in exports. Here, both positive and negative deviations of the threshold variable capacity utilization from its average value $c$ matter. This implies that as long as the deviation of the transitional variable from $c$ is small, there would be no or only small substitution effects from domestic demand to exports (band of inaction). However, if the capacity utilization variable is either significantly above or below its average value, we would expect substitution effects.

The main difference between these two forms of non-linear error-correction model refers to different deviations of the transition variable from its threshold value (its mean): the LSTR case distinguishes positive vs. negative deviations and the ESTR model large vs. small deviations from equilibrium. The former will therefore yield asymmetric results around the threshold, and the latter symmetric deviations above or below the threshold.

3. Empirical Results

3.1 Specification tests

Where $Wt=(Δddt,Δddt-1,…, Δddt-p,Δyt*,…,Δyt-p*,Δrt,…,Δrt-p, Δxt-1,…,Δxt-p,ε^t-1)$ and $φi=φi1,…,φiq'$ with $q$ equal to the number of regressors (i.e. the number of elements in $Wt$⁠).

The null hypothesis, which refers to the linear model being adequate, is tested as $H0: φi=0$with $i=2,3,4$against the alternative$H1$ where at least one $φi≠0$⁠, implying that the higher-order terms are significant (Teräsvirta 1998). The test statistic has an $χ2$ distribution with three degrees of freedom.¹³

This procedure also enables the choice of an adequate transition variable. In the case of the linearity hypothesis being rejected, a method for choosing the latter lies in computing the test statistic for several transition functions, that is different values of the lag order j, and selecting the configuration for which its value is maximized (van Dijk et al. 2002). Teräsvirta (1994, 1998) has proven that this procedure is adequate.

The decision rule to select the most adequate transition function introduced by Teräsvirta (1994) is as follows. If the rejection of $H03$ is the strongest one in terms of lowest p-value or largest test statistic respectively, the ESTR model should be chosen or otherwise the LSTR model should be preferred.¹⁴

Table 3 displays the empirical realizations of the non-linearity test statistics, while Table 4 provides the test statistic to distinguish between both configurations.

The common procedure behind selecting the lag length of the transition variable in the Teräsvirta testing and modelling cycle intuitively seems to pick the lags for which the chance to observe non-linearity is strongest (Belke et al. 2015). However, this would seem to artificially favour our prior which is to find non-linearity in the data.¹⁵

To react fully as possible to this important caveat, we provide findings from Table 3 on where a common lag order is just imposed for all countries to make the results more comparable.

The findings in Table 3 show that non-linearity is essentially never rejected for all lag orders. Table 4 shows that a distinction between both model configurations turns out to be difficult. Teräsvirta (1998) suggests estimating different models and choosing between the different specifications and different lag lengths only during evaluation of the estimation results. LSTR and ESTR models generally form very close substitutes. Tests as the ones above should thus be seen as a starting point for estimation instead of providing clear-cut outcome at this early stage of analysis. Taking the ambiguous findings into account, we therefore estimate both LSTR and ESTR models for all countries. To allow for a direct comparison of our findings, we always use a unified lag length of 2 for our transition function in our study.

3.2 Estimation

To evaluate our parameters, we estimate Equation (2) with non-linear least squares (NLS). Our main coefficient of interest $β$ depends on the transition function $F(zt - j,γ,c)$ as depicted in either Equation (4) or (5). To choose the final specifications, we examine our estimation results by simple judgment regarding the plausibility of the parameter values and the regimes which the models imply, the models’ convergence properties, goodness-of-fit measures, and a test of no residual autocorrelation. For this misspecification test, we apply a variant of the Breusch–Godfrey LM (BG) test suitable for non-linear estimation as suggested in Teräsvirta (1998). The test’s null hypothesis is that there is no $p$th order serial correlation in our residuals$ut$⁠. The test regresses $u∼t$ (the estimated residuals) on $u∼t-1,…,u∼t-p$and the partial derivatives of the regression function with respect to$γ$⁠.

Estimation results are found in Table 5 for countries with an ESTR specification and in Table 6 for countries with an LSTR specification. Our theoretical priors suggest a negative coefficient for the coefficient $β$⁠, that is, a substitution effect from domestic demand to exports during times of low or high capacity utilization. When estimating the ESTR model, coefficient $β1i$ for $Fzt - j,γ,c=0$ (i.e. the linear model) shows us results for capacity utilization levels around the threshold level. The joint coefficient $β1i+β2i$ for the case when$Fzt - jt - j,γ,c=1$ yields the results for positive and negative deviations from our threshold. In the LSTR case, $β1i$ represents low levels of capacity utilization, and $β1i+β2i$ high values of capacity utilization.

3.3 Estimation results

Let us first turn to the econometric specification based on an ESTR model (Table 5). For France, Greece, and Ireland, the effects for 1 or 2 lags display negative values for extreme levels of past capacity utilization, while negative contemporaneous effects are not identified except for the case of Italy. The contemporaneous coefficient is positive for Ireland; for Italy the results are ambiguous. This suggests a substitutive relationship between domestic and foreign sales when the economy is close to peak or trough. When capacity utilization is very low, firms react to a fall in domestic demand by increasing their efforts to export. Conversely, if the economy operates at high capacity utilization, capacity constraints imply that an increase in domestic demand triggers a reallocation of resources from external to domestic clients. A positive coefficient may imply that the short-run liquidity channel dominates, whereby the cash flow generated by exports is used to finance domestic operations, and the existence of increasing returns dominates the capacity constraints channel (Belke et al. 2015, and Berman et al. 2011). As argued above, also this general pattern is in line with the prevalence of hysteresis and the band of inaction due to switching costs for suppliers between serving the domestic and foreign market.

We now turn to our findings based on the LSTR specification (Table 6). The contemporaneous substitution coefficient is positive for France and Ireland but insignificant for the other countries in the first regime (beta 0, business cycle trough). For Portugal and France, the substitution coefficient becomes negative in a boom (which is reflected by the sum of both coefficients). While there is hardly any significance of the coefficients for a lag of two quarters, we find a positive coefficient for Spain and Portugal in case of negative capacity utilization (trough) and a negative one for Italy. However, the sum of both coefficients becomes positive for Italy, France, and Greece in case of positive capacity utilization (boom).

Overall, our empirical results presented in Tables 5 and 6 suggest that the relationship between domestic sales and exports depends on capacity utilization and the business cycle. Evidence of a substitutive relationship between domestic and foreign sales varies among countries and with different lag lengths. The findings are broadly in line with the gain in export market shares in several euro area (debt and banking) crisis countries during the subsequent recession. There is more diversity across countries during other stages of the business cycle suggesting that capacity constraints and the liquidity channel play a different role across countries and/or partly cancel each other out.

Seen on the whole, thus, more analytical rigour by imposing a common break point in all country-specific estimations and tests and common lags for the transition variable comes at a ‘cost’ in terms of less adaptation of country specifics and thus economically plausible results.¹⁶

3.4 Robustness check: taking stock of political uncertainty

In the following, we are reporting the results of an important robustness check of our estimations.¹⁷

As a final check, we included economic policy uncertainty in our analysis, a variable playing a prominent role in explaining band of inactions in the reaction of exports to its main drivers (see, for instance, Belke and Kronen 2016). More precisely, we rely on the change of European policy uncertainty in period t−1 (i.e. lagged one quarter) as a transition variable in Equation (3). As derived in Section 1, the empirical results of the hysteretic export equation may turn out to be more pronounced because the band of inaction gets larger with increasing uncertainty. The findings for the LSTR model are given in Table 7.

For the following interpretation of the estimation results, we have to keep in mind that the first coefficient holds if economic policy uncertainty decreases, while the sum of both coefficients is relevant for an increase in economic policy uncertainty. This interpretation is based on Equation (3) for both values of the transition function. The evolution of the transition function reflects the band of inaction. The transition function increases from 0 to 1 if economic policy uncertainty increases. The second coefficient in Table 7 always provides the additional effect once the transition function increases from 0 to 1. Hence, the overall effect for the highest increase in uncertainty is given by the sum of both coefficients which reflects the case where the transition function is 1. We focus on two potential effects in the following: the effect on the substitution coefficient and the one on the global demand coefficient. In each case, we analyse the impact with a delay of the regressor of 1 or 2 lags. Due to the rich number of models estimated, several coefficients are as usual insignificant. Nevertheless, a few important results are worth mentioning.

The original effect of domestic sales on exports is always either positive or insignificant in the regime with a decrease in uncertainty. However, the effect for an increase in uncertainty (measured by the sum of both coefficients) becomes negative for Italy (1 lag) and Greece (1 lag) and is strongly reduced for Portugal (2 lags). This points to a substitution effect as a result of higher uncertainty. Interestingly, the sum of both coefficients becomes positive for Spain. These results broadly confirm robustness of the results gained before with respect to the inclusion of a formerly omitted variable ‘policy uncertainty’. In other words, there is no omitted variable bias in our case. On the contrary, our model of export hysteresis presented in Section 1 is corroborated for some EMU member countries such as Italy, Greece and, a bit less so, also for Portugal. The ‘non-case’ of Ireland may be explained by the higher flexibility of the Irish economy compared to its Southern European counterparts. Flexible prices and immigration may have made capacity constraints less binding (see Belke et al. 2015).

For information only, differences between the two regimes are also observed with regard to the effect of world demand on exports. The effect of world demand on exports of France and Greece is negative in case of an increase in policy uncertainty. The opposite is observed for Portugal and Spain, where the effects on exports increase in case of higher uncertainty. While no effect is observed for Italy, the specific effect on Greece depends on the lag order, but the sum of coefficients for higher uncertainty is always significant. A useful extension for further research will be the consideration of country-specific economic policy uncertainty indices. However, these are not available over the full sample (see www.policyuncertainty.com).

4. Conclusions

In this article, we have analysed the relation between domestic demand and exports for six euro area countries using non-linear smooth transition estimations faced with a strong a priori restriction of common break points and common lags across individual country specifications. To illustrate the results gained in this article, it seems worthwhile to contrast them with those identified by us without these a priori restrictions.

Our empirical results based on individual and potentially different break point specifications which have been gained in a previous version of this article (available on request) clearly indicated that domestic demand behaviour is relevant for the short-run dynamics of several euro area member countries’ exports. The estimation results suggested that on an aggregated level, contemporary and lagged domestic demand developments can affect a country’s export performance significantly.

We found that in the cases of Spain, Portugal, and Italy, the symmetric non-linearity of the relation manifests itself in a contemporary substitutive relationship between domestic demand and export activity if deviations from average capacity utilization are large. This is somewhat independent of their sign, but we found stronger evidence for notably low levels of capacity utilization. In other words, the substitution effect from domestic demand to exports turns out to be stronger and more significant during more extreme stages of the business cycle. During periods of more average levels of capacity utilization, our empirical evidence pointed to a band of inaction in which the relation between domestic and foreign sales is complementary. On a micro level, theoretical reasons for these findings can be found in the sunk costs hysteresis approach. For France, the evidence for non-linearity was weaker. We found evidence of mostly complementary relationships. In the cases of Ireland and Greece, we detected an asymmetric non-linear relationship between domestic demand and exports. Domestic demand and exports are slightly substitutive during a business cycle trough and complements during normal times and in a boom.

Overall, our results mostly confirmed the short-run non-linear relationship between domestic and foreign sales depending on capacity constraints. A substitutive relationship with low capacity utilization shows up most clearly for Spain, Portugal, and Italy.

We also provided first ideas for why we believe there are valid reasons for the different findings in the other countries (such as the high number of multinational corporations in Ireland, the lower openness of the French economy, or the small Greek tradable sector). However, deriving more detailed explanations for these heterogeneous results for some countries in our sample provides an interesting area for future research. A further interesting avenue could lead to a more disaggregated, sectoral analysis to understand the underlying firm behaviour in more detail (Esteves and Prades 2017).

A final interesting avenue was taken in this article: we conducted all necessary estimations and tests based on a common break point implementation not to bias the results into the direction of the empirical model with the highest degree of non-linearity and on common lags for all countries to avoid the impression that the country-specific regression models were over-fitted till significance. The pattern of the results changed as follows. A substitutive relationship between domestic and foreign sales can now most clearly be found for France, Greece, and Ireland (ESTR model) and France, Portugal, and Italy (LSTR model), providing evidence of the importance of sunk costs and hysteresis in international trade in these EMU member countries.

What is more, our empirical results are robust to the inclusion of a variable measuring European policy uncertainty. In some cases (Italy, Greece, and Portugal) the results underscore the empirical validity of the export hysteresis under uncertainty model. While we do not feel legitimized to go more deeply into economic interpretations of the country-specific results due to the pronounced ex ante restrictions such as the imposition of common break points and of common lags across all country-specific empirical models, we would like to stress the finding of a general non-linear pattern of export activity of the Euro area member countries with a remarkable goodness of fit.

Seen on the whole, the macroeconomic perspective is able to offer insights on overall adjustment effects for euro area countries with previous imbalances. In recent years, the six countries under consideration which recorded large current account deficits before the European debt and banking crisis starting in 2010 have seen a significant correction of their external imbalances. This holds in particular for their trade balances, and exports have been a key adjustment factor. Our results provide one explanation for the rising exports besides standard competitiveness arguments; the observed increase in export market shares accompanying the reduction of the current account deficits could be due to non-price related factors, such a low domestic demand leading to survival-driven exports, instead of an increase in price competitiveness as expected by sustainable structural reforms. This argument appears to be especially relevant in the current period for the countries under consideration in which their capacities have been utilized only to a low degree and domestic demand has fallen strongly. Low domestic demand then did not only affect imports but at the same time exports and has thus strongly contributed to the external adjustment.

Regarding policy implications, our findings provide important insights for the discussion of macroeconomic adjustment and the reduction of imbalances in the euro area. Prima facie, our results for specific countries could suggest that domestic demand and exports are negatively related only in the short run, triggered by current economic conditions. To the extent that the closure of the output gap is driven by a pickup in domestic demand, a lot of the gains in export market shares of vulnerable euro area countries could be lost in the long run. In such a scenario, analyses of cyclically adjusted current account balances could possibly overestimate the structural adjustment to the degree that weak domestic economic conditions impact not only the import side of the net trade equation but also the export side.

On the other hand, at least three factors give rise to the hope that the gains in export market performance may be of a more long-run nature. First, domestic demand conditions in peripheral economies are likely to remain depressed as long as the debt burden of both private and public sector remains high. An extended period of deleveraging pressure increases the chances that the reallocation of resources to the export sector will also be more permanent, possibly also fostering increased export-oriented foreign direct investment into distribution networks and other hedging activities (Belke et al. 2013). This would make the hypothesized substitutive relationship between domestic demand and exports more long run. Secondly, our sunk cost hysteresis model suggests that once domestic producers have paid sunk costs for shifting production to exports, they remain in a band of inaction even as the business cycle improves. Reversing export market participation should not be expected as long as there are capacities for serving both domestic and foreign market. Thirdly, with increasing exports today and a pickup in domestic demand in the future, a complementary relation between domestic sales and exports might develop in the long run due to improvements in efficiency encouraged by learning-by-doing effects. In conclusion, the export increase could therefore be lasting and a substantial part of the gains in export market shares may not only a cyclical phenomenon, but indeed be of a more structural nature.

Acknowledgments

The authors would like to thank Gabriel Felbermayr and two anonymous referees for valuable comments and above all Joscha Beckmann but also Stéphane Dees, Philippe de Rougemont, Frauke Skudelny, Florian Verheyen, participants in the IMF research department’s brown bag seminar, the ECB’s DED seminar, and the CESifo-Delphi conference on ‘Current Account Adjustments’ for valuable comments and suggestions. The authors thank Christian Buelens for the provision of data on value-added exports.

References

Appendix