1 Introduction

In today’s global economy, domestic firms and workers in each country have become increasingly sensitive to international competition. This has induced the emergence of serious concerns about the labor market’s drawbacks of globalization. Such issues figure prominently in the agenda of policy makers around the world. A good illustration of this is the current political pressure, especially exerted by the US and the EU, towards a revaluation of the Chinese renminbi. Indeed, the Chinese government is accused of keeping the value of its currency deliberately undervalued, in order to grant Chinese firms a competitive advantage on the export markets. And yet, recent reports are showing that China has been displaying a substantial appreciation in real terms lately, due to the relatively high increase in wages, thus emphasizing the importance of looking at real exchange rates, besides nominal ones. Footnote 1 This paper studies the impact of real exchange rate movements on net and gross job flows at the industry level, focusing on the manufacturing sector of Belgium, for the time span 1996–2002.

In a context of increasing trade openness, firms are expected to become more responsive to variations in real exchange rates, which reflect the relative prices of competing goods on the international markets (Gourinchas 1998). Indeed, real exchange rates can be seen as synthetic indicators of the competitiveness of domestic firms relative to their foreign competitors. In particular, real exchange rates are influenced by the evolution of nominal exchange rates and by the relative price dynamics across countries. Thus, they capture the influence of monetary policy and currency trading on the financial market, but also the effects of relative costs and productivity dynamics, which are finally reflected in prices.

Until now, only a few papers have studied the relation between real effective exchange rates and job flows: Gourinchas (1998, 1999) on the US and France, respectively, Klein et al. (2003a) on the US, and Moser et al. (2010) on Germany. All these studies have found that real appreciations have a negative impact on net employment growth, and that this impact is magnified by increasing levels of trade exposure, at the firm and industry level. However, differences have been found with respect to the adjustment process, which may work mainly through lower job creation (France and Germany) or through higher job destruction (US), depending on the specific context. Building on this literature, this paper aims at deepening our understanding of the impact of real effective exchange rates on job flows, by focusing, for the first time, on a small open economy such as Belgium.

Belgium constitutes a very interesting case study for this research question. Indeed, it is one of the most open economies in the world, and has experienced the highest increase in trade exposure among the EU countries over the considered period. Footnote 2 Moreover it is characterized by strong labor market rigidities, which are typical of EU continental economies. Footnote 3 Finally, very good data are available at the firm level for the Belgian manufacturing sector, allowing to carry a micro-founded analysis by relying on a large data set. Footnote 4

The empirical analysis is based on the theoretical model by Klein et al. (2003a), where job flows are put in relation with industry-specific real exchange rate (RER) changes and openness to trade. The main prediction of the model is that a real appreciation induces lower job creation and higher job destruction at the industry level, and these effects are stronger the higher is the level of trade exposure in the industry. The main findings of my analysis can be summarized as follows. First, I find that real appreciations do have a negative impact on net employment growth, and this impact increases with the level of trade exposure. Second, concerning the margin of adjustment, the net employment effect is driven by an increase in job destruction, while job creation is not significantly affected. This result is robust to using different measures of job flows, and is stable across a number of different estimations. Such evidence of a job destruction-driven adjustment is in line with earlier findings by Klein et al. (2003a) for the US, and differs from what has been found for other European countries, in particular France (Gourinchas 1999) and Germany (Moser et al. 2010), where the adjustment to RER shocks was mainly driven by the job creation margin. Moser et al. (2010) attributed the difference between the US and Germany to the far-stricter employment protection legislation in Germany, which makes firing costly and thus prevents smooth adjustments to shocks through job destruction. The same explanation could be proposed for France, where employment protection is also high. And yet, according to the OECD Index for the Strictness of Employment Protection, the Belgian labor market displays a level of rigidity in line with Germany and France, and thus much higher than the US one (OECD 2004). Footnote 5 Hence, differences in labor market institutions are not likely to explain the different findings for Belgium on one hand, and France and Germany on the other. In my interpretation, these different findings are consistent with Belgium being a small open economy. Indeed, Belgian firms are operating in a much smaller domestic market than French or German firms, and they face, on average, much higher levels of trade exposure. As a result, they are forced to be more reactive to shocks hitting their international competitiveness. This may explain why Belgian firms adjust through the job destruction margin when they face a real exchange rate shock. In fact, even though destroying jobs can be costly for the firms, due to the strictness of employment protection, the cost of failing to adjust timely to the shock may be even higher in a small open economy. Overall, this highlights the importance of studying the effects of RER changes on job flows in different contexts, where results can be significantly different. Analyzing the case of a small open economy constitutes the main contribution of this study.

As previously mentioned, this paper builds upon previous work by Gourinchas (1988, 1999), Klein et al. (2003a) and Moser et al. (2010). More generally, it is related to the growing body of literature on the connections between international trade and the labor market, as reviewed by Klein et al. (2003b) and Crino’ (2009). Several studies have explored the impact of increasing foreign competition on net employment growth at the industry level. Net job losses have been found to be induced by lower import prices and RER appreciations in open industrialized economies. Footnote 6 More recently, other papers have started to study the implications of trade also on gross job creation and destruction flows. Footnote 7 Focusing on gross flows is important for assessing the adjustment costs implied by increasing trade integration, as resources get reallocated to their most productive uses. Indeed, trade-related adjustment costs are likely to be proportional to gross flows rather than net ones. Moreover, the same net variation in employment might be generated by different combinations of job creation and job destruction, with potentially diverse welfare implications, as discussed by Klein et al. (2003a, b). Footnote 8 Consistently, this paper focuses on the impact of RER movements both on net and on gross job flows.

My analysis starts in Sect. 2.1 with the computation of job flows, following the methodology developed by Davis and Haltiwanger (1992). In Sect. 2.2, I turn to the analysis of trade exposure at the industry level. In Sect. 2.3, I discuss the computation of real effective exchange rates, following the approach by Gourinchas (1998, 1999) and Klein et al. (2003a). In Sect. 3, I present a descriptive analysis of the correlations between job flows, trade exposure and RER movements. In Sect. 4, I first sketch the theoretical model by Klein et al. (2003a), focusing on its main predictions; then, I present the econometric analysis, and discuss the results. Finally, Sect. 5 concludes.

2 Job flows, trade exposure and real exchange rates

2.1 Firm level data and job reallocation analysis

The job reallocation analysis relies on firm level data for 14,599 Belgian companies, operating over the time span 1996–2002, in 82 NACE 3-digit manufacturing industries. Footnote 9 Data are drawn from the Amadeus database of the Bureau Van Dijk, a Belgian consultancy company. For all firms in the sample, the database provides comprehensive company accounts, including the full time equivalent number of employees, on a yearly basis. On average, firms in the sample account for 66 % of total official manufacturing employment (see Table 11 in the "Appendix"). The panel is unbalanced, as some firms enter the sample during the considered period, while others exit. For the purposes of my analysis, the year in which the first observation is recorded denotes a firm’s entry, while exit is assumed to take place in the year after which no new information is available in the data set. Footnote 10 And yet, a firm’s entry in the data set does not necessarily coincide with its market entry, and a firm’s exit from the data set does not necessarily correspond to its exit from the market. This implies that entry and exit can only be measured with an error. This shortcoming of Amadeus has been already discussed by previous studies, for instance Gómez-Salvador et al. (2004). In light of this, as a robustness check, the analysis will be performed both on “standard” measures of job flows, and on what will be referred to as “cont” job flows, i.e., job flow figures based on continuing firms only. The latter figures are computed by excluding, in each year, the contribution of entering and exiting firms to job creation and job destruction, in line with Gómez-Salvador et al. (2004). Footnote 11

For the job flows analysis I adopt the same methodology as in Davis and Haltiwanger (1992), which has been extensively employed in the literature. As a first step, employment growth rates (g ft ) at the firm level are computed as the difference in the number of jobs reported by the firm between year t and t − 1, over the average firm employment in years t and t − 1:

$$ g_{f\left( t\right) }=\frac{jobs_{f\left( t\right) }-jobs_{f\left( t-1\right) }}{x_{f\left( t\right) }} $$
(1)

where f denotes the firm, and \(x_{f\left( t\right) }=\frac{(jobs_{f\left( t\right) }+jobs_{f\left( t-1\right) })}{2}. \) Footnote 12

Then, the job creation rate for industry i at time t is obtained as the weighted summation of all the positive firm growth rates at time t:

$$ Job \,Creation_{i\left( t\right) }=\sum \limits_{f\in S_{i\left( t\right) }^{+}}\omega _{fi\left( t\right) }\ast g_{f\left( t\right) } $$
(2)

where \(S_{i\left( t\right) }^{+}\) denotes the subset of firms in industry i witnessing a positive employment growth at time t, and the weights (\(\omega _{fi\left( t\right) }\)) are defined as the ratio of each firm’s employment over total employment in the industry:Footnote 13

$$ \omega _{fi\left( t\right) }=\frac{x_{f\left( t\right) }}{\sum \nolimits_{f\in i}x_{f\left( t\right) }} $$
(3)

The “standard” measure of job creation in Eq. 2 can be split in two components: (1) job creation “cont”, that is accounted for by continuing firms only, and (2) job creation due to entry. Such a split is reported in the job reallocation tables.

The job destruction rate for industry i at time t is obtained as the weighted summation of all the negative firm growth rates at time t, in absolute levels:

$$ Job\,Destruction_{i\left( t\right) }=\sum \limits_{f\in S_{i\left( t\right) }^{-}}\omega _{fi\left( t\right) }\ast |g_{f\left( t\right) }| $$
(4)

where \(S_{i\left( t\right) }^{-}\) denotes the subset of firms in industry i witnessing a negative employment growth at time t, and the weights (ω fi(t)) are defined as above.

As for job creation, also job destruction in Eq. 4 can be split in two components: (1) job destruction “cont”, that is accounted for by continuing firms only, and (2) job destruction due to exit. The split is reported in the job reallocation tables.

The net employment growth rate (net flow) is obtained as the difference between job creation and job destruction:

$$ Net\,\,Flow_{i\left( t\right) }=Job\,\,Creation_{i(t) } \,\,-Job\,\,Destruction_{i\left( t\right) } $$
(5)

Instead, by summing job creation and job destruction one obtains the gross job reallocation figure (gross flow):

$$ Gross\,\,Flow_{i\left( t\right) }=Job\,\,Creation_{i\left( t\right) } \,\,+Job\,\,Destruction_{i\left( t\right) } $$
(6)

Finally, the excess flow can be obtained by subtracting the net flow, in absolute value, from the gross flow:

$$ Excess\,\,Flow_{i(t)}=Gross\,\,Flow_{i(t)}-|Net\,\,Flow_{i(t)}| $$

This is a measure of the job flows exceeding the amount that would be needed in order to just accommodate the net employment change.Footnote 14

Table 1 shows the results of the job reallocation analysis on the pooled sample of 14,599 Belgian firms. Contemporaneous creation and destruction of jobs is documented in each year. On average, the job creation rate is 6.3 %, while the destruction one is 4.5 %. Positive net flows are observed in all years but in 2002. The average gross flow is 10.8 %, meaning that, on average, around one job out of ten is either created or destroyed in the manufacturing sector every year. This magnitude of turbulence is not negligible, and is in line with earlier findings for Belgium and other European countries by Gómez-Salvador et al. (2004).Footnote 15 However it is significantly lower than what has been documented for the US (20 %) by Davis and Haltiwanger (1992). This difference is commonly attributed to the higher level of rigidity which characterizes the European labor markets (Gómez-Salvador et al. 2004; OECD 2004). Job creation due to the entry of new firms accounts on average for 22 % of the total creation (1.4 over 6.3 %). Instead, job destruction due to firm exit accounts for only about 9 % of total destruction (0.4 over 4.5 %). The latter low figure is not surprising, considering that the average size of exiting firms is only 4.3 employees, against a mean size of 15.5 for entering firms, and 32 for the whole sample (see Table 13 in the "Appendix").Footnote 16

Table 1 Job reallocation rates based the whole sample (14,599 firms)
Full size table

Table 15 in the "Appendix" reports the average (“standard”) job flows for each NACE 3-digit industry separately.Footnote 17 All industries display both job creation and job destruction, on average over the time span. The magnitude of job flows is generally considerable. For instance, both creation and destruction rates are lower than 1 % in only about 5 % of the yearly observations. Finally, Table 2 shows a set of summary statistics referring to all industry-specific average job flows. The standard deviations indicate the presence of substantial heterogeneity across industries.

Table 2 Summary statistics for industry-specific average job flow rates
Full size table

2.2 Trade exposure

This section analyzes the trade exposure of the Belgian manufacturing sector, at the 3-digit industry level, over the time span 1996–2002. Three different indexes of trade exposure are employed: (1) overall openness, (2) import competition and (3) export intensity. The overall openness index is computed as the sum of imports and exports over the sum of domestic production and imports, for each 3-digit industry. The import competition index is defined as in Davis et al. (1996): imports over the sum of domestic production and imports; analogously, the export intensity index is given by the ratio of exports over the same denominator.

For the computation of the indexes, I employ data sourced from the National Bank of Belgium (NBB). These data are based on the Eurostat Prodcom and Comext databases, for domestic production and trade figures, respectively. In both cases, industry level figures have been computed by the NBB starting from the Eurostat product level data, by mapping the product codes into the 3-digit industry codes of the NACE (Rev. 1.1) classification of economic activities.

The overall openness index averages 1.10, ranging from a minimum level of 0.42 to a maximum of 1.92. The import competition and the export intensity indexes average, instead, 0.53 and 0.57, respectively. Such figures are broadly consistent with the aggregate values of the trade(in goods)-to-GDP ratio provided by the OECD for Belgium: 1.13 in 1996, increasing up to 1.30 in 2002. The same index, in 2002, takes value 0.42 for France, 0.54 for Germany and 0.18 for the US. Overall, the 3-digit figures confirm, at a more disaggregated level, that Belgian firms face very high levels of trade exposure, which are typical of a small open economy.

In order to investigate the cross-industry variation in trade exposure more in depth, Fig. 1 shows the evolution of the overall openness index at the 10th, 25th, 50th 75th and 90th percentiles of the index distribution across the 82 3-digit industries. The graph confirms a trend of increasing openness over the time span. For instance, the median value of the index grows from 1.07 up to 1.15. In addition to that, there is evidence of considerable and increasing heterogeneity in openness across industries. Indeed, the distance between the 10th–25th and 90th percentiles is high at the beginning of the sample and increases over time, as the growth in trade openness for the more open industries is not matched by analogous dynamics for the relatively closed ones.Footnote 18

Fig. 1
figure 1

Overall openness index: percentiles evolution

Full size image

2.3 Real exchange rates

In theory, the real exchange rate is defined as the following ratio: the price of domestic goods over the price of foreign goods sold by the trading partners, both expressed in domestic currency. For the computation of industry-specific RER movements, I follow the same methodology as in Gourinchas (1998, 1999) and Klein et al. (2003a), which is based on the Wholesale Price Index (WPI). However, while they take into account only the major trading partners of each industry, I compute the real exchange rates with respect to a set of 73 foreign countries, which account, on average, for at least 95 % of total trade in each 3-digit industry. This is meant to improve the level of accuracy with respect to the above-mentioned studies. In fact, focusing only on the major trading partners, as in their approach, may lead to disregarding countries that account for up to 50 % of trade in each industry.Footnote 19 Recently, Moser et al. (2010) have employed an alternative measure of RER, which is based on the cross-country comparison of the average hourly wages, all denominated in the same currency. The main advantage of such an approach is that of focusing explicitly on labor costs, which are likely to be most relevant when studying firms’ employment decisions. The shortcoming, though, is that other factors’ costs and differences in productivity dynamics across countries are not taken into account.Footnote 20 A WPI-based measure of RER is supposed to be more comprehensive in this respect, as discussed by Gourinchas (1998).Footnote 21

For computing the industry-specific RER movements I proceed as follows. As a first step, I obtain the series of bilateral real exchange rates for Belgium with respect to each trading partner j, in each year t, as follows:

$$ E_{j(t)}=\frac{WPI_{Belgium(t)}}{NER_{j(t)}\ast WPI_{j(t)}} $$
(7)

where NER j(t) is the bilateral nominal exchange rate, and WPI denotes the Wholesale Price Index. Data are sourced from the International Financial Statistics database (IFS) provided by the IMF.Footnote 22

As an outcome of this first step, for every year I obtain a series of RER percentage variations with respect to each of the trading partners (\(\Updelta E_{j(t)}\)). Building on this, I then compute the change in each industry-specific RER (\(\Updelta E_{i(t)}\)) as a weighted summation of the bilateral RER percentage variations (\(\Updelta E_{j(t)}\)). I employ as weights the industry-specific trade shares of each trading partner. In particular, in order to smooth the series and avoid endogeneity problems in the econometric analysis, a lagged two years moving average of shares is adopted. In formulas:

$$ \Updelta E_{i(t)}=\sum \limits_{j=1}^{73}\omega _{j(t)}^{i}\ast \Updelta E_{j(t)} $$
(8)

where, as in Klein et al. (2003a), the weight of trading partner j, at time t, for industry i is defined as follows:

$$ \omega _{j(t)}^{i}=\left(\frac{1}{2}\right)\sum_{s=1}^{2} \left[\frac{ X_{ij(t-s)}+M_{ij(t-s)}}{\sum_{j=1}^{73}(X_{ij(t-s)}+M_{ij(t-s)})}\right] $$
(9)

where X ij and M ij stand for industry-specific exports and imports to/from country j, respectively.

A positive \(\Updelta E_{i(t)}\) constitutes a real appreciation. Conversely, a negative \(\Updelta E_{i(t)}\) indicates a real depreciation, with domestic goods becoming more competitive as compared to foreign ones. Since the 82 3-digit industries are characterized by different trade patterns, i.e., they trade more/less intensively with different trading partners, I find extensive cross-industry heterogeneity in RER movements. In particular, given the unique set of bilateral RER variations in each year (\(\Updelta E_{j(t)}\) in Eq. 8) some industries experience a real depreciation, while others face a real appreciation at the same time, due to differences in the composition of trade (i.e., the trade shares ω i j(t) in Eq. 8). The heterogeneity in RER movements across industries is illustrated in Fig. 2. In particular, for each year, the 10th, 25th, 50th, 75th and 90th percentiles of the \(\Updelta E_{i(t)}\) distribution are displayed. The diversity in RER dynamics, together with the analyzed heterogeneity in trade exposure, is expected to be relevant in explaining the cross-industry variation of job flows. In the following sections, the relation between real effective exchange rates, trade exposure and job flows dynamics is investigated, both at the descriptive and the econometric level.

Fig. 2
figure 2

Real effective exchange rates: percentiles of growth rates

Full size image

3 Job flows and international factors: descriptive evidence

In this section, raw correlations between job flows and international competition factors are explored at the descriptive level. In particular, in the spirit of previous studies by Levinsohn (1999) and Konings et al. (2003), job flow rates are compared across homogeneous groups of industries, based on average trade exposure and RER movements. The starting point for this analysis is thus the computation of industry-specific average figures for the following indexes: overall openness, import competition, export intensity and the change in the industry-specific RER. Based on the obtained average values, the 82 industries are aggregated into six different groups with respect to each of the four heterogeneity dimensions, each considered separately from the others. Cut-off values are the 10th, 25th, 50th, 75th and 90th percentiles of the cross-industry distributions of the four relevant indexes. The job reallocation analysis is then performed for each of the resulting 24 groups of industries (i.e., 6 groups * 4 heterogeneity dimensions), and average job flows are computed.

Panel (a) of Table 3 shows the results for the 6 groups of industries based on the average levels of overall openness to trade. When looking at the figures for the two most open groups, higher trade exposure seems to be associated with somewhat higher job destruction and lower net flows. Gross flows are also greater than the average for the most open industries.

Table 3 Average job flow figures for homogeneous groups of industries, based on overall trade openness, import competition, export intensity and average RER change
Full size table

In panels (b) and (c) the correlations with respect to import competition and export intensity are separately explored. Groups of industries characterized by higher levels of import competition display above average gross flows, resulting both from higher job creation and higher job destruction. Instead, average job flows do not display any evident correlation with respect to export intensity levels.

The results for industry groups based on RER movements are reported in panel d). The last group contains those industries experiencing the most favorable exchange rate variations, i.e., the largest average depreciations over the time span. The corresponding job flow figures seem to suggest that real depreciations are associated with higher than average net employment growth, resulting both from higher job creation and from lower job destruction.

Overall, these results need to be evaluated carefully. The one just discussed is indeed only a preliminary and descriptive analysis. Differences in job flows across groups are generally small, and could be determined by different factors than the ones explored. In the next section, the relation between job flows and international competition factors is further investigated through econometric analysis.

4 Econometric analysis

4.1 Theoretical background

Before presenting the econometric analysis, this section reviews the main elements and the predictions of the model by Klein et al. (2003a), which constitutes the theoretical background of the analysis.

In a context of openness to trade, international factors are modeled as affecting the demand equation for each firm’s output as follows:

$$ Q_{p}=A_{p}Y^{\beta }\Uppi _{j=1}^{k}\left[ E_{j}^{-\mu \Upomega _{i}}Y_{j}^{\ast \beta \Upomega _{i}}\right] ^{\omega _{j}^{i}} $$
(10)

where Q p is the demand for the output of firm p in industry i, and A p is the idiosyncratic demand shock faced by the same firm. Since output can be sold both domestically and abroad, the demand equation includes both Y, a measure of domestic income, and a multiplicative expression capturing foreign demand. Each term of the latter expression refers to a single country j and is influenced by several factors. In particular, the demand contribution of each foreign country is directly proportional to its income \(Y_{j}^{\ast }, \) and inversely related to the bilateral RER E j . The latter is in fact defined as the ratio of the price of domestic goods over the price of foreign goods (in domestic currency), as explained in Sect. 2.3. Thus, the higher E j , the lower the competitiveness of firm p output. The impact of \(Y_{j}^{\ast }\) and E j is directly proportional to \(\Upomega _{i}, \) an indicator for the level of trade openness of the industry, which does not vary across firms.Footnote 23 Finally, the contribution of each trading partner is weighted by its share in the total trade of the industry, ω i j , which is also common to all firms within the same industry.Footnote 24

Labor demand growth at the firm level has the following expression:

$$ \widehat{L_{p}}=-\left( 1-\alpha \right) \widehat{W_{p}}+\left( 1-\alpha \right) \widehat{G_{p}}+\widehat{A_{p}}+\beta \widehat{Y}-\mu \Upomega _{i} \widehat{E_{i}}+\beta \Upomega _{i}\widehat{Y_{i}^{\ast }} $$
(11)

where the notation \(\widehat{Z}\) stands for \(d\ln (Z), \) for any variable Z . The variable W p is the wage and the variable G p is the unit-cost of the non-labor input. The term \(\widehat{E_{i}}\) is the industry-specific, trade-weighted RER variation. Analogously, \(\widehat{Y_{i}^{\ast }}\) is the trade-weighted growth of foreign partners’ output.Footnote 25 Taking into account the general equilibrium effects of RER movements on wages (disregarding for simplicity variations in G, Y and \(Y^{\ast }\)), the final expression for firm level labor demand growth is as follows:

$$ \widehat{L_{p}}=\left( \widehat{A_{p}}-k\widehat{A_{i}}\right) -k\varepsilon \gamma \widehat{\Upgamma }-\left( 1+k\right) \mu \Upomega _{i}\widehat{E_{i}} $$
(12)

where A i is an industry-specific idiosyncratic shock, resulting from the summation of the firm-specific ones. The variable k is a parameter ranging between zero and one, γ is a measure of labor supply elasticity (γ > 0), \(\varepsilon \) is the cross-elasticity of labor supply between industry i and the rest of the economy (\(\varepsilon \geq 0\)), and \(\Upgamma \) is the prevailing wage in the rest of the economy.

Equation 12 implies that, ceteris paribus, a real appreciation (\(\widehat{E_{i}}>0\)) has a negative effect on labor demand at the firm level, and this effect is magnified by increasing levels of trade openness (\(\Upomega _{i}\)). Since industry-specific job flows are computed as weighted summations of the firm-specific ones, the final prediction of the model is that a real appreciation, ceteris paribus, induces lower job creation and higher job destruction at the industry level. These effects are stronger the higher the level of trade exposure in the industry. The intuition for this is straightforward: higher exposure to trade implies enhanced sensitivity to international competitive factors.Footnote 26

4.2 Empirical strategy

Drawing on the presented theoretical framework by Klein et al. (2003a), and building upon previous work by Moser et al. (2010), the baseline estimation equation for the econometric analysis is specified as follows:

$$ \begin{aligned} Job\_Flow_{i(t)} &=\alpha _{0}+\alpha _{1} Job\_Creation_{i(t-1)}+\alpha _{2}Job\_Destruction_{i(t-1)}\\ &\quad +\alpha _{3}\left( Openness\_Index_{i(t)}\ast \Updelta E_{i(t)}\right) +\alpha _{4}Z_{i(t)}+\alpha _{5} X_{(t)}+\alpha _{i}+\varepsilon _{i(t)} \end{aligned} $$
(13)

where i refers to 3-digit industries and t indexes years. The term \(\mathbf{Z}_{i(t)}\) is a vector of industry-specific control variables, while \(\mathbf{X}_{(t)}\) is a vector of macro-controls, which do not vary across industries. α i is a vector of 3-digit industry fixed effects, and \(\varepsilon _{i(t)}\) is the estimation error.

The dependent variable Job_Flow i(t) corresponds to, alternatively, one of the following four job flows: Net_Flow i(t), Job_Creation i(t), Job_Destruction i(t), and Gross_Flow i(t). The lagged values of job creation and job destruction are always included as regressors, in order to account for possible dynamic adjustments.

Openness_Index \(_{i(t)}\ast \Updelta \) E i(t) stands for the interaction between the change in the industry-specific RER (\(\Updelta \) E i(t)) and the overall trade openness index at the industry level (Openness_Index i(t)). This interaction variable is crucial with respect to the research question, and its inclusion follows directly from the theoretical framework presented in Sect. 4.1. The term \(\Updelta\, E_{i(t)}\) is computed as in Eq. 8, i.e., as a trade-weighted average of bilateral RER changes with respect to a set of 73 trading partners. Openness_Index i(t) is instead obtained as explained in Sect. 2.2, as the sum of imports and exports over the sum of domestic production and imports, for each 3-digit industry. In particular, a lagged two-year moving average of this ratio is employed, in order to avoid endogeneity problems in the estimation. In addition, I will also present regressions in which \(\Updelta\, E_{i(t)}\) is interacted with the alternative measures of trade exposure: the import competition index and the export intensity index, both computed as explained in Sect. 2.2. and both employed as lagged two-year moving averages.

The set of industry controls, \(\mathbf{Z}_{i(t)}, \) contains the following variables:

$$ \left[ \begin{array}{c} Openness\_Index_{i(t)}\ast \Updelta Y_{i(t)}^{\ast },RCA_{i(t)},Herfindahl\_Index_{i(t-1)}, \\ Relative\_price\_change_{i(t-1)}, Mean\_tangible\_assets_{i(t-1)} \end{array} \right] $$

\(Openness\_Index_{i(t)}\ast \Updelta\, Y_{i(t)}^{\ast }\) is the interaction between the openness index, constructed as above, and \(\Updelta \,Y_{i(t)}^{\ast }, \) which represents the industry-specific average percentage change in the real GDP of the trading partners.Footnote 27 The inclusion of this interaction variable is motivated by the theoretical model (see Eq. 11), and is meant to capture a second determinant of foreign demand, besides real exchange rates.Footnote 28 RCA i(t) represents an industry-specific measure of revealed comparative advantage. Following the standard definition by Balassa (1965), it is computed as the following ratio:

$$ RCA_{i(t)}=\frac{\frac{Export_{i(t)}}{Import_{i(t)}}}{\frac{\sum_{i=1}^{n} Export_{i(t)}}{ \sum_{i=1}^{n}Import_{i(t)}}} $$
(14)

where n stands for the total number of 3-digit industries (indexed by the subscript i).

The inclusion of this control follows from recent theoretical results by Bernard et al. (2007b), who have shown that job flows in an industry might be systematically related to the comparative advantage enjoyed by the same industry. In particular, Bernard et al. (2007b) have analyzed, at the theoretical level, the effects of increasing openness to trade on multiple domestic industries, characterized by heterogeneous firms and comparative advantages. Their model predicts that, in the adjustment to increasing trade exposure, comparative disadvantage industries display net job destruction, while comparative advantage ones enjoy net employment growth. However, higher trade exposure triggers simultaneous job creation and destruction in all industries, and this effect tends to be magnified in industries characterized by a relative comparative advantage. In fact, the market selection induced by trade is stronger for comparative advantage industries than for comparative disadvantage ones. The inclusion of RCA i(t) in the empirical model is meant to control for the latter effects.

Several studies have linked job flows with product market imperfections, market concentration and the pricing power of firms within an industry (Geroski et al. 1995; Bertrand and Kramarz 2002; Blanchard and Giavazzi 2003; Peoples 1998; Boeri et al. 2000). In particular, these studies have shown that high market concentration, associated with high markups and barriers to entry, may have a negative impact on employment growth. Following this literature, \(\mathbf{Z}_{i(t)}\) includes three additional industry level controls for the market structure: (1) Herfindahl_Index i(t-1), (2) Relative_price_change i(t-1), and (3) Mean_tangible_assets i(t-1). They are obtained as follows. (1) Herfindahl_Index i(t-1) is computed using firm level turnover data from the Amadeus sample. In particular, first the market share of each firm is computed, for each year, with respect to the total sample figure of turnover within the corresponding 3-digit industry (i.e., the sum of firm level turnover for each sample firm within the industry). Then, the Herfindahl Index is obtained as the sum of each firm’s squared market share, for each industry i and year t. (2) Relative_price_change i(t-1) is the relative change in the Producer Price Index (PPI) of industry i with respect to the average figure for the whole manufacturing sector in the same year. For instance, if the PPI in year t increases by 3 % in industry i, while the average increase in the manufacturing sector is 1 %, the relative price change for industry i will be 2 %. The construction of this variable is in line with Klein et al. (2003a), and its computation is based on Eurostat data on PPIs at the 3-digit industry level. (3) Mean_tangible_assets i(t − 1) is the average firm level value of tangible fixed assets within each given industry (and year). This variable is meant to control for barriers to entry (Geroski et al. 1995). It is computed based on the Amadeus firm level data, just as the Herfindahl Index, and all the job flow figures employed in the analysis.

The vector of macro-controls, \(\mathbf{X}_{(t)}, \) contains the following variables:

$$ \left[ \begin{array}{c} Real\_interest\_rate_{(t)},\,\, GDP\_growth\_rate_{(t)},\,\, \\ Real\_wage\_growth\_rate_{(t)},\,\, Tot\_employment\_growth \_rate_{(t)} \end{array} \right] $$

As already anticipated, all these regressors do not vary across industries but only through time, and their inclusion follows earlier work by Klein et al. (2003a) and Moser et al. (2010). These variables are meant to control for aggregate dynamics which could be correlated with real exchange rate movements, thus potentially leading to spurious findings of significant effects of RER changes on job flows. Real_interest_rate (t) is computed using IFS data for Belgium, as the prime lending rate minus the inflation rate. GDP_growth_rate (t) is the real GDP growth rate of Belgium, also obtained from IFS. Real_wage_growth_rate (t) is the average growth rate of wages in the manufacturing sector of Belgium, deflated using the Consumer Price Index (CPI). In order to construct this variable, the nominal average growth rate of wages is computed using Amadeus firm level data on total staff costs. In particular, the latter costs are first divided for the number of employees (full time equivalent), as to retrieve the average wage at the firm level in each year. Then, the yearly average is taken across all firms in the sample, and the nominal growth rate is computed over the years. Finally, the nominal figures are deflated using CPI data for Belgium, retrieved from the IFS. Tot_employment_growth _rate (t) stands for the net employment growth for the whole manufacturing sector. As explained in Sect. 2.1, this variable is obtained as the difference between the job creation rate and the job destruction rate, computed over the full sample of 14,599 Belgian companies, using Amadeus data. The resulting figures are presented in the third row of Table 1.

Concerning the econometric methodology, it is important to notice that the specification outlined in Eq. 13 is a dynamic panel model. In fact, depending on the employed dependent variable (Net_Flow i(t), Job_Creation i(t), Job_Destruction i(t), Gross_Flow i(t)), either the dependent variable itself, or its components of job creation and destruction, appear with a 1-period lag among the regressors. As shown by Nickell (1981), if the time dimension of the panel is small, as in my case, a fixed-effects estimator is inconsistent for dynamic models, due to the correlation of the lagged dependent variable with the group-mean of the error term: the so-called “Nickell bias”. In particular, as discussed by Moser et al. (2010), when only the first lag of the dependent variable is included as a covariate, a fixed-effects estimator will underestimate the coefficient of the lagged dependent variable. On the other hand, an OLS estimator will have an opposite bias, due to the endogeneity of the lagged dependent variable (Trognon 1978). In order to obtain consistent estimates, I employ the Blundell and Bond (1998) one-step system-GMM estimator, where Job_Creation i(t − 1) and Job_Destruction i(t − 1) are always instrumented using their second and higher-order lags.Footnote 29 Since such instruments are only valid if errors are not autocorrelated, the appropriate autocorrelation tests are reported. As these tests are referred to first-differenced errors, first-order autocorrelation is to be expected, while the absence of second-order autocorrelation is needed. In addition, for all the system-GMM regressions I also report the results of the Hansen test for overidentifying restrictions.

Given all the above considerations, in line with Moser et al. (2010), all the regressions are estimated in three different ways: (1) OLS with no industry fixed effects, (2) OLS augmented by 3-digit industry fixed effects, (3) Blundell and Bond (1998) system-GMM estimation. This allows to compare different estimates, and check whether the Blundell and Bond (1998) estimator is correcting biases as expected from theory. In all the estimations, standard errors are corrected for clustering within 3-digit industries. Results are discussed in the next section.

4.3 Estimation results

Table 4 reports the econometric results from the estimation of Eq. 13, using Net_Flow i(t) as the dependent variable. In particular, column 1 displays the outcome of the simple OLS regression, with no industry fixed effects. Column 2 reports the results from the OLS regression augmented by 3-digit industry fixed effects. Column 3 reports the results from the Blundell and Bond (1998) system-GMM estimation. Finally, column 4 reports system-GMM results obtained using the “cont” measures of Net_Flow i(t) instead of the “standard” ones. Such “cont” measures are computed as explained in Sect. 2.1, by excluding the contribution of entering and exiting firms to job creation and job destruction in each year. Column 4 thus provides a robustness check against possible biases related to firm entry and exit in the Amadeus sample. Tables 5, 6, and 7 are structured in the same way as Table 4, considering as a dependent variable Job_Creation i(t), Job_Destruction i(t) and Gross_Flow i(t), respectively. In all cases, the industry-specific RER variation (\(\Updelta \,E_{i(t)}\)) is interacted with the overall openness index (Openness_Index i(t)).Footnote 30 The discussion will mainly focus on the system-GMM results, where the dynamic nature of the specifications is properly addressed.

Table 4 Regression results for net flows, with RER changes interacted with the overall openness index
Full size table
Table 5 Regression results for job creation, with RER changes interacted with the overall openness index
Full size table
Table 6 Regression results for job destruction, with RER changes interacted with the overall openness index
Full size table
Table 7 Regression results for gross flows, with RER changes interacted with the overall openness index
Full size table

First, the results indicate that a real appreciation has a negative impact on net employment growth, and that this impact is magnified by increasing levels of overall openness to trade. Second, concerning the adjustment margin, the net employment effect is strongly driven by an increase in job destruction, while job creation is not significantly affected. The magnitude of the effects is relatively small, in line with earlier evidence by Gourinchas (1999), Klein et al. (2003a) and Moser et al. (2010). In particular, Table 8 shows the impact of a one standard deviation real appreciation (1.6 %) on net employment growth and job destruction, for industries characterized by different levels of openness to trade. The computation of the effects is based on the estimated coefficients for Openness_Index \(_{i(t)}\ast \Updelta \, E_{i(t)}, \) as reported in column 3 of Tables 4 and 6. When focusing on the median industry in terms of trade exposure, a one standard deviation real appreciation determines a decrease in net employment growth by 1 %, driven by a 0.7 % increase in job destruction (about 3,900 jobs lost, when evaluated on the whole manufacturing workforce in Table 11 in the "Appendix", just as an example). To have an idea of the growth in the effects as the level of trade exposure increases, we have that the impact for the industry at the 90th percentile of openness is more than twice the one for the relatively closed industry at the 10th percentile. The results do not qualitatively change when considering “cont” job flows instead of the “standard” ones: if anything, the coefficient for the impact on net flows is more precisely identified. However, the magnitude of the effects is slightly lower than for standard flows. This is consistent with the fact that a real appreciation may force some firms to exit, as found by Baldwin and Yan (2011) and Moser et al. (2010). Therefore, when focusing only on job destruction by continuing firms, I am capturing the lower bound of the total effect of a RER appreciation on job destruction and net employment growth. As an additional robustness check, in the spirit of Gourinchas (1999), I have re-estimated all the regressions on a subsample of relatively more open NACE 2-digit industries. This has been done by dropping the 21 3-digit industries belonging to the 25 % most closed 2-digit industries.Footnote 31 The main results, reported in Table 9, are qualitatively unaffected.

Table 8 Impact of a one standard deviation real appreciation (1.6 %) for industries at different levels of openness to trade
Full size table
Table 9 Regression results dropping the 21 NACE 3-digit industries most closed to trade, for both “standard” and “cont” flows, with RER changes interacted with the overall openness index
Full size table

As expected from theory, when looking at the estimated coefficients for Job_Creation i(t−1) (in the first three columns of Table 5) and Job_Destruction i(t−1) (in the first three columns of Table 6), we can see that the system-GMM estimates are in the middle between the OLS and the FE ones. In particular, the OLS coefficients are overestimated while the FE ones are underestimated. This is reassuring with respect to the appropriateness of the system-GMM estimation. Moreover, for each of the system-GMM regressions, the AR(2) test does not reject the null hypothesis of no second-order autocorrelation of the residuals in first differences, and the Hansen test is also supporting the validity of the instruments. Thus one can be confident that the dynamic panel estimator is well specified.

Overall, when focusing on the system-GMM results in column 3 across Tables 4, 5, 6 and 7, job creation and net flows do not appear to be related to lagged levels of job creation and destruction. Instead, job destruction and gross flows are positively related to past levels of job destruction, and thus may react weakly to past RER changes. Concerning the other industry level controls, higher levels of mean tangible assets seem to be related to lower job creation and net employment growth. Instead, the coefficients for the foreign trade-weighted GDP growth, the real comparative advantage, the Herfindahl Index and the relative price change are not found to be significant. As to macro-controls, job creation and gross flows are positively related to the real interest rate and GDP growth. Gross flows are instead negatively related with increases in real wages. Finally, increases in total manufacturing employment are positively related with job creation and net flows, and negatively related with job destruction at the 3-digit industry level.

The main results are not qualitatively changed when interacting the RER movements (\(\Updelta\,E_{i(t)}\)) with the alternative measures of trade exposure: import competition index and export intensity index (see Table 16 in the "Appendix"). As a further robustness check, all the estimations have been repeated by dropping the macro-controls and including year dummies. The results are unchanged.Footnote 32 Finally, results are not likely to suffer from endogeneity problems concerning the main variable of interest: \(Openness\_Index_{i(t)}\ast \Updelta \,E_{i(t)}. \) Indeed, as also discussed in previous studies, all the employed trade data are lagged, and the bilateral real exchange rates are determined at the country (and euro area) level, while job flows are studied at a highly disaggregated (and country-specific) 3-digit industry level. The results are discussed in the next section.

4.4 Discussion

The main finding of the econometric analysis is the following: an RER appreciation has a negative impact on net employment growth at the industry level, and this impact is magnified by increasing levels of trade exposure. As to the adjustment margin, the decrease in net employment growth is driven by an increase in job destruction, while job creation is not significantly affected by RER movements. Such evidence of a job destruction-driven adjustment is in line with earlier findings by Klein et al. (2003a) for the US. In contrast, my results depart from previous evidence for other EU countries. In fact, Gourinchas (1999) and Moser et al. (2010) have found that job creation plays a major role in the reaction to RER changes, in France and Germany, respectively. Footnote 33

Moser et al. (2010) attributed the different results for the US and Germany to differences in the labor market institutions, and a similar argument could be put forward for France. As a matter of fact, France and Germany display much higher levels of employment protection, which makes job destruction difficult and costly for firms. For instance, in 2002, the OECD Overall Index for the Strictness of Employment Protection takes value 3.05 for France and 2.09 for Germany, against a level of 0.21 for the US. In ordinal terms, the US rank first for low-strictness among the 28 surveyed industrialized countries, while France and Germany rank 24th and 17th, respectively. In the interpretation given by Moser et al. (2010), since firing is relatively more costly in France and Germany than in the US, French and German firms do not immediately react to a negative RER shock by destroying jobs, as their US counterparts do; rather, they adjust by creating less new jobs. Now, my results indicate that the adjustment to RER shocks in Belgium works through the job destruction margin, in line with the US and differently from France and Germany. Such differences between Belgium and the other two European countries are not likely to be driven by differences in the labor market institutions. Indeed, Belgium is also characterized by a relatively high level of labor market rigidities. For instance, the above-cited OECD Index in 2002 is equal to 2.18 for Belgium, slightly above the German figure. Hence, the explanation for the uncovered differences has to be found elsewhere.

In my interpretation, my findings may be explained by the fact that Belgium, differently from France and Germany, is a small open economy, where firms face much higher levels of trade exposure. Indeed, as discussed in Sect. 2.2, the trade(in goods)-to-GDP ratio is equal to 1.30 for Belgium in 2002, against a value of 0.42 for France and 0.54 for Germany (OECD data). In particular, focusing on the 82 3-digit manufacturing industries in my sample, the lowest yearly figures for both import competition and export intensity are around 0.21, i.e., well above the minimum thresholds adopted by Gourinchas (1999) for identifying tradable industries in France (0.13 and 0.125 for export intensity and import competition, respectively). Overall, Belgian firms operate in a much smaller domestic market than French or German firms, and are more exposed to international trade. As a result, they are forced to be more reactive to shocks hitting their international competitiveness. This may explain why the adjustment to real exchange rate appreciations occurs through the job destruction margin, despite the high level of employment protection. In fact, even though destroying jobs is costly for Belgian firms, the cost of failing to adjust timely to RER shocks is likely to be even higher, in terms of reduced profits, due to the very high level of trade exposure. Overall, these results highlight the importance of studying the adjustment to RER changes in different country-contexts, where job flow responses may be very differentiated. The main contribution of this paper, building on earlier studies, is that of shedding light on the differences between a small open economy and relatively larger EU countries, which are characterized by similar levels of labor market rigidity.

Some other results are worth discussing in terms of industry level controls, focusing on the system-GMM results. In particular, consistent with expectations from previous studies, I find some evidence of lower job creation and net employment growth in industries characterized by high barriers to entry, as measured by mean tangible assets. On the contrary, there is no evidence of the expected link between comparative advantage and job flows. This may be due to the employed measure of revealed comparative advantage, as introduced in Sect. 4.2. In fact, as domestic firms increase their imports of intermediates from abroad, in order to exploit cost and/or quality advantages in foreign countries, competitiveness gains within an industry may go hand in hand with a decrease in the employed RCA measure (Amiti and Konings 2007; Kasahara and Rodrigue 2008; Halpern et al. 2011). This calls for the development of refined indicators of revealed comparative advantage, a task that is beyond the scope of this paper.

5 Conclusion

This paper has investigated the impact of RER movements on job flows, focusing on the manufacturing sector of a small open economy characterized by significant labor market rigidities: Belgium. I have found that real appreciations have a negative impact on net employment growth through an increase in job destruction, while job creation is not significantly affected. While being in line with previous empirical evidence on the US, my results depart from earlier findings for larger EU countries, in particular France and Germany, where the adjustment to RER shocks has been found to occur mainly through the job creation margin. These differences may be attributed to the fact that Belgium is a small open economy, where firms face much higher levels of trade exposure.

Being able to identify the adjustment margin behind the net flow effects is of fundamental importance for welfare analysis. Indeed, numerically equivalent increases in job destruction and reductions in job creation may have very different welfare implications. For instance, while an increase in job destruction is likely to involve the displacement of high-wage and older workers, a reduction in job creation is more likely to slow down the accumulation of human capital, by increasing the duration of unemployment (Davis et al. 1996; Klein et al. 2003a, b). A thorough welfare analysis is beyond the scope of this paper, and constitutes an interesting avenue for further research.