![Sample our Economics, Finance,Business & Industry journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5931/TOC-Subject-Sample-2021-Economics-Finance-Business-Industry.jpg"})
ABSTRACT
By exploiting cross-country micro-aggregated CompNet data, this study investigates the main implications of firm heterogeneity for international trade of EU countries, distinguishing between old and new Member States. On the one hand, exporting firms are larger, more productive and pay higher wages than non-exporting firms, especially in new EU economies. Only the former firms are indeed able to bear export costs, which are higher in new EU countries and are related to various factors, such as the quality of the legal system, the restrictiveness of labour-market regulation and the degree of access to finance. Hence, only few enterprises actually export, and the intensity of aggregate export concentration within few firms varies across countries and sectors. On the other hand, opening to trade boosts individual firms’ productivity, via a number of channels (including GVC integration, which is particularly important for new EU countries), and also enhances allocative efficiency across firms, in turn increasing aggregate productivity growth. One of the main standard determinants of export growth, namely changes in the real effective exchange rate, impacts aggregate performance differently across countries and sectors, depending on sectoral composition and on firm characteristics, in both old and new Member States.
Acknowledgements
We thank three anonymous reviewers, Matteo Bugamelli, Alberto Felettigh, Pavlos Karadeloglou, Roberta Serafini, Luigi Federico Signorini, Joao Sousa, Helena Schweiger and all participants of the 2018 CEBRA conference in Goethe University, Frankfurt-Am-Main, the 2019 productivity workshop at the European Central Bank, Frankfurt-Am-Main and the 2019 First conference of CompNet users in France Strategie, Paris, for useful comments on previous versions of this article. Moreover, we are grateful to Antoine Berthou for sharing his data with us. Any errors however are responsibility of the authors. The views represented herein are those of the authors, and not of the institutions represented.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 The old Member States are defined from hereafter as countries that joined the EU by 1995 at the latest, namely Western Europe; the rest of EU countries included in the analysis will be labelled ‘new EU countries’.
2 The ‘Productivity’ module is available for the period 2001-2013, although it is unbalanced given delayed entry of some countries and availability of information until 2012 for countries obtained from the 4th vintage of CompNet data. This explains the different time coverage of different tables and charts, according to the indicators considered.
3 Trade-related information, which is used in this article, is only available for 23 2-digit industries of the manufacturing sector. Note, however, that the rest of the CompNet database covers 54 industries of the business economy, including manufacturing, services and construction.
4 CompNet considers five size-classes, following Eurostat: firms with 1–9 employees; 10–19; 20–49; 50–249; and 250 or more.
5 Although the methodology used to estimate firm-level TFP in the CompNet database is standard in the micro-applied literature, it is worthwhile recalling some important drawbacks. First of all, CompNet draws from administrative firm registries and therefore has no access to firm-level output prices. Hence, to deflate firm-level output the sector-level deflator is used instead leading to the estimation of the co-called ‘revenue TFP’. However, this choice has several implications. First, if firm-level price variation is correlated with input choice, this will result in biased input coefficients (Katayama, Shihua, and Tybout Citation2009; Van Beveren Citation2012). Moreover, in the presence of imperfect competition in input markets, it is likely that input prices are also firm-specific, further biasing the estimates. Second, the estimation procedure assumes that all firms operating in the same sector have the same production function. Although we consider sectors defined at the 2-digit level, firms operating in the same sector might be specialised in different tasks, be multi- or uni-product firms, belong to global supply chains or only produce goods for the domestic market, etc. All these differences would likely be reflected in different production functions, which our approach does not capture.
6 What we loosely refer to as ‘wages’ are nominal labour costs per employee, which actually include both nominal wages and employers' social security contributions.
7 The reason we include a contemporaneous, and not lagged, ‘new exporter’ dummy is the following. This takes the value one at time t when a firm did not export at t−1 and starts exporting at t and continues exporting at t+1. The objective of the exercise is to compare the productivity of new exporters with non-exporting firms at the time of entry, that is, at t (and not one year later). Correlating entry of new exporters at t−1 with productivity at t would contaminate the pure ‘self-selection effect’ with some ‘learning-by-doing effect’. Yet, since CompNet does not have firm-price data (as stated in footnote 5), the premium may also be reflecting a rise in mark-ups linked to entry into international markets, a caveat which also holds for Equation (6).
8 Note, however, that the premia of all countries would be significant at the 1 per cent level if no other firm characteristic were controlled for. For instance, Estonia’s labour productivity premium of new exporters turns insignificant when size is controlled for, because in this country only very large firms are productive enough to become new exporters. Moreover, the existence of a premium across all these dimensions is found also when comparing the whole set of exporting (i.e. also incumbent exporters) with non-exporting firms (results available upon request).
9 As regards labour productivity specifically, figures in Table are significantly more contained than those reported in Figure 1. Amongst various reasons, in Table it is noteworthy that productivity premia are estimated conditionally to a set of control variables, namely the size of firms and various fixed effects, whereas in Figure 1 computations provided are unconditional and therefore do not take into account observed and unobserved heterogeneity. This reduction in exporter premia after controlling for heterogeneity is standard in the literature (see, for example, Bernard et al. Citation2007; ISGEP Citation2008). Note also that Table refers to the productivity premia of new exporters relative to non-exporters. Productivity differences between continuing exporters and non-exporters are larger.
10 Clearly, it would be interesting to also test for the significance of the legal system and other institutional features in the destination country, in addition to the source country. Unfortunately, CompNet data do not include bilateral trade information and, therefore, the destination country cannot be identified.
11 We also tried to include country dummies in order to account for country-specific firm characteristics that affect the productivity distribution; these fixed effects, however, wipe out the statistical significance of the determinants of trade costs, which vary little over time, and are therefore not included in our baseline specification. Indicators of labour and product market regulation have also been included in the baseline specification, but are found to be statistically insignificant.
12 Amongst the controls, VA is statistically significant and positive in some regressions, implying that exporter premia are higher the larger the size of the manufacturing sector it operates in. Country-specific business cycle conditions, on the other hand, do not appear to be significant.
13 Finland notably has few exporting small and medium-sized enterprises (e.g. European Commission Citation2017), as will be seen in the next chart, and in the period under study was highly concentrated in the electronic and paper industries.
14 The elasticity of exports of euro-area countries to HCIs in macroeconomic analyses is generally smaller than one (Goldstein and Khan Citation1985; Christodoulopoulou and Tkačevs Citation2014; Giordano and Zollino Citation2016, Citation2017; Bugamelli et al. Citation2018).
15 In principle, trade elasticities computed using aggregated data should reflect the weighted average of the sectorial elasticities, as long as residuals are well-behaved. However, Imbs and Mejéan (Citation2015) among others show that this is not the case and results in aggregation bias. More concretely, the estimation of elasticities based on aggregate data presumes: (i) the same reaction of trade volume across goods, regardless of the degree of substitution between domestic and foreign goods (in reality, a homogenous good will react more strongly to exchange-rate movements); and (ii) that each good accounts for the same import share in a country’s basket. As in the real world this is not the case because of heterogeneity, residuals will be correlated with the regressors, thus biasing the coefficients, and therefore the estimated elasticities, downwards.
16 This is similar to a Bartik instrument, constructed on WIOD data, which exploits information about the initial export structure of each country-sector and foreign partners’ total imports in these sectors, following a similar methodology developed in Berman, Berthou, and Héricourt (Citation2015). In particular, foreign demand is computed as the weighted average absorption by a given country i’s export partners using i’s initial export shares as weights: , where d is the destination country, Y is GDP and M is imports.
17 The positive sign and/or statistical insignificance that comes up in some sectors could either be due to the lower number of observations (e.g. beverages) and/or to the fact that some sectors may be on average more dependent on imported intermediate goods, such as wearing apparel and fabricated metals, which can flip the expected negative sign of the REER elasticity.
18 Interestingly, this effect is contemporaneous. A similar analysis on the extensive margin, which we discuss further on, finds a significant negative correlation between this other margin and a REER depreciation, albeit with a one-year lag. Indeed, it is presumably easier and quicker for an already exporting firm to increase its foreign sales than for a non-exporting firm to start exporting, in the face of a REER depreciation.
19 Qualitatively similar results are obtained if the average, as opposed to the median, TFP is employed.
20 We had to drop the average REER depreciation from the specifications underlying columns 3 and 4 since the inclusion of up to triple interaction terms, in addition to fixed effects, led to multicollinearity issues.