European export performance

Abstract

Using an econometric shift-share decomposition, we explain the redistribution of world market shares at the level of the product variety and by technological content. We decompose changes in market shares into structural effects (geographical and sectoral) and a pure performance effect. We regard the EU-27 as an integrated economy, excluding intra-EU trade. Revisiting the competitiveness issue in such a perspective sheds new light on the impact of emerging countries on the reshaping of world trade. Since 1995 the EU-27 withstood the competition from emerging countries better than the United States and Japan. The EU market shares for high-technology products, as well as in the upper price range of the market, proved comparatively resilient, though less so since the crisis.

Appendix

1.1 Data description

The trade data used in this paper are from the BACI database, a database for the analysis of international trade at the product-level developed by Gaulier and Zignago (2010). BACI draws on the UN COMTRADE information, in which imports are reported CIF (cost, insurance and freight) and the exports FOB (free on board). BACI provides reconciled FOB data on trade flows: for a given product k and a given year t, exports from country i to importer j are equal to j imports from i. This reconciliation of mirror flows is performed for both values and quantities, and relies on estimated indicators of the reliability of import and export country reports. The quantity units are converted into tons, making possible the computation of homogeneous unit values. ²⁸

BACI covers trade between more than 200 countries, in the roughly 5,000 products of the 6-digit Harmonised System (HS6) classification. However, this study excludes intra-EU-27 trade flows. This choice must be borne in mind when it comes to market shares and changes therein. We also exclude mineral, specific and non-classified products.²⁹ Trade flows below USD 10,000 and involving non-independent territories and micro-states are also excluded in Sect. 2.1 and in Sect. 3 For the shift-share analysis in Sect. 3 we employ HS2 data obtained by aggregation of HS6 data. The motivation behind is to keep a larger share of trade flows in the intensive margin, the only component of the export growth discussed in that section.

Concerning the high-tech products, we use the classification in broad sectors proposed by Lall (2000), detailed in Table 9.

The availability of traded unit values at a very disaggregated level (country-partner-product-year) in the BACI database makes it possible to compute international trade price indices. Similar to Gaulier et al. (2008), we compute price indices as chained Törnqvist indices of unit values, but unlike them we compute an index for each pair of trading countries (exporter-importer) and HS2 heading. Data in 2,000 is taken as reference. We use these indices to deflate trade values (expressed in current USD in BACI) to obtain trade volumes expressed in terms of 2,000 prices. Since this exercise allows us to disentangle price effects, we refer to obtained data as volumes.

The world distribution of unit values for each HS6 heading allows us to classify each product-bilateral flow into three price segments, and to examine competition among the main world exporters within each of these segments. Trade flows are ordered according their unit values and classified as follows: flows with the lowest unit value form the bottom-range, the ones with intermediate unit values—the mid-market, and the ones with the highest unit value—the mid-range. We employ the technique developed by Fontagné et al. (2008) to construct the three market segments. There is also a small “non-classified” range of trade flows for which data on trade quantities is not available and unit values cannot be computed, but they represent <10 % of world trade.

Tables of this paper display results for countries accounting for more than 1 % of world exports from 1995 to 2010. Results for countries accounting for <1 % of world exports from 1995 to 2010 are aggregated within three groups: the Middle East and North Africa (MENA), Sub-Saharan Africa (SSA), and Rest of the World (RoW). Results for all other countries in the world are available in our online appendix. ³⁰

1.2 Additional results

See Tables 10, 11 and Figs. 4, 5.

Table 10 Extensive and intensive margins in 1995–2010 for world exports by country, %

Table 11 Shift-share decomposition of changes in world market shares, all products, 1995–2010: in volume terms

Fig. 4

Export market share growth rates used by the traditional and econometric shift-share decompositions, all countries in the sample

Fig. 5

Composition effects: econometric vs. traditional shift-share analysis, all countries in the sample

1.3 Computation of standard errors

One of the advantages of the econometric shift-share approach, with respect to the traditional approach, is the estimation of standard errors for each effect. Effects Perf _i , Geo _i , and Sect _i of the decomposition of market share growths, given by Eq. (13), are computed from exporter, importer and sector fixed effects α ^t _i , β ^t _j , γ ^t _k estimated with Eq. (7). Summarizing the computations in Sect. 3.2, we have:

$$ {Perf}_{i} = \exp \left( \sum\limits_t \hat{\alpha }_{i}^{t} + \sum\limits_{j,t} {w_j^t \hat{\beta }_j^t} + \sum\limits_{k,t} {w_k^t \hat{\gamma }_k^t} - \sum\limits_{i,t} {w_i^t \left(\!\hat{\alpha }_i^t \!+\! \sum\limits_j \!{w_j^t \hat{\beta }_j^t} \!+\! \sum\limits_k {w_{k}^{t} \!\hat{\gamma }_{k}^{t}}\!\right) } \right) - 1 $$

(17)

$$ Geo_i = \exp \left( \sum\limits_{j,t} {\frac{w_{ij}^t}{w_i^t }\; \left(\hat{\beta }_j^t -\sum_j{w_j^t } \hat{\beta }_j^t \right) }\right) - 1 $$

(18)

$$ Sect_i = \exp \left( \sum\limits_{k,t} {\frac{w_{ik}^t}{w_i^t }\; \left(\hat{\gamma }_k^t -\sum_k {w_k^t } \hat{\gamma }_k^t \right) }\right) - 1 $$

(19)

One can use the standard errors of these effects to predict the statistical significance of country-level export performance and structural effects. We compute standard deviations for terms Perf _i , Geo _i , and Sect _i using the so-called Delta method, as the diagonal of square matrices $S = \left( A \Upomega A^T \right)^{1/2}$, where A is the matrix of partial derivatives of effects Perf _i , Geo _i , or Sect _i with respect to the vector of all estimated effects ${\bf v} = (\alpha_i^{t} \; \beta_j^{t} \; \gamma_k^{t})$, and $\Upomega$ is the variance-covariance matrix of effects α ^t _i , β ^t _j , γ ^t _k estimated with Eq. (7). Note that the covariance of different type of effects is equal to zero: $\Upomega = \left(\begin{array}{ccc} \sigma(\alpha_i^t)^2 & 0 & 0 \\ 0 & \sigma(\beta_j^t)^2 & 0 \\ 0 & 0 & \sigma(\gamma_k^t)^2 \\ \end{array}\right)$.

Using the definition of Perf _i , Geo _i , or Sect _i (see above), we have:

$$ \begin{aligned} S({Perf}) = & \left( \left( \begin{array}{c} \vdots\\ \frac{\partial {Perf}_i}{\partial {\bf v}}\\ \vdots\\ \end{array} \right) \Upomega \left( \begin{array}{c} \ldots \frac{\partial {Perf}_i}{\partial {\bf v}} \ldots\\ \end{array} \right) \right)^{1/2}\\ S\left({Perf}\right)_{i,l} = & \sqrt{{Perf}_i {{\it Perf}}_l \left( \sum\limits_{t} \sigma(\alpha_i^t,\alpha_l^t) + \sum\limits_{j,d,t} w_j^t w_d^t \sigma(\beta_j^t,\beta_d^t) + \sum\limits_{k,m,t} w_k^t w_m^t \sigma(\gamma_k^t,\gamma_m^t) \right)}_{i,l} \end{aligned} $$

$$ \begin{aligned} S(Geo) = & \left( \left( \begin{array}{c} \vdots\\ \frac{\partial Geo_i}{\partial {\bf v}}\\ \vdots\\ \end{array} \right) \Upomega \left( \begin{array}{c} \ldots \frac{\partial Geo_i}{\partial {\bf v}} \ldots\\ \end{array} \right) \right)^{1/2}\\ S\left(Geo\right)_{i,l} \, = & \left(\sqrt{\sum\limits_{j,d,t}\frac{\partial Geo_i}{\partial\beta_j^t} \; \sigma(\beta_j^t,\beta_d^t) \; \frac{\partial Geo_l}{\partial\beta_d^t}}\right)_{i,l}\\ = &\left( \sqrt{Geo_i\cdot Geo_l\cdot \sum\limits_{j,d,t} \left(\frac{w_{ij}^t}{w_i^t}-w_j^t\right) \left(\frac{w_{ld}^t}{w_l^t}-w_d^t\right) \sigma(\beta_j^t,\beta_d^t)} \right)_{i,l} \end{aligned} $$

$$ \begin{aligned} S(Sect) = & \left( \left( \begin{array}{c} \vdots\\ \frac{\partial Sect_i}{\partial {\bf v}}\\ \vdots\\ \end{array} \right) \Upomega \left( \begin{array}{c} \ldots \frac{\partial Sect_i}{\partial {\bf v}} \ldots\\ \end{array} \right) \right)^{1/2}\\ S\left(Sect\right)_{i,l} \,= & \left(\sqrt{\sum\limits_{k,m,t}\frac{\partial Sect_i}{\partial\gamma_k^t} \; \sigma(\gamma_k^t,\gamma_m^t) \; \frac{\partial Sect_l}{\partial\gamma_m^t}}\right)_{i,l}\\ = & \left( \sqrt{Sect_i\cdot Sect_l\cdot \sum\limits_{k,m,t} \left(\frac{w_{ik}^t}{w_i^t}-w_k^t\right) \left(\frac{w_{lm}^t}{w_l^t}-w_m^t\right) \sigma(\gamma_k^t,\gamma_m^t)} \right)_{i,l} \end{aligned} $$

where $i,l=\overline{1,I}, \; j,d=\overline{1,J}, \; k,m=\overline{1,K}, \;$ and $\; t=\overline{1,T}$.

The standard errors of shift-share decomposition effects are obtained as follows:

$$ \begin{aligned} \sigma({Perf}_i) = & \sqrt{{Perf}_i^2\cdot\left( \sum\limits_{t} \sigma(\alpha_i^t)^2 + \sum\limits_{j,d,t} w_j^t w_d^t \sigma(\beta_j^t,\beta_d^t) + \sum\limits_{k,m,t} w_k^t w_m^t \sigma(\gamma_k^t,\gamma_m^t) \right)} \\ \sigma(Geo_i) = & \sqrt{Geo_i^2\cdot\sum\limits_{j,d,t} \left(\frac{w_{ij}^t}{w_i^t}-w_j^t\right) \left(\frac{w_{id}^t}{w_i^t}-w_d^t\right) \sigma(\beta_j^t,\beta_d^t)} \\ \sigma(Sect_i) = & \sqrt{Sect_i^2\cdot \sum\limits_{k,m,t} \left(\frac{w_{ik}^t}{w_i^t}-w_k^t\right) \left(\frac{w_{im}^t}{w_i^t}-w_m^t\right) \sigma(\gamma_k^t,\gamma_m^t)}. \end{aligned} $$