State Aid and Export Competitiveness in the EU

Abstract

Despite the proclaimed return of industrial policy state aid provided by EU Member States is at historically low levels. This fact is at least partly explained by the unique institutional arrangement in the EU which empowers the European Commission to monitor and restrict state aid activities of Member States. Making use of European state aid statistics over the period 1995–2011 we construct a measure for manufacturing state aid and estimate an expanded macro-economic export function to investigate the relationship between state aid for the manufacturing sector and Member States’ export performance. Since national governments can be expected to provide subsidies primarily to foster domestic value added we use value added exports as the export performance measure. Non-stationarity of the data confines us to investigating the short run relationships in which we only find limited evidence for a significant impact of state aid on manufacturing value added exports.

Appendices

Appendix 1: Methodology for calculating value added exports

This appendix explains in some more detail the methodology for calculating the value added exports used in this paper. The concept of value added exports was initially suggested by Johnson and Noguera (2012), though these expositions follow more closely the discussion in Stehrer (2012).

As explained in the main text, three components are required to calculate the value added exports. For any reporting country r, these components are the (industry-specific) value added requirements per unit of gross output, v ^r _i ; the Leontief inverse of the global input–output matrix, L; and the global final demand vector, f ^C _i , where i denotes the industry dimension and the subscript C indicates that the vector comprises the final demand of all countries c ∈ C.

Country r’s (industry-specific) value added coefficients are defined as ${v}_i^r=\frac{value\ adde{d}_i^r}{gross\ outpu{t}_i^r}$. The value added coefficients are arranged in a diagonal matrix of dimension 1435 × 1435 (41 countries × 35 industries). This matrix contains the value added coefficients of reporting country r for all industries along the diagonals. The remaining entries of the matrix are zero because the interest here is with the value added created in country r.

The second element is the Leontief inverse of the global input–output matrix, L = (I − A)^− 1 where A denotes the matrix of coefficients. In the WIOT this matrix of coefficients (and hence the Leontief matrix) is of dimension 1435 × 1435 which contains the technological input coefficients describing the domestic production process in country r in the diagonal elements and the technological input coefficients of country r’s imports (from a column perspective) in the off-diagonal elements.

The final building block is the global final demand vector. This vector is also industry specific and has the dimension 1435 × 1. Most importantly, for our purposes, final demand must be split into separate blocks indicating the origin of the demand for the final goods. This split of final demand by country of origin of that demand is, however, done within the elements in the column vector. As usual, each row is associated with one source of the final demand (i.e. one partner country providing this demand). In the 3-country-1-sector case, the full demand vector, f ^C _i , has the form

$$ {f}_i^C=\left(\begin{array}{c}\hfill {f}^{r,r} + {f}^{r,2} + {f}^{r,3}\hfill \\ {}\hfill {f}^{2,r} + {f}^{2,2} + {f}^{2,3}\hfill \\ {}\hfill {f}^{3,r} + {f}^{3,2} + {f}^{3,3}\hfill \end{array}\right). $$

where the subscript C indicates that the vector comprises the final demand of all countries c ∈ C. The typical element of this vector contains the final demand from all possible sources. For example, the element f ^r,3 captures the value of final goods that country 3 demands from country r. Since the idea of value added exports is that it comprises only value added that is created in one country but absorbed in another, the final demand from reporting country r itself needs to be eliminated for the calculation of country r’s VAX. Therefore we will work with an adjusted final demand vector, f ^c ≠ r _i in which reporting country r’s final demand (i.e. the first column in the above matrix) is set to zero. This vector has the form:

$$ {f}_i^{c\ne r}=\left(\begin{array}{c}\hfill 0 + {f}^{r,2} + {f}^{r,3}\hfill \\ {}\hfill 0 + {f}^{2,2} + {f}^{2,3}\hfill \\ {}\hfill 0 + {f}^{3,2} + {f}^{3,3}\hfill \end{array}\right). $$

Reporting country r’s value added exports can then be calculated as

$$ VA{X}_i^{r,*}={\boldsymbol{\upsilon}}_{\boldsymbol{i}}^{\boldsymbol{r}}\cdot \boldsymbol{L}\cdot {f}_i^{c\ne r} $$

(A1)

where VAX ^r,* are the sector specific value added exports of country r to all partner countries.

To further illustrate the calculation, we show the matrices in the three countries–two sector case, where country r acts as the model country and we label the industries with m (for manufacturing) and s (for services). Equation (A1) then takes the following form:

$$ \left(\begin{array}{c}\hfill VA{X}_{m,*}^{r,*}\hfill \\ {}\hfill VA{X}_{s,*}^{r,*}\hfill \\ {}\hfill 0\hfill \\ {}\hfill 0\hfill \\ {}\hfill 0\hfill \\ {}\hfill 0\hfill \end{array}\right) = \left(\begin{array}{cccccc}\hfill {\nu}_m^r\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill \\ {}\hfill 0\hfill & \hfill {\nu}_s^r\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill \\ {}\hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill \\ {}\hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill \\ {}\hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill \\ {}\hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill & \hfill 0\hfill \end{array}\right) \cdot \left(\begin{array}{llllll}{l}_{m,m}^{r,r}\hfill & {l}_{m,s}^{r,r}\hfill & {l}_{m,m}^{r,2}\hfill & {l}_{m,s}^{r,2}\hfill & {l}_{m,m}^{r,3}\hfill & {l}_{m,s}^{r,3}\hfill \\ {}{l}_{s,m}^{r,r}\hfill & {l}_{s,s}^{r,r}\hfill & {l}_{s,m}^{r,2}\hfill & {l}_{s,s}^{r,2}\hfill & {l}_{s,m}^{r,3}\hfill & {l}_{s,s}^{r,3}\hfill \\ {}{l}_{m,m}^{2,r}\hfill & {l}_{m,s}^{2,r}\hfill & {l}_{m,m}^{2,2}\hfill & {l}_{m,s}^{2,2}\hfill & {l}_{m,m}^{2,3}\hfill & {l}_{m,s}^{2,3}\hfill \\ {}{l}_{s,m}^{2,r}\hfill & {l}_{s,s}^{2,r}\hfill & {l}_{s,m}^{2,2}\hfill & {l}_{s,s}^{2,2}\hfill & {l}_{s,m}^{2,3}\hfill & {l}_{s,s}^{2,3}\hfill \\ {}{l}_{m,m}^{3,r}\hfill & {l}_{m,s}^{3,r}\hfill & {l}_{m,m}^{3,2}\hfill & {l}_{m,s}^{3,2}\hfill & {l}_{m,m}^{3,3}\hfill & {l}_{m,s}^{3,3}\hfill \\ {}{l}_{s,m}^{3,r}\hfill & {l}_{s,s}^{3,r}\hfill & {l}_{s,m}^{3,2}\hfill & {l}_{s,s}^{3,2}\hfill & {l}_{s,m}^{3,3}\hfill & {l}_{s,s}^{r,3}\hfill \end{array}\right)\cdot \left(\begin{array}{lllllllllll}0\hfill & +\hfill & 0\hfill & +\hfill & {f}_{m,m}^{r,2}\hfill & +\hfill & {f}_{m,s}^{r,2}\hfill & +\hfill & {f}_{m,m}^{r,3}\hfill & +\hfill & {f}_{m,s}^{r,3}\hfill \\ {}0\hfill & +\hfill & 0\hfill & +\hfill & {f}_{s,m}^{r,2}\hfill & +\hfill & {f}_{s,s}^{r,2}\hfill & +\hfill & {f}_{s,m}^{r,3}\hfill & +\hfill & {f}_{s,s}^{r,3}\hfill \\ {}0\hfill & +\hfill & 0\hfill & +\hfill & {f}_{m,m}^{2,2}\hfill & +\hfill & {f}_{m,s}^{2,2}\hfill & +\hfill & {f}_{m,m}^{2,3}\hfill & +\hfill & {f}_{m,s}^{2,3}\hfill \\ {}0\hfill & +\hfill & 0\hfill & +\hfill & {f}_{s,m}^{2,2}\hfill & +\hfill & {f}_{s,s}^{2,2}\hfill & +\hfill & {f}_{s,m}^{2,3}\hfill & +\hfill & {f}_{s,s}^{2,3}\hfill \\ {}0\hfill & +\hfill & 0\hfill & +\hfill & {f}_{m,m}^{3,2}\hfill & +\hfill & {f}_{m,s}^{3,2}\hfill & +\hfill & {f}_{m,m}^{3,3}\hfill & +\hfill & {f}_{m,s}^{3,3}\hfill \\ {}0\hfill & +\hfill & 0\hfill & +\hfill & {f}_{s,m}^{3,2}\hfill & +\hfill & {f}_{s,s}^{3,2}\hfill & +\hfill & {f}_{s,m}^{3,3}\hfill & +\hfill & {f}_{s,s}^{3,3}\hfill \end{array}\right) $$

The coefficients in the Leontief matrix represent the total direct and indirect input requirements of any country for producing one dollar worth of output for final demand. For example, the coefficient l ^r,r _m,s indicates the input requirement of country r’s services sector from country r’s manufacturing sector for producing one unit of output. Likewise the coefficient l ^r,3 _m,m indicates country r’s input requirement in the manufacturing sector supplied by country 3’s manufacturing sector.

The resulting elements in this example, VAX ^r,* _m,* and VAX ^r,* _s,* are the total value added exports of country r’s manufacturing respectively services sector to all other sectors of all partner countries. What is used in the econometric exercise as dependent variable is each country’s element VAX ^r,* _m,* .

As pointed out in the main text, there are 41 reporters (40 countries plus the Rest of World) and 35 industries, including 14 manufacturing industries in the WIOD. Therefore the actual dimension of the vectors and matrices used in these calculations is 1435 × 1 and 1435 × 1435 respectively. In the actual calculation the resulting 35 entries in the VAX ^r,* _i vector are treated as follows. The values of the 14 manufacturing industries are added up in order to get the manufacturing sector level value added exports, while the remaining industries (primary industries and services) are set to zero.

The VAX are calculated separately for each of the 27 EU Member States covered in the WIOD using this methodology.

Appendix 2: Unit root tests

This appendix provides additional information on the unit root tests and the problem of non-stationarity of the level-data discussed in the main text.

Despite the limited time series dimension of our panel data which consists of 17 years, there is still a need to check the variables for the presence of a unit root. This is because in case of a unit root present in any (or several) of the time series, the log-level regression results may be spurious (see Granger and Newbold 1974). Therefore we perform standard panel unit root tests opting for the methods suggested by Im et al. (2003) and the Fisher-type tests (see Maddala and Wu 1999). The results of the Im-Pesaran-Shin (IPS) and the Fisher-type panel unit root tests are summarised in Table 5.

Based on the inspection of the time series graphs we run the unit root tests with a time trend in the case of our dependent variable (i.e. the value added exports) as well as manufacturing aid, the foreign GDP and the wage level. In contrast, the unit root tests for government effectiveness and the real effective exchange rate are performed without a trend. In the latter case, the exchange rate seems to have a trend in several EU Member States but not in all. However, since the unit root tests without a trend suggests that the real exchange rates time series is non-stationary we cannot rule out the existence of a unit root problem. In Table 5 we then also report the results from the unit root tests without a time trend for the real exchange rate.

Both the IPS test and the Fisher-type tests have as the null-hypothesis that all panels contain a unit root against the alternative that some panels are stationary. For the IPS test we let the lag structure be determined by the Akaike-information-criterion (AIC), for the Fisher-type tests we include one-period lags. The test results in Table 5 suggest that there are severe problems of non-stationarity in both the dependent variable and the explanatory variables with the exception of the foreign GDP.

Table 5

Table 5 Panel unit root tests

Certainly this poses a problem for the econometric model we intend to estimate. It poses a problem because the time series dimension in the panel is rather short so it is hard to detect any co-integration between the dependent variable and the explanatory variables. The Westerland error-correction-based test for co-integration that we use (see Persyn and Westerlund 2008) can only be tested for a maximum of one lead and one lag. The null hypothesis of this test, which is that there is no co-integration between the variables, cannot be rejected for any of the pairwise tests of co-integration between the value added exports and the explanatory variables.³⁷ Hence, a dynamic OLS estimator is not an option for us and we resort to estimating our model in first differences.

The unit root tests for the first-differenced variables, with and without time trend, are shown in Tables 6 and 7 respectively.

In first differences all variables – or at least some panels within each of the variables – appear to be stationary. In particular, this is true for the dependent variable, Δvax, and the explanatory variable of main interest, Δaid, The only exception may be foreign GDP, Δgdp*, where the IPS test with time trend (Table 7) suggests a remaining unit root problem in all panels. However, the gdp* series in levels did not seem to be plagued by a unit root so we proceed with using the data in first differences.

Table 6

Table 6 Panel unit root tests for variables in first differences, no time trend

Table 7

Table 7 Panel unit root tests for variables in first differences, including time trend