**3. Methodology and Data**

The intention of this study was to make a proof-of-concept of the possibility to use the IIT as a proxy for eco-innovation measures. The methodology includes the following steps. Firstly, we collected data from Eurostat for the eco-indexes (Eco-Innovation Scoreboard, Eco-IS) for all EU countries. We utilised the Eurostat eco-innovation index to show how well each individual Member State performs in eco-innovations, as compared to the EU average. As the EU-28 average index is equal to 100, the value for each country depends on the relevant components(eco-innovation input, eco-innovation activities, eco-innovation output, environmental results and socio-economic results). Based on these data, we calculated a standard deviation of eco-indexes per country and per year. The degree of variation of the index allows predicting which countries would be potentially problematic in their eco-innovation performance. Other necessary data for our research are the import–export data (in thousands of dollars) from WITS (COMtrade) database of UN for the EU-28 countries. We utilised SITC Revision 1 classification, with one-digit level of aggregation ("Section") per year and per country, because it represents the most recognisable and used nomenclature of goods. This nomenclature includes the following groups: food and live animals, beverage and tobacco, crude materials and fuels, mineral

fuels and relevant materials, animal and vegetable oils and fats, manufactured goods classified chiefly by material, machinery and transport equipment, miscellaneous manufactured goods, commodities and other (not class) goods. All data were selected for the period 2010–2018 for the above-mentioned 10 sectors. Further, we extended the research for two-digit level ("Division"), comprising 60 sectors. In fact, we have a "mirror" database for the trade of each EU country, or in other words one country (Reporter country) has 27 trading partners and each of these partners have counter-standing trade position for this country. Furthermore, in our study, we utilised the classical Grubel–Lloyd formula, according to the data for the export of each EU country:

$$GL = 1 - \frac{\sum\_{k=1}^{n} |X\_k - M\_k|}{\sum\_{k=1}^{n} (X\_k + M\_k)} \tag{1}$$

where *Xk* and *Mk* are respectively the export and the import of the *k*-th sector.

We utilised the Grubel–Lloyd index as a measure of proximity in the countries' technology levels. Values closer to 1 indicate a bigger share of intra-industry trade (IIT) between countries, and vice versa. Intra-industry trade, also called two-way trade, consists of simultaneous import and export of products (including assemblies, subassemblies and components) coming from the same industry, which are close substitutes in the sphere of consumption, production, or in both areas [12].

The first IIT index was proposed by Balassa (1966) [34], later the major developments were made by Grubel and Lloyd (1975) [12], which is the most important contribution in the area. During the years, different authors, such as Aquino in 1978 [35], Henao-Rodrigez et al. in 2016 [36], Brulhart in 2009 [37], Greenaway in 1983 [38], Hamilton and Kniest in 1991 [39], Siggel in 2006 [40] and Glejser, Goossens and Eede in 1982 [41], made good overviews of measurement issues of IIT. When the Grubel–Lloyd index (GL) varies to -, it indicates pure inter-industry trade and when the index varies to 1, it indicates pure intra-industry trade. We calculate the average meaning of the GL indexes per country, per year. The average meaning represents a value for the country as a whole, thus the trade strengths with a specific commodity are eliminated (presumably, each country has its own natural, socio-economic or other advantages). The next step of our study was to make a regression of the average values of GL indexes with the eco-indexes in order to search for *p*-values < 0.05 in the ordinary least squares model (OLS). *P*-values show whether there is an interrelation between the values of the GL index and the eco-index. If there is an interrelation, we assume that we can use the GL rather than the eco-index, because it shows the eco-index validity. In this sense, the level of intra-industry trade between countries coincides with their ecological orientation. The more developed countries are expected to have more predominant intra-industry trade (IIT), and vice versa. IIT is a suggestive hint for the degree of technological development of the countries and acts also for a driver of their environmental development. We observed also *R*<sup>2</sup> value to be equal to or greater than 0.10, as suggested by Falk and Miller in 1992 [42]. The final step of our study was the clusterisation process, which served for grouping of countries and offered an initial benchmark for their eco-innovation level. On the basis of the Eco-Scoreboard (Eco-IS) database, we positioned the countries both in space (28 EU countries in the 8-dimensional space, according to the number of years relevant to this study) and in time (9 years in the 28-dimensional space, according to the number of countries). The purpose of the clusterisation was to find out whether the EU-28 countries differ in their environmental progress per year. Through the calculations made for the period 2010–2018, we prove the existing interrelation between eco-innovation and technological level in the EU-28 countries, as well as analyse the interaction between the two utilised indexes.

#### **4. Results**

To illustrate the groups of countries within the EU, we start with two dendrograms which represent the eco-innovation index of the 28 Member States of the European Union. The first dendrogram features two sub-clusters that outline the differences in eco-innovation levels in Europe (Figure 1). In clusterising the countries per values of the index in periods, the "East–West" division becomes very clearly visible. East European countries have traditionally less eco incentives, and thus the values of the eco-innovation index are lower in general. Thus, the situation for the 1990s, when the East European countries did not participate in international trade and generally lacked incentives, continues to be a particularity 20–30 years later. Environmental consequences are reminiscent of the situation in West European countries one or two generations ago [43].

**Figure 1.** Eco-innovation levels in EU, 2010–2018. Source: Own calculations based on Eurostat data.

By observing the clusterisation diagram (Figure 2) of the eco-innovation index, with the clusterisation made across the countries' dimension, we can also clearly distinguish between the two halves of the period—2010–2013, and 2014–2018, respectively. The main deviations are in 2010 and in 2013, as the first one can be related to the global financial crisis and the second with the recovery from the effects of the financial crisis including the delays in investing in eco-innovations. The clusterisation is in line with findings of other studies [44] that, during a crisis, the part of productivity growth which comes as a result of eco-innovations is reduced.

Our findings show that splitting the clusterisation in two—i.e., for 2010–2013 and 2014–2018—does not change the groups of countries significantly. Countries do not move substantially between sub-clusters, they tend to remain in their initial places in general.

The clusterisation diagrams (available on demand) of the eco-index for the EU countries show that countries such as Germany, Sweden and Finland in the periods 2010–2013 and 2014–2018 sustained their levels of the index. The eco-index values for those countries, which traditionally invest in eco-innovations, are also high. The comparison for the same two periods of the eco-index values of countries such as Bulgaria, Romania and Croatia shows instability and low levels, which is linked to the low investment in eco-innovation in these countries.

Further, we considered the standard deviations of the index, per country (see Figure 3).

The highest fluctuation of the value is for Bulgaria, followed by other relatively poor countries. The lack of experience in eco-innovation activities can be a plausible explanation for the relatively high standard deviation of the index. The lowest values are observed in the most developed countries—Germany and the Scandinavian countries—where the eco-oriented policies have been applied for many decades.

Next, we continue with the Grubel–Lloyd index.

**Figure 2.** Eco-innovation levels in EU, by period. Source: Own calculations based on Eurostat data.

**Figure 3.** Standard deviations of the eco-innovation index. Source: Own calculations based on WITS data.

We wanted to check the fundamental ability of IIT to be a proxy for the eco-innovations, therefore we used data for the EU countries. In this way, we wanted to handle the "data existence and quality" problem—for many countries, the data for eco-innovations are either unavailable or incomplete, or both.

We consider EU data suitable for our study, because of the relative comparability between the country data and the relatively wide sample, with regions including countries of different technology, climate and income.

For our purpose, the data for 2010–2018 were retrieved from the WITS (Comtrade) database. We took exports and imports in USD, mirror data for the 28 countries, split on SITC Revision 1, one-digit ("Section") and two-digit ("Division") level of aggregation. We utilised data for 2010–2018, i.e., for one-digit 2017 data, we took 7547 export/import pairs, for 2016, 7560 pairs, etc. We consider such division of data as informative enough, since our goal is to make a "proof-of-concept" study. Furthermore, the literature on VIIT vs. HIIT (vertical vs. horizontal IIT) suggests that HIIT is a relatively rare case, even between EU countries, and the "Section" division captures VIIT quite well.

We calculated the bilateral GL-indexes and also the average values per country per year (Table 1 represents the calculations for 2017 only; others are available on request from the authors). The data (Table 1) confirm that IIT is large for the trade between countries with relatively similar profiles (income per capita, consumers with similar demand, etc.). For example, Estonian GL-indexes have relatively big values for Finland (0.72, similar language and traditions) and for several countries with similar (totalitarian) past and income—Bulgaria (0.78), Romania (0.74), Lithuania (0.64), Latvia (0.69) and Portugal (0.61)—unlike for Luxembourg (0.05).

**Table 1.** IIT Grubel–Lloyd indexes for 2017 EU-28. Source: Own calculations based on WITS data.


Table 2 shows the average of the GL-index by country for the period 2010–2018. First, an index was calculated for each partner country, and then average values for each country were calculated. This gives a clearer picture of the values of intra-industry trade in EU countries.

The highest average values of the indexes are of Germany (0.68–0.70, Table 2).

Our next step was to search for linkages between the GL-index and the eco-innovations index. We used simple regression (ordinary least squares, OLS) because we aimed to obtain robust and easy understandable and replicable results.

Year by year regressions of the eco-index values on the average Grubel–Lloyd values (for each year we regress the GL index on the eco-index) have significant results for many of the years studied—there is a link between GL and the eco-index.

In the standard OLS models with an intercept, the *p*-values for the slope coefficients are significant at 5% in 2010, 2013, 2014, 2015, and 2018; for 2012, the coefficient is significant at 10% level with SITC one-digit and at 5% with SITC two-digit data. More specifically, the GL-index seems to be dependent on eco-innovation. In models without an intercept, *p*-values are very close to zero, indicating an even stronger interdependence. Our results confirm that for the bigger part of the years (2010–2018) R2 values are greater than 0.16 and the variance explained of our particular endogenous construct is deemed adequate.


**Table 2.** Average GL-indexes, 2010–2018, EU-28. Source: Own calculations based on WITS data.

*p*-values of the slope coefficient are represented in the following Table 3.

**Table 3.** OLS results—*p*-values of slope coefficients, 2010–2018, EU-28. Source: Own calculations based on WITS data.


We also calculated the standard deviation (coefficient of variation) of both the GL-indexes and the eco-innovation indexes, in order to explore whether there is a positive link between variances. The average standard deviation for the period, by country, regressed on the standard deviation of the eco-innovation index leads to significant results (F-statistic: 5.46 on 1 and 25 DF, *p*-value: 0.027), *p*-value for the slope coefficient is 0.027 (t = 2.306). We can claim the existence of a link between both variances, of the GL-index and of the eco-innovation index. Regressing the standard deviation of the eco-innovation index on the average values of the GL-index gives an even more interesting result—a *p*-value of 0.033 (Figure 3).

We argue that the level of the eco-innovation index is closely related to the level of intra-industry trade index. The use of the SITC Rev 1 two-digit level data (more disaggregated data for 60 sectors) only further supports our results, because p-values in the OLS regressions become smaller. In this case, the distinction between the polluting and non-polluting sector does not have a leading role, because it is country-specific and period-specific for each country and not a subject of cross-country analysis. Therefore, the measure for IIT (the GL-index) can be used as a proxy for eco-innovation level. The data for IIT have better availability, thus using this proxy makes sense.

#### **5. Discussion**

For a long time, the focus on innovations was concentrated on their dynamics in general. Among the large-scale efforts to measure innovation are: the European innovation scoreboard; OECD science technology and industry outlook; UNCTAD world investment report; OECD comparative innovation performance: countries and policies review; Eurostat CIS-2; etc. Specifically for the EU effort, the eco-innovation scoreboard, the calculations made are subject to large fluctuations by country, resulting in variations in the results obtained and might be misleading for the correct level of eco-innovation performance in a particular country. Measuring innovations is reasonable mainly for sustainability purposes, because many normal innovations are de facto eco-innovations. Our study confirms the need for prominent empirical tool to measure eco-innovation. Moreover, the degree of technological development of the countries depends on the degree of their trade exchange. For this reason, an alternative indicator is needed to act as a verifier of the eco-innovation index in order to ensure that this index is a suitable measure. One of the explanations for this fact is that the

existing data for the eco-innovation index have been available since 2010 (less than 10 observations per country, because it is an annual index, with data for 2010–2018), therefore unit root and co-integration techniques just do not work, even the panel ones—with such data length tests have virtually no power at all. Therefore, we have adopted the strategy to make several OLS regressions for each year in the N dimension (we have ca. 30 countries). To exclude the possible endogeneity, we make the OLS in the "space" dimension—among countries, for the same year. For this reason, the OLS cannot be considered a basic one, but it is the only applicable in our study (considering the observations in the T dimension). Therefore, we offer a GL index, exploring carefully the empirical work guided by theory. Our research of EU countries for the period 2010–2018 shows that, due to the interrelation between eco-innovation and intra-industry trade (IIT), the GL index is a proxy indicator for eco-innovation performance across countries. The results of our research show that the GL-index and the eco-index reveal the strengths and weaknesses of each country in terms of eco-innovation and IIT in a similar way. We note that the intra-industry trade, and the simultaneous import and export of similar types of goods or services, determines the level of eco-innovations in all EU countries. The results of the study allow us to also place particular attention to the eco-innovation level by country. It is confirmed that the EU-28 countries are technologically different and the Eastern European countries have a lower degree of environmental progress. Our study confirms some other research results for the European countries: countries with a lower level of socio-economic activity as compared to their supportive environment need to prioritise on using more direct measures such as appropriate technology transfer to promote and increase the eco-innovation activities [30]. Our analysis also argues that the higher is the income in a given country, the higher are the levels of ecological development and IIT. In this context, the eco-innovation is becoming a problematic indicator for lower-income countries because the rising incomes affect environmental quality in a positive way. Our study amplifies the knowledge on some other observations [45] that IIT is linked to FDI inflows, particularly in Eastern European "transition" economies, because the low level of the eco-innovation index in these economies is also due to weak foreign investments in sustainable production and technology. The proposal for a proxy indicator fully corresponds to the eco-innovation mission to realise a new policy learning in Europe, as suggested by Kemp in 2011 [46] and better identification of policy mix and their instruments for sustainable development as mentioned in the platform of the European Sustainable Development Network [47]. Within the EU-28, this indicator can be implemented by the national governments as opportunity measure for successful trade and technological experience.
