1 Introduction

Carbon dioxide emissions and the improvement of environmental efficiency in relation to global warming have become urgent issues throughout the world. Over the last two decades, economic growth has been associated with a 44 % increase in CO2 levels, and only a small number of countries have managed to decrease their emissions during this period.

The advancements in science and technology are considered to be key concerns in addressing environmental issues and confronting climate change (Abbott 2012), but there are several unanswered questions. For instance: “How exactly do technology and innovation affect carbon dioxide emissions?”, “Does technology innovation, especially environmental innovation, positively affect the reduction of emissions?”, and “How can the government act with respect to the policy on relevant innovations?” These are only some examples of questions raised by scholars and policymakers in the last decade.

Most of the literature has relied on firm-level data to test environmental innovation drivers. Among recent contribution, for example, in Berrone et al. (2013), the authors investigate if both institutional pressures and factors internal to the firm’s organisation as well as their interaction, have a positive effect on firm’s propensity to engage in environmental innovation. Based on a sample of 326 firms in the US, they found that institutional pressures can actually enhance eco-innovation adoption in firms, especially when they show higher rates of pollution with respect to the other firms in the sector. Secondly, the authors found that firm’s internal organization matters in terms of assets specificity and resource availability. Similarly, Cai and Zhou (2014) investigated the factors that influence the adoption of eco-innovation in Chinese firms. Their findings highlight that in this country environmental innovation is mainly triggered by pressures external to firms such as customer’s green demand and competitors in the sector. This branch of literature is very extensive, and a survey can be found in Del Río (2009) and Cecere et al. (2014).

Another branch of the environmental innovation literature focused on a sector-level viewpoint. This is a key perspective for two main reasons: first, while innovation occurs within the firm, technological change takes place only at the sector level (Dopfer 2012). Secondly, the sectorial level allows a more rigorous evaluation of the overall policy implications. Goulder and Schneider (1999) investigates the role of induced technological change on CO2 abatement expenditure at a sectorial level. The authors found that even if technological change can lower the cost of achieving a certain environmental target, it implies an higher gross cost of a given carbon tax. Del Río et al. (2011), investigated the drivers of environmental innovation on a panel of Spanish industries and concluded that technology investments are positively and strongly related to human and physical capital intensity and R&D and negatively related to the export intensity of sectors. In addition, they found that policy stringency played a relevant role in shaping the investment choices in environmental technologies. The empirical results of Carrión-Flores and Innes (2010), revealed a negative and significant bidirectional linkage between toxic air pollution and environmental innovation, by the estimation of a panel of 127 manufacturing industries over a 16-year period (1989–2004).

A third wave of research on environmental innovation and its effects on the actual reduction of polluting emissions goes beyond the economic agent perspective and considers a geographical viewpoint to discuss issues such as agglomerative effects and spatial features. Costantini et al. (2013) used NAMEA data to investigate the heterogeneous distribution of emissions across Italy. Considering differences in local factors affecting environmental innovation, they found an agglomeration effect that seems to influence environmental performance at a regional level. Moreover, they found that technological and environmental spillovers are relevant for sectorial environmental efficiency and that these factors can drive environmental efficiency more than internal innovation.

From a country perspective, many authors highlighted differences in pollution emissions trends across countries or group of countries. For example, Kim and Kim (2012) studied the CO2 emission trend in both OECD and non-OECD countries and found that, notwithstanding some variation within the two groups of countries, emissions are decreasing in OECD-countries such as European member states and the US, but they are increasing in countries such as India and China, which are experiencing a great economic growth.

Nevertheless, literature on the effect of technological changes, particularly those aiming to improve environmental conditions, is still rather scarce, particularly concerning regional and local points of view. This paper attempts to fill this research gap by taking a ‘local perspective’ through empirically testing the data of 95 provinces in Italy over the years 1990–2010. In particular, referring to the IPAT/STIRPAT framework we shed a light on the role played by green technological change on climate change mitigation. In particular we constructed a green patent stock and used it as proxy of technological change and we adopted two different measures of environmental performances, i.e. total CO2 emission and environmental productivity. The intuition here is that the development of green technologies at provincial level can induce firms to adopt greener production processes which have become less expensive, and this can lead to an overall increase in environmental performances. As a consequence we expect a strong connection between environmental productivity and the stock of green knowledge, the latter being strongly related to productivity. On the contrary we do not have a priori expectation on the effect of technological change on total CO2 production, given that other factors, like an increase in the scale of the economy or changes in consumption patterns can offset the positive effect of technological change.

The preliminary evidence (at the regional level) presented in Figs. 1 and 2 confirms previous expectations on North–South disparities, with several exceptions. Emissions tend to be more concentrated in more industrialised Northern provinces, while the South tends to produce, on average, less CO2. Puglia is a relevant exception, being the third highest polluter; similarly, Trentino-Alto Adige, a Northern region, is among the cleanest in the country. In particular, concerning CO2 emissions, Piemonte, Lombardia and Puglia are the three regions associated with a higher level of total CO2 production, whereas in the other areas, total emissions are on a homogeneous level. Notably, the regional ranking in regard to emission efficiency (Fig. 2) is fairly similar to that of total emission, but it shows a completely different trend over time. On the one hand, the total CO2 emission generally increases from 1990 to 2010 (with the exception of year 2010 in full economic crises); on the other hand, emission intensity is significantly decreasing, highlighting an overall gain in environmental efficiencies across Italian regions. Finally, Fig. 3 suggests that the overall increasing trend in green knowledge can be partially correlated to the gain in environmental efficiency, which is constantly increasing over time. Moreover, it should be noted that in the case of green patents, the North–South divide is very evident; patents are more prevalent in Northern regions such as Lombardia, Piemonte, Veneto and Emilia-Romagna. These considerations hold also if we scale patent data for provincial Value Added in order to account for differences in the size of Italian provinces (see Fig. 4).

Fig. 1
figure 1

The CO2 emission of 20 regions in Italy for five selected years (Unit: Mg)

Full size image
Fig. 2
figure 2

Emission intensity (CO2/VA) across 20 regions in Italy for five selected years (Unit: Mg/VA)

Full size image
Fig. 3
figure 3

Green patent stock for 20 Italian regions in the five selected years

Full size image
Fig. 4
figure 4

Number of patents per unit of value added, 20 Italian regions in the five selected years

Full size image

Several reasons justify the choice to conduct a territorial analysis of environmental topics. First, regional frameworks allow for focusing the investigation on structural and idiosyncratic features compared to national averages; second, a disaggregated approach provides useful insights on specific environmental and economic development dynamics, which might be useful for regional policymakers; third, this analysis has political economy implications, which can be differentiated across different regions and territories. This is especially relevant in a country like Italy, which is characterised by high disparities, such as the famous North–South divide. Moreover, it should be noted that this infra-country heterogeneity involves not only economic aspects but also environmental performances, which are highly heterogeneous within the country and tend to favour Northern industrial regions, as confirmed by previous studies based on the national accounting matrix for environmental accounts, known as NAMEA (see Mazzanti and Montini 2010). However, although several works at the national level based on hybrid environmental accounts are well established in the literature (De Haan 2004; Mazzanti and Montini 2010), analysis based on the sub-national/regional level is much rarer.

This paper investigates the role of innovation aimed at reducing carbon dioxide emission as a factor that compensates for economic growth and population growth effects. We test the effect of technology on carbon emissions within a STIRPAT framework, using Italian provincial data covering all 95 provinces over the period 1990–2010. Data are collected every 5 years during this period.

We first conduct the empirical analysis on the entire Italian territory, which is subsequently divided in two sub-samples that characterise the Northern Italian regions and the Southern Italian regions; the aim is to determine the different effects of the environmental innovation adoption on CO2 emissions taking into account the Italian North–South divide.

Our main finding is that the stock of green patents did not exert a significant effect on CO2 reduction; on the contrary, it had a significant and positive effect on environmental productivity (CO2/VA). Notably, this effect seems stronger in the Southern regions, suggesting that some technological effect is also emerging in that part of the country.

The remainder of the paper is organised as follows: Sect. 2 presents emissions’ main driving forces; Sect. 3 describes the empirical approach; Sect. 4 discusses the main results; and Sect. 5 concludes.

2 Driving forces

Contributions to literature in this field have discussed the main forces that drive CO2 emissions in specific countries, such as in Great Britain (Kwon 2005), China (Chong et al. 2012; Feng et al. 2012; Liu et al. 2012), OECD countries (Kerr and Mellon 2012), ASEAN countries (Borhan et al. 2012), and the former Soviet Union (Brizga et al. 2013). Some of these empirical analyses have applied the IPAT framework to build a model for polluting emissions (e.g.: Kwon 2005; MacKellar et al. 1995). Results have shown that many factors affect CO2 emissions, such as economic scale, population, industrial structure, energy consumption structure and the level of technology and management (Kaya 1990).

The following paragraphs will explain the most relevant factors in depth.

2.1 Population

Population has been found to play a significant role in determining emission levels; in a paper by Dietz and Rosa (1997), who developed a stochastic version of the IPAT model, they concluded that there are diseconomies of scale for the most populated nations that are not consistent with the assumption of direct proportionality (log-linear effects) common to most previous researches. Shi (2003), in a cross country analysis covering 93 different states, has shown that the effect of income on carbon dioxide emission varies across country groups, and that lower income countries have greater elasticity on population. A similar result is obtained by Cole and Neumayer (2004). Dietz and Rosa (1997) and York et al. (2003) found that the elasticity of a population with respect to income is less than 1, in the context of the IPAT model. Finally, researchers working with micro-level data have shown that activities such as transport and residential energy consumption vary according to age structure and household size (e.g., O’Neill and Chen 2002; Liddle 2004; Prskawetz et al. 2004; Zagheni 2011). Recently, studies using cross-country, macro-level data have shown a similar relationship (e.g., Liddle and Lung 2010).

2.2 Affluence

According to York et al. (2003), affluence can be defined as either per capita production or per capita consumption. Dietz and Rosa (1997) predicted that population and economic growth would exacerbate the problem of greenhouse gas (GHG) emission and estimated that the effects of affluence on CO2 emissions would reach a maximum at approximately $10,000 measured in per capita GDP and would decline at higher levels of affluence. Fan et al. (2006) found that the effect of GDP per capita on total CO2 emissions is greater for low income countries and found that the effect of energy intensity is strong in upper middle income countries by estimating the same model from different income levels.

The role of affluence, as an indicator of economic growth has been first introduced with Environmental Kuznets Curve (EKC), a model which considers the connection between environmental degradation and economic growth (Grossman and Krueger 1995). EKC shows that, the relation between economic growth and environmental degradation has an inverted U shape. This strand of literature highlight as there are three different driving forces behind this relationship: technology, composition and scale of the economy. The adoption of green technologies, increasing the environmental productivity of firms, tend to promote the emergence of a turning point in the relationship between economic growth and environmental degradation, similarly the shift towards a service society could also be able to foster a process of transition towards a greener society. On the contrary, an increase in the scale of the economy could slow down this process. For a recent review of this topic see Carson (2010).

Differently from this representation, the IPAT framework does not impose any shape on the relation between economic growth and environmental impacts.

2.3 Technology

Green technology is meant to play a central role in reducing the environmental effect of CO2 emissions and of other pollutants and to simultaneously enhance economic growth. However, although the economic effects of environmental innovations can be related to the economic effects of a more general type of innovation, there remains a lack of evidence on the effects that green technologies can exert on CO2 emissions. Recently, Wang et al. (2012), who investigated the relationship between innovation in the energy technology sector (proxied by the stock of patents) and CO2 emission in China, found that innovations that are oriented toward carbon-free technologies can significantly help lower CO2 level in China. In Gilli et al. (2014), where the complementarity between environmental innovations and general innovation is investigated, results shows that at least in the European manufacturing sector, the joint adoption of eco-innovation and product innovation can considerably affect environmental performance.

A frequent problem researchers face is the measurement of technology stock; several indexes have been developed and used since 1990, which include research expenditure, the amount of the research staff and patent data. Finally, some contributions have measured eco-innovation or other types of innovation through questionnaire surveys (e.g., Anton et al. 2004; Christmann 2000). Among these measures, patent applications are particularly appealing for researchers for many reasons.

First, patent data are easily available in terms of both time and country coverage, and second, they can be easily and efficiently related to technological fields. Each patent is, in fact, classified through an International Patent Classification (IPC) code, developed by the World Intellectual Property Organisation. This tree-like classification allows for creating technological fields at different levels of detail. For example, Section “D” contains all patents related to “textiles; papers”, and the subcategory “D 21” refers more specifically to “paper making and production of cellulose”, “D 21 F” refers to “Paper making machines; methods of producing paper thereon”, and, at the maximum level of detail, “D 21 F 11/06” refers to the hyper-specific field of patents related to “Processes for making continuous lengths of paper, or of cardboard, or of wet web for fibreboard production, on paper-making machines of the cylinder type”.

This coding allows for the creation of specific technological subcategories to identify specific fields of interest. For these reasons, patent data have long been considered a useful indicator of innovation for economic research (Griliches 1990). Moreover, as Dernis and Kahn (2004) suggested, in general, all the relevant inventions in economic terms are patented, and for this reason, patents may be used as a valuable indicator of innovative activities by firms, sectors or countries.

Nevertheless, patents also suffer from well-known criticalities. First, it is difficult to discern the value of different patents. An indicator created as the sum of patent counts per year by country certainly includes patents with a high commercial and/or technological effect and a patent with a lower value. Second, patent regimes and patent attitudes may be different across countries. This phenomenon may be partly due to legislative differences across countries and partly due to a different general propensity toward patenting (i.e., in some countries, firms might be more likely to patent new inventions than in others).Footnote 1

3 Empirical settings

The IPAT model initially originated from a controversy regarding environmental degradation’s driving factors between Commoner (1971) and Ehrlich and Holdren (1971), which included the three indicators of population (P), Affluence (A) and Technology (T) in the context of analysis to form the formula of. The result was a model that integrated the mutual effect that these three factors exert on environmental pollution I (Impact). Dietz and Rosa (1994) developed a stochastic framework to allow for inferences in the IPAT model. This stochastic model (STIRPAT), which is adopted in the present analysis, also allows for other influential factors to be added to analyse their influence on environmental performance.

Starting from these premises, in the present work, we estimate the following equation:

$${\text{CO}}_{2} {\text{ or }}\frac{{{\text{CO}}_{2} }}{\text{VA}} = \alpha_{i} + \tau_{i} + \beta_{1} \,{\text{population}}_{it} + \beta_{2} \,{\text{value added}}_{it} + \beta_{3} \,{\text{green}} \,K \,{\text{stock}}_{it} + \varepsilon_{it}$$
(1)

where \(\alpha_{it } \,{\text{and }} \tau_{it}\) are, respectively, provincial and year fixed effect, and ɛ it is the error term. The two-way fixed-effect model is estimated trough an ordinary least square estimator.Footnote 2 Dependent variables are CO2it and CO2/VA it which, according to the IPAT/STIRPAT framework, represent environmental effects and environmental productivity respectively, for province i in year t. CO2 in particular, reflects the total environmental effects of economic activities, and CO2/VA accounts for the size of the economy and it is a widely used indicator of environmental productivity (see, among others, Repetto 1990; Gilli et al. 2014). We believe that considering both dependent variables may provide interesting new insights to the literature, disentangling the effect that green technological change has in both relative and absolute terms.

The control variables, Population it and Value Added it are denoted by the terms P and A in the IPAT framework, i.e., the size of human population of the chosen economy (P) and its level of consumption (A), respectively.

Finally, Green K stockit and K Stock it represent the indicator of green technological change and general technological change, computed using data on patent applicationsFootnote 3 filed at the European patent office (EPO).Footnote 4 Because EPO applications are more expensive, Italian inventors typically first file a patent application in their home country and later apply to the EPO if they desire protection in multiple European countries. As a consequence, EPO patents are generally considered to be higher-quality than the national documents and tend to be more homogeneous in value. We believe that this choice partially mitigates the difficulty in disentangling the value of different patents in the stock. The above indicators are derived according to OECD classification.Footnote 5 Table 1 summarises the variables used and presents basic descriptive statistics.

Table 1 Descriptive statistics. Data available for years 1990–1995–2000–2005–2010
Full size table

Some final caveats on the empirical strategies are important. First, the empirical analysis is based on a balanced panel dataset of 475 observations. The dataset is built by merging the data sources of all 95 Italian provinces over the years 1990–2010, each wave of data covering a 5 year period (e.g., waves were available in 1990, in 1995, in 2000, and so on). It is important to note that the country changed its administrative configuration several times during the considered period; consequently, in 2010, there were 12 more provinces than in 1990. To ensure comparability, we refer in the paper to the 1990 configuration, harmonising data when needed.Footnote 6 Second, regressions are run first on the entire Italian territory and only secondly, the sample is split into two subsamples, i.e., Northern regions and Southern regions. The Northern regions include all Northwest and Northeast regions, and the Southern area includes Central and Southern regions and Islands. The purpose of this second set of regressions is to analyse the different patterns of the effect of green patents on CO2 emission intensity. Third, we did not include the flow of patent applications, but following Popp et al. (2011) we considered the stock of past knowledge. In fact, on the one hand, the effect of new technology on environmental performance is not instantaneous, and on the other hand, the effect of older technology is meant to decrease over time. Therefore, we need to discount the number of both total and green patents according to the following formula:

$$K {\text{ Stock}}_{i, t} = \mathop \sum \limits_{s = 0}^{\infty } e^{ - \beta 1(s)} \left( {1 - e^{ - \beta 2(s + 1)} } \right){\text{PAT}}_{i, j, t - s}$$
(2)

According to the previous literature (Popp 2002), the rate of knowledge obsolescence is set equal to 0.1 (β1 = 0.1) and the rate of knowledge diffusion to 0.25 (β2 = 0.25). The resulting knowledge stock varies by province and technology. According to Popp et al. (2011), year fixed effects have been included in all specifications to account for the tendency of knowledge stock to grow over time.

4 Results

Table 2 below presents regression results obtained from the estimation of the model in Eq. 2, using two different dependent variables (CO2 and CO2/VA, respectively) and applying five different specifications. In Specification I, in particular, we use the Green Knowledge Stock to account for technological change dynamics, whereas in Specification II, we employ the stock of total knowledge to test for the effect of overall patenting on our dependent variable. This approach enriches the first specification and provide complementary results. If the effect of green technologies on climate change mitigation is expected to be positive, we do not have a priori expectations of the total knowledge stock. Several countervailing forces play a role in this case: from the one hand an increase in total knowledge can be environmentally exacerbating, given that it could be connected with an intensified economic activity and thus an increase in CO2 emission, while on the other hand it can also be beneficial for the environment if the adoption knowledge is driven by the search for new processes/technologies which are good for the environment. Specification III restricts the sample to only Northern provinces to determine whether the results are driven by geographical disparities, whereas Specification IV studies the behaviour of Southern provinces only. Given the significant correlations among variables (see Fig. 5 and Table 3 in the appendix) a VIF is computed after each regression, in order to detect severe multicollinearity. However, the indicator do not highlight that collinearity as an issue in this analysis. Finally in Specification V we propose an alternative way to estimate the difference in the effect of the green knowledge stock in the two samples by interacting the stock variable with a dummy equal to one for northern regions.

Table 2 Estimation results
Full size table
Fig. 5
figure 5

Scatterplot matrix of the dependent and explanatory variables

Full size image
Table 3 Correlation matrix of the dependent and explanatory variables
Full size table

Specification I results show that technological change only exerts an effect on environmental productivity and that any effects is found with respect to CO2. In particular, column 2 shows a statistically significant and negative coefficient of Green K Stock, which confirms the hypothesis that an increase in a country’s green knowledge base, measured here by green patent stock, has a positive effect on environmental productivity. However, there is no evidence of a positive technological effect with respect to total CO2 emissions. Regarding the other covariates, population is not statistically significant in the Italian context, which is a reasonable result in an industrialised country like Italy, characterised by slowly changing demographic trends.Footnote 7 On the contrary, VA shows a significant and positive coefficient in column 2 but not in column 1. This latter result confirms the evidence found in previous EKC studies, which found no absolute delinking between CO2 or CO2/VA and economic indicators (Marin and Mazzanti 2010). Referring to the EKC context, Column 2 shows the presence of a monotonically increasing relationship (also known as relative delinking) between economic growth and CO2/VA. Overall, these results suggest that, roughly speaking, although green technological change has a positive effect on environmental productivity, it has not been able to shrink the total level of emission. From a macro perspective, a negative scale effect (partially confirmed by the significance of value added) seems to prevail on the positive technological effect. Regarding the quantification of results, a one standard deviation increase in the stock of green knowledge leads to a 0.39 standard deviation decrease in CO2/VA, and an increase of the same size in value added increases environmental productivity by a standard deviation of approximately 0.19.

The regression results of Specification II basically confirms previous evidence, and the magnitude of the coefficient is fairly similar (the standardised coefficient of knowledge stock is equal to −0.34). This phenomenon also suggests that employing a broader concept of technical change does not alter previous evidence. This is not an obvious result, considering that total knowledge stock also includes brown patents, which might have a negative effect on emissions if they increase the value added of pollution-intense sectors. (See Aghion et al. 2012, for a discussion of brown and green patents and their effect on the environment.)

Specifications III and IV show that the aggregate results also hold when splitting the full data set into the two subsamples of Northern and Southern regions of Italy. In this case, the primary evidence does not change, but the magnitude of the effects is much stronger in the South, where 1 standard deviation increase in the green knowledge stock leads to an increase in the dependent variable equal to 1.9 standard deviations, whereas the effect in the North is very similar to the national average.Footnote 8 This latter result—particularly if compared to the descriptive statistics of Figs. 1, 2, 3 and 4, which highlighted how the South tends to have a lower patent propensity—suggests that in these areas, even a small marginal increase in knowledge formation can have a strong effect on environmental productivity. Finally, the main evidence also holds in Specification V, where we interacted the Green K Stock with a north dummy. Interestingly the result confirms that the effect is much stronger for southern regions.

5 Conclusions

This paper has carefully examined primary main factors that may influence CO2 emissions according to the IPAT/STIRPAT framework exploiting an original dataset that covers 95 Italian provinces over the years 1990–2010.

The primary evidence shows that the stock of green patents did not exert a significant effect on CO2 reduction in Italy; instead, it improved overall environmental productivity. On the contrary, the growth in the scale of the economy, proxied here by Value Added, slowed environmental productivity by exerting more pressure on the environment. Overall, this evidence suggests that technology has not yet played a significant role in promoting environmental protection, although a scale effect seems to prevail. Notably, however, green technological change is positively correlated with environmental productivity, and this correlation is stronger in the South, which suggests that some technological effects are emerging in the country.