Environmental Kuznets Curves for Air Pollutant Emissions in Italy: Evidence from Environmental Accounts (NAMEA) Panel Data

Abstract

This paper provides new empirical evidence on delinking and Environmental Kuznets Curves (EKC) for greenhouse gases and other air pollutant emissions in Italy. A panel dataset based on the Italian NAMEA (National Accounts Matrix including Environmental Accounts) for 1990–2001 is analysed. The highly disaggregated dataset (29 production branches, 12 years and nine air emissions) provides a large heterogeneity and can help to overcome the shortcomings of the usual approach to EKC based on cross-country data. Both value added and capital stock per employee are used as alternative drivers for analysing sectoral NAMEA emissions. Trade openness at the same sectoral level is also introduced among the covariates. We find mixed evidence supporting the EKC hypothesis. The analysis of NAMEA-based data shows that some of the pollutants such as two greenhouse gases (CO₂ and CH₄) and CO, produce inverted U-shaped curves with coherent within-range turning points. Other pollutants (SO_X, NO_X, PM₁₀) show a monotonic or even N-shaped relationship. Macro sectoral disaggregated analysis highlights that the aggregated outcome should hide some heterogeneity across different groups of production branches (industry, manufacturing only and services). Services tend to present an inverted N-shape in most cases. Manufacturing industry shows a mix of inverted U and N-shapes, depending on the emission considered. The same is true for industry (all industries, not only manufacturing): although a turning point has been experienced, N-shapes may lead to increased emissions with respect to very high levels of the economic driver. In general, EKC evidence is more pronounced for greenhouse gases. The results suggest that analysis at macro sector (whole industry, manufacturing only and services) can be the most promising approach to future research on EKC.

Keywords:

• Delinking
• NAMEA
• air pollutants
• environmental Kuznets curves

Notes

For extensive evidence and discussion on the period prior to the 1990s see Tilton (1988, 1991) on metals/materials, Martin (1990) on energy, and Zoboli (1995). For recent thorough analyses of long run energy efficiency trends see Gruebler et al. (1999) and Ayres et al. (2004).

Among the early works on pollution, see Holtz-Eakin and Selden (1992), Ten Kate (1993), Grossman and Krueger (1994), and Selden and Song (1994).

The EKC hypothesis does not originally stem from a formalised theoretical model, but recent contributions have started showing how it may be included in formalised economic models. A seminal recent contribution in this direction is Copeland and Taylor (2004). See also Andreoni and Levinson (2001), who set the EKC within a microeconomic production function framework, showing that increasing returns from abatement are a key explanation of EKC shapes, and Chimeli and Braden (2005). Kelly (2003) finds that the EKC shape depends on the dynamic interplay between the marginal costs and benefits of abatement. Bella (2006) presents an endogenous growth model related to EKC reasoning. His work noticed that as countries develop, they always cease to produce certain pollution intensive goods, no matter their starting level of development; it appears an economy ‘where environmental concerns affecting the welfare of future generations enter the decision making problem of a green social planner’. Prieur (2007) measures the repercussions of irreversibility (of the pollution) on the relationship between growth and the environment. He found some equilibria that exhibit irreversible pollution that have the characteristics of poverty traps. Economic growth is accompanied by the accumulation of ecological debt, but due to the irreversible character of some pollution, the debt may be such that, once the economy engages in maintenance, the effort does not suffice to avoid the irrevocable degradation of the environment; this finally creates an economic recession. Cantore (2005) shows that the existence of a turning point is not a reliable criterion to extract useful policy information because the welfare, considered as a third dimension, is not considered in the world composed only by the two economic and environmental dimensions. Di Vita (2003) analyses EKC in relation to discount rate issues.

For a more detailed survey see Mazzanti et al. (2007). See also Stern et al. (1996), Cole et al. (1997), Ekins (1997), Yandle et al. (2002), Cole (2003), Dinda (2004, 2005), Stern (2004), Fonkych and Lempert (2005), and Managi (2006a, b) for extensive, critical surveys of the EKC literature.

The paper, which is strictly linked to that by Matthews et al. (2000), presents descriptive quantitative evidence on material, waste and emission flows, from the perspective of material input–output accounting.

The parametric analysis presents costs and benefits, with respect to semi- or non-parametric investigations; the latter do not, by definition, fully outperform parametric models (Greene, 1997, p. 904).

The authors claim that EKC studies have made two main significant contributions: they launched an agenda along the trade-environment links, and provided evidence that there exists an income effect that raises environmental quality.

Caratti et al. (2006) survey the availability of environmental data across different official international sources. Their investigation highlights that main added value could derive from studies that exploit newly available disaggregated data at national/regional level, and on specific realms such as waste.

This is true for all the EKC literature. Concerning air pollutant emissions, List and Gallet (1999) present evidence on the US using state-level SO₂ and NO_X emissions from 1929 to 1994. The large majority of states follow an EKC shape, predominantly in quadratic rather than cubic form, and with a larger share of states for NO_X. Then, turning points predicted by the traditional panel model are lower than the peaks observed state by state. Thus, traditional panel analysis may lead to overly optimistic conclusions, driven by the result that represents the average picture, hiding specific EKC dynamics by states or regions within countries. See also the recent evidence provided by Managi (2006a, b), on US and Japanese data, who supports the idea that analyses based at a more disaggregated geographical or sectoral level are needed for advancing the EKC literature.

See the works by Ike (1999), Keuning et al. (1999), Vaze (1999), among others, who provide descriptive and methodological insights on NAMEA for some major countries. Steenge (1999) provides an analysis of NAMEA with reference to environmental policy issues, while Nakamura (1999) exploits Dutch NAMEA data for a study on waste and recycling and input–output reasoning. We claim that the application of quantitative methods to NAMEA matrices may provide a contribution to advancements in EKC research.

On the private consumption side, Roca and Serrano (2007) recently analysed the impact of Spanish households on atmospheric pollution in 2000 by considering six greenhouse gases and three other gases. They integrated input–output tables, the Spanish NAMEA and the Spanish Household Budget survey by different typology of households' size but they do not use longitudinal data (only year 2000).

Carbon dioxide (CO₂), nitrous oxide (N₂O), methane (CH₄), nitrogen oxides (NO_X), sulphur oxides (SO_X), ammonia nitrogen (NH₃), non-methane volatile organic compounds (NMVOC), carbon monoxide (CO) and particulate matter (PM10). Lead (Pb) emissions were excluded from the analysis.

NAMEA data also include emissions derived from three household consumption activities (transport, heating and other, such as painting and solvent use) but we have excluded these source of emissions because our interest deals mainly with the productive activities disaggregation (for which the macro sectors are primary, industry and services).

Accounts were not available for the first years for all the 50 branches considered in the last years. Thus, we could not use the full breakdown as data losses would have been too large. We structured the panel assigning equal weight to temporal and cross-section heterogeneity, rather than biasing towards the latter by using a shorter run but larger dataset.

Other NAMEA matrices were recently published including the year 2002. However, 2002 was excluded due to heterogeneity of data between 1990–2001 and 2002. The update of NAMEA to a homogeneous 1991–2005 series will be available in early 2008, according to ISTAT.

See Figure 2 for trends concerning the main economic indicators at aggregate level.

The analysis is grounded on a typical ‘larger than longer’ panel database, exploiting relatively more cross-section heterogeneity. Much of the most recent empirical EKC literature criticises EKC evidence on the basis of flaws arising from not accounting for typical time series problems, such non-stationarity of non-linear models, cointegration issues, etc (Vollebergh et al., 2005; Galeotti et al., 2006a, b; Wagner, 2006). It has to be recalled, however, that standard panel analysis is framed on cross-section data that witnesses more waves/years (more than one, but typically not many).

Shobee (2004) suggests a third-order polynomial specification as a more realistic relationship between environmental degradation and income per capita.

We are aware of some critical arguments arising from the literature, in particular the causality effect, from economic driver to environmental impacts. Some argue the possibility that reverse causality may exist; that is, lower or higher environmental pressures may affect income/welfare. Although possible, we point out that, first, external effects are not included in GDP/VA measures; second, effects may be quantitatively marginal. Emissions are the ultimate effect of production activities causing limited feedbacks, but damages society as a whole. Then, the use of instrumental variables for the economic driver may be attempted, although data availability and reliability for instruments is a severe constraint. We nevertheless remind readers that our present context is a bit different to a typical EKC framework: we analyse here environmental ‘efficiency’ measures at sector level, per employee and not in per capita terms.

Following the procedure in Wooldridge (2002, p. 176), which tests serial first-order correlation by a t-test on the coefficient of the lagged fitted residual term in a regression that takes the fitted residual in time T and the vector of explanatory factors as the dependent variable. Lagged residuals are significant in both FEM and REM models; thus, the corrected correlation model, which does not consider time T for estimation, is indicated. We recall that the corrected correlation model reduces the number of observations since it is based on T-1 periods, unlike the time period effect model.

We estimate the EKC model using a least square dummy variable specification (LSDV), fixed effects (FE). We use a LSDV model since we are not interested specifically in estimating individual fixed effects, which may be inconsistently estimated when sample size increases. On the other hand, the alternative within-effects model does not present an intercept. Since no dummy is used, this model has a larger degree of freedom for error, resulting in incorrect (smaller) standard errors for the parameter of interest (Wooldridge, 2002).

In Mazzanti et al. (2007) an analysis of all the emissions geographically disaggregated is presented. It involves a panel dataset with the same nine emissions considered in this paper, 103 Italian provinces and 3 years (1990, 1995 and 2000) and includes all the emissions of the economy (both on the productive and the household side).

See Femia and Panfili (2005) for a descriptive analysis of eco-efficiency (emission on value added) on different sectors, using NAMEA 1995 and 2000 datasets. Agriculture is not considered due to insufficient data for a sound investigation.

Despite this, it has to be noted that the regressors' value added per employee in logarithm and squared logarithm are not correlated with the trade openness indicator.

With regard to the services, a specific consideration has to be made: the trade openness indicator can be obtained only for the services K and O (real estate, ICT, R&D, firm services and other public, social and personal services). The effect of trade openness is negative, in this case, and significant.

Recently Nansai et al. (2007) proposed a new environmental indicator for consumption that focuses on the relationship between consumption growth and technology advance (eco-velocity of consumption). Its calculated value for Japanese household consumption in terms of CO₂ emissions shows that, without a shift of consumption patterns, it may be very hard to reduce CO₂ emission and achieve the target set in the Kyoto Protocol.