2. Theoretical background

The positive relationship between the level of income and waste generation has been well established in the economic literature, showing that as disposable income and standards of living increase, consumption of goods follows the same trend and waste generation increases accordingly. The growth in waste production is highly associated with lifestyle and consumption habits, higher production of organic waste and the limited lifespan of electronic devices and equipment, among others. In this setting, the concept of decoupling (or delinking) has attracted special attention, in view of explaining the relationship between economic activity and its impact on the environment, with special focus on waste generation. Decoupling is when the volume of the environmental indicator, such as waste production, increases at a lower rate than the economic indicator used, normally income or added value or any other indicator measuring the standard of living. This result provides evidence in favor of relative decoupling, in contrast to the absolute decoupling that arises when waste production decreases (or stays stable) in absolute terms as economic activity expands (Mazzanti et al., Reference Mazzanti, Montini and Zoboli2008; Jaligot and Chenal, Reference Jaligot and Chenal2018). Relative decoupling is empirically identified by an inverted U-shaped relationship between the waste generation variable and the economic indicator used in the analysis, originally linked to the environmental Kuznets curve (eKc).Footnote ³ It is also associated with a decreasing ratio of material inputs with respect to an economic indicator, implying less waste production in the future.

Indicators of decoupling became popular for detecting and measuring improvements in environmental quality or resource efficiency with respect to economic activity. In this context, a large body of the literature has analyzed the relationship between income and environmental degradation (Grossman and Krueger, Reference Grossman and Krueger1995; Galeotti et al., Reference Galeotti, Lanza and Pauli2006), whereas studies examining the relationship between waste generation and income indicators are less common, especially at the regional or municipal level. Findings on the eKc are not clear and are controversial, since some studies do not support the bell-shaped relationship, while others offer evidence in favor of it, indicating a turning point for the waste Kuznets curve (wKc) (Stern, Reference Stern2004; Ercolano et al., Reference Ercolano, Gaeta, Ghinoi and Silvestri2018). Cross-country studies mostly reject the wKc premise, showing that waste generation monotonically increases with income. However, this kind of empirical research has crucial limitations (Johnstone and Labonne, Reference Johnstone and Labonne2004).Footnote ⁴ It estimates ‘average’ international curves, which might not hold when the analysis applies to sub-national units that possibly implement different policies on waste management. On the other hand, the definition of waste generation varies from country to country, so the results of cross-country analysis are biased. For these reasons, scholars have recently started using more refined single-country data, such as those concerning sub-national administrative units. Studies using municipal data are still few, although increasing, and most seem to support the inverted U-shaped hypothesis (Ichinose et al., Reference Ichinose, Yamamoto and Yoshida2015). The reason for observing a turning point when sub-national units are used is that municipal authorities implement different waste management policies, which cannot be viewed or captured when regional or country level data is used.

Recent literature exploits provincial data to overcome the limitations of cross-country studies. This is the case of Mazzanti et al. (Reference Mazzanti, Montini and Zoboli2009) who analyze panel data for 103 Italian provinces over the period 2000–2004. Their findings strongly support the wKc non-linear relationship between municipal solid waste generation and value-added per capita, used as a proxy for economic development. Another study that validates the wKc principle is conducted by Ichinose et al. (Reference Ichinose, Yamamoto and Yoshida2015), who employ cross-sectional data for the Japanese municipalities concerning the year 2005. They provide evidence that the turning point is significantly lower than the maximum income observed in the sample researched. Trujillo et al. (Reference Trujillo, Bermúdez, Charris and Inglesias2013) use Colombian panel data for 707 municipalities over the period 2008–2011. Their analysis validates the inverted U-shaped hypothesis between landfilled solid waste and economic development, whose turning point differs across regions. Nevertheless, in an earlier study, Cole et al. (Reference Cole, Rayner and Bates1997) found no evidence for an inverted U-shaped wKc curve concerning municipal waste data in 13 OECD countries over the period 1975–90. The environmental indicator measured by per capita municipal waste is found to monotonically increase with income over the observed period without exhibiting a turning point behavior. A study by Gui et al. (Reference Gui, Zhao and Zhang2019) uses panel data of 285 Chinese cities for the period 2006–2015 to explore spatial dependence of waste generation. Employing a spatial autoregressive term, they found that GDP per capita is linearly related with waste per capita instead of an inverted U-shaped pattern as expected by the wKc hypothesis. It is also found that road length, tertiary industry share, and urbanization rate are positively associated with waste while sanitation investment and education level have a moderate negative influence on waste. Although some attempts have been made to address the wKc hypothesis at the municipal level, this kind of work is not extensive and the results are not consensual, leaving room for further research.

According to scholars, a number of economic mechanisms would explain the decoupling phenomenon in waste generation translated into the wKc hypothesis:

1. (i) A scale effect through production or income can explain the wKc performance linked with the development level. At earlier stages of development, waste generation is higher because of higher levels of production and consumption through income increase. This stage explains the ascending trend of the wKc behavior, implying a positive and monotonic link between the development level of an economy and environmental degradation. At higher levels of development, countries have the means to develop more efficient environmental policies, among them an efficient system of waste management processing to preserve the environment. At this stage, the wKc curve changes its shape to a descending trend. An income-effect also explains the turning point of the waste generation curve, due to the diminishing marginal utility of consumption, as income expands (Swart and Groot, Reference Swart and Groot2015). Beyond a certain income threshold, people have satisfied their basic consumption needs and tend to satisfy other immaterial needs with less waste content involved.

2. (ii) A composition effect explains the non-linearity between economic development and waste generation, due to the productive structure of the economy. Countries with a higher share of manufacturing in the overall economy generate more waste than countries with a higher share of services. The former is more resource-intensive than the latter, generating more waste and environmental pollution. When the economies shift from manufacturing to services, there is a reduction in environmental degradation (Tsurumi and Managi, Reference Tsurumi and Managi2010). This structural shift in the economy describes the descending part of the bell-shaped curve of waste generation.

3. (iii) A technological effect can also explain the turning point of the waste generation curve by inventing new production methods with less resource-intensive techniques (Hettige et al., Reference Hettige, Mani and Wheeler2000). Technology can contribute to the reutilization of inputs and the recycling of previously-used production materials. As it is realized, the technological effect is directly related to the level of development, highlighting that more developed countries have the means to allocate financial resources to R&D which is responsible for fostering technological progress.

4. (iv) A preference effect is highly associated with environmental literacy and the public preference for consuming ‘green goods’ with less impact on the environment (Unruh and Moomaw, Reference Unruh and Moomaw1998). This preference effect is highly associated with the level of education, acknowledging that more educated people better understand the need to preserve the environment and adapt their habits to this goal.

5. (v) A policy effect is based on policy-related variables such as waste taxation, weight-based waste fees, and incentives for waste recycling and separation, among others (Hage et al., Reference Hage, Sandberg, Söderholm and Berglund2018).

Important drivers that are often used in the waste generation analysis include socioeconomic and demographic indicators, although there is no consensus on the type of explanatory factors introduced to explain waste generation performance.Footnote ⁵ Some studies focus only on the income variables as the main drivers for explaining the decoupling phenomenon (income per capita, income per household, household expenditure, value-added per head, etc.). Other studies consider socioeconomic variables, such as population age, number of single households, degree of urbanization, population density, education, and employment status, among others. Few studies include policy-related drivers, such as waste disposal options, waste sorting, and waste tax policies, due to the lack of data at the sub-national level. The choice of the explanatory factors mainly depends on the availability of data and the research objectives.

In the existing literature that employed within-country data at a disaggregated level, Jaligot and Chenal (Reference Jaligot and Chenal2018) use ‘tax point value’ as an independent income variable, representing the distribution of wealth per capita between districts in the canton of Vaud in Switzerland, in order to explain municipal waste generation over the period 1996–2015. Other explanatory factors include population density as a proxy for urbanization, fixed waste tax, and bag tax paid by all households. Using a Generalized Least Squares (GLS) estimation approach applied to panel data, they find evidence that waste generation tends to stabilize as income increases, not confirming the emergence of the wKc. Population density is not statistically significant, while the bag tax has an important effect on reducing waste generation in this geographical location.

Another study conducted by Ercolano et al. (Reference Ercolano, Gaeta, Ghinoi and Silvestri2018) considers average tax return per inhabitant, population density, population share above 65 years old, tourism accommodation and the share of foreign residents as the set of explanatory factors to explain waste generation per capita. The study uses municipal-level panel data from the Lombardy region in Italy over the period 2005–2011. The Generalized Method of Moments (GMM) estimation approach validates an inverted U-shaped relationship between economic development and waste generation, although few municipalities reach the turning point. It is also shown that population density generates less waste per capita, and the share of older people and tourism display a positive effect, while the share of foreign residents affects waste generation negatively.

In the same stream of studies, Ichinose et al. (Reference Ichinose, Yamamoto and Yoshida2015) provide evidence of the wKc by applying spatial econometrics to cross-sectional data at the municipal level in Japan. They divide waste generation into four types (total waste, household waste, business waste and landfill waste) and conclude that absolute decoupling has been observed for total waste and household waste only. The set of control variables comprises income per household, number of separated waste items, unit-based pricing of waste (dummy variable), population density, the ratio of single-person households over total households, the ratio of households composed of elderly couples to the total, and a commuter indicator (the ratio of commuters from outside the municipality over the number of people who commute from the municipality to elsewhere). It is shown that the amount of each type of waste significantly increases as the ratio of single-person households increases, and that the higher the ratio of elderly households the lower the amount of waste generated, but this hypothesis in not confirmed in the case of landfill waste, where the relationship is weak. The commuter indicator (as a proxy for economic activity) shows a positive and significant impact on all types of waste, while population density and unit-based pricing of waste decrease the amount of generated waste.

The study by Madden et al. (Reference Madden, Florin, Mohr and Giurco2019) considers municipal data across the Australian state of New South Wales over the 2011–2015 period. The authors use spatial econometric techniques based on a geographically and temporally weighted regression model to test the wKc hypothesis for municipal waste and examine the socioeconomic and urban characteristics of the municipalities included in the sample. To explain per capita waste generation, a set of control variables is used that includes mean household income, population density, number and size of households, proportion of waste recycled and distance to urban context. Their methodology allows them to identify the local government areas that exhibit relative rather than absolute decoupling, and the authors highlight the importance of considering variations in regional socioeconomic and structural characteristics when assessing the decoupling phenomenon.

Reviewing the empirical literature, the results confirm that within-country data may offer more insight than cross-country datasets and are also more valuable from the policy perspective. Evidence on decoupling (or delinking) is extremely important for detecting and measuring environmental improvements in relation to economic activity. Detection of the main fundamentals to explain waste generation and the incidence of decoupling is essential from the policy orientation perspective and for waste management coordination across provinces. Empirical evidence is still limited at the municipal level and our paper aims to fill this gap, considering the Portuguese municipalities as a case study.

3. Data analysis and descriptive statistics

According to data availability, our empirical analysis considers an unbalanced panel data sample, made up of 307 Portuguese municipalities. The time series data covers a recent ten-year period from 2009 to 2018, thus allowing a large sample data set to be considered, which is important for ensuring the asymptotic properties of the estimates. The sample of sub-national units considered is highly heterogeneous both in terms of territorial and population size, but also in terms of the productive structure and economic activity level. Table 1 explains the variables considered in the empirical analysis, the corresponding label, the expected sign of the covariates on the dependent variable (MWGpc), and the source of the data.

Table 1.

Variable definition, expected effect and data source

^a Since 2015 (inclusive), the values quantify the managed urban waste instead of the collected urban waste. Those values refer to the municipal waste sent for processing in the urban waste management system and not to all the collected waste. Considering that in general collected waste is always sent for treatment, this change in the accounting methodology does not create any structural break in the series on waste. This conclusion is reinforced by the fact that no abrupt changes are observed in the data on waste from the year 2014 to 2015.

The dependent variable is defined as municipal waste generated by households, measured in tons per capita on an annual basis. Economic activity is measured by the gross value-added produced in each municipality (non-financial sectors) in per head terms (GVApc), which serves as a proxy for the income indicator.Footnote ⁶ The decoupling result implies a positive sign of the coefficient of the GVApc variable and a negative sign of its squared value. The variables that express the productive structure of the municipalities are given by the share of the primary, secondary and tertiary sectors, labeled as GVA1r, GVA2r and GVA3r, respectively. These shares are obtained by dividing the gross value-added of each sector by the total value-added.

Population density (PopDen) is another covariate used as a proxy for the urbanization stage, reflecting population concentration in urban areas. As discussed before, its impact on waste generation is uncertain and, therefore, it is an empirical matter to look into. The share of elderly people (Pop65) is also used to test the hypothesis in the literature that older people consume less and therefore contribute less to waste generation. Tourism is also found in the literature to contribute positively to waste generation through the consumption effect. To express this variable, we use the overnight-stay ratio given by nights spent in tourist accommodation per 100 inhabitants (Tour). Finally, we use a dummy variable (D) distinguishing the more developed municipalities (with GVApc above the average, that take the value of one) from the less developed municipalities (with GVApc below the average, that take the value of zero). The purpose is to verify whether more developed municipalities behave differently from less developed municipalities with respect to the production of household waste.

Descriptive statistics on the variables used in the empirical analysis are reported in table 2, revealing the heterogeneity across the 307 Portuguese municipalities. The overall sample size is fairly large, varying from 2,297 to 3,080 total observations (this variation is due to missing values). It is shown that the mean annual value of municipal waste generation is 0.45 tons per head and the maximum value around 1.52 is more than triple the average. Differences in waste generation across municipalities can be seen in figure A1 of online appendix A. The map shows that the coastal areas of Portugal produce in general more waste (darker areas) than the municipalities of the interior (countryside), probably explained by the higher income level. Given the objectives of this paper, the gross value-added per capita (GVApc) is used as the income indicator for testing the decoupling hypothesis on waste generation. High heterogeneity is revealed in the production activity, which varies from 0.82 to around 44.03 (thousand euro per head), with a yearly mean value around 4.57. Differences in the production pattern between municipalities can be viewed in figure A2 of online appendix A. It can be pointed out that coastal areas show higher economic activity (darker areas) than countryside areas, which is in line with figure A1 which hints at more waste production in the coastal areas.

Table 2.

Descriptive statistics of variables (overall values)

Source: Authors' own elaboration.

The productive structure of the Portuguese municipalities indicates a higher share of the services sector in comparison to other sectors, as is the case in most developed economies. It is shown that, on average, 49.3 per cent of the value-added is produced in the services sector, 38.3 per cent in the industrial sector, and only 7.7 per cent in the primary sector.Footnote ⁷ Recall that the production structure is used as an income proxy to explain household consumption patterns, and therefore waste generation in the Portuguese municipalities. It is presumed that production in all sectors generates income for the working population, affecting waste production through household consumption. Observing the minimum and maximum values, we conclude that there is a high degree of heterogeneity in the production activity across sectors at the municipal level.Footnote ⁸

The mean value of population density (PopDen) is 295 inhabitants per km² and the range is between 4 and 7,604 inhabitants, which shows a high discrepancy across municipalities. The age structure of population, in turn, reveals the common problem of other developed countries, which is an ageing population. The average share of elderly population (65 years old or more) is 23.4 per cent, varying between 7.8 per cent and 45.6 per cent. In some municipalities almost half of the population is composed of elderly individuals with a lower income, which sets restrictions on the consumption pattern. The tourism variable shows an overall average stay of 523 nights per hundred inhabitants, ranging from 0 to 20,613.Footnote ⁹ This reveals high discrepancies across sub-national units, which differentiates municipalities that attract more tourists (especially the coastal areas) from other less attractive municipalities. According to the literature, tourism is a source of waste generation explained by the increase in population in the host municipalities. Finally, the mean value of the dummy variable (0.34) shows that most of the municipalities report production values (GVApc) below the average, over the period considered.

4. Model specification and estimation methodology

As has been noted, panel data at the municipal level is a preferable option for testing the decoupling hypothesis since local authorities follow different waste management policies and this characteristic cannot be captured when cross-country data is used. The model specification is also an important issue to address in the empirical analysis. Following the existing literature, evidence of the wKc can be derived from the following equation:

(1)\begin{equation}MWGp{c_{it}} = {\beta _1}GVAp{c_{it}} + {\beta _2}GVApc_{it}^2 + \delta ^{\prime}{Z_{it}} + {u_i} + {\epsilon _{it}}\end{equation}

where MWGpc_it is municipal waste generation per capita in the ith municipality at year t; GVApc_it stands for gross value-added per capita, reflecting the economic activity level; Z_it is a vector of k control variables; u_i is the municipal fixed effect; and ɛ_it is the error term.Footnote ¹⁰ The model to estimate is figured as a 2nd order polynomial (a quadratic form)Footnote ¹¹ with respect to the economic activity variable in order to test the inverted U-shaped hypothesis consistent with the wKc curve, which is verified if β₁ > 0 and β₂ < 0 and both are statistically significant.

More specifically, equation (1) allows us to test different relationships between waste generation and economic activity variables:

1. (i) β₁ = β₂ = 0 implies no relationship between the environmental variable (household waste generation) and the economic indicator (gross value-added).

2. (ii) β₁ > 0 and β₂ = 0 infers a linear relationship between the two variables, i.e., a monotonic rise of waste generation as economic activity increases. This is the ascending part of the wKc.

3. (iii) β₁ > 0 and β₂ > 0 implies a continuous increase in waste generation as economic activity increases and a persistent environmental degradation over time, with no turning point accomplishment.

4. (iv) β₁ > 0 and β₂ < 0 confirms the inverted U-shaped relationship consistent with the wKc and the relative decoupling hypothesis. Waste production by households increases at a lower rate than the increase in economic activity after the threshold point is reached. It must be noted that both coefficients should display statistical significance.

5. (v) β₁ < 0 and β₂ = 0 indicates a monotonic decreasing relationship between the two variables, offering evidence that absolute decoupling takes place. This suggests that the waste management policies are effective, enabling the waste production to be disconnected from economic activity.

6. (vi) β₁ < 0 and β₂ > 0 implies a U-shaped relationship and no decoupling effect.

Two standard estimation techniques can be applied to consistently estimate equation (1): assuming fixed effects or random effects. The fixed effects estimation approach uses the Least Squares Dummy Variables (LSDV) method, treating u_i as additional parameters to be estimated.Footnote ¹² In this way, the unit-dummy variables used capture differences between municipalities which are constant in time (such as territory size, geographical location, natural resources, administrative institutions, among others). It also assumes the orthogonality condition of no correlation between the control variables and the error term ɛ_it. The random effects estimation approach uses the GLS method to estimate the model, assuming that the individual effects u_i are not correlated with the explanatory variables.Footnote ¹³ While in the fixed effects model the municipality-level effects are observable, in the random effects model they are treated as unobserved random disturbances. Both fixed and random effects models can be estimated by considering robust standard errors to heteroskedasticity and autocorrelation to obtain efficient estimates. The Sargan-Hansen test can be performed to make the best selection between the fixed or the random effects models, which is appropriate when robust standard errors are used in the regressions.

Equation (1) can otherwise be estimated by introducing dynamics, by including the lagged value of the dependent variable as an additional regressor. In this way, we test whether past realizations of waste generation influence policies to control current waste production. However, a statistical problem arises as the lagged dependent variable might be correlated with the error term, thus producing biased and inconsistent estimates. To overcome this problem, Arellano and Bond (Reference Arellano and Bond1991) suggested the GMM approach, by taking the first differences in variables of the original model and instrumenting the first-differenced lagged values of the dependent variable by using previous lagged levels. It is useful to apply this estimation approach to panel data when the number of individual units is higher than the number of time periods used in the analysis (N > T), as in our case (see Baltagi, Reference Baltagi2005).

Besides gross value-added per capita, other control factors are included in equation (1) to explain municipal waste generation in Portugal. The production structure of municipalities is given by the shares of the primary, secondary and tertiary sectors, obtained by dividing the gross value added of each sector over the total gross value added. We expect differences in the production structure to reflect differences in income levels across municipalities, which will affect waste production through different consumption patterns. According to the literature, the shift in production composition to manufacturing and services is a feature of developed economies and this structural shift is assumed to explain the descending part of the inverted U-shaped curve in waste generation. To the best of our knowledge, this hypothesis has not been tested before in the empirical literature, at least at the municipal level.

Another covariate used is population density of the municipality, given by the number of inhabitants per km² (see table 1). Some authors claim that population density can be considered a proxy for the degree of urbanization. The evidence of the impact of this variable on waste generation is mixed. Ichinose et al. (Reference Ichinose, Yamamoto and Yoshida2015) found that population density has a negative effect on total waste and landfill waste considering municipal data for Japan. Population density was found to be not significant in the study conducted by Jaligot and Chenal (Reference Jaligot and Chenal2018), which tried to explain waste generation in the canton of Vaud in Switzerland. On the other hand, evidence of the positive effect of population density on waste is given by Madden et al. (Reference Madden, Florin, Mohr and Giurco2019), considering municipal data for the Australian state of New South Wales. Therefore, the expected outcome is uncertain, and more research is needed to identify the exact relationship between waste generation and population density.

The population age structure is also used to explain municipal waste generation in Portugal, as recommended by the literature. The share of elderly population appears as an explanatory factor in many studies, suggesting a negative impact on waste generation. Elderly people consume less than other generations, which justifies its negative impact on waste generation. In our study we use the percentage of population aged 65 and over to test this hypothesis.

Tourism is another factor that is expected to positively impact waste generation. It is argued that tourism inflows increase the population of the receiving locations and therefore waste increases due to higher consumption. Tourism inflows given by the overnight-stay ratio at the municipal level is used in our analysis to assess this hypothesis.

Finally, we divide municipalities into two groups having as reference the mean value of total gross value-added per head, as explained in table 1. The aim is to verify whether the group of more developed municipalities behaves differently from the less developed locations as far as solid waste production is concerned.

5. Empirical results and discussion

5.1 Evidence from the fixed effects model

The results of estimating the waste equation (1) with fixed effects are reported in table 3. A stepwise estimation process is employed, introducing each time additional covariates. This strategy makes it possible to control whether multicollinearity affects the regression results. Model 1 tests the decoupling hypothesis only, using total gross value-added per capita as the income indicator. Models 2 to 4 use as an additional driver the productive structure of the municipalities given by the shares of the primary, the secondary and the tertiary sectors. The importance of the share of the two latter sectors is tested separately due to high collinearity between them (see table A1, online appendix B). Model 5 tests the significance of the joint impact of the secondary and tertiary sectors, since the latter is not statistically relevant in the previous regression. Model 6 introduces other covariates, such as elderly population share, population density (as a proxy for urbanization) and the development dichotomy (dummy variable) to test their significance. Tourism inflows (overnight-stay ratio) are introduced as an additional driver in Model 7. All models are estimated assuming fixed effectsFootnote ¹⁴ and robust standard errors are used to ensure the small variance property of the estimates.

Table 3.

Results of the fixed effects model in municipal waste generation in Portugal, 2009–2018

Notes: FE is fixed effects, TP is turning point. Numbers in parentheses are robust standard errors, and numbers in square brackets are P values. **, *** indicate the statistical significance of coefficients at the 5 per cent and 1 per cent significance level, respectively. The F test assesses the joint significance of all slope coefficients and the Sargan-Hansen statistic allows us to select the appropriate model between the FE and random effects when robust standard errors are used.

Results show that the decoupling hypothesis consistent with the wKc framework is only confirmed in Models 6 and 7, when the full set of covariates is introduced in the waste equation. In these two models, the overall production coefficient is positive and statistically significant at the 1 per cent level, whereas its squared value coefficient is negative and statistically significant at the 5 per cent and 1 per cent level, respectively. This evidence suggests an inverted U-shaped relationship between waste generation and economic activity across the Portuguese municipalities, with a turning point at around 34.52 (thousand euros of gross value-added per head on a yearly basis)Footnote ¹⁵ much above the mean value of 4.6 (see table 1). The turning point in Model 7, where tourism inflows are considered, is smaller (26.02), but we should be aware that the sample size in this regression is substantially reduced due to missing data on this variable. Analyzing the data of variable GVApc, we see that only the Lisbon region (the capital of the country) overcomes the turning point 34.52, limiting therefore the decoupling performance to a specific geographical area characterized by a higher level of economic activity.Footnote ¹⁶ Considering the lower turning point (26.02) found in Model 7, besides Lisbon, three other municipalities exhibit decoupling behavior, namely: Oeiras, Sines (the last three years) and Castro Verde. Oeiras is a municipality that is very close to Lisbon, benefiting therefore from some externality effects from the capital city of the country. Sines is a municipality where economic activity is dominated by the oil-refinery, and in Castro Verde the extraction sector (mining) plays a dominant role in the local economy. The evidence of the fixed effects analysis on the decoupling phenomenon reveals that only a restricted number of municipalities satisfy this hypothesis consistent with the wKc principle, and therefore more effective policies must be taken by the rest of the Portuguese municipalities in order to reduce the volume of waste produced.

Our empirical analysis is also focused on assessing the importance of the productive structure on waste generation. As explained before, differences in the production structure will reflect differences in income levels across municipalities, which will affect waste production through consumption. The evidence displayed in table 3 reveals that the production share of the primary sector has a negative impact on waste generation, and this effect is statistically significant in all models at the 1 or 5 per cent level. More precisely, the effect obtained from Model 6 (where the full set of covariates is considered) suggests an average reduction of 0.138 tons of waste per capita for each one percentage point increase in the share of the primary sector, whereas everything else remains constant. Two main reasons can be called uponto explain this negative impact: the income effect and the preference effect. Income is relatively lower in the primary sector and less consumption contributes to lower waste production. The other argument is associated with the preference effect, especially in the agricultural sector where a part of the goods consumed is self-produced. The impact of the secondary sector ratio on waste generation is found to be positive and statistically significant at the 5 per cent level. Model 3 indicates that, on average, a one percentage point increase in the share of the secondary sector is responsible for 0.064 tons per capita increase in waste generation, everything else remaining unchanged. This is an expected result due to the income effect. Income in the secondary sector is higher, which translates into higher consumption patterns, therefore producing more waste. Nevertheless, an unexpected result is found in Model 4, i.e., the services sector share does not impact waste generation significantly. One plausible explanation of this outcome can be found in the high correlation between the secondary and the tertiary sector shares (−0.7031), as shown in table B1 of online appendix B. It is possible that the secondary sector captures the whole significance due to the existing complementarity between these two sectors. To avoid the effects of multicollinearity, a new interaction term is introduced in the waste equation defined as the product of the two sector shares and, as Model 5 highlights, it displays statistical significance at the 1 per cent level. The joint impact of the secondary and tertiary sectors indicates a 0.205 increase in waste generation (tons per capita), while everything else remains constant.

Elderly population (above 65 years) is also used as an explanatory factor in Model 6, which shows a negative significant effect at the highest 1 per cent level of significance. A reduction of 0.017 tons per capita in waste generation for every one percentage point increase in the share of elderly population is predictable, everything else being constant. This evidence is in line with previous findings in the literature, namely that the elderly population produces less waste as a result of lower consumption (see for instance Ichinose et al., Reference Ichinose, Yamamoto and Yoshida2015). The income effect and preference effect would justify the lower consumption pattern in this age group. On average, pensions in Portugal are low and do not guarantee comfortable living standards for the elderly, therefore limiting consumption. Additionally, older people show less appetite for consumer goods, as a consequence of different biological needs.

Population density, given by the average number of individuals per km², is an additional covariate in Model 6, and aims to measure the degree of urbanization of Portuguese municipalities. As can be seen from table 3, this variable does not make any important contribution to explaining waste generation, as it is statistically insignificant. Jaligot and Chenal (Reference Jaligot and Chenal2018) also found that population density is not relevant for explaining municipal waste in the canton of Vaud in Switzerland. One explanation of this irrelevance might be the fact that the economic indicator we use to test the decoupling hypothesis is expressed in per capita terms (GVApc), capturing therefore the population effect on waste generation per capita.

The dummy variable used to express the development dichotomy between municipalities is also not significant, although carrying the expected positive sign. Therefore, it is not possible to find different behavior of waste generation between the more developed and less developed municipalities, which is consistent with the decoupling evidence found before and verified in a restricted number of municipalities only.Footnote ¹⁷

Finally, the tourism inflows variable plays no role in explaining waste generation in the Portuguese municipalities, not confirming the findings in the literature that waste production increases with the increase in tourism flows, through higher consumption. As we explained before, we must interpret this result with caution, taking into account that data on this variable is incomplete and subject to different measurement methods applied over time.

Overall, the evidence obtained from estimating the equation of waste generation with fixed effects supports the decoupling hypothesis, although a few municipalities achieved this performance. The main drivers of waste generation are associated with the productive structure of the local economies and the behavior of the elderly population.

5.2 Evidence from the dynamic model

As mentioned before, it is important to test whether past realizations of municipal waste influence policies for controlling the current values of waste generation. To test this hypothesis, the lagged value of the waste per capita variable is introduced in equation (1), thus estimating a dynamic panel data model. However, a problem arises when estimating this type of model: the lagged dependent variable might be correlated with the error term, producing biased and inconsistent estimates, particularly if the sample has a small T and a large N (see Baum, Reference Baum2006). The u_i error component enters every value of the MWGpc_it by assumption, so the lagged dependent variable is not independent of the composite error term. To overcome this endogeneity problem, Arellano and Bond (Reference Arellano and Bond1991) suggested the GMM approach, which consists of taking the first variable differences (removing therefore the u_i term) and instrumenting the first-differenced lagged values of the dependent variable by using previous lagged levels (the second and third lags of MWGpc, for example). This technique specifies the model as a system of equations, one for each time period, and the equations differ only in the number of instruments used (the moment conditions).Footnote ¹⁸ The predetermined (lag-dependent variable) and other endogenous regressors (if any) in the first differences are instrumented with suitable lags of their own levels. Strictly exogenous regressors are also part of the set of the instruments. Nevertheless, a problem with the original Arellano-Bond estimator is that lagged levels are often shown to be poor instruments for first differences, especially for variables that are close to a random walk. In order to solve this problem, Arellano and Bover (Reference Arellano and Bover1995) and Blundell and Bond (Reference Blundell and Bond1998) suggested, in a later work, a modification of the original model, recommending the use of lagged levels and lagged differences as well. The original estimator is often called the difference GMM approach, whereas the augmented estimator is the system GMM. Both estimators have one-step and two-step variants, the latter being asymptotically more efficient, since a finite-sample correction is made to the standard errors of coefficients, as indicated by Windmeijer (Reference Windmeijer2005).Footnote ¹⁹

The dynamic version of the waste equation is estimated by using the GMM two-step difference system approach with robust standard errors (according to the Windmeijer (Reference Windmeijer2005) correction) and the results are reported in table 4. Model 1 includes the full set of explanatory variables considered in the fixed effects model (except the tertiary sector share to avoid the collinearity problem) and the lagged dependent variable of waste generation as an additional covariate. Model 2 is the same as Model 1, albeit excluding the secondary sector share. Model 3 replaces the secondary and tertiary sector shares with the product of the two (GVA2r* GVA3r) to test the statistical relevance of this interaction term.Footnote ²⁰ Model 4 adds the tourism ratio as a covariate as well, omitting the population density variable since it is statistically not significant. Model 5 is similar to Model 3, but introduces the dummy variable D ₂₀₁₅ to control for the methodological change to waste generation registration in 2015.

Table 4.

Results of the dynamic panel-data model, two-step difference GMM on municipal waste generation in Portugal, 2009–2018

Notes: Numbers in parentheses are robust standard errors according to the Windmeijer (Reference Windmeijer2005) correction. Numbers in square brackets are P values. *, **, *** indicate the statistical significance of coefficients at the 10 per cent, 5 per cent and 1 per cent significance level, respectively. The F-test assesses the joint significance of all slope coefficients, AR(2) is the Arellano-Bond test for second-order autocorrelation in first differences, the Hansen-test is used for testing over-identified restrictions (validating the instruments), and TP is turning point. We consider all the explanatory variables as exogenous, except the lagged dependent variable which is instrumented by its own lagged values, the lagged values of all other exogenous variables and the time trend variable.

A careful selection of the dynamic regression model is made, taking into consideration the diagnostic statistical tests that support the appropriateness of the estimated model. In particular, the F-statistic shows the joint significance of all slope coefficients, the AR(2) testFootnote ²¹ ensures the absence of second-order error autocorrelation, and the Hansen testFootnote ²² validates the exogeneity of the instruments used in the estimation approach for all models. Additionally, for higher robustness of the results, it is ensured that the number of instruments does not exceed the number of groups (municipalities) considered in the sample.

As can be seen, the lagged dependent variable is highly significant at the 1 per cent level, justifying that previous realizations of waste generation are important for explaining current values. The dynamic specification of the model supports the partial adjustment principle, suggesting that actual variation of waste is a fraction of the desired variation of waste, expressed as:

(2)\begin{equation}({MWGp{c_{it}} - MWGp{c_{it - }}_1} )= \theta ({MWGpc^{\ast}_{it} - MWGp{c_{it - }}_1} ),\end{equation}

where MWGpc*_it is the desired or optimal level of waste (the steady-state level) and θ the partial adjustment coefficient, which varies between 0 and 1. The closer θ is to zero, the slower the speed of adjustment; the closer θ is to one, the faster the speed of adjustment. The desired level of waste generation is unknown, but can be specified as follows:

(3)\begin{equation}MWGpc^{\ast}_{it} = {\beta _0} + {\beta _1}GVAp{c_{it}} + {\beta _2}GVApc_{it}^2 + \delta ^{\prime}{Z_{it}} + {u_i} + {\epsilon _{it}}\end{equation}

Substituting the long-run model (3) into the partial adjustment mechanism (equation (2)) and solving for MWGpc_it, the short-run model is derived, presented as:

(4)\begin{align}MWGp{c_{it}} &= \theta {\beta _0} + ({1 - \theta } )MWGp{c_{it - 1}} + \theta {\beta _1}GVAp{c_{it}}\notag\\ &\quad + \theta {\beta _2}GVApc_{it}^2 + \theta {\delta^{\prime} }{Z_{it}} + \theta ({u_i}+ {\epsilon _{it}})\end{align}

As can be seen from Models 1 and 2, the speed of adjustmentFootnote ²³ indicates that 62 per cent of the actual variation in waste generation is adjusted to its desirable level in the same year, which is a relatively fast adjustment. The speed of adjustment is even higher, around 63.4 per cent in Model 3 where the interaction term between the secondary and tertiary sectors is included. Evidence from Model 4, where tourism is an additional covariate, shows that the adjustment is even faster (around 70 per cent). The presence of the decoupling effect consistent with the wKc principle is again confirmed as in the fixed effects model, suggesting a non-linear relationship between waste generation and economic activity. The turning point is about 35 thousand euros of GVApc in Models 1 to 3, and lower in Model 4 (around 30) where tourism inflows are considered. Few municipalities achieve this threshold of gross production after which waste declines (Lisbon in the whole period, Sines in 2016–2017, Oeiras in 2009–2010 and Castro Verde in 2009–2011).

As in the fixed effects model, the primary production sector is found to negatively affect waste generation in all models, and it is statistically significant at the 5 and 10 per cent levels only. Considering the results of Model 3, it is predicted that in the short-run a one percentage point increase in the share of the primary sector is responsible for 0.19 tons per capita decrease in waste generation, while the long-run effectFootnote ²⁴ is even higher and equal to 0.29. Nevertheless, the secondary (in Model 1) and tertiary (in Model 2) sectors are not statistically relevant, and this may be due to the high correlation between the two variables, as explained before. To avoid the consequences of multicollinearity, we introduce an alternative interaction term (the product of both sectors) and, as Models 3 and 4 highlight, the joint impact turns out to be positive and statistically significant at the 10 and 5 per cent levels, respectively. These sectors contribute positively to the increase in municipal waste generation through higher patterns of consumption, by providing higher income to households.

Elderly population contributes negatively to waste generation and this impact is verified in all dynamic versions of the estimated model. This evidence is in conformity with the fixed effects model and the expectations from the literature that ageing people have lower consumption patterns due to less consumption needs and lower income. According to Model 3, the short-run effect predicts a 0.011 decrease in waste generation given a one percentage point increase in the share of the elderly group, while the effect in the long-run predicts a 0.017 fall in waste, ceteris paribus. As in the fixed effects model, population density and the production dichotomy variables are not statistically relevant. In turn, the impact of tourism inflows (over-night stay ratio) on waste generation is positive, as expected, and statistically significant at the 5 per cent level, but the marginal effect is small.Footnote ²⁵

A last issue we wish to address is whether the change in methodology of recording waste produced in the Portuguese municipalities affects the regression results.Footnote ²⁶ As explained in table 1, since 2015 the values represent managed waste instead of collected waste. One way to test this hypothesis is to introduce in the model a dummy variable D ₂₀₁₅ that takes the value of one in the year this change occurred and zero otherwise, and to check its relevance. Model 5 of table 4 reports the results of this regression, controlling for this potential break in the series. It is comparable to Model 3, since both have the same set of other covariates. It is shown that despite the significance of the dummy variable in Model 5, the results are very similar to Model 3 where this variable is absent, both in terms of the statistical significance of the covariates and their marginal impacts on waste. It is therefore confirmed that the methodological change in waste recording does not affect the robustness of the results of the determinants explaining waste generation in the Portuguese municipalities.

6. Concluding remarks

This paper aims to contribute to the waste literature that analyzes the decoupling phenomenon at the municipal level, where studies are still limited. Data from the Portuguese municipalities is used for the first time to evaluate the non-linear relationship between waste generation and economic activity, consistent with the wKc and showing an inverted U-shaped trend. Other socioeconomic factors related to the productive structure of the local economies, such as population age structure, population density, production dichotomy and tourism, are considered as driving forces to explain waste generation in the Portuguese municipalities.

To achieve these goals, panel data models are estimated assuming fixed effects and dynamic specifications, encompassing 307 municipalities over the 2009–2018 period. A careful selection of the regression models is made, aimed at ensuring the robustness of the results. The fixed effects estimation technique takes into account the heterogeneity among the Portuguese municipalities, ensuring the efficiency of estimates with the use of robust standard errors. The GMM system approach is employed with first differences in variables to estimate the dynamic models, as recommended by Arellano and Bond (Reference Arellano and Bond1991) to solve the endogeneity problem of the lagged dependent variable. The dynamic model specification is in line with the partial adjustment mechanism that allows one to distinguish the short-run and long-run effects of the covariates and the adjustment speed of the actual variation in waste generation to its desired level. Results obtained from the fixed effects and dynamic specifications are coherent with each other and in line with the findings from the literature.

The empirical analysis shows that the decoupling hypothesis is confirmed through all estimated models. However, only a few municipalities (Lisbon and three others) achieve the threshold from which waste generation declines as economic activity expands. From the policy point of view, this result suggests that more effective waste management policies are needed to prevent waste generation, pointing to enhance environmental protection. The high value of the turning point highlights the need for more development of the local economies and better waste management in order to reduce waste generation and therefore preserve the environment. Technological advances and higher educational standards can help to achieve these goals.Footnote ²⁷

The productive structure of the local economies is shown to be important in explaining waste generation performance. The fixed effects model suggests that waste production is adversely affected by the share of the primary sector, apparently explained by lower income and consumption patterns in this sector. The combined shares of the secondary and tertiary sectors impact municipal waste generation positively, presumably through higher income and higher consumption patterns. The complementarity between the two sectors may explain the joint effect, which is statistically significant, in contrast to the individual effect which is not relevant.

Among other drivers, the population aged 65 or over is found to contribute negatively to waste generation in the Portuguese municipalities, corroborating other findings in the literature that hint at lower consumption in this age group due to lower consumption needs. Population density reflecting the urbanization stage of the local economies is not important to the formation of municipal waste, nor is the development dichotomy distinguishing municipalities with gross production above or below the average. Tourism inflows exhibit a positive and significant effect on waste formation in the dynamic model only. This is in line with findings from the literature that highlight growing local population due to tourism, which contributes to more municipal waste production through consumption. However, its impact on waste is of a low size. Finally, the dynamic model provides evidence that waste generation in the past contributes to current levels of waste produced and that the process of adjusting actual waste generation to the desired level is relatively fast.

By exploring the driving forces behind waste generation, this study provides policymakers with a valuable tool for designing waste management strategies towards prevention and reuse. This type of analysis can be particularly useful for developing countries, where waste management policy framework is still at a very early stage of development. Although the current study refers to a developed country, some hints could be offered to developing countries as well. Waste generation is a dynamic process that requires the identification and implementation of effective waste prevention policies. As our study shows, the volume of waste depends on the structure of the economy and as production shifts away from the primary sector, waste production tends to increase (through the income effect), thus challenging the waste management system. Population in developing countries is younger, so higher consumption patterns are to be expected as the standard of living improves, which could create additional problems for the waste generation process. The growth policy based on tourism promotion, while bringing important revenue to these countries, has environmental costs that put pressure on the treatment of extra waste produced. The decoupling effect is expected to be achieved at higher levels of development, since the disassociation between waste generation and economic activity will imply the development of efficient waste management systems, which involves high costs and technology-driven processes. New studies on waste generation performance should be conducted taking into account the specific characteristics of the developing countries. The existing literature provides little empirical evidence from these countries, due to the lack of statistical information on waste generation.