This research adopts a case study perspective considering alternative investments in emissions abatement technologies in a nonferrous metal factory, to evaluate their cost efficiency through the NPV obtained in a CBA. Since the assessment of environmental impacts needs to be placed in a specific geographical or local context and related to a specific manufacturing activity, this study is focused on the production of aluminium in a fictitious facility placed in Spain. The case study is built on the experience acquired by the authors in the development of different research projects in industrial plants to assess the feasibility of environmental investments. Technical and economic data were defined with the advice of experts and managers of similar facilities and considering the descriptions of industrial processes contained in the related BREF document (Cusano et al., 2017). The plant does not currently meet the BAT-AELs associated with BAT 60 concerning fluoride and dust (particulate matter or PM) emissions. To renew its IED permit conditions, installation of a dry scrubber is required as BAT to replace its current wet scrubber at a specific emission point. The high CAPEX associated with this investment motivates the consideration of alternative measures to achieve a similar or higher environmental performance, applying for derogation. These alternative measures involve retrofitting the plant to use natural gas instead of fuel oil and installing starbags in other plant sections. This process enables reductions in emissions of fluoride, PM, and other pollutants (SO2, NOX, CO2), similar or higher than those that would be obtained by installing the dry scrubber, allowing for compliance of the BAT-AEL in global computation in the whole plant and at lower investment and operation and maintenance costs. Therefore, this is a potential derogation to the IED that guarantees a high level of protection of the environment as a whole.

Appendix A reflects the technical and economic detailed data related to the expected effect (reduction) on air emissions and the private costs (capital and operating expenditures) of the two alternative compliance options considered. The next section and Appendix B provide more detailed information on how the potential values of the other parameters (WACC, SDR, damage costs and uplift factor) involved in the CBA of the two alternatives and the Monte Carlo simulations were defined. These parameters are, then, presented as conditions for the fsQCA in the last section, where the alternative models to test the effect of these conditions on the defined outcome (SNPV) will be specified.

The analysis of investment decisions related to environmental assets has traditionally been approached through the CBA methodological framework. CBA is considered a relevant tool in environmental public and private decision-making (Alam, 2008; Feuillette et al., 2016; Pearce, Atkinson & Mourato, 2006) because it sets the decision to spend scarce public or private financial resources under a social accounting framework (Cordes, 2017). It has been applied to many different fields (Pascal et al., 2018; Paulrud & Laitila, 2013; Piñeiro-Chousa, López-Cabarcos, Romero-Castro & Vázquez-Rodríguez, 2021), but its presence in the scientific literature related to IED derogations is practically non-existing.

In a CBA on environmental investments, all the relevant private and social benefits and costs should be estimated and included in the analysis, requiring the valuation of environmental goods or services that have no market value (Pascal et al., 2018; Pearce et al., 2006). To conduct this study, various CBA frameworks were taken as a reference, some specifically related to the IED and the analysis of the proposals for derogations to the IED (EA (Environment Agency), 2017; EPA (Environment Protection Agency), 2016; European Commission, 2006, 2015; HM Treasury, 2018). The economic valuation of the different abatement alternatives includes:

- The private or internal costs or benefits: those that have a direct impact on the company, including the investment costs in the technologies or assets that make it possible to comply with the BAT-AELs (costs of acquisition of equipment, installation and start-up, opportunity costs, financing costs, etc.), and operation and maintenance costs (personnel, materials consumption, energy, etc.) in incremental terms compared to the business-as-usual (BAU) situation (higher or lower costs), in addition to any additional benefits (savings or higher income) that may derive from the options considered. Appendix A explains how the private costs and benefits of the two alternative compliance options considered were defined based on data from experts and Cusano et al. (2017).

- The social or external costs or benefits: specifically, those focused on assessing the impact on the environment and health of changes in emissions into the atmosphere derived from the options considered compared to the BAU situation. Estimates of the damage costs associated with the emissions of various types of pollutants (SO2, NOX, fluoride, particles and CO2) have been generated from various sources of proven quality and solvency (de Bruyn et al., 2010; EA (Environment Agency), 2017; EEA (European Environment Agency), 2014; Institute of Energy Economics and Rational Energy Use (University of Stuttgart), 2020). Appendix B provides further explanations on how these values were derived from these sources, distinguishing between central values, applied in the base case CBA, and minimum and maximum values that are considered to create uniform probability distributions for each pollutant damage cost to be introduced in the Monte Carlo simulation tool.

With all the estimations on benefits (B), costs (C), and investments (I), the NPV of the two considered compliance options (BAT and derogation) was calculated as the discounted value of the annual cash flows (CF) at the SDR (Eq. (1)). In addition to other private and social costs, ΔCt includes the private financial costs (which depend on the WACC), while both ΔCt and ΔBt include the social costs or benefits derived from changes in air emissions, which depend on the uplift factor and the value of the damage costs.

For the discounting of the CFs, a real SDR of 5% was considered following Campos, Serebrisky and Suárez-Alemán (2015), who reported a 4%−6% range for the SDR employed by the Spanish government in the evaluation of investment projects across different sectors. Accounting for the debate on the convenience of applying higher or lower SDRs, a range of 2.5%−7.5% was set to perform the simulation analysis.

Table 1 summarizes all the estimates and parameters considered to perform the CBA of the BAT and derogation options.

All costs and benefits (investment, operation, and external) are presented in real terms (€2020) without considering the effect of inflation. Consequently, the appropriate interest rates to determine both the financial costs of investments and the present value of future costs or benefits are also rates in real terms.

After obtaining a first NPV calculation for each compliance option based on the central values considered for the damage costs, WACC, SDR and uplift factor, two Monte Carlo analyzes with 1000 iterations were run with Crystal Ball Excel add-in to create the cases to be analyzed through fsQCA. Monte Carlo simulations are usually applied to perform a risk assessment on the calculation of a project's NPV (Deng et al., 2013). Taking the surplus NPV (SNPV in Eq. (2)) as the target result, a first simulation (SNPV1) considered SDR, WACC, uplift factor and a multiplying factor as uncertain variables that move all the damage costs between their minimum and maximum values. A second simulation (SNPV2) considers as uncertain variables only the six damage costs involved in the analysis (HF, CO2, CO2-traded, PM, NOX-PM and SO2), defining uniform probability distributions around their respective minimum and maximum values. Each simulation provided 1000 scenarios that are treated as observations or cases in alternative fsQCA model specifications.

Charles Ragin (1987, 2000, 2008) proposed qualitative comparative analysis (QCA), based on Boolean algebra, to identify combinations of causal conditions (configurations or paths) that are necessary or sufficient to explain a specific level of an outcome (Fiss, 2007, 2011). Although QCA was originally proposed to conduct small-N studies, its application to larger N studies is widely acknowledged (Fiss, 2011; Ott, Williams, Saker & Staley, 2019; Slager, 2012) and has contributed to its increased use in management research (Roig-Tierno, Gonzalez-Cruz & Llopis-Martinez, 2017). fsQCA has been applied to the study of very different areas (Damian & Manea, 2019; Kusa, Duda, & Suder, 2021; Maggetti, 2014; Muñoz & Cohen, 2017; Ott et al., 2019), but to the best of our knowledge, it has not been applied in the analysis of environmental investment decisions yet.

As shown in Table 1, the parameters considered in the CBA as more subject to uncertainty and discretionary choices are taken as conditions to analyze their relationship with the SNPV through fsQCA. These conditions are WACC, SDR, damage costs’ uplift factor and the damage cost estimates related to six types of emissions: hydrogen fluoride (HF), nitrogen oxide free of particulate matter (NOX-PM), particulate matter (PM), sulphur dioxide (SO2) and carbon dioxide, distinguishing between its shadow price (CO2) according to The World Bank estimates, and the market price of CO2 emission allowances (CO2-t).

To assess the degree of membership of each observation or case in each set of conditions and outcomes, a calibration process is conducted based on continuous fuzzy sets with membership scores ranging from 0.0 to 1.0 (Ragin, 2008; Schmitt, Grawe & Woodside, 2017). Emmenegger, Schraff and Walter (2014) suggest that calibration with percentiles can be an adequate strategy for continuous variables, while for categorical data calibration should be based on theoretical and contextual knowledge. Given that all conditions and outcomes are continuous variables in this study, calibration is performed with the help of the calibration function provided by fsQCA 3.0 software (Ragin & Davey, 2017), using the 75th, 25th and 50th percentiles (Muñoz & Cohen, 2017; Tóth, Thiesbrummel, Henneberg & Naudé, 2015) to determine for full membership (score of 1), full nonmembership (score of 0) and the crossover or point of maximum ambiguity (score of 0.5), respectively. Pappas and Woodside (2021) consider that the choice of direct calibration leads to more rigorous studies which are easier to be replicated and validated.

To decide which paths to include in the final fsQCA solution (Skarmeas, Leonidou & Saridakis, 2014), this study sets the recommended cutoff consistency to 0.8 (Ragin, 2008) and a minimum frequency of 5 cases to identify sufficiency solutions using the truth table algorithm of the fsQCA 3.0 software. This frequency threshold ensures that after deleting all causal combinations where there are fewer than 5 cases, at least 80% of all cases were retained as recommended by Ragin (2008). A condition is considered necessary when its consistency value was above 0.9 (Fiss, 2007; Ragin, 2008; Schneider & Wagemann, 2012).

A solution consistency above 0.75 is usually the minimum acceptable value (Ragin, 2008). The exploratory nature of this study and the high uncertainty affecting the causal conditions considered advice against making simplifying assumptions to create parsimonious and intermediate solutions based on easy and difficult counterfactuals, so the complex solution was chosen to report the results (Fotiadis, Yeh & Huan, 2016; Skarmeas et al., 2014).

Following Slager (2012), necessity and sufficiency analyzes were developed both concerning the presence or absence of a high SNPV, respectively identified as SNPV and ∼SNPV, where “∼” represents the absence of a condition or outcome. The two simulations create two alternative sets of cases, so the necessity and sufficiency analyzes were performed four times on four outcomes through the fsQCA 3.0 estimation software Ragin & Davey, 2017). Regarding the sufficiency analysis, Eqs. (3) and ((4) correspond to Models I and II concerning the presence or absence of a high SNPV1, and Eqs. (5) and (6) correspond to models III and IV concerning the presence or absence of a high SNPV2. Letters “fs” indicate that the fuzzy-set calibrated values of both outcomes and conditions are considered.

Accounting for the private and social costs and benefits of the BAT investment and the derogation alternative, the CBA revealed a negative NPV (−13,897,043.77 €) for the BAT option and a positive value (25,233,802.13 €) for the derogation option, resulting in an SNPV of 39,130,845.90 €. Table 2 summarizes the main components of these results. Therefore, the BAT compliance option shows disproportionate costs that could justify the concession of a derogation based on alternative investments that derive higher social benefits at lower private costs.

Building on this base scenario, a first Monte Carlo simulation with 1000 trials was performed considering four uncertain variables: WACC, SDR, uplift factor and a multiplying factor that ranged between 40% and 160% and was applied to the central values estimated for the damage costs associated with air emissions. The SNPV (SNPV1) was always positive, ranging between a minimum value of 25,146,593.70 € and a maximum of 61,040,079.66 €. The second Monte Carlo simulation considers the six damage cost estimates as uncertain variables. The SPNV (SNPV2) was also always positive, yet the range revealed by the 1000 iterations was narrower (29,416,299.47 €–48,847,975.91 €) since the values of the WACC, SDR and uplift factor remain constant in their base estimates in this case. Appendix C includes a detailed summary of results from the two simulations performed, also showing the 25th, 50th and 75th percentiles applied in the calibration of the variables through the fsQCA 3.0 calibration function and descriptive statistical data on the calibrated variables.

After calibrating the data, the analytical phase is entered, conducting a necessity and a sufficiency analysis (Maggetti, 2014). The necessity analysis on the four conditions considered in the first simulation (Table 3) revealed that none of them was necessary for either the presence or the absence of a high SNPV1, since none reaches a consistency of 0.9.

Tables 4 and 5 show the results of the complex solutions from the sufficiency analysis on the presence (Model I, Eq. (3)) and absence (Model II, Eq. (4)) of a high SNPV1. Solution consistency and coverage were high in both models: respectively, 0.900 and 0.725 in Model I, and 0.911 and 0.707 in Model II. Three paths were included in each model's solution, and they were mirror opposites. This concept was quite straightforward since the outcome was related to the conditions through a mathematical model. Thus, the results are jointly analyzed for the two models’ solutions. It is concluded that the more cost-efficient and environmentally sound derogation option can be the most valued option (Model I, high SNPV1, Table 4) under different combinations of conditions: in path 1 (exhibiting the higher consistency and raw coverage) higher damage costs that consider more pathways of impacts on health and ecosystems were combined with lower SDRs that imply a lower discounting of environmental benefits in the future; in path 2 (consistency 0.898, raw coverage 0.309) higher damage costs were not needed since the combination of a high uplift factor with a low SDR and a high WACC (that penalizes the more capital intensive BAT option) also favour the derogation alternative; and in path 3 (consistency 0.923, raw coverage 0.305) a low SDR was not necessary since the derogation option was favoured by the combined effect of applying in the CBA higher damage costs, uplift factor and WACC.

The mirror opposite conclusions can be derived from the analysis of the results for Model II (low SNPV1, Table 5).

Moving to the second set of cases derived from the Monte Carlo simulation, that considers the six damage cost estimates as uncertain variables to derive alternative SNPV2 calculations, the necessity analysis (Table 6) again shows that no condition was necessary to produce either the presence or absence of a high SNPV2, since none exceeds a consistency of 0.9. However, it is worth noting that the highest consistency value corresponds to the damage cost for SO2, with a higher value favouring the derogation option (consistency of 0.860 for the presence of a high SO2 related to the presence of a high SNPV2) and a lower value favouring the BAT option (consistency of 0.844 for the absence of a high SO2 related to the absence of a high SNPV2). This is due to the increase in SO2 emissions associated with the BAT options, which involves the substitution of a wet scrubber to install the dry scrubber.

Tables 7 and 8 show the results of the complex solutions from the sufficiency analysis on the presence (Model III, Eq. (5)) and absence (Model IV, Eq. (6)) of a high SNPV2, with high consistency and coverage in both models: respectively, 0.904 and 0.804 in Model III, and 0.885 and 0.841 in Model IV. The alternative paths included in both solutions revealed the important role played by the damage cost of SO2, as derived from the necessity analysis, so it seems to be a ‘quasi’ necessary condition since it is present in all the combinations of conditions leading to a high SNPV2 and four of the six leading to a low SNPV2. To favour the derogation option (Model III, Table 7), a high SO2 damage cost can be combined with either a low PM in path 1 (since PM reductions were higher in the BAT option), a high shadow price of CO2 in path 2 or a high NOX damage cost in path 3 (since the BAT alternative does not affect CO2 and NOX emissions). The three paths show high raw coverages above 0.5 (0.531 for path 1, 0.522 for path 2 and 0.509 for path 3). It is interesting to note that the HF damage cost does not appear in any configuration favouring the derogation option since both compliance alternatives derive similar effects on HF emissions.

The analysis of the conditions leading to a low SNPV2 (Model IV, Table 8) revealed 6 different paths that involve low damage cost estimates for all the emissions considered except for PM (higher reductions in the BAT option benefited from a higher value of the associated damage cost). Both high and low damage costs of HF and CO2 emission allowances (CO2t) are present in different paths. Paths 1, 2 and 3 are the natural mirrors opposite of the three paths in Model III, showing similar raw coverages above 0.5. Paths 4, 5 and 6 include more sophisticated configurations (3 or 5 conditions involved) but show lower raw coverages. Path 4 (consistency 0.906, coverage 0.285) indicates that low damage cost estimates for SO2, HF and CO2 emission allowances favour the BAT option, since it generates an increase in SO2 emissions and a lower HF reduction compared with the derogation alternative, and it does not have any significant impact on the private costs of the plant (purchase of CO2 emission allowances). Path 5 (consistency 0.877, raw coverage 0.120) also involves a low price of CO2 emission allowances and a low HF damage cost estimate, but instead of combining them with a low SO2 damage cost, it shows high PM, low NOX and low CO2 damage costs. Path 6 (consistency 0.876, raw coverage 0.115) may be more interesting since it combines high values of the HF damage cost and CO2 emission allowance price, showing that although the BAT option generates lower HF reductions and no relevant impact on CO2 emissions that could lower the associated private costs (purchase of emission allowances), the higher reduction in PM compensates for the reductions of CO2 and NOX of the derogation option thanks to a low valuation of their associated damage costs.

As a robustness test, an alternative calibration method with percentiles 90th, 10th and 50th was tried, yielding the same configurations or paths and similar consistency and coverage values.