Science policy frameworks for a post-pandemic green economic recovery

As current production and consumption patterns exceed planetary boundaries, many leaders have stressed the need to adopt green economic stimulus policies in the aftermath of the COVID-19 pandemic. This paper provides an integrated multi-stakeholder framework to design an economic recovery strategy aligned with climate stabilisation objectives. We first employ quantitative energy and economic models, and then a multi-criteria decision process in which we engage social actors from government, enterprises and civil society. As a case study, we select green recovery measures that are relevant for a European Union country and assess their appropriateness with numerous criteria related to climate resilience and socio-economic sustainability. Results highlight trade-offs between immediate and long-run effects, economic and environmental objectives, and expert evidence and societal priorities. Importantly, we find that a ‘return-to-normal’ economic stimulus is environmentally unsustainable and economically inferior to most green recovery schemes.


Introduction
Since mid-2020, despite the persistence of the COVID-19 pandemic, the response of governments around the world has partly moved from the provision of immediate relief to the design and implementation of economic recovery measures for the short and medium term. Leaders of international organisations have stressed the importance of adopting green economic stimulus policies in line with the United Nations Sustainable Development Goals (UN SDGs) and the Paris agreement on Climate Change, as greener economies are more resilient to climate change, social unrest, and epidemics [1][2][3][4]. Global economic support for relief and recovery from the pandemic has risen to significant levels since spring 2020 -but as regards the conformity of such stimulus measures with climate compatible growth, the picture is mixed [5,6].
The lockdown measures to contain spread of the pandemic led to a serious economic downturn in Cyprus, with GDP contracting by 11.9% in the second quarter of 2020 [7]. As in other world regions, economic stimulus packages had to be designed for rapid implementation and to contribute to positive growth and employment impacts in the short term, keeping in mind long-term development and decarbonisation objectives, such as the EU's 2030 energy and climate goals and the envisioned carbon neutrality for 2050, as foreseen in the 'European Green Deal' [8].
In this paper we develop a novel integrated assessment framework for the design of an economic recovery strategy that could have promising impacts on climate resilience and socio-economic sustainability. This involves multi-criteria decision analysis, which incorporates both quantitative data derived from models and qualitative input provided by several stakeholders (Fig. 1). The adopted quantitative models are opensource and transparent, which allows for their rapid adaptation to the needs of policymakers in a specific country; the appropriateness of such models for rapid policy formulation has been demonstrated very recently [9]. The use of qualitative input is not only necessary because models cannot adequately simulate all possible impacts; it is also essential for increasing the likelihood of social acceptance of the proposed recovery interventions, by reducing dependence on knowledge silos of academic experts or policy makers. This is in line with the need for broader mobilization of society for the transition to sustainability [10]. Although we applied this approach in the case of Cyprus, an EU member state in the Eastern Mediterranean, the framework has general application, and the underlying tools and processes can be adopted in other national or regional contexts.
This science-policy framework involves regular interaction with societal actors, which can help highlight the importance of thinking beyond purely short-term recovery measures and considering investments and reforms that may take time to materialise, but are essential for meeting medium-and long-term climate objectives [11]. A short-term recovery plan, no matter how green it is, cannot deliver the low-carbon transition by itself; it has to be complemented by structural reforms that can deliver environmental and economic benefits over the longer term [12]. Our approach highlighted this aspect to national stakeholders, some of whom were understandably interested in mitigating the immediate impacts of the pandemic and paid less attention to long-term green reforms. Based on quantitative evidence which is shown in this paper, we also emphasised the mediocre performance of a 'return-to-normal' economic stimulus, not only in environmental but also in economic terms. To encourage ownership of the measures by national decision makers we built on policies and measures that are largely based on existing national plans so that stakeholders are familiar with such interventions. This increases the likelihood for adoption of these measures in the national recovery strategy.
This novel combination of a) building on familiar national plans, b) employing open-source models that can be rapidly adapted to national particularities, c) comparing the performance of green versus businessas-usual stimulus plans and d) obtaining immediate qualitative input from national stakeholders can serve as a model for many countries around the world, which have requested assistance from international organisations on designing recovery packages during late 2020 and early 2021. Section 2 of the paper describes the main stages of this approach, the recovery measures that were considered, and the sustainability criteria used for assessing these measures. Section 3 presents the results of the energy and economic models employed, while it also provides an overview of the multi-criteria assessment, which makes use of input from a targeted group of stakeholders. Finally, Section 4 outlines the lessons learnt from this study and their relevance for science-policy interactions in other national contexts.

Workflow
Post-pandemic stimulus packages have to promote short-term economic growth without compromising long-term energy and climate goals. To address these multiple requirements, it was necessary to assess promising green economic recovery measures that had already been identified in the public discourse.
We started from existing plans announced by the Finance Minister in May 2020 as well as from measures included in the National Energy and Climate Plan (NECP) submitted to the European Commission in January 2020 [13]. Similarly, if the framework is applied in a non-EU country, the approach might begin with a breakdown of the measures included in a country's Nationally Determined Contribution (NDC) to the UN Framework Convention on Climate Change. In summary, the approach involves proposing interventions which expand measures already announced or planned by national authorities; receiving feedback from stakeholders to obtain a first reality check; assessing measures with proper criteria that account for multiple sustainability objectives in the short (up to 2022) and long term (up to 2030); and prioritising measures on the basis of this assessment, considering budget availability.
Above all, it was important to ensure active participation of decision makers in the process, by using transparent workflows. This enables Ministry officers to provide input at various stages of this work, and ensures overall process accountability. In so doing we tend towards U4RIA guidelines. U4RIA is an acronym for Ubuntu, 1 Retrievability, Reusability, Repeatability, Re-constructability, Interoperability and Auditability. It aims to further good governance and sound scientific principles to energy modelling for policy support [14].
Specifically, the workflow consisted of the following stages: 1. Screening and preliminary assessment of potential green stimulus measures. In April 2020 we published a policy brief, alerting policymakers on the need to ensure that economic stimulus measures would enable the green transition [15]. We identified a first list of measures that could be implemented quickly, with positive employment and environmental impacts.

Dissemination of the list of measures and stakeholders' feedback.
The initial list of green recovery measures was circulated to Ministries of Finance, Environment, Energy and Transport, European Commission officials, NGO representatives and academics. Stakeholders focused on the need to ensure low administrative burden to allow fast implementation, and the importance of aligning the proposals with measures announced by the national government. 3. Interaction with national business associations. In mid-May 2020, after the first wave of the pandemic had been contained and public discussions focused on the 'return to normality', the national Federation of Employers and Industrialists (OEB) set up a working group on the green restart of the economy, in which we participated. OEB  After the model-based energy and economic assessment of the measures was completed, qualitative assessments followed for the rest of the criteria. Stakeholder input was sought, and a specific workshop was organised with representatives of different governmental departments, the private sector, and NGOs. Each stakeholder provided a score of each recovery measure by criterion and a weight for all criteria. 8. Prioritisation of measures. Stakeholder data were processed and, along with model-based results, provided a final ranking of the proposed interventions. These were communicated to all decision makers (DMs). Governmental authorities are able to use these to prioritise recovery measures.

Energy modelling
The assessment of the effect of green interventions on energy use and air emissions was carried out with OSeMOSYS, a long-term cost-optimisation energy systems model [18]. Some of the input to OSeMOSYS is provided by a separate energy forecast model [19], which projects final energy consumption across the economy and the related energy expenditures of households and businesses. The application of OSeMOSYS for the energy system of Cyprus is described in detail in previous studies [17,20].
For this specific study, OSeMOSYS was used to quantify impacts related to the implementation of existing Sustainable Urban Mobility Plans (SUMPs) (M8), the construction of the tram line in the capital city of Nicosia (M9), the scrappage scheme for old cars to be replaced with battery electric vehicles (M10), and the gradual implementation of a fiscally neutral carbon taxation system for sectors that do not fall within the EU Emissions Trading System (ETS) (M13). The first two of these measures are included in the country's National Energy and Climate Plan [13], while the latter two interventions have already been under consideration by the government. The model output focused only on benefits related to energy savings and carbon emission reductions; improvements in air quality and congestion, which are essential benefits of sustainable mobility measures, were not considered explicitly but were included in a qualitative manner in the criterion 'other environmental impacts'. Implementation of SUMPs entails an effort to achieve a considerable shift away from private vehicles to sustainable modes of transport. The techno-economic characteristics for each technology option in these modes is available in the existing base literature [21]. The development of the tram line in Nicosia, which is planned to come into operation in 2028, is closely related to SUMPs; that project will further enable the adoption of sustainable mobility and is assumed to reduce the annual mileage of private passenger cars and motorcycles by 540 and 16 million vehicle-kilometres respectively. According to the feasibility study conducted by national authorities, the tram line will have an upfront cost of approximately €225 million and annual operation and maintenance costs of €12 million [22].
The third measure that relates to the transport sector is the scrappage of old passenger vehicles and their replacement with battery electric vehicles. For a maximum replacement rate of 200 vehicles per year with a grant of €5000 per vehicle, this measure will require public funds of €1 million annually. Overall, it is assumed that 400 vehicles will be replaced through this scheme. Finally, a carbon tax on fuels for sectors that do not fall within the EU ETS is assumed to be implemented gradually and reach €120 per tonne of CO 2 by 2025. This will encourage adoption of energy efficiency measures and increase the attractiveness of low-carbon technologies, such as heat pumps in the heating and cooling sector and electric vehicles in the transport sector.

Economic modelling
Input-Output (IO) analysis is a quantitative technique for studying the interdependence of production sectors in an economy over a stated time period [23,24]. Here we applied a continuous demand-driven IO model with disequilibrium adjustment processes to assess the economy-wide effects of the selected energy-related economic recovery measures. Projected annual expenditures, including capital investments and operation and maintenance costs, from the OSeMOSYS model are introduced to the IO model to reflect changes in the investment demand of economic sectors as a result of each of the measures. Expenditures are classified in seven categories: industrial equipment, power generation technologies, electricity storage technologies, gas infrastructure, public transport, private transport, and buildings (including energy efficiency measures, heat pumps, solar water heaters etc.). Projected annual energy consumption expenditure of households is introduced to the IO model to estimate the multiplier effect of changes in private consumption. Tables in Appendix C present the distribution of annual spending associated with investments and private consumption by sector of economic activity for each recovery measure. The shares of spending for the development and operation of the recovery measures to the individual economic sectors have been allocated based on information obtained from relevant literature [25,26]. Taliotis et al. [17] describe in detail the assumptions behind the distribution of spending for each measure.
The initial static equilibrium conditions of the IO model, which serve as the reference case, are based on the latest available national symmetric IO table of Cyprus for the year 2016. The national table, which includes 65 sectors of economic activity, was aggregated into 20 economic sectors and is presented in Appendix C. The demand growth rates for the economic sectors are defined based on the GDP projections for the period up to 2030, including the impact of COVID-19 pandemic, and were obtained from the Ministry of Finance. For 2020 we assumed a decline in the growth of the Cypriot economy (− 7.4%) across all economic sectors, and for 2021 a strong economy-wide recovery (+6.1%). From 2023 onwards, growth rates return to usual levels following the official national macroeconomic outlook. Some of the recovery measures listed in Table 1 involve energy savings and hence induce a decrease in private consumption for energy, traded products, and services. We assumed that this reduction of spending, after accounting for household savings (we assumed a household saving rate of 2.4% of disposable income in Cyprus, in line with Eurostat [27]), will return to the economy and induce a rise in consumer demand for goods and services in line with the current consumption expenditure of Cypriot households [28].
Finally, apart from the thirteen selected green recovery measures, we explored the macroeconomic effects of a counterfactual scenario, which would be to provide uniform economy-wide demand stimulus. In this scenario, we aggregated the capital investments of all measures, amounting to 395 million Euros'2020 up to 2022 as shown in Table 1, and reallocated them in the economy according to the current sectoral shares of final demand.

Overview of the assessment framework
Multi-criteria decision analysis (MCDA) has been developed to support DMs, especially when facing decisions involving multiple and potentially competing objectives [29]. In the last decades, several methods have been developed for many types of decision problems. MCDA techniques have been widely applied in a variety of fields, including energy and environment [30][31][32][33][34]. Two of the most common methods used are the Analytical Hierarchy Process (AHP) method and the Preference Ranking Organisation Method for Enriching Evaluation (PROMETHEE) [35,36]. There are several examples of applications of AHP and PROMETHEE in the fields of energy planning, selection of energy projects, and sustainable supply chains [37][38][39]. PROMETHEE has also been used in the development and evaluation of scenarios for energy planning [40,41] and for evaluating market opportunities for renewables [42].
AHP and PROMETHEE can be combined, as shown by several studies [43][44][45]. AHP can be used to produce the weights of each criterion for each DM, which would be used as an input for applying PROMETHEE to produce the ranking of the actions. A similar framework has been developed in Matlab® for this paper, using a PROMETHEE Group Decision Support System (GDSS) approach.
Appendix D provides technical information about the application of these methods. In summary, AHP is a pairwise comparison method which uses a ratio scale that does not require any units. DMs express their preferences for one alternative over another one, using a 1-9 scale, which is assumed to offer the appropriate flexibility. In the framework of this study the 23 criteria shown in Appendix B have been divided into two broad categories of short-and long-term impacts and further subdivided into two subcategories, namely: (i) environmental criteria, and (ii) economic/social criteria.
For the evaluation and ranking of the alternatives the PROMETHEE method has been applied. The independent experts were asked to provide a score of each alternative recovery measure (or action according to PROMETHEE terminology) for each criterion in a typical 1-5 scale ranging from 'very low' to 'very high' impact. As several DMs provided input, the PROMETHEE GDSS was then implemented to combine the scores of individual DMs and produce a global evaluation that leads to the final ranking of measures.

Stakeholder input
A variety of stakeholders were invited to act as DMs and provide input. The group of DMs consisted of ten stakeholders: three economic planning officers from the Finance Ministry, one tax officer from the Finance Ministry, one officer from the Ministry of Energy, one officer from the Ministry of Transport, one from an energy NGO, one from an environmental NGO, and two from the national Federation of Employers representing the private sector. At the workshopthe participants were informed in detail about the list of recovery measures and the evaluation criteria, and were then provided with the respective tables to fill in (Appendix E). Weighting of the different criteria according to each DM's preferences, in line with the AHP method, was carried out through a piecewise comparison between each criterion. All input required for the application of the AHP method was checked for consistency. In all cases, the consistency ratio was calculated and was found to lie below the threshold of 10%, so that the input of all DMs was considered to be consistent.
Next, the PROMETHEE II method was applied, in which each DM evaluated the performance of all green interventions with a score in the scale 1-5, indicating an evaluation ranging from 'very low' to 'very high'. Scores for each recovery measure by criterion, in line with this method, were provided by each stakeholder. It should be noted that the scores for four criteria (energy savings, carbon emission savings, economic multiplier, and new jobs created) have been calculated through simulations with the relevant aforementioned models, therefore stakeholders could not change these scores.
Before arriving at the final results, it was necessary to calculate the net flow from the input of each stakeholder. The final step was the application of PROMETHEE GDSS to calculate the global ranking of all measures, assuming that each one of the ten stakeholders has equal weight. Obviously, the ranking of alternatives is affected by both the weights assigned by each DM and their respective scoring by measure and criterion.

Sensitivity analysis
The PROMETHEE method provides the ranking among a set of alternatives by considering the preferences of individual DMs, which is a result of their respective evaluations and weights. Considering that the weights have a significant impact on the final ranking, four alternative scenarios were examined. First, an analysis was carried out where the values assigned to the weights for each criterion varied for each DM (scenario 1). Given the large size of this problem, this analysis was carried out by implementing an iterative approach where the weights follow a uniform distribution in the interval [0,1] in 1000 iterations. Second, a similar approach was used where the values assigned to the weights for each DM in the PROMETHEE GDSS were changing, in order to examine the impact each individual DM has on the final ranking (scenario 2). The results are evaluated via a "robustness index", which shows how many times the measures ranked as top five are repeated. Third, only the input from the six government officials of our sample is considered to produce the ranking of measures (scenario 3). Fourth, the input from the four economists of the sample was used to produce an alternative ranking (scenario 4).
The aim of the first two sensitivity cases (scenarios 1 and 2) was to examine how sensitive the results are to a change in the weights of the criteria. The top 5 measures had high robustness indices and a very high probability of being ranked as top 5 even when the weights of the DMs change. For instance, the first two measures of the original ranking (M13 and M5) have a probability of 98% and 99.2% of being in the top five, respectively. Scenario 3 results are the same with those of the base GDSS (i.e. all stakeholders), although with a different ranking. However, results are very different in Scenario 4. In this scenario, M7 (grants to enterprises) ranks first instead of M13 (carbon taxation), which now ranks sixth. Also, two other measures appear in the top five list (M6: subsidies to business loans with an environmental management system; M3: grants for energy renovations of buildings to upgrade to NZEB).

Results and discussion
After extensive deliberations with stakeholders from the public and private sector since the outbreak of the pandemic, we arrived at thirteen recovery measures to evaluate further. We used a wide array of sustainability criteria for assessing these thirteen interventions, adapted from a comprehensive checklist that was developed by the World Bank [16]. As the EU decided in 2019 to explicitly include the seventeen UN SDGs in its regular macroeconomic monitoring procedure, and due to the universality of SDGs [46], Appendix B also includes an indication of the SDGs addressed by each sustainability criterion used in this analysis.

Energy and economy model results
Some of the quantitative criteriathose related to impacts of green measures on energy use, carbon dioxide emissions, economic output, and jobswere evaluated through specialised models that had been already used for national impact assessments and were known to public authorities. The energy model provided projections for energy and carbon emission savings per million of Euros invested in each measure. According to model results, some of the best-performing measures in the short-term are also the preferred ones for the longer term (up to 2030); these are carbon taxation (a regulatory measure with very low implementation costs), virtual net billing (which is also regulatory and will enable a faster deployment of decentralised solar power generation), and implementation of urban mobility plans. In the short run, measures on energy efficiency upgrades also have good carbon abatement potential per unit of investment.
To simulate the impact of each recovery measure on economic growth and employment, an economic input-output model was used. The impact of a policy depends on how much investments in a sector affect demand for goods/services in other sectors, what part of intermediate inputs of a sector takes place in the country, which production activities are displaced, and how labour-intensive are the sectors affected by new investments, compared to that of displaced activities. Note that the reduction of household consumption in response to energy efficiency improvements is returned in the economy and further respent through a rise in consumer demand for goods and services. Fig. 2 illustrates the differences in economic effects versus (a) employment and (b) environmental effects of the modelled interventions. The impact is relatively higher in the short run due to the front loaded allocation of investment expenditures in 2021-22, both because this is what a recovery plan focuses on, and because a strong economy-wide rebound was assumed for these two years.
It is interesting to compare Fig. 2a with Fig. 2b, in order to assess the performance of measures across the key economic and environmental criteria. Even though several measures, such as car scrappage, installation of smart meters and energy renovations, perform quite well in terms of economic multiplier and employment, as indicated in the top right corner of Fig. 2a, these same measures do not lead to high carbon dioxide emission reductions. In fact, no measure appears in the top right quadrant of Fig. 2b, despite the emission reduction effectiveness of some of the measures focusing on improving efficiency. This might be an indication that the pool of measures does not include individual measures that manage to perform ideally across all three of these categories (i.e. economic output, employment, emission reductions), which makes the selection of measures with diverse benefits even more imperative.
Results confirm once more the conclusion that Barbier [12], Popp et al. [47], and Strand and Toman [11] have drawn on the basis of ex-post assessments around the world: measures performing best in the short run are partly different from those with the largest positive effect in the longer term. With regard to economic output generation, in the short-run measures M4 and M5 (installation of smart electricity meters and virtual net billing respectively) create the highest economy-wide effects relative to the reference scenario; for every million Euro (M€) invested for these interventions, the total output of the economy increases by 1.45 M€ and 1.44 M€ respectively in 2022-23. Two measures that could boost short-term output, M9 and M12, are not included -M9 because the construction of the tram line is expected to start after 2023, and M12 because tree planting has not been simulated with the IO model due to lack of data. In the long run, virtual net billing (M5) creates the highest economy-wide effects.
The impacts on employment are similar but not identical to the effect on economic output. In the short run, virtual net billing and smart meters create the highest positive effects on national employment; for every million Euro invested in either measure, about 14 new jobs are created throughout the economy. In the long-run, the virtual net billing measure still creates the highest economy-wide employment effects. On the contrary, sustainable mobility interventions M8 and M9 do not perform well in terms of employment generation, thanks to their success to shift mobility from private cars to public transport; this causes household spending for fuel and car purchases as well as for car maintenance to drop considerably, and the affected economic activities (imports of fuels and vehicles) which are labour-intensive seem to have a negative impact on the overall employment. However, such model-based seemingly undesired effects of energy efficiency measures have to be treated with caution. They should not be interpreted as suggesting to avoid energy efficiency investments that improve the economy-wide productivity of energy use. A feature of input-output economic models is the assumption of fixed technical coefficients: the combinations of inputs are employed in fixed proportions. This assumption implies that there is no substitution among the inputs and no technological progress, which is plausible when immediate policy impacts are simulated, but becomes less realistic when the impacts over a longer time horizon are modelled. The effect of the changing technical coefficients -due to changes in production technologies -on GDP, employment and CO 2 emissions using either modelling techniques or expert judgements [48,49] can be pursued in future enhancements of this work. To the extent that the reduced economic output and employment in sectors such as trade of vehicles and fuels is compensated through re-training of workers and re-orientation of business activities, negative economic impacts of sustainable mobility can be overcome. More broadly, the ability of an economy to transform itself and use the resources saved to grow new sectors, or to divert saved resources to export-oriented activities as a result of increased business competitiveness, will depend on factors like people's skills, availability of financing and policy decisionsaspects that are insufficiently accounted for by this kind of economic models.
It should also be noted that the economic model does not distinguish between employment categories, so our approach cannot include the separate impact on low-skilled and high-skilled workers, which would be important for evaluating the effect of each intervention on long-term growth prospects. Such considerations are important in view of the findings of studies that examined stimulus measures applied after the 2008-2009 economic downturn, which highlight the importance of observing differences in skills in order to properly compare alternative policies [50]. These are aspects that merit further investigation. As such, a disaggregation of employment IO multipliers can be pursued in the future to better portray the effects of measures on low-and high-skilled employment and subsequently on overall long-term growth. Moreover, job calculations do not account for the possibility that supply of skills in some sectors may not suffice to meet growing demand. For example, a construction boom due to energy renovations may be limited by a lack of skilled technicians in the country, at a time where a 'renovation wave' in buildings is foreseen across Europe. Therefore, before deciding on the extent of implementation of a recovery measure, a skill mismatch analysis would be needed to ensure that human resources are available for realising this intervention.
It is particularly interesting to observe the results of the counterfactual scenario, which assumes a 'return-to-normal' economic stimulus, where all recovery funds are allocated uniformly to households and businesses, and consumption continues as before. As indicated by a red dot in Fig. 2, a business-as-usual economic recovery is clearly not the preferable option; with an economic multiplier of just 0.83 it performs better than only two out of all the green measures. It also has a mediocre effect in terms of employment generation, with 6.3 new jobs per M€, whereas four green measures have more than double job benefits. This clearly indicates that a uniform demand stimulus is far from the most effective strategy for increasing employment in the short term. Such a finding, which has been explored by very few studies so far, is confirmed by Pollitt [51] for some major EU economies (through a macro-econometric model) and by the IMF [52] for the global economy (through a computable general equilibrium model).

Multi-criteria assessment results
The transition to sustainability requires implementation of ambitious but actionable measures, and stakeholder participation for codeveloping a green stimulus plan is crucial. Therefore, a variety of stakeholders were invited to provide input for this assessment. They were selected in order to be representative of public authorities, businesses, and civil society, and participated in a dedicated workshop, held in October 2020. Specifically, ten stakeholders participated in the evaluation: one representative from the Ministry of Transport, one representative from the Ministry of Energy, two representatives from the Ministry of Finance, two representatives from the Directorate General Growth, one representative from an environmental NGO, one representative from an NGO promoting sustainable energy, one partner from the National Association of Employers and Industrialists, and one partner from academia. There, participants were presented with the recovery measures and the evaluation criteria and were then provided with the respective tables to fill in, applying the adopted multi-criteria methods.
Preferences of workshop participants varied significantly by criterion, highlighting the different priorities of each stakeholder, as indicated in Appendix F. For example, representatives of private enterprises valued short-term criteria more strongly than long-term ones, in contrast to other stakeholders. On the other hand, governmental stakeholders provided a higher weight to long-term environmental criteria compared to short-term ones. Overall, most decision makers assigned a higher importance to the long-than to the short-term, whereas there was no consistent preference to environmental versus economic/social criteria. Fig. 3 displays intermediate results of the evaluation, before applying the final weighting of all criteria. It illustrates the relative strengths and weaknesses of each measure in terms of short-or long-term economic and environmental performance, as perceived by the participating stakeholders. For instance, the carbon tax reform (measure M13) received a high score for its environmental performance in both the short-and the long-term, and actions related to sustainable mobility (M8 and M9) also had a good score on long-term environmental performance. Conversely, measures M5, M6 and M7, which mainly target businesses, were assigned by participants the highest scores regarding long-term economic effectiveness. Visualization of this variability in the perceived economic and environmental effectiveness of the various measures highlights the contrast between the respective criteria and could provide useful insights before policy implementation.
After weighting the scores for all groups of criteria (environmental and socio-economic, short-and long-term), the final ranking shows that the best green recovery intervention is considered to be the fiscally neutral carbon tax reform (M13) ( Table 2). This measure had the best results in terms of energy savings and carbon savings for the short-and long-term, which were calculated using quantitative models, but received also high evaluations from the stakeholders in several criteriait ranked first in the individual ranking of five out of ten stakeholders. In addition, stakeholders valued measures M5, M2 and M7 (implementation of virtual net billing, energy renovations of buildings and grants to reduce the carbon footprint of enterprises) as the next most important for the sustainable development of Cyprus.
These results offer interesting insights when compared to evaluations performed informally by some of the authors, who applied a simple assessment approach to obtain a rapid overview of the preferred policies. Although the expert view of some of the authors had considered long-term infrastructure modernisation and decarbonisation interventions as important priorities, those were not among the preferred measures of stakeholders. For example, the only nature-based measure (M12 -tree planting) and the two sustainable mobility measures (M8 and M9) rank average or below-average in stakeholder preferences; this may be attributed to concerns by some stakeholders, which were orally communicated during the workshop, that these measures are unrealistic, too costly, or can only have limited effects. Another reason may be that most stakeholders considered the period to 2030 as 'long-term' and   [53,54]. Nonetheless, such discrepancies are not surprising; it has been recognised in the literature that stakeholder preferences diverge from those of experts [39]. Even when they contradict modelling results and expert opinions, these views need to be taken seriously, considering the direct experience of stakeholders and decision-makers. Some measures may have lower social acceptance than experts believe, and may require more in-depth work to take into account stakeholders' concernsfor example, by considering more carefully the risks of some measures to social equity and through proper information and communication campaigns to target audiences and the broader public.
At the same time, it is interesting that a seemingly unpopular measure (carbon taxation, even if framed as fiscally neutral) received the top score among recovery measures. In view of the extensive discussions about the social acceptance of such pricing schemes worldwide [55,56], this seems to be a surprising but also encouraging result, as carbon pricing is widely considered by economists as a necessary ingredient of effective decarbonisation policies. In the context of the current pandemic, Engström et al. [57] call this kind of reform "excellent climate policies [which] also help deal with the coronavirus crisis by allowing reductions to labour taxes". A plausible explanation for the high score of this measure among Cypriot stakeholders is that this kind of green tax reform (comprising an increase in environmental taxes to be compensated by reductions in labour taxation) has been promoted in Cyprus by some experts since 2015, with a consistent attempt to inform governmental authorities, NGOs and trade unions about its advantages [58]. The resulting top performance in this assessment may be an indication that targeted and well-supported information flows to diverse stakeholders have been effective and may lead to societal acceptance of such a reform in the near future.
To check the robustness of the above findings, we ran a number of sensitivity cases. It turned out that the ranking of measures shown in Table 2 is quite robust to variations of the weight of individual criteria. Similarly, the preferences do not substantially change according to whether the stakeholders are government officers or work in the private sector or NGOs. Conversely, it seems that economists in our sample (four out of ten decision makers) had considerably different input than engineers and natural scientists who formed the rest of the sample. Economists preferred three types of measures: a regulatory change with no cost (virtual net billing -M5), grants/loans to businesses (M6 and M7) and energy renovations of buildings (measures M1 and M3). It seems that their earlier policy experience has led them to conclude that building renovations are the most realistic technological interventions, whereas they are sceptical towards carbon taxation, probably in view of its political cost. It is thus impressive that non-economists strongly consider the carbon tax as their preferred measure, so that it ranks first in the overall ranking despite the scepticism of economists.

Conclusions
A set of policy recommendations can be extracted from the findings of the analysis. Firstly, it is evident that a variety of measures is needed to achieve economic growth without compromising environmental sustainability. Even though the carbon tax reform has received the top score, it does not mean that implementation of this measure on its own can have the desired impact across the entire set of criteria. At the same time, this measure's high ranking, which has been in discussion with local stakeholders already for several years, highlights the importance of information campaigns to increase awareness of the potential benefits of a certain action. This helps to build consensus that permits gradual implementation and social acceptance of reputedly unfavourable policies. Additionally, the weights and scores given by the stakeholders indicate that great significance is given on the perceived plausibility of actual implementation of certain measures. As such, measures that can lead to considerable reductions in carbon dioxide emissions, such as the Nicosia tram or the Sustainable Urban Mobility Plans, perform relatively poorly. The success of these measures will be determined to a great extent to the political will of the responsible authorities, as they will be quite costly, their implementation will require several years, while their benefits can be expected primarily in the long-term.
Our analysis illustrates the importance of combining simple methods with more sophisticated models for an assessment of recovery measures that can provide meaningful support to policymakers. Moreover, our findings clearly demonstrate trade-offs between the short-term (2022), the long-term (2030) and the climate neutrality (2050) targets, as well as the superiority of many green measures in comparison to business-asusual demand stimulus. Some immediate measures with attractive shortterm impact have short-lived benefits and turn out to be inferior in both economic and environmental terms by 2030. This is in line with the finding of Barbier [12] that a different policy mix is required for short-term interventions as compared to a medium to long-term strategy for a recovery leading to a green transition. Institutional or regulatory changes, such as the gradual implementation of carbon pricing or the reform of electricity rules to enable decentralised power generation, may have long-term impacts with low cost. Blunt economy-wide demand stimulus measures are not only environmentally unsustainable but also economically mediocrethey perform worse in promoting economic growth and employment than most of the examined measures. This provides solid quantitative evidence (which can be found in very few studies in the literature) against a 'return-to-normal' stimulus. The final ranking of appropriate measures, benefiting from input by diverse societal stakeholders, contributes to the 'democratisation' of the policy formulation process [29] and enables ownership of the measures by national decision makers. Linking the sustainability criteria with the UN SDGs facilitates the alignment of national recovery programmes with the international policy agenda.
This analysis provides useful insights into the opportunities to green the post-pandemic recovery. However, it needs to be complemented by a broader look at the full package of recovery measures, including those that are not directly related to energy and climate change. For instance, public investments will be directed to health and social care infrastructures as well as information and communication technology, and green and climate considerations will need to be included in the design of these investments. Modernisation of hospitals and schools increases welfare of the population and, if it includes for instance energy efficiency renovations and proper management of hazardous waste, can as well improve resilience and reduce carbon emissions.
It is always possible to do more analyses and mobilise more sophisticated models to cover more dimensions of the problem and refine the policy prioritisation process. In particular, future research could analyse the economy-wide rebound effects of energy efficiency improvements in terms of CO 2 emissions through the application of an energy IO model. At the same time, a crisis by definition is a situation where timeliness is critical, and it would not make sense to take years to provide the ideal recovery package. Our two-step approachcombining modelling when tools and data are available, and experts' and stakeholders' opinion to complement quantified results and prioritise actionis one solution to find a compromise between timeliness and confidence, and to ensure that decision makers do not have to apply blindly the result of a model. As demonstrated recently [59], free and open-source energy, environmental and economic models can be rapidly developed and adapted to a new national context [9]. The speed and transparency of such modelling systems allows for transparent auditing and model improvement, which helps tailor them to national needs. Complementing this open modelling environment with qualitative data can help overcome issues of missing data and fragmented models; hence this framework can be applied in multiple contexts for sustainable economic recovery and climate compatible growth, including where data and models are less easily available than in the EU.
Future applications of the presented framework can be tested in countries or regions beyond the EU. However, this may require potential adjustments to make the methodology more applicable to the local context. Additionally, the approach does not need to be limited to postpandemic recovery, but can be adopted to address economic recovery in a broader context. For instance, in regions hit by recent conflicts, such as Ukraine, the list of criteria can be expanded to include security aspects or take into account the need for reconstruction of crucial infrastructure.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability
Data are available in an open website, and we have declared this in the manuscript. and the Republic of Cyprus, in the frame of the project "Supporting the Economic Recovery of Cyprus with a View to Energy and Climate Policy" (grant no. CONCEPT-COVID/0420/0008). Mark Howells' input was supported by the FCDO Climate Compatible Growth Program of the UK government. The authors are grateful to stakeholders who participated at a workshop on October 5, 2020 and provided input for the multicriteria assessment. Subsidy to loans of businesses certified with an environmental management system

Appendix A. List of Green Recovery Measures
As an economic stimulus measure, Interest rates of business loans will be subsidised by the government. It is proposed that the interest rate to be subsidised is 0.5% higher for those firms that have adopted or will adopt the EU Eco-Management and Labelling Scheme EMAS by December 31, 2021. The certification process can be subsidised with a small grant of the order of €2000 per firm.
This measure will cost approximately €2 million and can yield fast improvement in energy efficiency and/or environmental performance of businesses. New proposal, currently not included in National Energy and Climate Plan.

M7
(continued on next page) Total cost of implementation is estimated at €85 million.

M13 Fiscally neutral carbon taxation for economic sectors out of the EU Emissions Trading System
Tax up to €120/tonne of carbon dioxide on non-ETS sectors, i.
e. on all fossil fuels except those used for power generation and by cement plant and brick factories.
Gradual introduction within 5 years. Expected public revenues in full implementation: €100-150 million/year.
Tax revenues could be rebated to all households to increase political acceptance.
Administrative cost estimated at €500,000.  Appendix C. Input-Output model

D.1. AHP Method
AHP is a pairwise comparison method which uses a ratio scale that does not require any units. DMs express their preferences for one alternative over another one. The number of comparisons is n 2 − n 2 , expressed in an n × n pairwise comparison matrix. Typically, DMs express their preferences using a 1-9 scale, which is assumed to offer the appropriate flexibility.
One important aspect of this method is that the pairwise comparison matrix needs to be consistent, which becomes more difficult for matrices with large dimensions. This can be checked via the consistency ratio as shown in eqs. (1) and (2).
Where CI(X) the consistency index, RI n a real number that estimates the average CI from a large dataset of randomly generated matrices of size n and estimations can be found in literature, and λ max the maximum eigenvector [60]. It is suggested that matrices with CR > 0.1 are inconsistent. The priorities can be calculated typically by three methods, namely: (i) the eigenvector method, (ii) the normalised column sum method, and (iii) the geometric mean method. In this paper the geometric mean method has been applied, where the priority vector is calculated as the geometric mean of the elements on a row, over the respective normalisation term in order for the sum of the weights to be equal to 1, as shown in eq. (3): In the framework of this study the 23 criteria shown in Table 2 have been divided into two broad categories of short-and long-term impacts and further subdivided into two subcategories, namely: (i) environmental criteria, and (ii) economic/social criteria, as shown in Figures D1 to D.3.

D.2. PROMETHEE Method
For the evaluation and ranking of the alternatives the PROMETHEE method has been applied. The independent experts were asked to evaluate each alternative (or action as it is called in PROMETHEE terminology) for each criterion. The remaining criteria are qualitative, and the DMs were asked to express their evaluation in a typical 1-5 scale ranging from very low to very high. In PROMETHEE each action is compared to (m − 1) other actions in order to calculate the positive and negative outranking flow of each action as a number between 0 and 1. These values express how much this action is preferred over all the other ones as shown in eqs. (4) and (5): While the PROMETHEE I method offers a partial ranking between the alternatives, the PROMETHEE II method was used, which can offer a complete ranking amongst all the actions. In PROMETHEE II the net flow needs to be calculated in order to rank the actions, according to eqs. (6) and (7): ] w j (6) and For the pairwise comparison a Type 5 V-shape preference function has been used, which considers a preference (p) and indifference (q) threshold as shown in Figure D.4.   Fig. D.4. The type 5 preference function [61] As there are several DMs, the PROMETHEE GDSS is implemented. At the final stage of this method a global evaluation takes place; after the individual evaluation from each expert a global matrix is constructed with the rows being the alternatives and the columns the flow values calculated by the DMs. We assumed that the weights of the DMs are equal and that the preference function is of the same type. It is noted that the DMs have agreed to the preference and indifference thresholds. Then the PROMETHEE method is applied which gives the final ranking. Tables   Table E1. Pairwise comparison between short-and long-term impacts using the AHP scale (please fill only the blank cells)

Short-term impacts
Long-term impacts Short-term impacts 1 Long-term impacts 1 Table E2. Pairwise comparison between short-term "Energy/Environmental" and "Economic/ Social" criteria using the AHP scale (please fill only the blank cells)