Funding and Project Details – SIMPOL Project




Traditional models of financial contagion fail to incorporate the effects of commonly used instruments and practices such as derivatives and re-hypothecation, which can give rise to very complex patterns of interdependent payoffs. Further, despite the growing awareness among policy makers of the urgency to assess the systemic impact of climate change and climate policies on the financial system, traditional models do not offer suitable tools to do so. In this context, the first goal of SIMPOL has been to fill these gaps by working closely with regulators involved in the process of elaborating novel tools to inform policy recommendations. The project has pursued and achieved three objectives in this regard: O1) to develop new-generation models of financial networks, O2) to provide tools to quantify the accuracy in the estimation of systemic risk to better inform the discussion about financial information disclosure, and O3) to develop methods to identify key actors in the climate-finance nexus so to contribute to the design of a more sustainable climate-financial system.

The stability of the financial system and its role in the transition to a low-carbon economy has many societal implications that go beyond the stakeholders in the financial industry. For this reason, the vision of the SIMPOL project is that the civil society should play a larger role in the debates in the policy areas of finance and climate-finance. This vision is consistent with the EU Better Regulation agenda[1], which intends to foster the transparency of the policy making process and a more balanced representation of the economic interests in the policy outcome with the goal to increase the political cohesion of EU citizens. The platforms to engage EU citizens in the policy cycles include the public consultations[2], the Refit platform and the European citizens’ initiatives[3]. The participation of citizens to these initiatives has been limited so far. In particular, the Better Regulation agenda intends to change a diffused perception that, in several policy areas, the policy cycle is largely influenced by networks representing specific economic interests and industry players and that the public interest remains underrepresented in the process.

In this context, the second goal of the SIMPOL project has been to provide new insights into the barriers and enablers for citizens and societal stakeholders to engage more actively in the policy making process. The project has pursued and achieved three objectives in this regard: O4) Develop an ICT infrastructure to enable the data collection and analysis of policy-relevant socio-economic networks, including through crowdsourcing tools, O5) Develop an ICT infrastructure to gather semantic web data for policy relevant socio-economic networks, and O6) Improve our understanding of how to empower citizens with collective mapping of networks of influence in policy making processes in the climate-finance regulation arena.




Project Objectives

Project Objectives

1. Develop new-generation models of financial networks incorporating complex financial instruments that are prone to generate unforeseen systemic effect

2. Provide new tools to quantify gains in the accuracy of risk estimation from future regulations and information disclosures, in particular with respect to non-conventional banking

3. Identify key actors and instruments to contribute to the design of a more sustainable climate-financial system.

4. Develop an ICT knowledge sharing platform to enable the crowd-sourcing of data collection and data analysis regarding policy-relevant socio-economic networks.

5. Develop an ICT infrastructure to gather semantic web data for policy relevant socio-economic networks.

6. Demonstrate the possibility to empower citizens by enabling real-time collective mapping of networks of influence in policy making processes.



The overall work plan of the SIMPOL Project consists of three main work packages (WP) whose activity covers the goal of our project.

WP1 – Financial Networks for policy modelling. In the first instance we believe it is necessary to develop financial network models that incorporate complex instruments in order to provide insights to regulators regarding contracts data disclosure and capital requirements. More specifically we want to develop network models of shadow banking by incorporating also network dependencies related to climate finance and those obtained via crowdsourcing. At this point we have the mathematical instruments to measure and investigate the robustness of different financial system architecture in order to contribute to the public policy debate.

WP2 – Modelling Climate Finance Policies. At the same time, we want to analyse the interactions among the various actors and instruments involved in climate finance. Since climate change is a paradigmatic case of global issue with planetary scale and a long-term temporal horizon, the nexus Finance-Climate is a paramount challenge in policy modelling. Moreover, the complexity of financial innovations often hampers the transparency that is needed in climate related agreements. In this respect, we aim to develop network models of climate finance that incorporate financial instruments not taken into account to far. We also develop macro models of climate finance and models of networks of influence revolving around climate-finance policies.

WP3 – Collective Mapping of Influence Networks. Finally, we build empirical socio-economic networks from a large variety of resources available. Because the sustainability of the financial system is a societal issue, the input of civic society cannot be neglected. The aim of this WP is to demonstrate the feasibility empowering citizens and increase decision-making transparency around financial regulations. It is not necessary to engage civic society in the discussion of the details of policies and their effects. Instead, it is possible and socially desirable to involve civic society in uncovering the network economic interests revolving around policies. To avoid cold-start problems, we follow a mixed strategy whereby we both engage users (with the incentive that they can in exchange gain privileged access to the platform) and we mine existing open datasets. To this end, we will first develop a simple infrastructure, based on the semantic web, to construct, visualize, explore and extend the networks. The networks will be initially constructed by crawling public, open data on the web. Next, the interested public will be able to explore and contribute partial data. This will be used to construct new economic networks and to enrich those that already exist in the database. In this WP, we will also monitor streams of big data ranging from news to financial blogs and social media (such as Twitter) to extract and aggregate the public pulse about socio-economic issues, thus providing additional links and properties to the networks.

WP4 Management. We define the instruments to set up collaboration with nodes and to deliver the objectives of our project.

WP5 Dissemination. We make a plan on the dissemination instruments we shall use in the course of the activity, including scientific papers, papers targeted at the broader public, conferences and schools.



The following table lists the partner institutions to the original SIMPOL project (i. e., the project funded by European FP7-FET grant no. 610704 called SIMPOL). For a list of the partners of the current SIMPOL Project Initiative, please visit our Who We Are page.

Institution Country
University of Zurich Switzerland
IMT Alti Studi Lucca Italy
Global Climate Forum Germany
Université Paris I Panthéon-Sorbonne France
Institute Jozef Stefan, Ljubljana Slovenija
London Institute for Mathematical Sciences UK

Final Results and Potential Impact

Final Results and Potential Impact

The potential impact of the work of SIMPOL in scientific, technological and societal terms includes:

1. Enriching the policy toolbox with network models of complex financial markets. The results on the methodology of “Flow of Risk” demonstrated how to derive policy insights of financial stability in the context of financial derivative instruments such Credit Default Swaps. Moreover, our results on the fundamental limitations on pricing arising in the presence of Credit Default Swaps will inform the policy discussions (within EU central banks and other institution concerned with financial stability such ESRB, ECB and EBA) about financial derivative markets and in particular about benefits and limitations from the disclosure of information about contracts and the migration towards centrally cleared markets.

2. Providing mathematical foundations to the policy debate on financial interconnectedness. We have provided better mathematical foundation to the discussion about the relation between the structure of financial networks and their resilience. These results will inform the policy discussions about market integration and financial stability within EU central banks and other institutions concerned with financial stability such as ECB, ESRB, EBA, ESM) about financial interconnectedness, market integration and macro-prudential policies.

3. Responding to the G20 FSB call for new tools to assess climate-related financial risks. We have filled an important gap in the literature in the context of climate risks and climate policy risks, by providing financial institutions and investors with a tool to assess the potential impact of climate risks and climate policy risks on the financial system. The methodology and the results of the climate stress-testing methodology has attracted the interest of representatives from several relevant policy makers including international organizations such as the World Bank, the Climate Investment Funds, the European Investment Bank, the G20 Task Force on climate-related financial information disclosure, the German Environmental Ministry.

4. Testing the idea of ICT-enabled crowdsourcing for complex societal issues. The Digital Single Market Agenda’s Pillar VII “ICT-enabled benefits for EU society”[1] aims address through ICT’s complex societal issues such as climate change mitigation. In this context, we have contributed to better understanding benefits and limitations from the possible use of crowdsourcing through the use ICT’s in order to address more complex tasks than those usually considered in the literature.

5. Identifying barriers and enablers for citizens’ engagement. In the context of the EU agenda on Better Regulation, focusing as a case study on the public consultations on the EU 2030 climate-energy framework, we have identified as two major barriers the legal complexity of the themes proposed in the consultations, and the perception of disempowerment in the face of more influential advocacy groups. The policy recommendations emerging from this exercise point towards improving the communication around the EU public consultations and towards engaging in dialogues on new institutional channels in which civil society could make its voice better heard.

6. Providing new tools to analyse the public sentiment about policy topics in social media. Creating feedback loops from the sentiment of the EU public about policy discussions and the impact of policies being already implemented in order to understand how to increase citizens and confidence is a crucial element for both the EU Consumer Agenda[2] and the Digital Agenda[3]. In this context, we have provided new tools to analyze the dynamics of sentiment in social media in relation to policy topics. These tools are documented on open access publications that other researchers can exploit.

7. Providing new tools to analyse the spread of misinformation on policy topics in the social media. Preventing misinformation on the objectives and the impact of EU policies is also a common objective of the EU Better Regulation Agenda, Consumer Agenda and Digital Agenda. In this context, we have provided new tools and cases studies to analyse the spreading of misinformation, as well as on the dynamics of belief reinforcing.

8. Uptake of methods and results across other current and future EU projects. The output of SIMPOL has already been taken up in a number of other EU projects such as H2020 FET project DOLFINS[4] (grant no. 640772), the climate action project GREENWIN[5] (grant no. 642018), the FP7 project IMPRESSIONS[6] (grant no. 603416). Results have being serving as a building block for new submissions to research projects on European Climate Risk Assessment.




[3] see IP/10/581, MEMO/10/199 and MEMO/10/200




Research Output

Research Output

Task 1.1 OTC Derivatives Networks

Thanks to the collaborations with ECB and ESRB, we were successful in accessing first to the global data on CDS market from DTCC (see the publication D’Errico et al 2017). Later, based on the interest raised by the results, we were also invited by the ESRB to contribute to the processing and analysis of the EMIR data. We have been producing two models of distress propagation in this context:

  • In Battiston et al. 2016 “The Price of Complexity in Financial Network”, published on PNAS we have developed a model that allows to resolve the problem of computing the default probability of financial institutions in the presence of CDS derivative contracts[2]. We study the accuracy over the estimation of default probability as function of data disclosure/errors.
  • In Schuldenzucker et al. 2016 we have succeeded to develop an extension in the presence of CDS of the pioneering model of Eisenberg and Noe 2001 model, which is widely used as a basis for stress-tests. We have analyzed the problem of determining the set of defaulting banks in a financial network consisting of both debt and OTC derivatives like credit default swaps (CDSs). We show that these networks are exposed to a new kind of systemic risk unnoticed so far in the literature: the very question of which bank is in default may have several answers (multiple consistency) or no answer at all (inconsistency). These phenomena might cause severe legal, liquidity, and pricing-related problems, thus demanding a further systematic analysis.

Further, in the paper D’Errico et al. 2017, in addition to the empirical analysis of the DTCC, we also developed a methodology to compare the systemic impact of banks in different locations of a financial network of CDS contracts. We have demonstrated analytically that the systemic risk is higher in a bow-tie architecture than in more fragmented architectures.

Task 1.2 Re-hypotecation Networks

Because of the persistent lack of data on rehypothecation and the unique opportunity that came up at the beginning of the project to access confidential regulatory data on CDS markets, in agreement with the reviewers at M24 review, we had decided to reallocate part of the effort from Task 1.2 to Task 1.1 Derivative Networks. Nonetheless, in collaboration with the group of Mauro Napoletano in OFCE Sophia-Antipolis (external to the SIMPOL consortium), we have developed a model of rehypothecation in which agents establish contracts among each other and have the possibility to reuse the collateral (see Luu et al. 2017). We found that the liquidity in the system (measured here by the amount of collateral that is endogenously created in the system) increases with the density of the network, revealing an important effect of market integration and diversification processes on collateral and liquidity creation. In addition, as such processes expand, the emergence of long chains and especially cyclic structures can create a level of collateral that is much larger than the initial total proprietary collateral of banks in the network. Furthermore, we extend the model by introducing the possibility for agents to hoard dynamically part of the liquidity and we study the cascade of liquidity hoarding and consequent loss in total collateral as a result of shocks hitting a selection of banks in the system under different network structures. The work also provides a method for the estimation of the impact of a shock due to collateral losses in presence of increasing uncertainty, modelled as changes in the perceived value of Value-at-Risk used by the agents to determine their level of liquidity hoarding (Luu et al. 2017).

Task 1.3 Shadow Banking

As discussed with the reviewers on M24, the effort initially planned in the above Subtasks has been reallocated to Task 1.1 and Task 1.4 in view of the unique opportunities to collaborate with ECB, ESRB and BoE emerged during the project. We were however able to contribute to an empirical study in a joint ESRB-EBA policy-oriented paper on the topic of shadow banking, which has also been covered in a press release[3]. Indeed, the ESRB Secretariat has invited one of the consortium’s researcher from the University of Zurich to work on the analysis of an exclusive dataset resulting from a one-off data collection exercise on Shadow Banking performed by the European Banking Authority (EBA). The paper represents a policy report of the analysis. Due to the confidentiality of the data, we cannot foresee at the moment in what form the paper could be published. The methodologies adopted in the analyses build on network techniques and a large part of the empirical analysis is based on the “Leverage Network Framework”, extended to take into account specific risk overlaps arising from the peculiar nature of interconnectedness between banks and shadow banks. The paper was further presented at two different meetings of the ESRB Joint Expert Group on Shadow Banking of the ESRB.

Task 1.4 Financial System Architectures

  • In Bardoscia et al. 2016, we systematically investigated the financial stability of different network architectures (Erdos-Renyi, core-periphery and scale-free networks, gradually increasing the density to locate the phase where instabilities start to appear. We demonstrated the occurrence of a novel phenomenon that we call “pathways to instability”, i.e. trajectories in the space of evolving networks along which the network move from stable to unstable. We show that a pathway to instability can occur in the process of market integration as well as in the process of increasing portfolio diversification. This finding has precise policy implications because it shows that micro-prudential policies aimed at lowering risk at the level of individual institution can have unintended consequences in terms of increased systemic risk.
  • In D’Errico et al. 2016 we compared the aggregate losses that, conditional to a shock occur in various possible architectures. We demonstrated analytically that the architecture known as “bow-tie” (which contains a large “strongly connected component”) leads in many conditions to larger systemic risk in terms of potential losses than alternative, more fragmented, market architectures. We also showed that the empirical networks in the CDS market have in most cases a bow-tie structure, a finding of significant policy relevance.
  • In Battiston et al. 2016 (“The Price of Complexity”), we analytically studied the effect of complex architectures on the uncertainty of the estimation of systemic risk. In particular, by studying simple benchmark structures, i.e. the ring, the chain, and tree structure, we have been able to show the increase in the uncertainty of the determination of aggregate losses.
  • In Roukny at al. 2016, we provide necessary and sufficient conditions on the structure of a financial network for the emergence of multiple equilibria on the default state. The policy implication is that cyclic network structures generate a fundamental source of uncertainty depending on the belief of the market players.
  • In Schuldenzucker et al. 2016 we provide sufficient conditions on the structure of the network of bonds and CDS such that CDS do not cause problems in the pricing (i.e. the determination of a clearing vector has a unique solution). In terms of policy implications, we show that this condition is equivalent to prevent the trading of naked CDS having as reference entities other banks in the system. Further, we also show that moving CDS to a centrally cleared market does not guarantee to resolve the pricing indetermination.
  • In Battiston et al. 2017 “A Climate Stress-test of the Financial System), we develop a novel methodology that building on insights from work in previous EU-funded projects (e.g. the DebtRank methodology) as well as the work carried out in SIMPOL Tasks 1.1, allows to assess the Value at Risk of financial institutions in the face of shocks on climate policy shocks and in particular to climate-finance instruments.


[1] In 2012 the EU adopted the European market infrastructure regulation (EMIR) with the aim to increase transparency in the OTC derivatives markets, mitigate credit risk, reduce operational risk

[2]The section regarding derivatives is available in the ssrn version of the paper at and it was excluded from the published version by the editors for reasons of space. We will make a future separate publication on this section.



Task 2.1 Network Models of Climate Finance

This task has led to the development of three main lines of research.

  • A network-based analysis of the EU carbon market, which has been left aside following the recommendation of the reviewers after the M12 review. The main result was the identification of a relationship between the structure of the network of transactions on the one hand and market inefficiencies and price volatility on the other hand. This relationship, can be used as an early-warning system for failures in the European carbon system. This line of work has led to a publication in international proceedings (Karpf and al. 2015) and a journal paper is still under review.
  • The development of a climate stress-test framework for the financial system, which was regarded as very promising by the reviewers. This work combines a detailed analysis of the relevance of existing sectoral classification systems in view of climate policy and models of distress propagation in financial networks developed in WP1. The work is described in details below and has led to a publication in one of the leading journals in climate research (Battiston et al, 2017).
  • Data analysis on the state of the climate finance field. In the first year of the project, we had hence reviewed existing instruments in climate finance with a particular emphasis on green bonds. The development of our climate stress-test framework has led us to refocus this effort in the last year of the project on collecting and analyzing data about the exposures of the financial system to climate policy complementary to the one that can be obtained from existing databases. We have hence conducted a detailed analysis of the exposures to the European electricity market in view of the integration of our stress-testing methodology with the modeling work developed in task 3.2.


Task 2.2 Macro-economic Models of Climate Finance

This task has led to the development of four main lines of research.

  • The comparative analysis of macro-economic analysis in terms of the types of climate policies they are able to account for. In particular, we point out how, by construction, commonly used model structures find mitigation costs rather than benefits. We then describe mechanisms that, when added to these model structures, can bring win–win options into a model’s solution horizon, and which provide a spectrum of alternative modelling approaches that allow for the identification of such options. This line of work had led to the publication Wolf et al. (2016).
  • The development of network-based models of production networks in view of understanding the potential impact of climate finance and of climate policy on firms’ demographics and the direction of technological change. A first publication has focused on the development of a model of the formation of production networks with out-of-equilibrium dynamics (Gualdi and Mandel, 2016). A second paper (Gualdi and Mandel, 2017) has focused on the impacts of policies such as feed-in tariffs or preferential market access for renewables on endogenous technological change, firms’ demographics and the long-term evolution of global production networks. These models have been taken up by the FP7 project IMPRESSIONS, which investigates the impact of high-end climate change on the European Economy, in order to analyze the restructuring of production networks that could be induced by long-term climate change.
  • The development of an agent-based model of the European Electricity Market. The objective of this line of research is to develop tools to simulate the impacts of climate policy at the utility level in order to provide an assessment of the potential shocks induced on the largest utility companies in Europe. These shocks can then be used to quantify the risks related to the European energy transition for the financial system using our climate stress-test methodology. This line of work has been given lower priority in the last year following the reviewers’ comments and the necessity to invest further resources in the collaboration with WP3. Current status of development is reported in Balint and Mandel, 2017.
  • In view of a better understanding the interface between finance and climate finance and of the properties of the supply of climate finance to the real side of the economy, we have pursued a fourth line of research on green bonds during the last year of the project (M25-M40). This has led to a working-paper (Karpf and Mandel, 2017) in which we analyze a database of 2 millions transactions on the U.S municipal bonds market. We show that there is a positive spread between brown and green bonds, which can be explained by differences in their fundamental characteristics. However the market does not fully account for the fact that emitters of green bonds are in general more creditworthy and impose, all other things being equal, a negative premium on green bonds. These results are detailed below.


Task 2.3 Modelling Networks of Influence in Finance and Climate Finance

This task has led to the development of three main lines of research.
  • In collaboration with WP3, we have developed a crowdsourcing platform in order to construct a database about the networks of influence revolving around the EU 2030 climate and energy framework. This has led to the development of a methodology for the analysis and the visualization of European public consultations. This line of research is extensively described in the Deliverable 3.3_Update.
  • We have built on the wealth of data made available by European institutions in the context of their transparency policies in order to draw maps of key policy networks, and produced the following works: The set of lobby organizations (Zeng and Battiston, 2016), the global ownership network (Glattfelder and Battiston, 2017), the EU parliament (Cherepnalkoski and al. 2016), and more recently the interactions between lobby groups and E.U commissioners (Karpf and Mandel, in progress). The main results of these activities are summarized below.
  • In line with the objective of using algorithmic game-theory models of the influence process, we have pursued a more theoretically oriented line of research. We have investigated both algorithmically and analytically the strategic behavior of influencers embedded in a network in which the majority of agents update their opinions in a boundedly rational manner on the basis of the opinions of their neighbors. The main innovations of our approach are to consider heterogeneous type of agents (boundedly rational and strategic) and to account for the presence of multiple interests via a game-theoretic perspective whereas the bulk of the existing literature in computer science has focused on the behavior of a single influencer (in view of marketing applications). This has led to a first publication in one of the leading journals in the field (Grabisch et al. 2017). Our second contribution (Mandel and Venel, 2017), which is highlighted below, has further introduced a dynamic perspective to the analysis of influence strategies. In this framework, we have emphasized the duality between agents adopting a forward looking perspective on influence, i.e. these that focus on the forward propagation of their opinion, and agents adopting a backward looking perspective, i.e. these that focus on counteracting the influence of their opponents.


Task 3.3 Crowd-sourcing the Engaged Public

In the second review meeting some suggestions on how to improve the crowdsourcing activities were made. In line with these recommendations, for the remainder of the project, the consortium worked on:

  1. Improving the engagement and visualization tools by:
    • Enriching the data from the consultation with additional financial and company data (from the Orbis database) as well as data from the transparency register, such as lobbying expenses by the organization.
    • Developing a general conceptual framework (Policy Network Map – see figure 1 below) to analyse the contrasting positions of different actors with respect to policy issues and a methodology to visualize the data based on a novel network layout algorithm.
    • Improving the visualization of large networks
  2. Improving the outreach through different types of online and offline engagement.
    • Dedicated sessions with students and conference participants at the Conference on Sustainability and Financial Stability in Zurich, January 2017
    • Individual meetings and calls/webinars with NGOs working on engaging civil society organisations in EU public consultations.
  3. Conceptual work to identify fundamental barriers and drivers to citizens engagement in the policy making process


Task 3.4 Mining Streams of News and Blogs

Research questions
The research questions addressed here are relevant for the Objective 4: Develop an ICT knowledge sharing platform to enable the crowd-sourcing of data collection and data analysis regarding policy-relevant socio-economic networks, and Objective 5: Develop an ICT infrastructure to gather semantic web data for policy relevant socio-economic networks. The questions are: How to pinpoint news articles relevant for a specific policy (such as Climate finance)? How to incorporate news articles relevant for a specific policy in a crowdsourcing scenario?

Activities and results
The workflow for acquiring and processing news articles builds upon the pipeline developed in the European project FIRST. We have adapted the pipeline according to the specific needs of SIMPOL by extending the list of news sources being followed to include those relevant to the domain of climate finance. We have changed the database for storing the articles from SQL to a high-performance document store ElasticSearch, which is designed for efficient retrieval of documents from very large document collections using a comprehensive query language. We developed a web-based front end which employs the querying capabilities of ElasticSearch to retrieve news articles based on search queries (similar to Google News). We constructed a filter for separating news articles relevant to the domain of climate finance from those that are irrelevant and developed a web-based interface that can be accessed at

News acquisition and preprocessing pipeline
The SIMPOL news acquisition and processing pipeline is based on a news acquisition and processing pipeline that was developed in the European project FIRST. Here we give a high-level overview of the steps involved in the FIRST pipeline and then explain the upgrade to the SIMPOL pipeline.

The FIRST pipeline consists of (i) data acquisition components, (ii) data cleaning components, (iii) natural-language preprocessing components, and (iv) semantic annotation components.

Environmental news sources
The acquisition pipeline requires a file with RSS sources to work. These sources are periodically polled for content. The file format is relatively simple and contains several lists of RSS sources, one for each Web site.

SIMPOL News database
The news articles acquired and processed by the SIMPOL pipeline are stored in an Elasticsearch database. Elasticsearch ( is a search engine based on Lucene. It provides a distributed, multitenant-capable full-text search engine with a HTTP web interface and JSON documents. Elasticsearch is developed in Java and is released as open source under the terms of the Apache License.

Filtering environmental news
The acquisition pipeline described in this report collects over 20,000 news articles per day. The collected news articles are spread across a variety of topics that are typical for media outlets including politics, economy, culture, sports, and many others.

News searching interface
The database of news can be searched through the web-based interface located at: The search interface supports a comprehensive query language which is transformed into native Elasticsearch queries and sent via the REST API to the Elasticsearch database.

Understanding financial news with multi-layer network analysis
We addressed the question “What is in the news?” by constructing and comparing multi-layer networks from different sources. We examined the overlap of the most important links in the constructed networks, and compare their structural similarity by node centrality and main k-cores. A comparative analysis reveals that the co-occurrences of countries in the news most closely match their geographical proximity, while positive sentiment links most closely matches the trade connections between the countries. Correlations between financial indicators have the lowest similarity to financial news.

Temporal multi-layer network construction from major news events
We present an approach which extracts interesting events from thousands of daily news. We construct a time-varying, three-layer network where the nodes are entities of interest in the news. We demonstrate the news network evolution over a period of four years in an interactive web portal.

Twitter crowdsourcing based on news articles
Complementary to the work in Task 3.3: Crowdsourcing the engaged public, we have set up an infrastructure that uses Twitter as a medium for crowdsourcing. Instead of filling-in a survey-like questionnaire, users interact with a friendly user interface to post tweets that report the stances of organizations towards various environmental issues stemming from the 2030 framework.


Task 3.5 Mining social media

Research questions
The research questions addressed here are relevant for Objective 5: Develop an ICT infrastructure to gather semantic web data for policy relevant socio-economic networks. The main research question is: How to extract and aggregate the public pulse about socio-economic issues from streams of big data from social media?

Activities and results
Much of the work reported in this document has been carried out in the last part of the project, during M25-M40. Understanding people’s reaction to policy topics such as sustainability is of vital importance for planning policies regarding e.g. climate change. It is especially important to understand how people acquire information on a given topic and develop an opinion. Recent research on social dynamics has shown that social media play a key role in the information acquisition process, by acting as “disintermediators”, which enable a direct path from producers to consumers of contents. Figuring out how such ideas diffuse through social media may be key to reach the public and create solid basis for citizens empowering. A thorough study of users’ activity on Twitter and Facebook can thus reveal the way people “process information” on a series of topics of interest.

One of the main messages stemming from the stream of work on social media dynamics within the SIMPOL project is that information campaigns appear to be necessary steps to be taken before the adoption of any policy on topics like climate change. These campaigns should be intended to neutralize the effect of misinformation on the same topics: they, thus, should focus on the way people approach information, by making citizens aware of the effects of their background knowledge on the acquisition of new knowledge.

These results can be seen as complementary to some of the findings from WP2 on the networks of influence in climate-finance. For instance, in the domain of climate, networks of influence reflect existing economic interests. While, according to the popular view, the “disintermediation” role played by social media would allow people to bypass the influence of lobbies (which, supposedly, extends to traditional information media), in turn allowing citizens to access genuine, “unfiltered” information, this turns out to be false. Social media, in fact, just act as reinforces of one’s beliefs. As a consequence, in order for any policy to be successfully accepted by people, a preliminary information campaign must be conducted.

In order to achieve accurate sentiment predictions on Twitter, we construct domain and language specific sentiment models trained on manually annotated data. These models achieve predictive performance close to the inter-annotator agreement level. We also build the first emoji sentiment lexicon, which we propose as a European language-independent resource for automated sentiment analysis. We further monitor the Twitter activities of members of the European Parliament. The resulting retweet networks reveal real-world relationships, as well as general insights into the cohesion and voting behavior of political groups.

Additionally, we study on Facebook how information related to very distinct narratives (on the one hand mainstream scientific news, and, on the other hand, conspiracy news, such as the one known as “Chemtrails” conspiracy theory) are consumed by users and shape communities. Polarized, well separated communities (echo chambers) emerge around distinct types of contents, and users’ engagement on a specific content correlates with the number of friends having similar consumption patterns. For both science and conspiracy contents, the longer the discussion the more the negativity of the sentiment grows. We finally propose an Unbounded Confidence Model to explain the coexistence of two stable final opinions, as well as a percolation model of rumor spreading which demonstrates that homogeneity and polarization are the main determinants for predicting the size of information cascades.

The work performed using Twitter includes:

  • Sentiment analysis
  • Sentiment of emojis
  • Retweet networks of the European Parliament
  • Voting and tweeting in the European Parliament

The work performed using Facebook includes:

  • Dataset construction
  • Viral misinformation: The role of homophily and polarization
  • Content and sentiment analysis for echo chambers
  • Determinants of misinformation spreading and cascades

Task 5.1 Publications and Presentations

The Dissemination activity of the SIMPOL project is on track.
  1. We have produced several publications (41 published papers, 15 among conference proceedings, books and book chapters, and 9 working papers).
  2. We have organized several conferences and workshops in which we have involved both academics and policy makers.
  3. We have started 3 new collaborations with ECB, ESRB and BoE, which have lead to the uptake of Global Systems Science (GSS) tools at policy level in the area of financial stability.
  4. We have also initiated a series of discussions and partnerships with stakeholders interested in stress-testing exercises applied to the context of climate policy, including the World Bank, the BIS and the WWF and the G20 study group on climate-finance.
  5. We co-organized several events associated with the climate negotiations COP21 in Paris and COP22 in Marrakesh.
The work of the SIMPOL project has resulted in several top journals peer-reviewed publications: i.e. 3 PNAS, 1 Nature Communications and 1 Nature Climate Change. In particular, we would like to emphasize that the Climate Stress-test of the Financial System has been the result an entirely new line of research initiated only during the SIMPOL project and despite its recent development it made into one of the most influential journals in the area of Climate Change, a journal that is regularly followed also by policy makers.

We would also to emphasize that the consortium was successful in reaching the academic community of economics and finance through:

  • Several publications in journals such as Journal of Financial Stability (2), Journal of Economic Dynamics and Control (2), Ecological Economics (1), Journal of Network Theory in Finance, Statistics & Risk Modeling (1).
  • Several collaborations with researchers within policy making bodies (e.g. ECB, ESRB, BoE)
  • The co-organization of policy-relevant events (e.g. conference with Journal of Financial Stability)
  • Several presentations at policy meetings (e.g. EBA and ESRB workshops)
  • Presentations at mainstream economics conferences (e.g. co-organization of session at the annual conference of the American Economic Association in Chicago, January 2017)


Task 5.2 Dissemination to a Wider Audience

Reaching the policy makers community. We have disseminated the policy implications of our results among policy makers in presentations at policy relevant conferences as well as meetings at policy making bodies and by co-organizing events involving policy makers.

Reaching the mainstream economics community. The fruitful collaboration with the ESRB has resulted also in the participation/co-organisation of a series of events. In particular, ESRB and UZH researchers have submitted a proposal to organise a whole session based on financial micro-data at the next American Economic Association annual meeting in Chicago (January 2017).

Interact with other FET projects in the same proactive call and other calls. We have established/continued collaborations with other FET projects such as MULTIPLEX and DOLFINS as well as non-FET EU projects in the domain of economic policy such as ISIGROWTH, in the domain of global system science such as CoeGSS, and in the domain of climate-finance such as SEIMETRIC.

Contribute to events related to the Global System Science. We have contributed to various GSS events such as Genoa Sept 2015, CCS2015 and CCS2016 satellites, OECD NAECS Complexity and Policy.

Lectures and events for the general public. Following the work plan in the project proposal we have been undertaking a series of activities aimed at disseminating the results of the project to a wider audience.

Policy Jams and crowdsourcing the analysis of policy network maps. We have developed a crowdsourcing tool with the purpose to increase the transparency on EU public consultations. The specialized user interface allows the interested public to explore the policy relevant networks and contribute new portions of the data. In particular, the interface enables the users to explore relationships among certain economic entities regarding their views on different policy issues and provides additional data on their lobbying activity.

Global Systems Science MOOC. Franziska Schütze (GCF) became an educator in the massive open online course (MOOC) “Global Systems Science and Policy: an Introduction”, organized by Jeff Johnson via the Future Learn platform. The crowdsourcing campaign was promoted in the course as an example for “citizens in the policy loop” and “citizens as policy makers”.

Discussion and meetings with stakeholders. We also organized meetings with representatives from NGO’s and other stakeholders with the purpose of 1) gaining insights into barriers and enablers for citizens’ participation to our crowdsourcing experiments, and 2) disseminate the results of the SIMPOL project activities to a broader audience.


Task 5.3 Schools and Conferences

We have organised or co-organised a series of conferences, particularly for M25-M40:

  • Workshop “Statistical Physics of Financial and Economic Networks” (Paris, France, 15-16 July 2016).
  • Summer School “Complex networks: from socio-economic systems to biology and brain” (Lipari, Italy, 29 August – 02 September 2016).
  • “The First Workshop on MIning DAta for financial applicationS” (Riva del Garda, Italy, 19 September 2016).
  • Satellite meeting to CCS2016: “Dynamics of Multilevel Complex Systems” (Amsterdam, The Netherlands, 21 September 2016).
  • “Conference on GSS & Policy” (London, United Kingdom, 27-28 November 2016).
  • “First Conference on Financial Networks and Sustainability” (Zurich, Switzerland, 11-13 January 2017).
  • “International Conference on Synthetic Populations” (Lucca, Italy, 22-23 February 2017).


Diagrams illustrating and promoting the work of the project


Illustrations from the paper “A climate stress-test of the financial system”, available at:

Diagram illustrating the reclassification of sectors from NACE Rev2 codes into climate-policy-relevant sectors. Paper available at:


Equity holdings in EU and US listed companies in 2015 (data from Bureau Van Dijk Orbis). a, Exposures to climate-policy-relevant sectors of aggregate financial actors worldwide. b, Exposures to climate-policy-relevant sectors of selected investment funds worldwide (top 15 by size of equity portfolio in the data). c, Exposures to climate-policy-relevant sectors of selected banks worldwide (top 15 by size of equity portfolio in the data). Paper available at:


First- and second-round losses in banks’ equity for the 20 most-severely affected EU listed banks, under the Fossil fuel + Utilities 100% shock. Subsidiaries have not been taken into account. Paper available at:


Individual banks’ value at risk under green and brown investment strategies. Value-at-risk at the 5% significance level of the 20 most-severely affected EU listed banks in the dataset, under the scenario that they follow the green investment strategy (a) or the brown investment strategy(b). Darker colour refers to VaR(5%) computed on the distribution of first-round losses only, while lighter colour refers to VaR(5%) computed on the sum of first- and second-round losses. Paper available at:



Illustration from the paper “Pathways towards instability in financial networks”, available at:

Stability of the network of the top 50 European banks using data from their 2013 balance sheets. Y-axis: largest eigenvalue of the interbank leverage matrix associated with the top 50 largest EU banks. Interbank exposures are estimated based on BvD Bankscope data. X-axis: network density measures the fraction of pairs of banks engaging in interbank contracts out of all the possible pairs, along the simulation trajectory. Paper available at:


Illustration from the paper “The price of complexity in financial networks”, available at:

Example of a contract network among three banks. Each bank balance sheet consists of assets and liabilities. The assets of one bank are liabilities of another bank. Credit contracts are represented as arrows (pale red) from the lender to the borrower. Pale blue arrows represent investment in assets issued by entities external to the banking system. Paper available at



Bow-tie structure and flow-of-risk. Stylised representation of the bow-tie structure in the CDS market: risk flows from the IN component (the URS) to the SCC set (dealers) and eventually ends up in the OUT component (the URB). Paper available at: available at:  

CDS network visualisations (aggregate network) for the four snapshots. Ultimate Risk Sellers (URS) and Ultimate Risk Buyers (URB) are ordered from the top to the bottom by notional traded at first by each sector and then within the sector. Paper available at: available at:



Further illustrations:

(c) the interlocking links between different groups. (d) The scatter plot of the lobby cohesiveness versus the lobby money within groups. The size of the bubbles in (a)(b)(c) is proportional to the number of organizations in the group. The size of the link between two groups is proportional to the number of links from the organizations in one group to the organizations in the other group in the original network. The size of the points in (d) is proportional to the total number of lobbyists of the group. Source: “The Multiplex Network of EU Lobby Organizations”, An Zeng, Stefano Battiston (2016). Paper available at:




Illustration from the ESRB Working Series paper “Sentiment leaning of influential communities in social networks”, available at:

Twitter communities and sentiment around climate issues. Sluban et al. 2015. Paper available at:



Below are screenshots from the SIMPOL knowledge base platform. A full-text search engine has been implemented. In the example searching for a given organization yields the results in one or more subgraphs in the Neo4j database. The relations are combined across subgraphs and the resulting graph can be exported.