ARTICLES

PUBLISHED ONLINE: 27 MARCH 2017 | DOI: 10.1038/NCLIMATE3255

A climate stress-test of the financial system

Stefano Battiston1*, Antoine Mandel2, Irene Monasterolo3, Franziska Schütze4 and Gabriele Visentin1

The urgency of estimating the impact of climate risks on the financial system is increasingly recognized among scholars and
practitioners. By adopting a network approach to financial dependencies, we look at how climate policy risk might propagate
through the financial system. We develop a network-based climate stress-test methodology and apply it to large Euro Area
banks in a ‘green’ and a ‘brown’ scenario. We find that direct and indirect exposures to climate-policy-relevant sectors
represent a large portion of investors’ equity portfolios, especially for investment and pension funds. Additionally, the portion
of banks’ loan portfolios exposed to these sectors is comparable to banks’ capital. Our results suggest that climate policy
timing matters. An early and stable policy framework would allow for smooth asset value adjustments and lead to potential
net winners and losers. In contrast, a late and abrupt policy framework could have adverse systemic consequences.

A
ssessing the impact of climate risks and climate policies on several types of financial instruments. This analysis is meant as a
the financial system is currently seen as one of the most tool to support further investigations of the potential impact and
urgent and prominent policy issues1,2 . In particular, there is the political feasibility of specific climate policies20,21 . To go beyond
a debate on whether the implementation of climate policies to meet the mere exposure to the fossil-fuels extraction sector, we remap an
the 2 ◦ C target generates systemic risk or, instead, opportunities for existing standard classification of economic sectors (NACE Rev2)
low-carbon investments and economic growth. However, data are according to their relevance to climate mitigation policies, and
scarce and there is no consensus on the appropriate methodologies we analyse empirical microeconomic data for shareholders of
to use to address this issue. The magnitude of so-called stranded listed firms in the European Union and in the United States.
assets of fossil-fuel companies (in a 2 ◦ C economy) has been We find (see Supplementary Table 6) that while direct exposures
estimated to be around 82% of global coal reserves, 49% of global via equity holdings to the fossil-fuel sector are small (4–13%
gas reserves and 33% of global oil reserves3 . Moreover, several across financial actor types), the combined exposures to climate-
studies have investigated the role of stranded assets in specific policy-relevant sectors are large (36–48%) and heterogeneous. In
sectors and countries4–9 . By investing in fossil-fuel companies, addition, financial actors hold equity exposures to the financial
financial institutions hold direct ‘high-carbon exposures’, which for sector (13–25%), implying indirect exposures to climate-policy-
European actors have been estimated to be, relative to their total relevant sectors.
assets, about 1.3% for banks, 5% for pension funds and 4.4% for
insurances10 . One can compute the value at risk (VaR) associated Results
with climate shocks11 in the context of integrated assessment By targeting the reduction of greenhouse gas (GHG) emissions,
models12 in which aggregate financial losses are derived top-down climate policies can affect (positively or negatively) revenues and
from estimated GDP (gross domestic product) losses due to physical costs of various sectors in the real economy with indirect effects on
risks resulting from climate change. Yet, assessing the financial risk financial actors holding securities of firms in those sectors. However,
of climate policies (often referred to as transition risks) requires the existing classifications of economic sectors such as NACE
estimations of the likelihood of the introduction of a specific Rev2 (ref. 22) or NAICS (ref. 23) were not designed to estimate
policy. However, the likelihood that a climate policy is introduced financial exposures to climate-policy-relevant sectors. Therefore, we
depends on the expectations of the agents on that very likelihood. define a correspondence between sectors of economic activities at
Thus, the intrinsic uncertainty of the policy cycle undermines the NACE Rev2 4-digit level and five newly defined climate-policy-
reliability of the probability distributions of asset returns, also due relevant sectors (fossil fuel, utilities, energy-intensive, transport and
to the presence of fat tails13 . Further, it is now understood that housing) based on their GHG emissions, their role in the energy
interlinkages among financial institutions can amplify both positive supply chain, and the existence in most countries of related climate
and negative shocks14–16 and significantly decrease the accuracy policy institutions (see Methods and Fig. 1).
of our estimation of default probabilities in an interconnected The exposures of financial actors (classified according to the
financial system17 . As a result, calculations of expected losses/gains standard European Systems of Accounts, ESA (ref. 24)) can
from climate policies carried out with traditional risk analysis be decomposed along the main types of financial instruments:
methodologies have to be taken with caution. Here, we develop equity holdings (for example, ownership shares including both
a complementary approach, rooted in complex systems science, those tradable on the stock market and those non-tradable),
and consisting of a network analysis of the exposures of financial bond holdings (for example, tradable debt securities) and loans
actors18,19 to all climate-policy-relevant sectors of the economy, as (for example, non-tradable debt securities). By combining the
well as the exposures among financial actors themselves, across breakdown of exposures across instruments with the reclassification

1 Department of Banking and Finance, University of Zurich, Andreasstr. 15, 8050 Zürich, Switzerland. 2 Université Paris 1 Panthéon-Sorbonne, Centre
d’économie de la Sorbonne, Maison des sciences économiques, 106-112 Boulevard de l’hôpital, 75647 Paris Cedex 13, France. 3 Frederick S. Pardee Center
for the Study of the Longer Range Future, Boston University, 67 Bay State Road, Boston, Massachusetts 02215, USA. 4 Global Climate Forum, Neue
Promenade 6, 10178 Berlin, Germany. *e-mail: stefano.battiston@uzh.ch

NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange 1

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLES NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3255

Table 1 | Absolute (first row, in US$ billions) and relative (second row, percentage of aggregate equity portfolio) exposure of each
financial actor type to each sector.
OCIs GOV Individuals Banks IPFs OFSs NFCs IFs
(955) (125) (33,733) (798) (6,392) (3,081) (14,851) (5,124)
Fossil-fuel 31.17 66.17 98.17 173.29 230.21 185.15 377.30 549.85
(767) 6.02% 11.43% 3.77% 6.34% 7.09% 5.33% 8.06% 6.05%
Utilities 19.32 63.58 21.16 77.02 55.53 65.46 93.09 249.32
(216) 3.73% 10.99% 0.81% 2.82% 1.71% 1.88% 1.99% 2.74%
Energy-intensive 172.84 147.53 766.33 708.30 865.87 1,019.84 1,408.65 2,701.69
(3,956) 33.40% 25.49% 29.47% 25.92% 26.68% 29.36% 30.08% 29.71%
Housing 13.26 15.88 100.57 59.07 85.28 76.60 146.72 189.36
(797) 2.56% 2.74% 3.87% 2.16% 2.63% 2.21% 3.13% 2.08%
Transport 11.43 18.48 55.38 47.67 54.48 69.96 106.67 173.02
(224) 2.21% 3.19% 2.13% 1.74% 1.68% 2.01% 2.28% 1.90%
Finance 127.01 95.33 419.63 684.72 609.11 669.82 702.44 1,532.08
(2,659) 24.54% 16.47% 16.14% 25.06% 18.77% 19.29% 15.00% 16.85%
Other 142.44 171.80 1,139.53 982.46 1,345.08 1,386.27 1,847.40 3,698.41
(6,259) 27.53% 29.68% 43.82% 35.95% 41.44% 39.91% 39.46% 40.67%
Numbers in brackets indicate the number of firms in this group of actors or sectors. OCIs, Other Credit Institutions; GOV, Government; IPFs, Insurance and Pension Funds; OFSs, Other Financial
Services; NFCs, Non-Financial Corporations; IFs, Investment Funds.

Reclassification of economic sectors from Reclassification of asset portfolios

NACE Rev2 into climate-policy-relevant sectors
Asset portfolio: Asset portfolio by
NACE Rev2 Climate-policy- breakdown by climate-policy-
codes relevant sectors instrument relevant sector

B Fossil-fuel

Equity
Utilities

C
Bonds
Energy-intensive

D
Loans
Housing
F

H Transport

Figure 1 | Diagram illustrating the reclassification of sectors from NACE Rev2 codes into climate-policy-relevant sectors. For more information see the
Methods and Supplementary Table 3.

of securities, we compute the total direct exposure of a given methodology, we construct the portfolio of each shareholder and
financial actor to each climate-policy-relevant sector (see Methods). we compute its exposure to each climate-policy-relevant sector. To
gain insights into the magnitude of indirect exposures we further
Direct financial exposure through equity holdings classify equity holdings in companies belonging to the financial
To provide empirical estimates of exposures to climate-policy- sector. We group shareholders by financial actor type to include,
relevant sectors, we apply our methodology to recent available besides the institutional financial sectors from the ESA classification
data sets. Despite their relevance for policy purposes, data about (that is, Banks, Investment Funds, Insurance and Pension Funds)
securities holdings of financial institutions, in particular to climate- also Individuals, Governments, Non-Financial Companies, Other
policy-relevant sectors, is generally scarce, inconsistent or even Credit Institutions and Other Financial Services (Table 1).
undisclosed. Along the three main instrument types mentioned Figure 2a shows the result of the aggregated exposures in terms
above (equity, bonds and loans), at the level of individual institutions of equity holdings in listed companies for each financial actor
only some data of equity holdings are publicly available. type. The combined shares of equity holdings held by the financial
We thus first analyse a sample obtained from the Bureau Van Dijk sector (that is, Investment Funds, Insurance and Pension Funds,
Orbis database covering all EU and US listed companies and their Banks, Other Credit Institutions, and Other Financial Services)
disclosed shareholders (14,878 companies and 65,059 shareholders) amount to about 32.4 trillion US dollars, equivalent to 58.7% of total
at the last available year, that is, 2015. On the basis of our market capitalization.

2 NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3255 ARTICLES
a Investment Sectors
funds Fossil-fuel
Industrial Utilities
companies Energy-intensive
Housing
Other financial
Transport
services
Finance
Insurance and Other
pension funds

Banks

Individuals

Governments

Other credit
institutions

0 1 2 3 4 5 6 7 8 9
Equity exposure (US$ trillion)

b Blackrock Sectors
Vanguard Group
Fossil-fuel
State Street Corp
Utilities
FMR LLC
T Rowe Price Group Energy-intensive
T Rowe Price Assoc. Housing
Sun Life Financial Transport
Franklin Resources Finance
Morgan Stanley Other
UBS Group AG
Goldman Sachs G.
Geode Capital Hold.
Prudential Financial
Power Corp. of Canada
Schroders PLC
0 200 400 600 800 1,000 1,200 1,400
Equity exposure (US$ billion)

c Jpmorgan Chase Sectors

Bank of NY Mellon
Fossil-fuel
Northern Trust
Utilities
Bank of America
Deutsche Bank Energy-intensive
Wells Fargo Housing
BPCE Transport
Toronto-Dom. Bank Finance
BNP Paribas Other
R. Bank of Canada
PNC Financ. Serv.
Société Générale
Bank of Montreal
Norges Bank
Lazard
0 50 100 150 200 250 300
Equity exposure (US$ billion)

Figure 2 | 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).

The following findings emerge. First, the relative equity portfolio exposures to climate-policy-relevant sectors. Within each financial
exposures of all financial actors types to the fossil-fuel sector are actor type, the standard deviation of exposures across individuals
limited (that is, ranging from 4.4% for Individuals to 12.9% for (see Supplementary Table 6) reflects the level of heterogeneity across
Governments) (see Supplementary Table 6). Second, their relative individuals’ portfolio compositions. Examples of individual equity
equity portfolio exposures to all climate-policy-relevant sectors are holdings’ compositions are shown in Fig. 2b,c for the twenty largest
large (that is, ranging from 45.2% for Insurance and Pension Funds, players among investment funds and banks.
to 47.7% for Governments), and mostly accounted for by the energy-
intensive sector. Third, since financial actors’ exposures to the Climate stress-testing EU largest banks
financial sector itself range from 13% for Industrial Companies up Several quantitative estimates exist for the macroeconomic impacts
to 25.8% for Other Credit Institutions, they bear additional indirect of climate change and climate policies25,26 , as well as for the value

NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange 3

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLES NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3255

Deutsche Bank
Crédit Agricole
Commerzbank
Natixis
Nordea Bank
Crédit Industriel et Co..
Standard Chartered
Raiffeisen Bank Intern.
UniCredit
Dexia
Société Générale
Skand. Enskilda Banken
Barclays
Delta Lloyd
Aareal Bank
Jyske Bank
KBC Groep
Monte Paschi
BNP Paribas
Svenska Handelsbanken
0% 5% 10% 15% 20% 25% 30%

First round Second round

Figure 3 | 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.

of stranded assets6 . Accordingly, probabilistic estimates of the in a severe scenario, there is no systemic impact when considering
climate VaR can be carried out from an aggregate perspective11 . only the equity holdings channel.
However, these estimates are too broad to define shock scenarios More refined scenarios, allowing one to compute a VaR for each
for individual institutions. At a more granular level, estimates of bank, require one to have distributions of shocks across climate-
the value of stranded assets are available in the literature but their policy-relevant sectors, which are not available in the literature
sectoral coverage is currently too narrow to inform an analysis of at this stage. As a first step in this direction, in our second
systemic impacts. exercise, we construct distributions of shocks for the fossil-fuel and
To overcome these limitations, we extend the stress-test method- utility sectors based on the economic impact assessment of climate
ology developed in refs 27,28, which allows one to disentangle the policies provided by the LIMITS database26 and we consider several
two main contributions to systemic losses. First-round losses are scenarios of banks’ exposures to climate-policy-relevant sectors
defined as losses in banks’ equity due to direct exposures to shocks. (see Methods).
Second-round losses are defined as indirect losses in banks’ equity In particular, we interpret scenarios (2) and (4) in terms of
due to the devaluation of counterparties’ debt obligations on the distributions of losses suffered by a ‘representative’ (average) bank
interbank credit market. The magnitude of second-round effects can adopting one of two different investment strategies:
vary significantly. Traditional methods (based on ref. 29), yielding ˆ (2) a ‘green’ bank having all its equity holdings in utilities invested
small second-round effects, are appropriate only under specific in renewables-based utilities and having no equity holdings in the
market conditions (that is, full recovery from counterparties’ asset fossil-fuel sector,
liquidation and no mark-to-market valuation of debt obligations). ˆ (4) a ‘brown’ bank having all its equity holdings in utilities
In general, instead, second-round effects can be comparable in invested in fossil-fuel-based utilities and keeping its equity
magnitude to first-round effects15,27,28,30 . holdings in the fossil-fuel sector.
We illustrate how our methodology can be used to conduct a Supplementary Table 10 reports the main statistics on the global
climate stress-test of the banking system based on microeconomic relative equity loss in the banking system. The results of the two
data at the level of individual banks, by carrying out two exercises exercises are consistent: the system’s VaR in the brown scenario
on the set of the top 50 listed European banks by total assets is less than 1% of the total banks’ capital. Supplementary Table 3
(see Methods). reports the statistics for the ‘representative’ brown and green
In the first exercise we aim to determine an upper bound on the bank: depending on whether their exposure to utilities is mainly
magnitude of the losses induced by climate policies by considering concentrated on renewables-based utilities or on fossil-fuel ones and
a set of scenarios in which the whole equity value of the firms in if they are exposed to the fossil-fuel sector, banks might face very
the shocked sector would be lost. We can then compute for each different impacts from climate policies. Further, Supplementary
bank the ratio of the exposures to climate-policy-relevant sectors Fig. 6 shows the distribution of first-round losses: the brown
over the banks’ capital (that is, banks’ equity on the liability side bank incurs more losses than the green one, but these losses are
of their balance sheets). Different scenarios consist of different small in comparison with the equity of the average bank (that is,
combinations of sectors as indicated in Supplementary Table 8, by US$32 billion) and with its total asset (that is, US$604 billion).
increasing levels of shocks’ severity. For instance, in the second Finally, Fig. 4a,b reports the VaR for the 20 most affected banks both
scenario, 100% of the market capitalization of listed firms both in the in the brown and in the green scenario.
fossil-fuel sector and in the utilities sector is lost. Figure 3 shows the The limited magnitude of banks’ losses in this exercise is due to
losses as a percentage of the banks’ capital across the 20 most affected the fact that Euro Area banks bear little equity holdings compared
banks as a result of the second scenario from Supplementary Table 8. with their balance sheet (about 1.2T EUR, that is, 3.8% of total assets
Light (dark) grey bars indicate the losses from the first- (second-) and 48% of capital), probably due to higher capital requirements for
round shocks. Notice that some banks have no first-round losses equity holdings31 . However, banks bear larger exposures on loans to
but have important losses at the second round. None of the largest non-financial corporations (about 4.8T EUR = 13.8% of total assets
banks could default solely due to their exposures to climate-policy- and 192% of their capital). Unfortunately, Euro Area banks’ loans
relevant sectors on the equity market. This result implies that even are only available at 1-digit NACE Rev2 aggregation32 . At this stage,

4 NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3255 ARTICLES
a Deutsche Bank Orbis), the following findings emerge for the Euro Area. First,
Crédit Agricole the major direct exposures to climate-policy-relevant sectors
Commerzbank
Natixis of investment funds and pension funds are concentrated in
Nordea Bank equity holdings, while for banks they are concentrated on loans.
Crédit Industriel et Co.
Standard Chartered
Interestingly, bond holdings are only a minor channel of direct
Raiffeisen Bank Intern. exposure to climate-policy-relevant sectors because outstanding
Unicredit bonds issued by non-financial firms in the Euro Area amount to
Dexia
Société Générale about 1 trillion Euro, that is, about only one-fifth of the values
Skand. Enskilda Banken of equity shares issued by the same type of firms. Indeed, only
Barclays
Delta Lloyd
less than 7% of bonds are issued by firms in the real sectors, with
Aareal Bank roughly 40% issued by governments and another 45% issued by
Jyske Bank financial institutions.
KBC Groep
Monte Paschi Second, financial actors bear also indirect exposures to climate-
BNP Paribas policy-relevant sectors. For instance, pension funds hold an
Svenska Handelsbanken
exposure of about 25% of their total assets in equity shares of
0 100 200 300 400 investment funds, which in turn have an estimated exposure of
VaR(5%) on asset losses (US$ million) about 25% of total assets in equity holdings of climate-policy-
relevant sectors. Pension funds also hold an exposure of 15% of
their total assets in bonds and loans to banks, which, on the basis
b Deutsche Bank
Crédit Agricole of the previous section, hold an estimated exposure of about 14%
Commerzbank of total assets to climate-policy-relevant sectors. In contrast, the
Natixis
Nordea Bank
direct exposure of pension funds to climate-policy-relevant sectors
Crédit Industriel et Co. through equity holdings is about 8% of total assets. These findings
Standard Chartered imply that shocks on the fossil sector and increased volatility on
Raiffeisen Bank Intern.
Unicredit asset values in the other climate-policy-relevant sectors could affect
Dexia non-negligible portions of pension funds’ assets through both direct
Société Générale (8.3%) as well as indirect exposures (about 8%).
Skand. Enskilda Banken
Barclays
Delta Lloyd Conclusions
Aareal Bank
Jyske Bank By remapping the existing classification of economic activities
KBC Groep (NACE Rev2) into newly defined climate-policy-relevant sectors,
Monte Paschi
BNP Paribas
we find that direct and indirect exposures to such sectors represent
Svenska Handelsbanken a large portion of financial actors’ equity holdings portfolios (in
0 1,000 2,000 3,000 particular for investment funds and pension funds). Moreover,
VaR(5%) on asset losses (US$ million) exposures represent a portion of banks’ loan portfolios comparable
to banks’ capital. Further, we develop a network-based climate
Figure 4 | Individual banks’ value at risk under green and brown stress-test methodology that can be used to derive statistics of losses
investment strategies. Value-at-risk at the 5% significance level of the for individual financial actors, including VaR. We illustrate the
20 most-severely affected EU listed banks in the data set, under the methodology on a sample of the top 50 largest EU banks taking
scenario that they follow the green investment strategy (a) or the brown into account first- and second-round effects of shocks to their
investment strategy (b). Darker colour refers to VaR(5%) computed on the equity portfolios.
distribution of first-round losses only, while lighter colour refers to Our findings suggest that the implementation of climate
VaR(5%) computed on the sum of first- and second-round losses. mitigation policies is key, both in terms of timing and expectations.
The extent to which financial exposures will translate into shocks
we cannot compute individual exposures of banks to climate-policy- depends on the ability of market participants to anticipate climate
relevant sectors via their loans. Sector level data for 2014 from the policy measures. If climate policies are implemented early on and
ECB Data Warehouse provide the following aggregate estimations in a stable and credible framework, market participants are able
for the banks’ exposures on loans as a fraction of banks’ capital: to smoothly anticipate the effects. In this case there would not be
11.4% for fossil and utilities; 28% for energy-intensive; 16% for any large shock in asset prices and there would be no systemic
transportation; 73% for housing. We also need to consider banks’ risk. In contrast, in a scenario in which the implementation of
loans to households (presumably mostly granted for mortgages), climate policies is uncertain, delayed and sudden2,10 (for example,
which add a further 208% of exposures in the housing sector as a as a reaction to increased frequency of extreme weather events and
fraction of capital. to align with the COP21 agreement), market participants would
Better disclosure of climate-related financial exposures33 would not be able to fully anticipate the impact of policies. In this case,
allow one to improve calculations for individual banks. The above given the large direct and indirect exposures of financial actors
considerations suggest that banks would not default solely due to to climate-policy-relevant sectors, this might entail a systemic risk
their loan exposures to firms in the fossil-fuel and utilities sectors. because price adjustments are abrupt and portfolio losses from the
However, if climate policies imply higher volatility of loans’ values fossil-fuel sector and fossil-based utilities do not have the time to be
in the energy-intensive and transport sector or in the housing compensated by the increase in value of renewable-based utilities.
sector and for mortgages, this would translate into volatility of large These two scenarios and their corresponding VaR are illustrated by
portions of banks’ assets, relative to their capital (16% + 28% = 44% the loss distributions for a ‘green’ and a ‘brown’ investing strategy in
and 73% + 208% = 281%, respectively). our climate stress-test on EU banks.
Moreover, the fact that financial actors bear large exposures
Indirect exposures of European financial actors to climate-policy-relevant sectors implies that climate mitigation
By cross-matching aggregate balance sheet information for financial policies could increase volatility on large portions of their portfolios.
actors (from ECB Data Warehouse) with equity holdings (from Climate mitigation policies are commonly thought to have an

NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange 5

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLES NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3255

adverse effect on the value of assets in the fossil-fuel sector5 , as 17. Battiston, S., Roukny, T., Stiglitz, J., Caldarelli, G. & May, R. The price of
well as an adverse effect on the whole economy (see Ch. 6 of complexity. Proc. Natl Acad. Sci. USA 113, 10031–10036 (2016).
ref. 25). However, a transition to a low-carbon economy could 18. May, R. M., Levin, S. A. & Sugihara, G. Complex systems: ecology for bankers.
Nature 451, 893–895 (2008).
also have net positive aggregate effects34 . Overall, the effects of 19. Haldane, A. G. & May, R. M. Systemic risk in banking ecosystems. Nature 469,
climate policies are likely to vary across firms and sectors: for 351–355 (2011).
example, the renewable energy and the energy efficiency sectors 20. Rogelj, J. et al. Emission pathways consistent with a 2 ◦ C global temperature
are expected to increase massively in market share (see ref. 35, limit. Nat. Clim. Change 1, 413–418 (2011).
IEA report 2015; IRENA Annual Review 2016), while real-estate 21. Peters, G. P. The ‘best available science’ to inform 1.5 ◦ C policy choices.
assets can increase or decrease in value, depending on their energy Nat. Clim. Change 6, 646–649 (2016).
22. NACE Rev. 2—Statistical Classification of Economic Activities (Eurostat, 2008);
performance (see Supplementary Table 6.7 in ref. 25). Further, stock http://ec.europa.eu/eurostat/web/nace-rev2
price volatility in climate-policy-relevant sectors can increase as a 23. North American Industry Classification System (United States Census Bureau,
result of: technological innovation36,37 , increased competition38 and 2017); http://www.census.gov/eos/www/naics
policy uncertainty39 . Therefore, climate policy could lead to winners 24. European System of National and Regional Accounts (Eurostat, 2010);
and losers (in absolute terms) across financial actors, depending on http://ec.europa.eu/eurostat/web/esa-2010
the composition of their portfolios. 25. IPCC Climate Change 2014: Mitigation of Climate Change
(eds Edenhofer, O. et al.) (Cambridge Univ. Press, 2015).
Overall, our network analysis of financial exposures highlights 26. Kriegler, E. et al. What does the 2 ◦ C target imply for a global climate
that financial actors’ portfolios are both interdependent and largely agreement in 2020? The LIMITS study on Durban Platform scenarios.
exposed to the outcome of the climate policy cycle. This implies the Clim. Change Econ. 4, 1340008 (2013).
possibility of multiple equilibria without a clear way to assign ex ante 27. Battiston, S., Puliga, M., Kaushik, R., Tasca, P. & Caldarelli, G. DebtRank:
probabilities for each equilibrium to occur. Therefore, while climate- too central to fail? Financial networks, the FED and systemic risk. Sci. Rep. 2,
related financial information disclosure is crucial for risk evaluation, 541 (2012).
28. Battiston, S., Caldarelli, G., D’errico, M. & Gurciullo, S. Leveraging the network:
a stable policy framework is necessary to resolve the multiplicity a stress-test framework based on DebtRank. Stat. Risk Model. 33, 1–33 (2016).
of possible outcomes. To this end, a network-based, conditional 29. Eisenberg, L. & Noe, T. H. Systemic risk in financial systems. Manage. Sci. 47,
VaR approach represents an advancement in the analysis of climate- 236–249 (2001).
policy risks and their implications for the financial sector. 30. Battiston, S., D’Errico, M. & Visentin, G. Rethinking Financial Contagion
Working paper series no. 2831143 (2016).
31. Basel Committee on Banking Supervision: Capital Requirements for Banks’
Methods Equity Investments in Funds (Bank for International Settlements, 2013);
Methods, including statements of data availability and any http://www.bis.org/publ/bcbs266.pdf
associated accession codes and references, are available in the 32. Statistical Data Warehouse (European Central Bank, 2017);
online version of this paper. http://sdw.ecb.europa.eu
33. Recommendations of the Task-Force on Climate-Related Financial Disclosure
(Financial Stability Board, 2016); https://www.fsb-tcfd.org/publications/
Received 13 July 2016; accepted 24 February 2017;
recommendations-report/
published online 27 March 2017 34. Wolf, S., Schütze, F. & Jaeger, C. C. Balance or synergies between environment
and economy a note on model structures. Sustainability 8, 761 (2016).
References 35. Towards Green Growth (OECD, 2011).
1. Carney, M. Breaking the tragedy of the horizon–climate change and financial 36. Shiller, R. J. Market Volatility (MIT, 1992).
stability Lloyd’s (29 September 2015); http://www.bankofengland.co.uk/ 37. Mazzucato, M. & Semmler, W. Economic Evolution Learning, and Complexity
publications/Pages/speeches/2015/844.aspx (Springer, 2002); http://doi.org/fckmx3
2. Too Late, Too Sudden: Transition to a Low-Carbon Economy and Systemic Risk 38. Irvine, P. J. & Pontiff, J. Idiosyncratic return volatility, cash flows, and product
(ESRB Advisory Scientific Committee, 2016). market competition. Rev. Financ. Stud. 22, 1149–1177 (2009).
3. McGlade, C. & Ekins, P. The geographical distribution of fossil fuels unused 39. Brogaard, J. & Detzel, A. The asset-pricing implications of government
when limiting global warming to 2 ◦ C. Nature 517, 187–190 (2015). economic policy uncertainty. Manage. Sci. 61, 3–18 (2015).
4. Meinshausen, M. et al. Greenhouse-gas emission targets for limiting global
warming to 2 ◦ C. Nature 458, 1158–1162 (2009).
5. Leaton, J. Unburnable Carbon—are the World’s Financial Markets Carrying a
Acknowledgements
The authors would like to thank J. E. Stiglitz and A. C. Janetos for fruitful comments on
Carbon Bubble 1–36 (Carbon Tracker Initiative, 2012).
an early version of the paper, M. D’Errico for precious suggestions on macro-network
6. Robins, N., Keen, A. & Night, Z. Coal and Carbon Stranded Assets: Assessing the data from the ECB Data Warehouse, and J. Glattfelder for help on equity holdings data
Risk (HSBC, 2012). extraction from Orbis. We also would like to thank A. Barkawi, P. Monnin and M. Tanaka
7. Fleischman, L., Cleetus, R., Deyette, J., Clemmer, S. & Frenkel, S. Ripe for for their comments during the Bank of England conference on Climate Change and
retirement: an economic analysis of the US coal fleet. Electr. J. 26, 51–63 (2013). Central Banking. S.B. acknowledges financial support from the Swiss National Fund
8. Caldecott, B. & Robins, N. Greening China’s Financial Markets: The Risks and Professorship grant no. PP00P1-144689. All the authors acknowledge the support of the
Opportunities of Stranded Assets Briefing Paper 1–29 (Smith School et Enquête European Projects Future and Emerging Technologies (FET) SIMPOL (grant no. 610704)
du PNUE, 2014). and DOLFINS (grant no. 640772), and the European Project SEI Metrics (grant
no. 649982).
9. World Resource Institute and UNEP-FI Carbon Asset Risk Discussion
Framework Tech. Rep. 1–67 (2015).
10. Weyzig, F., Kuepper, B., van Gelder, J. W. & Van Tilburg, R. The Price of Doing Author contributions
Too Little Too Late Tech. Rep. 1–69 (Green European Foundation, 2014). All authors contributed to the writing of the manuscript, as well as material and analysis
11. Dietz, S., Bower, A., Dixon, C. & Gradwell, P. Climate value at risk of global tools. G.V. and S.B. also performed the data analysis.
financial assets. Nat. Clim. Change 6, 676–679 (2016).
12. Nordhaus, W. D. Rolling the ‘DICE’: an optimal transition path for controlling
greenhouse gases. Resour. Energy Econ. 15, 27–50 (1993). Additional information
13. Nordhaus, W. D. The economics of tail events with an application to climate Supplementary information is available in the online version of the paper. Reprints and
change. Rev. Environ. Econ. Policy 5, 240–257 (2011). permissions information is available online at www.nature.com/reprints. Publisher’s note:
14. Battiston, S., Caldarelli, G., Georg, C.-P., May, R. & Stiglitz, J. Complex Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations. Correspondence and requests for materials should be
derivatives. Nat. Phys. 9, 123–125 (2013).
addressed to S.B.
15. Battiston, S. et al. Complexity theory and financial regulation. Science 351,
818–819 (2016).
16. Battiston, S., D’Errico, M. & Gurciullo, S. DebtRank and the network of Competing financial interests
leverage. J. Altern. Invest. 18, 68–81 (2016). The authors declare no competing financial interests.

6 NATURE CLIMATE CHANGE | ADVANCE ONLINE PUBLICATION | www.nature.com/natureclimatechange

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3255 ARTICLES
Methods directly the bankruptcy of a financial institution that then defaults on its
Identifying climate-policy-relevant sectors in the real economy. Many climate obligations towards its financial counterparties. Second-round effects refer to
policies target the reduction of GHG emissions (in particular in non-carbon financial contagion effects including, but not necessarily, further defaults. More
neutral processes). To identify the climate-policy-relevant sectors we group generally, the accounting practice of mark-to-market implies that the deterioration
economic activities with the following logic. We start from the top sectors by direct of the balance sheet of a financial institution has a negative impact on the market
GHG emissions according to Eurostat (scope 1 CO2 equivalent), which includes value of its obligations held by its counterparties. Mark-to-market and, in
activities across sectors such as utilities, transports, agriculture, manufacturing and particular, credit valuation adjustment, is recognized as a major mechanism of
households. We also include the mining sector, although it has small direct financial distress propagation; during the 2007/2008 financial crisis, it accounted
emissions according to the scope 1 classification, because all the emissions of the for two-thirds of losses among many financial institutions (see ref. 42). More
three above sectors derive directly or indirectly from the fossil-fuel extraction when formally, in the breakdown of total assets, we can distinguish the securities issued
accounting from the supply side40 . We then take into account the so-called carbon by firms in the financial sectors (whose values depend on their own assets’ values)
leakage risk classification, which according to the EC Directive 201441 identifies from those issued by firms in the climate-policy-relevant sectors to obtain
activities (mostly within manufacturing) for which either costs or competitiveness  
is heavily affected by introduction of a carbon price. It can be easily verified that X Equity
the traditional NACE Rev2 (but the same holds for NAICS) classification of Ai =  αij (Aj ) + αijBond (Aj ) + αijLoan (Aj )
economic activities is not well-suited for a climate-policy analysis. For instance, j∈F

some activities classified under B-Mining and quarrying, such as ‘B7.1-Mining of  

iron ores’, are not so relevant for climate policies. In contrast, some activities X
α +α +α
Equity Bond Loan
classified under C-Manufacturing, such as ‘C19.2-Manufacture of refined + ik ik ik
 + Ri (2)
petroleum products’ or transport ‘H49.5-Transport via pipeline’, are more relevant k∈A/F

to the fossil-fuel sector from the criterion of economic scenarios resulting from
climate policies. Furthermore, some activities that pertain to the housing sector where A denotes the set of all actors and, again, F denotes the set of institutional
from a policy perspective fall into different NACE Rev2 sectors such as financial actors. When we consider the above equation for many financial actors
F—Construction and L—Real estate. simultaneously, equation (2) becomes a system of coupled equations in the asset
All the considered economic activities can be divided into three categories: values. In the spirit of analysing the short-term effects of a deviation in the values
(1) suppliers of fossil fuels, (2) suppliers of electricity (3) users of either fossil fuels from an initial face value of the securities, the terms αijInstrument (Aj ) can be written as
or electricity. We can further divide the third category according to the traditional the product αij0 fij (Aj ), where αij0 represents the face value of the security at the initial
policy areas: transport, housing and manufacturing. While suppliers of fossil fuels time and fij (Aj ) represents the valuation of the security with respect to its face value.
are mostly negatively affected by GHG emission reduction policies, the other While the exact functional form of fij depends on the instrument type and the
categories can be affected positively or negatively depending on the energy source pricing model used for the valuation of the security, it is possible nevertheless to
utilized (fossil fuel versus renewable). On the basis of all the above information, we infer certain useful properties. Consider for instance a chain of exposure in which
can finally remap all the economic activities from the 4-digit NACE Rev2 the financial actor i holds bond securities issued by the financial actor j, who in
classification into the following climate-policy-relevant sectors: fossil, utilities, turn holds securities issued by a firm k in the climate-policy-relevant sector. From
transport, energy-intensive, housing. The complete mapping from NACE Rev2 the equations above it follows that
4-digits codes is provided in Supplementary Information.
∂Ai (Aj (Ak )) ∂Ai (Aj ) ∂(Aj ) ∂fij ∂fjk
Assessing direct exposures of financial actors. Since our goal is to assess the = = αij0 αjk0 (3)
exposure of financial actors to the climate-policy-relevant sectors in the real ∂Ak ∂Aj ∂Ak ∂Aj ∂Ak
economy, we group financial actors into financial institutional sectors
according to the standard ESA classification: banks, investment funds, insurance Without loss of generality, in line with widely used pricing models such as those
and pension funds. The exposures of each financial actor can be decomposed based on the Merton model for the value of debt obligations, the functions fij are
along the main types of financial instruments: equity holdings (for example, non-decreasing in the value of the assets of the issuer j, that is, dfij /dAj ≥ 0, because
ownership shares including both those tradable on the stock market and those the ability of the issuer to pay either dividends or interest rates to its creditor
non-tradable), bond holdings (for example, tradable debt securities) and loans (for generally increases with the issuer’s total assets, everything else the same.
example, non-tradable debt securities). More formally, denoting by Ai the total It follows that, as long as the terms dfij /dAj are not too small and comparable
assets of financial actor i, and by S the set of climate-policy-relevant sectors, we across instruments, the indirect exposure to a climate-policy-relevant sector along
can write chains of financial actors is determined by the product of the face value of the
  exposures along the chain, αij0 αjk0 , where each exposure corresponds to the strength
XX of the link between the two nodes. The result can be generalized to longer chains,
αij + αijBond + αijLoan  + Ri
Equity
Ai =  (1) although we focus on length two in this work. Therefore, the problem of identifying
S∈S j∈S the largest indirect exposure of a given path length is mathematically equivalent to
the graph-theoretical problem of finding the path(s) with the largest product of link
where the terms αij denote the monetary values of the exposures of i in the weights along the path in a weighted graph.
securities associated with economic actors j for the different types of instruments
and Ri is a residual accounting for the exposure to other sectors and instruments Distribution of shocks. To infer a distribution of shocks on the fossil-fuel and
not considered in our analysis. utilities sector we use the LIMITS database26 , which provides economic impact
Although instrument types have different risk profiles, it is informative to look assessments of climate policies using a set of economic models and several
at the total exposure of financial actors to a given sector across all instruments. For scenarios that take into account the stringency of climate policy and the timing of
instance, we can compute in this way the full exposure of a given bank to the fossil its implementation. Results are reported as time series of forecasted production
sector, by summing up all of its equity holdings, bonds and loans exposures to this level for each sub-sector with a five-year interval up to 2050. In particular we
sector.PIf we denote by αiS the total exposure of actor i to sector S, we can write analyse the estimated time series of the share of fossil fuels and renewables in
αiS = j∈S αij + αijBond + αijLoan .
Equity
primary and secondary (electricity) energy consumption. Out of the time series,
In addition to the exposures of individual financial actors, we are also interested one can infer a distribution of shocks by considering each change in market share
in the aggregate exposure of an entireP financial institutional sector F to a given from one period to the next as corresponding to an observation of a shock for the
climate-policy-relevant sector, AFS = i∈F αiS . Finally, the total direct exposure of respective sub-sector. Hence, one obtains one shock per period per scenario and
the financial
P system P in the totality of climate-policy-relevant sectors is per model, for a total of 5,421 shocks. From an economic viewpoint, interpreting
AF S = F∈F i∈F αiS , where F denotes the set of institutional financial actors. these shocks on market shares as shocks on equities amounts to make the following
simplifying assumptions. First, the share of nominal expenses on energy is constant
Assessing indirect exposures of financial actors. A large portion of total assets (that is, the demand elasticity of substitution is 1). Second, the value of equity in a
held by financial institutions are in fact securities issued by other financial sub-sector is proportional to total income. Third, market valuation is based on
institutions (for example, about 40% for banks in the Euro Area). Moreover, about one-period (five years) ahead expectations. The shocks can then be interpreted as
25% of total market capitalization is invested in equity issued by companies in the the impact on market valuation of a previously unanticipated policy measure. The
financial sectors, and about 40% of the bond market is represented by outstanding extent to which these shocks will materialize depends on the ability of agents to
obligations issued by financial institutions. anticipate policy measures. The shock scenario we describe in the paper
As a result, there is a potential systemic risk that can materialize through the corresponds to a setting in which informational imperfections prevent agents from
so-called second-round effects16,17 . For instance, first-round effects may induce smoothly adjusting their expectations. The alternative scenario emphasized in the

NATURE CLIMATE CHANGE | www.nature.com/natureclimatechange

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLES NATURE CLIMATE CHANGE DOI: 10.1038/NCLIMATE3255

conclusion corresponds to a situation where a stable policy framework would allow (http://ec.europa.eu/eurostat/statistics-explained/index.php/Greenhouse_gas_
financial actors to smoothly adjust their expectations. In this case, climate-induced emission_statistics). Data on financial exposures at the sectoral level have been
systemic risk would not materialize. Supplementary Fig. 6 shows the resulting obtained from the ECB Data Warehouse (http://sdw.ecb.europa.eu).
distribution of the variation in asset value for a brown bank (investing in fossil-fuel
primary sector and fossil-fuel-based utilities) and a green bank (investing in the Data availability. The data that support the findings of this study are available
renewable utilities sector only). from Bureau Van Dijk (Orbis database) but restrictions apply to the availability of
these data, which were used under licence for the current study, and so are not
Data. Data on equity holding were obtained through the Bureau Van Dijk Orbis publicly available. Data are however available from the authors on reasonable
database. We collected a sample covering all EU and US listed companies and their request and with permission of Bureau Van Dijk.
disclosed shareholders with voting rights as of the end of the last available year, that
is, 2014. After some consistency checks, we end up with 14,878 companies and References
65,059 shareholders. By grouping the exposures by investor we thus reconstruct 40. Erickson, P. & Lazarus, M. Accounting for Greenhouse Gas Emissions Associated
portions of their equity holding portfolios, within the limitations of the available with the Supply of Fossil Fuels Tech. Rep. 1–4 (2013).
data. Further details on the data set and the methodology are provided in the 41. European Commission: Decision of 27 October 2014 determining, pursuant to
Supplementary Information. Data on the balance sheets of the top 50 listed Directive 2003/87/EC of the European Parliament and of the Council, a list of
European banks are obtained from the Bureau Van Dijk Bankscope database. Data sectors and subsectors which are deemed to be exposed to a significant risk of
include for each bank its total lending and borrowing to other banks. Exposures of carbon leakage, for the period 2015 to 2019 (notified under document
a bank to individual other banks are not publicly available and have been estimated C(2014) 7809); http://data.europa.eu/eli/dec/2014/746/oj
on the basis of existing methodologies (see literature in ref. 28). Data on GHG 42. FSA The Prudential Regime for Trading Activities Tech. Rep. (Financial Services
and CO2 emissions of sectors have been obtained from Eurostat statistics Authority, 2010).

NATURE CLIMATE CHANGE | www.nature.com/natureclimatechange

© 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.