A Finance Approach To Climate Stress Testing

A finance approach to climate stress testing*
Henk Jan Reinders

Rotterdam School of Management, Erasmus University; De Nederlandsche Bank;
The World Bank
Dirk Schoenmaker
Rotterdam School of Management, Erasmus University; CEPR
Mathijs van Dijk

Rotterdam School of Management, Erasmus University
April 2020
Abstract
There is increasing interest in assessing the impact of climate policies on the value of financial
sector assets, and consequently on financial stability. Prior studies either take a “black box”
macro-modelling approach to climate stress testing or focus solely on equity instruments –
though banks’ exposures predominantly consist of debt. We take a more tractable finance
(valuation) approach at the industry-level and use a Merton contingent claims model to assess
the impact of a carbon tax shock on the market value of corporate debt and residential
mortgages. We calibrate the model using detailed, proprietary exposure data for the Dutch
banking sector. For a €100 to €200 per tonne carbon tax we find a substantial decline in the
market value of banks’ assets equivalent to 4-63% of core capital, depending on policy choices.
*
E-mail addresses: reinders@rsm.nl, schoenmaker@rsm.nl, and madijk@rsm.nl. The authors are grateful to Dion
Bongaerts, Gianfranco Gianfrate, Jakob de Haan, Jean-Stéphane Mésonnier, Erik Roelofsen, Claus Schmitt, as
well as participants at the 2019 Innovations and New Risks conference at the Bank for International Settlements
(Basel), the 2019 IWFSAS conference at the University of Victoria, and the 2019 GRASFI conference at the
University of Oxford for stimulating discussions and useful suggestions. The opinion in this paper is those of the
authors and does not necessarily coincide with that of the DNB or the Eurosystem.
Electronic copy available at: https://ssrn.com/abstract=3573107

“Changes in climate policies, new technologies and growing physical risks will prompt
reassessments of the values of virtually every financial asset.”
Mark Carney, Governor of the Bank of England 1
1. Introduction
There is a substantial gap between Green House Gas (GHG) emission paths that are consistent
with keeping global warming well below two degrees Celsius and emissions paths that result
from current climate policies (Rogelj et al., 2018). In recent years, this gap between practice
and policy goals has led central banks, financial supervisors, and researchers to investigate the
financial risks stemming from climate change (“physical risks”) and from transitioning towards
a low-carbon economy (Campiglio et al., 2018; Nieto, 2019). The latter “transition risks” may
primarily arise if much more stringent climate policies are implemented by governments to
follow up on the 2015 Paris Agreement (Batten et al., 2016).
Decarbonizing the economy is expected to come at an economic cost, at least in the
short run (Acemoglu et al., 2012; Nordhaus, 1992). Depending on policy choices, some of these
costs will likely be borne by owners of financial assets, including banks, insurance companies,
and pension funds (Smale et al., 2006; Scholtens and Van Der Goot, 2014). For example, higher
taxes on GHG emissions can lead to additional costs for firms in GHG-intensive sectors and
more rapid write-off of their capital investments (i.e., “stranded assets”), reducing firms’
market values and increasing their credit risk. These costs can be substantial and range across
a wide variety of sectors and asset classes (Leaton, 2011). Estimates of the implied prices of
CO2-equivalent (CO2e) emissions needed to limit global warming to below two degree Celsius
range from $15-$360 per tonne in 2030 and from $45-$1000 per tonne in 2050 (Stiglitz et al.,
2017).
1
Speech at the 2019 Task Force on Climate-related Financial Disclosures TCFD Summit, Tokyo (Oct. 8, 2019).

From the perspective of both financial institutions and regulators concerned with
financial stability, a key question is what the potential impact of such climate policies is on the
balance sheets of financial institutions. Several studies have started to explore how climate
policies adversely affect the value of financial institutions’ balance sheets and, for extreme
scenarios, can cause systemic financial crises. 2 Battiston, Mandel, Monasterolo, Schütze, and
Vistentin (2017) perform a climate stress test for Eurozone banks, looking at the impact of
climate policies on the equity exposures of European banks while at the same time allowing
for second-round effects resulting from exposures between financial institutions. Vermeulen,
Schets, Lohuis, Kölbl, and Jansen (2019) propose a framework for climate stress tests that
builds on macro-economic stress test methods. This work fits into a broader literature on
financial sector stress testing (see, for example, Upper, 2011; Henry et al., 2013; Ong, 2014).
Our main contribution to the emerging literature on climate stress tests is two-fold.
First, our research contributes to the climate stress test literature by tractably modelling the
industry-level impact of climate policies on the value of financial sector holdings of both equity
and debt instruments. The literature to date either uses indirect and less tractable (“black box”)
macro modelling to determine financial sector impacts or focuses exclusively on equity
instruments while assuming that the value of certain financial instruments evaporates
completely. 3 Especially for banks, it is crucial to understand the effects of climate policies on
debt instruments, since the majority of bank assets are subordinated in nature. For example, in
the euro area, at least 85% of all banking assets consist of debt, while only 2% is equity (see
Table 1). Second, because we take an industry-level approach, we are able to investigate the
2
From a policy perspective several central banks have started performing climate stress tests, or have announced
to start doing them. These include the Dutch central bank, the Bank of England, and the Banque de France
(Lehmann, 2020).
3
Macro financial stress testing typically estimates the impact of scenarios on GDP and other macro-variables first
and then uses those variables as inputs into financial risk models. Final results are obtained through several layers
of modelling employing a large and complex set of equations, increasing the potential for modelling error and
making it harder to intuitively interpret results.

differences in impact between industries and across specific carbon taxation scenarios in detail
(e.g., different price levels and policy design choices). Further research along these lines could
inform the alignment of climate policies, and carbon taxes in particular, with the objective of
financial stability.
To assess the impact of climate policies on the financial sector assets, we build on the
option valuation and structural credit risk modelling literature (Black and Scholes, 1973;
Merton, 1974). Specifically, we employ the ideas by Merton (1974), who models the structural
factors that determine the market value of debt. Merton’s key insight is that equity can be
viewed as a residual claim on assets after the debt has been repaid. This implies that the equity
holder has a call option on the value of the firm’s assets, where the payoff is either zero (in
case of default) or the value of assets minus the face value of the debt. Conversely, the debt
holder has a risk-free bond and is short a put option of the firm’s assets. Overall, Merton’s
contingent claims approach implies that a negative asset valuation shock will affect the value
of both equity and debt in a non-linear way.
We modify the standard Merton (1974) model to account for mortgages that have
additional safeguards built-in for the lender. Specifically, the traditional Merton model assumes
that default occurs when at maturity the value assets V lies below the face value of debt L. For
corporates, this is likely a valid approximation, although some extensions have been proposed
to relax this assumption. 4 However, for mortgages, which represent an important asset class
for banks, default is more complicated since these often have additional safeguards built-in for
the lender, such as recourse to the wealth and income of the borrower. This implies that, in a
Merton setting, we need to adjust the default trigger for residential mortgages. Following Sy
(2014), we take an approach where mortgage default is conditional on both insolvency (i.e.,
4
For example, Black and Cox (1976) look at the case where restructuring already occurs before V falls below L.

the value of the house falling below the value of the mortgage) and delinquency (i.e., not having
sufficient liquidity to make the periodical payment on a loan).
Our empirical analysis focuses on the banking sector in the Netherlands, for which we
use detailed and proprietary loan and debt exposure data from the Dutch central bank (DNB).
This includes residential mortgage data from the DNB loan-level database and corporate loan,
debt, and equity data that are aggregated to 4-digit NACE sector level. Our exposure dataset
includes the three largest banks in the Netherlands that collectively cover 79% of total assets
in the Dutch banking sector. Total assets in the Dutch banking sector were €2,381 billion in
2017. Furthermore, we calibrate a Merton contingent claims model to allocate asset valuation
losses to junior (equity-like), and senior (debt-like) claim holders of the asset. For corporates,
we use representative samples of firms from the Orbis Bureau van Dijk database to obtain
estimates of leverage and asset volatility by sector. For listed firms, we link these samples to
Thomson Reuters Datastream to obtain market estimates for asset volatility. For non-listed
firms, we estimate asset value volatility using a cross-sectional regression model. For
residential mortgages, we disaggregate exposures to different types of dwellings (i.e.,
apartments, terraced houses, and detached houses) per loan-to-value (LTV) bucket and per
remaining maturity bucket. We also obtain estimates for house price value volatility and the
probability of delinquency. We then estimate asset valuation shocks by valuing both the
negative cash flows of the carbon tax and the total cash flows of representative firms and real
estate, which leads to valuation shocks per industry and real estate segment. These shocks are
calibrated using carbon intensity (scope 1 emissions or “carbon footprint”) data obtained from
Eurostat, which are available for most 2-digit NACE sectors. For this reason, we further
aggregate our exposure data to the same 2-digit NACE level. With minor exceptions, all our
data are for 2017.

To perform the stress test, we define four carbon taxation scenarios that differ in their
severity and likelihood. Our main scenarios are based on the introduction of a €100 carbon tax,
which lies well within current estimates of implied carbon prices that are needed to achieve the
goals in the 2015 Paris Agreement. We differentiate our scenarios by assuming either an abrupt
(overnight) or smooth (10-year phase-in period) of the tax and by assuming either a regional
application (no cost pass-through from firms to consumers) or a global application (50% cost
pass-through from firms to consumers). Some of these policy scenarios are severe but not
entirely implausible. They hence serve the purpose of investigating extreme policy scenarios,
which is a common practice in financial sector stress testing (Cihák, 2007).
Our findings indicate that, depending on the policy scenario, market value losses range
from €4.5 billion to €35.8 billion following the implementation of a carbon tax of €100. In the
most severe scenario, in which carbon taxation is applied abruptly, and there is no pass-through
(e.g., due to regional application), losses amount to 30% of the available Common Equity Tier
1 (CET1) capital in the Dutch banking system and to 1.5% of total assets. When carbon taxation
is instead phased-in over ten years, the losses as fractions of CET1 capital and total assets
decline to 14% and 0.7%, respectively. For €200 per tonne carbon taxes, the market value
losses increase exponentially, ranging from €17.9 billion to €75.2 billion. In the most severe
scenario, this equals 63% of CET1 capital and to 3.2% of total assets. When carbon taxation is
instead phased-in over ten years, the losses as fractions of CET1 capital and total assets decline
to 47% and 2.4%, respectively. Some of our €100 and €200 carbon tax scenarios show market
value losses that are substantially (two to three times) higher than those obtained under
traditional banking sector stress tests, of which the adverse scenarios typically result in declines
in core capital around 20%. 5 This may warrant increased attention of prudential policy makers
5
For example, the 2018 stress test by the European Banking Authority (EBA) finds that, on aggregate, the CET1
capital of EU banks declines by 19.2% in their adverse scenario. Furthermore, the Federal Reserve found that, on
aggregate, the Tier 1 leverage ratio of US banks declines by 19.8% in their severely adverse scenario in their 2019
stress test.

to climate-related scenarios. It also underscores the importance of designing climate policies
in such a way that they both achieve decarbonization in line with international agreements and
safeguard financial sector soundness.
Furthermore, our findings shed light on vulnerable asset classes and sectors. We find
that first-order market value losses for Dutch banks are primarily driven by exposures to
corporate loans and debt, and to a lesser extent by residential mortgages and equity. 6 Principal
reasons for this finding are the low exposures of Dutch banks to equity instruments in
transition-sensitive industries (i.e., less than 1% of total assets) and the low net present value
of carbon taxes for most types of housing compared to their valuation, combined with recourse
to a borrowers income on top of the recourse to the underlying real estate (which puts market
value losses on real estate mostly as a burden to households and to a lesser extent on the banking
sector). In the €100 carbon tax scenarios, the largest absolute contributions to market value
losses are, in declining order, obtained for electricity, gas, steam and air conditioning supply
(D.35), the manufacture of coke and refined petroleum products (C.19), water transport (H.50),
the manufacture of basic metals (C.24), and air transport (H.51). Taken together, these five
industries drive between 83% and 91% of the total market value losses, depending on the choice
of scenario.
Overall, our results point to the substantial impact that climate-related policies can have
on the market value of assets on the balance sheets of banks in the Netherlands. By tractably
modelling the vulnerability of financial assets to carbon taxation, our analysis highlights the
importance of debt exposures for their contribution to overall losses, and in particular the
corporate loans and debt portfolio. In severe scenarios, climate policies may lead to substantial
losses in the banking sector and our research, therefore, underlines the importance of
6
Note that second-round losses on mortgage portfolios might be higher, for example if unemployment rates and/or
interest rates were to rise. The effect of climate policies on such macro-variables is highly uncertain, however,
and depends amongst others on additional policy choices (e.g., re-employment programs and industry
compensation schemes).

adequately addressing the interlinkages between climate policies and financial stability (e.g.,
steering away from investments in long-term assets that are not compatible with a low carbon
economy). Since our analysis focuses on direct (i.e., scope 1) carbon emissions and does not
include second-order effects (e.g., through changing unemployment and interest rates), our
outcomes are likely conservative, and we encourage further research to investigate further
channels through which carbon taxes can affect financial institutions’ balance sheets.
2. Financial stress test model
In this section, we develop a tractable impact model that underlies our stress test. We set out a
general approach which we further develop in the subsequent two sections. We also discuss
assumptions and possible extensions.
2.1 General set-up
The primary goal of our modelling is to determine the impact of climate policy stress scenarios
on the market value of debt and equity portfolios of banks. We do this by separating the
aggregate exposures of banks into equity exposure and debt exposure in groups of assets k
(from hereon: segments) that share similar vulnerability characteristics (e.g., carbon intensity).
In principle, k can be at the level of an individual firm or real economy asset. However, given
the exposure data that we have available, we will take k to represent either an industry (for
firms) or real estate of a similar dwelling type. Total market value loss is given by:
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 = ∑𝑛𝑛𝑘𝑘=1 𝜗𝜗𝐸𝐸,𝑘𝑘 ∗ 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝐸𝐸,𝑘𝑘 + 𝜗𝜗𝐷𝐷,𝑘𝑘 ∗ 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝐷𝐷,𝑘𝑘 , (1)
with stress test coefficients 𝜗𝜗𝐸𝐸,𝑘𝑘 for equity and 𝜗𝜗𝐷𝐷,𝑘𝑘 for debt defined as follows:
𝑀𝑀𝑀𝑀𝐸𝐸,𝑘𝑘 ∗ 𝑀𝑀𝑀𝑀𝐷𝐷,𝑘𝑘 ∗
𝜗𝜗𝐸𝐸,𝑘𝑘 = and 𝜗𝜗𝐷𝐷,𝑘𝑘 = (2)
𝑀𝑀𝑀𝑀𝐸𝐸,𝑘𝑘 𝑀𝑀𝑀𝑀𝐷𝐷,𝑘𝑘

In the above formula, 𝑀𝑀𝑀𝑀𝑘𝑘 represents the market value of a group of assets k. An asterisk (*) is
used to denote the market value after the scenario shock has been applied. Using this definition
gives us the fraction of the market value of the portfolio that remains after the stress scenario
is applied. Hence, the expected market value loss per unit of exposure can be written as 1 – 𝜗𝜗. 7
Section 2.2 provides the modelling of 𝜗𝜗𝐷𝐷,𝑘𝑘 and 𝜗𝜗𝐸𝐸,𝑘𝑘 as functions of a set of calibration
parameters ϴ𝑖𝑖 and an asset value shock 𝜉𝜉𝑘𝑘 . We follow Merton’s (1974) structural credit risk
model, which we extend to take into account more complicated default conditions that are
characteristic of European mortgages with double recourse. The basic idea is to distribute the
asset valuation shock 𝜉𝜉𝑘𝑘 to holders of equity (E) and debt (D). This part of the stress test model
can hence be viewed as covering the financial (right-hand) side of the corporate balance sheet.
Since we take the calibration parameters to differ per asset i and calculate the stress test
coefficients per segment k, we sum the outcomes of n individual assets making up a segment
using a weighting parameter 𝜔𝜔𝑖𝑖 .

𝑛𝑛
𝜗𝜗𝐸𝐸,𝑘𝑘 , 𝜗𝜗𝐷𝐷,𝑘𝑘 = � 𝜔𝜔𝑖𝑖 ∗ 𝑓𝑓𝑀𝑀𝑒𝑒𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 (𝜉𝜉𝑘𝑘 , 𝛳𝛳𝑖𝑖 ) (3)

𝑖𝑖=1
In section 2.3, we put forward a stylized discounted cash flow model to determine a valuation
shock 𝜉𝜉𝑘𝑘 per segment. We model 𝜉𝜉𝑘𝑘 such that it ranges between zero (no losses) and one (full
loss of sector value). This can be viewed as the real (left-hand) side of the corporate balance
sheet, representing the value of a physical bundle of assets. We take 𝜉𝜉𝑘𝑘 to be a function of the
scenario variable 𝜏𝜏𝑘𝑘,𝑡𝑡 , which represents the euro tax amount per tonne of CO2e emissions over
7
Since we are interested in the consequences for the market value of the bank balance sheet, we estimate the
expected loss in risk-neutral terms (i.e., the probability of default is adjusted to reflect market participants’ risk
preferences).

time, and a set of vulnerability parameters 𝛺𝛺𝑘𝑘,𝑡𝑡 which we differentiate per segment k and which
may vary over time t. 8
𝜉𝜉𝑘𝑘 = 𝑓𝑓(𝜏𝜏𝑘𝑘,𝑡𝑡 , 𝛺𝛺𝑘𝑘,𝑡𝑡 ) (4)
2.2 Merton’s structural credit risk model for asset valuation shocks
In a standard Merton structural debt framework, the market value of debt 𝑀𝑀𝑀𝑀𝐷𝐷 can be written
as its risk-free value minus the risk-neutral expected loss (the latter being equivalent to a put
option on the value of the assets). Following the notation of Giesecke (2002):
𝑀𝑀𝑀𝑀𝐷𝐷 = 𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) − 𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) �𝑁𝑁(−𝑑𝑑2 )� − 𝑉𝑉𝑡𝑡 𝑁𝑁(−𝑑𝑑1 ) (5)
with
𝑉𝑉 𝜎𝜎 2
ln ( 𝑡𝑡 )+(𝑟𝑟+ 𝑉𝑉 )(𝑇𝑇−𝑡𝑡)
𝑑𝑑1 = 𝐿𝐿 2
𝜎𝜎𝑉𝑉 √𝑇𝑇−𝑡𝑡
𝑉𝑉 𝜎𝜎 2
ln ( 𝑡𝑡 )+(𝑟𝑟− 𝑉𝑉 )(𝑇𝑇−𝑡𝑡)
𝑑𝑑2 = 𝐿𝐿 2
,
𝜎𝜎𝑉𝑉 √𝑇𝑇−𝑡𝑡
where N is the probability of the standard normal density function below d. Hence 𝑀𝑀𝑀𝑀𝐷𝐷 can be
expressed as a function of asset value V, contracted repayment L, time to maturity T-t, the
standard deviation of asset value 𝜎𝜎𝑉𝑉 and the risk-free interest rate r. Furthermore, under the
assumption that asset values follow a geometric Brownian motion, the volatilities of the firm
and its equity are given by:
𝑉𝑉
𝜎𝜎𝐸𝐸 = 𝑁𝑁(𝑑𝑑1 )𝜎𝜎𝑉𝑉 (6)
𝐸𝐸
For our purposes, we will assume an instantaneous shock ξ on asset value such that immediately
after the shock asset value V* is given by:
8
Note that all our scenarios assume that the carbon tax applies equally to all segments k, hence for our analysis
we can suffice by writing 𝜏𝜏𝑡𝑡 . This is however not a necessity; in practice climate policies often differentiate
between industries. Vulnerability parameters in our analysis are the carbon footprint, the capacity to pass-on the
carbon tax to consumers, adaptive capability, a sector specific discount rate, and (for mortgages) the probability
of delinquency.

𝑉𝑉 ∗ = (1 − 𝜉𝜉)𝑉𝑉, (7)
which gives the market value of debt after the shock as:
𝑀𝑀𝑀𝑀𝐷𝐷 ∗ = 𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) − 𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) �𝑁𝑁(−𝑑𝑑2∗ )� − 𝑉𝑉𝑡𝑡∗ 𝑁𝑁(−𝑑𝑑1∗ ) (8)
Replacing 𝑉𝑉 ∗ with (1 − 𝜉𝜉)𝑉𝑉, defining the ratio of contracted repayment to asset value (leverage
ratio) as R = L / V and dividing by the discounted exposure 𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) we find that:
𝑀𝑀𝑀𝑀𝐷𝐷 ∗ = 1 − �𝑁𝑁(−𝑑𝑑2∗ )� − ((1 − 𝜉𝜉)/𝑅𝑅𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) )𝑁𝑁(−𝑑𝑑1∗ )
with
𝜎𝜎 2
(1−𝜉𝜉)
ln ( 𝑅𝑅 )+(𝑟𝑟+ 𝑉𝑉 )(𝑇𝑇−𝑡𝑡) (9)
𝑑𝑑1∗ = 2
𝜎𝜎𝑉𝑉 �(𝑇𝑇−𝑡𝑡)
(1−𝜉𝜉) 𝜎𝜎 2
ln ( 𝑅𝑅 )+(𝑟𝑟− 𝑉𝑉 )(𝑇𝑇−𝑡𝑡)
𝑑𝑑2∗ = 2
𝜎𝜎𝑉𝑉 �(𝑇𝑇−𝑡𝑡)
Hence,
𝑀𝑀𝑀𝑀𝐷𝐷 ∗ 1−�𝑁𝑁(−𝑑𝑑2∗ )�−((1−𝜉𝜉)/𝑅𝑅𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) )𝑁𝑁(−𝑑𝑑1∗ )

𝜗𝜗𝐷𝐷 =
𝑀𝑀𝑀𝑀𝐷𝐷
=
1−�𝑁𝑁(−𝑑𝑑2 )�−(1/𝑅𝑅𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) )𝑁𝑁(−𝑑𝑑1 )
(10)
Thus, given a risk-free interest rate r, 𝜗𝜗𝐷𝐷 is a function of the asset valuation shock 𝜉𝜉, the
leverage ratio R, asset value volatility 𝜎𝜎𝑉𝑉 and the time to maturity T-t. Moreover, equations (6)
and (9) can be solved simultaneously in order to determine V and 𝜎𝜎𝑉𝑉 from E and 𝜎𝜎𝐸𝐸 . In a similar
fashion, the Merton equation for equity is given by:
𝑀𝑀𝑀𝑀𝐸𝐸 = 𝑉𝑉𝑡𝑡 𝑁𝑁(𝑑𝑑1 ) − 𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) 𝑁𝑁(𝑑𝑑2 ) (11)
And following the same line of reasoning as for debt, we find that:
𝑀𝑀𝑀𝑀𝐸𝐸 ∗ (1−𝜉𝜉)𝑁𝑁(𝑑𝑑1∗ )−𝑅𝑅𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) 𝑁𝑁(𝑑𝑑2∗ )

𝜗𝜗𝐸𝐸 = = (12)
𝑀𝑀𝑀𝑀𝐸𝐸 𝑁𝑁(𝑑𝑑1 )−𝑅𝑅𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) 𝑁𝑁(𝑑𝑑2 )
2.2.1 Merton model limitations and extension to mortgages
10

The Merton (1974) model is based on several assumptions, some of which have been
challenged in subsequent research. 9 One key assumption is that asset value follows a geometric
Brownian motion, which implies that in a short interval of time, asset value can only change
by a small amount (Merton, 1976). Several authors have noted that this is inconsistent with
empirical observation, namely that in a short interval of time there can be large changes in
stock prices or “jumps” (e.g., Cai and Kou, 2011). Moreover, in most segments, there are
substantial costs associated with a default that is not captured by the Merton model (i.e., the
model assumes that there are no specific costs resulting from triggering default). To account
for these costs, some authors explicitly introduce recovery values (e.g., Benos and
Papanastasopoulos, 2007; Longstaff and Schwartz, 1995). Both of these assumptions in the
standard Merton model lead to potentially higher losses as a result of an asset valuation shock,
implying that our model is more likely to underestimate than to overestimate potential losses.
On the other hand, specifically for mortgages, the Merton model may overestimate
losses due to the recourse nature of most European mortgages. Recourse entitles the creditor to
other household assets besides the value of the secured real estate, including other assets and
future income. In contrast to some American mortgages, this implies that households are less
prone to default on their mortgages in the face of asset valuation losses, even if the value of the
real estate is lower than the value of the mortgage. 10 To account for recourse, we model a more
stringent default condition, rewriting equation (8) by dividing by the discounted exposure
𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) and multiplying its last term by 𝑁𝑁(−𝑑𝑑2 )/ 𝑁𝑁(−𝑑𝑑2 ):
𝑉𝑉𝑡𝑡 𝑁𝑁(−𝑑𝑑1 )
𝑀𝑀𝑀𝑀𝐷𝐷 /𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) = 1 − 𝑁𝑁(−𝑑𝑑2 )(1 − ∗ ) (13)
𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) 𝑁𝑁(−𝑑𝑑2 )
9
For a full list of assumptions, see Merton (1974).
10
We note here that although a mortgage may legally be full-recourse, in practice this full-recourse is not always
(fully) applicable. An example is the case of Ireland where in the aftermath of a housing crisis the central bank
implemented regulations that severely restricted the ability of banks to contact or harass delinquent borrowers,
making the Irish residential mortgages de facto limited recourse contracts (Connor and Flavin, 2015).
11

In this equation, 𝑁𝑁(−𝑑𝑑2 ) is the risk-neutral probability of default and 𝑁𝑁(−𝑑𝑑1 )/𝑁𝑁(−𝑑𝑑2 ) is the
expected discounted recovery rate (Sy, 2014). For residential mortgages, we can then introduce
a more strict default trigger by replacing the Merton probability of default 𝑁𝑁(−𝑑𝑑2 ), which
could be thought of as representing insolvency, by a more broad probability of default that is a
multiplication of 𝑁𝑁(−𝑑𝑑2 ) and the probability that a household will not have sufficient wealth
and/or income to pay their instalment P(delinquent). If we assume that there is no correlation
between 𝑁𝑁(−𝑑𝑑2 ) and P(delinquent), equation (13) can then be rewritten as:
𝑀𝑀𝑀𝑀𝐷𝐷 𝑉𝑉𝑡𝑡 𝑁𝑁(−𝑑𝑑1 )

= 1 − 𝑁𝑁(−𝑑𝑑2 ) ∗ 𝑃𝑃(𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑) ∗ (1 − ∗ ), (14)
𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) 𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) 𝑁𝑁(−𝑑𝑑2 )
which leads to:
𝑀𝑀𝑀𝑀𝐷𝐷 ∗ 1−𝑁𝑁(−𝑑𝑑2∗ )∗𝑃𝑃(𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑)∗ ∗(1−(1−𝜉𝜉)/𝑅𝑅𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) ∗𝑁𝑁(−𝑑𝑑1∗ )/𝑁𝑁(−𝑑𝑑2∗ ))

𝜗𝜗𝐷𝐷,𝑀𝑀 = = (15)
𝑀𝑀𝑀𝑀𝐷𝐷 1−𝑁𝑁(−𝑑𝑑2 )∗𝑃𝑃(𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑)∗(1−1/𝑅𝑅𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) ∗𝑁𝑁(−𝑑𝑑1 )/𝑁𝑁(−𝑑𝑑2 ))
Similar reasoning can be applied to determine the impact of the market value of the equity
portion of mortgage exposures (or in general, exposures where default is triggered by combined
but uncorrelated insolvency and delinquency).
2.3 Asset valuation shocks
To determine asset valuation shocks, we take the yearly CO2 emissions connected to the
segments’ activities 𝛾𝛾𝑘𝑘 (i.e., the sector-specific carbon footprint) and multiply this by the
carbon tax 𝜏𝜏𝑡𝑡 . The total valuation impact of the tax can then be determined by discounting the
tax-related cash flows, most of which occur in the future, into a net present value using an
appropriate discount rate per segment 𝑟𝑟𝑘𝑘 . Without any response from any of the actors involved
(such as adjustments in the production process, the quantity or the price of products and/or
making energy efficiency investments in real estate), an unanticipated shock to the carbon tax
rate would lead to a reduction in the value of the bundle of assets that is equal to the present
value of the additional (negative) cash flows. Assuming that there are no net tax effects, the
12

impact of the tax shock on the net present value of a physical asset or firm can be thought of
as follows:
NPVtax, k = ∑𝑇𝑇𝑡𝑡=0 (1 − 𝑟𝑟𝑘𝑘 )𝑡𝑡 ∗ 𝛾𝛾𝑘𝑘 (−𝜏𝜏𝑡𝑡 ) (16)
Of course, it can be expected that firms and households respond in an attempt to offset
the potential loss in their value after a carbon tax is announced. We account for this in two
ways. First, one response that is well-documented in the literature is the pass-through of
increasing costs for firms (in this case the carbon tax) into product prices (Fabra and Reguant,
2013; Smale, Hartley, Hepburn, Ward, and Grubb, 2006). This increase in price could partially
offset the initial tax burden on producers. However, for most goods, an increasing price reduces
the size of the market, which potentially leads to firms exiting the market or lowering their
production volumes. 11 Therefore, in some of our scenarios, we allow for a non-zero amount of
pass-through that can change over time (e.g., due to contract renewals after certain periods),
denoted by 𝜑𝜑𝑘𝑘,𝑡𝑡 . 12 Second, our model takes into account the possibility that firms and
households adjust their physical assets and their use over time, for example by substituting
inputs (e.g., green for brown electricity) and by making additional investments (e.g., energy
savings technologies and technologies that avoid atmospheric emissions such as carbon filters).
We do this by allowing the carbon intensity per segment 𝛾𝛾𝑘𝑘 , and hence the tax burden, to change
over time. We hence add a subscript t. We then arrive at the following expression for the
valuation impact on the value of a physical asset or firm as:
NPVtax,k = ∑𝑇𝑇𝑡𝑡=0 (1 − 𝑟𝑟𝑘𝑘 )𝑡𝑡 ∗ 𝛾𝛾𝑘𝑘,𝑡𝑡 (1 − 𝜑𝜑𝑘𝑘,𝑡𝑡 )(−𝜏𝜏𝑡𝑡 ) (17)
11
Note that this could not only lead to stranded assets (e.g., oil reserves and specialized capital goods) as often
referred to in the literature, but also stranded business (i.e., future earnings that are priced into firm value but are
not expected under the new climate policy regime).
12
We will define two sets of scenarios with respect to pass-through. One will assume no pass through, which
resembles the situation in which carbon taxation is only applied regionally and there is free trade between regions.
In such a case, producers that are taxed are (for most products) expected not to be able to pass on the cost to
consumers. Another set will assume 50% pass through, resembling the situation in which the tax is applied more
widely.
13

Finally, we relate the net present value of the tax shock to the total asset value, which gives us
the fraction of the total asset value that is lost due to the carbon tax:
𝑁𝑁𝑁𝑁𝑁𝑁𝑡𝑡𝑡𝑡𝑡𝑡,𝑘𝑘
𝜉𝜉𝑘𝑘 = (18)
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑘𝑘
3. Data and calibration
To perform the stress test, we need data to combine data on the exposure of financial
institutions to different asset segments (i.e., industries and types of real estate), data on the
vulnerability of these segments to a carbon tax (e.g., carbon intensity), and data to calibrate the
contingent claims model (e.g., leverage and asset volatility). We describe our data in the next
three sub-sections. Furthermore, as part of the calibration, we estimate a model to predict the
non-observable asset volatility of a representative sample of non-listed firms. This is important
since non-listed firms are the main beneficiaries of bank funding, while their asset volatility
cannot directly be observed.
The choice for our data sources is, to a large extent, driven by the availability of
breakdowns that match the segment classification of our exposure data. For corporate
exposures, these are industries according to a 2 or 4-digit NACE industry classification, while
for mortgages we use a segmentation based on the type of dwelling (e.g., terraced houses,
detached houses and apartments). All data is for 2017, except the sectorial exposures to equity
for which we only have 2016 data available. For corporate exposures, we include all industries
at the 2-digit NACE division with a carbon intensity of more than 0.5 kg CO2e / euro gross
operating surplus. This includes most subsectors within agriculture, forestry and fishing (A),
mining and quarrying (B), manufacturing (C), electricity, gas, steam and steam conditioning
supply (D), water supply, sewerage, waste management and remediation activities (E), and
transportation and storage (H). The choice of these industries is in line with other papers (e.g.,
14

Battiston et al. 2017; Vermeulen et al. 2019). We refer to this group as “transition-sensitive
industries”. 13
3.1 Bank sector exposure data
We use two proprietary datasets of the Dutch central bank that provide a detailed breakdown
of corporate and residential real estate exposures. Combined, these two asset classes make up
59% of the balance sheet of the Dutch banking sector. Other major asset classes that are outside
the scope of our analysis are government loans and debt (11%) and loans and debt to financial
institutions (14%) – see Table 1.
For corporate debt exposures, we use a 2017 dataset on the industry classification of
the asset holdings of Dutch banks. This dataset is obtained as part of a 2017 climate exposure
survey and includes the exposures of the three largest banks in the Netherlands, which together
hold 79% of total assets in the Dutch banking sector. In this dataset, corporate loans and debt
are categorized using a 4-digit NACE classification. For each 4-digit NACE class, the dataset
provides total exposure and average remaining maturity. For our analysis, we aggregate these
exposures to the 2-digit NACE division, in order to match them with Eurostat data on carbon
intensity. We calculate the average remaining maturity for each 2-digit NACE division based
on the exposure-weighted average remaining maturity..s. For equity exposures, we use a
dataset on the sectorial classification of debt and equity, which was part of a 2016 survey on
climate-related exposures. This dataset includes the same set of banks as the 2017 data.
Based on these datasets, the largest three Dutch banks hold €208 billion worth of
corporate debt and equity in transition-sensitive industries, which equals 11.1% of their total
assets. Looking at the 2-digit NACE sectors in transition-sensitive industries, the largest three
Dutch banks have, in declining order, the highest exposure to agriculture (A.01), food
13
The full list of transition-sensitive industries can be found in Table 2.
15

manufacturing (C.10), water transport (H.50), electricity and gas (D.35), and the extraction of
crude petroleum and natural gas (B.06). For a full summary of this data, see table 2.
For residential real estate loan exposures, we use 2017 loan-level data on residential
mortgages from the Dutch Central Bank. This dataset covers 67% of the mortgages in the Dutch
banking sector. In this dataset, we segment loans according to the type of building (e.g.,
apartment, town-house, detached). The database includes, at loan-level, the loan-to-value
(LTV), remaining maturity, and the last transaction price of the house. The total exposure in
the dataset is €497 billion, which equals 32% of total assets in the Dutch banking sector. With
respect to residential real estate, the majority of the exposure consists of detached or semi-
detached houses which make up about three quarters (76%) of the total residential real estate
portfolio. For a full summary of this data, see table 3.
3.2 Vulnerability data

For corporates, we estimate the vulnerability to a carbon tax based on the carbon intensity
variable in the Eurostat Structural Business Statistics (SBS) database. This variable provides
the amount of Green House Gas emissions as a fraction of gross value added, at the 2-digit
NACE division. It includes emissions of CO2, N2O and NH4 in CO2-equivalents (CO2e). We
adjust the Eurostat carbon intensity variable by subtracting personnel costs from gross added
value, to obtain an as close as possible estimation of yearly profitability. Data on personnel
costs are obtained from the same SBS database to ensure consistency. The resulting variable
provides the kilogrammes of CO2e emissions per euro gross operating surplus, and thereby an
approximation of the fraction of yearly profits that are eroded by each unit of carbon taxation.
For real estate, we estimate the vulnerability to a carbon tax based on the energy use
relative to the value of the real estate. For each segment, we calculate the average house price
and the associated energy use. We use the average natural gas (per M3) and electricity
16

consumption (in kWh) per dwelling type for 2017 from the Dutch Statistical Office (CBS)
Statline database. We combine these data with emission factors of 1.9 kg CO2e/M3 for natural
gas and 0.355 kg CO2e/kWh for electricity. 14 The reported carbon intensities are based on
annual capital costs, which we derive from average sales prices per dwelling type, also obtained
from CBS Statline, at a 3% per annum interest rate (which represents the typical mortgage rate
in the Netherlands for a 100% loan-to-value mortgage). Hence, we multiply the average sales
price by 0.03. We exclude residential mortgage exposure for which there is no classification
for the type of dwelling, or the type of dwelling is of an uncommon nature (e.g., land-only and
bungalows). The omitted exposure equals €18,145 million (3.6% of total reported exposures).
3.3 Model calibration
To calibrate the Merton model, we need estimates per industry and real estate segment for four
parameters: leverage, asset value volatility, remaining time to maturity, and the risk-free
interest rate. In addition, for mortgages, we need an estimate of the probability of delinquency.
The remaining time to maturity is available as part of the exposure data for both corporate and
residential real estate exposures as an exposure weighted average. Also, the loan-to-value
(LTV) ratios are available as part of the exposure data for residential real estate, providing the
exposure per segment in 10 LTV buckets. We assume a constant risk-free interest rate of 2%.
The remaining challenge is then to obtain estimates for leverage and asset volatility for
corporate exposures (both public and private firms), and to obtain estimates for asset volatility
and the probability of delinquency for residential real estate exposures.
3.3.1 Corporate exposures
14
CE Delft report, Emissiekentallen elektriciteit, 2015 (https://www.ce.nl/publicaties/download/1786) and RVO
report, Berekening van de standaard CO2-emissiefactor aardgas, 2018.
17

Neither leverage nor asset value volatility is part of the corporate exposure data at the Dutch
central bank. To obtain estimates of the leverage per industry, we create a representative
portfolio of firms obtained from the Orbis van Dijk database. Since most of the exposures of
the Dutch banking sector are in the Netherlands, we obtain the full sample of Dutch firms that
are present within the Orbis van Dijk database and then restrict the sample to those firms that
have a non-zero and positive amount of long-term debt (which is the closest indicator we could
find in the database of firms being bank-funded). For all these firms, we divide total debt by
total assets to obtain a measure of the leverage ratio. Moreover, we obtain, for each firm in the
sample, their respective 4-digit NACE industry classification code. This allows us to link the
sample data to the exposure data.
A key challenge in the calibration process is that the volatility of market values is not
directly observable for most firms in the Orbis van Dijk sample of Dutch firms. For each firm
in the sample that is publicly listed, we determine the implicit asset value volatility based on
the observable volatility of its listed stock. We do this by linking the listed firms in the Orbis
van Dijk sample to Thomson Reuters Datastream using their ISIN-codes and obtaining the
standard deviation of yearly total equity returns including dividends (computed using the
Datastream return index) between 2006 and 2017. We choose this 12-year period to minimize
the number of firms for which there is no complete time series available, while still including
the variation caused by the global financial crisis in 2007 and 2008. We then transform the
standard deviation of equity into implied asset volatility by using the Merton equations in
section 2.2 (i.e., simultaneously solving for equations (6) and (9)). We exclude firms for which
there are more than three missing values in the 12-year period.
For non-listed firms, we cannot observe the standard deviation of yearly total equity
returns. Excluding non-listed firms from the sample can be problematic, however, as this will
likely lead to sample bias (non-listed firms are typically smaller, and may make different
18

choices with regards to risk-taking and leverage). Specifically, smaller firms that are typically
non-listed may have higher asset volatility than larger firms that are more often listed, due to
fewer diversification opportunities within their business boundaries. To obtain an as valid
estimate for asset volatility as possible for non-listed firms, we estimate a predictive model for
the asset volatility of non-listed firms based on relevant financial and size characteristics that
are available within the Orbis database. We first estimate the model using data for listed firms
and the used to predict the asset volatility of non-listed firms.
We estimate four models based on a sample of 2,346 listed firms in the EU-15 in
transition-sensitive industries, including size, profitability, leverage and liquidity as predictors
of asset volatility. All variables except asset volatility are directly obtained from the Orbis
database, while asset volatility is obtained via Datastream using the same methodology as for
the listed Dutch firms only (described in the previous paragraph). We also include country and
industry fixed effects (based on 2-digit NACE industries).
The results for four variants of the OLS-regression are provided in Table 4. For each
model, we report the estimated coefficient as well as its t-value. We start with a simple model
(model 1) that includes only the natural logarithm of total assets as an explanatory variable, as
well as country and industry fixed effects. 15 Total assets are found to be negatively related to
asset volatility (-0.031), implying that smaller firms indeed have higher asset volatility than
larger firms. This relationship is statistically significant at the p<0.01 level. We then introduce
several other potentially relevant variables: the return on assets in model 2 and the leverage
and liquidity ratios in model 3. Model 4 provides a full model that includes total assets, return
on assets, the leverage ratio and the liquidity ratio. Profitability (i.e., return on assets) is
significant in the full specification (model 4) but does not add substantially to the explained
15
An F-test confirms the significance of the fixed effects. The F-statistic for the country fixed effects is 2.15 (p-
value = 0.0077) and for the industry fixed effects F = 4.20 (p-value = 0.0000).
19

variance – the R2 of model 4 and model 3 are similar at 0.37, while model 3 is simpler by
excluding return on assets. For this reason, we use model 3 as our baseline model to estimate
the asset value volatility of the non-listed firms.
Table 5 reports the summary statistics for the standard deviation of assets (for the
combined set of listed and non-listed firms) and leverage for each sector. The summary
statistics are based on a sample of all firms in the Orbis database that are registered in the
Netherlands and that are funded by a non-zero amount of long-term debt (this is the category
under which the majority of the bank exposures in our analysis would be accounted for). This
results in a sample of 6,595 both listed and non-listed firms in transition-sensitive industries.
To exclude outliers, we use the winsorization technique for the 5% lowest and highest predicted
values. This results in the standard deviation of assets between 0.20 (C.11 Manufacture of
beverages) and 0.44 (B.06 Extraction of crude petroleum and natural gas). With respect to
leverage, we find that mean values per industry range between 0.48 (B.09 Mining support
service activities) and 0.74 (D.35 Electricity, gas, steam and air conditioning supply).
3.3.2 Residential real estate
To obtain a measure the volatility of real estate assets, we use indices of average sales prices
that are obtained from the Dutch statistical office (CBS) Statline database. This dataset
provides house price indices from 1995 to 2018 with yearly intervals. For the Netherlands as a
whole, the average house price has an annual standard deviation of 6.1% over that period. Since
this is an aggregate index, we do not measure the idiosyncratic component of asset volatility.
For this reason, we also look at a set of house price indices in the same Statline database for
the 12 capital cities of the Dutch provinces. The average annual standard deviation of these
indices over the 1995-2018 period is 6.6% (with a cross-sectional standard deviation of 1.1%
across the 12 cities), which is the number we use in our further analysis.
20

To estimate the probability of delinquency, we base ourselves on the historical default
rates on Dutch mortgages. We believe that this provides a sensible estimate since defaults on
mortgages in the Netherlands are only triggered in case of (prolonged) delinquency and not in
case of insolvency. We take the long-run annual probability of default for the Dutch mortgages
to be 0.96% (Stanga, Vlahu, and de Haan, 2017). We multiply this with the time to maturity of
individual mortgages to obtain an estimate of the probability of default over the lifetime of an
average mortgage.
4. Climate stress test of the Dutch banking sector
After developing the stress test model and its calibration, in this section we turn to the results
of our stress test. We first define a set of scenarios which we employ in our stress test (section
4.1). As is common in financial sector stress tests, we aim to investigate a set of scenarios that
is severe, but still plausible. We identify our scenarios based on three characteristics: the level
of the carbon tax (i.e., the price per tonne of CO2e emissions), the timing of the tax (e.g.,
overnight versus phased-in over time), and the scope of its application (e.g., in a confined
region versus globally). We then apply these shocks to exposure data of the Dutch banking
sector and describe and interpret our findings (section 4.2). We conclude this section with a
sensitivity analysis (section 4.3).
4.1 Shock scenarios
As a starting point, our analysis considers the introduction of a €100 per tonne CO2e carbon
tax, in line with assumptions in Vermeulen et al. (2019). Additionally, we investigate the
introduction of a €50 per tonne CO2e carbon tax and the introduction of a €200 per tonne CO2e
carbon tax. For each level of carbon taxation, we define four policy scenarios. In the first
scenario (I), the carbon tax is applied overnight, and we assume that regional application will
21

prevent firms from passing through the tax to consumers (i.e., there is no room for increasing
prices, and firms are inter-regional price takers). In the second scenario (II), the carbon tax is
phased-in during a 10-year period in which the carbon tax increases linearly. Similar to the first
scenario, there is no pass-through. In a third scenario (III), the carbon tax is applied overnight,
but we allow firms to pass-through 50% of the tax to consumers (representing a situation where
an inter-regional level playing field is maintained, limiting the impact on businesses). Finally,
in a fourth scenario (IV), the carbon tax is phased-in during a 10-year period, and we allow
firms to pass-through 50% of the tax to consumers.
For all scenarios, we assume that the carbon tax comes on top of the set of climate
policies that are already expected (and priced in by the market). In other words, we assume that
the shock is unanticipated. Furthermore, we assume that the introduction of the carbon pricing
policy does not affect market expectations about further climate policies to follow (i.e., this is
a one-shot fix). This assumption is relevant since changing expectations could alter the
(expected) future asset volatility within segments and could, in turn, affect the market value of
debt. Also, for all scenarios, we assume that the carbon pricing policy is applied to all emissions
at their source (e.g., at the point where fossil fuels are burned). We hence focus on our analysis
on scope 1 emissions.
To link the scenarios to our model, we relate our scenario assumptions to the parameters
in equation (17). Specifically, we vary the path of carbon prices prices 𝜏𝜏𝑡𝑡 and the share of the
tax that firms are assumed to be able to pass-on to consumers 𝜑𝜑𝑘𝑘,𝑡𝑡 . For scenarios I and III
(overnight application) we set 𝜏𝜏𝑡𝑡 equal to the level of the carbon tax for all years t. For scenarios
II and IV we set 𝜏𝜏𝑡𝑡 to increase linearly from zero to the level of the carbon tax during the first
10 years, and equal to the level of the carbon tax for all years thereafter. For scenarios I and II
we set 𝜑𝜑𝑘𝑘,𝑡𝑡 to zero for all industries k and all time periods t. For scenarios III and IV we set
𝜑𝜑𝑘𝑘,𝑡𝑡 to 0.5 for all industries k and all time periods t except for t equals zero where we assume
22

no pass-through (𝜑𝜑𝑘𝑘,𝑡𝑡 = 0), hence implying a one-year period before repricing can occur. For
real estate segments k the pass-through parameter 𝜑𝜑𝑘𝑘,𝑡𝑡 is zero in all cases. However, to ensure
consistency in the scenarios, we do adjust the taxation costs for households. For the scenarios
without cost pass-through (I and II), we base the taxation costs that are linked to real estate on
the use of natural gas only, since the full burden of the tax for electricity would fall onto the
electricity producers. For the scenarios with 50% cost pass-through (III and IV), we base the
taxation costs that are linked to real estate on the sum of the carbon costs of burning natural
gas and 50% of the carbon costs related to the use of electricity. In all scenarios, we assume
the interest rate 𝑟𝑟𝑘𝑘 to be constant at 2% and the adaptation parameters 𝛾𝛾𝑘𝑘,𝑡𝑡 to linearly increase
from 0% to 10% or 20% over a period of 5 years, the maximum reduction depending on the
potential for electrification and the potential to cost-effectively capture emissions. 16 Details are
provided in Table A1 in the appendix.
Finally, to arrive at shocks per unit of asset value, we estimate total asset value for each
segment k, using equation (18). For corporates, we base ourselves the gross operating surplus,
obtained from Eurostat, which gives an as close approximation as possible of current
profitability. We extrapolate this profitability by calculating the perpetuity value of a cash flow
of the same magnitude, using the same discount rate (6%) as used in the NPV calculation of
the carbon tax. For real estate, we base ourselves on the average sales prices per housing type,
obtained from CBS Statline. This results in a vector of asset valuation shocks 𝜉𝜉𝑘𝑘 that represents
the net present values losses in each industry k as a fraction of total firm value. These vectors
fully quantify our scenarios, and hence any given scenario that produces the same vector 𝜉𝜉𝑘𝑘
will produce the same stress test outcome. 17
16
Since we lack the data to estimate the adaptation potential over time in all industries, we make conservative
assumptions. We also note that the adaptation parameter reflects the net effect of both savings due to lower carbon
emissions (i.e., less tax) and additional costs to achieve those savings.
17
We note here that this opens up the possibility to use the same approach for other type of scenarios (both climate
and non-climate related) that lead to valuation shocks within industries.
23

The resulting asset valuation shocks per industry are provided in Tables 6 and 7. These
tables report the asset valuation shocks 𝜉𝜉𝑘𝑘 that are calculated using equation (18), per industry
(rows) and scenario (I to IV, columns). We find that, under a €100 carbon tax, corporate
valuation shocks range between minus 2% (0.02) and minus 89% (0.89). For each industry,
shocks are greatest under scenario I (overnight application, no pass-through), followed by
scenario III (overnight application, 50% pass-through), scenario II (10 years phase-in, no pass-
through, and scenario IV (10 years phase in, 50% pass-through). Industries that are
experiencing the highest decline in asset value are the manufacture of basic metals (C.24), the
manufacture of coke and refined petroleum products (C.19), and electricity, gas, steam and air
conditioning supply (D.35). Shocks for these industries range between no less than 0.83 and
0.87. For the real-estate segments, asset valuation shocks range between a relatively modest
0.018 and 0.038. The greatest asset valuation shocks are observed under scenario III, which for
detached or semi-detached houses yields an asset valuation shock of 0.038, followed by
terraced houses (0.035) and apartments (0.028). We note that even modest real estate asset
valuation losses can result in substantially increased default risk since most households finance
their homes with significant leverage.
4.2 Estimates of the impact of a carbon tax on Dutch banks’ balance sheets
Our main results are shown in Table 9. This table reports the stress test results for our four
carbon tax scenarios, at three carbon tax levels: €50 / tonne, €100 / tonne, and €200 / tonne.
We report both the market value losses for the sample of the three largest Dutch banks, as well
as an extrapolation based on market share for the entire Dutch banking sector (using a factor
of 1.27). 18 The table also presents the market value losses per major asset class, including
18
The largest three Dutch banks cover 79% of the total assets in the Dutch banking sector. The total Common
Equity Tier 1 (CET1) capital for the entire Dutch banking sector was €120 billion in 2017. Total assets were
€2,381 billion.
24

corporate loans and debt, corporate equity, and residential mortgages. Finally, the table reports
the market value as fractions of total Common Equity Tier 1 (CET1) capital in the Dutch
banking sector and the total assets in the Dutch banking sector.
From the table, we distil three major findings. First, for the main estimates for a €100
per tonne carbon tax, total market value losses for the whole Dutch banking system range
between €4.5 billion and €35.8 billion, depending on policy choices made. In the most severe
scenario (I), in which carbon taxation is applied abruptly, and there is no pass-through (e.g.,
due to regional application), losses amount to 29.9% of CET1 capital and to 1.5% of total
assets. When carbon taxation is instead phased-in over ten years (scenario II), the losses as
fractions of CET1 capital and total assets decline to 14.4% and 0.7%, respectively. When
carbon taxation is applied abruptly and allowing for 50% pass-through (scenario III), the losses
as fractions of CET1 capital and total assets are 7.7% and 0.4%. Finally, for the least severe
scenario (IV), in which carbon taxation is phased-in over ten years and allowing for 50% pass-
through, the losses as fractions of CET1 capital and total assets are 3.8% and 0.2%. These
losses are substantial: comparing outcomes to regular stress test exercises by financial
regulators, we find that the market value loss under the most severe policy assumptions
(scenario I) is of the same order of magnitude (29.9%) as the impact on CET1 capital in the
most severe scenarios employed in regular financial sector stress testing. For example, the 2018
stress test by the European Banking Authority (EBA) found that, on aggregate, the CET1
capital of EU banks declines by 19.2% in their adverse scenario, while the EBA also stated that
the 2018 stress test was more severe than any previous EU-wide exercise. 19 Additionally, the
19
https://eba.europa.eu/risk-analysis-and-data/eu-wide-stress-testing/2018. In the adverse scenario, the fully
loaded CET1 capital declines from 1,176 billion to 950 billion, which is equivalent to a decline of 19.2%. Key
features of the adverse scenario of the EBA 2018 stress test were: a cumulative fall in GDP over 3 years by 2.7%,
unemployment reaching 9.7% in 2020, cumulative inflation over 3 years standing at 1.7%, and a cumulative fall
in residential and commercial real estate prices over 3 years of 19.1% and 20% respectively.
25

2019 stress test by the Federal Reserve found that, on aggregate, the Tier 1 leverage ratio of
US banks declines by 19.8% in their severely adverse scenario. 20
Second, losses for the Dutch banking sector are primarily driven by exposures to
corporate loans and debt. In the €100 carbon tax scenarios, the fraction of losses that are driven
by residential mortgages and equity is only between 1% and 2% of total losses. Principal
reasons for this finding are the low exposures of Dutch banks to equity instruments in
transition-sensitive industries (i.e., less than 1% of total assets) and the low net present value
of carbon taxes for most types of housing compared to their valuation, combined with recourse
to a borrowers income next to recourse to the underlying real estate (which puts market value
losses of real estate mostly as a burden to households and not to the banking sector). Looking
at the corporate loans and debt exposure in more detail, the majority of losses for our €100 /
tonne estimates are driven, in declining order, by exposure to electricity, gas, steam and air-
conditioning (D.35), the manufacture of coke and refined petroleum products (C.19), water
transport (H.50), the manufacture of basic metals (C.24), and air transport (H.51). Together,
these five industries drive between 83% and 91% of total market value losses in the corporate
and mortgage portfolios across the four scenarios. The market value losses for all sectors are
reported at a 2-digit NACE aggregation level in Table 10.
Third, the market value losses for Dutch banks’ balance sheets increase exponentially
with the euro amount of the tax. By looking at the same four scenarios but with lower (€50)
and higher (€200) carbon prices, we investigate the sensitivity of our outcomes to the
magnitude of the carbon tax. For example, for scenario I we find that at a price level of €50 per
tonne, the total market value losses amount to €8 billion. This increases non-linearly to €36
20
https://www.federalreserve.gov/publications/files/2019-dfast-results-20190621.pdf. In the adversely severe
scenario the Tier 1 leverage ratio declines from 8.6% to 6.9%, which is equivalent to a decline in CET1 capital of
19.8% if other liabilities are constant. The scenario assumes a global recession with the U.S. employment rate
rising by more than 6 percentage points to 10 percent, accompanied by a large decline in real estate prices (-25%
for house prices and -35% for commercial real estate) and elevated stress in corporate loan markets.
26

billion for a price level of €100 per tonne (a factor 4.5 compared to €50 per tonne) and to €75
billion for a price level of €200 per tonne (a factor 2.1 compared to €100 per tonne). The
exponential increase in market value losses is counteracted by some industries reaching the
point where the full market value of the asset is lost. We find the latter to be the case for four
industries in Scenario I and III at a price of €200 per tonne: the manufacture of coke and refined
petroleum products (C.19), the manufacture of basic metals (C.24), electricity, gas, steam and
air conditioning supply (D.35) and air transport (H.51). Figure 1 provides a graphical
representation of the total market value losses as a function of the price per tonne CO2e of the
carbon tax.
Besides our main results, we provide the results for a specific scenario that estimates
the market value losses for the Dutch banking sector in case a major share of fossil fuel assets
becomes stranded. To this end, we perform an additional analysis where we investigate
stranded assets in the fossil fuel extraction industries covering coal, lignite, oil and natural gas
extraction (B.05 and B.06). Since the direct (scope 1) emissions of these sectors are limited,
they do not play a large role in driving the main stress test results. However, in case of more
stringent climate policies, it is highly likely that these sectors are also affected, either by
reduced demand or by other types of climate policies. For example, McGlade and Ekins (2015)
find that, globally, a third of current oil reserves, half of the gas reserves and over 80 per cent
of coal reserves should remain unused from 2010 to 2050 in order to meet the Paris Agreement
target of keeping global warming below two degrees Celsius. We use these estimates for our
industry valuation shocks as reported in Table 8: 0.85 for the mining of coal and lignite (B.05),
0.34 for the extraction of crude petroleum (B.061), and 0.50 for the extraction of natural gas
(B.062). Results are reported in Table 11. We find that market value losses for this specific
scenario amount to €2.1 billion. This equals 1.8% of total CET1 capital in the Dutch banking
27

sector and 0.1% of total assets. Hence an unburnable carbon scenario alone does not seem to
affect the Dutch banking sector severely.
4.3 Sensitivity analysis
In this subsection, we test the sensitivity of the outcomes of our stress test to changes in key
assumptions. Results are reported in Table 12. In this table, we present the outcomes of our
€100 carbon tax scenarios, based on alternative assumptions relating to the risk-free interest
rate, the ease with which firms can adapt to the carbon tax (e.g., by investing in carbon
abatement technologies), and allowing for firm-level variation in leverage and asset volatility
(instead of taking the average leverage and asset volatility for the entire industry). The first two
represent parameters in our modelling that are relatively hard to determine, either because they
are not known yet (i.e., the future risk-free interest rate) or because the estimates of the
parameters are not readily available (i.e., the industry-specific abatement curves that determine
the rate with which firms can reduce their carbon footprint 𝛾𝛾𝑡𝑡 ). The latter represents a
refinement of our modelling, which we introduce to show that within industry variation in the
calibration parameters matters for the outcome (and loosening our previous assumption of
representative firms in an industry).
The first panel (A) of Table 12 reports outcomes based on a risk-free interest rate that
is constant over time at a rate of 0%, instead of 2% which is assumed in our main analysis. This
assumption reflects a ‘low-for-long’ scenarios with respect to risk-free interest rates in the euro
area. We find that lowering the risk-free interest rate to 0% increases the market value losses
in all four scenarios; however, not to the same extent. Losses as a fraction of CET1-capital are
32.5% (scenario I), 16.6% (scenario II), 9.8% (scenario III), and 5.3% (scenario IV). These
represent increases over the baseline scenario of 8.7%, 15.6%, 27.7%, and 40.7%, respectively.
28

These increases are, to a large extent, driven by market value losses for residential mortgages,
and is primarily due to their relatively long duration k.
The second panel (B) of Table 12 reports outcomes based on an increased ability of
firms to cost-effectively reduce their carbon footprint. Specifically, we double the adaptation
parameters presented in Table A1. This results in adaptation parameters between 20% and 40%,
depending on the potential for electrification and the potential to capture emissions. We find
that increasing the ability to reduce carbon footprints decreases the market value losses for
banks. Losses as a fraction of CET1-capital are 19.4% (scenario I), 9.8% (scenario II), 5.8%
(scenario III), and 2.8% (scenario IV). These represent decreases over the baseline scenario of
34.9%, 32.1%, 24.0%, and 25.6%, respectively. This result is primarily driven by exposures to
corporate loans and debt.
Finally, the third panel (C) of Table 12 reports the outcomes where we introduce firm-
level variation in leverage and asset volatility in the corporate loan portfolio. One challenge to
our results is that we employ a “representative firm” approach, by defining average parameter
values for leverage and asset volatility per industry (e.g., a 2-digit NACE industry). To check
whether this has a substantial impact on results, we estimate our model for each of the 6,595
firms in our sample individually. These include all firms in the Orbis database that are
registered in the Netherlands and that are funded by a non-zero amount of long-term debt
(which is the category under which the majority of bank lending would be accounted for). We
estimate an asset valuation shock for each of these firms, and then aggregate these shocks by
taking the long-term debt weighted average for all firms within an industry. We find that this
approach increases the market value losses for banks. Losses as a fraction of CET1-capital are
34.5% (scenario I), 18.1% (scenario II), 10.1% (scenario III), and 5.2% (scenario IV). These
represent increases over the baseline scenario of 15.4%, 25.6%, 31.5%, and 37.0%,
respectively. We thus show that a representative firm approach may underestimate shocks in a
29

Merton setting, which is at least partly driven by the non-linear increase in market value losses
when solvent firms approach their default point.
5. Conclusion and discussion
Current trajectories of carbon emissions lead to a global warming scenario of three to four
degree Celsius (Rogelj et al., 2013). That is well beyond the safe boundary of keeping global
warming below two degree Celsius. A sudden tightening of climate policies is therefore
possible. Using the Merton methodology to assess the impact of the introduction of a carbon
tax on equity- and debt-type assets allows us to calculate the impact on bank assets. Current
studies of climate stress tests that take a industry-level approach look primarily at losses on
equities and thus underestimate carbon risk, while macro-econometric approaches are
intractable and rely on strong assumptions regarding GDP channels.
Using our novel climate stress testing methodology, we find that 3.8% to 29.9% of the
available Common Equity Tier 1 (CET1) capital of the Dutch banking system is wiped out in
first-round losses following the implementation of a sizeable carbon tax of €100, depending on
the geographical scope of application and abruptness of the policy. These estimates can be seen
as a lower bound, as second-round effects could lead to further losses. Moreover, first-round
losses increase exponentially with the size of the carbon tax. A carbon tax of €200 leads to
first-round losses of 14.9% to 62.6% of the available CET1 capital of the Dutch banking
system. Losses in some of the more severe scenarios are of the same order of magnitude (for
€100 per tonne) and higher (for €200 per tonne) as those obtained under the severe scenarios
in more traditional financial sector stress tests. This indicates that climate-related scenarios are
of relevance to prudential supervisors and may warrant increased alignment between prudential
policymakers and climate policymakers. This to avoid severe losses in the financial system
30

while at the same time achieving a decarbonization of the economy in line with international
climate agreements.
Our analysis has several limitations. First, we focus our attention to market value effects
that are an immediate result of asset valuation shocks. We hence do not account for general-
equilibrium effects, such as potentially increasing unemployment, as well as other second-
round losses due to exposures between financial institutions. Second, the Merton model
assumes no additional costs at bankruptcy and no sudden jumps in asset value. Both of these
limitations are likely to result in a conservative estimation of total losses, implying that our
results reflect a lower bound estimate of total market value losses in the investigated scenarios.
Third, our scenarios do not include potential valuation changes in industries that are not
necessarily carbon intensive but that are dependent on carbon-intensive value chains (such as
the traditional, fossil-fuel based, car industry) or that tend to benefit from climate policies (such
as renewables and electric car producers). Incorporating the potential valuation shocks to such
industries is, in our view, an important avenue for future research. Finally, our analysis assumes
that, besides the asset value shock, the parameters in the Merton model remain constant. We
hence implicitly assume that our scenario shocks do not alter asset value volatility and/or the
risk-free interest rate.
Our findings are relevant to macro-prudential supervisors, micro-prudential
supervisors, and other financial sector participants. For macro-prudential supervisors, our
results show that strong carbon pricing policies have the potential to substantially alter the
market value of a broad range of assets on banks’ balance sheets. This is in line with previous
findings in Battiston et al. (2017) and Vermeulen et al. (2019). As a consequence, from a
systemic mandate, macro-prudential authorities may wish to engage in an ongoing dialogue
with climate policymakers in order to achieve orderly decarbonization of the economy. For
micro-prudential supervisors, our results point to those assets and industries that are of a
31

heightened risk of losing their value in an energy transition. This could have implications for
the risk scoring of individual financial institutions that are, to a greater extent, exposed to these
industries (e.g., specialized banks such as agricultural banks). And finally, our results have
implications for financial institutions in their assessment and pricing of transition-related
financial risks. In particular, it provides estimates of market value losses in the tail-end of the
distribution. This can help financial institutions set risk limits, and could provide input into
their loan origination, investment, and pricing decisions. For the latter, however, a key missing
parameter is the probability of scenarios occurring and having a full set of scenarios. We would
also encourage further research into the pricing of climate risk based on forward-looking
scenarios.
32

References
Acemoglu, B. D., Aghion, P., Bursztyn, L., & Hemous, D. (2012). The Environment and
Directed Technical Change. American Economic Review, 102(1), 131–166.
Batten, S., Sowerbutts, R., & Tanaka, M. (2016). Let’s Talk About the Weather: The Impact
of Climate Change on Central Banks. Bank of England, 603.
Battiston, S., Mandel, A., Monasterolo, I., Schütze, F., & Visentin, G. (2017). A climate
stress-test of the financial system. Nature Climate Change, 7(4), 283–288.
Benos, A., & Papanastasopoulos, G. (2007). Extending the Merton Model: A hybrid approach
to assessing credit quality. Mathematical and Computer Modelling, 46(1–2), 47–68.
Black, F., & Cox, J. C. (1976). Valuing Corporate Securities: Some Effects of Bond
Indenture Provisions. The Journal of Finance, 31(2), 351–367.
Black, F., & Scholes, M. (1973). The pricing of options and corporate liabilities. Journal of
Political Economy, 81(3), 637–654.
Cai, N., & Kou, S. G. (2011). Option Pricing Under a Mixed-Exponential Jump Diffusion
Model. Management Science, 57(11), 2067–2081.
Campiglio, E., Dafermos, Y., Monnin, P., Ryan-Collins, J., Schotten, G., & Tanaka, M.
(2018). Climate change challenges for central banks and financial regulators. Nature
Climate Change, 8(6), 1–13.
Cihák, M. M. (2007). Introduction to applied stress testing (Issues 7–59). International
Monetary Fund.
Connor, G., & Flavin, T. (2015). Strategic, unaffordability and dual-trigger default in the
Irish mortgage market. Journal of Housing Economics, 28, 59–75.
Fabra, N., & Reguant, M. (2013). Pass-through of emission costs in electricity markets.
American Economic Review, 104(9), 2872–2899.
Giesecke, K. (2003). Credit Risk Modeling and Valuation: An Introduction. Ssrn, 607, 1–67.
33

Henry, J., Zimmermann, M., Leber, M., Kolb, M., Grodzicki, M., Amzallag, A., Vouldis, A.,
Hałaj, G., Pancaro, C., Gross, M., Baudino, P., Sydow, M., Kok, C., Cabral, I., &
Żochowski, D. (2013). A Macro Stress Testing Framework for Assessing Systemic
Risks in the Banking Sector. ECB Occasional Paper, 152.
Leaton, J. (2011). Unburnable Carbon – Are the world’ s financial markets carrying a carbon
bubble? In Carbon Tracker Initiative, Investor Watch.
Lehmann, A. (2020). Climate risks to European banks: a new era of stress tests. Bruegel
Blog.
Longstaff, F. A., & Schwartz, E. S. (1995). A Simple Approach to Valuing Risky Fixed and
Floating Rate Debt. The Journal of Finance, 50(3), 789–819.
McGlade, C., & Ekins, P. (2015). The geographical distribution of fossil fuels unused when
limiting global warming to 2 C. Nature, 517(7533), 187.
Merton, R. C. (1974). On the pricing of corporate debt: The risk structure of interest rates.
The Journal of Finance, 29(2), 449–470.
Merton, R. C. (1976). Option pricing when underlying stock returns are discontinuous.
Journal of Financial Economics, 3(1–2), 125–144.
Nieto, M. J. (2019). Banks, climate risk and financial stability. Journal of Financial
Regulation and Compliance, 27(2), 243–262.
Nordhaus, W. D. (1992). An optimal transition path for controlling greenhouse gases.
Science, 258(5086), 1315–1319.
Ong, M. L. L. (2014). A guide to IMF stress testing: methods and models. International
Monetary Fund.
Rogelj, J., McCollum, D. L., Reisinger, A., Meinshausen, M., & Riahi, K. (2013).
Probabilistic cost estimates for climate change mitigation. Nature, 493(7430), 79–83.
Rogelj, J., Popp, A., Calvin, K. V, Luderer, G., Emmerling, J., Gernaat, D., Fujimori, S.,
34

Strefler, J., Hasegawa, T., & Marangoni, G. (2018). Scenarios towards limiting global
mean temperature increase below 1.5 C. Nature Climate Change, 8(4), 325.
Scholtens, B., & Van Der Goot, F. (2014). Carbon prices and firms’ financial performance:
An industry perspective. Carbon Management, 5(5–6), 491–505.
Smale, R., Hartley, M., Hepburn, C., Ward, J., & Grubb, M. (2006). The impact of
CO2emissions trading on firm profits and market prices. Climate Policy, 6(1), 31–48.
Stanga, I., Vlahu, R., & de Haan, J. (2017). Mortgage arrears, regulation and institutions:
Cross-country evidence. De Nederlandsche Bank Working Paper, 580.
Stiglitz, J., Stern, N., Duan, M., Edenhofer, O., Giraud, G., Heal, G. M., la Rovere, E. L.,
Morris, A., Moyer, E., & Pangestu, M. (2017). Report of the high-level commission on
carbon prices.
Sy, W. N. (2014). A Causal Framework for Credit Default Theory. In SSRN Electronic
Journal. Quantitative Finance Research Centre, University of Technology, Sydney.
Upper, C. (2011). Simulation methods to assess the danger of contagion in interbank markets.
Journal of Financial Stability, 7(3), 111–125.
Vermeulen, R., Schets, E., Lohuis, M., Kölbl, B., & Jansen, D. (2019). The Heat is on : A
framework measuring financial stress under disruptive energy transition scenarios. DNB
Working Paper, 625.
(Giesecke, 2003)(Black & Cox, 1976)(McGlade & Ekins, 2015)(Vermeulen et al., 2019)
35

Table 1 – Aggregate assets in the euro area and Dutch banking sector
This table reports the aggregate balance sheet of banks in the euro area and the Netherlands. The shaded
area shows the assets for which we have granular exposure data available and that are in the scope of
our analysis. These assets together represent 59% of the aggregated balance sheet of the Dutch banking
sector. All data is for 2017 and obtained from the ECB Statistical Data Warehouse.
Euro area Netherlands

Percentage Percentage
€ trillion € trillion
of total of total
Equity exposures 0.54 2% 0.02 1%
Corporate loans and debt 5.08 23% 0.59 26%
Residential mortgages 3.84 17% 0.74 32%
Consumer loans (non-mortgage household loans) 1.88 8% 0.05 2%
Government loans and debt 2.82 13% 0.25 11%
Financial corporate loans and debt 3.38 15% 0.32 14%
Central bank loans and debt 1.84 8% 0.16 7%
Other 2.91 13% 0.15 7%
Total 22.30 100% 2.28 100%
36

Table 2 – Carbon footprint and exposures in the corporate loans and debt portfolio
This table reports exposures and carbon intensity per industry. For our analysis, we focus on a subset
of two-digit NACE industries that have carbon footprints higher than 0.5 (we refer to these industries
from hereon as transition-sensitive industries). Carbon footprint data is based on Eurostat carbon
intensity per gross value added and includes emissions of CO2, N2O and NH4 in CO2-equivalents
(CO2e). We adjust the Eurostat carbon intensity data to reflect the profitability of industries as close as
possible by subtracting personnel costs from gross value-added, in order to arrive at carbon emissions
per euro gross operating surplus. Data to make this adjustment is however not available for Agriculture,
forestry and fishing (A.01 to A.03). For these industries we estimate the carbon intensity based by taking
the ratio of gross operating surplus and gross value added from the Dutch national accounts provided
by the Dutch Statistical Office (CBS). Moreover, not all Eurostat carbon intensity data is available at a
2-digit NACE industry level. Industries for which only a higher level of aggregation is available are
denoted with an asterisk (*). Exposure amounts include debt and equity and are based on a sample of
the three largest Dutch banks, covering 79% of the assets in the Dutch banking sector. All figures are
for the Netherlands and for 2017. Data is obtained from the Dutch central bank (DNB) and Eurostat.
Carbon footprint
Exposure
(kg CO2e / euro
(€
gross operating
million)
surplus)
A.01 Crop and animal production, hunting and related service activities 2.75 65,793
A.02 Forestry and logging 0.65 2,946
A.03 Fishing and aquaculture 1.39 1,117
B.05 Mining of coal and lignite 0.56* 0
B.06 Extraction of crude petroleum and natural gas 0.56* 11,307
B.07 Mining of metal ores 0.56* 0
B.08 Other mining and quarrying 0.56* 827
B.09 Mining support service activities 0.56* 9,404
C.10 Manufacture of food products 0.71* 26,499
C.11 Manufacture of beverages 0.71* 6,996
C.12 Manufacture of tobacco products 0.71* 1,018
C.17 Manufacture of paper and paper products 1.26 3,546
C.19 Manufacture of coke and refined petroleum products 9.51 7,153
C.20 Manufacture of chemicals and chemical products 3.24 10,109
C.23 Manufacture of other non-metallic mineral products 2.79 3,076
C.24 Manufacture of basic metals 9.75 3,427
D.35 Electricity, gas, steam and air conditioning supply 10.08 20,434
E.37 Sewerage 5.02* 22
E.38 Waste collection, treatment and disposal activities; materials recovery 5.02* 1,778
E.39 Remediation activities and other waste management services 5.02* 134
H.49 Land transport and transport via pipelines 1.56 9,272
H.50 Water transport 3.96 20,932
H.51 Air transport 8.82 2,284
37

Table 3 – Carbon footprint and exposures in the residential mortgages portfolio
This table reports residential mortgages exposure and carbon footprint per type of dwelling. Exposure
amounts are based on a sample of 9 Dutch banks, covering 67% of the total aggregated residential
mortgages exposure on the balance sheets of Dutch banks. To calculate the carbon footprint, we use the
average natural gas (per M3) and electricity consumption (in kWh) per housing type for 2017 from the
Dutch Statistical Office (CBS) Statline database. We combine these data with emission factors of 1.9
kg CO2e/M3 for natural gas and 0.355 kg CO2e/kWh for electricity. The reported carbon intensities
are based on the annual capital cost which we derive from average sales prices per housing type, also
obtained from CBS Statline, assuming a 3% per annum mortgage interest rate (hence multiplying the
average sales price by 0.03). We exclude residential mortgage exposure for which there is no
classification for the type of dwelling or the type of dwelling is of an uncommon nature (e.g., land-only
and bungalows). The omitted exposure equals €18,145 million (3.6% of total reported exposures).
Carbon footprint (kg CO2e / euro annual Exposure (€ million)

capital cost)
Based on natural gas Based on total
consumption energy consumption
Flat/Apartment 0.25 0.36 65,022
Terraced House 0.31 0.45 51,210
Detached or semi-detached 0.36 0.48 363,103
38

Table 4 – Prediction model for asset volatility of non-listed firms
This table reports the OLS-regression results for different models to predict the yearly standard
deviation of asset value (asset volatility). We base our analysis on a sample of 2,346 listed firms in the
EU-15 in transition-sensitive industries, obtained from the Bureau van Dijk Orbis database. All
variables except asset volatility are directly taken from the Orbis database. We also obtain for all firms
their ISIN code, which we use to obtain the yearly standard deviation of equity value (based on the
return index of stock prices between 2006 and 2018) via Thomson Reuters Datastream. We then
transform the standard deviation of equity into asset volatility by using the Merton equations as put
forward in section 2.2 (i.e., simultaneously solving for equations (5) and (8)). We exclude firms for
which there are more than three missing values in the 12-year period based on which we calculate the
standard deviation of equity value. Furthermore, we exclude firms with the 1% largest and smallest
values for asset volatility and 1% of firms with the largest leverage. This results in an estimation sample
of 1,548 firms. We perform F-tests to confirm the significance of the sets of dummy variables in the
full model. The F-statistic for the country dummy variables is 2.15 (prob > F = 0.0077) and for the
industry dummy variables 4.20 (prob > F = 0.0000), based on the full model (Model 2). T-values are
reported within brackets, *** denotes a significance-level of 1%.
Model 1 Model 2 Model 3 Model 4

-0.031*** -0.016*** -0.019*** -0.031***
Total assets (natural logarithm)
(-9.54) (-4.88) (-6.32) (-10.02)
-0.000 -0.001***
Return on assets - -
(-0.30) (-2.64)
-0.502*** -0.486***
Leverage ratio - -
(-13,88) (-13.66)
0.005*** 0.005***
Liquidity ratio - -
(3.41) (3.50)
Country fixed effects Yes Yes Yes Yes

Industry fixed effects Yes Yes Yes Yes
R-squared 0.26 0.37 0.37 0.26

N 1,532 1,521 1,537 1,548
39

Table 5 – Summary statistics of corporate loans and debt calibration sample
This table reports summary statistics by industry for the sample of firms that we use to calibrate the Merton model. We base ourselves on all firms in the Orbis
database that are registered in the Netherlands and that are funded by a non-zero amount of long-term debt (this is the category under which the majority of the
bank exposures in our analysis would be accounted for). This results in a sample of 6,595 listed and non-listed firms in transition-sensitive industries. For the
non-listed firms, we estimate the standard deviation of assets based on Model 3, as reported in table 4. We winsorize the 5% lowest and highest predicted values,
which results in a range of predicted values for asset variation of individual firms in the sample between 0.05 and 0.49. Note that, in line with the exposures of
the largest three Dutch banks, there are no registered Dutch firms in the NACE industries B.05 and B.07 in the Orbis database.

Standard deviation of assets
Leverage
(estimated based on Model 3)
Asset Asset
Standard Standard
N Mean weighted Mean weighted
deviation deviation
mean mean
A.01 Crop and animal production, hunting and related service activities 2,316 0.23 0.13 0.19 0.56 0.25 0.54
A.02 Forestry and logging 18 0.22 0.15 0.23 0.58 0.27 0.50
A.03 Fishing and aquaculture 102 0.40 0.09 0.36 0.55 0.24 0.47
B.05 Mining of coal and lignite 0 - - - - - -
B.06 Extraction of crude petroleum and natural gas 18 0.44 0.08 0.37 0.54 0.34 0.70
B.07 Mining of metal ores 0 - - - - - -
B.08 Other mining and quarrying 42 0.43 0.08 0.43 0.61 0.24 0.56
B.09 Mining support service activities 56 0.38 0.11 0.32 0.48 0.26 0.37
C.10 Manufacture of food products 548 0.30 0.11 0.20 0.60 0.23 0.46
C.11 Manufacture of beverages 39 0.20 0.09 0.07 0.59 0.20 0.63
C.12 Manufacture of tobacco products 4 0.23 0.14 0.21 0.52 0.34 0.58
C.17 Manufacture of paper and paper products 107 0.29 0.10 0.16 0.63 0.21 0.70
C.19 Manufacture of coke and refined petroleum products 12 0.22 0.13 0.12 0.64 0.24 0.79
C.20 Manufacture of chemicals and chemical products 200 0.25 0.12 0.17 0.56 0.23 0.53
C.23 Manufacture of other non-metallic mineral products 172 0.26 0.11 0.11 0.63 0.21 0.71
C.24 Manufacture of basic metals 76 0.36 0.10 0.32 0.59 0.22 0.45
D.35 Electricity, gas, steam and air conditioning supply 310 0.24 0.13 0.25 0.74 0.25 0.49
E.37 Sewerage 24 0.23 0.13 0.20 0.57 0.27 0.57
E.38 Waste collection, treatment and disposal activities; materials recovery 235 0.22 0.12 0.17 0.66 0.24 0.63
E.39 Remediation activities and other waste management services 80 0.24 0.13 0.20 0.67 0.24 0.65
H.49 Land transport and transport via pipelines 1,649 0.24 0.11 0.16 0.67 0.21 0.57
H.50 Water transport 574 0.31 0.13 0.28 0.66 0.25 0.63
H.51 Air transport 13 0.26 0.14 0.07 0.68 0.26 0.90
41

Table 6 – Estimated asset valuation shocks by industry in €100 carbon tax scenario
This table reports the asset valuation shocks 𝜉𝜉𝑘𝑘 estimated by industry in four different scenarios. All
shocks are reported as net present value losses as a fraction of total firm value, using equations (17) and
(18). The scenarios differ based on the path of carbon prices 𝜏𝜏𝑡𝑡 and the share of the tax that firms are
assumed to be able to pass-on to consumers 𝜑𝜑𝑘𝑘,𝑡𝑡 . Scenarios I and II reflect the situation where there is
no pass-through of costs to consumers (which can be thought of as more regional application, without
a level-playing field), while scenarios III and IV reflect the situation where 50% of the cost of the tax
is passed through to consumers (which can be thought of as more global application, where a level-
playing field is largely maintained). Furthermore, scenarios I and III are based on an overnight
application of the tax, which then remains constant at €100 / tonne CO2e. Scenarios II and IV are based
on a linear phase-in of the tax over a period of 10 years, after which it remains constant at €100 / tonne
CO2e. Carbon emissions are obtained from the Eurostat Air Emissions Accounts (AEA) database, while
total asset value is based on the gross operating surplus as obtained from the Eurostat Structural
Business Statistics (SBS) Database. All data is for the Netherlands and for 2017.
I II III IV
A.01 Crop and animal production, hunting and related service activities 0.25 0.13 0.18 0.09
A.02 Forestry and logging 0.06 0.03 0.04 0.02
A.03 Fishing and aquaculture 0.13 0.07 0.09 0.05
B.05 Mining of coal and lignite 0.05 0.03 0.04 0.02
B.06 Extraction of crude petroleum and natural gas 0.05 0.03 0.04 0.02
B.07 Mining of metal ores 0.05 0.03 0.04 0.02
B.08 Other mining and quarrying 0.05 0.03 0.04 0.02
B.09 Mining support service activities 0.05 0.03 0.04 0.02
C.10 Manufacture of food products 0.06 0.03 0.05 0.02
C.11 Manufacture of beverages 0.06 0.03 0.05 0.02
C.12 Manufacture of tobacco products 0.06 0.03 0.05 0.02
C.17 Manufacture of paper and paper products 0.11 0.06 0.08 0.04
C.19 Manufacture of coke and refined petroleum products 0.87 0.46 0.64 0.32
C.20 Manufacture of chemicals and chemical products 0.30 0.16 0.22 0.11
C.23 Manufacture of other non-metallic mineral products 0.25 0.14 0.19 0.09
C.24 Manufacture of basic metals 0.89 0.48 0.66 0.33
D.35 Electricity, gas, steam and air conditioning supply 0.83 0.45 0.60 0.31
E.37 Sewerage 0.46 0.25 0.34 0.17
E.38 Waste collection, treatment and disposal activities; materials recovery 0.46 0.25 0.34 0.17
E.39 Remediation activities and other waste management services 0.46 0.25 0.34 0.17
H.49 Land transport and transport via pipelines 0.14 0.07 0.09 0.05
H.50 Water transport 0.36 0.18 0.24 0.12
H.51 Air transport 0.80 0.43 0.59 0.30

Table 7 – Estimated asset valuation shocks for residential real estate in €100 carbon tax scenario
This table reports the asset valuation shocks 𝜉𝜉𝑘𝑘 estimated for residential real estate in four different
scenarios. All shocks are reported as net present value losses as a fraction of total real estate value,
using equations (16) and (17). The scenarios differ based on the path of carbon prices 𝜏𝜏𝑡𝑡 and the amount
of the tax that firms are assumed to be able to pass-on to consumers 𝜑𝜑𝑘𝑘,𝑡𝑡 . In the scenarios (I and II)
where there is no pass-through of costs to consumers, we base the asset valuation shock on the use of
natural gas only (since electricity is assumed not to increase in price). In the scenarios (III and IV) where
there is 50% pass-through of costs to consumers, we base the asset valuation shock on the total use of
energy (natural gas and electricity). Furthermore, scenarios I and III are based on an overnight
application of the tax, which then remains constant at €100 / tonne CO2e. Scenarios II and IV are based
on a linear phase-in of the tax over a period of 10 years, after which it remains constant at €100 / tonne
CO2e. All data is for 2017 and obtained from the Dutch statistical office (CBS).
I II III IV
Apartment 0.023 0.018 0.028 0.022
Terraced house 0.028 0.022 0.035 0.027
Detached or semi-detached house 0.033 0.026 0.038 0.030
43

Table 8 – Asset valuation shocks for unburnable carbon in fossil fuel stranded assets scenario
This table reports asset valuation shocks for the extractive industries (coal, lignite, crude petroleum and
natural gas) that are based on the fraction of fossil fuel reserves that cannot be burned if global warming
is to be limited to two degrees Celsius, as reported by McGlade and Ekins (2015). We take the average
value for scenarios with and without Carbon Capture and Storage (CCS).
2-degrees
alignment of
fossil fuel
extraction
B.05 Mining of coal and lignite 0.85
B.061 Extraction of crude petroleum 0.34
B.062 Extraction of natural gas 0.50
44

Table 9 – Market value losses for different carbon tax scenarios, in € million
This table reports the stress test results for our four carbon tax scenarios. Total market value losses are
reported for the sample of the three largest banks and extrapolated to the entire Dutch banking sector
(market estimate). The total Common Equity Tier 1 (CET1) capital for the Dutch banking sector was
€120 billion in 2017. Total assets were €2,381 billion.
Scenario I Scenario II Scenario III Scenario IV

• Regional • Regional • Global • Global
• Abrupt • Phase-in • Abrupt • Phase-in
€50 / tonne carbon tax
Corporate loans and debt 6,259 3,372 1,926 1,073

Corporate equity 0 0 0 0
Residential mortgages 73 57 85 66
Total (three largest banks) 6,332 3,428 2,011 1,140
Total (market estimate) 8,041 4,354 2,555 1,447
% of CET1 capital 6.7% 3.6% 2.1% 1.2%
% of total assets 0.3% 0.2% 0.1% 0.1%

% of CET1 capital 29.9% 14.4% 7.7% 3.8%
% of total assets 1.5% 0.7% 0.4% 0.2%

% of CET1 capital 62.6% 46.7% 34.8% 14.9%
% of total assets 3.2% 2.4% 1.8% 0.8%
45

Table 10 – Market value losses per industry, in € billion (€100 / tonne carbon tax)
This table reports the contribution of individual industries to total market value losses in the four main
€100 / tonne carbon tax scenarios, for the total sample of the three largest Dutch banks. The largest
absolute contributions to market value losses are, in declining order, obtained for electricity, gas, steam
and air conditioning supply (D.35), the manufacture of coke and refined petroleum products (C.19),
water transport (H.50), the manufacture of basic metals (C.24), air transport (H.51) and crop and animal
production, hunting and related service activities (A.01).
I II III IV
A.01 Crop and animal production, hunting and related service activities 1.53 0.85 0.50 0.30
A.02 Forestry and logging 0.01 0.01 0.01 0.00
A.03 Fishing and aquaculture 0.02 0.02 0.01 0.01
B.05 Mining of coal and lignite 0.00 0.00 0.00 0.00
B.06 Extraction of crude petroleum and natural gas 0.17 0.13 0.09 0.06
B.07 Mining of metal ores 0.00 0.00 0.00 0.00
B.08 Other mining and quarrying 0.01 0.01 0.01 0.00
B.09 Mining support service activities 0.03 0.02 0.01 0.01
C.10 Manufacture of food products 0.02 0.01 0.01 0.01
C.11 Manufacture of beverages 0.00 0.00 0.00 0.00
C.12 Manufacture of tobacco products 0.00 0.00 0.00 0.00
C.17 Manufacture of paper and paper products 0.05 0.03 0.02 0.01
C.19 Manufacture of coke and refined petroleum products 5.90 3.76 2.12 0.92
C.20 Manufacture of chemicals and chemical products 0.15 0.06 0.03 0.01
C.23 Manufacture of other non-metallic mineral products 0.09 0.04 0.02 0.01
C.24 Manufacture of basic metals 2.53 0.97 0.39 0.17
D.35 Electricity, gas, steam and air conditioning supply 12.78 4.86 2.13 0.92
E.37 Sewerage 0.00 0.00 0.00 0.00
E.38 Waste collection, treatment and disposal activities; materials recovery 0.28 0.13 0.06 0.03
E.39 Remediation activities and other waste management services 0.02 0.01 0.01 0.00
H.49 Land transport and transport via pipelines 0.04 0.02 0.01 0.01
H.50 Water transport 2.67 1.39 0.91 0.57
H.51 Air transport 1.75 1.18 0.74 0.39
- Detached or semi-detached house 0.12 0.09 0.14 0.11
- Apartment 0.02 0.01 0.02 0.02
- Terraced house 0.02 0.01 0.02 0.02
Total 28.22 13.60 7.25 3.57
46

Table 11 - Market value losses for unburnable carbon, in € million
This table reports the outcome for a partial stress test in a fossil fuel stranded assets scenario based on
the shocks presented in Table 8. Total market value losses are reported for the sample of the three largest
banks and extrapolated to the entire Dutch banking sector. The total Common Equity Tier 1 (CET1)
capital for the entire Dutch banking sector was €120 billion in 2017. Total assets were €2,381 billion.
We note that our sample does not include any exposure to B.05 (mining of coal and lignite).
2-degrees alignment of
fossil fuel extraction
Corporate loans and debt 1,680
Total (three largest banks) 1,680
Total (market estimate) 2,134
% of CET1 capital 1.8%
% of total assets 0.1%
47

Table 12 – Sensitivity analysis
This table reports market value losses for the €100 carbon tax scenarios, using alternative assumptions.
Panel (A) reports outcomes based on a risk-free interest that is constant over time at a rate of 0% instead
of 2%. Panel (B) reports outcomes assuming double the potential for cost-effective carbon abatement
(20% or 40%). Panel (C) reports outcomes by running the Merton model at the firm level in the
corporate portfolios, thereby allowing for variation in the leverage and asset volatility parameters. The
underlined percentages report the difference compared to the main estimate outcome in Table 9.

• Regional • Regional • Global • Global
• Abrupt • Phase-in • Abrupt • Phase-in
(A) risk-free interest rate at 0%

Residential mortgages 897 698 1,060 823
% of CET1 capital 32.5% 16.6% 9.8% 5.3%
% of total assets 1.6% 0.8% 0.5% 0.3%
Difference with main estimate +8.7% +15.6% +27.7% +40.7%
(B) +100% abatement potential

% of CET1 capital 19.4% 9.8% 5.8% 2.8%
% of total assets 1.0% 0.5% 0.3% 0.1%
Difference with main estimate -34.9% -32.1% -24.0% -25.6%
(C) including firm-level variation in leverage and asset volatility

% of CET1 capital 34.5% 18.1% 10.1% 5.2%
% of total assets 1.7% 0.9% 0.5% 0.3%
Difference with main estimate +15.4% +25.6% +31.5% +37.0%
48

Figure 1 –Market value losses in the Dutch banking sector as % of total CET1 capital
This figure shows the total market value losses (as a percentage of total CET1 capital) in the four main
scenarios as a function of the price per tonne CO2e of the carbon tax. In the first scenario (I), the carbon
tax is applied overnight, and we assume that regional application will prevent firms from passing
through the tax to consumers (i.e., there is no room for increasing prices, and firms are inter-regional
price takers). In the second scenario (II), the carbon tax is phased-in during a 10-year period in which
the carbon tax increases linearly. Similar to the first scenario, there is no pass-through. In a third
scenario (III), the carbon tax is applied overnight, but we allow firms to pass-through 50% of the tax to
consumers (representing a situation where an inter-regional level playing field is maintained, limiting
the impact on businesses). Finally, in a fourth scenario (IV), the carbon tax is phased-in during a 10-
year period, and we allow firms to pass-through 50% of the tax to consumers. Discontinuities in the
curve occur when an industry reaches the point where the full market value of the assets is lost. Total
market value losses are based on estimates for a sample of the three largest Dutch banks and
extrapolated to the entire Dutch banking sector. The total Common Equity Tier 1 (CET1) capital for the
entire Dutch banking sector was €120 billion in 2017.
70%
60%
50%
40%
30%
20%
10%
0%
€0 €20 €40 €60 €80 €100 €120 €140 €160 €180 €200
49

Appendix
Table A1 – Parameter assumptions for abatement potential
This table reports the assumptions for abatement potential over time, which we operationalize as a
decline in carbon footprint 𝛾𝛾𝑡𝑡 over time. For all industries, we take a linear reduction of carbon footprint
over five years. We conservatively assume the maximum reduction to be 10% for all industries unless
the industry has strong potential for electrification (land and water transport) or strong potential to
capture emissions (electric power generation). In those cases, we take abatement potential to be 20%.
NACE Abatement
Rev. 2 Industry potential
A.01 Crop and animal production, hunting and related service activities 10% (5 yr)
A.02 Forestry and logging 10% (5 yr)
A.03 Fishing and aquaculture 10% (5 yr)
B.05 Mining of coal and lignite 10% (5 yr)
B.06 Extraction of crude petroleum and natural gas 10% (5 yr)
B.07 Mining of metal ores 10% (5 yr)
B.08 Other mining and quarrying 10% (5 yr)
B.09 Mining support service activities 10% (5 yr)
C.10 Manufacture of food products 10% (5 yr)
C.11 Manufacture of beverages 10% (5 yr)
C.12 Manufacture of tobacco products 10% (5 yr)
C.17 Manufacture of paper and paper products 10% (5 yr)
C.19 Manufacture of coke and refined petroleum products 10% (5 yr)
C.20 Manufacture of chemicals and chemical products 10% (5 yr)
C.23 Manufacture of other non-metallic mineral products 10% (5 yr)
C.24 Manufacture of basic metals 10% (5 yr)
D.35 Electricity, gas, steam and air conditioning supply 20% (5 yr)
E.37 Sewerage 10% (5 yr)
E.38 Waste collection, treatment and disposal activities; materials recovery 10% (5 yr)
E.39 Remediation activities and other waste management services 10% (5 yr)
H.49 Land transport and transport via pipelines 20% (5 yr)
H.50 Water transport 20% (5 yr)
H.51 Air transport 10% (5 yr)
(Sy, 2014).
50

A Finance Approach To Climate Stress Testing

Uploaded by

Copyright:

Available Formats

A Finance Approach To Climate Stress Testing

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Finance Approach To Climate Stress Testing

Uploaded by

Copyright:

Available Formats

A finance approach to climate stress testing*

Henk Jan Reinders

Mathijs van Dijk

Electronic copy available at: https://ssrn.com/abstract=3573107

reassessments of the values of virtually every financial asset.”

Mark Carney, Governor of the Bank of England 1

follow up on the 2015 Paris Agreement (Batten et al., 2016).

Decarbonizing the economy is expected to come at an economic cost, at least in the

Electronic copy available at: https://ssrn.com/abstract=3573107

macro modelling to determine financial sector impacts or focuses exclusively on equity

Electronic copy available at: https://ssrn.com/abstract=3573107

of both equity and debt in a non-linear way.

Electronic copy available at: https://ssrn.com/abstract=3573107

sufficient liquidity to make the periodical payment on a loan).

residential mortgages, we disaggregate exposures to different types of dwellings (i.e.,

data are for 2017.

Electronic copy available at: https://ssrn.com/abstract=3573107

which is a common practice in financial sector stress testing (Cihák, 2007).

Electronic copy available at: https://ssrn.com/abstract=3573107

safeguard financial sector soundness.

Electronic copy available at: https://ssrn.com/abstract=3573107

2. Financial stress test model

assumptions and possible extensions.

2.1 General set-up

Electronic copy available at: https://ssrn.com/abstract=3573107

using a weighting parameter 𝜔𝜔𝑖𝑖 .

𝜗𝜗𝐸𝐸,𝑘𝑘 , 𝜗𝜗𝐷𝐷,𝑘𝑘 = � 𝜔𝜔𝑖𝑖 ∗ 𝑓𝑓𝑀𝑀𝑒𝑒𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 (𝜉𝜉𝑘𝑘 , 𝛳𝛳𝑖𝑖 ) (3)

Electronic copy available at: https://ssrn.com/abstract=3573107

may vary over time t. 8

𝜉𝜉𝑘𝑘 = 𝑓𝑓(𝜏𝜏𝑘𝑘,𝑡𝑡 , 𝛺𝛺𝑘𝑘,𝑡𝑡 ) (4)

𝑀𝑀𝑀𝑀𝐷𝐷 = 𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) − 𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) �𝑁𝑁(−𝑑𝑑2 )� − 𝑉𝑉𝑡𝑡 𝑁𝑁(−𝑑𝑑1 ) (5)

and its equity are given by:

after the shock asset value V* is given by:

Electronic copy available at: https://ssrn.com/abstract=3573107

𝑀𝑀𝑀𝑀𝐷𝐷 ∗ = 𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) − 𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) �𝑁𝑁(−𝑑𝑑2∗ )� − 𝑉𝑉𝑡𝑡∗ 𝑁𝑁(−𝑑𝑑1∗ ) (8)

𝑀𝑀𝑀𝑀𝐷𝐷 ∗ = 1 − �𝑁𝑁(−𝑑𝑑2∗ )� − ((1 − 𝜉𝜉)/𝑅𝑅𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) )𝑁𝑁(−𝑑𝑑1∗ )

𝑀𝑀𝑀𝑀𝐷𝐷 ∗ 1−�𝑁𝑁(−𝑑𝑑2∗ )�−((1−𝜉𝜉)/𝑅𝑅𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) )𝑁𝑁(−𝑑𝑑1∗ )

fashion, the Merton equation for equity is given by:

𝑀𝑀𝑀𝑀𝐸𝐸 = 𝑉𝑉𝑡𝑡 𝑁𝑁(𝑑𝑑1 ) − 𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) 𝑁𝑁(𝑑𝑑2 ) (11)

𝑀𝑀𝑀𝑀𝐸𝐸 ∗ (1−𝜉𝜉)𝑁𝑁(𝑑𝑑1∗ )−𝑅𝑅𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) 𝑁𝑁(𝑑𝑑2∗ )

2.2.1 Merton model limitations and extension to mortgages

Electronic copy available at: https://ssrn.com/abstract=3573107

𝐿𝐿𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) and multiplying its last term by 𝑁𝑁(−𝑑𝑑2 )/ 𝑁𝑁(−𝑑𝑑2 ):

Electronic copy available at: https://ssrn.com/abstract=3573107

could be thought of as representing insolvency, by a more broad probability of default that is a

𝑀𝑀𝑀𝑀𝐷𝐷 𝑉𝑉𝑡𝑡 𝑁𝑁(−𝑑𝑑1 )

which leads to:

𝑀𝑀𝑀𝑀𝐷𝐷 ∗ 1−𝑁𝑁(−𝑑𝑑2∗ )∗𝑃𝑃(𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑)∗ ∗(1−(1−𝜉𝜉)/𝑅𝑅𝑒𝑒 −𝑟𝑟(𝑇𝑇−𝑡𝑡) ∗𝑁𝑁(−𝑑𝑑1∗ )/𝑁𝑁(−𝑑𝑑2∗ ))

but uncorrelated insolvency and delinquency).

2.3 Asset valuation shocks

Electronic copy available at: https://ssrn.com/abstract=3573107

NPVtax, k = ∑𝑇𝑇𝑡𝑡=0 (1 − 𝑟𝑟𝑘𝑘 )𝑡𝑡 ∗ 𝛾𝛾𝑘𝑘 (−𝜏𝜏𝑡𝑡 ) (16)

valuation impact on the value of a physical asset or firm as:

NPVtax,k = ∑𝑇𝑇𝑡𝑡=0 (1 − 𝑟𝑟𝑘𝑘 )𝑡𝑡 ∗ 𝛾𝛾𝑘𝑘,𝑡𝑡 (1 − 𝜑𝜑𝑘𝑘,𝑡𝑡 )(−𝜏𝜏𝑡𝑡 ) (17)

Electronic copy available at: https://ssrn.com/abstract=3573107

3. Data and calibration

non-observable asset volatility of a representative sample of non-listed firms. This is important

cannot directly be observed.

Electronic copy available at: https://ssrn.com/abstract=3573107