Quantifying the Impact of Climate on Corporate Credit Risk

Abstract

Moody’s Analytics Climate-Adjusted EDF™ (Expected Default Frequency) framework provides a consistent, transparent, and customizable means for analyzing physical and transition risks’ impact on public companies’ credit risk. This paper explains the functionality, methodology, and underlying data driving the Climate-Adjusted EDF framework for our physical and transition risk models. Resulting Climate-Adjusted EDF credit measures can be used for stress-testing, loan origination, credit monitoring, asset allocation, and disclosure.

The Physical Risk-Adjusted EDF model forecasts both direct and indirect effects of weather and climate events on businesses’ infrastructure, operations, and markets. Resulting credit measures account for acute physical events (e.g., hurricanes, wildfires, and floods) and chronic physical effects (e.g., sea level rise, heat stress, and water stress). For a range of possible warming paths, the methodology analyzes differential climate exposures based on the time horizon, location of a firm’s physical assets and operations, and firm financial characteristics.

The Transition Risk-Adjusted EDF model forecasts the risks associated with the transition to a lower carbon economy. The effects of carbon transition on a firm’s financial and credit health may be driven by policies such as carbon taxation, variable technological growth, and/or socioeconomic trends. To capture the complicated economic drivers affecting firms, we employ an Integrated Assessment Model (IAM) to understand how a given transition future affects sector-level prices, quantities sold, and costs. We augment the IAM with a model of firm-level competition to understand the effects of transition over time on firm earnings, firm valuations, and, ultimately, firm credit risk.

The Climate PD Converter allows users to adjust their own baseline PDs via the Climate-Adjusted EDF methodology. Based on a firm’s non-climate adjusted probability of default (PD) inputs and other characteristics, the Converter outputs a full set of credit metrics conditional on a range of climate scenarios. This can be useful for analyzing unlisted names, measuring generic sectoral or regional risk, and adjusting internal ratings. The converter allows the user to input asset value, sales revenue, carbon footprint, and scope 1 and 2 emissions. If this information is not available, it imputes these variables.

1. Introduction

Moody’s investment in climate risk analysis is motivated by scientific consensus that global warming poses a major risk to the stability of the international financial system and the world economy. Already, we see that a 1°C global warming above pre-industrial levels affects nearly every facet of the global economy—from infrastructure, agriculture, and real estate to human health and labor productivity.¹ Although fossil-free energy use is on the rise, and many governments have introduced carbon taxes and pledged to achieve net zero greenhouse gas emissions by 2050, the past decade was the warmest on record.²

The increasing pace of climate change and the increasing likelihood of major transition-related policy action have important implications for the financial health of firms across the globe. Firms may suffer direct damages to their physical assets or disruption to their business models due to weather and climate events. Carbon taxation, changing energy prices, and new green technologies confront corporations with risks as well as rewards. Financial institutions with exposure to these firms must understand and carefully measure climate-related risks during loan origination, monitoring, asset allocation, stress-testing, and disclosure.

To quantify the corporate credit risks associated with climate change, Moody’s Analytics has developed a climate-adjusted version of its Public Firm EDF (Expected Default Frequency) model. The Public Firm EDF model is a structural model of credit risk that has been used by global banks, insurers, corporates, and asset managers for more than 30 years. During that time, continuous updates and validations have shown the model’s ability to accurately predict default events in diverse economic environments. The Public Firm EDF model provides a robust framework for understanding the effects of structural climate shocks on corporate credit risk.

To augment the Public EDF framework to account for climate risk, Moody’s Analytics has developed a methodology to account for the effect of climate on the underlying drivers of EDF metrics. The climate-adjusted model integrates climate scenarios devised by the Network for Greening the Financial System (NGFS)³ and state-of-the-art data and assessment tools from the Moody’s MESG team to project the physical and transition risk metrics related to global warming and their impact on credit risk.

Adjusted EDF models allow the user to vary which damage functions are selected to calculate global physical damages in each scenario or to provide customized physical damage paths. Our Climate-Adjusted EDF models provide users with the following functionality:

• 30-year EDF term structures conditional on a climate scenario: For the 40,000 distinct names in the CreditEdge universe, the Climate-Adjusted EDF model produces a probability of default (PD) term structure that forecasts credit risk over time for each climate scenario. We currently offer output based on the NGFS I & II scenarios, a consensus set of possible climate futures employed by many central banks and financial supervisors. We also produce scenarios by the Monetary Authority of Singapore (MAS).
• Conditional valuation metrics: For each bond issuer in the CreditEdge universe, we leverage on the Moody’s EDF model and on the conditional PD term structures and adjust these to the physical and transition climate risks to produce the climate-adjusted EDF term structure as well as a host of intermediate outputs, such as earnings, assets, and implied ratings. The relevant output metrics are listed in Table 1.

• Climate adjustment of user-supplied PDs via the Climate PD Converter: The Climate-Adjusted EDF (CEDF) model is an attractive framework for understanding climate scenario-conditioned credit risk, even when baseline (unconditional) PDs are derived from a model other than the Public EDF model. The Climate PD Converter module allows users to input a baseline PD and key characteristics of a custom entity, such as asset value, sales, carbon footprint, and scope 1 and 2 emissions. The Converter in turn returns climate-adjusted term structures and associated credit metrics for the name. The tool is useful for running unlisted and private names through the model, understanding generic sectoral and regional risks, and adjusting internal or reduced form PD models to account for climate risk.

In the following sections, we first summarize the key elements of the CEDF framework. Second, we describe the Global Change Analysis Model (GCAM) which is used to generate climate scenarios. We then provide details regarding our physical risk methodology where we leverage on firm-specific MESG physical scores to derive firm-level damages from a global damage function. Transition risk employs the sectoral output from GCAM combined with firm-level scope 1 and 2 emissions. These are used to adjust firm-level costs. Using these costs and a model of oligopolistic competition, we disaggregate sectoral output to the firm level. The impact of the two risks is combined via the path of asset values. We generate a joint output and also illustrate aggregated risks in a portfolio of corporate bonds. Finally, we show how customized inputs, such as emissions plus assets and sales from financials can be used to produce climate-adjusted EDFs for both private and public companies.

2. Climate-Adjusted Expected Default Frequency Framework

Moody’s Analytics CEDF models follow the established taxonomy of climate risk, which falls into two broad categories: physical risk and transition risk—see Figure 1. The CEDF model final output is a term structure, the corporate probability of default, one of the credit risk metrics.

• Physical risk encompasses the costs and risks arising from the physical effects of climate change on businesses’ operations, workforce, markets, infrastructure, raw materials, and assets. Physical risks are further delineated as acute (e.g., extreme weather-driven events such as cyclones, hurricanes, or floods) or chronic (e.g., longer term climatic shifts that may cause changes in temperature, precipitation, water stress, or sea level rise).
• Transition risk encompasses the costs and risks associated with the transition to a lower carbon economy, and can include policy changes, such as carbon taxes or cap and trade, new regulations on goods and services, reputational impacts, and shifts in market preferences, norms, and technologies.

Levels of physical and transition risk can vary dramatically between firms. Firms with facilities in South East Asia, the Middle East, and the Caribbean—areas with high exposures to warming-related climate and weather events—will have relatively high physical risk. Firms in industrial sectors such as Coal, Oil & Gas, and Electricity Generation—which are highly exposed to carbon transition—will have relatively high transition risk.

Current best practice, influenced by both stress-testing conventions and climate science methods, is to discretize the continuous distribution of possible economic and climate futures into several representative climate scenarios. Each scenario represents a joint path of economic growth, emissions, and warming over a long period of time (typically, to the year 2100). By analyzing the effects of climate under these scenarios, practitioners can gain an understanding of the plausible physical and transitional impacts that global warming may have on the risk profiles of their exposures.

Once a specific climate scenario is selected for analysis, we carry out several modeling steps to better understand the financial and credit effects of the future climate path in question. First, we model the relationship between raw climate risk drivers and the economic environment at any point of time within a given climate scenario path. Next, we analyze how modeled economic environments affect each firm’s financial health within these environments. Finally, we translate each firm’s financial position into credit risk forecasts at any point within the scenario. The impact is then aggregated at portfolio level.

We forecast the effects of physical and transition risk on the financial drivers of the Moody’s Public EDF model. The Public EDF model is employed to translate the firm’s financial position into its credit risk—see Figure 2. It is a Merton-type structural model of credit risk. Structural credit risk models are defined by explicitly modeling the total firm asset value (enterprise value) process over time and by estimation of the likelihood of a firm’s asset value falling below a lower bound (the default point) within a certain time horizon. If the firm asset value does in fact fall below this default point, the firm is considered insolvent, and it goes into default. Therefore, the likelihood of the firm’s asset value falling below the default point is also the probability of default denoted as the Public EDF (Expected Default Frequency) value. The Climate-Adjusted EDF methodology provides credit risk and valuation metrics for all firms in the CreditEdge universe (i.e., nearly all global publicly traded firms, over 40,000 listed companies).

3. Climate Scenarios

3.1 Attributes of Climate Scenarios

Raw scenario assumptions are converted to a path of future emissions via a climate-focused economic model called an integrated assessment model (IAM). IAMs explore the development of technological, socioeconomic and policy pathways based on anthropogenic emissions defined by the Representative Concentration Pathways (RCPs).⁴ The RCP scenarios refer to radiative forcing and include future trajectories of concentrations of greenhouse gasses and other air-pollutants in the atmosphere as well as changes in the global land use/land cover. They are expressed at watts per square meter. The scenarios incorporate multiple levels of anthropogenic emissions, including low (RCP1.9 and 2.6), intermediate (RCP4.5 and 6.0) and high pathways (RCP7.0 and 8.5). They are further associated with temperature projections up to 2100. Based on the severity of each scenario, a likely range of temperatures is related to each RCP relative to the pre-industrial period. Global surface temperature change for the end of the 21st century (2081–2100) is projected to exceed 1.5°C for RCP2.6 and is likely to reach temperatures above 2°C for RCP4.5 and higher emission scenarios.⁵

Emissions are mapped into global temperatures by use of the Model for the Assessment of Greenhouse-gas Induced Climate Change (MAGICC). Since the mapping of emissions to temperatures is associated with uncertainty, NGFS provides us with a distribution of future temperature paths. We use the median (50th percentile) expected temperature path for two Orderly (Net Zero 2050 and Below 2°C), two Disorderly (Divergent Net Zero and Delayed Transition), and two Hot House World (Nationally Determined Contributions and Current Policies) scenarios. Future temperature paths are then mapped into future damage in percentage of global GDP paths via aggregated damage functions. There is a variety of research studies informing this mapping. In our model, we make sure that the damage path we use for the analysis matches a consensus view of the research community.

The scenarios dependent on the carbon prices as defined by NGFS II narratives (NGFS II, 2021)⁶ can be described as follows:

• Net Zero 2050: Assumes that ambitious climate policies are introduced immediately. Global warming is limited to 1.5°C through stringent climate policies and innovation, reaching global net zero CO2 emissions around 2050. Some jurisdictions such as the US, EU, and Japan reach net zero for all GHGs.
• Below 2°C: Assumes that climate policies are introduced immediately and become gradually more stringent though not as high as in Net Zero 2050. A 67% chance of limiting global warming to below 2°C is given.
• Divergent Net Zero: Assumes that climate policies are more stringent in the transportation and buildings sectors. Net zero is achieved around 2050 but with higher costs due to divergent policies introduced across sectors leading to a quicker phase out of oil use.
• Delayed transition: Assumes new climate policies are not introduced until 2030 and annual emissions do not decrease until then. Strong policies are needed to limit warming to below 2°C. CO2 removal is limited.
• Nationally Determined Contributions (NDCs): Assumes that the moderate and heterogeneous climate ambition reflected in the conditional NDCs at the beginning of 2021 continues over the 21st century (low transition risks). Includes all pledged policies even if not yet implemented.
• Current Policies: Assumes that only currently implemented policies are preserved, leading to high physical risks.

Figure 3 shows the scenario-specific mapping from the carbon price to emissions by scenario. Emissions imply the corresponding temperature pathway, which in turn determines the global damages, according to Kalkuhl & Wenz (2020).⁷

3.2 Shared Socioeconomic Pathways

Shared Socio-economic Pathways (SSPs) were developed to complement the RCPs with varying socioeconomic challenges to adaptation and mitigation. Based on five narratives, the SSPs describe alternative socio-economic futures in the absence of climate policy intervention, comprising sustainable development (SSP1), middle-of-the-road development (SSP2), regional rivalry (SSP3), inequality (SSP4), and fossil-fueled development (SSP5)—see Figure 5. The combination of SSP-based socioeconomic scenarios and Representative Concentration Pathway (RCP)-based climate projections provides an integrative framework for climate impact and policy analysis.

All economic assumptions are taken from the shared socioeconomic pathway 2 (SSP2), designed to represent a “middle-of-the-road” future development:

• Social, economic, and technological trends do not shift markedly from historical patterns.
• Development and income growth proceeds unevenly, with some countries making relatively good progress while others fall short of expectations.
• Global population growth is moderate and levels off in the second half of the century.
• Income inequality persists or improves only slowly.

3.3 Global Change Analysis Model (GCAM)

In assessing climate risk impact, we rely on GCAM that provides us with detailed sectoral and regional analysis. GCAM (see Technical documentation NGFS Phase II)⁸ is a global model that captures interactions among the energy system, water system, agriculture and land use, the economy, and the climate (Figure 6).⁹ GCAM has the following features:

• It consists of a data input framework and the GCAM core.
• The GCAM data system reconciles a wide range of different data sets and systematically incorporates a range of assumptions.
• GCAM generates a dataset with historical and base-year data for calibrating the model along with assumptions about future trajectories such as GDP, population, and technology.
• The GCAM core models capture economic decisions (e.g., land use and technology choices) and their joint dynamics with various human and Earth systems.
• GCAM processes assumptions from the data system to create a complete scenario including future projections of prices; energy and other transformations; and commodity and other flows across regions and into the future.
• GCAM provides the greatest disaggregation spatial dimension, temporal dimension, as well as demand sectors and subsector detail when compared with the other Integrated Assessment Models used by NGFS.
• GCAM allows for the estimation of global and regional mitigation cost, for the analysis of emissions pathways, for the associated land use and energy system transition characteristics, and for quantification of investment.

Figure 7 presents the total amount of energy consumed in transportation industry for the 32 regions defined in GCAM under the delayed transition scenario. International shipping and aviation, passenger and freight are included in this sector. All energy-related outputs in GCAM are reported in EJ/yr (exajoule/year). An EJ is equal to 10¹⁸ (one quintillion) joules. For example, the 2011 Tōhoku earthquake and tsunami in Japan had 1.41 EJ of energy according to its rating of 9.0 on the moment magnitude scale. Furthermore, the annual U.S. total energy consumption amounts to roughly 94 EJ. In Figure 7, China is presented to have the largest energy consumption with a peak around 2050. Other large regional energy consumers, the US and selected European countries,¹⁰ continuously reduce their energy needs, especially after 2030 when climate policies are adopted in the specified scenario.

Figure 8 presents the fuel cost projections for global transportation. Costs in GCAM are shown in $1975/GJ, where one gigajoule (GJ) equals 10⁹ (one billion) joules. A GJ of natural gas is about 25.5 cubic metres at standard conditions, which is approximately equivalent to 27 litres of fuel oil, 39 litres of propane, 26 litres of gasoline or 277 kilowatt hours of electricity. The graph presents increased prices of electricity and hydrogen-generated energy, median prices for oil and petroleum products (refined liquids) and stable low costs for gas and coal.

4. Physical Risk

In 2005, Hurricane Katrina destroyed the equivalent of roughly 9% of the GDP of Louisiana, Florida, Georgia, and Alabama in the United States. Under various climate scenarios, additional global annual physical damage is projected to be close to 3% of GDP by 2050. This is equivalent to a Hurricane of the size of Katrina every three years, everywhere in the world. The exposure to physical risk varies dramatically between regions. Analyzing the exposure of business facilities of a firm to physical risk is therefore paramount for assessing the impact of physical climate risk on a firm’s financial health and its probability of default. In this section, we present a five step methodology to capture physical risk within the Moody’s Public EDF model leveraging Moody’s ESG (MESG) climate scores. We find that accounting for physical climate risk can significantly drive up EDFs.

Our five-step methodology concentrates on the impact of physical climate risk on a firm’s financial position and its probability of default. Physical climate risk enters the framework of Moody’s Analytics Public Firm EDF (Expected Default Frequency) model. Physical climate-adjustment is carried out by adjusting the statistical moments describing the earnings, and thus, asset value process. We quantify the effect of physical risk on these firms’ moments within scenario analyses as follows:

1. Global physical damages. In Section 3, we employ the outcomes of an Integrated Assessment Model (IAM), a Global Circulation Model (MAGICC), and Aggregated Damage Functions (ADF) to obtain the expected global damages paths in percentage of GDP under various climate scenarios.
2. Firm location-specific physical damages. Firms with facilities in areas with high exposures to warming-related climate and weather events will have relatively high physical risk-related damage. We leverage Moody’s ESG (MESG) country and firm-level scores to determine how global damages are distributed across firm locations. We also leverage on the UN Sustainability Index Data and EM-DAT Database for data on natural disasters by country and the World Bank Data for GDP.
3. Firm-specific hazard frequency. We translate firm location damage (in % of GDP) in a hazard frequency. To do so, we calculate the average damage ‘dose’ of significant climate events. We then compute the number of average dosed climate events, which is equivalent to the associated damage of the firm location. More firm-level damage is represented by a higher frequency of average dosed climate hazard events.
4. Firm impact per event in financial terms. We translate economic damages (% of GDP) into financial damages (change in asset value). For this, we leverage the results of a case study on historical public asset return of climate events (Ozkanoglu, 2020)¹¹ to assess the impact on firms’ earnings and asset value upon realization of adverse climate hazard events. As a result, we obtain the total expected impact on the statistical moments describing a firm’s asset value process.
5. Effect on firm EDF drivers. We translate physical climate risk-related earnings shocks into climate-adjusted asset value processes, and therefore the associated EDFs.

Next, we describe each of the five steps of our methodology in more detail.

4.1 Firm Location-specific Physical Damages

We determine how a specific firm’s damage compares to the global average damage. Global average damage is measured in percent of global GDP. We construct measure of a firm’s damage (in percent of a firm’s GDP) and then use our MESG climate risk scores to estimate the relationship between global average damage and a firm’s (facilities) damage. These scores reflect an assessment on the locations of a firm’s production facilities and physical assets. The application matches facility location data against climate model forecasts of relative geospatial exposure to different climate hazards in a warming world. The output is a relative physical risk exposure score from 0 (low risk) to 100 (high risk). MESG physical risk climate scores comprise weighted information of three key sub-components: 1) Operations risk (70%), 2) Supply chain risk (15%), and 3) Market risk (15%), see Figure 9. The MESG score can be considered an ordinal score, ranking relative risk across different firms but not relating proportional risk between two firms based on the ratio of their scores.

Operations risk aggregates the overall risk exposure of a firm’s facilities with regard to specific hazards. Exposure is assessed at the facility-level, which includes screening thousands of facilities. In total, the exposure is assessed to six different climate hazards: Floods, Sea Level Rise, Water Stress, Hurricanes & Typhoons, Heat Stress, and Wildfires. The scores account for the fact that different facilities at the same location may be affected very differently by specific hazard events. The six hazards are complemented by a measure reflecting socioeconomic risk score, reflecting the commitment to ESG-related goals of the countries in which the firm’s facilities are located. Figure 10 showcases the distribution of the Floods sub-score across the facilities for Petroleo Brasileiro SA where the facilities’ exposure to Floods is ranked on a scale between 0 (low risk) and 100 (high risk).

Market Risk depends on climate risk exposure of sales, and Weather Sensitivity measures the sensitivity of economic output to climate variability for a given industry. Supply Chain Risk consists of two indicators. Countries of Origin measures the climate risk in the countries that export the commodities a firm is dependent on for production and delivery of products and services. Demand for resources measures the industry-level dependence on climate-sensitive resources, such as water, land, and energy across the supply chain.

As the data on damage in terms of a firm’s location GDP is not available, we establish the relationship between damages of country k (in % of country GDP) relative to global damage (in % of global GDP) leveraging on MESG country climate scores:

We employ data derived from the United Nations Sustainable Development Goal Indicator Database, the EM-DAT international disaster database, and the World Bank to calculate economic damages associated with weather and climate on a yearly basis by country from 2000 to the present. GlobalDamage_t is the observed average global level of yearly damages as a percentage of world GDP.

Future FirmDamage_i,t (as a % of the firm’s GDP) is a stochastic function of expected global damage and the MESG firm scores. Figure 11 illustrates the projected distribution of damages for Petroleo Brasileiro SA (MESG firm score of 79). We display the expected global damage path under the Below 2°C scenario (grey line). The expected firm damage of Petroleo Brasileiro SA is then calculated as

Note that due to the specification in Equation (1), FirmDamage_i,t follows a lognormal distribution. We plot the realization of 50 randomly drawn events using Equation (2) for each year between 2021 and 2100. Comparing the individual event distribution with the expected damage of Petroleo Brasileiro SA for a given year, one sees that the distribution of damage size of events is heavily skewed towards small damages sizes, i.e. with a high likelihood the damage associated with an event realization is small, and with a small likelihood the damage associated with an event realization is very large. In several realizations, the event damage size will exceed the expected damage by far. For instance, the largest event realization of the 50 simulations for the year 2040 is 9.44% (of firm GDP) – exceeding by far the expected global damage of 1.67%.

4.2 Firm-specific Hazard Frequency

4.3 Financial Impact of Hazard Events

From the previous steps, we have a distribution of hazard-specific frequencies of climate events in any given year. The final step to calibrating asset shocks is forecasting the financial impact upon realization of a climate hazard event. To achieve this, we again turn to the event study estimates reported in Ozkanoglu, et al. (2020). An advantage of using the EDF model to understand climate shocks is that because it has been in use for 30 years, we have estimated EDF asset returns for public firms over that entire time period. The event study looks at the asset returns for firms affected by large climate events over this period, and compares those returns to asset returns of similar firms in the same time period not affected by climate events. The result is a measure of the negative excess asset returns associated with climate events, which we treat as the financial magnitude of asset shocks associated with the climate events.

Asset shocks associated with climate events are assumed to follow a hazard-specific normal distribution, N(μh, σ²_h), where μh describes the mean and σ²_h describes the variance of the normal distribution. In our methodology, we exploit the fact that moment parameters characterizing respective normal distribution depend on a firm’s baseline asset volatility σB—not yet adjusted for physical climate risk. We leverage the Moody’s Analytics event study research to identify the empirical relationships of μh := μh(σB) and σh := σh(σB), and hence N(μh(σB), σh(σB)).

First, we find that the higher the baseline asset volatility σB, the larger the expected negative return μh after a hazard event. A theoretical underpinning for this observation is that a low baseline asset volatility reflects a firm’s general resilience with regard to shocks. Arguably, firms that display a higher resilience in general may also be expected to be more resilient with respect to climate shocks. Second, we find that the higher the baseline asset volatility σB, the more uncertain the actual size of the impact of the hazard event on the asset value. For instance, if the baseline asset volatility is very low, the measured hazard induced asset shock size may be very accurate. For a larger baseline asset volatility the dispersion of measured hazard-induced asset shock may increase since several other factors may contribute to the variability in the asset value.

4.4 Effect on Firm EDF Drivers

We proceed in several steps to introduce physical climate risk in the Moody’s Analytics’s Expected Default Frequencies (EDF) model. First, we briefly outline the baseline CreditEdge Public EDF model, which abstracts from physical climate risk. Second, we show how the EDF term structure is modeled, and illustrate that taking into account physical climate risk amounts to adjusting the firm’s asset value process. Finally, we show how the adjustment of the asset value process is implemented.

4.4.1 Public EDF Model (Current EDF)

The CreditEdge Public EDF model¹² is a Merton-type structural model of credit risk. Structural credit risk models are defined by explicit modelling of the total firm asset value (enterprise value) process over time and by estimation of the likelihood of a firm’s asset value falling below a lower bound (the default point) within a certain time horizon. If the firm asset value does in fact fall below this default point, the firm is considered insolvent, and goes into default. Figure 12 illustrates the asset value process of a firm over a one-year period represented by 50 possible paths of the firm asset value. Using the structural model, we can calculate the percentage of possible asset value paths that fall below the default point within a year. This percentage is the probability of default, labeled in CreditEdge as EDF.

4.4.2 Physical Risk Adjustment EDF Term Structure

The structural approach builds upon a one-year probability of default EDF_B,t=1. Our baseline EDF term-structure over T years is denoted EDF_B,t∀t ∈ {1, ..., T }. Our climate-free approach is taking into the stage of the credit cycle, region, industry, and firm-specific shocks. The main drivers are the asymptotic default tendency (ADT) that reflects the long-term default risk for a specific firm, the aggregate factor (AF) that captures the systemic risk in the credit cycle (conditioned on region), and the firm-specific factor (FF) that captures the firm’s credit risk when compared to the current credit cycle within the corresponding region. We illustrate the interaction between the AF and FF in Figure 13. In panel (a), the aggregate factor (AF) pushes the EDF below the ADT. At the same time the firm-specific factor (FF) pushes the EDF above the ADT, see panel (b). As the impacts of both factors fade away at different speeds, a hump-shaped baseline term structure arises, panel (c).

4.4.3 Climate Adjusted Cash Flow and Market Asset Value Processes

As a next step, we determine how V_B,t and σ_B,t change due to physical climate risk to get V_C,t and σ_C,t. The market asset value of a firm can be thought of as a firm’s ability to generate future cash flow.

The physical climate-adjustment is carried out by adjusting the statistical moments describing the cash flow, and subsequently the asset value process. In panel (a) of Figure 14, we highlight the impact of physical climate change on one realization of the cash flow path. The green line reflects cash flows without climate change, the orange line depicts cash flows after taking into account physical climate risk. In panel (b), we show the impact of physical climate risk for the full sample of realizations of the cash flow process. Note that the climate-induced uncertain shocks affect both the expected value as well as the dispersion of cash flow paths. In panel (c), we plot the transmitted effect on the asset value process, reflecting the discount factor and investors’ expectations. Both the expected asset return as well as the dispersion of the asset value change. Compared to a situation in which investors do not take into account physical climate risk, investors taking into account physical climate risk expect that physical hazard events have an effect on the company’s future abilities to generate cash flows. Since the asset value is the sum of discounted expected future cash flows (see Equation (5)), this causes a drop in the asset value today.

4.5 Case Study: Petroleo Brasileiro SA

Petroleo Brasileiro SA is relatively exposed to physical climate risk as expressed in the value of the MESG composite climate score of 79 (see Table 2). The shape of the baseline EDF term structure in Figure 15 is a result of the baseline model abstracting from climate physical risk.

Figure 16 illustrates the impact of the physical risk on the EDF drivers and on the EDF. In the top-left panel, earnings are affected differently across climate scenarios. Generally, the larger the damages associated with a scenario, the lower the earnings. Since earnings (cash flows) are directly mapped into asset values, larger damages associated with a scenario imply lower asset values. Note that in addition to earnings, asset values are driven by the discount factor and the investors’ expectations. The change in asset values has a direct impact on the distance-to-default, and thus, EDF, as in the bottom-left panel. Introducing physical climate risk generally leads the EDFs to increase. Finally, on the bottom right graph is displayed the difference between baseline forward EDF and climate-adjusted forward EDF (in percentage).

5. Transition Risk

As with the physical risk methodology, our approach to measuring transition risk is to calculate its effects on the drivers of the Public Firm EDF model. We explicitly model the fundamentals driving the asset value process of the firm over time. The fundamentals that give a firm value are the discounted cash flows expected to be accrued. By modeling transition-adjusted cash flows, we can use properly discounted future earnings expectations to model transition-adjusted asset value processes. These asset value processes, in turn, dictate the expected paths of the Public Firm EDF drivers—see Figure 17 for an illustration of the process.¹³

In contrast to the Moody’s Analytics physical risk-adjusted EDF methodology, the explicit forecasting of earnings requires an additional level of modeling. The necessity of the extra step of modeling earnings is twofold. First, in case of physical risk, we have access to the historical analog of the direct effect of past climate events on firm asset values, which can be used to calibrate the effect of future physical damage on the asset process. No such clear historical analog exists to analyze future transition scenarios, so a more theoretically driven calibration method is necessary. Second, the economic effects of climate-driven transition are complex due not only to their heterogeneous and often opposing effects on firms and sectors, but also due to the long-time dimension over which the effects take place. To understand the consequences of transition on firm valuation requires careful modeling of the earnings paths underlying firm asset values.

To understand how earnings are affected by transition, it is necessary to model a scenario’s effect on competitive equilibria both between the production of different good types and on firms producing the same good within each market. Figure 18 shows a more detailed look at how we move in three steps from a transition scenario to earnings paths, the asset value process, and finally transition risk-adjusted EDF metrics.

Step 1: Earnings projections on sectoral/regional level

To understand the effect of the scenario on earnings in each sector/region combination, we leverage on the Global Change Analysis Model (GCAM) discussed previously. GCAM is distinctive among Integrated Assessment Models used by the NGFS to generate scenario pathways for its highly detailed modeling of regions and industrial sectors, providing prices, production quantities, and itemized costs for over a thousand interlinked production technologies. This data enables us to calculate sector-level earnings that account for scenario-conditioned supply and demand shocks arising from each transition future. Table 3 lists our aggregated GCAM Transition Sectors for reference.

Step 2: Earnings projections on a firm level converted into Asset Value Projections

To forecast earnings on a firm level, we augment the GCAM framework with a model of competition within each market. Firms with potential differences in average as well as marginal costs set profit-maximizing output prices, converging to equilibrium market shares and earnings. Other than the calibration based on current market shares, the main driver of heterogeneity between firms are the emissions-intensity and energy-intensity of production. These different intensities, derived through firms’ scope 1 and scope 2 emissions from the MESG dataset, cause relative costs to change over time as emissions and energy costs (typically) rise within a transition scenario. The result is a forecast on how each firm’s market share and earnings will change over time as a result of its new level of economic competitiveness. These are converted into projections of asset values. We employ standard discounted cash flow(DCF) techniques, while giving users full ability to vary discount factors to their own calibrations.

Step 3: Measuring the effect of the scenario-conditional asset value process on EDF metrics

The new asset value process links directly to the drivers of EDF, in turn giving a full term structure of scenario-conditional EDF metrics.

5.1 Sectoral Analysis: Integrated Assessment Model

GCAM is at its heart a complex demand framework, where both downstream producers and end consumers choose between different goods based on both the relative costs of the goods and non-price relative preferences between them. As the transition scenario affects costs of production through emissions taxes, energy prices, capital investment, and technological growth, downstream buyers adjust their consumption share of each good accordingly. Figure 19 shows a simplified diagram of the GCAM product demand structure. On the left-hand side, resource producers extract raw energy materials from each region’s energy reserves. Downstream, energy transformation industries convert these raw energy products into final energy types. Final energy is transmitted through energy carriers to end-product sectors that consume the final energy. Demand of final products are a function of population, income growth, and the price of final goods. Note the locations on the graph where multiple arrows point to one industry. This represents multiple production technologies that are competing within a broad industry, each taking different inputs upstream. Downstream producers and end consumers choose among technologies, creating market shares for each technology. Because of the multi-level demand structure, consumer choice between two products several levels downstream can affect demand for each product’s inputs at the beginning of the production chain.

To see how upstream and downstream product choice fits together to create equilibrium sectoral prices and shares, observe the GCAM example market structure laid out in Figure 20. In this heuristic example, assume electricity can be generated only by burning coal or using solar power. Coal-powered and solar-powered electricity generation technologies each take two generic inputs, with these inputs’ costs determined by the cost of their production in upstream markets. Via the cost equations, the cost of producing each type of electricity is a function of the price of these inputs, as well as the intensity (input-output ratio) of their use in production of electricity output. Note that the price of each technology is equivalent to its cost of production.

By solving for the general equilibrium of the GCAM, we recover several objects for each sector: the price, the quantity, as well as the costs of different kinds of inputs (they are the prices of upstream goods). Furthermore, each GCAM object is dynamic—we observe its current realization as well as the future ones. Nevertheless, granular as the markets in the GCAM can get, they remain sector-level outputs. While sectors—such as solar and coal energy—compete against each other, there is no strategic interaction within sectors. Therefore, to utilize GCAM sector-level output for making firm-level predictions, another model is needed. For this purpose, we develop an oligopoly model and leverage on the concept of Nash equilibrium. The model is described in the next sub-section.

5.2 Firm-Level Analysis: Moody’s Oligopoly Model

The perfect/monopolistic competition assumption implicit to the GCAM framework implies free entry and exit. This means that any excess profits earned by firms in the industry are quickly whittled away by entry, ultimately resulting in zero economic profits in equilibrium (price equals average cost for each firm). Contrary to this result, observed empirical profit margins are not only very different across firms and sectors, but also these differences remain persistent over long periods of time. A more realistic representation of these markets is therefore oligopolistic competition. This market structure, assumes entry barriers and firm heterogeneity, and results in a finite number of firms with the ability to set profit-maximizing prices. The Moody’s oligopoly model thus involves differentiated products, costs, and prices across firms. The costs depend on the scope 1 and 2 emissions, which determine the differentiation among firms. We first describe the key output metrics.

5.2.1 Revenues and Earnings

The model has two core outputs—climate-adjusted revenues and earnings, defined as follows:

5.2.2 Quantities

There are two parameters for the market demand function. α_ƒ is firm-specific and β^m is market-specific. Assuming we know the two demand parameters and the firm price, we can obtain Q_ƒ as follows:

• Use prices P_i , i ∈ F to compute the market share S_ƒ.
• Weighting the price for each firm by its market share, get the market price P^m.
• Assuming we know demand elasticity (to be discussed later), we obtain the output quantity for each market Q^m given the price P^m.
• Finally, use the market shares again to get firm quantities from the market quantities Q^m.

5.2.3 Costs

For each technology sector and period, GCAM generates multiple costs variables. They can be aggregated into objects that correspond to average (i.e., per unit) costs associated with direct scope 1 emissions c₁(t), indirect scope 2 emissions c₂(t), and costs not associated with emissions c_other(t). We can construct firm-level average costs as follows:

Intensities are either provided directly by MESG whenever available or can be calculated using scope 1 and 2 emissions as input. If scope 1 and 2 emissions are not available, we use industry averages, i.e. a₁ = a₂ = 1. We suppress the t subscript for the sake of brevity in the subsequent derivation.

For example, for Chevron in the early policy scenario for oil and natural gas extraction in 2045 we have a₁ = .9603, a₂ = .9720 and (normalized) cost is C_f = 100.389.

5.2.4 Prices

It follows from the firm profit maximization problem and the Nash equilibrium for the oligopoly model that prices are determined by solving the following system of equations for {P_i}_i∈F such that:

Note that both P^m and S_i are functions of {P_i}_i∈F and π^m is a parameter associated with the market price P^m. The detailed derivation of (15) is provided in Appendix A.3. The elasticity will be defined and discussed explicitly in the subsequent sections.

5.2.5 Parameters

Assume that the base period has been realized which implies that we know the market prices P^m and shares S_i∈F in the equation (15). We find β^m using a standard shooting algorithm. For some initial guess β^m, we solve the above system for prices P_i∈F and check if they are consistent with the observed market shares and prices. If yes, the algorithm concludes, if not, we update our guess to another value β^m′ and iterate until convergence. For each firm ƒ,

as market shares S_ƒ are given in the base period.

5.2.6 Market Elasticities Calculation

The last key missing parameter to be defined and estimated is the market elasticity π^m. We have previously remarked that GCAM produces a set of variables for narrowly defined sectors for each one of the NGFS climate scenarios. Let us now elaborate. The output incorporates multilevel results on different regions, sectors, subsectors and technologies. A market represents the combination of regional and sector information. The term for a specific goods market in GCAM is a technology. Each technology is bundled in a set of similar technologies called a subsector, and each subsector with a set of similar subsectors called a sector. For each of these levels, GCAM provides a series of estimated future projections from which prices (costs), supply, demand, CO2 emissions, CO2 sequestration, and emissions costs are used. Consumers or downstream producers can choose not only between different sectors, but also between different subsectors within a sector, and different technologies within a subsector. Grouping technologies in this way allows for rich substitution patterns between technologies as customers choose between options.

We need the substitution elasticities and equilibrium market shares at each of these levels to calculate the full market demand elasticity for each technology. As shown previously, they determine the output for each market.

To estimate market demand elasticity, the entire path of demand for each good needs to be known. This passes out of a firm’s sector, downstream to the next sector/subsector/technology, and then potentially onto several other the next sector/subsector/technology combinations, until it reaches a the final good market (i.e. the market for a good that is not an input). The first step of achieving this is to line up the full stream of output-input relationships downstream from goods produced within each technology. This is possible because we have the input good and the output good for every technology provided in the GCAM output. Since we have that:

an increase in price of a technology (P_tech) will not only decrease the market share of the technology within the subsector, but also put upward pressure on subsector and sector prices respectively. This means that the sector quantity sold will decrease, as technologies downstream that use this good will see their prices increase, and thus lower their market share in the same way:

and iterating across all chains until each has reached the end node. σ can be interpreted as the non-monetary component of the demand function - the higher it is, the greater the quantity sold at each price. One could think about it as of an intercept—if we assume Q = αP^β then log Q = log α + β log P. In the system of equations determining prices, the parameter cancels out and therefore does not affect the end result.

Assuming constant demand elasticity makes computing the effect of changing price of the final good on its demand straightforward. However, suppose that the price of an intermediate good needed to produce the final good changes. Then, we have to figure out how the price of the final good is affected in order to determine the final quantity impact. To do so, we first compute the extent to which the producer of the final good will substitute away from the more expensive input—that is, the change in the market share of the producer of the intermediate good. Knowing the cost of the new combination of inputs used to produce the final good, we compute its new price and quantity. With this in hand, we can determine the final demand for the intermediate good—given its altered share in the market for inputs. If the price of a second-tier intermediate good changes (i.e., a good necessary to produce the intermediate good), we proceed the same way, except that the chain has now two links. Some supply chains in GCAM can have as many as 28 links.

5.2.7 Scenario Adjustment

Information necessary to run GCAM (including the model itself), NGFS assumptions, as well as the NGFS database of the IAM output that corresponds to each climate scenario, are publicly available. However, when regulators conduct exercises intended to measure resilience of various institutions to climate shocks, the exact data used as GCAM (or an alternative IAM) inputs are not widely distributed. Due to the relatively long run times, it is not feasible to reverse engineer these inputs by repeatedly running the model. Therefore, to ensure consistency between the GCAM output for each climate scenario run locally and the public NGFS dataset, a we designed a scenario adjustment procedure.

The granular information of GCAM output at the market, sectoral, subsectoral and technology levels is initially aggregated into the respective broader categories that match the regulator’s grouping. A comparison between the two datasets allows for calculation of adjustment factors for carbon prices, emissions, prices, and quantities in the GCAM output to match the NGFS output at the broader categories level. Some of the changes that are applied to incorporate the NGFS directives, which vary between different releases of the climate scenarios and their corresponding IAM outputs, are future projections of exogenous technological growth assumptions, and socioeconomic growth. As we run GCAM scenarios locally, we get more output variables, i.e. detailed sectoral and subsectoral information, than are typically provided by the NGFS or other regulatory bodies. For all default scenarios to date, we attempt to match the assumptions of a predefined scenario, such as that of the NGFS. We calculate adjustment factors such that GCAM aggregate sectoral, subsectoral and technological groups follow the same growth path over time as the regulator’s aggregate groups. These newly calculated adjustment factors are then applied to the original GCAM output, representing the more granular grouping that is used by our model.

5.3 From Climate Scenarios to EDF

The effect of each transition scenario on EDF metrics is modeled by forecasting the firm’s expected earnings path conditional on a given scenario, and then converting that earnings path to a firm asset value process. These steps are detailed in Figure 21. We first forecast future earnings that equal to a product of a profit margin in each period and quantity sold. As discussed previously, this is achieved by GCAM at the sectoral level and the Moody’s oligopoly model at the firm level. This process allows us to calculate firm level costs over time, the optimal prices each firm sets in equilibrium, the market share of each firm as a function of these prices, and ultimately, each firm’s earnings path. Next, we take the expected earnings paths for the firm and use corresponding discounting factors to estimate the expected firm asset value path today and at any point in the future. The discount factors take into account both the time value of money and the risk aversion to the volatility of earnings around its expected value. Finally, we convert the asset value path to its effect on EDF drivers and calculate the term structure of EDFs from these conditional driver paths.

5.4 Case Study: Chevron Corp

Figure 22 shows the paths for earnings, assets, and climate adjusted EDF (including changes with respect to the baseline) for Chevron Corp. The figures reveal that disorderly scenarios are particularly stressful. For instance, one can see that in the divergent net zero scenario, lower earnings (a) translate through discounted cash flow methods into lower asset values (b). Lower asset values, being a main EDF driver increase the EDF across the tenor, (c) and (d).

6. Portfolio Analysis

We have so far analyzed firms in isolation. Next, we are going to analyze groups of firms (i.e., portfolios). This allows us to identify patterns and firm characteristics that drive the exposure to physical and transition risk.

6.1 Physical Risk

Figure 23 depicts the average relative increase in 30-year EDFs driven by physical risk for our sample of companies. These firms face different level of exposure to physical risk, which is captured by the climate risk scores.

6.2 Transition Risk

Table 4 presents the distribution of the relative increase in 1-year and 10-year EDFs due to transition risk. The results are presented at the sector level considering a large sample of strongly energy-related companies.

6.3 Combined Risk

Figure 24 presents the combined risk, taking into account both the transition risk and the physical risk. Figure 25 shows the distribution of rating changes at the portfolio level.

7. Climate PD Converter

To further increase the coverage and usability of the framework, Moody’s Analytics has built the Climate PD Converter. Figure 26 shows the converter’s general approach. Users provide a baseline unconditional 1-year PD for each name, as well as some important characteristics of the firm associated with its climate exposure and long-term baseline risk. Location and industry are mandatory inputs while the company’s size and emissions are optional. Based on this data, the converter constructs a custom entity, calculating expected values of any missing characteristics given the information provided. The custom name is then run through the physical and/or transition risk models to create full climate-adjusted PDs (and associated credit metrics) for the entity. The output of the tool includes the CEDF term structure and implied ratings. While not a part of the tool output, various bond metrics can be generated as well.

7.1 Use Cases

The adopted approach is useful for the following use cases:

• Climate-adjusted risk for unlisted private names and small and medium enterprises (SMEs): For private names, the structural Public EDF framework provides a more advantageous methodology for understanding the effects of climate change than many of the reduced form credit models standardly employed for the asset class. To take advantage of the Climate-Adjusted EDF methodology, Moody’s Analytics provides climate-adjusted RiskCalc™ EDF credit measures for private names using the Climate PD Converter.
• Custom entities and “generic firms” for an industry/region: Users may be interested in understanding the relative climate-induced credit risk for hypothetical names based on certain characteristics. This can be especially useful for modeling on the sectoral and regional level, allowing measurement of average risk increase for a pool of similar exposures.
• Employing user-provided unconditional PDs in the framework: Users may have their own internal or reduced form credit model and may wish for the output of climate-stressed conditional PDs to be consistent with this input. The Climate PD Converter allows them to input their own baseline/unconditional PD and stress these risk values based on the values provided.

Table 5 shows inputs for a company located in Singapore in the sector of Nuclear Fuel Generation. The EDF for Transition Risk illustrates how the company actually benefits from transition as the EDF actually declines in the Delayed Transition scenarios—see Figure 27.

7.2 Custom Entity Construction

The Custom Entity construction is at the core of the Climate PD Converter as it allows to extend the domain of the previously described methodology designed originally for public companies. As we have seen, the value and volatility of a firm’s assets are sufficient to infer its EDF in this case, assuming constant values of liabilities (the default point). Volatility tells us how much asset value could change in the future, rendering the probability that it will be insufficient to cover the firm’s liabilities, in which case a default occurs. For the publicly traded firms, we know the quoted market price for its equity at any point in time. As the market value of assets and equity are interlinked via elementary accounting rules, we can infer how asset value of a public firm has evolved historically. That is indicative of its future evolution. This idea underlies the use of Black-Scholes-Merton type models for credit risk. The primary limitation of this approach is that there are relatively few public firms compared to the ones held privately. For private firms, equity prices are not quoted on a regular basis, so we know very little about how the market value of their assets evolve. However, the methodology for public companies is to extract the path for the value of assets. With that accomplished, we can make the same kind of adjustments for physical and transition climate risk as we do for public firms. Thus, we obtain the climate-adjusted EDF term structure.

As discussed above, the PD converter relies on the following firm-level data:

1. The markets in which the firm operates (countries and industries),
2. Firm size (measured by assets and sales),
3. Exposure to climate change (physical score and reliance upon carbon emissions),
4. Other financial figures related to the firms’ credit risk (initial PD, asset volatility, etc.).

Not all inputs are required. Many items can be imputed based on information we have on the typical values for firms operating in the same region and/or industry, as discussed below. The first step is to approximate the missing values. If sales or asset value are missing, we infer them based on industry averages. If the PD is missing, we leverage ratings provided by the user to approximate it. With the PD and asset value available, we can evaluate the current asset volatility, as the structural model renders a connection between the three values (EDF, asset value and volatility). Then, with region, industry, sales and asset value in place, we can estimate the firm’s Central Default Tendency, i.e., its EDF in the very long run. This is critical, as it allows us to utilize a proprietary model that exploits the relationship between long-term default probability and asset volatility to estimate the latter. That gives us a long-term asset volatility. By interpolating between the current asset volatility from the Black-Scholes-Merton model and the long-run asset volatility implied by the Central Default Tendency, we get a volatility term structure. With asset value, sales (an excellent proxy for earnings) and volatility (over time) we are in position to approximate the EDF term structure of the firm (not yet climate adjusted) using the Black-Scholes-Merton framework—see Denghan, Nazeran, and Dwyer (2017)¹⁴ and Nazeran, P., and Dwyer, D. (2015).¹⁵

7.3 Climate Risk Adjustment

7.3.1 Physical Risk

A firm’s exposure to physical risk is captured by MESG physical scores, which is a way of incorporating expected global damages into the firm’s financials. In the PD converter, firms are assigned the MESG sovereign score from their respective country. When a firm operates in multiple countries, the score becomes an average of the relevant sovereign scores, weighted by the share of revenue the firm gets from each country. From this point onwards the model proceeds just like for public firms. Note that industry has no impact on physical risk as the sovereign score does not take it into account. In practice, one would want exposures to differ across industries, even for the same country. For example, some facilities may be located closer to water than others on average. Nevertheless, the physical risk model partially compensates for this by assuming that firms with more volatile assets tend to be hit more heavily by acute weather events.

7.3.2 Transition Risk

The integrated assessment model that underlies our transition risk framework—GCAM—provides us with output at very granular market levels (regions and technologies). The PD Converter maps the clients’ firms to the markets in which they operate through the user-provided countries and industries. Carbon emissions are the key drivers of transition risk. They are used to construct intensities (emissions over revenue) at the sector-region level for each firm. Leveraging on the user-provided information on the company’s industry and country, the PD Converter separates the company’s revenue and emissions in all relevant region-market combination (country -> NERJ region and industry -> transition industries). If scope 1 and scope 2 emissions are not provided, the PD converter has two ways of imputing them:

• If the user provides only the total amount of carbon emissions, scope 1 and scope 2 emissions are calculated using the industry-level average share of each scope in the total.
• If no information on emissions is provided, we assume that the company has the same intensity as the industry average. Hence,its emissions-related costs increase at the same rate as they do for the entire industry.

A. Appendix

A.1 Abbreviations

A.2 Firm-specific Hazard Frequencies Example

A.3 Derivation of the System of Equations that Determines Prices

A standard maximization problem of an arbitrary firm is:

Note that the numerator and the denominator of this term have the same functional form as the constant elasticity of the demand function. Note the important difference, though: π_m is an inter-market parameter, measuring the extent to which consumers will substitute towards alternative goods should the market price increase. β^m is the intra-market elasticity that measures the extent to which consumers will substitute away from an individual firm’s products in case of a price hike. Part of this substitution is towards firms in the same market, but as each firm’s price affects the market price, both effects act simultaneously.

The derivative of the market share of each firm with respect to its price is: