Monetary authorities have started requesting financial institutions account for changes in credit risk due to climate change. We combine an acute climate event with a chronic, physical risk climate change scenario and analyze the impact on credit risk losses for UK residential mortgages via PD and LGD projections.
We combine an acute flooding event with the Bank of England’s 2021 Climate Biennial Exploratory Scenario (CBES), to stress credit losses for UK mortgages. The flood event affects House Price Index (HPI) projections at the property level, based on Moody’s ESG climate hazard scores. The HPI affects loss given defaults (LGDs). We then estimate the direct impact of flooding on probability of defaults (PDs), while controlling for national and regional drivers using an IFRS 9/stress testing model embedded in Moody’s Analytics Mortgage Portfolio Analyzer. We add the impact on PD to the effect on LGD to calculate climate-adjusted credit losses.
Many global financial regulators have either already conducted climate stress tests or are planning to do so in the near future. This paper focuses on how climate changes affect residential mortgages, using the UK market as an example. Two key challenges complicate assessing the impact of the climate on risk metrics such as PD or LGD:
- Climate change macroeconomic scenarios that capture only chronic physical risk typically imply minor economic stress.
- The impact of acute physical risk must be incorporated and measured at the location level to accurately reflect a property’s exposure.
The chronic physical risk embedded in climate change scenarios such as CBES 2021 typically has only a small impact on traditional credit risk drivers. Transition to a carbon-free economy, generally achieved via imposing a carbon tax, can induce stress on firms. While transition risk does affect economic drivers of credit risk, our main objective is to fully capture the physical risk’s impact. We follow the route taken by regulators such as the Hong Kong Monetary Authority, which includes an acute climate stress event in addition to incorporating chronic physical risk.
We first investigate which climate hazards have the most impact in the UK, leveraging Moody’s ESG scores for heat stress, floods, hurricanes and typhoons, sea level rise, water stress, and wildfires. Flooding is the primary risk for UK mortgages. We start by estimating the impact on inputs employed in standard IFRS 9/stress testing models that link risk parameters to economic drivers. We consider GDP, unemployment rate, and HPI. Only HPI is significantly affected (in statistical terms) by flooding. In the next step, we search for the effect on default frequency directly. This implies focusing on model output (versus input).1
We use the quantified impact of flooding to construct a super flood for the UK. This climate event affects all regions where flooding has occurred in the past. The imposed regional impact reflects historical severity. We combine the acute event with the CBES climate change scenarios. We further stress property-level HPIs based on postcode-associated flooding hazard scores and calculate projections of PDs and LGDs, with PDs adjusted to flooding risk at the regional level. This exercise generates meaningful impact on credit losses, which are concentrated in properties with the greatest flooding hazard risk.
Our approach to quantifying the impact of climate change on credit risk parameters leverages a framework used by financial institutions for stress testing and IFRS 9 reporting. The framework consists of models linking macroeconomic drivers to PD and LGD. Figure 1 describes the Moody’s Analytics UK Mortgage Portfolio Analyzer off-the-shelf tool (UK MPA), an example of such a framework. This set of models is enhanced by adding climate change scenario and location-specific information regarding climate hazards.
We use a representative sample of 10,000 UK mortgages, total balance £955.7 million, December 2020. There are 123 underperforming accounts representing 1.08% of the total exposure. Moody’s ESG dataset provides facility-level climate risk scores for the following hazards: heat stress, floods, hurricanes and typhoons, sea level rise, water stress, and wildfires. In the UK, flooding poses the primary risk to the performance of mortgage portfolios. Figure 2 maps flooding risk scores2 for the mortgage portfolio and Table 1 presents the summary statistics.
We utilize projections of macroeconomic drivers from the 2021 CBES scenarios that correspond to Early Action (E), Late Action (L), and No Action (N) scenarios. We implement these into the Moody’s Analytics UK MPA to obtain scenario-specific credit risk metrics. The No Action scenario captures chronic physical risk, and it shows economic performance deteriorates gradually. In contrast to the impact of transition risk in the Late Action scenario, the increase in average PD and LGD is minor as compared to the Early Action scenario (Figure 3).
To account for acute physical risk, we construct a hypothetical flood event calibrated to historical data. The flood event hits the UK in Q4 2021. Two impact channels affect the risk parameters. The first channel works via altering the CBES HPI to reflect location-specific climate risk. We quantify the impact the flood event has on HPIs at the Local Authority Unit (LAU) level and translate this effect to the property level using Moody’s ESG flood climate scores. The second channel captures the impact of climate risk on the functional form of PD, while controlling for the effect of the macroeconomic environment and loan-specific performance and application metrics.
We estimate the impact of flooding on HPI using data related to 74 flood events observed in the UK from 1989. The dataset contains information for location, start and end date of each event, duration, severity, magnitude, cause (heavy rain, tropical cyclone, rain and snowmelt, torrential rain, or storm surge), the number of casualties, and the number of displaced persons. Figure 4 presents the areas impacted by a flood event between December 2013 and February 2014 as an example. For each flood observed and for every LAU affected, a dummy variable is triggered when at least 40% of the LAU is affected for up to two quarters since the event. We control for historical macroeconomic drivers when estimating impact. The maximum quarterly decline for a given flood in affected areas is 1%.
We use the greatest estimated coefficient of the flood dummy for each LAU to produce the stress projections at the LAU level using CBES scenarios as the starting point. This implies that we construct a super-flood event where the hypothetical event takes place in all areas where flooding has occurred historically at least once. Next, we aggregate these projections to obtain NUTS 1 forecasts accounting for the acute flooding event (available geographic data in the UK MPA is at the NUTS 1 level).
We further employ Moody’s address-specific ESG flooding hazard scores to adjust the HPI path for each property. By design, accounts with a Flood Risk Score ≤ 27 are not at risk of experiencing a flood event, and the cumulative HPI change is equivalent to the HPI NUTS 1 projection without the acute-event adjustment. Accounts with a Flood Risk Score > 27 are at risk of experiencing a flood event. We stress the HPI projections based on the estimated impact of flood events in HPI and the account-specific flood risk score. Figure 5 depicts the projections for a selected NUTS 1 region. Note, once the HPI projections constructed in this way for each property are aggregated, the HPI at NUTS 1 level corresponds to NUTS 1 forecasts accounting for the acute flooding event (black dashed line). Figure 6 illustrates the absolute and relative increase in LGD due to the flooding event.
The second impact channel for climate risk adjustment implies ex-post adjustment of the PD. We investigate if flooding has explanatory power, in addition to the predicted default rate from the IFRS 9/stress testing model in Moody’s Analytics UK MPA, conditional on macroeconomic variables. The flooding dummy variable is defined as in the case of HPI. The climate-adjusted PD is then calculated for a given event up to several quarters for each region. Figure 7 contrasts the climate-adjusted PD with the standard PD. It also maps regions flooded in the past. Assuming the likelihood of default (PD) is 1% in a given region for a given month, it increases to 2% for red regions, 1.9% for amber regions, 1.3% for yellow regions, and 0% for white regions, respectively. The PD can double for the riskiest areas.
Finally, we compare expected losses over the period from January 2021 to December 2025. We calculate the standard PD and LGD and compare them with losses calculated using the climate-adjusted PD and LGD (Figure 8). The difference between standard and climate-adjusted losses is, respectively, 0.261 basis points (bp) for the Early Policy scenario, 0.232 bp for the Late Policy scenario, and 0.324 bp for the No Policy scenario. Note, the transition to a carbon-free economy in the Late Policy scenario occurs only in 2031 and, hence, the impact on losses is smallest for the period up to 2025.
1In terms of LGDs, we could not conduct a similar analysis, as the default data at our disposal do not include geographic location.
2 The floods score measures the severity and frequency of historical floods, the frequency of future heavy rainfall events, and the intensity of prolonged periods of heavy rainfall.