Integrating physical climate risk into credit risk modelling

Extreme weather catastrophically affects both borrowers and lenders. Statistics show that 40% of small businesses do not reopen after an inundation, and another 25% fail within a year. Additionally, homeowners fail to pay mortgages following a flood, especially if they have no flood insurance [1].

Borrowers may default due to financial shocks caused by extreme weather events. These events can damage property used as collateral (immovable assets), significantly impacting a bank’s residential and commercial mortgage portfolio.

According to the Basel Committee on Banking Supervision (BCBS), banks should evaluate the potential impacts of extreme weather events on the property values of collateralized loans [2]. Banks also should establish credible criteria to measure the correlation between defaults and property damage resulting from perils. Credit risk models used in bank climate stress testing are designed to estimate standard risk measures in the face of potential shocks, including expected losses. In the banking sector, these expected losses are determined by three main parameters: probability of default (PD), loss given default (LGD), and exposure at default (EAD). These are the main components when calculating the expected loss.

Probability of default (PD) assesses the likelihood that a borrower will fail to repay their loan, with mortgage default generally defined as a loan being 90 days overdue.

Exposure at default (EAD) represents the total value a bank is at risk of losing when a borrower defaults.

Loss given default (LGD) estimates the amount a bank loses when a borrower defaults on a loan, which can be expressed as a percentage or in monetary terms.

While loss given default focuses on the percentage of loss a lender might incur after accounting for recoveries, exposure at default quantifies the total monetary amount at risk at the time of default.

For example, consider a borrower who takes a €250 000 household loan for 18 years. The probability of flooding in this municipality is estimated to be once in 100 years, or 0,2%, indicating a chance of the borrower defaulting. After making payments for 14 years, a flood damages the property. At this moment, the outstanding loan balance is €70 000. This is the amount of money that is still owed on a loan. Due to the flood, the borrower fails to pay the loan during the next two years. The bank will need to sell this damaged property for only €25 000, which is only 10% of the initial property valuation. This updated property value of €25 000 is the expected recovery from the collateral.

In this case, the Exposure at Default (EAD) is €70 000 (the outstanding loan balance). The Loss Given Default (LGD) in monetary terms is calculated as EAD minus the recovery amount from collateral: €70 000 - €25 000 = €45 000. Alternatively, Loss Given Default (LGD) can be represented as a percentage, calculated as LGD in dollars / EAD *** 100 : €45 000 / €70 000 100 = 64%.

A Loss Given Default (LGD) of 64% means that if a borrower defaults on a loan, the bank is expected to lose 64% of the exposure at default (EAD). In other words, under this scenario, after accounting for any recoveries from collateral or other sources, the bank expects that 64% of the outstanding loan balance amount can not be recovered. In this example, with the EAD of €70 000, the lender would expect to lose €45 000 in the event of default.

In many countries, loans are contingent upon having insurance. What do insurance companies say about flood risk? What is the order of magnitude for climate-induced losses?

According to global insurance companies, in the past years climate perils increased in the frequency and severity. They have never recorded so many climate disasters as in 2023 (37 in 2023 compared to 30 in 2020, the previous record). In France, according to France Assureurs, the volume of climate-related claims reached €10.6 billion in 2022 and €6.5 billion in 2023 [3].

Here are several numbers [4, 5, 6] for climate related costs in France :

Ciaran and Domingos storms in the northwest in 2023 : €1.3 Billion

Floods in Northern France in 2023 : €640 Mln

Cyclone Belal in Reunion in 2024 : €100 Mln.

With all the evidence there are residential and commercial mortgages among these affected properties.

It’s also important to highlight that these numbers published by insurance companies do not reflect any property devaluation, any business interruptions, nor the collateral risk incurred by banks either. It would be very interesting to calculate the magnitude of the corresponding Exposure at default (EAD). After reading the annual corporate filings of the major banks, it seems to me that retail banks do not disclose any EAD numbers.

To manage physical climate risks effectively, banks and insurance companies should evaluate the potential financial impacts of weather perils on a borrower’s ability to pay back. In our example here above, we demonstrated such calculations. To hedge climate risk, insurance companies continuously raise the premiums and banks charge higher interest rates for mortgages on properties with greater exposure to extreme weather events.

Every year, insurance companies incorporate new flood observations into the repricing of insurance premiums for all mortgages. In other words, the cost of insurance is indicative of the actual risk.

Visionary asset managers need to be more informed about climate risks associated with a property's location when taking business decisions.

Behind these dry economic figures lie significant cascading social and environmental effects. Municipalities affected by floods and severe droughts see businesses relocate to safer areas, reducing tax revenue. Potential new businesses also may avoid these areas due to the history of devastating floods. Blacklisted municipalities face job losses, increased unemployment, and amplified social inequalities. These revenue and opportunity losses are rarely included in Expected Loss calculations. For instance, when climate perils are reported as a percentage of GDP loss, only direct property losses are considered, not these broader cascading impacts.

Thank you for reading all the way to the end! I would appreciate your thoughts and feedback. Please feel free to share any suggestions and observations with me directly and in the comments below.