How Climate Resilience Ratings can help agricultural banks

Climate and credit risk models

Bank credit risk models primarily focus on two variables to manage their loan
portfolios:

• Probability of Default (PD): The likelihood that a business will fail to meet its debt obligations within a specific time frame, typically one year.
• Loss Given Default (LGD): The percentage of the loan the bank expects to lose if a default does occur, after accounting for collateral recovery.

According to a survey recently published by United Nations Environment Programme Finance Initiative (UNEP 2025), 61% of banks are currently incorporating climate risk (including both physical risk and transition risk) into PD modelling; and 43% are currently incorporating climate risk in their LGD measurements. Of banks operating specifically in the agriculture sector, 61% are assessing physical risk and 67% are assessing transition risk.

Challenges to more effective credit risk modelling

It appears that not all banks are investigating climate physical risks with conviction. Doing the minimum to meet regulatory requirements or voluntary, climate-related sustainability targets will not help to address genuine and growing credit risks that are emerging in the agriculture sector. However, modifying risk models requires convincing evidence and robust data.

Proof of the relationship between climate resilience and PD/ LGD is needed before existing models can be adjusted. To derive statistically significant relationships that are sufficient to justify an adjustment of loan terms, granular, farm-level data is required. Gathering such data takes planning and effort but is achievable. Collection of some resilience data can be included in existing customer due-diligence processes. Responsibility for gathering additional data can be outsourced to organisations who have existing relationships with farming communities and can manage the process cost-effectively.

Clearly, maintaining a positive relationship with existing farmer-customers is a key part of the process. If additional data collection efforts are to gain traction, the approach to farmers must be supportive. Provision of data by farmers should be reciprocated. Some farmers may be interested in benchmarking against others in their area; others may be interested in solutions to climate problems that are being effectively deployed elsewhere. These data on resilience and adaptation methods can be shared in a useful form without disclosing personal or confidential information. Incentives for engagement need to go beyond the possibility of favourable loan terms, because not all farmers will qualify initially and will need time to increase their resilience.

Data used in agricultural credit risk models

From these statistics, it appears that a majority of banks have climate physical risk covered. However, the climate metrics used in most credit risk models tend to be spatially coarse and therefore weakly correlated with farm level impacts.

For example, the climate physical risk variables used typically relate to the exposure to climate hazard (such as drought, or flooding) and frequency of occurrence in a given geographic area; but variables explaining interaction of the hazard with the local physical environment, crop cultivation techniques, local drainage and irrigation systems, knowledge and social support networks – i.e. the things that really determine whether a climate hazard actually has a detrimental effect on a farm business – are often missing. The resilience of farm businesses to climate events is highly variable. Neighbouring farms may have entirely different climate risk profiles. To explain the differences, credit risk models need granular data. However, the UNEP survey indicates that granular data for climate risk assessment are largely absent from the analysis. In contrast, traditional credit risk models use highly granular financial data. Sources include:

• Financial data from farm businesses, including historical repayment data (credit history, timeliness of payments), financial statements (revenue, operating expenses, debt, cash flow), collateral value and characteristics of the loans provided by the bank.
• Agricultural data on commodity prices (and their effects on revenue), farm type (crops, livestock) and historical yield (often compared to regional averages).

Banks are currently struggling to access climate physical risk data with similar granularity, but this need not be the case.

How Climate Resilience Ratings can help

The climate resilience surveys of farm businesses that are carried out by companies like Resilience Constellation provide granular data points that can be used in statistical PD and LGD models. The surveys, which gather data directly from farmers, make it possible to determine the resilience of farm businesses to locally relevant climate hazards. Resilience Constellation’s resilience rating produced is based on metrics of:

•  Exposure: Measures of the farmers’ experience of extreme weather events (including drought, flood, heat stress), their frequency and recent trends at each farm location. This is corroborated by empirical weather data from the last decade.
•  Sensitivity: Measures of the susceptibility of the farmland, farm equipment, infrastructure and cultivation methods to physical damage arising from a particular climate hazard.
• Adaptability: Measures of the farmer’s ability to cope with a climate hazard or, more specifically, the ability of farmers to a) recover the climate hazard and b) to make incremental changes in response to climate hazards (also referred to as adaptation).
•  Transformability: Measures of the farmer’s ability to transform an agricultural system or business by making radical changes to cope with a climate hazard, such as switching to an alternative crop or changing the business model.

Climate resilience surveys generate data of particular relevance to credit risk models, including:

•  Land management practices, such a data on use of cover crops, no-till farming, water efficiency (practices that reduce physical risk).
•  Water management practices, such as data on drainage, water storage, abstraction rights, irrigation systems.
• Crop/ livestock diversification and other off-farm revenue streams that reduce reliance on a single, possibly climate sensitive commodity.
• Forward-planning, including farmers understanding of adaptation options, their willingness to commit to climate-smart agriculture and any plans that are already evolving or being implemented.

Integrating resilience data into a PD model

The resilience rating (RR), or desired components, can be added into a bank’s Probability of Default (PD) model such that:

PD = f(Cash Flow, Debt Ratio, RR)

A farm business with a “Highly Resilient” or “A” rating category should have experienced fewer loan defaults during extreme weather events than farm businesses in other less resilient categories. Assuming the model shows a statistically significant relationship between RR and historical low default rates, then a bank could potentially offer better loan terms to more resilient farmers.

Integrating resilience data into a LGD model

Loss Given Default (LGD) can be calculated as:

LGD = 1 – (Recovery Rate + β.RS)

where β is a coefficient that represents the change in recovery associated with a resilience score (RS) for relevant assets. It will be positive if the resilience score is above average, negative if not. The implication is that farm businesses can mitigate collateral depreciation by using land management techniques that protect land value. Climate resilience data can be used in the LGD calculation to determine those that have better collateral. For example, the Resilience Constellation surveys provide data on:

• Soil health. Healthy soils with high organic matter and microbial activity are more resilient to drought and (to some extent) erosion and maintaining their productive capacity => increase in expected land value
• Erosion control. This involves use of terracing, contour ploughing, permanent cover crops and other practices that ensure the land is less susceptible to physical degradation, thereby maintaining the land’s value =>less depreciation in expected land value for pro-active farm businesses.

Timeframes for analysis

When assessing climate physical risks, shorter and medium time-horizons of 0–3 years and 3–10 years are the most commonly used, with about 70% of banks selecting them (UNEP 2025). Taking a short time horizon is the most sensible approach, for two reasons:

• Evidence indicates that climate is already changing in line with the climate model projections produced before 2010 (e.g. Hausfather et al. 2020). As a result, the climate trends established in the recent past are very likely to be a good guide to trends in the near future. We do not need to interpret complex climate models in order to predict what we (or farmers) can already see is happening.
• Assessing physical risks across the longer-term time-horizons of 30 years or more requires the use of a large number of highly speculative assumptions. Future climate may be relatively predictable (albeit for different/ less predictable emissions pathways) but other factors that influence the financial health of a farm business are not predictable. Farm businesses will all adapt (albeit at different speeds) and their production methods and crop selection will change accordingly over this time frame. Technology will evolve and markets will also change. Given all these variables, little that affects the resilience of farm businesses is predictable.

Conclusion

Could your portfolio benefit from granular resilience data?

Explore how our climate resilience ratings can be tailored to your organisations needs.