Flood Risk Model Governance

?Flood Risk in the UK is no small matter.? The UK housing value is over ï¿¡8.7tn (3.8x GDP), and the mortgage outstanding debt stands at ï¿¡1.6tn.? The UK Environmental Agency's 2021 report stated that " at least one in six properties is at risk of flooding in England."

Several exciting advancements have been made in technologies used to measure and assess a catchment area's ability to cope with precipitation in recent years.

?1. Satellite Imagery. Remote sensing has dramatically enhanced our ability to monitor and analyse land cover with increasing granularity

?2. LiDAR (Light Detection and Ranging). Technology has become increasingly crucial for the detailed mapping of catchment topography and features.

?3. In-situ Sensor Networks. Advanced sensor networks provide real-time data on various hydrological parameters:

?4. Advanced Weather Radar. It provides improved precipitation estimates and can distinguish between different types of precipitation.

?These technologies have significantly improved our ability to measure, monitor, and model catchment areas' responses to precipitation. They provide more accurate, timely, and comprehensive data, enabling better water resource management and flood prediction.

?With such advancements, it is opportune to understand how this technology and the modelling that manipulates the raw data work together to assess Flood Risk for any location in the British Isles.? Given that such output is now being used in the financial industry in areas such as mortgage applications, we must now migrate what are primarily academic models into finance that require Model Risk Governance oversight.

?This document presents the finance industry's challenges to gaining sufficient assurance that such models can be used to support financial risk decision-making based on the additional consideration of Flood Risk.

Modelling Flood Risk in the UK

?The modelling of Flood Risk combines sequential processes of measuring, calibration, and simulation. The output of an earlier model becomes the input to the next model in the sequence. The preparation stage divides the country into catchment areas—regions where water naturally drains to a common point, typically a river or a lake.

?1. Measure Typical Flow Rates.? Officers measure typical flood events in each area, noting how the land responds to rainfall. They manually define scores for pre-defined factors like how quickly water flows, how much is absorbed, and how much runs off.

2. Calibrate a Distribution of Flows.? Using the data measured in the first step, statisticians create a distribution of potential flows, including how variable a maximum can be and its frequency.

3. Estimate the Frequency of Flood Flows.? The next step combines the catchment profiles with the statistical models to estimate the relative increase in water flow for less frequent events than typical flow rates.

?4. Calculate Flood Heights by Location.? Modellers then take the flow estimates under different probabilities and adapt fluid dynamic simulations to measure how a burst of rainwater leads to a rise of water in each location.

Model Risk Governance

?The Flood Risk model stack is impressive and results from decades of effort from government agents and commercial entities.? A flood risk map provides excellent insight that can support decisions around significant investments such as infrastructure and highlights clusters of assets exposed to be uninsurable.?

?Applying the lens of Model Risk Governance, here are a couple of initial observations worth highlighting: -

1. Inter-Model Dependency.? Flood Risk combines four model instances: direct measurement, statistical, proxying, and fluid dynamics. The question is their compatibility regarding their underlying assumptions around the variability of maximum flows and frequency.? In model parlance, statistical models are shoehorned into data, which is prone to human inconsistency.

2. Amplification of Errors.? The final simulator takes the output from the first three steps, which are perfectly smooth and error-free distributions of future flow severity and frequency.? The simulation then adds its assumptions on how the landscape will respond. ?Such model setup in banking tends to create an amplification of errors

3. Granularity. ??No matter how granular and sophisticated the flow models are, like weather prediction, the flood simulation will struggle with minor differences between two locations that can lead to different results. ?All those affected by floods point out how some properties avoid being affected while those nearby are inundated.

The saying goes, " All models are wrong, but some are useful.”? The model stack for Flood Risk is undoubtedly valuable for supporting decisions around major infrastructures and clusters of physical assets, including whole portfolios of mortgages.

?There is a difference between models built by academic bodies and those that feed into financial firms' risk origination, management, disclosure, and capital processes. Academics can be sanguine about their embedded assumptions and model weaknesses. In contrast, those used in production by a financial firm under regulation must undergo a rigorous review under a Model Governance framework.

?With the financial world outside of insurance starting to integrate Flood Risk into their risk supply chain, we have a window of opportunity to adopt Model Risk Governance from the outset rather than wait until the industry lurches into a new financial panic.

?