## wpiea2024197-print-pdf

## Source details

**Canonical URL:** [wpiea2024197-print-pdf](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024197-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024197-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024197-print-pdf.pdf.json)

---

### 1. INTRODUCTION — Overview and key findings
- Floods are among the most destructive natural hazards in terms of economic losses: global insured losses from floods in the decade through 2022 were more than 88 billion US dollars, and these losses were more than 30% higher than those during the previous decade.
- Climate change can lead to more severe and frequent flood events, potentially increasing damages and affecting the financial sector.
- Flood scenarios are used to assess how flood damages propagate to the economy and the banking sector through a macro-financial stress-testing framework.
- Key banking-sector findings:
  - Overall, the banking sector can cope with credit losses from flood events, but impacts vary importantly across scenarios.
  - Floods due to breaches in primary defenses and in the lower river courses are more damaging to the banking sector than floods in unembanked areas and other regions.
  - Floods in neighboring countries (Belgium and Germany) do not have large impacts on Dutch banks in the scenarios considered.
  - With climate change (higher hydraulic loads), the adverse impact of floods on the economy and the banking sector is stronger.
  - Accounting for planned reinforcements of flood defenses reduces physical capital damage and mitigates banks’ credit risks.
- Extreme (EDO) scenarios:
  - EDO (Ergst Denkbare Overstromingen; worst credible floods) produce significantly larger banking sector losses, in the range of 30-60 basis points capital deviation from the baseline.
  - EDO scenarios are very severe and may have a return period possibly larger than 1 in a million years, but are still considered credible.
- Contribution and novelty:
  - First to explicitly capture impact of flood protections and their reinforcements using detailed Dutch data and a legislated, forward-looking risk-based adaptation plan.
  - Includes climate change impacts and adaptation in flood scenario calibration; incorporates multiple-breach floods and EDO scenarios.
- Policy relevance:
  - Reinforcement of defenses as legislated in the Netherlands serves as an absorber of adverse economic impacts from floods and mitigates bank credit risks.
  - Framework informed the IMF 2023 Netherlands FSAP assessment.

### Framework and methodology (macro-financial stress-testing pipeline)
- Scenario design and aggregation:
  - A total of 77 flood scenarios calibrated varying by geographical regions, flood types, climate conditions, return periods, and flood defenses and their reinforcements (adaptation).
  - Damages from flood scenarios estimated using detailed data on economic exposures (residential and commercial buildings and infrastructure).
  - Damages aggregated to nation-wide levels and used as inputs to the IMF Global Macro-financial Model (GFM) to generate macro scenarios that incorporate flood impacts.
  - Macro scenarios feed a three-year banking sector stress test estimating banks’ credit losses and resulting capital losses.
- Physical risk propagation channels included:
  - direct destruction of physical capital,
  - impact on total factor productivity (TFP),
  - shock to house prices.
- Scope of banking stress test:
  - Focus on credit risk for six Dutch Systemically Important Institutions (SIs) and loan categories: mortgage, corporates, other retails, financial institutions, government, and qualifying loans.
  - Loans distinguished by country of exposure including the Netherlands, Germany, and Belgium.
  - Banks’ data as of June 2023 from confidential regulatory reporting (COREP and FINREP); PD historical series from DNB.

### 3. Flood scenarios — design, regions, flood types, climate and protection
- Regions (from Ten Brinke et al. (2010) EDO scenarios):
  - Region I: Southwest and Central Coast
  - Region II: Wadden Sea Coast
  - Region III: Rhine and Meuse Rivers
  - Region IV: Lower River Courses
- Flood types and focus:
  - Type A: Flooding in unembanked areas (cannot be privately insured)
  - Type B: Breaches in primary flood defenses (cannot be privately insured)
  - (Other types defined but analysis focuses on A and B)
  - EDO scenarios aligned with type B (breaches) included for contingency planning and can be independent of likelihood.
- Climate conditions:
  - Current climate conditions and future climate conditions under the Dutch W+ scenario (KNMI’14), broadly aligning with RCP 8.5 (IPCC 5th Assessment Report).
  - KNMI’23 was published in October 2023 but water level/hydraulic load estimates based on KNMI’23 were not available at the time of analysis.
- Treatment of flood type B scenarios (three cases):
  - B1: Failure Probability (Reinforcement) = Current Situation; Hydraulic Loads = Current Situation (Readily available flood water depth map)
  - B2: Failure Probability = 2050 Safety Standard; Hydraulic Loads = 2050 (W+) (Combined effects of reinforcement and climate changes)
  - B3: Failure Probability = Current Situation; Hydraulic Loads = 2050 (W+) (Impact of climate changes on current failure probability)
- Return periods considered:
  - Flood type A: 1-in-10, 1-in-100, 1-in-1,000 and 1-in-10,000 years.
  - Flood type B: 1-in-100, 1-in-1,000 and 1-in-10,000 years.
- Breach numbers by region and return period (maximum simultaneous breaches, from Kolen and Nicolai (2023)):
  - Region I: Southwest and Central Coast — Return Period 100: 0; 1,000: 4; 10,000: 7
  - Region II: Wadden Sea Coast — Return Period 100: 0; 1,000: 4; 10,000: 7
  - Region III: Rhine and Meuse Rivers — Return Period 100: 1; 1,000: 3; 10,000: 4
  - Region IV: Lower River Courses — Return Period 100: 1; 1,000: 3; 10,000: 3
- Cross-border combined scenario:
  - For combined Netherlands–Germany–Belgium scenarios only floods with a return period of 1-in-100 years are taken into account due to data limitations in Germany and Belgium.
  - Water depth maps for Germany and Belgium from Jupiter Intelligence.

### 4. Damage estimation — SSM (Deltares Standard Method 2017) and capital shock derivation
- Damage estimation method:
  - Employed the Deltares Standard Method 2017 implemented in the Schade Slachtoffer Module (SSM).
  - SSM uses granular geolocated object data (residences, businesses, infrastructure) at resolutions of 5m, 25m, 50m and 100m grids.
  - Source data: BAG 2022 for buildings, NWB-Wegen 2022 for roads, Top10NL (2022) for railways, plus other object registries.
  - SSM computes expected casualties using a mortality function with inputs including water depth and flow rate.
- Direct damages focus:
  - Computed total direct physical damages for each flood scenario (indirect damages such as business interruption are excluded).
- Damage calculation formula (as used):
  - Damage_s = sum_{i=1}^{N} alpha_{i,s} * n_{i,s} * S_i
    - alpha_{i,s}: damage factor of category i given a certain water depth
    - n_{i,s}: number of objects or m^2 in category i affected by floods
    - S_i: maximum damage per object or m^2 in category i
- Examples of maximum direct damage per category (SSM2017 v4.1 values; Table 4 excerpts):
  - Business — Office: 1,607 m^2; Retail and Commerce: 1,796 m^2; Industries: 1,420 m^2
  - Residential — Single family houses - Structure: 1,295 m^2; Furnishing: 81,985 obj.
  - Infrastructure — Regional roads: 2,243 m; Motorways: 1,520 m; Railroads – electrified: 1,710 m; Railroads – unelectrified: 6,842 m
  - Other — Vehicle: 10,491 obj.; Pumping stations: 1,177,853 obj.; Waste/water treatment plants: 17,107,030 obj.
  - Note: Maximum direct damage in Table 4 is damage from floods within dikes only.
- Capital shock (damage rate) computation:
  - Capital shock = percentage of estimated direct physical damage to pre-damage total capital value.
  - Total capital value proxy: generate a hypothetical flood map with 10 meters of water depth (entire Netherlands submerged), compute damage with SSM, and interpret this damage amount as proxy for total capital value.

### 5. Floods in neighboring countries — methodology, caveats, and integration
- Methodology for Belgium and Germany (Fornino et al. (2024) approach):
  - Flood depths and fraction flooded data: Jupiter Intelligence for a 1-in-100 years return period under SSP5 RCP 8.5 scenario in 2050.
  - Damage functions for floods in Europe: Huizinga et al. (2017).
  - Gridded GDP: Murakami et al. (2021).
  - Aggregate country-level damage rate Dc computed as:
    - Dc = ∑ (di,c * GDPi,c / GDPc) where di,c = fraci,c × dffloods(depthi,c)
  - GDP exposure split into built-up and non-built-up areas using CGLOPS-1 for 2019; built-up uses equally weighted residential, commercial, industrial damage functions; non-built-up uses agriculture and infrastructure functions.
- Disaggregation and selection:
  - Subbasins along Rhine and Meuse divided into level-9 subbasins; flood depth and gridded GDP within each subbasin used to compute damage rates.
  - A damage rate selected from distribution across subbasins, considering historical event sizes; simultaneous flooding of all subbasins considered unrealistic.
- Caveats and limitations:
  - Damage rate is in GDP losses while macro model uses capital stock damage rate as input to a non-linear production function; mismatch may cause underestimation of macro impacts.
  - Regional macro dynamics and regional distribution of banks’ loans are not captured due to lack of granular loan-level data and a dedicated regional model.
  - Germany and Belgium flood scenarios are based on less granular data and rely on proxies.

### Integration into macro-financial and banking stress tests — shock calibration and PD/LGD modelling
- Three types of macro shocks calibrated from aggregated country-level flood damages:
  1. Direct destruction of physical capital: total damage rate aggregated at country level serves as immediate direct shock to capital stock.
  2. Impact on TFP: calibrated at twice the total damage rate and assumed persistent.
  3. House prices shock: calibrated using ratio of direct damages for all residences relative to maximum damages for residences multiplied by the number of residences; adjusted for regional-to-national price elasticity. House price shock applied only to the Netherlands.
- PD modelling:
  - For mortgage, other retail, qualifying revolving, and corporate portfolios: Panel ARDL with logit-transformed PDs:
    - ln(PDi,t / (1−PDi,t)) = αi + λ ⋅ ln(PDi,t−1 / (1−PDi,t−1)) + ∑ βi,s P s=0 zt−s + ui,t
    - zt include economic growth, interest rate, housing price growth, and real wage growth; αi are country fixed effects.
  - Government and financial institution PDs computed via a structural model:
    - PDi,t = (Credit spreadi,t / (1 − Recovery Rate))
  - System-wide PDs transformed to bank-specific PDs by assuming constant differential of risk between aggregate system and individual bank (2022 distance-to-default differences held constant).
- LGD modelling:
  - Secured portfolio LGDs derived using bank-specific LTV projections and other cost factors.
  - Unsecured portfolio LGDs modelled as a function of future PDs.
- IFRS 9 provisioning:
  - Bank-specific transition matrices by sector estimated from historical loan movements and supervisory statistics.
  - Transition probabilities adjusted with scenario-conditional PDs using a “beta-linking” approach.
  - 12-month provision for stage 1 loans and lifelong provision for stage 2 and 3 loans computed. Write-off rate assumed to be zero.
- Analysis focuses on credit risk channel; interest rate and market risk channels excluded.

### Damage magnitudes, selected scenarios, and stress-test results (key statistics preserved)
- Overall capital shocks across 77 scenarios:
  - Range from 0.001 percent of total capital under the A1 flood scenario with 1-in-10 years return period in regions I and II
  - Up to 0.912 percent under the B3 flood scenario with 1-in-10,000 years return period in region III
- House price damages:
  - Range from 1.46 to 24.71 percent under the A1 flood scenario with 1-in-10 years return period in region I and III respectively.
- Twelve scenarios selected for stress tests; four stress-test exercises summarized:
  1. Impact of climate change in unembanked area (Region IV)
     - Mortgage PDs under future climate rise to 2.30 and 2.36 by 2025, relative to 2.15 and 2.07 under current climate and baseline.
     - Flood under current climate causes an additional 11 percent credit losses compared to baseline in 2023; magnitudes increase to 13.5 in 2050(W+) and to 14.7 in 2100(W+).
     - In the 2100(W+) scenario, floods reduce the bank capital ratio by 0.09 percentage points in 2023 relative to baseline.
     - Capital and House Price Shock by Scenario (Return Period: 1,000 years):
       - A1 Current: Capital Shock 0.020; House Price Shock 1.26
       - A2 2050(W+): Capital Shock 0.032; House Price Shock 1.48
       - A3 2100(W+): Capital Shock 0.044; House Price Shock 1.70
  2. Impact of climate change and reinforcement (adaptation) in embanked area (Region II)
     - Example: 1-in-10,000-year flood in Region II under current conditions generates 0.2 percent destruction in capital stocks.
     - Under 2050 conditions, damage rate expected to decrease by 0.138 percentage point thanks to dike reinforcements.
     - In B1 scenario, bank capital losses increase by 18.95 percent relative to baseline in 2023.
     - Climate change adds losses of 0.28 percentage points (B3 effect); lower defense failure probability in 2050 absorbs losses by 0.81 percentage points (adaptation effect).
     - Net result: capital loss rate lower by 0.53 percentage point in total under B2 scenario.
     - Capital and House Price Shock by Scenario (Return Period: 10,000 years):
       - B1 Current: Capital Shock 0.200; House Price Shock 9.6
       - B2 2050(W+): Capital Shock 0.062; House Price Shock 10.5
       - B3 2050(W+) / Current failure probability: Capital Shock 0.241; House Price Shock 9.5
  3. Impact of extreme flood scenarios (EDOs)
     - EDO inundated areas:
       - Region I inundated area: approximately 4,300 km2
       - Region II inundated area: approximately 4,600 km2
       - Together these add up to nearly 26 percent of total land area in the Netherlands.
     - Capital and House Price Shock by Scenario (EDO):
       - EDO-I: Capital Shock 5.4; House Price Shock 13.9
       - EDO-II: Capital Shock 2.4; House Price Shock 13.9
       - EDO-III: Capital Shock 1.9; House Price Shock 17.7
       - EDO-IV: Capital Shock 2.4; House Price Shock 17.4
     - In extreme flood scenarios, bank capital ratio drops by 0.3–0.6 percentage points relative to baseline in the first year, but remains above requirement.
     - Sensitivity: applying a higher house price shock (regional loss to national house value without elasticity adjustment) adds 0.1 percentage point decline in bank capital ratio in year 1.
  4. Impact of floods in neighboring countries (Region III; 1-in-100 years)
     - Dutch banks’ average exposures:
       - 7.6% of total exposures is in Germany.
       - 5.8% of total exposures is in Belgium.
     - Capital and House Price Shock by Scenario (Return Period: 100 years):
       - NLD: Capital Shock 0.124; House Price Shock 10.5
       - DEU: Capital Shock 0.041; House Price Shock -
       - BEL: Capital Shock 0.13; House Price Shock -
     - Under combined Region III 1-in-100 scenario, floods in neighboring countries increase banks’ capital loss rate by 0.06 percentage point relative to scenario with floods only in the Netherlands.
     - Conclusion: floods in Germany and Belgium add small additional bank capital losses but do not materially transmit additional credit risks in modeled scenarios.

### Limitations, uncertainties, and caveats emphasized
- Large uncertainties inherent in climate risk analysis and expert judgement in selecting/calibrating future-climate flood scenarios.
- Aggregation step:
  - Damages calibrated at granular level for the Netherlands are aggregated to the country level to feed the macro model; this may understate sectoral or bank-level heterogeneity.
  - Macro aggregation may produce smaller estimated capital losses than micro borrower-level approaches (example comparison: Caloia et al. (2023) report 40-110 basis points under EDO scenarios versus 30-60 basis points in this macro approach).
- Neighboring-country scenarios (Belgium, Germany) are less granular, do not calibrate specific dike breaches, and rely on proxies due to data gaps; improved granular flood and collateral/exposure data would refine spillover estimates.
- Stress-test scope excludes other channels (interest rate, market risk), possibly underestimating total impacts.

*Source: IMF working paper — chapters 1, 3, 5, and 7.4 from wpiea2024197-print-pdf.*

### 1. INTRODUCTION __________________________________________________________________ 3

### 1. INTRODUCTION

### Overview
- Floods are among the most destructive natural hazards in terms of economic losses: global insured losses from floods in the decade through 2022 were more than 88 billion US dollars, and these losses were more than 30% higher than those during the previous decade.
- Climate change can lead to more severe and frequent flood events, potentially increasing damages and affecting the financial sector.
- Flood scenarios are used to assess how flood damages propagate to the economy and the banking sector through a macro-financial stress-testing framework.

### Framework and methodology
- The framework designs a comprehensive set of flood scenarios varying by:
  - geographical regions,
  - flood types,
  - climate conditions,
  - return periods,
  - flood defenses and their reinforcements (adaptation).
- Damages from the flood scenarios are estimated using detailed data on economic exposures (residential and commercial buildings and infrastructure).
- Damages are aggregated to nation-wide levels and used as inputs to the IMF Global Macro-financial Model (GFM) to generate macro scenarios that incorporate flood impacts.
- Macro scenarios feed a three-year banking sector stress test estimating banks’ credit losses and resulting capital losses.
- Physical risk propagation channels included:
  - direct destruction of physical capital,
  - impact on total factor productivity,
  - shock to house prices.

### Scenarios and scope
- A total of 77 flood scenarios were calibrated, covering multiple geographical regions, flood types, climate conditions, flood protection standards, and return periods.
- Main flood types emphasized:
  - type A: flooding in unembanked areas (cannot be privately insured),
  - type B: breaches in primary flood defenses (cannot be privately insured).
- Analysis covers four independent geographical areas in the Netherlands (coastal and river regions) and also calibrates flood scenarios for neighboring countries (Germany and Belgium) with less granular hazard data.

### Key findings
- Banking sector resilience:
  - Overall, the banking sector can cope with credit losses from flood events, but impacts vary importantly across scenarios.
- Scenario sensitivity highlights:
  - Floods due to breaches in primary defenses and in the lower river courses are more damaging to the banking sector than floods in unembanked areas and other regions.
  - Floods in neighboring countries (Belgium and Germany) do not have large impacts on Dutch banks in the scenarios considered.
  - With climate change (higher hydraulic loads), the adverse impact of floods on the economy and the banking sector is stronger.
  - Accounting for planned reinforcements of flood defenses reduces physical capital damage and mitigates banks’ credit risks.
- Extreme (EDO) scenarios:
  - EDO (Ergst Denkbare Overstromingen; worst credible floods) produce significantly larger banking sector losses, in the range of 30-60 basis points capital deviation from the baseline.
  - EDO scenarios are very severe and may have a return period possibly larger than 1 in a million years, but are still considered credible.

### Contribution to literature and practice
- Novelty:
  - First to explicitly capture the impact of flood protections and their reinforcements using detailed Dutch data and a legislated, forward-looking risk-based adaptation plan.
  - Includes climate change impacts and adaptation in flood scenario calibration; incorporates multiple-breach floods and EDO scenarios.
- Methodological positioning:
  - Uses a “macro approach” aggregating damages at the country level and mapping to macro-financial scenarios via the IMF GFM, complementary to borrower-level “micro approaches”.
  - Comparable approaches cited: Hallegatte et al. (2022) for tropical storms in the Philippines and Dolk et al. (2023) for Mexico.
- Scope for transferability:
  - Framework can be adapted to other countries and financial sectors using country-specific data and flood characteristics.

### Limitations and caveats
- Large uncertainties inherent to climate risk analysis, including expert judgement in selecting and calibrating future-climate flood scenarios.
- Aggregation step:
  - Damages are calibrated at a granular level for the Netherlands but then aggregated at the country level to feed the macro model; this may understate sectoral or bank-level heterogeneity.
  - Compared to micro approaches mapping damages directly to banks’ balance sheets, the macro approach may produce smaller estimated capital losses (example: Caloia et al. (2023) report 40-110 basis points under EDO scenarios versus 30-60 basis points in this macro approach).
- Neighboring-country scenarios (Belgium, Germany) are less granular and do not calibrate specific dike breaches; damage computation relies on proxies due to data gaps.

### Policy relevance and application
- The analysis supports the view that adaptation (reinforcement of defenses as legislated in the Netherlands) serves as an absorber of adverse economic impacts from floods and mitigates bank credit risks.
- Framework informed the IMF 2023 Netherlands FSAP assessment: despite sizeable flood-prone land area, the banking sector exhibits resilience, though climate change can amplify losses over the long run; government reinforcement plans could help mitigate some anticipated losses.

### Structure of the paper (as provided)
- Subsequent sections cover: Dutch flood risk management (Section 2), flood scenario calibration (Section 3), damage computation for the Netherlands (Section 4) and neighboring countries (Section 5), banking sector stress test methodology (Section 6), results (Section 7), discussion (Section 8), conclusions (Section 9), and annexes with additional scenario maps and trajectories.

*Source: IMF working paper — 1. INTRODUCTION (from wpiea2024197-print-pdf).*

### 3.  Flood Scenarios

### 3.  Flood Scenarios

### Overview of scenario design
- A range of flood scenarios was chosen, encompassing various regions, flood types, climate conditions, and flood protection for different return periods.
- Flood maps for each scenario were designed in collaboration with Dutch climate experts from HKV, in partnership with the Ministry of Infrastructure and Water Management (MIENW).
- HKV and MIENW provided information on breach locations and the number of breaches occurring at the same time, for different return periods.
- Based on the scheme described, a total of 77 flood scenarios have been designed.
- Water depth maps for the Netherlands are retrieved from the LIWO (National Water and Flood Information System) database for each breach location.
- If there are multiple breaches in a scenario, water depth maps are manually combined using GIS software; in cases of overlapping inundated areas, the maximum water depth is selected.

### 3.1 Regions
- Flood scenarios focus on four independent geographical areas in the Netherlands, selected from Ten Brinke et al. (2010) EDO scenarios:
  - Region I: Southwest and Central Coast
  - Region II: Wadden Sea Coast
  - Region III: Rhine and Meuse Rivers
  - Region IV: Lower River Courses
- Selection rationale:
  - Regions represent flood-prone areas based on different threats (sea, rivers, lakes) and where floods would cause the largest damage due to higher population and economic activity density/concentration.
  - Two coastal regions: Southwest and Central Coast; Wadden Sea Coast. A storm surge in the Straits of Dover can affect both the southwest region and the central coast; a more northerly storm surge can affect the Wadden Sea coast. These coastal regions are treated as independent because the likelihood of a flood occurring across the entire coastal zone simultaneously is low.
  - Two river regions: Rhine and Meuse Rivers; Lower River Courses. Rhine and Meuse floods can extend to neighboring countries (Germany and Belgium), as observed during the 2021 Limburg flood. Flood scenarios for Region III are enhanced by incorporating flood scenarios for Germany and Belgium near the Rhine and Meuse basins.

### 3.2 Flood Types
- Two flood types are considered for each region: Type A and Type B.
- Flood types classification (Table 1):
  - Type A: Flooding in unembanked areas
  - Type B: Breaches in primary flood defenses
  - Type C: Breaches in regional flood defenses
  - Type D: Flooding from bank overflow by regional water bodies
  - Type E: Water on streets due to extreme rainfall
- Focus rationale:
  - Analysis focuses on flood types A and B because they cannot be privately insured and are potentially the most damaging for the banking sector.
  - De Nederlandsche Bank (DNB) also focused on types A and B in their physical risk stress testing (Caloia and Jansen, 2021); results indicate flood type B are more damaging relative to flood type A, though DNB did not incorporate geolocational-specific scenarios or future climate conditions.
- EDO scenarios:
  - EDO scenarios are included as a separate flood type, categorically aligning with type B (breaches in flood defenses).
  - EDO scenarios are designed for contingency planning and can be independent of likelihood, possibly extending to 1-in-1,000,000 years or more.
  - Worst credible scenarios were defined using expert judgment on hydrodynamics (water level, wave height, duration) and possible number, locations, and size of breaches.

### 3.3 Climate Conditions and Flood Protection
- Climate conditions considered:
  - Current climate conditions.
  - Future climate conditions under the Dutch W+ scenario (KNMI’14), broadly aligning with RCP 8.5 (IPCC 5th Assessment Report).
- Note on climate scenarios:
  - KNMI’14 scenarios convert IPCC 5th AR global emissions to the Netherlands up to 2100.
  - KNMI’23 was published in October 2023 but water level/hydraulic load estimates based on KNMI’23 were not available at the time of analysis.
- Treatment by flood type:
  - Flood type A:
    - Water levels under higher return periods are considered as those under future climate conditions.
    - Kolen et al. (2022) find return periods of water levels in most water systems decrease by approximately a factor of 3 in 2050 W+ and by about a factor of 10 in 2100 W+ (i.e., exceedance probability increases by factors of 3 and 10 respectively).
    - Flood maps for 2050W+ and 2100W+ are generated by applying current flood depth maps with different return periods: 1-in-10, 1-in-100 and 1-in-1,000 years (higher return periods than 1-in-10,000 years are not available for current climate).
  - Flood type B:
    - Must account for both impact of future conditions on hydraulic loads and future reinforcement of flood defenses (breach-origin floods).
    - By legal mandate, primary flood defenses in the Netherlands will be reinforced to meet flood safety standards at the latest in 2050 to account for climate change and socio-economic developments.
    - Scenarios incorporate future climate conditions in 2050 with and without safety standards reinforcements.
    - Strength of flood defenses is expressed in terms of return periods (acceptable failure probability), varying by location and depending on impact of flooding and reinforcement costs.
- Scenarios for flood type B (Table 2):
  - B1: Failure Probability (Reinforcement) = Current Situation; Hydraulic Loads (Climate Conditions) = Current Situation; Description: Readily available flood water depth map
  - B2: Failure Probability (Reinforcement) = 2050 Safety Standard; Hydraulic Loads (Climate Conditions) = 2050 (W+); Description: Combined effects of reinforcement and climate changes
  - B3: Failure Probability (Reinforcement) = Current Situation; Hydraulic Loads (Climate Conditions) = 2050 (W+); Description: Impact of climate changes on current failure probability

### 3.4 Return Periods and Breach Numbers
- Return periods considered:
  - Flood type A: 1-in-10, 1-in-100, 1-in-1,000 and 1-in-10,000 years.
  - Flood type B: 1-in-100, 1-in-1,000 and 1-in-10,000 years.
  - These return periods are suggested by HKV and considered in the Netherlands for flood protection standards.
- Breach numbers by region and return period (Table 3):
  - Region I: Southwest and Central Coast — Return Period 100: 0; 1,000: 4; 10,000: 7
  - Region II: Wadden Sea Coast — Return Period 100: 0; 1,000: 4; 10,000: 7
  - Region III: Rhine and Meuse Rivers — Return Period 100: 1; 1,000: 3; 10,000: 4
  - Region IV: Lower River Courses — Return Period 100: 1; 1,000: 3; 10,000: 3
- Assumptions and notes:
  - For each region and return period, the maximum number of possible and realistic simultaneous breaches are specified based on Kolen and Nicolai (2023).
  - These maximum breach numbers are assumed consistent across the three flood type B cases (B1, B2, B3).
  - In case of a 1-in-100 years return period in the Southwest and central coast and Wadden sea coast, no breaches occur, resulting in no damages.
  - In B2 (2050 safety standard + 2050 W+), lower failure probabilities lead to different breach locations compared to B1; return periods are higher than B1 (adjusted to 2050 W+ conditions).
  - In B3 (current failure probability + 2050 W+), breach locations are the same as B1 but return periods are higher.
- Cross-border combined scenario:
  - In combined scenario involving the Netherlands, Germany and Belgium, only floods with a return period of 1-in-100 years are taken into account due to data limitations in Germany and Belgium.
  - Water depth maps for Germany and Belgium were obtained from private data vendor Jupiter Intelligence (detailed local expert data not available).

### 4.  Damages Estimation (methodology summary)
- Damage estimation framework:
  - The Deltares Standard Method 2017 (implemented in the Schade Slachtoffer Module, SSM) is employed to estimate direct flood damage and casualty.
  - SSM software contains granular data on real estate/objects at each geographical grid (geographical resolutions of 5m, 25m, 50m and 100m grids).
  - Source data:
    - Business and residences: Basic Registration of Addresses and Buildings (BAG) 2022 and buildings and residence objects (ESRI file geodatabase).
    - Infrastructure: Regional roads from National Road File (NWB-Wegen 2022) via national georegister Netherlands; railway data from Top10NL (2022).
  - SSM also considers special objects (vulnerable objects, national monuments, IED installations, WFD protected areas) — numbers/areas reported but no damage calculated for these.
  - The SSM software also calculates expected casualty using a mortality function with water depth, water flow rate, rate of ascent, inhabitant data as inputs.
- Direct damages focus:
  - Analysis computes total direct physical damages for each flood scenario (physical capital loss from direct contact with flood).
  - Indirect damages (e.g., business interruption) are excluded.
- Damage calculation formula:
  - Damage_s = sum_{i=1}^{N} alpha_{i,s} * n_{i,s} * S_i
    - alpha_{i,s}: damage factor of category i given a certain water depth
    - n_{i,s}: number of objects or m^2 in category i affected by floods
    - S_i: maximum damage per object or m^2 in category i
    - N: total number of categories
  - Damage factor alpha_i,s determined from damage functions varying across exposure categories and subcategories, calibrated for the Netherlands (damage factor increases from 0 to 1 with water depth).
  - Floods outside dikes have different damage functions/assumptions (e.g., structural adaptations such as no basement, stone floors) leading to adjustments in the outside dike method.
- Maximum direct damage per category (SSM2017 v4.1 values; Table 4 excerpts):
  - Business:
    - Meeting facilities: Maximum Direct damage (€/unit) = 194 m^2
    - Office: 1,607 m^2
    - Health services: 2,689 m^2
    - Industries: 1,420 m^2
    - Education facilities: 1,228 m^2
    - Sport facilities: 113 m^2
    - Retail and Commerce: 1,796 m^2
  - Residential:
    - Single family houses - Structure: 1,295 m^2
    - Single family houses - Furnishing: 81,985 obj.
    - Ground floor apartments - Structure: 1,295 m^2
    - Ground floor apartments – Furnishing: 81,985 obj.
    - First floor apartments – Structure: 1,295 m^2
    - First floor apartments – Furnishing: 81,985 obj.
    - Higher floor apartments - Structure: 1,295 m^2
    - Higher floor apartments – Furnishing: 81,985 obj.
  - Infrastructure:
    - Regional roads: 2,243 m
    - Motorways: 1,520 m
    - Other roads: 414 m
    - Railroads – electrified: 1,710 m
    - Railroads – unelectrified: 6,842 m
  - Other categories:
    - Agriculture: 2.36 m^2
    - Green house: 63.1 m^2
    - Recreation intensive: 17.22 m^2
    - Recreation extensive: 13.98 m^2
    - Urban area: 76 m^2
    - Airport: 185 m^2
    - Vehicle: 10,491 obj.
    - Pumping stations: 1,177,853 obj.
    - Waste/water treatment plants: 17,107,030 obj.
  - Note: Maximum direct damage in Table 4 is damage from floods within dikes only.
- Netherlands’ capital shock (damage rate) estimation:
  - Compute total direct physical damage under each scenario.
  - Capital shock = percentage of estimated direct physical damage to pre-damage total capital value.
  - Total capital value proxy:
    - Generate a hypothetical flood map with 10 meters of water depth, assuming the entire surface of the Netherlands is submerged.
    - Calculate damage from this hypothetical flood using SSM; interpret this damage amount as a proxy for total capital value in the Netherlands.

*Source: IMF Working Paper — chapter "3. Flood Scenarios" (wpiea2024197-print-pdf).*

### 5.  Floods in the Neighboring Countries

### 5.  Floods in the Neighboring Countries

### Methodology for estimating flood damages
- Damages for Belgium and Germany are estimated using the methodology developed in Fornino et al. (2024).
- Flood inputs:
  - Flood depths and fraction flooded data: Jupiter Intelligence for a 1-in-100 years return period under SSP5 RCP 8.5 scenario in 2050.
  - Damage functions for floods in Europe: Huizinga et al. (2017).
  - Gridded GDP: Murakami et al. (2021).
- Aggregate country-level damage rate of country c (Dc) computed as:
  - Dc = ∑ (di,c * GDPi,c / GDPc) (sum over i = 1 to n)
  - Where:
    - di,c: the damage rate for location i in country c, calculated as di,c = fraci,c × dffloods(depthi,c)
    - dffloods: damage function in Europe from Huizinga et al. (2017)
    - fraci,c: fraction of flooded area within location i from Jupiter Intelligence
    - depthi,c: flood depth in location i from Jupiter Intelligence
    - GDPi,c: gridded GDP in location i from Murakami et al. (2021)
    - GDPc: the total GDP of country c
- GDP exposure split into built-up and non-built-up areas using Copernicus Global Land Operations “Vegetation and Energy” (CGLOPS-1) for 2019 (Buchhorn et al. (2020)).
  - Built-up: residential, commercial, and industrial damage functions combined by equally weighting each function.
  - Non-built-up: agriculture and infrastructure damage functions applied.

### Selection of inundated locations and disaggregation
- Due to limited granular flood data in Germany and Belgium:
  - Subbasins along the Rhine and Meuse rivers are divided into different levels of granularity.
  - Flood depth and gridded GDP within each subbasin (level 9) are inputs to damage functions to calculate damage rate for each subbasin.
  - Damage rates across subbasins are obtained, but simultaneous flooding of all subbasins is considered unrealistic.
  - A damage rate is selected from the distribution across subbasins, considering size of damages from historical events.
- Note on "location i": refers to an area where both flood depths and gridded GDP data are available at a certain level of granularity (e.g., each subbasin treated as a location).

### Caveats and limitations
- The damage rate is calculated in terms of GDP losses, while the macro model uses a damage rate of capital stock as input to a non-linear production function.
  - Implication: GDP loss rate may differ from capital damage rate unless the production function is perfectly linear.
  - Potential consequence: underestimation of impacts on macro variables.
- Regional macro dynamics and regional distribution of banks’ loans are not captured by the country-level aggregation approach, due to lack of granular loan-level data and a dedicated regional model.

### Integration into macro-financial and banking stress tests
- Aggregated country-level flood damages (from Section 5 methodology) are used to calibrate macro-financial scenarios in the IMF Global Macro-financial Model (GFM) over three-year horizons.
- Three types of shocks (following Hallegatte et al. (2022) and Donk et al. (2023)):
  1. Direct destruction of physical capital: total damage rate aggregated at country level for the Netherlands, Belgium, and Germany serves as immediate direct shock to the capital stock.
  2. Impact on total factor productivity (TFP): calibrated at twice the total damage rate and assumed persistent.
  3. House prices shock: calibrated using ratio of direct damages for all residences relative to maximum damages for residences multiplied by the number of residences; adjusted for regional-to-national price elasticity. House price shock applied only to the Netherlands due to regional data limitations in Germany and Belgium.
- Banking sector stress test focuses on credit risk for six Dutch Systemically Important Institutions (SIs) and loan categories: mortgage, corporates, other retails, financial institutions, government, and qualifying loans.
  - Loans distinguished by country of exposure: the Netherlands, Germany, Belgium, UK, United States, Australia, and the rest of the world.
  - Banks’ data as of June 2023 from confidential regulatory reporting (COREP and FINREP); PD historical series from DNB.
- PD and LGD modelling:
  - PDs projected using historical relationships with macro variables via panel regression models.
  - For mortgage, other retail, qualifying revolving, and corporate portfolios: Panel Autoregressive Distributed Lag (ARDL) model with logit-transformed PDs (Equation (1)):
    - ln(PDi,t / (1−PDi,t)) = αi + λ ⋅ ln(PDi,t−1 / (1−PDi,t−1)) + ∑ βi,s P s=0 zt−s + ui,t
    - Explanatory zt include economic growth, interest rate, housing price growth, and real wage growth; fixed effect αi captures country-specific characteristics.
  - Government and financial institution PDs computed using a structural model (Equation (2)):
    - PDi,t = (Credit spreadi,t / (1 − Recovery Rate))
  - System-wide PDs transformed to bank-specific PDs by assuming constant differential of risk between aggregate banking system and individual bank (distance-to-default differences from 2022 held constant).
  - LGDs:
    - Secured portfolio LGDs derived using bank-specific LTV projections and other cost factors.
    - Unsecured portfolio LGDs modelled as a function of future PDs.
- IFRS 9 provisioning implemented:
  - Bank-specific transition matrices by sector estimated from historical loan movements and supervisory statistics.
  - Transition probabilities adjusted with scenario-conditional PDs using a “beta-linking” approach.
  - 12-month provision for stage 1 loans and lifelong provision for stage 2 and 3 loans computed. Write-off rate assumed to be zero.
- Analysis focuses on credit risk channel; other channels (interest rate, market risk) excluded, possibly underestimating total impacts.

### Damage magnitudes across scenarios (summary of results)
- Damages estimated for all 77 scenarios (Section 3); reported in Annex II as capital and house price shocks.
- Capital shocks:
  - Range from 0.001 percent of total capital under the A1 flood scenario with 1-in-10 years return period in regions I and II
  - Up to 0.912 percent under the B3 flood scenario with 1-in-10,000 years return period in region III
  - Local nature of floods limits overall damage to physical capital relative to country’s total capital stock.
- House price damages:
  - Range from 1.46 to 24.71 percent under the A1 flood scenario with 1-in-10 years return period in region I and III respectively.
- General observations across scenarios:
  - For the same 1-in-1000 year return period and same geographical region, damages for floods type A tend to be significantly smaller than floods type B.
  - Damages tend to be higher under future climate conditions relative to current.
  - Adaptation (e.g., dike reinforcements for flood type B) tends to reduce damages.
- Twelve scenarios selected from the 77 to cover all flood types, regions, climate conditions, and reinforcement at least once for stress test exercises. Using these 12 scenarios, four stress test exercises considered:
  1. Impact of climate change in the unembanked area (Region IV)
  2. Impact of climate change and reinforcement (adaptation) in the embanked area (Region II)
  3. Impact of extreme flood scenarios (EDOs) across all regions
  4. Impact of floods in the neighboring countries (Region III)
- Overall conclusion: credit losses from floods are modest for the Dutch banking sector, with significant heterogeneity across flood types and scenarios.

### Regional and scenario-specific impacts
- 7.1 Impact of climate change in the unembanked area (Region IV)
  - Region IV selected due to higher capital damages for the 1-in-1000 year return period.
  - Floods in Regions III and IV typically result from both storm surge and river floods, with longer durations and higher damages.
  - Under future climate scenarios, banks’ PDs for the mortgage portfolio rise to 2.30 and 2.36 by 2025, relative to 2.15 and 2.07 under current climate and the baseline scenarios (see Figure A-4 in Annex III).
  - The flood under current climate causes an additional 11 percent credit losses compared to the baseline scenario in 2023; magnitudes increase in future climate scenarios to 13.5 in the 2050(W+) and to 14.7 in the 2100(W+) scenarios.
  - In the 2100(W+) scenario, floods reduce the bank capital ratio by 0.09 percentage points in 2023 relative to the baseline.
  - No bank’s capital ratio falls below the capital requirement in all scenarios.
  - Capital and House Price Shock by Scenario (Return Period: 1,000 years):
    - A1 Current: Capital Shock 0.020; House Price Shock 1.26
    - A2 2050(W+): Capital Shock 0.032; House Price Shock 1.48
    - A3 2100(W+): Capital Shock 0.044; House Price Shock 1.70
  - Conclusion: Type A floods in these areas do not lead to significant banking sector capital depletion.

- 7.2 Impact of climate changes and reinforcement (adaptation) in the embanked area (Region II)
  - Transition comparisons:
    - B1 to B2 measures difference between current and future conditions under reinforcement.
    - B2 to B3 captures effect of lower failure probability due to dike reinforcement (keeping climate constant).
    - B1 to B3 measures climate change effect.
  - Example: A 1-in-10,000-year flood in Region II under current conditions generates 0.2 percent destruction in capital stocks.
    - Under 2050 conditions, damage rate expected to decrease by 0.138 percentage point thanks to dike reinforcements.
    - In B1 scenario, bank capital losses increase by 18.95 percent relative to baseline in 2023.
    - Climate change adds losses of 0.28 percentage points (B3 effect); lower defense failure probability in 2050 absorbs losses by 0.81 percentage points (adaptation effect).
    - Net result: capital loss rate lower by 0.53 percentage point in total under B2 scenario.
  - Capital and House Price Shock by Scenario (Return Period: 10,000 years):
    - B1 Current / Current failure probability: Capital Shock 0.200; House Price Shock 9.6
    - B2 2050(W+) / 2050 failure probability: Capital Shock 0.062; House Price Shock 10.5
    - B3 2050(W+) / Current failure probability: Capital Shock 0.241; House Price Shock 9.5
  - Conclusion: Reinforcement plan can substantially absorb capital losses from climate change.

- 7.3 Impact of extreme flood scenarios (EDOs)
  - EDOs assume simultaneous breaches of multiple dikes; coastal storm surge scenarios in Region I and II affect largest areas:
    - Region I inundated area: approximately 4,300 km2
    - Region II inundated area: approximately 4,600 km2
    - Together these add up to nearly 26 percent of total land area in the Netherlands.
  - Region I’s higher population density leads to larger GDP declines despite comparable inundated areas elsewhere.
  - Capital and House Price Shock by Scenario (EDO):
    - EDO-I: Capital Shock 5.4; House Price Shock 13.9
    - EDO-II: Capital Shock 2.4; House Price Shock 13.9
    - EDO-III: Capital Shock 1.9; House Price Shock 17.7
    - EDO-IV: Capital Shock 2.4; House Price Shock 17.4
  - In extreme flood scenarios, bank capital ratio drops by 0.3–0.6 percentage points relative to baseline in the first year, but remains above requirement.
  - Factors limiting capital ratio impact relative to GDP reductions:
    - Heterogeneity in impact size across banks.
    - PD model for the Netherlands suggests relatively small impact of GDP changes on PD trajectories.
    - Exclusion of other risk channels (interest and market risks) from analysis.
  - Sensitivity: A higher house price shock (applying regional loss to national house value without elasticity adjustment) adds 0.1 percentage point decline in bank capital ratio, amplifying first-year impacts.

*Source: IMF Working Paper — Chapter 5, "Floods in the Neighboring Countries" (from provided PDF).*

### 7.4 Impact of floods in the neighboring countries

### 7.4 Impact of floods in the neighboring countries

### Event description and observed damages
- In July 2021, heavy rains across Belgium, Germany, the Netherlands, and many other western European countries caused streams and rivers to overflow their banks in many locations. Some affected regions experienced rainfall "of this magnitude not seen in the last 1,000 years."
- The floods are estimated to have caused a minimum of 10 billion euros in total damage, with particularly severe damage to infrastructure in Belgium and Germany.
- According to the international disaster database (EM-DAT), the damage per GDP in Germany was approximately three times larger than that in Belgium.
- Damage rates (capital shocks) for Germany and Belgium used in the analysis are selected based on evidence from the 2021 flood event.

### Dutch banks’ exposures to neighboring countries
- Average of the total exposures of the Dutch banks in the sample:
  - 7.6% is in Germany.
  - 5.8% is in Belgium.
- Floods along Rhine and Meuse River area in Germany and Belgium have minimal spillover impacts to Dutch banks despite this exposure.

### Simulation results and quantified impacts
- Under the scenario for Region III and a 1-in-100 year return period, floods in neighboring countries increase banks’ capital loss rate by 0.06 percentage point relative to the scenario with floods only in the Netherlands.
- Figure 9 — Capital and House Price Shock by Scenario (Return Period: 100 years)
  - NLD: Capital Shock 0.124; House Price Shock 10.5
  - DEU: Capital Shock 0.041; House Price Shock -
  - BEL: Capital Shock 0.13; House Price Shock -
- Overall conclusion from the simulation: while floods in Germany and Belgium have negative impacts on banks’ capital in the first year, the impact is very small and not large enough to transmit additional credit risks to Dutch banks in the modeled scenarios.

### Data granularity and model caveats
- The flood scenarios for Germany and Belgium were based on less granular data than for the Netherlands.
- The analysis notes that adopting more granular flood and collateral data from Germany and Belgium would help refine the assessment of damage and spillover impacts on banks.
- Damage rates used for neighboring countries are selected from the 2021 event evidence; limitations in granularity and scope are acknowledged.

### Implications for risk assessment
- Limited spillover: Given the modeled exposures and selected damage rates, floods in neighboring Germany and Belgium add bank capital losses but with limited impacts on Dutch banks.
- Need for improved data: More detailed flood maps and geolocated collateral/exposure data from Germany and Belgium would improve precision of spillover estimates.
- Interaction with broader findings: These neighboring-country results fit within the study’s broader conclusion that the local nature of floods limits overall damage to the banking sector, but extreme scenarios and limited data at borrower level can lead to underestimation of risks.

*Source: wpiea2024197-print-pdf - 7.4 Impact of floods in the neighboring countries*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024197-print-pdf.pdf_