Impact of floods on LTV of Dutch mortgage portfolios
One year ago, heavy rainfall in the Netherlands caused high water levels in the Maas and its affluents. In Limburg this led to large floods with a total of 2300 properties damaged, of which 700 badly. Due to the changing climate, the probability and intensity of heavy rainfalls both increased. Heavy rainfalls can cause rivers to overflow and since the Netherlands is a delta - large European rivers flow into the sea through our country - this is a serious threat. Next to this, one-third of the Netherlands’ surface is below sea level. The sea level rises because of global warming, which makes the probability of floods in the Netherlands even higher.
Impact for banks
A decrease in the market valuation of the properties in a flooded area may happen due to two reasons: the flood can cause significant physical damage to the property, and the demand for the property decreases when potential buyers start considering the area as too risky to live in.
A decrease in the price of a property may lead to an increase in an important risk driver for banks: the client’s loan-to-value (LTV), which is the ratio between the outstanding amount of the mortgage (risk) and the value of the collateral (risk mitigant). Having a high LTV is undesired from a risk management perspective since it increases the loss expectations of the bank in case of a default. Not incorporating flood risk in the credit risk analysis for mortgage portfolios can thus lead to underestimation of the actual risk, especially when many properties are located in high-risk areas.
Next to that, a stress test of the European Central Bank (ECB)[1] found that banks in the Netherlands are not including climate risk sufficiently in their internal models and stress tests. Because of that, the ECB stated in their regulations that from 2022 onwards banks need to report information about climate risk and manage the climate risk data gaps.
RiskQuest’s analysis
Rijkswaterstaat developed an (open-source) tool containing data on the consequences of climate change in the Netherlands[2]. From this source, we use data on the probabilities of floods with different depths. Aggregating this data on a postal code level resulted in the following plots.
RiskQuest has created a dashboard to analyze the potential effects of a flood on the LTV. The aim of this dashboard is simple; provide insights to banks on how a flood can affect the LTVs in their mortgage portfolios, and thereby come to a better understanding of the potential climate impact on credit risk. The tool simulates a mortgage portfolio using adjustable inputs such as location and urbanity. Ultimately the simulated portfolio data is combined with climate risk data to re-calculate LTVs after a property value drop. The analysis done by the tool is visualized in the flow chart below.
The various inputs in the dashboard determine the LTVs of the individual loans that make up the simulated portfolio. For instance, property values are based on average selling prices per postal code area. Since the average LTV per postal code is not public data, we combine data on the average LTV per age category[3] with the age distribution in each postal code area[4] to construct the weighted average LTV of the simulated portfolio. Alternatively, the user may override the simulation and manually select the average LTV of the portfolio.
After the mortgage portfolio is simulated, the tool evaluates the change in the weighted average LTV after incorporation of a flood scenario. The change is based on a loss function – which estimates the drop in collateral value after a flood – obtained from research by ABN AMRO[5]. Since this loss function is dependent on the flood depth, a flood depth of either 50 or 200 centimeters can be chosen in the tool.
Extreme cases
From the geographical plots in Figure 1, one can derive that many parts of the country have a nonzero probability of flooding. One of the high-risk areas in the Netherlands is the province of Flevoland, which is on average 5 meters below sea level. Simulating a mortgage portfolio with properties located in Flevoland results in an increase of the average LTV from 0.64 to 0.94 (+47%) after a flood of 200 centimeters. This results in over 20% of the clients in the portfolio getting an LTV higher than 1, which is an important marker since the risk becomes significantly higher once the outstanding is larger than the property value.
The high-risk areas around the rivers in the provinces Gelderland, Noord-Brabant, and Limburg is another alarming example. The simulated mortgage portfolio in the postal code areas, specifically with the highest risks in these provinces, shows an increase in the average LTV from 0.58 to 0.85 (+47%) after a flood of 200 centimeters. Since the probabilities of floods in the area of this simulated portfolio are higher than once every 30 years, there are serious risks in this area. That being said, there are large differences to be found between provinces. For instance, in a low-risk province like Drenthe, the LTV is expected to only rise from 0.57 to 0.61 (+7%) after a flood of 200 centimeters. The popular saying in real estate also applies here; location matters!
To conclude
Data availability is often mentioned as the blocking issue for quantitative climate risk analyses. Thanks to the open data-source of Rijkswaterstaat this is less of an issue for flood risk in the Netherlands. By developing our dashboard, a quick insight into the potential change in LTV after a flood scenario can be provided with only a few assumptions. First results confirm the importance of flood risk on LTV and how this is largely influenced by the portfolio's location and concentration. We understand that these first results are extreme cases and banks might also have mortgages in less risky areas, therefore the inputs in the dashboard can be adjusted such that the simulation replicates a more realistic mortgage portfolio of a particular bank. Ideally, the analyses would be tailored to a non-simulated actual Dutch mortgage portfolio to get the most relevant insights.
On top of that, more accurate forecasts can be made once more data about floods and their consequences on property prices becomes available . The analysis is expandable to different climate risk hazards (e.g. wildfires, droughts) and different portfolios (e.g. corporates, agriculture loans). For example, drought can also have an impact on the flood risk since the dike walls may become unstable due to lack of rain. RiskQuest will keep its eyes on the innovations regarding climate risk and new insights in the data to improve the quantitative tools.
[1] ECB 2022 climate risk stress test, can be accessed via the link Climate Stress Test report by Banking supervision.
[2] The mentioned tool can be accessed via www.klimaateffectatlas.nl.
[3] This data is obtained from De Nederlandse Economie 2011 (chapter 9, figure 9.7 on page 215) by CBS. This paper can be accessed via https://www.cbs.nl/nl-nl/publicatie/2012/36/de-nederlandse-economie-2011.
[4] This data is obtained from CBS via https://www.cbs.nl/nl-nl/dossier/nederland-regionaal/geografische-data/gegevens-per-postcode.
[5] The research mentioned is the paper Economische impact van toekomstige overstromingen in Nederland from ABN AMRO (7 december 2020).