It’s windy when it’s wet why UK insurers may need to reassess their modelling assumptions

Giorgis Hadzilacos, Ryan Li, Paul Harrington, Shane Latchman, John Hillier, Richard Dixon, Charlie New, Alex Alabaster and Tanya Tsapko

The 2015/16 storms caused the most extreme flooding on record, with parts of the UK impacted by heavy precipitation and extreme wind over a four-month period. These extreme weather events occurred in quick succession, hindering relief efforts and accruing £1.3 billion in insured losses. Without adequate mitigation, such events may result in claims handling strain and capital risk for insurers. Recent research finds that above-average windstorm seasons are typically accompanied by above-average inland flooding. That raises a challenge for insurers: should they have adequate risk mitigation measures in place for periods that are both windy and wet? We argue that insurers need to reassess their model assumptions, especially as climate change might make wet years more frequent than in the past.

Introduction

UK insurers use models that simulate extreme weather events to inform their pricing, manage their accumulation of risk, and decide how much capital they need to operate from both an economic and – in many cases – regulatory viewpoint. But there are two fundamental challenges that arise when it comes to modelling. The first is ensuring insurers are including in their models all the material risks they are exposed to – for instance freeze, subsidence, flood and wind risks for UK properties. The second is to ensure that any likelihood of adverse scenarios occurring together (ie ‘correlate’) across those risks is appropriately depicted in the models – a recent example is when a pandemic event impacts not only healthcare but also property business interruption. When either risks – or their correlations – are under-represented in an insurer’s model, it could lead to under-capitalisation and therefore a weakened financial protection for policyholders. It is insurers’ responsibility to periodically reassess their model assumptions especially as both the underlying risk and their exposures change over time. Our work indicates that flood and wind is one of many correlations that are under-represented in many insurers’ models.

Tendency for major windstorms to co-occur with inland floods during the winter season

But let’s take things one step at a time. Why look at inland flood and windstorms to start with? For retail property risk, windstorms pose the largest single natural peril risk. We have more understanding of the relationship between windstorm and coastal flooding but less so for the relationship with inland flood, a risk that is expected to change its patterns under a changing climate.

How do we know that inland flood and windstorm events occur close to each other (ie correlate) in the UK? By analysing historic time series, a recent scientific study undertaken in collaboration with the Universities of Loughborough and Reading, better quantified the correlation between wind and rain extremes, whilst also linking the precipitation to observed inland flood losses. Chart 1 illustrates how different parts of the UK experience this correlation based on historical information: the closer the correlation figure is to 100%, the higher the confidence that when inland flood events occur then windstorms will have also arisen in the same season. What this means in practice is that above-average windstorm seasons are more often accompanied by above-average floods – and vice versa.

Chart 1: Map of Europe, adapted from our recent scientific study, showing the intra-seasonal winter-time dependency between wind hazard and precipitation, which is linked to inland flood losses

(a): Map of Europe, adapted from our recent scientific study, showing the intra-seasonal winter-time dependency between wind hazard and precipitation, which is linked to inland flood losses. The darker shades of red show a higher dependency (higher seasonal wind hazard coinciding with a higher seasonal rainfall and a lower seasonal wind hazard coinciding with a lower seasonal rainfall). The white areas with the open circles show where the results are not statistically significant. The Pearson correlation coefficient (r) is shown at selected sites (W, E and C). Overall, our work indicates correlations in UK typically range between 30% and 70%.

(b) and (c): Scatter plots for site W: (West Scotland) and site C: (Central England) showing the coincidence of above-average seasonal wind hazard and above average seasonal rainfall (at a local level) being evident at both sites but more likely in West Scotland than in Central England.

How does the windstorm – inland flood correlation impact insured losses?

Identifying that major windstorm events tend to co-occur with inland floods within a season is one thing. Quantifying the potential financial impact to an insurance portfolio is another. To investigate the weather losses, we used a commercial catastrophe model, a type of software used by insurers to quantify the potential losses to their portfolios. Most catastrophe (or ‘cat’) models assume windstorm losses to be independent of inland flood losses and it is insurers’ task to combine those – along with other perils in their portfolios. If two perils are correlated, then the aggregated risk is raised and adjustments to their models may be required.

Chart 2 illustrates the impact of implementing general rank correlations between UK windstorm and inland flood using a ‘what if’ analysis of a more moderate range of 20% and 40%. The identified correlation could impact insured losses to a UK market insurance portfolio compared to their being considered to be independent.

For more extreme correlation strengths this study found the impact to the Aggregate Exceedance Probability (AEP) curve to be an increase of up to 6% before reinsurance, and 10% net of reinsurance. In other words, the increase in assumed correlation between wind and inland flood is pushing up the size of both the gross and net industry losses, whereas the exceedance probability (AEP) is fixed in each case at 1-in-200. Nevertheless, there are several caveats to these analyses such as:

  • approximations used in implementing hazard-based correlations in the catastrophe model (eg using precipitation as a proxy for inland flood);
  • correlations derived from research using winter-only analyses potentially overestimating the correlation factors applied to year-round losses;
  • insufficient historic time series to assess the nature of the correlations for more extreme return periods (ie events rarer than 1 in 50 years);
  • use of different correlation methods can significantly alter the impact to gross and net losses. For instance we found that the use of T-Copula instead of general rank correlation can double the impact on net 1 in 200 year AEP;
  • the impact is dependent on the correlation assumed which is a function of the geographic distribution of the portfolio; and
  • assumptions on how insurers can recover from subsequent weather events can materially impact the net loss assumptions (ie whether reinsurance is assumed to protect against a number of catastrophic events).

The indicative results are presented before (gross) and after (net) reinsurance protection is applied, assuming a typical reinsurance protection. The results are dependent on the reinsurance programme assumed, the magnitude of the interperil correlation, and the tail-behaviour (eg copula) chosen. What it is possible to securely infer, however, is the broad insight that there is a potential for the identified correlation to be material to modelled losses. Are they material enough to impact insurers’ capitalisation or risk management framework?

Chart 2: Indicative impact on a 1-in-200 year return period for gross and net AEP using correlation factors of 20% and 40%

Note: AEP stands for Aggregate Exceedance Probability and represents the probability of experiencing total annual losses of a particular amount or greater. For further details refer to the LMA guidance.

How does the windstorm – inland flood correlation impact insurers’ capital?

Insurers’ internal models are complex as they depict the range of potential risks that an insurer is exposed to. Hence a change to a single model element may – or may not – impact the overall capital position. To answer the question of whether the windstorm-inland flood correlation can have a material impact on an insurer’s capitalisation, we undertook a stress-test exercise using an idealised insurer’s internal model. Table A demonstrates the potential impact the windstorm-inland flood correlation can have on an insurer’s capital under an extreme but plausible event (1-in-200 year return period) – also known as solvency capital requirement (SCR). We undertook this analysis using two different correlation assumptions for the perils in question.

Table A: Indicative impact to an internal model’s SCR from wind-flood correlation for an insurance portfolio exposed to UK property risk

Percentage change Percentage change
Impact to Model variables 20% correlation40% correlation
Cat AEP Net (as per Chart 2) +5%+11%
Premium risk +2%+4%
SCR +1%+2%

In this illustrative example, the correlation between windstorms and inland floods led to a capital impact in the low single digit percentages. That severity could be different depending on insurers’ individual characteristics such as the proportion of property catastrophe risk (eg versus their motor book), the diversification of the business (eg life exposures next to non-life exposures) as well as the relative severity of flood risk to windstorm risk. Table B illustrates how different characteristics could result in the assumed correlation producing a more material capital impact.

Table B: Sensitivity analysis on the indicative impact to an internal model’s SCR from wind-flood correlation to a large, retail, non-life solo, well diversified insurer, showing the potential range of SCR impacts that can arise from a single correlation assumption

Impact of 40% flood-wind correlation on SCR (Table A)
UK solo GI insurer well-diversified: base case2.4%
Non-cat risk reduces by 50% so cat becomes more dominant2.6%
Non-cat risk reduces by 80% so cat becomes more dominant3.0%
Other risks reduce by 50% so premium risk becomes more dominant3.9%
Other risks reduce by 80% so premium risk becomes more dominant4.9%
Non-cat risk reduces by 80%, other risks reduce by 80%7.6%

The assumptions that lead to the above results are indicative for a UK retail insurance portfolio; however, as each insurance portfolio is unique, the severity of the impact will vary depending on the individual portfolio characteristics.

This study has focused on insurers’ capital requirements that are based on plausible but extreme events (ie 1-in-200 year return period). However the findings are even more pertinent to insurers’ risk management frameworks. For instance, the increased likelihood of multiple smaller flood and windstorm events impacting insurers’ retained losses net of reinsurance protection could impact earnings volatility in the near-term more significantly than that implied in Table B (eg at the 1-in-5 or 1-in-10 year return period).

Conclusion

All models are by design a simplification of the real world and insurers need to decide carefully which aspects of the real world to simplify. UK property is exposed to weather risk but only a few insurers assume that the tendency for major windstorms to co-occur with inland floods during the winter season needs to be reflected within their model. This pilot study challenges this assumption, providing an initial indication that the correlation between windstorms and inland floods is underrepresented in insurers’ models. Our test case showed that the neglected correlation might plausibly result in a low single digit underestimation of insurers’ capital allowance. This is not alarming by itself but indicates that an aggregation of underrepresented correlations could raise risk management concerns – if not capital ones. Given these findings, insurers should reassess the materiality of assuming peril independence in their internal models especially as insurance exposures to flooding are increasing and the underlying atmospheric perils’ severity and frequency could be altering as the climate changes.


Giorgis Hadzilacos, Ryan Li and Paul Harrington work in the Bank’s Insurance Division, Shane Latchman works at AIR worldwide, John Hillier works at the University of Loughborough, Richard Dixon works at the University of Reading, Charlie New, Alex Alabaster and Tanya Tsapko work at Aon.

If you want to get in touch, please email us at bankunderground@bankofengland.co.uk or leave a comment below.

Comments will only appear once approved by a moderator, and are only published where a full name is supplied. Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.

One thought on “It’s windy when it’s wet why UK insurers may need to reassess their modelling assumptions

  1. What differs ‘ex ante’ or ‘post ante’ may vary according to axiomatic grounding.
    As a consequence – insurer CAT modelling may encounter uncertainty without recourse to relevant capital requirements that are also matched by up to the minute metrics.. As 1 Governor famously put it; “What we can measure, we can manage”. That at least is the principle, the remainder follows of course, or not at all.

Leave a Reply