Counterfactual Analysis: A History of Volatility
Analysing the impact of historical European windstorms on current market exposure using the Eurowind™ historical catalogue isolates the direct impact of windstorm hazard (e.g., location of impact, intensity) and how it has varied over time. In other words, this approach removes the volatility due to increases and geographic shifts in insured building stock by using a constant, modern view of exposure across the European model domain (Figure 1).
Figure 3 shows the time series of expected event losses (blue circles) produced by the Eurowind™ 63-year historical catalogue. This analysis used the CoreLogic 2023 Insured Exposure Database (IED), which represents insured building, contents, and business-interruption (BI) values for all countries within the model domain (Figure 1). Events are shown by season. Storm Daria, for example, occurred in January 1990, i.e., in the 1989-1990 windstorm season. The green line represents the 5-year centred mean of annual maximum loss. Storm Daria leads to the associated peak in the mean around 1989. The light blue line is the 5-year centred mean of annual aggregated loss.
Figure 3: Losses due to the set of Eurowind™ historical windstorms ordered by season. Losses have been calculated using CoreLogic’s 2023 Industry Exposure Database (IED) and the RQE financial model. Source: CoreLogic, 2024
Both event losses and the average annual mean loss exhibit a large degree of variability over time, with a notable peak of activity in the 1990s. Moreover, Kyrill (2007) is the largest single event loss in the last 16 years (€9.5Bn in today’s exposure levels), and there has been a significant decrease in both annual maximum and mean losses since then.
Several of the most damaging windstorms occurred within the 1990s. The 1989-1990 season leads as the largest annual loss, with storms Daria, Vivian and Wiebke contributing to the year-end total. The December 1999 storms Lothar, Martin and Anatol, lead to 1999-2000 being the second costliest season.
Both Daria and Vivian are examples of a severe and large (i.e., in terms of width) type of extratropical cyclone which affect at least half a dozen major European countries. Daria has the largest event loss in the historical catalogue (Figure 3), with high maximum gusts spread over a wide area (Figure 4), causing significant damage across multiple countries.
Storm Vivian’s footprint, which occurred one month later and has an associated current-day loss of around 50% of Daria’s, also affected a large swath of locations.
Prior to the 1990s, Capella (1976) and The Great Storm of 1987 (also known as 87J) led to significant losses (Figure 3).
Figure 4: The Eurowind™ maximum gust footprint produced by storm Daria (1990). Source: CoreLogic, 2024
How Location Impacts Insured Losses
European windstorm insured loss depends upon both the strength and spatial extent of the gusts of a storm footprint as well as the location of the footprint with respect to insured exposure. A major storm affecting an area of low insured exposure will not cause a significant loss to the market. This prompts the question: to what degree is loss volatility driven by storm location, rather than variation in the frequency and intensity of storms affecting Europe as a whole?
One way to quantify a windstorm’s intensity and compare one event against another is to use an index. The storm severity index (SSI) is a measure of a storm’s damage potential regardless of the underlying insured exposure and accounts for the maximum windspeeds, the area impacted by the most damaging winds and the time duration.
The annual maximum of storm severity index (SSI) calculated using the Eurowind™ historical footprints is shown by the bars in Figure 5 for the 63-year period. The SSI peaks in 1990 due to Storm Daria, and the maxima due to Capella (1976) and Kyrill (2007) are also clearly seen.
Figure 5: Maximum seasonal SSI over land in the Eurowind™ domain for latitudes less than 55N. Data from higher latitudes has been excluded as the historical record of surface observations is incomplete. Source: CoreLogic, 2024
Annual frequencies were assigned to the more extreme events within the historical catalogue using climatological simulations. According to the Eurowind™ Analytical Model, the hazard return periods of Daria, Vivian and Capella are higher than the empirical estimate of 1-in-63 years and therefore are regarded as outliers in the time series shown. The impact of the 1975-1976 and 1989-1990 seasons on the 5-year centred mean is shown by the red line, whilst the blue line shows the effect of excluding these years from the running mean.
The latter displays a small decreasing trend in SSI over the 63-year period. However, the historical record of damage potential of European windstorms is dominated by natural variability, with seasons 2017-2018 and 2019-2020 reaching values not attained since Kyrill (2007) and, before that, the 1999-2000 and 1993-1994 seasons.
Counterfactual Analysis using Eurowind™ Perturbations
In this section, the time period spanning from the 2006-2007 season, when Storm Kyrill occurred, to season 2021-2022 was used in the first of three counterfactual analyses. Here, the goal is to highlight the effectiveness of using a perturbation scheme to create realistic stochastic windstorm events. By considering the stochastic families of historical events that occurred during the 16 years after Kyrill, it is possible to determine the counterfactual probability of an event leading to loss equivalent to that of Kyrill.
Figure 6 shows key historical event losses (grey circles) over the period from 2006-2007 to 2021-2022, normalised to the Kyrill loss.
Figure 6: Historical and counterfactual (IED) losses of events spanning seasons 2006-2007 to 2021-2022. Source: CoreLogic, 2024.
The largest of the most recent losses was due to Friederike 2018 (Figure 7), which caused approximately 30% of Kyrill’s loss.
Figure 7: Maximum gust (in meters per second) footprint of storm Friederike (2018). Source: CoreLogic, 2024
Also shown for each event in Figure 6 are the 50th, 75th and 95th percentile losses of the event families (red, green and blue, respectively). The 95th percentile loss of the Friederike 2018 family is equal to the Kyrill historical loss. This magnitude of loss is obtained by perturbing the parent Friederike event southwards by 200km so that the footprint lies over an area of higher insured exposure. The wind speeds were also increased. Similar explanations apply to the other historical events shown. Conversely, Kyrill’s actual loss lies above the 75th percentile, implying that there was a large counterfactual probability of smaller losses occurring.
Even the slightest changes in historical European windstorm characteristics such as location and track could have resulted in much different events. The recent historical insured losses inflicted upon (re)insurers covering Europe have been manageable, but small changes to the storm’s track could have resulted in an industry-altering loss event.