What if Typhoon Jebi had been stronger and taken a different path?

Source(s): Swiss Reinsurance Company

By Vineet Kumar

An industry-wide review of Japan's typhoon history was triggered following the 2018 and 2019 seasons, which led to a firmer inclusion of events happening before official recording started. This allowed risk modellers and the industry to build a more robust perspective on Japan Typhoon occurrence at return periods of up to 50-70 years. At the same time, events such as Jebi and Hagibis can also be considered near-misses, where small and credible variations to their occurrence patterns would have led to very different, even more catastrophic outcomes. Addressing this in a more scientific and data-driven manner is the core domain of probabilistic cat modelling. Building on the data and learning from real observations, these models allow an extrapolation to a "what-if" scenario loss, using meteorological insights to derive credible occurrence probabilities for such scenarios.

With many urban and industrial clusters exposed to direct hits from complex and intense typhoons in a changing climate risk landscape, insured losses of between USD 25 to 30bn are not a far possibility in Japan. For instance, Typhoon Vera (1959) is used for regulatory solvency margin calculations in Japan. However, return periods between 200-250 years are typically considered by insurers for typhoon risk in Japan. It is thus important to have a robust view of tail risk beyond standard historical benchmarks.

Typhoon Jebi and Hagibis as wake-up calls

Typhoon Jebi hit the Osaka region of Japan in 2018 as the strongest tropical cyclone to hit the country in over 25 years with insured losses estimated over USD 12 billion. Other known key events hitting the Osaka region were the 1938 Muroto and 1961 Nancy. In 2019, Typhoons Faxai and Hagibis swept across the Greater Tokyo area, highlighting the vulnerability of urban regions and severity of flood risk in Japan. Jebi and Hagibis became wake-up calls for the insurance industry to reassess the level of typhoon risk confronting Japan. A re-evaluation of the long-term historical past illustrates that 2018 was, in fact, not so exceptional after all. When we look at typhoon activity throughout the 20th century we can see typhoons with the strength of Jebi or more reoccur at a probability of 1 in 25 to 35 years. Jebi and Hagibis are not the most damaging typhoons that hit the region. Many old events such as Typhoon Vera (1959) and Ida (1958) would result in higher losses if they had occurred today. Being exposed to typhoons in the most active basin in the world, a far higher loss potential exists in Japan with its huge concentration of assets.

Why history is not good enough

While a good understanding of low-to-moderate loss events are important in estimating earning loss potential for a company, a view of severe-to-very severe loss scenarios is used for capital adequacy measures including reinsurance purchases. Historically, Japan uses a reoccurrence of the 1959 Typhoon Vera as a regulatory benchmark for solvency margin calculations to evaluate the industry's capital adequacy. This is done in a scenario-based perspective with little relation to its occurrence frequency. Although historical references provide important information for risk management, probabilistic cat models can help by enforcing the inclusion of individual scenarios into a quantitative context, encompassing perspectives in both severity and probability of occurrence. Embedding both Typhoon climatology guided by history and meteorology, and a data-based insured value distribution, it provides the foundation for a sound  view of risks, for more frequently-occurring events while putting the tail loss risk into a quantitative perspective. Recently there are efforts towards using probabilistic approach in solvency margin calculations in Japan.

How bad can it get – and what we can learn from others

The potential of losses from an event that could happen in a region depends on several factors. Concentration and value of assets at risk, maximum potential intensity and complexity of the perils, as well as the response from the affected physical and socio-economic environment are some key factors. For example, Hurricane Katrina (2005) losses were the result of an intense storm with devastating storm surge flooding the city of New Orleans, causing over USD 86 bn insured losses (at 2020 value). These losses were materially amplified due to how the physical and social-economic environment responded, both before the event (e.g., adequacy of mitigation measures in place) and after the event (e.g., emergency response and insurance practice to settle complex wind/flood claims, post-event litigation).

In Japan, Typhoon Jebi highlighted the complexity when typhoons interact with both physical environment (e.g., wind speeds amplification in the downtown area) and the social environment (e.g., changing consumer behaviour), which in turn influenced the final losses. While massive flood protection systems successfully prevented major damage in the denser areas of Greater Tokyo from Typhoon Hagibis, the many breaches and overflowing rivers showed that a substantial part of the local flood risk is only partially mitigated. Regardless of their strength, structural measures can fail, leading to catastrophic consequences and an event does not necessarily need to be an intense for things to go wrong (e.g., Hurricane Sandy 2012 in New York).

"Hit or miss" remains an important factor influencing the outcome in terms of insured losses. There are different qualities to the "hit" – a slight, and fully credible variation in meteorological terms could have led to much more severe outcomes, not only for Typhoon Jebi, but also typhoons like Faxai. The model has helped us to shape a curve beyond a history of 50 to 70 years, where drivers including modifications to storm tracks, value concentrations, urban density with a scaling potential of typhoon losses  - as seen with Jebi – indicate that a worst-case storm has yet to be seen in history.

Figure 1 shows how the tail of a loss frequency curve (i.e., losses vs its exceedance probability) would look like for different regional risk profiles. With Japan's risk profile, it is expected to see losses continue to rise (beyond what had been experienced) with higher and higher potential losses before the curve flattens out. Swiss Re's model suggests a similar loss profile as the middle curve in Figure 1. The top loss curve would have made more sense if the strong structural mitigation measures around key urban regions were not in place in Japan.

Probabilistic models as a tool for tail loss risk view

A 50-to-100y typhoon experience helps to develop a reasonable risk view at low-to-moderate frequencies but it is not sufficient to extrapolate this information to get a tail risk view. Probabilistic models have been the core tools, controlled with scientific understanding of physical phenomenon (e.g., maximum cyclone intensity based on sea surface temperature), to develop a view about tail losses. In catastrophe models, regional climatology and physics of the phenomenon have been used to create a larger set of plausible events to develop a holistic view of risk. Figure 2 shows a set of plausible events in the Osaka region based on perturbations of Typhoon Jebi. These probabilistic events (or daughter events) have varying typhoon characteristics such as track geometry, forward speed,  intensity, and size. These events result in a range of losses both lower and higher than the original Jebi experience, depending on their characteristics and impacted area.

For example, one of the daughter events (highlighted in Figure 2) takes a path closer to Osaka and Kyoto regions (about 20 km right from the original event) and is a Category 4 typhoon relative to original Category 3 Typhoon Jebi. Category 4 and 5 typhoons and are very much plausible in the region as observed in the history (e.g., Typhoon Nancy (1961), Muroto (1938), Typhoon Vera (1959)). This event results in about the doubled losses of the original event and comes about 150-year return period in the loss frequency curve. As expected, the probabilistic event is less likely than original Typhoon Jebi (~1 in 30-year) but very much in the range where important risk management decisions need to be made. Further, perturbations of other past events like Typhoon Vera lead to even worse loss scenarios. A stronger Category 4 storm with a track closer to Nagoya region results in about double the original Vera losses which comes about 250y RP.

Recent Observations from Swiss Re's model update

Building on standard modelling principles, revaluation of the Japan Typhoon track record, and exploiting claims experience from 2018/19 seasons, Swiss Re has rebuilt its risk view for Japan Typhoon, with a focus to better understand the tail risk where historical reference is not sufficient.

Where a Jebi-type loss reaches a 30-year probability, depending on the portfolio, the model results put a double of Jebi-type loss at about 150-year probability, with a Vera-type reoccurrence in between (~70 year). This suggests that managing Japan Typhoon tail risk based on historical benchmarks is highly limited.


The industry's proven probabilistic model technique allows us to put a sound and principles-based approach around the probability of more severe scenarios. On the back of a few "what-if" scenarios, it becomes obvious that historical benchmarks such as Typhoon Vera (1959) provides little robustness, and certainly an overly optimistic picture of the possible extent as well as the probability of occurrence of significantly more severe events. Japan's risk profile along with model outputs suggest a loss potential that is much higher than Jebi and Vera, is not too far in the future.

Is this page useful?

Yes No Report an issue on this page

Thank you. If you have 2 minutes, we would benefit from additional feedback (link opens in a new window).

Please note: Content is displayed as last posted by a PreventionWeb community member or editor. The views expressed therein are not necessarily those of UNDRR, PreventionWeb, or its sponsors. See our terms of use