Swiss Re: A wake-up call on typhoon risks in JapanOctober 29 2019
A wake-up call on typhoon risks in Japan
Weather related events continue to dominate natural catastrophe losses globally. In Asia, Japan was particularly hit with multiple catastrophes including a series of typhoons in 2018. Typhoon Jebi, which hit the Osaka region in September 2018, was Japan’s costliest typhoon since Typhoon Mireille (1991), with insured losses estimated at about USD 12bn. That’s more than double of the USD 6bn that catastrophe models have initially estimated. Also, flooding from Typhoon Prapiroon (2018) in western Japan was a reminder of the potential severity of flood risk in Japan as experienced in history with Typhoon Kathleen (1947) and Typhoon Ida (1958). This year, Typhoons Hagibis (2019) and Faxai (2019), whose impact is currently being assessed, further underscore the high vulnerability of urban regions to both typhoon wind and flood risks despite structural mitigations.
Jebi was a wake-up call for the industry to review the adequacy of underwriting for this peak peril in Japan. It is a reminder that as an industry, we need to constantly monitor the various physical and socio-economic factors that can influence losses for large events, which may not be well-addressed by current cat modelling technology.
Large events such as Typhoon Jebi continue to surprise the industry
The storm’s adverse loss developments surprised the industry in 2018, but it is not alone. This phenomenon has been repeatedly observed globally resulting in optimistic bias for large events. Secondary effects of primary perils, changes in land use with urbanization, climate change, demand surge, social inflation are not well captured in catastrophe models. For example, failure of the levees in New Orleans during Hurricane Katrina (2005) resulted in flooding of over 80% of the city. More recently Hurricane Irma (2017) surprised the industry with a large loss creep due to assignment of benefits (AOB) lawsuits, amongst other factors. While the surprises come in different flavours each time, the outcome is the same – a heavy underestimation of losses for large events.
In the case of Typhoon Jebi, a mix of factors potentially led to the initial loss underestimation. Figure 1 hypothesize the breakdown of potential loss drivers based on available data and anecdotal information.
One possible factor was the low wind speeds estimated just after the event when the uncertainty was high. A recent Kyoto University simulation showed that Osaka city’s urban landscape with multiple tall buildings contributed to the build-up of eddies of wind within the city. This is, however, not a new phenomenon when typhoons interact with tall buildings in an urban centre – Hurricane Ike (2008) in Houston showed a similar experience, and it will happen again. Nevertheless, increased wind speeds alone could not explain the complete gap between modelled and actual losses.
Typhoon Jebi also took place at a time when Japan had yet to recover from the impact of earlier catastrophes. A magnitude-6.1earthquake hit Osaka in June and in early July, torrential rains ravaged western Japan. These events resulted in stress on resources and impacted claims settlement practices. In addition, Osaka region had not experienced high wind speeds for many decades. Changes in construction practices and customers’ behaviour towards insurance might have taken place. While these factors are hard to quantify, it would be imprudent to ignore these factors in underwriting.
déjà vu – A recurring occurrence in history
With large-loss weather events like Jebi, Faxai, Hagabis, and Prapiroon, it is tempting to point to climate change as a driving risk factor. However, a look into Japan’s typhoon history tells a rather different story. The irony is that Typhoon Jebi is not an unexpected event. Over the last century, there have been at least four events of similar scale in Japan (Muroto 1934, Vera 1959, Nancy 1961, Jebi 2018). In fact, Typhoon Muroto, Nancy, and Jebi took very similar tracks impacting the Osaka region.
Figure 2 shows the modelled loss estimates for the major events in the past century. Where available, trended original insured loss data (1991 and later) were used, and for earlier events, physical hazard characteristics have been matched to events of today, and remodelled in our risk assessment model. Assuming an 85 or 100-year observation window, at least 4 events have exceeded the JPY 1000bn level – making this a 1:20 or 1:25 years observation.
Figure 2: Japan typhoon loss experience over the last 85 to 100 years
Large gap between typhoon risk view in market and loss experience
Third, it shows the results of Figure 2, i.e. an experience analysis reviewing the past 85-100 years of typhoon history in Japan, focusing on three market loss levels and the observed exceedance over that period. While the analysis technique contains some uncertainties of past events and their absolute loss levels, the exceedance frequencies at JPY 500bn, JPY 750bn and JPY 1000bn levels derived from experience are reasonably well-defined.
Despite market perceptions that such large-scale events are a 1-in-50 year occurrence or longer, historical data shows a much shorter 25 to 30-year period. Comparing the loss experience at various loss levels with the risk view (lower shaded curve range) used in trading risk in the market, the figure shows a material gap between model risk views used in pricing Japan typhoon risk and loss experience.
The figure is a simplistic comparison and the conclusion may be more complex. Based on limited available information about vendor models, it appears that top-down risk views of vendor models are in better alignment with the historical loss experience than risk view used in pricing Japan risk, where a JPY 1000bn loss level sits at 45+ years. Consequently, there is a larger disconnect between the top-down view of vendor models and risk view as used in the in the reinsurance market, based on vendor models only.
Figure 3: Loss experience versus indicative market risk views based on vendor models
The gap can be attributed to multiple reasons including different exposure data assumptions, model options, among others. Models are generally calibrated based on a view of market exposure and loss benchmarks. As an example, if high replacement values in market exposure relative to cedant’s data are used for a top-down model calibration, it can lead to an underestimation of vulnerability in the model so as to match same-loss benchmarks. When applied to cedant’s portfolio, this will result in lower losses. It is critical to reconcile these differences to ensure effective use of models. At the same time, the historical loss experience, in relative terms, also holds for any widely distributed portfolio across Japan. It provides an essential reality check of model performance and risk management. For example, an insured loss level of about 80% of Jebi, which is equivalent to a JPY 1000 bn market loss level, has occurred at least 4 times in the past 85 to100 years, indicating a 20 to 25-year return period.
An adequate risk view is a key enabler for insuring resilient societies
One in three natural disasters in the world takes place in Asia, and they can literally bring societies to their knees. Asia, a region with large catastrophe protection gap, provides us an important opportunity. Underwriting catastrophe business sustainably means having an adequate risk view that accounts for long-term historical experience and on-going physical and socio-economic trends. More importantly, as an industry, this is a wake-up call to review our underwriting assumptions and develop more robust modelling tools, so we can provide the right protection at the right price, to ensure sustained resilience of our societies. While it is difficult to conclude on long-term trends for typhoon activity in Japan, the last few years of typhoon activity indicates that we might be in a high activity phase like 1960’s and 1980’s periods. Japan’s typhoon and flood history provides robust quantitative cornerstones for typhoon risk over a century. 2018/19 typhoon risk level is not a complete surprise, but rather a continuation of an active past which needs to be the basis of a sound and sustainable risk assessme