When a black swan is not coming alone – a lesson learnt during my recent trip to Japan
How likely is it that a twice in a century typhoon falls together with a magnitude five earthquake? Until recently, I would have said very unlikely. Yet, it is exactly what happened while I was in Tokyo two weeks ago.
Typhoon Hagibis was the most destructive typhoon hitting the greater Tokyo region since 1958. For me personally, the disaster caused apart from some minor inconveniences no harm and I feel sorry for the many victims.
The heavy rainfalls accompanied by Typhoon Hagibis resulted in considerable damage. But luckily, the almost concurrent earthquake did not create much harm and did also not magnify the impact of the typhoon. The latter is not self-evident, as earthquakes are among other things feared for triggering landslides. Together with the heavy rainfalls in the hours before, this could have ended much worse. And this made me to rethink certain aspects of risk management.
Prior to this experience, I would clearly have argued against evaluating two orthogonal stress scenarios jointly. For two rare and independent events like a century typhoon and a strong earthquake, I would have intuitively considered the risk of both materializing simultaneously as unrealistically low. However, the fact that I just witnessed such a joint event told me that something could be wrong with my first intuition. And this was indeed the case. While the frequency of an extraordinarily strong typhoon like Hagibis is only low, earthquakes of magnitude five are quite common in Japan and happen several times per year. Thus, when the rare event of a century typhoon materializes on a certain day there remains the same risk of experiencing a strong earthquake as on any other day. And because this risk is not low in Japan, the probability of observing both events jointly is not so much lower than the probability of a destructive typhoon to happen alone. Or in metaphorical terms, it is well possible that a black swan appears together with a more common friend.
Moreover, I realized that there is despite the independence of the two events a common link across them. This because the risk of landslides being triggered by an earthquake magnifies if soil is soaked. Thus, the heavy rainfalls associated with a typhoon make an earthquake on the same day much more dangerous than an earthquake that occurs on a normal day.
What are the consequences of this for stress testing? First, I see a need to complement macroeconomic stress tests with idiosyncratic risk components with moderate to high probability of materializing during the scenario time horizon, albeit there is no causal link to the scenario. Second, special attention needs to be given to risks which are likely to magnify if both events occur simultaneously.