Just as the events at the Fukushima nuclear plant following the Tohoku earthquake of 11 March 2011 pointed to new directions in nuclear plant safety assessment (see my earlier blogpost), so also the property/casualty losses following the quake point to logical next steps in earthquake CAT loss models.
The recent Swiss Re Report on Lessons from Recent Major Earthquakes highlighted a number of ways that CAT Models could improve their loss estimates for portfolios insured against earthquakes. Emphasizing first of all that 2011 set the record both for total economic losses from earthquakes ($ 226 B) and for insured claims ($ 47 B), it underlined that the Tohoku earthquake of 11 March 2011, with insured claims of $35 B, was the most expensive natural CAT of all time, not just among earthquakes, but all natural CATs.
Next, the report turned to perceived inadequacies in current generation CAT loss estimation models. The Swiss Re report pointed out that while most CAT models used by property/casualty underwriters appeared to have adequately modeled property losses following from ground shaking alone, they typically underestimated (if they modeled them at all) the losses resulting from secondary loss agents – (i) the tsunami(s) following, (ii) the seismic aftershocks, (iii) soil liquefaction (iv) business interruption (BI) and (v) contingent business interruption (CBI). Losses due to fires following earthquakes, another secondary loss agent, however, appear to be well modeled.
Tsunamis Where CAT modelers had considered tsunamis following quakes, the height, consequent inland penetration, and damaging force of the tsunami were underestimated. This was true both in the Tohoku quake in Japan, as well as with the recent earthquakes in Chile.
Seismic Aftershocks A major seismic event is often followed by aftershocks for a considerable period afterward. In some cases, a single aftershock can be more damaging than the original event; and very often the cumulative impact of the aftershocks is greater than that of the original event. Such cumulative effects and the clustering of smaller magnitude events following the original quake are important contributions to total losses, and need to be modeled more carefully.
Soil Liquefaction This is a phenomenon where, after an earthquake, the soil loses its normal resistance to plastic deformation, and begins to flow like a fluid with a temporal and spatially variable viscosity. This was observed both in the aftermath of the Tohoku quake and in the recent Christchurch quakes in New Zealand, although in the Tohoku quake the tsunami damage far overwhelmed damage from soil liquefaction. As a secondary loss agent, soil liquefaction impacts total property replacement costs in the following ways: by damage from subterranean flooding, costs of land restoration, and in the case of large structures, damage from differential settlement (caused by spatial viscosity variations in the liquefied soil). In geospatial modeling of soil liquefaction potential, important factors to consider include the existence of a shallow ground water table; properties built on reclaimed land, near a river bank, or on poorly consolidated sandy soils that are most prone to liquefaction. Many of these risk factors are easily satisfied in urban areas where large commercial properties are usually built.
Business Interruption (BI) losses are usually underestimated by models, because they underestimate the time period over which production facilities could remain damaged; and Contingent Business Interruption (CBI) losses are usually underestimated by models because they capture insufficiently well supply chain dependencies, location, and geographic factors.
These considerations point to logical next steps for Earthquake CAT loss estimation model developers to undertake as improvements in their models. While the Tohoku earthquake and the tsunami that followed was the most devastating CAT in history, it is worth remarking that, from the modelers' perspective, it was also geophysically the most well-recorded CAT of all time. Following the devastating Kobe quake of 1995, a large, dense, high-bandwidth, high-connectivity and high-sensitivity network of ground motion sensors was set up. This network spanned the area which was impacted by the Tohoku quake and tsunami, both on the ground and in the ocean, thus generating significant amount of data relative to the spatial distribution and magnitude of ground shaking intensity following the quake,both above ground as well as on the ocean-bed. In addition, following the Sumatra-Andaman earthquake-tsunami of 2004, a network of tsunami sensors and deep-water pressure gauges was also set up. The result is that a rich dataset is now available, which modelers can use to calibrate their ground motion loss estimate modules, and the correlation between earthquake moment magnitude with the size of the tsunami it can generate. However, on a larger scale, the lesson of the Tohoku tsunami quake is likely still to be that the historical record of tsunamis following earthquakes is as yet too sparse to enable confidence about the correlation between seismic moment magnitude and the temporal return periods. Nevertheless, CAT modelers can still proceed to remove the underestimation bias in the loss potential from secondary loss agents that was seen following recent earthquakes.
