By Dr. Jayanta Guin
In his 2007 best-selling book, The Black Swan: The Impact of the Highly Improbable, author Nassim Nicholas Taleb holds that the course of human history is shaped by unforeseen events, both positive and negative, whose probabilities are underestimated because they are too difficult to comprehend. Such events are called black swans, a metaphor for a seemingly impossible event that might later prove to be possible. (All swans were assumed to be white until a European explorer traveling to Australia in 1697 noted the existence of black swans there.)
Taleb, who is credited with predicting the 2008 financial crisis, described severe flaws in the risk management practices of banking and trading institutions that exposed them to calamitous losses not contemplated in the mathematical models on which they relied. Black swan events also exist in the realm of insurance risk management. After two quiet decades of tropical cyclone activity in the United States, 1992’s Hurricane Andrew caused insured losses previously considered by many to be impossible, catapulting the catastrophe modeling industry into prominence.
What may be a black swan to society at large may have limited insurance impact; likewise, some events that cause catastrophic losses may not seem extreme from other perspectives.
Since then, each successive large and unexpected catastrophe has prompted a reexamination of existing risk management practices. Because traditional statistical tools are not able to capture either the frequency or severity of black swan events, how should companies prepare for their impact?
Catastrophe events can be classified as known-knowns, known-unknowns, or unknown-unknowns. Those events for which there is abundant data and historical precedence are considered known-knowns, and they can typically be accounted for using past experience.
Known-unknown events are rarer and have more severe repercussions. Companies employ catastrophe models to prepare for these events. While there is inherent uncertainty associated with known-unknowns, a robust model uses what has occurred in the past to infer what is possible in the future. For example, though it is not known when or exactly where the next inevitable major earthquake will strike in California, catastrophe models can account for the probabilities associated with a full spectrum of loss outcomes.
For even more severe events, knowledge deteriorates precipitously with decreasing probability of occurrence. Black swan events belong to this last category, the unknown-unknowns. Their probability and severity are not possible to estimate with any degree of accuracy because they are often unimaginable until they actually occur. However, it is important to note that perspective matters. The number of fatalities caused by Hurricane Katrina was a surprise to most people, but the economic losses were within expectations for a major hurricane. And while the magnitude 9 Tohoku earthquake and tsunami in 2011 caught most seismologists by surprise, the level of insured loss fell well within the range to which prudent executives manage. What may be a black swan to society at large may have limited insurance impact; likewise, some events that cause catastrophic losses may not seem extreme from other perspectives.
Black swan events are unexpected either from an intensity perspective (they are not thought to be physically plausible) or a loss perspective (the damage inflicted was not thought possible even if the physical event itself had been contemplated). Knowledge of the underlying physical processes at work will always remain imperfect, so there is significant uncertainty as to what is truly an extreme scenario. For example, there is significant uncertainty about the physical processes surrounding climate change. How it will affect the occurrence of hurricanes, severe thunderstorms, and other hazards represents an even higher order of uncertainty. Climate change can surprise us in many ways.
Interactions between different physical processes — each of which may be unexceptional individually — can also lead to unexpected results. Hurricane Katrina, for example, triggered the catastrophic failure of the levee system in New Orleans. The Christchurch, New Zealand, earthquakes of 2010 and 2011 caused unexpectedly severe liquefaction. The 1923 Great Kanto earthquake, the deadliest in Japan’s history, struck almost concurrently with a typhoon that fanned the fires spawned by the earthquake into vast conflagrations.
In terms of losses, the value of the assets themselves constrains direct damage to physical assets, but the indirect effects can be near limitless. Often, the exposure for business interruption (BI) coverage, particularly contingent BI, is not well understood by the insurer or the insured, and losses can result from various feedback loops. The 2011 floods in Thailand have caused an estimated 15 billion USD in insured losses. The automobile and hard-disk industries were particularly hard hit as manufacturing output came to a standstill at hundreds of inundated, inaccessible, or powerless factories, revealing the extreme vulnerability of global supply chains to natural disasters.
Particular political and regulatory environments can inflate losses. For example, documentation since the 1920s has shown asbestos to be harmful, but not until a regulatory regime that was sympathetic to the public in the 1980s and ’90s did widespread litigation lead to extensive losses for the insurance industry.
Decisions based on incorrect data and faulty assumptions can lead to ineffective risk management practices or a false sense of security, setting the stage for events like the 2008 financial crisis.
Economic and social factors are also sometimes at play. A long lull in hurricane activity during the 1970s and 1980s led to complacency in disaster preparedness. That resulted in less stringent building standards and a relaxed attitude toward building maintenance that coincided with a construction boom — all of which contributed to the high losses caused by Hurricane Hugo in 1989 and Hurricane Andrew in 1992.
Finally, it should be noted that the misuse of risk management models can lead to black swan events. Decisions based on incorrect data and faulty assumptions can lead to ineffective risk management practices or a false sense of security, setting the stage for events like the 2008 financial crisis. Further, incorrect or misguided interpretation of model results — for example, focusing on one particular loss metric or ignoring the tail of a loss distribution — can leave companies ill-prepared for plausible loss scenarios.
Catastrophe models use statistical methods, physical models, and scientific and engineering expertise to provide a wide range of potential scenarios of what might be experienced in the future. Because those models are probabilistic, black swan events present a particular challenge. A Category 5 hurricane hitting the Northeast is extremely unlikely but perhaps not entirely unimaginable. Should it be assigned a probability of 0.0001 percent? 0.00001 percent? No one can say.
Still, there is much that can be done. AIR Worldwide is developing sets of physically possible extreme loss scenarios that can be considered deterministically using its next-generation software platform, TouchstoneTM. While it is not possible to create an exhaustive set of black swan events, and indeed the scenarios will likely fall closer to the known-unknown category, they will nevertheless be ones that stochastic modeling techniques are unlikely to capture. Risk managers can use these deterministic scenarios to perform a critical analysis of the peril and exposure to inform loss mitigation strategies.
Modeling companies should also explore other ways to address currently unmodeled sources of loss that can contribute to black swan situations. This includes secondary perils that may not yet be modeled (for example, the forthcoming update for the AIR Japan earthquake model will include a fully probabilistic tsunami model); currently unmodeled asset classes within existing models (like losses to insured infrastructure); and indirect and secondary losses, which can far exceed damage to physical assets.
It is important to remember that knowledge is imperfect and constantly evolving. No matter how sophisticated and detailed catastrophe models become, they will never encompass the entire realm of what is possible. Ultimately, hazard is unpredictable. While catastrophe models have become essential tools in sound risk management, it is prudent to be aware of their limitations and to be resilient to imperfect models.
Jayanta Guin, Ph.D., is senior vice president of research and modeling at catastrophe modeling firm AIR Worldwide. The author would like to acknowledge the help of Vineet Jain and Nan Ma for their contributions to this article.