Catastrophe Models: The Currency for Transferring Risk
July 25, 2012
Imagine that two pieces of U.S. currency, a $100 bill and a $1,000,000 bill, are offered to you. Which would you choose? Most people would take the $100 bill because they believe that it is actually worth $100—and they know that the $1,000,000 bill is only worth the paper it's printed on, if that.
The concept here is confidence, or trust. Money is worth something only to the extent that people have confidence that it has value and can be exchanged for goods and services. The same is true of model output, which are just numbers, perhaps with a lot of zeros too. Model output has become a currency for understanding and quantifying risk in insurance and reinsurance transactions, and it too has value only to the extent that the market has confidence in the numbers.
A 2008 paper by Harvard economists Robert Barros and Jose Ursua noted that since 1870, there have been 148 economic crises in which a country experienced a decline in GDP of at least 10%, and 87 crises in which consumption suffered a fall of 10% or more. The course of economic history is a roller coaster ride of booms and crashes that, despite their inevitability, continue to surprise people. The prominent early 20th century British economist John Maynard Keynes argued that much of what happens in economic systems is inherently unpredictable and beyond the realm of statistical analysis. In his Treatise on Probability Keynes wrote, "About these matters there is no scientific basis on which to form any calculable probability whatever. We simply do not know."
Much like financial shocks, natural disasters are defining moments in human history, unpredictable, and often unexpected. However, there is sound scientific and mathematical basis for managing natural disaster risk probabilistically, and it is for this reason that catastrophe modeling emerged 25 years ago. So in both financial and catastrophe risk management, we can understand the thinking that life is defined by moments; it is thus essential to prepare for the unexpected. Although models will never predict when or where the next big disaster will strike—or for any emerging disaster, what the final tally will be—they can give a sense of the range of possible losses and their likelihoods.
Catastrophe modeling has undoubtedly come a long way in its 25-year history, but uncertainty in models and in model output has been a persistent topic of discussion whenever we meet with insurance company executives. This is where the concepts of model output and currency converge, so I'd like to further explore some of the parallels in this article.
We live with constant uncertainty about the economy. Investors try to anticipate growth rates in different sectors; economists and policymakers are concerned with unemployment, inflation, interest rates, and trade imbalances; citizens worry about the value of their homes and retirement portfolios. One among these many concerns is uncertainty about the future value of money. For the purposes of this discussion, let's focus on currency volatility, which results from the change in the value of one country's currency against that of another.
For example, the Asian financial crisis of 1997 started with the collapse of the Thai baht after the government did not have enough foreign reserves to defend the fixed exchange rate. The crisis quickly cascaded to other Asian currencies in the region—including that of Indonesia, South Korea, the Philippines, and Malaysia—in an effect called financial contagion. These countries not only saw a precipitous decline in their currency values, but also experienced significant declines in their stock markets and GDPs, business failures, and increases in unemployment and civil unrest. Recovery took several years and required major bailouts and financial reforms.
Reductions in currency value are not necessarily bad. While the sovereign debt crisis in Europe pushed Greece toward a potentially chaotic exit from the Euro and threatened the future of Spain's once solid banking system, the story of Iceland is much brighter. Iceland too fell victim to the 2008 financial crisis, but being outside of the eurozone, the government has been able to implement unilateral monetary policy to stabilize the economy. The central bank devalued the Icelandic króna by approximately 50% against the Euro, driving up exports and tourism significantly. And while this reduced wages and decreased the import of expensive foreign goods, domestic consumption strengthened, turning Iceland's trade deficit into a surplus. While the devaluation has caused short-term hardships and not every sector has recovered, economic growth is resuming and employment has been steadily increasing.
Model output does not remain constant either and, like currency, there are two sides of the coin. Significant model changes can be disruptive to your business, so the modeler's decision to incorporate the latest published science and data should be weighed against stability in model output and informed by a determination of whether new findings represent a scientific consensus or simply the latest hypothesis.
For certain parts of the world where catastrophe experience is scarce or when an unexpected event occurs, the market actually seeks a "revaluation," or an update to the model following certain high impact catastrophes. For example, after the two earthquakes in Christchurch, New Zealand, in late 2010 and early 2011 that caused a combined insured loss of more than 15 billion USD, and after the March 11, 2011 Tohoku earthquake and tsunami in northern Japan, model users expect models to be updated to incorporate the newly learned lessons.
However, the market is not inclined to embrace changes equally to all models, nor should it. The historical record of U.S. hurricane activity, for example, is quite rich; a robust model should not require significant changes in response to a single event or single hurricane season. Model users are thus justified in expecting a consistent and stable view of risk for regions and perils where there is ample hazard, vulnerability and loss experience data.
Market Price, Technical Price
The value of money is not absolute, as inflation and deflation can change its purchasing power. Habsburg kings in 16th and 17th century Spain brought back so much silver from the New World to finance wars against the Dutch and English that the metal declined in value, driving up the price of goods—a previously unheard of economic phenomenon. Europe learned that money is worth only what someone else is willing to offer for it, which depends largely on market forces and the law of supply and demand.
There is a market price, too, for the transfer of catastrophe risk. What is actually paid for insurance and reinsurance is usually not the same as the price indicated by an actuarial formula, only one term of which is derived from catastrophe model output. Even assuming a perfect model, the price reflecting the "true risk" can differ from the market price for several reasons.
After a large event, market price can immediately increase, as was the case with reinsurance rates in New Zealand following the Christchurch earthquakes. Interestingly, prior to the event, AIR modeled losses for the region were considered by many to be too high, but AIR model results have since become well accepted.
Then there is the case of the model following the market. After the 2004 and 2005 hurricane seasons, the market believed that significantly increased rates were justified by a "new normal" in tropical cyclone activity. Accordingly, some model vendors released a "near-term" model with a 40% increase in hurricane frequency—an artificial inflation in response to sentiments in the industry—even though such a view was not borne out by more than 100 years of historical experience. AIR adopted a more measured approach, using objective data to develop an alternate, but still stable and robust, view of risk conditioned on warm sea surface temperatures as a supplement to the standard view based on all years of activity.
Companies that rely on model output should be aware of which way the feedback is flowing: is the model change reflective of objective new data or research, or is the model chasing the market?
Let's return to the concept of trust, which is the foundation of our financial system. For example, if you deposit money into a U.S. bank account, you assume that it will be there 20 years later, even if inflation will have diminished its value. Despite hundreds of bank insolvencies since 2008, deposits protected by the Federal Deposit Insurance Corporation (FDIC) are guaranteed up to a prescribed limit, with the full faith backing of the U.S. Treasury should the FDIC need credit.
Furthermore, the use of fiat money (representative currency with no intrinsic value, but that is declared by the government to be a legal means of payment) in itself requires a fundamental confidence in the financial system and the government behind it. Let's take this one step further. Most transactions today are virtual, with paychecks direct-deposited, bills paid electronically, stock trades completed in the blink of an eye. Even in times of economic uncertainty, most people trust that this invisible money holds value. What used to be marked on clay tablets, silver, and paper are now just some pixels on your computer screen.
That takes us back to catastrophe modeling results, which is our best formulation of what kind of losses can be expected in the future. While it took thousands of years to establish political and economic systems stable enough for us to be confident in money we can't see, in only 25 years, companies base multimillion-dollar and even billion-dollar decisions around equally intangible model outputs.
There is undoubtedly an increasing trend toward the use of analytics in our industry as businesses get more competitive and as investors and executive boards more closely scrutinize a company's return on capital. It is therefore especially important for the market to have confidence and trust in model output.
Catastrophe model output is the currency by which catastrophe risk is transferred. However, to be a viable currency, the model developer must earn your trust and confidence. How? By making sure that a given model is transparent and backed by sound science and the best quality data, and validated both using rigorous internal processes and by external independent experts. A model is only as good as its ability to help you make financially sound decisions, so a modeling company must make every effort to thoroughly verify the individual theoretical components of its models and ensure that it assembles into a coherent whole. However, ultimately,it is your due diligence to question whether the models that you rely on make sense in the real world, for your own business practices.
So trust, but verify. Question the assumptions and ask the model developer to help you understand the inner workings and limitations. Models are quantitative representations of complex physical phenomena, and as such, they will never be perfect. Just as there are concerns about the stability and value of money—concerns that have been around for 5000 years—there are uncertainties in model output that we will never stop thinking about. But also like currency, catastrophe models allow the market to operate efficiently by creating a common means of understanding and transferring risk.
Just as we would not want to go back to a barter system for the exchange of goods, the industry has advanced far beyond the days of using pin maps to control catastrophe exposure. While they may not provide the exact answer, models give both parties in a transaction an objective and scientifically sound measure of the risk that is being exchanged. To the extent that both parties have confidence in the underlying model, a trade is made possible.
Catastrophe modeling has become an invaluable tool in managing large and unpredictable losses across the insurance industry and beyond. We trust that the next 25 years will see models becoming even more transparent, flexible, and reliable.