Insurance underwriters in Latin America have to make a large number of decisions every day, just like their counterparts in the U.S., but often with much less information at their fingertips. And when the combination of time pressure and thinking come together, what behavioral economists and psychologists refer to as “cognitive bias” kicks in. How does this affect how underwriters manage and price catastrophic risk in Latin America? Chubb expert David Heard explores these questions and more.
Nobel economist Daniel Kahneman, in his best-selling work Thinking, Fast and Slow, introduced the concept of “System 1” and “System 2” thinking.
“System 1” thinking, according to Kahneman, is the brain’s fast, automatic, intuitive approach. It developed over millennia as humans evolved in a more dangerous world where thinking quickly could literally make the difference between life and death. “System 2” refers to the mind’s slower, analytical mode, where reason dominates. Kahneman’s thesis is that to this day, System 1 has the upper hand in human thinking and steers System 2 to a very large extent.1
Insurance underwriters are subject to the same kinds of cognitive biases that affect rest of humanity. According to the fields of evolutionary psychology and behavioral economics, “cognitive bias” refers to the “systematic pattern of deviation from the norm or rationality in judgment, whereby inferences about other people and situations may be drawn in an illogical fashion."2
Many people are familiar with the aphorism “Perfect is the enemy of the good,” which has been around much longer than Kahneman’s fascinating work. Underwriters, like skilled professionals in many other occupations, unconsciously use “heuristic techniques” that help them to solve problems with methods that are not guaranteed to be optimum or perfect, but are sufficient to meet their immediate goals. Where finding an optimal solution is impossible or impractical, heuristic methods are often automatically used to speed up the process of finding a satisfactory solution. Heuristics can be mental shortcuts that ease the cognitive load of making a decision. Examples of this method include using a “rule of thumb,” such as an educated guess, an intuitive judgment or just plain common sense.3 Underwriters often refer to it as “using their gut”.
But what does this have to do with how underwriters manage and price catastrophic risk in Latin America?
The most significant catastrophic exposure in Latin America is from earthquakes, followed by hurricanes and floods. Since earthquakes have caused extensive economic damage and loss of life all the way down the spine of Latin America, that is what we will focus on here.
In any given Latin American market, killer earthquake events are few and far between. For underwriters in any given country, you can say they happen once in a generation or even two generations (25-50 years).
Prior to the September 2017 Puebla earthquake, the last generation of underwriters in Mexico who lived through a highly destructive earthquake causing many deaths and a large amount of economic damage were those on the job 32 years earlier in 1985 when an 8.1 magnitude earthquake caused upward of 10,000 deaths, 50,000 injured, 100,000 homeless and $5 billion in economic damages. For underwriters, and even some senior managers, the damages from the 1985 earthquake were a distant memory compared to the competitive pressure they were facing to hit their 2017 production goals. Could this be why underwriters at some companies were quoting earthquake rates in the highest hazard zones of Mexico City and at an 80% – 90% discount to tariff rates promulgated by the Mexican Association of Insurance Companies (AMIS)?
It is ridiculously difficult for an underwriter, an actuary, or anyone else for that matter, to calculate the “correct” rates for an exposure such as earthquake that is characterized by such low frequency and high severity. Going back to heuristics, underwriters often use experience, intuition and observation of competitors in make decisions on earthquake pricing. One of the most common mistakes is to see underwriters evaluate the health of their portfolio based on a blended loss ratio that sums short term exposures (fire) with long-term exposures (earthquake). I have even seen underwriters pay profit shares on mortgage portfolios covering earthquake or windstorm exposure, failing to consider the long-term nature of catastrophic risk.
Breaking down a blended loss ratio of 45%, it would not be surprising to see a fire loss ratio of 90% accompanied by an earthquake loss ratio of 5%. The blended loss ratio looks fine, the fire loss ratio looks bad, and the earthquake loss ratio looks too low. The key difficulty for underwriters is that that the two exposures operate on two very different time horizons—for fire exposure you can look at 1-3 years of experience, whereas for earthquake the best practice approach is to use a model that can simulate multiple events over 10,000 years and then calculate an average annual loss (AAL) spread out over all of those years. Even though the actual reported loss ratio for earthquake is coming in at 5%, a modeled loss ratio using the same premiums would more likely be showing 65 – 70%, recognizing that the “big one” will happen some day in the future. An insurance company that wants to remain solvent needs to collect premium for that future exposure today.
Commercially available catastrophe models (often referred to as ‘cat models’) have only been in existence for the last 25 years, and saw an intense jump in usage after Hurricane Andrew slammed into Florida in 1992 and caused the insolvency of 11 insurers. As a result, insurers increased their adoption of cat models at an exponential rate, viewing them as a more sophisticated and reliable approach to catastrophe risk assessment.4 Hurricane Andrew had laid bare the deficiencies in traditional actuarial approaches to quantifying catastrophe losses.
How do earthquake Cat Models work? Most models have three critical modules:
In the particular case of hurricane exposure, cat modelers are increasingly challenged with the hazard module because climate scientists believe the frequency and severity of large windstorms will increase due to global warming. Since the global climate system is extremely complex, however, it is difficult to predict what may happen over any 10-20 year time period in a given part of the world. Outputs of the hazard module are also driven by projections of sea-level rise and the number of feet a risk location is currently above sea-level. In the last 110 years (between 1901 and 2010), sea-levels are calculated to have risen by approximately 19 cm – more than what they had increased in the preceding 2,000 years. Experts have attributed approximately 30% of the damages from Superstorm Sandy (2012) to this 19 cm increase in sea-levels.
Cat models today are highly sophisticated and have been developed by thousands of PhD’s over millions of hours in the fields of geophysics, materials science, structural engineering, geology, climatology, economics and others. The models we have today could not have been run 25 – 30 years ago because we did not have the computing power.
So why the question about how underwriters price cat risks in Latin America? How does a reasonably intelligent underwriter compete with the millions of hours of work that have gone in to building cat models?
As good and powerful as they are, the cat models that have been developed over the past three decades cannot actually provide an underwriter with any certainty. The models often produce results that are different from what actually happens in reality. One reason is poor data quality. Data quality has been a challenge for underwriters in the U.S., and is even more of one in Latin America. For cat models to work, they require accurate information about risk location (a huge number of insured risks in Latin America are not accurately geo-coded), age of construction (the building code in force at the time a property was built is highly predictive of the structure’s vulnerability to stress from an earthquake or hurricane), and construction type (e.g. reinforced concrete vs. masonry vs. wood frame). Cat models are just as subject to the “garbage in, garbage out” phenomenon as models used in any other discipline.
Models can still be inaccurate even when good quality data is available given the high level of complexity in modeling future events and their economic impacts in the physical world. One of the phenomena that early models missed was “demand surge,” which is the inflationary impact caused from excess demand for materials and supplies following a catastrophic event. For some insurance companies, the additional losses generated by demand surge can actually lead to insolvency. Following the 1994 earthquake in Northridge, California, for example, 20th Century Insurance Company was almost bankrupted due to demand surge impact reported at 20%.5
Remember that underwriters also operate in competitive markets. When an underwriter sees a competitor charging lower rates for cat coverage, the natural question comes to mind: “How can they do that?” If the underwriter has seen evidence of a cat event where the model did not predict the expected outcome, it puts additional doubt in his or her mind. Maybe the insurance company is not using the right model? Several modeling firms compete for the business of insurers, reinsurers and governments, including RMS, EQECAT, AIR and ERN. Whereas modeling firms refer to real world deviations from their model as “model-miss,” the underwriter simply says the model doesn't work. “PhD’s in Silicon Valley may understand the fault lines of California, but they sure don’t seem to understand Peru,” the underwriter may be thinking.
The frustration faced by underwriters in Latin America is no different from that of professionals in other disciplines that also rely on models. To cite one of the great statistical minds of the 20th century, George E.P. Box, “Essentially all models are wrong, but some are useful.”
At Chubb, we have an understanding that models can be wrong, but that they are also highly useful. The challenge an underwriter faces is not dissimilar to that of the mutual fund manager trying to outperform other fund managers by climbing to the top of the investment return leaderboard. He or she knows from internal models that the stock market is over-valued and will self-correct at some point in the future, but when? If a fund manager pulls back now and moves money into different asset classes in preparation for a drop in the stock-market, he or she may miss the next 5 points of market run-up and see his or her performance ranking drop relative to his peers. It is a very lonely place to put a stake in the ground and say “no more” when everyone else is charging ahead.
Well-managed insurance companies know the pressures underwriters face every day—making thousands of decisions that affect the company’s top line, bottom line and risk profile every year. Beyond training and the development of increasingly sophisticated decision-support tools, one approach the best insurers use to ensure that System 1 thinking is counterbalanced with slower, fact-based decision making is to establish levels of underwriting authority at the country, regional and global level. In this way, even when an underwriter is tempted to succumb to market pressures and unconsciously use heuristics (aka their “gut”) to help make a decision, underwriting authorities can require an underwriter to slow down and obtain approval from a higher level in the organization before dropping a rate on earthquake exposure to a price that is more than x% below what the model indicates. There is a trade-off of course, because this can slow down the decision-making process and periodically generate some tension between country, region and home office. Sometimes business opportunities are lost. The flip side is that this approach, when well-managed, helps guaranty the solvency and continuing ability of insurance companies to service their customers at their time of greatest need following major catastrophic events.
1 Harvard Gazette: Layers of Choice, Colleen Walsh, Feb 5, 2014. news.harvard.edu
2 Haselton, M.G.; Nettle, D. & Andrews, P.W. (2005), The Evolution of Cognitive Bias
4 Lloyd’s: Catastrophe Modelling and Climate Change, Ralph Toumi and Lauren Restell, 2014
5 What we know about Demand Surge: Brief Summary, Anna H. Olsen and Keith A. Porter, 2011, American Society of Civil Engineers
David Heard is Senior Vice President, Personal Lines & Agency, Chubb Latin America; this article originally appeared on LinkedIn.