To what end does one build a cat model?
The running myth is that catastrophe models are built without considerations of climate change and with nothing much more in mind than the "the best science." While a bold face creative interpretation of geophysical science and the importance of client demands it is actually quite rare to find printed documentation contrary to the running myth. But while looking for something else, I came across two examples this morning.
In this document (2014), Dr. Rick Thomas on behalf of Willis Re, suggests that near term estimates should use inflated estimates of hurricane risk because it is difficult to tell the difference between the affects of AMO and changes in SST due to anthropogenic climate change. And so he concludes: Meh, what ever floats your boat...
Medium-term rates conclusions
If it is climate change, the rates of extreme storms are up permanently
If it is AMO related, maybe we will start to see a fall in frequency (given Met office forecasts)
Either way, upwards adjustment to long-term rates to reflect elevated current rates of storms seems justifiable
Size of adjustment needs to be weighed carefully, and models run to estimate losses for multiple possible scenarios
The second document, has a couple of quotes useful for showing the link between climate change and cat models, and modelers and clients. All emphasis below are mine and more mental notes than much else.
Catastrophe (Cat) Modeling has developed almost exclusively in the private sector, driven by the needs of financial institutions such as insurers and reinsurers to understand and quantify their risk. It is currently both a hundreds of millions of dollars industry, and an innovative and exciting area of scientific research.... Cat modeling companies today employ more PhDs than many academic departments, and hire dozens of academic consultants from a wide range of disciplines.
The occurrence rates and intensities of many weather-related perils vary on multi-year to decadal multi-decadal timescales. Driven by the needs of their clients, cat modellers today are primarily interested in quantifying risk in the next 1-5 years. In academic realms of atmospheric science / geophysical fluid dynamics, the emphasis is on understanding and modeling of the phenomena. In cat modeling, the emphasis is on trying to produce a reasonable prediction, whatever the current level of understanding. This is particularly difficult in two instances: prediction of tropical Atlantic sea surface temperatures which influence hurricanes, and north Atlantic circulation patterns which are related to storm activity over Europe.
In cat modeling, the emphasis is on prediction. In order to make accurate predictions, it is necessary to ensure that statistical models are not overfitted to observations. Early generation cat models were overfitted, and thus represented historical data very well but had poor forecasting performance. The danger of overfitting is very high in cat models, because of their large numbers of parameters, and the small amounts of available data. There is a current trend, therefore, towards models that are built very rigorously to avoid overfitting in statistical components. Such methods are computationally intensive, and have only recently become feasible.
With growing demand for such output there are clearly many exciting opportunities for partnership between the insurance and cat modelling community and the climate modelling community. Building climate change into cat models can be achieved in a myriad of ways. Determining which is best, is where the challenge lies.
The roundabout argument here is that there are a bunch of experts making predictions thought to be accurate though the underlying understanding is lacking (and irrelevant) and historical data is lacking. Explicit regard of past models as overfitted though they have long been heralded as accurate predictions gives little comfort in current claims of increasing accuracy. Moreover, the time scales needed to judge accuracy of predictions are too long.
It is still just a man sitting behind a curtain pulling levers and pushing buttons to seem all knowing. The difference is instead of Munchikins he's got a empire of doting PhDs.