Frankenstorm and the Risk of Models
October 29, 2012
The Frankestorm is upon us, and by all accounts our friends in New York and along the east coast seem to be doing just fine at this stage.
The models that assumed Hurricane Sandy would hit New York initially on Sunday night triggered a shut down of the entire NYC transit system, and the decision to close US financial markets.
Reports from our friends in New York this morning say things are rainy, but that nothing crazy is happening.
Here’s a live cam of the Wall Street Bull http://www.earthcam.com/usa/newyork/wallstreet/chargingbull/ – at the time of this writing, there wasn’t much going on! Few cars, no tourists.
Will this be a ‘storm of the century/decade/year’ scenario, or will the models cry wolf one more time and cause people to be more complacent facing the next storm? Time will tell.
The next question is, should the failings of weather models make us question the predictive ability of all models?
Certainly the housing price models that failed to predict the burst of the housing price bubble and mortgage security debacle could potentially be forgiven, until we found that price declines had not been contemplated and were not compatible with those models.
Even the simplest models of interpolating or extrapolating numbers can lead people astray. One of my favourite examples (first brought to my attention by the folks at Brandes circa 2004) is this:
(San Francisco Chronicle, October 27th, 1993)
- When Elvis Presley died in 1977, there were an estimated 37 Elvis impersonators in the world.
- By 1993, there were 48,000 Elvis impersonators, an exponential increase.
- Extrapolating from this, by 2010 there will be 2.5 billion Elvis impersonators.
- The population of the world will be 7.5 billion by 2010.
- Every 3rd person will be an Elvis impersonator by 2010
Those who walked the streets in 2010, may have noticed that less than 1 in 3 people were actually Elvis impersonators, but it seems that every time companies or sectors or markets reach bubble valuations it is the errors of estimation that overestimate societal shifts in behaviour.
All models suffer from the risk of bad assumptions. And for many, particularly investors, it is the risk of buoyant assumptions that ultimately cost us money.