The Greek philosopher Thales of Miletus is said to have become the first options trader based on an accurate weather prediction. Expecting a good harvest, he cornered the olive press market and ended up making a fortune. Weather science has come a long way since ancient Greece and so too has its intersection with markets.
Weather forecasting may be both art and science, but the artistry of it can be quantified, says Stephen Bennett, founding partner and chief science and products officer at EarthRisk Technologies.
Bennett's experience as a meteorologist includes a stint with Enron Corporation, the collapsed Houston-based energy company, which then led him to take on a role at hedge fund Citadel. Citadel, he says, applied a highly quantitative, analytical approach to weather and EarthRisk now uses that mind-set in its weather forecasting approach.
Throughout his career, he has seen an increasing level of sophistication for predicting weather in the one- to two-week window, though the numerical weather prediction models begin to break down at the longer end.
"Models put the earth in a big computer programme and it simulates the movement of every air molecule on the globe. That works pretty well to about a week, but the errors grow over time," he explains.
The high degree of accuracy in the shorter end of the time scale is a major factor in the successful use of weather data streams to calibrate predictions in energy prices. Most energy trading shops have deep-level access to many different weather data sources and use the information as often as six or seven times a day to revise their expectations, he says.
Daily temperature fluctuations drive energy demand, so if it is a little bit warmer today than yesterday, the air conditioners are going to be running longer for example.
Though weather prediction has a long history with using probabilistic methods, the application of those outcomes in financial markets is relatively new. Bennett expects that it is going to be one of the biggest changes for how people work with weather data over the next five years.
"(A weather forecast) needs to be treated as a distribution, not as a single point, and that is a major delineating factor right now in its successful use," he says. In other words, expect the weather forecast is going to be wrong, opt for a distribution of possible outcomes and understand what portion of those possibilities link to specific events in the commodity being traded.
"That is the way of the future as more shops understand the nature of distribution and apply that to their decision-making as opposed to a deterministic forecast," he says.
He adds, however, that any automated strategy would need to take into account not just forecasting, but how much to weight any weather strategy day-to-day. "There are long periods of time that go by where factors such as geopolitical events, fundamental events such as power plant outages, are far more important. I don't think you could build an algorithmic, automated strategy based only on weather, let it run for a year and be profitable. You have to know when to turn it on and off."