Tuesday, March 13, 2007

ExChange in the Weather

Greetings, rain-dodgers.

A few posts ago, I made a case for future-proof policy; that is policy which automatically keeps up with the best that today's technology has to offer. I've advocated the use of results-based prizes for rewarding the discovery of useful medical treatments, since they tend to align public and private interests. Today I'm going to talk about another way we could make our policy future-proof by harnessing the free market: have our government meteorological systems switch over to prediction market-based weather prediction.

The Status Quo

Today, typically large institutions or governments hire meteorologists whose full-time job it is to interpret computer models based (largely) on publicly-available data. It takes a relatively long time for new weather-prediction models to gain acceptance: each one must be academically-verified and promoted, and the uptake of better weather-prediction techniques seems to be a patchwork affair.

At the same time, there are hundreds of math and physics geeks with computer power to spare who like to try their hand at predicting just about anything. Even the private sector has been unable to tap this latent talent pool, as is evidenced by the fact that the Netflix prize has been claimed.

Netflix Prize Exhibits Geek Talent

The Netflix Prize rewards people for discovering new ways of predicting the ratings people give their movies based on which other movies they liked. The contest started on October 2nd, 2006, and by October 15th one team had already beaten Netflix's predictions enough to claim a prize. If even a private-sector firm is unable to efficiently harness the best numerical prediction methods out there, what hope does a government agency have of keeping cutting-edge?

Prediction Markets for Weather Prediction

Imagine instead that any math nerd with a computer and an Internet connection could instantaneously profit by predicting weather better than rivals without having to apply for meteorology jobs. There are already small-scale weather prediction contests (like the WxChallenge), but these are still mostly for bragging rights, not for general-purpose weather and climate prediction.

Prediction markets are like stock markets. You can already buy and sell shares in, for example, Hillary Clinton becoming the Democratic nominee, in an online prediction market. The shares are worth $1 if the event takes place, and nothing if it doesn't. The fact that Hillary's shares trade at about 40¢ means there's a market consensus that there's a 40% chance Hillary will be nominated. Prices fluctuate with every factor that may influence her nomination probability.

If we did the same for weather (and maybe even sweetened the pot a bit to provide incentives for high-volume trading and good predictions) we could find the geeks' market consensus over the chances of it raining tomorrow. Other predictions could be made too, like the total rainfall in a season, or any other season-related information which might be economically, socially or environmentally relevant. Storm warnings could be automatically posted through regular weather channels when the price of storm stocks rose above some (low) threshold, like 20¢.

Probably what would happen is that a few centralized weather servers would emerge which would make predictions about weather at many different locations, while local "old salts" who have a sense for the weather could also make a quick buck while letting the world in on their secret, quasi-instinctual privileged weather-sense.


There is no method as efficient as the anarchy of the market to predict the value of a commodity. If we commoditize knowledge about the weather, we will automatically harness all the disparate knowledges about our turbulent atmosphere to reward the weather-seers and keep the rest of us dry under umbrellas when appropriate.

Stay dry!


PS I have to add a final caveat: if some foreign power (like a government) had deep enough pockets and had a desire to manipulate the market (by, for instance, ruining 4th of July plans by buying shares in it raining everywhere), they could do so as long as they were prepared to sustain a virtually unlimited financial loss. Perhaps a good safeguard in the system would be to include automatic "bizarreness detectors" which would sound an alarm if some fishy market activity starts.


Knaldskalle said...

I don't get this. Just because I'm willing to put my money where my 'aching elbow' is, that doesn't mean that I'm any good at predicting the weather, it just means that I'm willing to gamble with my money.

Even if we put all the best weather stations in the world to predict the weather, that doesn't mean that they'll get it right - nor does it mean that because they get it wrong that they're useless, it just means that weather might bee too complex a phenomenon to reliably predict (unlike political elections).

Just because Hillary's stock is currently at 40%, doesn't mean that there's a 40% chance that she'll win, it means that the fools(?) who're playing that game overall think that there's a 40% chance, but if the number of people playing is small there's a very good chance that the "predictions" they're making is wrong - especially if non-US people can play, since they'll be able to skew the results (being able to play, unable to vote)... See what I mean?

LeDopore said...

You're right that the predictions are going to reflect a bias if the people willing to bet systematically have a skewed viewpoint.

However, if that's the case, then you'd be able to make money by betting against the skew. Therefore, as long as there are at least a few greedy but minimally-skewed people out there willing to bet, they're going to be able to make money and improve predictions at the same time by betting against the skew.

You make a good point that this method is never going to overcome chaos theory and give perfect predictions. It's just a way to publicize the best predictions out there by rewarding the most accurate forecasters.

Knaldskalle said...

Amazingly, I just found a newspaper website that has a "newsgame", where you can "bet" on (buy stocks in) a coming headline. "Temperature will drop below -10 centigrades in January" or "Obama wins Democratic candidacy" and "Hillary wins Democratic candidacy". If you get it right (within a certain time limit) you win a fixed amount of currency, that you can then buy more "stocks" with (no actual money is involved). The current standings are 50, 68 and 60 respectively. These standings reflect the cost of a share, so the more certain or popular a given event is, the more expensive it is to bet on. Unfortunately it's not an English speaking paper...

Knaldskalle said...

Nope, I was wrong, it's not a fixed amount, you earn the difference between the standing you bought it at and 100. So if you bought a "newsstock" at 45 and the event happens, you earn 65 credits. Obviously the stock cannot rise above 100 or fall below 1.