Temperature Trading

Over on iPredict, New Zealand's event stock market, they're running contracts on this year's global temperature anomaly as measured by the HadCRUT3GL series.

The first contract, TEMP.2009, pays out at $1 if the 2009 global temperature anomaly is higher than that of 2008. The 2008 global temperature anomaly was 0.324.

The second contract, TEMP.2009.HIGH, pays out at $1 if the 2009 global temperature anomaly is the highest ever. The largest anomaly in the series is 1998, when it hit 0.546.

Six months of data are now out. So half of what goes into making up this year's average is known. The average of the six months so far is 0.403. So, just on raw numbers it looks pretty unlikely that 2009 will be warmer than 1998 or colder than 2008. But we can do better than that. Throw the 160 year time series into STATA (or your favorite regression package) and run a simple forecasting model. The simple one I use regresses average temperature on a year trend, some lagged temperatures, and the temperature readings for January through June; I weight observations by the average number of temperature recording stations in the given year. From that, we can easily get a predicted temperature anomaly for 2009 and, more importantly, a measure of the year by year residuals. The residuals tell us how far the actual temperature varied from the model's predicted temperature.

With the residuals, I can look back over the history of the time series and count the number of instances that the model gave a prediction so far out that, if we had an error that large again this year, we'd either be warmer than 1998 or colder than 2008. Doing that, I find that there's no instance where the the model was sufficiently out to get us a temperature this year warmer than 1998, and only one instance where it was sufficiently out to get us a temperature this year colder than 2008. So the price of both contracts should reflect a smaller than 1% probability that we'll either be warmer than 1998 (price of TEMP.2009.HIGH should be < $0.01) or colder than 2008 (price of TEMP.2009 should be > $0.99).

Current trading price of TEMP.2009.HIGH is about $0.065; TEMP.2009 trades at $0.88. The former gives you a 7% return over the period to end-January next year; the latter, about 14% over the same period. If my model is right. I've consequently dumped a bit of money into both. Unfortunately, iPredict operates under regulation from the Securities Commission requiring that folks deposit no more than $1000 per 6 month period. So I can't put in any more.

One possibility is that there's an informed trader with access to better forecasting technologies that incorporate more than just the observed temperature series.

But I tend to think rather that we're being hit by the deposit limit. There's only a small subset of traders who

1. watch the market closely enough to note the low risk, long time horizon trades
2. who also don't watch it closely enough to prefer trading the contracts that are higher variance and end quickly
3. and who trust the market enough to make substantial deposits.

Why get a 14% return over 6 months when you can get a 50% return over a couple of weeks in some of the other contracts? Absent the deposit restrictions, that small subset of folks could move their retirement accounts over to iPredict and earn a decent return.

However, that doesn't explain the difference in pricing between TEMP.2009 and TEMP.2009.HIGH. One explanation is that TEMP.2009 only became such a sure thing with the release of the June temperature figures: prior to that, the price on that contract was close to right. If TEMP.2009 quickly moves to approximate (1-TEMP.2009.HIGH), then the liquidity preference story might not be far out.

Full disclosure: I have big positions on both contracts, and would have larger positions if I were allowed to deposit more. This should not be interpreted as investment advice.