ForecastEx prediction market probabilities for Hurricane Isaias are appearing live on Bloomberg TV.

Crowdsourced feedback has greatly enhanced my analysis of the accuracy of the ForecastEx daily temperature prediction market. The latest revamped version of my manuscript is now posted on EarthArXiv. Key points: Prediction markets have a documented history of distilling dispersed information into well-calibrated probabilities (the wisdom of the crowd). However, prediction markets' value proposition is clearest for forecasts of, e.g., elections, where the prediction targets human behavior, information is subjective and scattered, and human biases play a large role. It is far less clear that prediction markets should improve predictions on questions in physical science, such as daily high-temperature forecasts, where competing forecasts are not pontificators with agendas but physical and statistical models refined over decades and supervised by expert meteorologists. However, this study finds that the ForecastEx daily high-temperature prediction market had the lowest errors among 20 established sophisticated scientific forecasting systems (at 24 US locations over 5,405 location-days). - Q: Was the prediction market advantage simply a result of the competing systems not targeting the same observations that the market settles on? A: No. The market advantage exists against systems that are specifically tuned to target the same observational data as the market at the same timescale, and the advantage persists when alternatives are scored against their own standards that better represent what each system targets. - Q: Was the prediction market advantage just a result of a quicker response to live observations during the day? A: No. The prediction market advantage over alternatives emerges immediately upon contract listing on the day before the target date and persists through the target date. - Q: Was the prediction market advantage just a result of it being ‘live’ whereas alternatives embed latency? A: No. The prediction market advantage over alternatives persists even when I sample forecasts only at the moments when alternatives were freshly published. - Q: Was the prediction market advantage a result of some idiosyncratic advantage at a few geographic locations? A: No. The prediction market advantage over alternatives was broad and consistent across locations. - Overall, the ForecastEx prediction market has consistently lower errors than physical models teamed with statistical corrections and lower errors than expert human meteorologists with access to the same information as the market. It produces these better forecasts not by any particular design, but as an emergent property driven by the direct financial incentives to be correct.

Prediction markets can represent the best collective knowledge about our energy and climate future. interactivebrokers.com/campu…