John Tierney’-s column the Science section of today’-s New York Times discusses a method for forecasting difficult to predict events. The Copernican Method, advocated by Princeton physicist Richard Gott, allows one to generate confidence intervals that an event will occur using only the duration time until now (that is, how long the event has been at risk but has not occurred). Using the often not realistic assumption that there is nothing special about today, one can derive the ninety-five percent confidence interval for the time until the event occurs,
(1/39)*t_past <- t_future <- 39*t_past
where t_past is the duration time so far and t_future is the stochastic time until the event occurs.(*) Intuitively, events for which we have not observed a failure for a long-time are more likely to persist than ones which have only been in existence for a short time period. The article (along with the original Gott (1993) piece) give many examples of his formula at work such as how long Stonehenge will remain standing to how long political leaders will stay in power.
I remember reading the New York Times column in 1993 which first discussed this approach (sorry may be gated) and finding this to be not very convincing. Think about the Doomsday case. Of course today is quite different from the past: the events which could have led to man’-s extinction in the past (largely exogenous natural events) are quite different from the dangers of today and the future (man-made events). But I always find data convincing. The NYT article claims that Gott made accurate forecasts of political tenure and the closing date of Broadway plays though I have been unable to track down the original predictions myself.
Well I doubt this will be of any use to folks investing in prediction markets. It has been about seven years since the last Democratic president. Applying Gott’-s formula, this means with ninety-five percent accuracy we can say that the next Democratic administration will begin at least two months from now and no more than 273 years from now. I think we do not need a formula to figure that out.
(*) See Monton and Kierland (2006) for a derivation
Previous blog posts by Koleman Strumpf:
- Prediction Markets in the Classroom: Inkling Markets
- Slides of presentations from Conference on Corporate Applications of Prediction/Information Markets (1 November), Kansas City
- Summary of Conference on Corporate Applications of Prediction/Information Markets (1 November), Kansas City
- Reminder: Corporate Applications of Prediction Markets Conference (1 November)
- Conference: Corporate Applications of Prediction/Information Markets (Thursday, 1 November 2007)
- Win Justin’s Money? (re: Is there manipulation in the Hillary Clinton Intrade market? Redux.)
- Is there manipulation in the Hillary Clinton Intrade market?