The Good Judgement Project is using the Brier score to rate participants forecasts. This is not LW’s usual preferred scoring system (negative log odds); Brier is much more forgiving of incorrect assignments of 0 probability. I checked the maths, and you’re expected score is still minimised by honestly reporting your subjective probabilities, but are there any more subtle ways to game the system?
Perhaps it encourages one to make long-shot bets? If you aren’t penalized too badly for P=0 events happening, this suggests that short-selling contracts at ~1% may be better than it looks.
The Good Judgement Project is using the Brier score to rate participants forecasts. This is not LW’s usual preferred scoring system (negative log odds); Brier is much more forgiving of incorrect assignments of 0 probability. I checked the maths, and you’re expected score is still minimised by honestly reporting your subjective probabilities, but are there any more subtle ways to game the system?
Perhaps it encourages one to make long-shot bets? If you aren’t penalized too badly for P=0 events happening, this suggests that short-selling contracts at ~1% may be better than it looks.