|This article is work in progess. The author is working on it and it is not yet ready for review.|
A post-mortem (in forecasting) is an analysis of a forecast, typically after it resolved. It is commonly used by professional forecasters to improve their reasoning for future forecasts.
Consider, for example, two forecasts on whether X will be reelected:
- Alice looks at recent polls, base rates for reelection, actuarial tables for people of X's demographic, unemployment numbers and energy prices as proxies for the public content, etc. and arrives at a 78% chance of X being reelected.
- Bob checks social media and whether Mars is aligned with Saturn to arrive at a 1% chance of X being reelected.
Assume now that X ends up not getting reelected. Despite being "correct" Bob should not stop questioning his approach and Alice should not (completely) abandon hers. Alice should, however, start thinking about whether she could have done better. So Alice's post-mortem might consist of looking into questions like
- Did I want X to win and thus discard/overlook some evidence against X getting reelected? (See Cognitive Biases.)
- Did I consult a biased set of polls?
- Did I rely too much on the inside view or the outside view?