Sunday, February 7, 2016

Book Review: Superforecasting by Tetlock and Gardner

The theme of this book is very similar to Nate Silver's The Signal and the Noise. The first few chapters are worth reading, while as it drags on the final one-third of the book is less appealing, as it indulges in corny talks of team works etc.. The new ideas I got from this book, in addition to those already listed in the review above, include:
  1. The use of Brier score for rating forecast, because 'good' forecast should be in probabilistic terms
  2. As a corollary, one has to make enough forecasts in order to have a statistically significant score (=/= good score)
  3. Update predictions as frequently as appropriate
  4. There is a vicious cycle going on in the 'Mainstream' of our society: people are not aware of the importance/option to think probabilistic -> they crave for binary, determined answers -> the media are aware of this demand -> the media cultivate the guru/pundit culture -> the people get more addicted to this way of thinking
 All in all, it is not as readable a book as Silver's, but I would still recommend it to anyone who is serious about improving and understanding the art of forecasting.

No comments:

Post a Comment