3 ms·

When you take wildly changing vote count estimates (from +10 to +2 within two weeks) and turn that into more than 80% of probability of a specific candidate win

• mful 10y ago
  Nate Silver had an excellent response to this on one of the "Model Talk" editions of the FiveThirtyEight Elections Podcast[1]. TL;DR of the problems with Taleb's critique: 1. "Volatility" in the model is not random, as Talib asserts -- it happens at predefined points in the time. For example, the accuracy of polls jumps greatly after the conventions complete. 2. Volatility at a previous point in time does NOT mean there must be similar volatility going forward, as Talib implies. 3. Taleb's quickie models that he posted for explanation don't remotely match empirical data. I respect Taleb, but he seems to have fallen into a trap really smart people often fall into: being right with limited data, but wrong with complete data (in this case, he creates assumptions from his knowledge in other areas like options pricing, which are wrong in this domain). [1] Skip to 13:30 remaining http://fivethirtyeight.com/features/what-makes-a-tipping-point-state/ http://fivethirtyeight.com/features/what-makes-a-tipping-poi... EDIT: formatting