Sunday, October 18, 2009

Prediction with game theory

I am going to this talk on Thursday so I am reading the book Predictioneer by Bruce Bueno De Mesquita. He has a ted talk video on his prediction ideas.

The use of game theory in forecasting seems based on the axiom that people are rational. The problem is they are not, here is a list of the different ways they are not. If you know how they are going to be irrational you can alter your models to take this into account. However if you fail to do this you will end up with less accurate models.

In chapter 2 the book claims people are rational and particularly that we are transitive in our preferences. "to be rational.. their preferences must not go in circles. For instance if I like chocolate ice cream better then vanilla and vanilla better then strawberry i presumably like chocolate ice cream better than strawberry." So no rock paper scissors can exist in human preferences.

The problem is they do. There is loads of evidence that people can have intransitive preferences. For example May showed in 1952 that people can have intransitive preferences for wealth, looks and intelligence in a partner. So how can you have a mathematical system modeling the world where one of your axioms is false?

While on the subject of intransitive never play dice with Warren Buffet. There are loads of stories of how he tries to con people with a set of intransitive dice. Edward Thorpe for example. Or Bill Gates as described in Bill Gates Speaks: Insight from the World's Greatest Entrepreneur By Janet Lowe

Buffet once attempted to win a game of dice with Bill Gates using intransitive dice. "Buffet suggested each would choose one dice and discard the other two. They would bet on who would roll the highest number most often. Buffet offered to let Gates pick first. This suggestion instantly aroused Gates curiosity. He asked to examine the dice after which he demanded buffet choose first."