The Signal and the Noise (2012) by Nate Silver is a book on forecasting, statistics and analysis. Silver writes the popular New York Times blog 538 that focuses on US electoral forecasting. The book is also full of information about sport statistics, betting, using computers to play chess and other subjects. Silver also looks at the rise of Bayesian statistics as compared to a frequentalist approach.
The first section starts off by looking at political forecasting by pundits and others. Here, people who ignore the betting odds and the polls are just putting forward what they want to happen. Silver quotes Tetlock’s Expert Political Opinion which shows how little we can predict about political events in general. Silver describes how his models can beat the betting odds but only slightly.
Silver then looks at baseball and covers much of the ground that the excellent Moneyball covers. But Silver extends on that book and points out that today the performance models and scouts and their judgement have been melded.
Weather forecasting is then described and the remarkable accuracy of a few days of weather forecasting is explained. The improvements in hurricane forecasting are well described. Silver then contrasts weather forecasting the endless failing search for some predictive skill in forecasting earthquakes.
Silver then looks at the models and the forecasts for infectious disease and describes the challenges with them. The next chapter looks at how some people make money in sports betting which Silver describes as using large amounts of information and computing power to beat betting spreads by a few percent. Silver then shifts to looking at how computers were created that could win at chess and Jeopardy and then again puts forward his view that computers coupled with experts can improve over raw odds by a few percent. Continuing looking at probabilistic games the online poker bubble is described.
Shifting back to looking at economics Silver looks at economics where he points out that quite a few people had pointed out that there was a housing bubble in the US but there were incentives for many people to keep the bubble going and when it collapsed being with the herd was safer than making a call that there was a problem and being wrong. The book has a really good discussion of the efficient market hypothesis and it’s varying degrees of application.
Silver looks at climate predictions where he misses points made by others on the lack of data compared to weather models and also he completely misinterprets ‘hide the decline’ from the climategate emails as being about current temperature rather than hiding data that was relevant to paleoclimatology.
Following that there is a chapter on surprise attacks that looks at Pearl Harbour and terrorism where he puts forward model similar to those on earthquakes that looks at the frequency of attacks that kill varying numbers of people.
Finally Silver describes why he thinks we should be Bayesian, or fox-like, shifting and changing our predictions and making them using varying data sources and respecting the odds. He also points out that making predictions and seeing where your own biases lay can improve your understanding.
The book is really good. It’s a little long and a chapter or two could have been removed but the message that we should be Bayesian is put forward in an informative, fun way.