Review: The Signal and The Noise

I've been a big fan of 538 and Nate Silver since the 2008 elections. The Signal and the Noise is his book about predictions, the state of the art, how they work, how they don't work, and the intersection of those concepts along with his interests. Nate Silver's a very smart guy, which means that his interests are broad and fascinating.

What comes through the book is Silver's humility --- he claims that he was extremely successful because political predictions as set by TV pundits and Fox News sets a particularly low bar that's easy to beat if you just do a pretty good job. Note that Silver's models, however, beat the so-called prediction markets like InTrade, for instance, which means that not only did Silver beat the TV pundits, but he also beat fairly size-able markets with real money sitting on the line.

What's fascinating about the book is we see how Silver sees the world through the prediction lens. He even treats Chess as a prediction problem, and his write-up of the Kasparov vs. Deep Blue matches gave me insight into the matches that I didn't know prior to reading his book.

Silver also covers epidemic modeling, weather forecasting (including hurricane forecasts), earthquakes, stock markets, economic bubbles, sports, poker, and global climate change. His explanation of Bayesian reasoning, its' history, and application in the modern world is as clear and enlightening as any I've seen, and his considered understanding of our prediction failures is profound and insightful. Fundamentally, weather forecasting has been the most successful of the disciplines examined in the book, and Silver explains why.

The biggest weakness of the book is one that a perceptive reader will see as a theme over and over again. Fundamentally market incentives skew predictions. For instance, the political pundits have incentives to make big predictions and tell stories that are wrong, because they're more entertaining to the masses than Silver's nuanced analysis. Economists, mutual fund managers, and all have incentives to be wrong conventionally than to be right unconventionally. It is because of these distortionary effects that many of our predictions fail, rather than because we do not have the tools to do the job correctly, or because Bayesian reasoning isn't widely utilized.

Having said that, the book is fascinating, interesting, and enlightening. It's the best book I've read all year, and hence comes highly recommended.