Book Review: The Signal and the Noise by Nate Silver

Predictions are hard, especially for the future. However, there are people that make their living out of it. Some are good in it, others are successful and few are successful and famous. One of the latter kind is Nate Silver - a blogger and columnist in New York Times. His star rose with the accuracy of his predictions of the latest presidential vote in US and many of us have been willing to know the secret of his success. The thirst has been quenched by his book The Signal and the Noise.

What's in it
The 13 chapters of book take the reader through the building and implementing of forecasting models in a great variety of fields - sports, weather, economy, stock exchange, climatology and many others. The coverage is extensive and I believe almost every reader would find an example that is close to his business or other activity and thus could understand the key concepts much better. The chapters are not that related to each other and could be read separately but I would recommend to follow the author's line for greater pleasure and clearer introduction of the concepts. The book takes the reader through stories and characters to crystallize the core concepts of the book one after another and to emphasize the important points. It does not cover any math specifics about models and focuses on the implementation and the reasons for their success or fail. The reader would learn about the difficulties of distinguishing the signal and the noise and the often mistake of take one for the other, dangers of over-fitting, the incentives of forecasters, the assumptions for implementing a model, the criteria for model performance and all the other things that come with forecasting process. The author makes the case of wider implementation of Bayes statistics and abandoning of some traditional hypothesis testing methods employed currently. It is bit controversial but controversy is good in the pursuit of the better.

What I think about it
I expected that the book is just an attempt to cash-in on the wave of Nate Silver's popularity but I liked it. It is easy going book, turning one page after another with nice stories and characters in it. It skips on heavy terminology and math complexities. Mr Silver lays the foundations that constitute a successful predictive model. No surprises in it here and anyone with some exposure to using or building forecasts would get a "I-knew-it" smile when reading some. Some of practitioners would also have "I told them so!" moment as well. I also like that the book looks at the process of model building in its full complexity and I did feel something is missing. It is very contemporary - the book is published this year and many of the examples are very recent. Because of the wide variety of application forecasting covered in the book, there were lots things new for me - like the story of online poker or the climate models for example. I also got a better idea of Bayes theorem application and the deeper meaning of it. There are details from Silver's personal history that shed some light on the factors that made him a success and the success drives have been of interest for me since Gladwell's Outliers, so good points for this as well. I really liked the story about economic data and forecasts based on it - maybe because it is closely related to my current work but I believe many readers would relate to a story in the book.

Who could be interested in reading it
The book would be a good choice for the wide public due to the covered topics and the easy writing style. Modeling and forecasting practitioners would find it useful for reviewing and improvement of their work - despite the fact that everybody knows everything, there are things to learn and reconsider even for everyone in  this business. Managers and everybody else involved with forecasts in some way would get better understanding of the processes, assumptions and people producing the lines on the charts that go into the future as well as they would get better at using the forecasts and setting expectation from them. Of course, it would be insane to expect the attitude toward predictive analytics to change dramatically after this book but it is another small step toward better models and better use of them.