The art and science of prediction (Nate Silver's 'The Signal and the Noise')

“This book is less about what we know than about the difference between what we know and what we think we know”, writes Nate Silver somewhere at the end of his book. He could not have phrased it more beautifully. Our brain is a remarkable instrument that, when it comes down to predicting almost anything, contains a substantial amount of techniques to mislead us. Of course, these techniques often helped us to survive in tougher times, but they fail us miserably in the area of future-watching. Just as an example: apparently we have a bad habit of neglecting certain risks just because they’re too hard to measure…. Not sure how far this will get us in terms of survival, however…

The solution? Statistics. According to Nate Silver, that is. He’s not saying statistics can predict everything (or, for that matter, anything) –‘all statistic models are faulty, but some are useful’- but at least it can provide us with a more accurate picture of the (possible) future(s).

Chapter by chapter Nate applies this thought on very specific subjects, ranging from the American elections (his specialty), to weather forecasting, earthquakes, economic growth and climate change, to more tangible topics such as poker, chess, or beating the stock market (which, he concludes, is impossible).

Fascinating material, and some conclusions linger on for quite some time. For instance: the level of uncertainty of prediction increases when the system we study is non-linear and dynamic (which sounds intuitive at first, but is very often forgotten, for instance when governments predict next year’s GDP), or when the data we start with is not accurate enough (like with weather forecasting). We also need to watch out not to ‘overfit’ our models (adapting it with the data we have at hand), as we often do with predicting earthquakes. Sometimes we can use models developed for one system onto another system, like applying the power-law distribution we use for earthquakes onto the event of future terrorist attacks (which results in a somewhat bewildering conclusion).

The central thought Nate Silver uses across his book is the formula of Bayes which, for the curious people among you, looks like this:

[xy] / [xy + z(1-x)]

Where

X = the initial likeliness you would attribute to an event.

Y = the ‘positive possibility’ (the likeliness of an event without taking x into account)

Z = the ‘negative possibility’ (the likeliness of the event not taking place).

Believe it or not, but with this formula you can even calculate the likeliness that your partner is cheating on you! The key figure in the formula is the ‘x’, the likeliness you grant to an event accuring –which implies that the overall likeliness gets more accurate as you gather more experience or knowledge about the event (if your partner cheated on you in the past, the ‘x’ would increase, and hence the overall likeliness in a particular event as well).

Strangely enough this thought process can be applied to virtually any system (or prediction), though dependent on a number of factors it can produce totally different outcomes, with different levels of accuracy. The world will never be predictable, but with a sensible use of statistics it can become a little more so…

Fascinating reading…