Prediction and forecasting have always been interesting to me, both as personal intellectual curiosity, and a professional concern for a data analyst. Below are the notes I took as I read two books on the subject written for a general audience: Nate Silver’s The Signal and The Noise and Superforecasting by Philip Tetlock and Dan Gardner.

The Signal and The Noise (2012)

Nate Silver is known, among other things, for his work in forecasting elections through the now-defunct FiveThirtyEight website.

• The difficulty of assessing aggregate risk of bundled mortgage-backed financial assets precipitated the 2018 financial crises since these assets’ market performance were often dependent on each other. Credit rating agencies used house prices going back 2 decades to assess their assets’ risk. However, house prices 1980s - 2000s were always stable or increasing, precluding the possibility of small declines, let alone crashes. This is an example of a out-of-sample problem.

• Political forecasters in the US are often fixated on broad stereotypes of what kinds of voters each of the two parties attract, rather than drawing on issue-specific polling to dissect smaller trends that sometimes run counter to overall ideology. To me, Silver seems to endorse a Bayesian approach of continually collecting information about the data, and updating inferences accordingly. FiveThirtyEight operate via three principles: think probablistically, good predictions change and look for consensus.

Superforecasting: The Art and Science of Prediction (2016)

Philip Tetlock is a professor at the University of Pennsylvania who along with his wife and research partner Barbara Mellers was responsible for The Good Judgement project.

Compared to The Signal and The Noise, Superforecasting is written in a more academic style. Interestingly, Nate Silver had met Tetlock and devoted a section in his book in The Signal and The Noise to discuss Tetlock and his ideas.

• Prior to World War II, many medical procedures were carried out despite dubious (or absence of) evidence for their effectiveness. Medical practice did not become evidence-based until randomized controlled trials were first seriously attempted in the late 1940s, advocated by Archie Cochrane.