Nat Silver author of this book got his notoriety or even fame in very unusual way. He specializes in political predictions and his prediction proved to be highly correct in last few election cycles. Since my own political prediction were more often then not incorrect it make lots of sense for me to pay attention to what he has to say and he has a lot to say about art of forecasting and prediction overall.
First of all he gives a nice common sense definition of difference between forecast and prediction: Prediction is a definite and specific statement about what will happen whereas Forecast is a probabilistic statement about future.
Also there is a very nice discussion about risk versus uncertainty with Risk being a quantifiable representation of possible outcome whereas Uncertainty is unquantifiable. “Risk greases the wheels of free-market economy; uncertainty grinds them to a halt.”
Then Silver goes to review multiple areas of prediction and forecasting dedicating a separate chapter to many of them:
• Political forecasting with heavy accent of its failures as practiced by pundits and experts and documented by Tetlock in his book.
• Sports forecasting
• Weather Forecasting
• Earthquakes Forecasting
• Even Poker game review as exercise in forecasting and prediction
Out of all these reviews comes out a number of rules that makes predictions and forecasting more or less viable:
1. Think probabilistically
2. Keep changing forecast as soon as new data come in – it is a dynamic process
3. Look for consensus – aggregate of forecasts is usually 15 to 20% more accurate then the individual ones.
4. Beware Magic Bullet forecast – too much certainty based on historical record could hurt
5. Weighting Qualitative information
6. Do everything possible to control for bias – objectivity is material for good forecast and is very difficult to maintain
7. Avoid overfitting – mistake of perception of noise for a signal. Often happens when correlation is taken for causation.
There is also quite nicely intersection with Taleb’s Antifragility and Randomness notions – a nice discussion of nonlinearity of the future and role Chaos theory (this nice little butterfly which can cause a huge hurricane due to nonlinearity of cause) effect sequences.
There is also an important discussion on Self-fulfilling and Self-cancelling predictions that provides a good reason to try taking into account the impact of prediction itself. Everything that we do has impact on the future events.
Silver also goes into nice set of details about Bayesian statistical methods and successes and failures of computer models and overall computer versus human issues in prediction and / or forecast development.
Overall a very useful book for me.