Nate Silver’s “The Signal and the Noise”

Overall this is a good book. By and large, this book has little math but a great deal of common sense. The points are easy to grasp. But I think that Silver has conveyed little of what makes him a good forecaster. Silver is a good forecaster because he is unusually smart in a broad sense. You can’t write down smart. What he does write down is very useful and sometimes fun to read. He is clearly right about many reasons for bad predictions.

The Kindle ‘locations’ go up to 11,820 for this book. The end-notes begin at 7661. Page numbers may be proportional.

As I read the book:

A quibble already in the introduction:

Perhaps ‘technology’ is what people make but earlier information mechanisms might be thought to start with:
  1. Cilia on bacteria which seem to have evolved, in part, to sense the environment,
  2. Nerves which are cells to move information about in a multicellular organism,
  3. Synapses by which nerves influence other nerves, and provide boolean logic,
  4. Nerves which respond to external patterns,
  5. Concept nerve clusters depicting external and internal patterns,
  6. Language which conveys concepts to conspecifics,
  7. Written language,
  8. Gutenberg.
I like Silver’s take on the biblical quote containing “nothing new under the Sun”—the idea that the stuff had been forgotten and thus the sense of lack of progress. In that light the quote seems an important and elegantly phrased observation.

Silver says:

I have not read his references but I have a distaste for his negative take on this. It sounds so much like what Kauffmann describes as expansion into the ‘adjacent possible’. It is a burst of cultural evolution much like the burst of species during the centrifugal Cambrian explosion.

(@1100) I remark now that sometimes one does better if he has a model. Predicting weather from historical data is useful. You can do better with current data from afar, a computer and a model. Weather models improve as new physics is added. They improve even more as new sources of current data are added.

Silver’s weather tale is well told. He might have added that Richardson had been doing the right arithmetic, which is no mean feat. Almost all of his equations are right.

Silver suggests that you can look at a chart graphing Observed frequency (of rain) against Forecast probability. Perhaps you are to see how far from the y=x line the dots lie. Suppose that a forecaster’s every forecast was that the probability of rain was the long term probability of rain. He would be right on, except perhaps for climate change. Perhaps you must somehow include the gamut of forecasts in the figure of merit. No obvious figure of merit comes to my mind. If the format of a prediction is a forecast of how much rain there will be tomorrow, then the above cheat fails.

(@2046) Silver tells the Richardson story well. Just about Richardson’s only mistake was numerical stability which I learned of in 1950 from Professor Bohnenblust. Richardson was over optimistic on this as well. A factor of one quadrillion is very useful. I have seen Richardson’s book on the book shelves of several weather theorists.

(@3065) Silver, discussing economic forecasts, has not yet mentioned the thought that forecasters may have—“We are leading expectations.”. This might cause them to lie about what they think. They may think it their duty.

(@3188) Goodhart law: “Once policy makers begin to target a particular variable, it may begin to lose its value as an economic indicator.”. My image is of someone trying to move a gelatinous mess—you push here and here moves but the bulk stays where it was.

(@3839) Silver remarks that Weather prediction is one of the few complex and successful models. I think he is excluding orbit calculation on some implicit categorical basis. There are all sorts of ‘engineering’ calculations that are as good as weather. They are so good that we seldom call them predictions.

(@4243) The posterior calculation is 98.71%.

(@4458) In summary, an excellent description of Bayesian vs. Frequentist.

(@4479) Silver says that Poe said: “if this chess-playing machine were real, it must by definition play chess flawlessly; machines do not make computational errors.”. Silver does not buy this but it remains today a common error, and is part of Penrose’s peculiar notions of intelligence. At @4540 Silver points out how a machine won at chess by abandoning certainty.

This is the link in footnote 22.

Starting at about @4600 Silver tells the story of the last chess game Kasparov won against Deep Blue. It is very well told with an interesting twist at the end. I found the section on chess the most interesting of the whole book.

(@5340) Silver relates the poker term “fish” for poor players who lose the money that keeps the good players in the game. This reminds me of high frequency trading where, I think, there are a few winners and many losers and most of what the winners win is what the losers lose. It is a slightly positive sum game however.

Silver’s ‘Bayesland’ (@5577) is fascinating. He fails to make a strong point for his method, a point that is lost to many others who push prediction markets: “When you bet money and are consistently wrong, you are removed from the market and no longer contribute to ‘the wisdom of crowds’.”. Those who can ‘see the picture’ are encouraged to form and advertise more predictions. Hansen made this point and few others seem to have noticed it. It seems to me that that is the principle virtue of the method. Thru this mechanism the crowd gets smarter even if individuals don’t.

At @5694 he speaks of “irrational pricing”. If so then that is a business opportunity unless there are limits on the betting.

(@5950:6000) This section is an argument not to invest via investment funds.