The British historian Niall Ferguson isn’t the first person to link John Maynard Keynes’ sexual preferences with his economic beliefs. Asked recently about Keynes’ famous remark that “in the long run we’re all dead”, Ferguson said that it was evidence of Keynes’ short-termism, and that it was a result of his being gay and having no children. There are various things wrong with this (and Ferguson apologised for the implication that gay people can’t be far-sighted), but the view has a pedigree. In his obituary of Keynes, the Austrian economist Joseph Schumpeter wrote: “He was childless and his philosophy of life was essentially a short-run philosophy.”
But Keynes’ views were perhaps subtler than his critics suggest, and the result of something other than his sexuality or childlessness.
Big Data is all the rage at the moment, and it promises to use information to help predict the future. Its pin-up is Nate Silver, the man who predicted the results of the 2012 presidential election in all 50 states – after doing so in 49 in the previous election. In his book The Signal and The Noise, Silver applies his predictive powers to climate change, economics, earthquakes, terrorism and (of course, being American) poker and baseball. Silver admits that in some areas more information hasn’t helped us forecast any better – earthquakes, for instance, are still hard to predict – but in general his position is that more information leads to better prediction.
Keynes, I think, would disagree. His “long-run” comment was a throwaway one, but reveals something about his outlook. He didn’t mean, I am sure, that he didn’t care about the future – nobody with nephews, nieces, godchildren, or indeed an imagination, would seriously say that. I think that he actually thought something subtler: that we can’t know enough about it to make any accurate predictions.
There are two reasons that you might believe the future is unpredictable. One is that you think you can’t get enough data, or the right data – as Silver might put it, that there is too much noise for us to isolate the signal. The second is the more radical one, which is that we can’t be sure the future will resemble the past sufficiently that things you learned in the past will still be relevant. In philosophy this is called the problem of induction, and Keynes was very familiar with it – he was friends with the philosopher FP Ramsey, one of the century’s most influential writers on the subject.
Studying induction can make you very sceptical about prediction. You soon come to understand that we observe lots of correlation, but know precious little about causation, and that all forecasts make the tacit assumption that conditions tomorrow will be the same as yesterday. Forecasts tell you a lot about how things were in the past, but there’s no guarantee they can tell us anything about the future.
Our view of the future, you might say, is fuzzy. So is there a fuzzy way or predicting it that might suit it better? There is, and it’s called experience. It’s only when theories get tried in practice that you find out what’s wrong with them. Silver refers to a paper that claimed two-thirds of the findings of medical trials can’t be replicated in real life. In those cases, data predicted outcomes that experience contradicted. Information is a proxy for experience, and often a poor one.
The bad news about experience is that it takes time to accumulate. The good news is that it is transferable. One of the most effective organisations for passing on the fruits of experience is the family. People who have grown up hearing about the history of a family, a town, a country, or a business, can have the benefit of other people’s experience. Experience is passed on through stories.
In his book Silver says some interesting things about the American and northern European outlook. He suggests that because of the legacy of Protestantism, people from those places value book-learning and data above other sorts of knowledge. It follows that such people would be inclined to believe that number-crunching and analytics will yield useful results, while they might undervalue fuzzier ways of getting knowledge, like story-telling.
You might say that the problem of induction still holds. But if you look at the successes of Silver-esque data-juggling you find that it is good at predicting results in what you might call “closed conditions” where the rules are stable and there can be no unexpected inputs, like poker or elections. In what you might call “open conditions”, like economics or predicting where terrorists will strike, where there are lots of variables and unknown unknowns, the problem of induction is heightened. In a poker game the rules aren’t going to change, while in economics they often do. In open conditions data is less effective, and fuzzy methods might be useful. Maybe, when it comes to seeing the future, grandad’s old stories can be better than newfangled numbers.