“Machines for rushing to judgement”
Ben Broadbent, member of the Monetary Policy Committee of the Bank of England and former Goldman Sachs economist, gave a fine talk on “forecast errors” at the Mile End Group at Queen Mary, University of London, last night. The text has lots of great graphs in it, including this one of difference between the MPC’s own forecasts of GDP growth one year before and what actually happened.
It shows that the MPC has been too optimistic about GDP for the last three years running, but three years is far too soon to tell if this is a persistent bias. (Broadbent cited studies showing that people asked to mimic a random sequence of heads and tails fail to realise how often sequences of three, four or more heads or tails in a row would actually occur.)
We have 15 years of consistent forecasts of GDP growth, he said, but that is not enough data to say with 95 per cent certainty that any one of 20 forecasters is better than any other. (That is not actually in the text of his speech, although he implies it at the bottom of page 6 and top of page 7.)
Indeed, “it is a bit odd”, as Broadbent said, that the average of all forecasts is a better predictor (although not by much) of GDP growth than any individual organisation’s forecast. Which is why he was asked afterwards why the Office for Budget Responsibility should not simply use the average rather than going to the trouble of making its own forecast. (His answer was that the OBR has to forecast lots of other things as well, such as tax receipts.)
The main points of the speech were these:
First, at least under some circumstances, distinguishing between economic models, and their forecasting performance, can take a long time. The greater the degree of true randomness in the world, the rarer the event being modelled and the stronger one’s prior belief about the structure of the economy, the longer it takes to be confident that any given forecast process is flawed.
Yet – and this is the second point – we are, all of us, genetically under-endowed with the patience it can require to make these distinctions. We are instead, as the psychologist Daniel Kahneman puts it, “machines for rushing to judgement”, biased judgement at that. We are naturally too inclined to see structure in what is actually random. We are also too inclined to view a forecast “error” as precisely that: someone’s mistake.
He also made an interesting point – in answer to questions rather than in the text – about whether something fundamental had changed in the economy in 2008. He said that GDP and employment had tracked each other closely for a long time, but had started to diverge five years ago. At that point GDP fell but employment fell by less than would have been expected and has since risen by more than would have been expected, while GDP has been more or less flat. (The same discontinuity can be seen in this new graph of under-employment by Bell and Blanchflower.) There is a one in 500 chance that this divergence is mere random variation, Broadbent said.
The trouble is that, although we can be pretty sure that something structural has changed, we cannot know whether that change will persist.Tagged in: economics
Latest from Independent journalists on Twitter