The UK’s national weather service, the Met Office, is to get a £1.2bn computer to help with its forecasting activities. That is a lot of silicon. My instinctive response was: when do we economists get one?
People may grumble about the weather forecast, but in many places we take its accuracy for granted. When we ask our phones about tomorrow’s weather, we act as though we are gazing through a window into the future. Nobody treats the latest forecasts from the Bank of England or the IMF as a window into anything.
That is partly because politics gets in the way. On the issue of Brexit, for example, extreme forecasts from partisans attracted attention, while independent mainstream forecasters have proved to be pretty much on the money. Few people stopped to praise the economic bean-counters.
Economists might also protest that nobody asks them to forecast economic activity tomorrow or even next week; they are asked to describe the prospects for the next year or so. True, some almanacs offer long-range weather forecasts based on methods that are secret, arcane, or both — but the professionals regard such attempts as laughable.
Enough excuses; economists deserve few prizes for prediction. Prakash Loungani of the IMF has conducted several reviews of mainstream forecasts, finding them dismally likely to miss recessions. Economists are not very good at seeing into the future — to the extent that most argue forecasting is simply none of their business. The weather forecasters are good, and getting better all the time. Could we economists do as well with a couple of billion dollars’ worth of kit, or is something else lacking?
The question seemed worth exploring to me, so I picked up Andrew Blum’s recent book, The Weather Machine, to understand what meteorologists actually do and how they do it. I realised quickly that a weather forecast is intimately connected to a map in a way that an economic forecast is not.
Without wishing to oversimplify the remarkable science of meteorology, one part of the game is straightforward: if it’s raining to the west of you and the wind is blowing from the west, you can expect rain soon. Weather forecasts begin with weather observations: the more observations, the better.
In the 1850s, the Smithsonian Institution in Washington DC used reports from telegraph operators to patch together local downpours into a national weather map. More than a century and a half later, economists still lack high-definition, high-frequency maps of the economic weather, although we are starting to see how they might be possible, tapping into data from satellites and digital payments. An example is an attempt — published in 2012 — by a large team of economists to build a simulation of the Washington DC housing market as a complex system. It seems a long way from a full understanding of the economy, but then the Smithsonian’s paper map was a long way from a proper weather forecast, too.
Weather forecasters could argue that they have a better theory of atmospheric conditions than economists have of the economy. It was all sketched out in 1904 by the Norwegian mathematician Vilhelm Bjerknes, who published “The problem of weather prediction”, an academic paper describing the circulation of masses of air. If you knew the density, pressure, temperature, humidity and the velocity of the air in three dimensions, and plugged the results into Bjerknes’s formulas, you would be on the way to a respectable weather forecast — if only you could solve those computationally-demanding equations. The processing power to do so was to arrive many decades later.
The missing pieces, then: much better, more detailed and more frequent data. Better theory too, perhaps — although it is striking that many critiques of the economic mainstream seem to have little interest in high-resolution, high frequency data. Instead, they propose replacing one broad theory with another broad theory: the latest one I have seen emphasises “the energy cost of energy”. I am not sure that is the path to progress.
The weather forecasters have another advantage: a habit of relentless improvement in the face of frequent feedback. Every morning’s forecast is a hypothesis to be tested. Every evening that hypothesis has been confirmed or refuted. If the economy offered similar daily lessons, economists might be quicker to learn. All these elements are linked. If we had more detailed data we might formulate more detailed theories, building an economic map from the bottom up rather than from the top down. And if we had more frequent feedback, we could test theories more often, making economics more empirical and less ideological.
And yet — does anyone really want to spend a billion pounds on an economic simulation that will accurately predict the economic weather next week? Perhaps the limitations of economic forecasting reflect the limitations of the economics profession. Or perhaps the problem really is intractable.
Written for and first published in the Financial Times on 21 February 2020.