Predicting the Future

The future is open and uncertain. Philosophers have devoted little attention to this realisation, so banal does it seem. However, according to Immanuel Kant, people do have free will; you may dictate in advance what an individual should do, but you cannot predict what they actually will do (see "The Conflict of the Faculties", 1794, Second Section; I thank Otfried Höffe for the reference). In short, the future is not predictable. There are of course laws of nature, such as gravity. If you throw a ball, for example, you can be fairly certain that it will fall to the ground – unless someone intercepts it (volleyball players, of course, could write a book on this subject).

In this way, forecasters are expected to perform the impossible, quite in line with the saying that "the impossible will be accomplished immediately; miracles take a bit longer". Paradoxically, although many people know that the future is not predictable, they consult forecasts on a regular basis: about the weather, traffic, the economy – not to mention speculations of extremely varied origin, from the ancient prophecies of Delphi to the soothsayers of the Internet age. Prognoses made by futurologists concerning what will happen in ten or twenty years enjoy particular popularity, for the forecasting horizon is short enough to stir the public interest, yet long enough so that no one remembers whether the predictions were erronerous or correct.

Prognoses of a scientific nature deal with uncertainty by assigning probabilities to possible future events. Confidence bands and intervals are used to graphically demonstrate the uncertainty associated with a forecast. In television weather forecasts, for example, shaded bands show the temperature in coming days. As the projection extends into the future, the bands spread farther apart. Economic forecasts work in a similar way. Yet as useful as confidence bands may be, their predictive power is quickly exhausted. The informational value of a forecast that annual GDP will grow between one and five per cent is absurdly low, even if it’s supported by hard data. Journalists, of course, are not big fans of confidence intervals, and regularly simplify forecasted figures. A prognosis that growth will be between "1.5 and 2%", for example, is simply reported as "1.75%" – thus suggesting an even greater degree of accuracy than if the prognosis were 1.8%. Even more unpopular among journalists is a growth forecast of 1 ¾% – which indicates a confidence interval of 1.65 to 1.84%. This method of notation is not well known outside of the field of statistics, and when it reaches the newspaper’s typesetting department, it’s invariably transcribed into 1.75%.

Forecasts are snapshots of a given moment in the economy that attempt to assess all available and relevant information. New developments can often lead an original forecast to be revised, and, in extreme circumstances, may invalidate it completely, particularly when relevant historical experience is lacking – as has been the case with the current financial and economic crisis. Such revisions are not motivated by a desire among economists to play games. In other situations, an economic forecast – e.g. of an impending recession – may confirm or disprove itself depending on the policy actions that are taken.

The foregoing remarks should be considered when the next wave of economic forecasts rolls over the land this autumn. Such forecasts are required for planning purposes, as firms and public authorities can scarcely afford to work from day to day without a thought to the future. The tremendous scientific effort that goes into generating these forecasts is deserving of our respect. Yet one should take them with a pinch of salt, for they seek to achieve something impossible, perhaps even miraculous. And as mentioned, achieving the miraculous is not such a simple matter.