Economic Forecasting and Science

Paul Samuelson

[An abridgement, reprinted from: Michigan Quarterly Review, October 1965]

If prediction is the ultimate aim of all science, then we forecasters ought to award ourselves the palm for accomplishment, bravery, or rashness. We are the end-product of Darwinian evolution and the payoff for all scientific study of economics. Around the country, students are grinding out doctoral theses by the use of history, statistics, and theory -- merely to enable us to reduce the error in our next year's forecast of GNP from $10 billion down to $9 billin.

Actually, though, I am not sure that the ultimate end of science is prediction -- at least in the sense of unconditional prediction about what is likely to happen at a specified future date. ...Some of our best economists seem to be fairly poor forecasters of the future. ...

A good scientists should be good at some kind of prediction. But it need not be flat prediction about future events. ...If he is a master scientist, his hunches about experiments never yet performed may be very good ones.

Similarly, a good economist has good judgment about economic reality. To have good judgment means you are able to make good judgements -- good predictions about what will happen under certain specified conditions. This is different from having a model that is pretty good at doing mechanical extrapolation of this year's trends of GNP to arrive at respectable guesses of next year's GNP. ...

This means that the master economist must peice together, from all the experience he has ever had, hunches relevant to the question at hand. ...

...The essence of science is mastering the art of filtering out the obsolete patterns of the past and filtering in the patterns of persistence.

As an economic scientist, I take economic forecasting with deadly seriousness. I hate to be far wrong. Every residual is a wound on my body. ...The reason is not vanity --because forecasting serves a purpose: each dollar of error costs something in terms of corporate or national policy; and if the "loss function" or "social welfare function" is a smooth one in the neighborhood of its optimum, it will be the square of the error of forecast that gives a rough measure of local error.

"The Puzzling Failure of Ecoomics" has a lead article inside that says, "I don't care who writes a nation's laws ... if I can write its economics." Paul Samuelson, The Economist, 25 August 1997]