November 12, 2015

Prediction does not work? Try retrodiction first

Comment on Lars Syll on ‘Macroeconomic forecasting’


You point out that “In New York State, Section 899 of the Code of Criminal Procedure provides that persons ‘Pretending to Forecast the Future’ shall be considered disorderly ... and liable to a fine of $250 and/or six months in prison.” (See intro)

According to this law we should see all Wall Street gurus in prison and some academic economists, too.

New York State law is in full accordance with science because “The future is unpredictable” (Feynman, 1992, p. 147). Feynman was a scientist and what he told the world is that science makes no predictions.

Hold on, is that not the whole purpose of science?

This is a slight misunderstanding that stems from astrology/astronomy. If, in some special cases, motion turns out to repeat periodically as with some comets, then the next occurrence can be predicted with high precision. This prediction, though, does not require the knowledge of the law of gravitation or other laws of motion. The Mayas and others were good at predicting astronomical events long before Newton. This theory-free predictions were based on meticulous observations over long time spans and a bit of extrapolation.

Apart from these periodically recurring phenomena, physics does not predict single historical events. The physicists’s predictions are of a different sort. For example: from the equivalence of energy and mass, his famous E=mc2, Einstein deduced that the path of light is deflected by large bodies. He calculated the deflection and asserted that it would be observable on occasion of solar eclipses. The rest of the story is well-known: “This eclipse was photographed from the expedition of Sir Arthur Eddington to the island of Principe. Positions of star images within the field near the Sun were used to test Albert Einstein’s prediction of the bending of light around the Sun from his general theory of relativity.” *

The point is that scientists use the word prediction in a quite different sense from everyday usage. And this leads to the paradox that while the future is ‘unpredictable’ certain aspects may be ‘predictable’ with high precision. Loosely speaking, the laws of physics allow for conditional predictions.

What does this mean for economics? As soon as we have economic laws, we are in the position to make conditional predictions. There is a snag here which is specific to economics because there is no such thing as laws of human behavior, yet there are structural laws of the economic system.

To make a long analysis short, this is the First Economic Law for the elementary consumption economy.

This fundamental economic law contains the interrelation between the consolidated business sector’s cost/profit situation rhoF, the real rhoX and the nominal rhoE side of the product market, and the income distribution rhoD. So, the interrelations between firm, market, and distribution are encapsuled in the formula.

This law allows for retrodiction (see Suppe, 1977, p. 621), that is, when we measure the ratios rho for past periods and insert them into the formula then the result must be 1 for all past periods. The formula holds in every single period from past to the future (2014, eq. (12)). In other words, we have a testable economic law. Test it twenty years hence and you will find out that it is true. Where, then, does the difficulty with prediction come in? The crucial point is that the variables that underlay the four rhos are random variables.

Roughly speaking, we can make a precise conditional prediction but the future conditions are not known. This allows for retrodiction but not for prediction except for one case. If we know/control three of the four ratios the forth can be predicted with absolute certainty — under the condition that the retrodiction tests have been successful.**

Conclusion: When economic theory is built upon a behavioral axiom like constrained optimization then no prediction of any sort will ever be possible, but when the theory is built upon objective structural axioms, then conditional prediction becomes possible.

Conditional predictions meet all scientific criteria and therefore the New York Code of Criminal Procedure is not applicable. It is still applicable, though, to the forecasts of standard economics which are derived from a logically and materially inconsistent theory, which in turn has been based on unacceptable green cheese behavioral assumptions.

Egmont Kakarot-Handtke

Feynman, R. P. (1992). The Character of Physical Law. London: Penguin.
Kakarot-Handtke, E. (2014). The Synthesis of Economic Law, Evolution, and History. SSRN Working Paper Series, 2500696: 1–22. URL
Suppe, F. (1977). Afterword–1977. In F. Suppe (Ed.), The Structure of Scientific Theories, pages 615–730. Urbana, IL, Chicago, IL: University of Illinois Press.

* See Wikipedia
** For an advanced application see the following equation which conditionally predicts the unemployment rate and is therefore useful for employment policy. For details see 'Keynes’s Employment Function and the Gratuitous Phillips Curve Disaster'.

Relates to 'Predictably confused'