How to create a good economic model?

Modern economists use a range of statistical methods, ranging from extremely complex ones to simple displays of historical data. It is generally accepted that through statistical correlation, historical data can be transformed into useful information, which in turn can serve as a basis for evaluating the economy as a whole.

Unfortunately, this is not as easy as it seems. For example, there is evidence that a decline in the unemployment rate leads to a general increase in the prices of goods and services. Should we conclude from this that a decline in unemployment is a major cause of price inflation? There is also evidence that price inflation is related to changes in the money supply. It has also been found that changes in wage levels and price inflation are highly correlated.

How do we judge which economic model is correct?

We are faced with not one but three competing “theories” of inflation. How do we decide which one is right? According to the conventional wisdom, the criterion for choosing a theory should be its predictive power. On this issue, Milton Friedman wrote:

"The ultimate goal of positive science is the creation of a theory or hypothesis that provides valid and meaningful predictions about phenomena that have not yet been observed" [1]

If the model (theory) "works", then it is considered valid. The moment it is refuted, we proceed to a new one. For example, the economist believes that consumers' spending on goods and services is determined by their disposable income. Once this hypothesis is confirmed by statistical methods, it is used as a tool for assessing the future direction of consumer spending. If the model does not provide accurate forecasts, it is either replaced or changed.

It de facto does not matter what the basic assumptions in the model itself are, as long as it can make good predictions. According to Friedman:

“The relevant question to ask about the assumptions of a theory is not whether they are realistic, because they never are, but whether they come close enough to the intended goal. The answer can only be given after seeing whether the theory works, which means whether it makes sufficiently accurate predictions.” [2]

The conventional wisdom that a model's predictive ability is a criterion for its acceptance is questionable. Even the natural scientists that mainstream economics tries to follow do not validate their models in this way. For example, a theory used to build a rocket predicts certain conditions that must be present for it to be successful. One of the conditions is favorable weather. Should we judge the quality of a rocket propulsion theory based on whether it can accurately predict the launch date of the rocket?

The quality of a model does not depend on whether it can predict the future

The assumption that a launch will take place on a certain date in the future will only happen if all the conditions hold. We have no way of knowing in advance whether this will be the case. For example, it may rain on the planned launch day. All that the theory of rocket propulsion can tell us is that if all the necessary conditions hold, then it will be successful. The quality of the theory is not impaired by the impossibility of making an accurate prediction of the launch date.

The same logic applies in economics. We can confidently state that, other things being equal, an increase in the demand for bread will increase its price. This conclusion is a priori true. The answer to the question of when exactly it will rise – tomorrow or some day after that – cannot be given by the theory of supply and demand. Should we then reject the theory as useless because it cannot predict the future price of bread?

Consider a situation where the stock market has been in a bullish trend for several years. As a result, an analyst has found that it is possible to get better returns by actively investing on the principle of dog barking. If the dog barks three times, it is a sign to buy, and if it barks once, it is a sign to sell. This is entirely possible - statisticians know more absurd statistical correlations. But should such a hypothesis be accepted as a valid theory because it has good predictions?

Contrary to popular belief, the criterion for choosing a model is not how well it has worked in the past, but whether it is theoretically correct.

1. Milton Friedman, Essays in Positive Economics, Chicago: University of Chicago Press, 1953.
2. There again.

The article was originally published on the author's blog.

Did you like it? Take a minute to support the EKIP on Patreon!