Is macroeconomics a science?

Macroeconomics can get a pretty bad rap at times, perhaps unfairly. Some of its practitioners are so politically influential on such familiar topics as unemployment and economic growth it's easy for the non-expert with an opinion to get a bit jealous. Few would dispute the merits of the latest winner of the Nobel in physics. But the Higgs boson is pretty inscrutable even to most physicists. It's only natural that economists get more attention -- and criticism -- when Nobel prize winners like Paul Krugman write popular columns for the New York Times.

Yet, even the noted economist Paul Romer has offered the caustic remark that the field is in "...a general failure mode of science that is triggered when respect for highly regarded leaders evolves into a deference to authority that displaces objective fact from its position as the ultimate determinant of scientific truth."

Ouch. So maybe macroeconomists are our modern day equivalent of medieval High Priests. Economists' theoretical models didn't predict the economic crash of 2008. Nonetheless, economists don't seem particularly troubled, certainly not troubled enough to consider that their models might be profoundly off course. From their perch, why should they?

Confirmation bias -- seeing only that which supports existing beliefs -- can be brushed off as the sort of normal human arrogance that we are all susceptible to. But being able to falsify a result lies at the core of the scientific method. It must be possible to set up a test that could lead to a model being discarded. 

For a comparison of professions, imagine if meteorologists predicted sunny days rather than the landfall of a hurricane. And then, because respected NASA scientist James Hansen was himself unconcerned, they put little effort into preventing such a thing from happening again.

That's not what happens. Instead, in meteorology, the validity of forecast models is constantly tested by performing what is known as "hindcasts" -- starting a model sometime in the past to see how well it predicts the present. Aside from the fact that the models are built on basic physics to the greatest extent possible, various model flavors are ranked according to their hindcast accuracy. It's the job of a professional meteorologist to both understand the model workings and know which models do best in which situations to communicate to the public the best forecast possible.

I can find no evidence of the economics profession doing something similar. Traditional macroeconomic models employ equations for the GDP, or “production functions”, that are  “tuned” to match past observations of labor and capital. It is not possible to falsify these moving theoretical targets because they are always made “right” by adding layers of social complexity or by tweaking the production function exponents until a decent fit is obtained. If conditions change and the formula no longer works, economists just tune again and call it a “structural break”.

This is cheating! At least if the goal is understanding how things work. It would be abhorrent to imagine a basic physics equation being adjusted as time progresses for the situation at hand. The speed of light in a vacuum doesn’t get to be different for you than for me or for last year versus this year.

Let's take for example the basic Cobb-Douglas production function used by economists as a starting point for relating economic production Y to labor L and capital K. The quantity A is a “total factor productivity” that has been thought -- largely due to efforts by Paul Romer -- to be related to innovation.

Y = A L^α Κ^1-α

Here the parameter α is tuned to past data in order to reproduce values of Y. In economic studies, when the inelegant Cobb-Douglas function (or whatever is used as a replacement) doesn’t work well, for whatever reason, the approach is not to ask whether or not something might be fundamentally wrong about the premise behind the fit, but rather to add ever more bells and whistles until once again a sufficient fit is obtained, totally independent of any consideration of dimensional self-consistency.

For example, maybe a constant exponent α doesn’t provide a good fit unless A is allowed to change too according some equally complex function. Paul Romer introduced government stimulus of R&D to obtain this sort of example of complexity:

So many free parameters! With such a complex function one could replace labor with the historical population of rodents in Calcutta and tune A, α and β in such a manner that the Cobb-Douglas function would still reproduce beautiful timelines for Y. As John Von Neumann quipped With four parameters I can fit an elephant, and with five I can make him wiggle his trunk.

This is not what sophistication should look like! Making things ever more mathematically complex does not make things more true, if anything less so. It feels akin to astrology, a highly complex, self-consistent model based on un-physical nonsense. Totally convincing to those who are looking to believe that the world has order and explanation, and that they alone have the years of training required to understand it, but completely lacking in any means for falsifiability.

It gets worse. The production functions lack the simple element of dimensional self-consistency. Take a basic physics equation, Newton's F=ma, or Force equals mass times acceleration. Mass has units of mass, obviously, and acceleration has units of distance per time squared. So the units of force are mass times distance per time squared. The equation would be totally bogus if force were declared to have any other sorts of units.

Now compare Newton's F = ma with the Cobb-Douglas function. There is nothing fundamental about the free parameter α since it is just a number. In fact, it can have any value depending on the statistical fit, the country, or the period considered. Suppose for the moment that α = 0.3. If A is just a number, labor has units of worker hours, and capital units of dollars, then Y would necessarily have the absurd units of worker hours to the 0.3 power and dollars to the 0.7 power. This has nothing to do with the real units of economic output which are dollars per time!

A couple years ago I had the opportunity to discuss economic growth models with well-known environmental economist Robert Ayres on a visit to Paris where he lives. He was quite adamant that I was wrong about everything. I don't think he had actually bothered to read anything I had done, which was too bad given the condition for the meeting (his idea) was that I buy and read his latest book. I tried to be patient, but eventually raised this units issue with him. His response was "only a physicist would care about units"!

Perhaps, I have been too harsh -- everybody is trying their best -- but it looks like fluency in Latin in the Catholic Church, where established macro-economists need something sufficiently opaque in order to maintain their high-priesthood. More generously, economics is complicated and economists just don’t yet know yet how to describe it without such detailed dimensionally inconsistent fits; even in physics, similar fits are occasionally used to describe interactions of particles with turbulence, for example, simply because the underlying physics can be rather challenging.

And maybe my rant is just another one of those pot-shots from non-economists, I have however tried to do better, by creating an economic growth model with no bells and whistles that can be easily tested and discarded. It is founded on a proposed constant relationship between energy consumption rates and a very general representation of total inflation-adjusted wealth (analogous to capital K) and is borne out by observations. Further evaluation of the model has been done by performing hindcasts, asking whether we predict the present with a deterministic model that is initialized at some point in the past. Again, in this case it appears we can: current global rates of energy consumption growth and GWP growth can be accurately predicted based on conditions observed in the 1950s, without appealing to any observations in the interim, with skill scores >90%.

For myself, there's adequate contentment in simply understanding some of the power of thermodynamics. But that is balanced by some abhorrence with certain aspects of macroeconomics.