Friday, May 7, 2021

Why (Again) Economics Cannot Be Considered As A Bona Fide Science.

 As debate continues to rage in the financial press (e.g. FT yesterday) on whether inflation has returned,  I am reminded of a WSJ piece some ten years ago entitled: 'Is the Dismal Science Really a Science?' One of the knocks noted against "the dismal science" being : "theories that were once discredited surge back into favor". This seldom if ever happens in the rigorously quantified sciences, such as physics, astrophysics and plasma physics. The Big Bang - having been vindicated via the discovery of the 2.73 K microwave background radiation-   is not going to disappear all of a sudden and be replaced by the steady-state theory again. And I can't see us chucking out Einstein's General theory of relativity to use Newton's theory of gravitation exclusively (of course, the latter remains useful for relatively small space-time domains).

My objections to economics being referred to as a bona fide science are bound to be criticized by the quants (mainly former physicists, or former physics grads) who work on Wall Street.  But that doesn't mean my objections are wrong.  To me, Economics has never been a science, it only adopts some scientific window dressing, a few of the mathematical methods (mostly statistics) and some jargon. But no serious empiricist would regard it as "science" - by which I mean it hits the normal criteria for being identified so.  

In particular,  the powerful tool we use in actual science is known as successive approximations. You have preliminary information (say from earlier research efforts) that leads to the above hypothesis 'X'. You then continue the process, gathering more data, and accessory information which leads to some predictive result which yields a more refined hypothesis:

X' = x (n+1) = x + P(x)

where x(n +1) denotes an improvement via iteration, with P(x) the predictive confirmation (of x(n +1) from x) that allows it. Later, after more refined data become available:

x(n + 2) = x(n + 1) + P'(x + 1)

Now, what if at this stage of testing we have:

x(n + 2) = x(n + 1) + P'(x + 1) + E(x + 1)

where E(x + 1) is an ERROR term.  And:

E(x+ 1) > > P'(x + 1)

Then we have a falsification introduced which must be corrected.  This is why we say that science is a self-correcting process. For example, the original Big Bang theory - because of new data - had to be corrected to incorporate cosmic inflation.  This has also enabled more accurate predictions.  By contrast, modern economic theory missed the boat in predicting the 2008-09 financial credit meltdown and crisis.  Why?  Because the theory did not take into account the credit default swaps (CDS) buried in a host of financial instruments including bonds - which were then given false credit ratings, see e.g.

Even now - today -   the majority of economists cannot agree on whether inflation is being fueled by spending, e,g, Covid relief spending, or not.  Much of this has to do with there being no standard definition of inflation.  So the Fed will always disagree with mom and pop citizens, for example, who are seeing it in their grocery purchases, or meds or rents.   This arises because of the Fed definition and use of 
"core inflation" (omitting costs of food, health care, housing). This one allows political interests to deliberately lowball the inflation rate in order to keep their pet numbers jiggered in the right way - especially so that the powers-that-be don't have to cough up too much in Social Security COLAs!

 Even in the realm of  job statistics economists  tend to fudge  based on political expediency. For example the BLS stat for unemployment is intentionally lowballed by dropping workers off the rolls who are still unemployed after 6 months. Quite a tidy trick that! It keeps the unemployment rate below 6% most of the time, unless wide swatches of workers are let go at once-   as with the Covid disruption, lockdowns last year. By contrast, Barbados and most of the EU (European Union) count workers as unemployed so long as they really are. They aren't working or producing so they're unemployed - not "discouraged". This latter is an artificial category concocted by the 1994 (mainly Repug) congress to conceal more than inform. Yet Economists employ it.

Nor does Economics make quality predictions based on said models, say like atomic physics. (Prediction of specific spectra - emission or absorption- using the Schrodinger equation, and the correct energy eigenvalues for the atom chosen). If it had such capabilities, it would have seen the 2008 credit meltdown coming well in advance and warned everyone!

Granted, economists do have a pale imitation of science, and who knows - maybe in a few hundred years they'll join the full club- after their objects of inquiry and methodology acquire greater maturity. When we see their first bona fide model, as opposed to a statistical macguffin that they inaccurately call a "model".

Let me get another point clear:  even the most rudimentary student of statistics understands that distributions, linear regressions etc. may show correlations, not causation. Because you have correlation, say between high subsidy Section 8 areas and crime does not imply the latter is caused by the former. The confusion of correlation with causation  is the single most recurrent contributor to errors in economic models. And thus mistaking economics for any kind of "Science". (Since models must be founded on seeking out and exposing causation, not merely correlation or coincidences)

Let me give some concrete examples, from my field, of solar physics. In two 1984 papers, one in Solar Physics, the other in The Journal of the Royal Astronomical Society of Canada I provided the basis to show how a solar flare trigger is feasible, and the theoretical conditions that might apply. In the former paper this was based on using a pure statistical model, with Gaussian (normal) distributions of flares, and multiple regression analyses. In the second paper, it was based on a consistent model that incorporated actual, physical features and observations.

What I did is to show how my mechanism, embodying a flare trigger depending on multiple agents (twisted fibrils, deformed magnetic polarities, magnetic gradients > 0.5 Gauss/cm, sunspot areas > 15000 msh or millionths of a solar hemisphere) , caused flares via a magnetic shear process. And thence, I showed which agents gave rise to which class flares. The final computer model allowed one to calculate the resident magnetic free energy, the generating twist angle of fibrils, the magnetic intensity and other physical indices directly from the quantities and mathematics of the model. (I used several partial differential equations, in tandem with Green's functions - and used Laplace transforms to obtain solutions to the physical equations) Statistics can crudely describe - in a correlative fashion- some primitive aspects of an empirical system.

For example, as in my first referenced paper (noted above) I wanted to obtain the correlation between the incidence (frequency) of a certain type of solar flare (generating importance 2+ sudden ionospheric disturbances) and delta class sunspot groups. I did a linear regression to find a fit for F (flare frequency) vs. T(S) type of sunspot group. I obtain a statistical function of the form:

F = T(s) + C

Where C is some quantity that intersects the particular (y-axis), in this case the one for F. This is a statistical representation of an empirical relationship. However, and this is important, the statistical relationship (F = T(s)) + C) tells me nada - nothing, about the underlying physics that produces flares!

The mathematical model for that might be written(to start):

Curl B = Ñ X B  =  aB  = (m
  / ) B      (Force-free field)

where 'curl' is a complex mathematical operation using partial derivatives, and it is applied to B - the magnetic induction. Or strength of the magnetic field in the sunspot, or group. (We measure B directly using magnetometers, which produce vector magnetographs for specific regions, sunspot groups) Further working through this model yields:

Ñ (B) = (a) B

Ñ   (DIV)' is another math operator called the "divergence". From this, we can obtain the partial differential equation for the magnetic field (B - which applies to the sunspots in x, y and z dimensions):

 xx  + B yy   + B zz   + (a) B = 0

which can then be solved - with the various quantities (
a, B and currents obtained from these) compared to the actual observed properties in the sunspots. This is distinct from any statistics - as it uses and incorporates actual physical properties of a particular model. (Given exclusive use statistics omits all these details.)  More to the point, as my own research has disclosed, we can accurately trace variations in the Earth’s magnetosphere to plasma eruptions and changes on the Sun proper. For example, registered in terms of species of protons captured at the Earth's magnetosphere, and the variable intensity - in watts per square meter- of soft x-ray fluxes.

In plasma tokamak devices, we can actually test solar flare theories directly via the kink and other induced instabilities - that shed direct light on our models and expected behavior.   Can the economists do the same with existing credit default swaps still floating around, say in auto loans, or student loans? I doubt it, since if they have they would have by now.  (And for which we'd be seeing greater oversight of all derivatives used in trading)

Meanwhile, the illustrious economists - in their naive and incomplete statistical "modeling" -are prepared to ignore (as in dismiss from consideration) an entire raft of variables associated with what they call "externalities". These glaring and inexcusable ecosystem omissions (as global monetary values), were assayed for one particular year in the study Putting a Price Tag on Nature's BountyScience, Vol. 276, p. 1029).   

If economics was a true science it would have replaced GDP by now with Prof Herman Daly's Index of Sustainable  Welfare.   The reason is that the latter index factors in most externalities. The irony in all this is if economics ever did evolve to become an actual science we'd behold a quantum leap in rational outcomes,  starting with getting rid of "trickle down" or supply side formulas once and for all.  We would also surely see the end of the Pareto efficiency model and its distribution, i.e.

With these strides we'd almost certainly see better predictions as well. 

No comments: