Home > Uncategorized > Econometrics — a matter of BELIEF and FAITH

## Econometrics — a matter of BELIEF and FAITH

from Lars Syll Everybody who takes regression analysis course, studies the assumptions of regression model. But nobody knows why, because after reading about the axioms, they are rarely mentioned. But the assumptions are important, because if any one assumption is wrong, the regression is not valid, and the interpretations can be completely wrong. In order to have a valid regression model, you must have right regressors, the right functional form, all the regressors must be exogenous, regression parameters should not change over time, regression residuals should be independent and have mean zero, and many other things as well. There are so many assumptions that it is impossible to test all of them. This means that interpreting a regression model is always a matter of FAITH – we must BELIEVE, without having any empirical evidence, that our model is the ONE TRUE VALID model. It is only under this assumption that our interpretations of regression models are valid …

Nonsense Regressions: If a regression model OMITS a significant regressor then it is INVALID; we may call such regressions “nonsense regressions”.

This formulation highlights the major mistake in modelling that is common. The regressors which are EXCLUDED by a regression model are just as important as the ones that are included. Thus the simple model C not only states that FDI determines GDP, it also states that no other variable has any effect on GDP, since no other variable is included in the model. It is this exclusion which is seriously questionable.

1. February 17, 2020 at 3:10 am

Econometrics, including regression method, produces tentative outcomes, this should be known by a learner at his/her beginning stage. To some extent, a tentative outcome can be deemed “subjective”, “believable”, or “faithful” despite using “objective” data and “objective” method, just like the data “saying” “subjectively” as a person. This inspires us what is “subjectivity”, and what is “objectivity”. The Algorithmic approach has exactly answered the question. However, in my opinion, econometrics should not be criticized in this way, although this is an extremely inductive method, it is honest, transparent, never boasting about something it has no. For example, Newton and Einstein’s physical theories are all extracted by the way similar to econometrics, thus we can also criticize that they did not know why their formulas appears in that way. But, the critics as such would be meaningless. This is exact the true face of science, the criticism on this just reflects philosophical ignorance.

2. February 17, 2020 at 1:27 pm

You are quite correct in your comments on Newton and Einstein. To find meaning the forms of equation used must be restricted by the requirements of the quantity calculus. If they are not the fitted equations are just another arbitrary equation. My analysis of abstract production theory surprised me in that the maintenance effort introduced itself as being necessary to satisfy the quality calculus. My early experiments in the computer simulation of production did not consider maintenance as a separate entity but demonstrated regularities which indicated an analytical solution was possible. It was. Hogan mentioned the possibility of comparing capital and depreciated capital using Solow’s method. Solow dismissed the idea as being of little importance. I wonder if the differences had been examined, would econometrics have delivered abstract production theory in its stark simplicity.
References:

Hogan, W. P. (1958). ‘Technical progress and production functions’. In: The Review of Economics and Statistics 40, pp. 407–411.

Solow, R. M. (1958). ‘Technical progress and production functions: reply’. In: The Review of Economics and Statistics 40(4), pp. 411–413. url: http://www.jstor.org/stable/1926346.

3. February 17, 2020 at 1:31 pm

Similar to BinLi, aren’t all regression equations properly presented with an error term? If so, this is the ‘fudge factor.’ This is what enables an investigator to claim both that the chosen variables are most important, and that whatever was missed that might be important is captured by the error term. So it is not completely correct to say that some variables are excluded, i.e that only the included variables have an effect. Their effects are captured by the error term, and this has a tremendous impact on how an investigator interprets the results of a regression model.

4. February 17, 2020 at 5:49 pm

Economic modeling should always be BOTH a mentally integrative (inductive/deductive) process AND and an hierarchical one as well so far as primary-ness/essence of cause is concerned. That is to take the paradigmatic viewpoint on analysis.

It should also be willing and able to consider that exogenous/obscurantist factors like parasitism could be skewing the modeling. “Emergent” qualities are sometimes the result of not giving this consideration its due. To paraphrase Sam Clemens: ‘What gets us into trouble is not ONLY what we don’t know. It’s ALSO what we know for sure that just ain’t so.’