Axiom of Correct Specification
A regression model has validity ONLY if it is specified correctly. That is, all the regressors are correctly specified, the functional form is correct and remains stable over time, the error terms have the required properties, and the exogeneity and endogeneity assumptions made are valid. This is a HUGE requirement. In practice, students IGNORE failure of this assumption. Even though the model in use can be PROVEN to be wrong, they pretend that the regression model is correct, and then interpret it accordingly. This leads to false and extremely misleading results and conclusions. The lecture attached below discusses this issue in detail. I will try to break up this lecture into smaller pieces and provide them here on this page.
This material is being copied from original website where it was placed earlier -- copy not complete yet: PIDEVCAsadZaman Econometrics Lectures
Video of Lecture 1: Introduction – Problems with Standard Econometric Methodology [On My Website) YouTube Version Now Available, see below
Original first lecture given to PIDE Faculty in Sep 2015 as part of mini-course on Applied Regresson Analysis - See YouTube: EM L01 Flaws in Conventional Econometric Methodology -- Original Lecture covers a lot of material, not discussed in later simplified versions, which focus on Choice of Regressors.
REVISED and SIMPLIFIED LECTURE now available from: http://bit.do/azreg
A regression model can be interpreted ONLY if it is correct -- that is, all the ASSUMPTIONS of the regression model are fulfilled. If any ONE of them fails, the regression is not valid, and the results are wrong. Consider explaining Economic Growth in Pakistan by -- Health Variables, IPR -- Intellectual Property Rights, FDI Foreign Direct Investment.
We have (actually) three papers --
P1: EG as function of Health
P2: EG as function of IPR
P3: EG as function of FDI
People think that ALL three models are correct and each one can be used -- P1 can be used to find effect of health, P2 to find effects of IPR and P3 to find effects of FDI.
ONLY ONE of the three can be correct at most. Because P1 SAYS that IPR and FDI are irrelevant (by excluding these regressors) and P2 says that Health and FDI are irrelevant and P3 says that Health and IPR are irrelevent
So these three models CONTRADICT each other.
Suppose that the true model is EG depends on X1 X2 X3 where these are all different variables. Then P1 P2 and P3 will ALL give wrong and misleading results -- the variable which is included acts as a proxy for the ones which are missing and picks up significance even if it is not at all important.
Even if EG is a function of all three Health IPR and FDI each of the regressions P1 P2 P3 will give HIGHLY MISLEADING results -- This because in P1 Health WILL ALSO PICK UP effects of IPR and FDI. -- if positively correlated it will seem THREE times as important as it really is.
DETAILS are available in the lecture write up attached below (written up by Ateequr Rahman
EM L01 Flaws in Conventional Econometric Methodology - Explains the Axiom of Correct Specification, and how it makes life difficult for econometricians
Choosing the Right Regressors - Islamic WorldView Blog Post explaining the Axiom of Correct Specification, how econometrics hides its flaws.