emilioalzamora1
Well-Known Member
Hi All, Hi David,
I would like to reach out with a perhaps very straightforward/easy question but I want to be reassured about the following:
Linearity of the OLS regression is not one of the Gauss-Markov conditions, however, non-linearity often happens in fields like labour economics: age vs. income (over many years your income will stagnate - let's say prior to have been awarded a degree and then income increases rapidly leading to a nonlinear relationship).
(This can be tested using Ramsey's RESET test).
We aware that the regression must be linear in the parameters (alpha and beta) but NOT necessarily in the variables.
In case I have a two variable regression with Y and X, let be Y (Disney) and X (S&P500) and use any sort of available return data for these two variables, it is most likely that beta ranges between 0 and 1 (> 1) and alpha is most probably not significantly different from 0.
However, if I randomly change one value of the time series to an exorbitantly high/low value, I will get a signifcantly negative/positive value for the intercept.
Does a negative value for the intercept after running a fitted regression already points towards non-linearity OR do I actually have to run the RESET test to be sure about non-linearity?
I would like to reach out with a perhaps very straightforward/easy question but I want to be reassured about the following:
Linearity of the OLS regression is not one of the Gauss-Markov conditions, however, non-linearity often happens in fields like labour economics: age vs. income (over many years your income will stagnate - let's say prior to have been awarded a degree and then income increases rapidly leading to a nonlinear relationship).
(This can be tested using Ramsey's RESET test).
We aware that the regression must be linear in the parameters (alpha and beta) but NOT necessarily in the variables.
In case I have a two variable regression with Y and X, let be Y (Disney) and X (S&P500) and use any sort of available return data for these two variables, it is most likely that beta ranges between 0 and 1 (> 1) and alpha is most probably not significantly different from 0.
However, if I randomly change one value of the time series to an exorbitantly high/low value, I will get a signifcantly negative/positive value for the intercept.
Does a negative value for the intercept after running a fitted regression already points towards non-linearity OR do I actually have to run the RESET test to be sure about non-linearity?
Last edited:
