Recursive residuals
Posted: Sun Jan 26, 2020 3:48 am
Hi,
I have a question on recursive estimation, concerning residuals and coefficients.
The example concerns a simple OLS with two independent variables (no intercept).
In recursive residuals estimation, y_t-x_t*beta_{t-1} is used according to the documentation (equation 25.42 at http://www.eviews.com/help/helpintro.ht ... stics.html and the text above the equation) . Since two observations are needed for beta, the first recursive residual that can be calculated is for t=3, i.e. the third observation. Yet, when I "Save results as series" in calculation, there is a recursive residual for t=2, while I identical residuals from t=3 and onwards can be manually replicated (in Matlab). Am I missing something central here?
Further, in recursive coefficient estimation, I cannot replicate coefficient estimates by manually running the regressions with expanding samples. Is there a rescaling or something similar going on in the recursive calculations? Further, something seems to be interchanged in the labels of the recursive estimates when output as new series.
Best,
Krille
I have a question on recursive estimation, concerning residuals and coefficients.
The example concerns a simple OLS with two independent variables (no intercept).
In recursive residuals estimation, y_t-x_t*beta_{t-1} is used according to the documentation (equation 25.42 at http://www.eviews.com/help/helpintro.ht ... stics.html and the text above the equation) . Since two observations are needed for beta, the first recursive residual that can be calculated is for t=3, i.e. the third observation. Yet, when I "Save results as series" in calculation, there is a recursive residual for t=2, while I identical residuals from t=3 and onwards can be manually replicated (in Matlab). Am I missing something central here?
Further, in recursive coefficient estimation, I cannot replicate coefficient estimates by manually running the regressions with expanding samples. Is there a rescaling or something similar going on in the recursive calculations? Further, something seems to be interchanged in the labels of the recursive estimates when output as new series.
Best,
Krille