Splitting Sample versus Threshold Model
Posted: Tue Jun 07, 2016 10:17 pm
Dear Tom,
I have a balanced panel data set of 61 economies from 1970 to 2013. The research involves two empirical methodologies: sample splitting approach and threshold modeling.
Equation (1) Y = B0 + B1*X + e (if D = 0)
Equation (2) Y = B2 + B3*X + e (if D = 1)
or
Equation (3) Y = A1 + A2*(1-D)*X + A3*D*X + e
where Y is the dependent variable, X is the regressor, and D represents a dummy variable. B1, B3, A2, and A3 are the parameters of interest. Theoretical prediction says B3 > B1 (sample splitting approach) or A3 > A2 (threshold modeling). My empirical result supports the prediction using the sample splitting approach, but the threshold approach fails to support the prediction (In fact, A2 > A3). I am wondering if there is something wrong with my code. Is there a reason that the two methods produce inconsistent results? Please advise. Thank you very much.
I have a balanced panel data set of 61 economies from 1970 to 2013. The research involves two empirical methodologies: sample splitting approach and threshold modeling.
Equation (1) Y = B0 + B1*X + e (if D = 0)
Equation (2) Y = B2 + B3*X + e (if D = 1)
or
Equation (3) Y = A1 + A2*(1-D)*X + A3*D*X + e
where Y is the dependent variable, X is the regressor, and D represents a dummy variable. B1, B3, A2, and A3 are the parameters of interest. Theoretical prediction says B3 > B1 (sample splitting approach) or A3 > A2 (threshold modeling). My empirical result supports the prediction using the sample splitting approach, but the threshold approach fails to support the prediction (In fact, A2 > A3). I am wondering if there is something wrong with my code. Is there a reason that the two methods produce inconsistent results? Please advise. Thank you very much.
Code: Select all
linreg(title="Sample 1",smpl=D=0) Y
# constant X
linreg(title="Sample 2",smpl=D=1) Y
# constant X
set D_X1 = (1-D)*X
set D_X2 = D*X
linreg(title="Threshold") Y
# constant D_X1 D_X2