When the independent variables are collinear with each other, and there is no option to reduce them using step wise regression or factor analysis as all of them are important. Then in this case ridge regression can be used to estimate the model by force and adjust the biasness caused by the collinearity
**** Ridge regressiion clear use “D:\UMT notes\Applied Econometrics\lectures\lecture 5\ridge reg\kleingoldberger.dta”, clear **this example us using klein goldberger data ** the model is domestic consumption = f(wage income, nonwage nonfarm income, farm income) ** source Griffiths, Hill and Judge, Learning and Practicing Econometrics, 1993, Wiley, (Table 13.1, p.433) ** Judge, Hill, Griffiths, Lutkepohl and Lee, Introduction to the Theory and Practice of Econometrics, 1988, Wiley, (Table 21.1, p. 861). label variable y “Domestic Consumption” label variable x1 “Wage Income” label variable x2 “non-farm income” label variable x3 “farm income”
reg y x1 x2 x3
r square is high
but majority variables insigificant
signs are wrong correlate x1 x2 x3
high correlation between x1 and x3 estat vif
vif does not dipict presence of multi
according to partial and part correlations and
eigen value and condition index calculated in
spss there is multi in variable x2 and x3 // ssc install ridgereg ridgereg y x1 x2 x3 , model(orr) kr(0.27) mfx(lin) lmcol diag ** here model(orr) means
simple ridge regression process ** here k(#) is
the degree of buiasness must be allowed to estimate under multi. ** mfx(lin) means that it will generate the marginal corefficients using linear method ** lmcol is the code to test multicollinearity ** diag means it is providing model selection criteroin ridgereg y x1 x2 x3 , model(grr1) mfx(lin) lmcol ** above one is the generalized rigde regression method ridgereg y x1 x2 x3 , model(grr2) mfx(lin) lmcol ridgereg y x1 x2 x3 , model(grr3) mfx(lin) lmcol