Hello,
Some help on PCA regression procedures would be highly appreciated. I have a problem with multicollinearity between a few variables that I like to keep in the regression. I've searched extensively on the forum and documentation, but I am not able to find any help. An excerpt of my program:
'Original equation
equation orgreg.ls @log(AMORT) C ln_loans ln_income LTV
'PCA
group pca_vars ln_loans ln_income LTV
pca_vars.makepcomp w1 w2 w3
pca_vars.pcomp(eigvec=V, eigval=eigenval)
sym L=@makediagonal(eigenval)
equation pcareg.ls @log(AMORT) C w1 w2 w3
'Transformation back to original dimension...
vector gamma=pcareg.@coef
vector gamma=@subextract(gamma,2,1,4,1)
'...Beta(g)s
vector Bhat=V*gamma
'...SEs Beta(g)s
matrix SEBs=((pcareg.@se)^2)*V*@inverse(L)*@transpose(V)
for !j = 1 to 3
scalar seb{!j}=sqr(SEBs({!j},{!j}))
next
matrix SEBs= @fill(seb1,seb2,seb3)
The problem (or the more obvious problem) is that my transformed standard errors become very large, rendering the coefficients insignificant. The PCA should correct the errors but I don't see why they would come up so large and I think there's something wrong with the code. Any suggestions would be highly appreciated.
Best, Matt
PCA in regression
Moderators: EViews Gareth, EViews Moderator
Who is online
Users browsing this forum: No registered users and 49 guests