I have problems with using the cluster option of MAXIMIZE. I use MAXIMIZE for the estimation a nonlinear equation (GARCH-M) for panel data, and I want to compute clustered standard errors. Clustering on individuals is no problem, as the option LWINDOW=PANEL works fine. However, I would like to do the clustering on another variable (time), so I used the option CLUSTER=time (see below). This specification does not seem to give me clustered standard errors. The output tells me that Rats has computed HAC Standard Errors with FLAT window. In addition, the standard errors do not change when I change the variable I am clustering on.
Any help would be appreciated!
Best regards,
Niko
declare series u
declare series h
NONLIN VC VB VD BD BE BE_ BF BG BK BL BM BN BO BQ BR BS BT VE LPBE
FRML RESID = X - LPBE*lpcds{1}-0.6*%cdf(BD+(BE+BK*dbe+BL*des+BM*dfi+BN*dfr+BO*dgr+BQ*die+BR*dit+BS*dnl+BT*dpt)*y{1}+ BE_*y_{1}+BF*w{1}) - BG*h
FRML HF = VC + VB*U{1}**2 +VD*U{1}**2*%if(u{1}<0, 1, 0)+VE*h{1}
FRML LOGL = (H(T)=HF(T)), (U(T)=RESID(T)), -0.5*(LOG(H(T))+U(T)**2/H(T))
LINREG(NOPRINT) X / U
# CONSTANT
COMPUTE VC=0.0001,VB=0.9,VD=-0.6,VE=0.45,BD=-1.4,BE=8,BE_=-0.001,BF=1,BG=0.55,BK=6,BL=6,BM=4,BN=-1,BO=14,BQ=4,BR=12,BS=-5,BT=8,LPBE=0.7
SET H = %SEESQ
MAXIMIZE(METHOD=Bfgs,ITERATIONS=500,ROBUSTERRORS,LAGS=6,CLUSTER=time) LOGL 9 2840