Multivariate EGARCH

[jomens51]

Hi

My name is Jones and a new user of eviews. I have been trying to estimate a Multivariate EGARCH model to analyse the spillover effect across different stock markets. I have searched the entire forum but haven't com across any code for such estimation. Would kindly like to request , if any has written the code and will like to share. Its quite urgent please.

Thanks

[yixiaoyang]

Code: Select all

'set sample pagestruct(start=3/30/2010, end=9/14/2011) * smpl @all series y1=dlog(ap) series y2=dlog(hp*exc) sample s1 3/30/2010 9/14/2011 sample s2 4/2/2010 9/14/2011 'set series 'declare coefficients coef(2) beta0 beta0(1)=0.001 beta0(2)=0.001 coef(4) beta beta(1)=-0.2 beta(2)=0.3 beta(3)=0.2 beta(4)=-0.2 coef(2) lambda lambda(1)=0.2 lambda(2)=0.2 coef(2) alpha0 alpha0(1)=0.001 alpha0(2)=0.001 coef(2) gamma gamma(1)=0.5 gamma(2)=0.5 coef(4) alpha alpha(1)=0.001 alpha(2)=0.3 alpha(3)=0.001 alpha(4)=0.2 coef(2) delta delta(1)=-0.1 delta(2)=-0.01 coef(2) ef ef(1)=0.2 ef(2)=0.1 coef(1) c0 c0(1)=0.01 coef(1) ci ci(1)=0.04 coef(1) g g(1)=0.7 coef(9) dum 'dum(1)=0.001 'beta(2) 'dum(2)=0.001 'lambda(1) 'dum(3)=0.001 'beta(3) 'dum(4)=0.001 'lambda(2) 'dum(5)=0.001 'alppha(2) 'dum(6)=0.001 'ef(1) 'dum(7)=0.001 'alpha(4) 'dum(8)=0.001 'ef(2) 'dum(9)=0.001 'c(1) !pi = @acos(-1) !absz=sqr(2/!pi) 'set presample values of expressions in logl smpl s1 series res1=y1-beta0(1)-beta(1)*y1(-1)-beta(2)*y2(-1) series res2=y2-beta0(2)-(beta(3))*y1(-1)-beta(4)*y2(-1) !var1=(@stdev(res1))^2 !var2=(@stdev(res2))^2 !correlation=@cor(y1,y2) series h11=!var1 series h22=!var2 series rho=!correlation series h12=2*rho*sqr(h11)*sqr(h22) series deth=h11*h22-h12*h12 series invh11=h22/deth series invh12=-h12/deth series invh22=h11/deth 'set up EGARCH likelihood smpl s2 logl ll1 ll1.append @logl logl @byeqn ll1.append res1=y1-beta0(1)-beta(1)*y1(-1)-beta(2)*y2(-1) ll1.append res2=y2-beta0(2)-beta(3)*y1(-1)-beta(4)*y2(-1) ll1.append h11=exp(alpha0(1)+gamma(1)*log(h11(-1))+alpha(1)*(abs(res1(-1)/sqr(h11(-1)))-!absz+delta(1)*((res1(-1)/sqr(h11(-1)))))+alpha(2)*(abs(res2(-1)/sqr(h22(-1)))-!absz+delta(2)*((res2(-1)/sqr(h22(-1)))))) ll1.append h22=exp(alpha0(2)+gamma(2)*log(h22(-1))+alpha(3)*(abs(res2(-1)/sqr(h22(-1)))-!absz+delta(2)*((res2(-1)/sqr(h22(-1)))))+(alpha(4))*(abs(res1(-1)/sqr(h11(-1)))-!absz+delta(1)*((res1(-1)/sqr(h11(-1)))))) ll1.append rho=@cor(res1,res2) ll1.append h12=rho*sqr(h11)*sqr(h22) ll1.append deth=h11*h22-h12*h12 ll1.append invh11=h22/deth ll1.append invh12=-h12/deth ll1.append invh22=h11/deth ll1.append likelihoods=log(deth)+(res1^2)*invh11+(res2^2)*invh22+2*res1*res2*invh12 ll1.append logl=-log(2*!pi)-0.5*likelihoods ll1.append z1=res1/sqr(h11) ll1.append z2=res2/sqr(h22) ll1.append z11=(res1^2)/h11 ll1.append z22=(res2^2)/h22 ll1.append z12=z1*z2 ' estimate and display output smpl s2 ll1.ml(showopts,m=100,c=1e-5) show ll1.output

this might be helpful for you. i am still testing the program, if anyone know any pitfall that I made here, please tell!

[garchiemanning]

Might be a fun idea to alter the correlation function to allow conditional corr (DCC). I'll see what I come up with.

[bilal]

Can anyone tell me if this code is valid . and can you make it to the bivariate version?