regime-switch VAR-GARCH-M
Posted: Tue Mar 29, 2011 4:46 am
I'm trying to combine the VAR-GARCH-M with regime-switching.
But I met some problems.
The code below is estimate the VAR-GARCH-M in 3 regimes depending on dSP{1}
But when running the code
I got error message
""## MAT15. Subscripts Too Large or Non-Positive
Error was evaluating entry 1""
I tried to divide the sample in advance, but the error message
""## MAT15. Subscripts Too Large or Non-Positive"" is still appeared.
Could anyone help me with this?
my whole code is shown below.
The attachment are the data and the whole code.
But I met some problems.
The code below is estimate the VAR-GARCH-M in 3 regimes depending on dSP{1}
But when running the code
Code: Select all
*
* This does a simple GARCH(1,1) model for the variance.
*
dec symm VA1(n,n) VA2(n,n) VA3(n,n) VB1(n,n) VB2(n,n) VB3(n,n) VC1(n,n) VC2(n,n) VC3(n,n)
compute VA1=WWW1,VA2=WWW2,VA3=WWW3,VB1=%const(0.50),VB2=%const(0.50),VB3=%const(0.50),VC1=%const(0.05),VC2=%const(0.05),VC3=%const(0.05)
nonlin(parmset=garchparms1) VA1 VB1 VC1
frml Gf1 = VA1+VC1.*G1{1}+VB1.*uu1{1}
nonlin(parmset=garchparms2) VA2 VB2 VC2
frml Gf2 = VA2+VC2.*G2{1}+VB2.*uu2{1}
nonlin(parmset=garchparms3) VA3 VB3 VC3
frml Gf3 = VA3+VC3.*G3{1}+VB3.*uu3{1}
set LOGLL = logl1 + logl2 + logl3
maximize(trace,parmset=meanparms+garchparms1+garchparms2+garchparms3,pmethod=simplex,piters=10,method=bfgs,iters=400) LOGLL * gend
""## MAT15. Subscripts Too Large or Non-Positive
Error was evaluating entry 1""
I tried to divide the sample in advance, but the error message
""## MAT15. Subscripts Too Large or Non-Positive"" is still appeared.
Could anyone help me with this?
my whole code is shown below.
The attachment are the data and the whole code.
Code: Select all
*open the file (m)代表月 data後面接起始日與結束日
open data TD5_final.xls
CALENDAR(irregular)
data(format=xls,org=columns) / sp f1m
set dsp = sp - sp{1}
set df1m = f1m - f1m{1}
*graph SP(畫好多張圖) vfields = 幾張 footer = 標題名稱
print / dsp
spgraph(vfields=2,$
footer="Figure 2. the changes")
graph(vlabel="spot freight rate")
# dSP
graph(vlabel="1-month forward freight rate")
# dF1M
spgraph(done)
******************************************************************************
***********************************VAR-GARCH-M Model**************************
******************************************************************************
set dSP1 = dSP{1}
set Lam1 2 1578 = (dsp1>-22.and.dsp1<4.)
set Lam2 2 1578 = (dsp1>4.)
print / Lam1
dec series[vect] La1
dec series[vect] La2
gset La1 = ||Lam1||
gset La2 = ||Lam2||
compute n=2
compute gstart=2,gend=1578
*
dec series[vect] Y
dec frml[vect] R1 R2 R3
dec vect[series] U1(n) U2(n) U3(n)
*
*
* The paths of the covariance matrices and uu' are saved in the
* SERIES[SYMM] named H and UU. UX and HX are used for the current values
* of the residual vector and H matrices
*
declare series[symm] G1 G2 G3 uu1 uu2 uu3
*
* ux is used when extracting a u vector
*
declare symm Gx1(n,n) Gx2(n,n) Gx3(n,n)
declare vect ux1(n) ux2(n) ux3(n)
*
* These will be the parameters for the mean equations. These are adjusted to add
* variance or covariance terms as needed.
*
dec vect A1(n) A2(n) A3(n)
dec rect B1(n,n) B2(n,n) B3(n,n) C1(n,n) C2(n,n) C3(n,n)
nonlin(parmset=meanparms) A B C
*
* Mean model = VAR(1) with sqrt(G) "M" term
*
frml R = Y-A1-A2*La1-A3*La2-B1*Y{1}-B2*La1*Y{1}-B3*La2*Y{1}-C1*%xdiag(G1)-C2*La1*%xdiag(G2)-C3*La2*%xdiag(G3)
*
gset Y = ||dsp,df1m||
*
* Run preliminary VAR(4) to get estimates of residuals
*
*
linreg(smpl=dsp1<-22) dsp / u1(1)
# constant dsp{1 to 4} df1m{1 to 4}
linreg(smpl=dsp1>-22.and.dsp1<4.) dsp / u2(1)
# constant dsp{1 to 4} df1m{1 to 4}
linreg(smpl=dsp1>4) dsp / u3(1)
# constant dsp{1 to 4} df1m{1 to 4}
linreg(smpl=dsp1<-22) df1m / u1(2)
# constant dsp{1 to 4} df1m{1 to 4}
linreg(smpl=-22<dsp1<4) df1m / u2(2)
# constant dsp{1 to 4} df1m{1 to 4}
linreg(smpl=dsp1>4) df1m / u3(2)
# constant dsp{1 to 4} df1m{1 to 4}
*
linreg(smpl=dsp1<-22) dsp
# constant dsp{1} df1m{1} u2(1){1} u2(2){1}
compute A1(1)=%beta(1),B1(1,1)=%beta(2),B1(1,2)=%beta(3),$
C1(1,1)=%beta(4),C1(1,2)=%beta(5)
linreg(smpl=dsp1>-22.and.dsp1<4.) dsp
# constant dsp{1} df1m{1} u1(1){1} u1(2){1}
compute A2(1)=%beta(1),B2(1,1)=%beta(2),B2(1,2)=%beta(3),$
C2(1,1)=%beta(4),C2(1,2)=%beta(5)
linreg(smpl=dsp1>4) dsp
# constant dsp{1} df1m{1} u3(1){1} u3(2){1}
compute A3(1)=%beta(1),B3(1,1)=%beta(2),B3(1,2)=%beta(3),$
C3(1,1)=%beta(4),C3(1,2)=%beta(5)
linreg(smpl=dsp1<-22) df1m
# constant dsp{1} df1m{1} u2(1){1} u2(2){1}
compute A1(1)=%beta(1),B1(1,1)=%beta(2),B1(1,2)=%beta(3),$
C1(1,1)=%beta(4),C1(1,2)=%beta(5)
linreg(smpl=dsp1>-22.and.dsp1<4.) df1m
# constant dsp{1} df1m{1} u1(1){1} u1(2){1}
compute A2(1)=%beta(1),B2(1,1)=%beta(2),B2(1,2)=%beta(3),$
C2(1,1)=%beta(4),C2(1,2)=%beta(5)
linreg(smpl=dsp1>4) df1m
# constant dsp{1} df1m{1} u3(1){1} u3(2){1}
compute A3(1)=%beta(1),B3(1,1)=%beta(2),B3(1,2)=%beta(3),$
C3(1,1)=%beta(4),C3(1,2)=%beta(5)
vcv(matrix=WWW1)
# u1
vcv(matrix=WWW2)
# u2
vcv(matrix=WWW3)
# u3
*
* These are used to initialize pre-sample variances.
*
gset G1 * gend = WWW1
gset G2 * gend = WWW2
gset G3 * gend = WWW3
gset uu1 * gend = WWW1
gset uu2 * gend = WWW2
gset uu3 * gend = WWW3
set u1(1) = 0.0
set u2(1) = 0.0
set u3(1) = 0.0
set u1(2) = 0.0
set u2(2) = 0.0
set u3(2) = 0.0
*
declare frml[symm] Gf1 Gf2 Gf3
*
frml logl1 = $
Gx1 = Gf1(t) , $
G1(t) = Gx1, $
ux1 = R1 , $
uu1(t) = %outerxx(ux1), $
%pt(u1,t,ux1),$
%logdensity(Gx1,ux1)
frml logl2 = $
Gx2 = Gf2(t) , $
G2(t) = Gx2, $
ux2 = R2 , $
uu2(t) = %outerxx(ux2), $
%pt(u2,t,ux2),$
%logdensity(Gx2,ux2)
frml logl3 = $
Gx3 = Gf3(t) , $
G3(t) = Gx3, $
ux3 = R3 , $
uu3(t) = %outerxx(ux3), $
%pt(u3,t,ux3),$
%logdensity(Gx3,ux3)
*
* This does a simple GARCH(1,1) model for the variance.
*
dec symm VA1(n,n) VA2(n,n) VA3(n,n) VB1(n,n) VB2(n,n) VB3(n,n) VC1(n,n) VC2(n,n) VC3(n,n)
compute VA1=WWW1,VA2=WWW2,VA3=WWW3,VB1=%const(0.50),VB2=%const(0.50),VB3=%const(0.50),VC1=%const(0.05),VC2=%const(0.05),VC3=%const(0.05)
nonlin(parmset=garchparms1) VA1 VB1 VC1
frml Gf1 = VA1+VC1.*G1{1}+VB1.*uu1{1}
nonlin(parmset=garchparms2) VA2 VB2 VC2
frml Gf2 = VA2+VC2.*G2{1}+VB2.*uu2{1}
nonlin(parmset=garchparms3) VA3 VB3 VC3
frml Gf3 = VA3+VC3.*G3{1}+VB3.*uu3{1}
set LOGLL = logl1 + logl2 + logl3
maximize(trace,parmset=meanparms+garchparms1+garchparms2+garchparms3,pmethod=simplex,piters=10,method=bfgs,iters=400) LOGLL * gend
*
* This does a BEKK model for the variance
*
dec rect VBB1(n,n) VBB2(n,n) VBB3(n,n) VCC1(n,n) VCC2(n,n) VCC3(n,n)
compute VBB1=.05*%identity(n),VBB2=.05*%identity(n),VBB3=.05*%identity(n),VCC1=.50*%identity(n),VCC2=.50*%identity(n),VCC3=.50*%identity(n)
nonlin(parmset=garchparms1) VA1 VBB1 VCC1
nonlin(parmset=garchparms2) VA2 VBB2 VCC2
nonlin(parmset=garchparms3) VA3 VBB3 VCC3
compute VA1=%decomp(WWW1),VA2=%decomp(WWW2),VA3=%decomp(WWW3)
frml Gf1 = VA1*tr(VA1)+VCC1*G1{1}*tr(VCC1)+VBB1*uu1{1}*tr(VBB1)
frml Gf2 = VA2*tr(VA2)+VCC2*G2{1}*tr(VCC2)+VBB2*uu2{1}*tr(VBB2)
frml Gf3 = VA3*tr(VA3)+VCC3*G3{1}*tr(VCC3)+VBB3*uu3{1}*tr(VBB3)
set LOGLL = logl1 + logl2 + logl3
maximize(trace,parmset=meanparms+garchparms,pmethod=simplex,piters=10,method=bfgs,iters=400) LOGLL * gend