MV-EGARCH with spillovers

[ibrahim]

Sorry about that, it is my miscoding,
Thank you so much Tom,
But after defining std residuals, for LB(Q) test I enter the below code

Code: Select all

corr(qstats,number=12) z1 

and then I got the below error

## SR10. Missing Values And/Or SMPL Options Leave No Usable Data Points

[TomDoan]

Works fine for me. Did you put this after the MAXIMIZE instruction?

[TomDoan]

There's an updated version which includes diagnostics at

http://www.estima.com/forum/viewtopic.php?f=8&t=1940

[ibrahim]

thank you so much Tom,
Best

[pitsikiou]

Dear Tom,
i follow Koutmos code to estimate the spillover among 4 countries but instead of a var - egarch model i have to estimate an ecm multvariate egarch model.
What should i add in this code so as to estimate the ecm-mv-egarch ?

Thank you very much

[TomDoan]

The Koutmos paper does a VAR on the differences. The difference between that and a VECM is that the VECM requires an error correction term (or terms, depending upon the number of common trends). For instance, the following would create a single common trend for the four exchange rates:

(You can use any sets of pairs of the exchange rates in this).

set ect1 = 100*log(fra/ger)
set ect2 = 100*log(ger/ita)
set ect3 = 100*log(ita/uki)

equation meaneq *
# constant y(1){1} y(2){1} y(3){1} y(4){1} ect1{1} ect2{1} ect3{1}

[pitsikiou]

You mean that i have to also write
set ect1 = 100*log(fra/ger)
set ect2 = 100*log(ger/ita)
set ect3 = 100*log(ita/uki)
set ect4 = 100*log (fra/ita)
set ect5 = 100*log (fra/uki)
set ect6 = 100*log (ger/uki)

in order to have the equation meaneq *
# constant y(1){1} y(2){1} y(3){1} y(4){1} ect1{1} ect2{1} ect3{1} ect4{1} ect5{1} ect6{1}???
I am sorry for the persistent quetions but i am working on my own the mv-ecm- egarch model so as to find the spillovers from each country to the other.
Thank you very much for your time

[TomDoan]

You mean that i have to also write
set ect1 = 100*log(fra/ger)
set ect2 = 100*log(ger/ita)
set ect3 = 100*log(ita/uki)
set ect4 = 100*log (fra/ita)
set ect5 = 100*log (fra/uki)
set ect6 = 100*log (ger/uki)

in order to have the equation meaneq *
# constant y(1){1} y(2){1} y(3){1} y(4){1} ect1{1} ect2{1} ect3{1} ect4{1} ect5{1} ect6{1}???
I am sorry for the persistent quetions but i am working on my own the mv-ecm- egarch model so as to find the spillovers from each country to the other.
Thank you very much for your time

No. Half of those are redundant. If FRA-GER and GER-ITA are stationary then FRA-ITA has to be stationary as well. I was also just giving an example based upon the Koutmos data; if you have a different VECM, you would set it up differently.

Have you tried estimating the VECM without the GARCH errors? That's the place to start.

[pitsikiou]

i completed the cointegration and found out that ger-uki: 1 cointegrating vector,ger-fra: 2 cointegrating vectors, ger-ita:2 cointegrationg vectors, fra-ita:2 coint.vect, fra-uki:2 coint. vec, ita-uki:2 coint vec

[TomDoan]

i completed the cointegration and found out that ger-uki: 1 cointegrating vector,ger-fra: 2 cointegrating vectors, ger-ita:2 cointegrationg vectors, fra-ita:2 coint.vect, fra-uki:2 coint. vec, ita-uki:2 coint vec

You're doing a four variable system, so you want to be looking at cointegration of the four variable system, not pairwise.

Also, don't you have theoretical values of the cointegrating vectors?

[pitsikiou]

i did the cointegration for the 4 countries and johansen showed 2 cointegrated vectors but Max-eigenvalue test indicated no cointegration.
The ecm model gave me the following results

Code: Select all

  Vector Error Correction Estimates				
 Included observations: 95 after adjustments				
 Standard errors in ( ) & t-statistics in [ ]				
				
Cointegrating Eq: 	CointEq1			
				
DEN(-1)	 1.000000			
				
ESP(-1)	 0.172437			
	 (0.21176)			
	[ 0.81429]			
				
FRA(-1)	-2.786610			
	 (0.50725)			
	[-5.49357]			
				
GER(-1)	 1.524009			
	 (0.42338)			
	[ 3.59961]			
				
C	-2.120413			
				
Error Correction:	D(DEN)	D(ESP)	D(FRA)	D(GER)
				
CointEq1	-0.059121	 0.033942	 0.179201	-0.096106
	 (0.05115)	 (0.10068)	 (0.08450)	 (0.08959)
	[-1.15573]	[ 0.33714]	[ 2.12075]	[-1.07276]
				
D(DEN(-1))	 0.311850	-0.026134	-0.016336	-0.025765
	 (0.16200)	 (0.31883)	 (0.26760)	 (0.28372)
	[ 1.92497]	[-0.08197]	[-0.06104]	[-0.09081]
				
D(DEN(-2))	-0.024185	 0.154542	 0.087055	 0.348286
	 (0.14035)	 (0.27622)	 (0.23184)	 (0.24580)
	[-0.17232]	[ 0.55950]	[ 0.37550]	[ 1.41697]
				
D(ESP(-1))	 0.020153	 0.617986	 0.103380	 0.317499
	 (0.08371)	 (0.16476)	 (0.13828)	 (0.14661)
	[ 0.24073]	[ 3.75090]	[ 0.74759]	[ 2.16557]
				
D(ESP(-2))	 0.165803	-0.070239	 0.232540	 0.116802
	 (0.08412)	 (0.16556)	 (0.13895)	 (0.14732)
	[ 1.97102]	[-0.42426]	[ 1.67350]	[ 0.79283]
				
D(FRA(-1))	 0.009473	-0.103016	 0.257007	 0.089151
	 (0.13044)	 (0.25672)	 (0.21547)	 (0.22844)
	[ 0.07263]	[-0.40128]	[ 1.19278]	[ 0.39025]
				
D(FRA(-2))	-0.055491	-0.151846	-0.196705	-0.035067
	 (0.11648)	 (0.22924)	 (0.19241)	 (0.20399)
	[-0.47640]	[-0.66238]	[-1.02233]	[-0.17190]
				
D(GER(-1))	 0.159824	 0.045931	 0.208637	-0.123134
	 (0.10868)	 (0.21389)	 (0.17952)	 (0.19033)
	[ 1.47063]	[ 0.21475]	[ 1.16220]	[-0.64695]
				
D(GER(-2))	-0.135151	-0.314002	-0.254719	-0.309460
	 (0.10334)	 (0.20338)	 (0.17070)	 (0.18098)
	[-1.30785]	[-1.54393]	[-1.49220]	[-1.70992]
				
C	 0.166465	 0.162745	-0.074771	-0.067585
	 (0.39885)	 (0.78497)	 (0.65884)	 (0.69852)
	[ 0.41736]	[ 0.20733]	[-0.11349]	[-0.09676]

[TomDoan]

You're planning to do a multivariate GARCH model with only 95 data points? What kind of data do you have?

[pitsikiou]

monthly data from 2006/01 - 2014/02 for 4 countries (98 observations for each country)

[TomDoan]

I seriously doubt that you're going to get reasonable results from running a complicated GARCH model with so little data. Have you run univariate GARCH models to see if there's even a GARCH effect to model?

[pitsikiou]

yes i tested that and found out that garch effect exists in univariate garch for each country