Koutmos JBFA 1996 Multivariate EGARCH

[TomDoan]

The following estimates a multivariate E-GARCH model with a VAR(1) mean model and spillovers in the GARCH specification. This is from G. Koutmos(1996) "Modeling the Dynamic Interdependence of Major European Stock Markets", Journal of Business Finance and Accounting, Vol. 23, pp. 975-988. The program should require relatively little change to be adapted to other data.

This program is covered in considerable detail as part of the ARCH, GARCH and Volatility e-course.

Version 9 of RATS adds the option VARIANCES=KOUTMOS to allow direct estimation using the GARCH instruction.

vegarch.rpf

Program file

(6.24 KiB) Downloaded 1320 times

vegarch9.rpf

Program for Version 9 or later

(3.12 KiB) Downloaded 1127 times

stock9.rat

Data file

(117.25 KiB) Downloaded 1177 times

[prashantj]

I am using two stock market series of japan and hongkong. When I have modified the programme of koutmos to suit my need. When I have run the programme, I have got the following message:

## REG20. GARCH Cannot Be Used with Gaps/Missing Values

could you please recommend me the solution? Please.
Prashant

[TomDoan]

You can't have any gaps (missing values) within the range that GARCH uses. Apparently you do. You'll have to figure out how you plan to deal with that and fix it.

[prashantj]

Dear Tom,
I have been using Koutmos JBFA 1996 Multivariate EGARCH programme for my data set. Both programme and the data set are attached. I have encountered a serious problem when I use it with RATS which
is as follows: My version of RATS is 8.3 standard.

When I run the programme with modification as per my requirement, the processing take unusual time to such an extent that RATS cannot generate output even after 50 minutes. I fail to figure out solution for the problem. Could you please help me? and oblige, please.

Looking forward for your reply,

With regards,
Prashant

[TomDoan]

Did you let it run longer? That has 80 parameters, a relatively long data set and a fairly complicated likelihood function. It's going to take a while. If you want some feedback, add a TRACE option to the MAXIMIZE instruction.

[prashantj]

yes, i allowed to let it go longer to more than one hour but of no avail. still RATS is processing it though The number of data set or observation are way less than what Koutmos's paper included. Could you please suggest the way out?
Looking forward for your reply,
Prashant

[TomDoan]

If you're concerned, cancel it and re-run it with the TRACE option on.

You're using five variables rather than four. The increase in the number of parameters is more significant than the somewhat shorter data set.

[TomDoan]

It's actually 85 parameters---you forgot to add the 5th variable to the mean equation, which is supposed to read:

equation meaneq *
# constant y(1){1} y(2){1} y(3){1} y(4){1} y(5){1}

You also have almost no contemporaneous correlation among your variables---many of the CC's are less than .01 in absolute value.

[prashantj]

If there are no contemporaneous correlation among variables, will it be worth examining volatility spillover using koutmos EGARCH model? I have read somewhere that even if there is poor correlation or no correlation, there can be a possibility of volatility spillover among the markets. looking forward for your reply,
Prashant

[TomDoan]

Is it possible? Yes. Is it unlikely that there's spillover when there's such a weak connection? I think also yes.

[prashantj]

Dear Tom,
I run the programme with little change to adapt to my requirement. I have got R(2,1) in table 1 and R(2,2) in table 2. But I could not get both at the same time in both the tables. I would like to know what does R indicate? Does it mean correlation coefficient? Does it make any any difference if R(1,1) and R(2,1) omitted in Table 1 and Table 2 respectively?
Another thing is t-statistics for coefficients of B(1) and B(2) in table 1 and G(1) and G(2) in table 2 , B(1) and B(2) in table 3 are very high even more than 100. Does it convey something wrong with the model or nothing to worry. I have attached the output, programme and data set for your reference, please.
Looking forward for your reply,
Prashant

[TomDoan]

Don't do 100 simplex iterations. You're already at the optimum by the time you're done with that, so the BFGS estimates of the curvature (used for the standard errors) are wrong.

The correlations are modelled using the packed subdiagonal matrix. GARCH does the same thing, but GARCH knows the proper way to label those. When you use MAXIMIZE, the correlation matrix has the form

1
R(1,1) 1
R(2,1) R(2,2) 1
....

so the R(1,1) is, in fact, the 2,1 correlation.

[f_ta]

Hi,

I am trying to use the VAR-EGARCH code of Koutmos, modified to bivariate with an exogenous variable in the mean model, on my data. I get the following warning:

NO CONVERGENCE IN 51 ITERATIONS
LAST CRITERION WAS 0.0000000
SUBITERATIONS LIMIT EXCEEDED.
ESTIMATION POSSIBLY HAS STALLED OR MACHINE ROUNDOFF IS MAKING FURTHER PROGRESS DIFFICULT
TRY HIGHER SUBITERATIONS LIMIT, TIGHTER CVCRIT, DIFFERENT SETTING FOR EXACTLINE OR ALPHA ON NLPAR
RESTARTING ESTIMATION FROM LAST ESTIMATES OR DIFFERENT INITIAL GUESSES MIGHT ALSO WORK

I read the forum and understood that this is only a warning, as also the vegarch.rpf-code states. However, how can I verify that my model actually converges if this warning appears? Sorry, I am new to RATS, and I would greatly appreciate any help with this.

Code and data below.

Best regards,
Amanda

[TomDoan]

Code: Select all

all 1 1292
open data h:\Datafnr.xls
data(format=xls,org=obs) / eurf usdf eursteep usdsteep
*
compute n=2
*
compute gstart=3:1292,gend=1292:1292
*
dec vect[series] y(n)
set y(1) = (eurf-eurf{1})
set y(2) = (usdf-usdf{1})
*
dec vect[series] s(n)
set s(1) = (eursteep-eursteep{1})
set s(2) = (usdsteep-usdsteep{1})

1. Not that it affects the results, you want

all 1292

and

compute gstart=3,gend=1292

This is an irregular time series data set, so you don't want to even try to use dates.

2. Don't you want

set y(1) = (eurf-eurf{1})
set y(2) = (usdf-usdf{1})

to be log differences? Level differences looks wrong.

3. This is probably what creates the problem:

set s(1) = (eursteep-eursteep{1})
set s(2) = (usdsteep-usdsteep{1})

Again, should those be log differences? But more important, should you have those in your mean model contemporaneously rather than with a lag? Those two have very large, very, very significant coefficients in the mean model, which probably would only happen if they were being determined simultaneously with the dependent variables.

[f_ta]

Dear Tom,

Thank you for your prompt reply. Changing the xsteep variables to contemporaneous still did not lead to convergence. I also tried leaving out the exogenous variable and did not get any valid results.

As I have several datasets some of which include negative values, I do not think that log-returns will work. I am trying to do a similar study as In (2007) Volatility spillovers across international swap markets: The US, Japan, and the UK, and Toyoshima & Hamori (2012) Volatility transmission of swap spreads among the US, Japan and the UK: a cross-correlation function approach, where both state that they are using first differences.

Is it possible to change the distribution in the maximize function to t-distribution?

Thank you for your help, it is highly appreciated.

Best regards,
Amanda