Structure Break in Variance Equation of MGARCH

[prashantj]

Hi Tom,

How can I find structural break in variance equation of MGARCH? is it the modified ICSS procedure in residuals of MGARCH or something different? Your help is appreciated.

Best,
Prashant

[TomDoan]

We deal with that in the GARCH course. While people have used ICSS, that's a really bad idea---ICSS and GARCH are substitutes, not complements. See the discussion in https://estima.com/forum/viewtopic.php?p=15075#p10266.

[prashantj]

Hi Tom,
XREGRESSORS with dummies for structural break. I understand that is useful for known break dates. but how to find break dates in variance when they are unknown and then create dummies? please.

Prashant

[TomDoan]

Use @FLUX which looks for structural breaks in the model you're estimating (unlike, say, ICSS which is looking for breaks in a completely different "model").

[prashantj]

Thanks, Tom. I appreciate it deeply.
One question may be very naive, but can you suggest crucial topics/ concepts that I should be referring from RATS programming manual (It is like bible-many topics) to get hold of fundamental concepts of RATS programming and build upon it for advanced programing?

[prashantj]

Is it possible to get the entry of the data point (observation number) with @flux procedure with MGARCH to know break point observation number?

Is there any way where I can delete data points by specifying the range i.e . data points with values in the range of 5-10 in a series?

[TomDoan]

Is it possible to get the entry of the data point (observation number) with @flux procedure with MGARCH to know break point observation number?

That's what the MaxBreak column is in the output.

Is there any way where I can delete data points by specifying the range i.e . data points with values in the range of 5-10 in a series?

I'm not sure what you mean.

[prashantj]

This is the command I used:
garch(mv=BEKK,rvectors=rd,hmatrices=hh,pmethod=simplex,asymmetric,stdresids=rstd,derives=dd,piters=20,iters=500,dist=T) / rbi rspx
@flux
# dd

The output is : It does not show observation number like at which entry it is statistically significant.
Test Statistic P-Value
Joint 15.2748530 0.00

1 0.1012890 0.56
2 1.5604437 0.00
3 1.1758595 0.00
4 0.6544222 0.02
5 7.6376021 0.00
6 1.9229347 0.00
7 0.6322445 0.02
8 0.3670913 0.09
9 2.7478954 0.00

[prashantj]

I want to delete observations whose value exceeds the values in the range i.e (5,10). For example, I have a series with 500 observations but I would like to delete all the observations whose value is greater than 5 or something like that. ( I am sorry I am sorry I am ignorant of using "quote" facility in the chat.

[TomDoan]

This is the command I used:
garch(mv=BEKK,rvectors=rd,hmatrices=hh,pmethod=simplex,asymmetric,stdresids=rstd,derives=dd,piters=20,iters=500,dist=T) / rbi rspx
@flux
# dd

The output is : It does not show observation number like at which entry it is statistically significant.
Test Statistic P-Value
Joint 15.2748530 0.00

1 0.1012890 0.56
2 1.5604437 0.00
3 1.1758595 0.00
4 0.6544222 0.02
5 7.6376021 0.00
6 1.9229347 0.00
7 0.6322445 0.02
8 0.3670913 0.09
9 2.7478954 0.00

There's a newer version of flux.src that includes that.

[prashantj]

Tom,
Finally, I have got the break point with dates as follows:

garch(mv=BEKK,rvectors=rd,hmatrices=hh,pmethod=simplex,asymmetric,xreg,stdresids=rstd,piters=20,iters=500,dist=T) / rb rs
@flux
# dd

Test Statistic P-Value MaxBreak
Joint 5.25339073 0.00 478

1 0.06902302 0.74 246
2 0.38262427 0.08 356
3 0.50392663 0.04 236
4 0.09000025 0.62 576
5 0.25960250 0.17 212
6 0.44114746 0.06 252
7 0.03917130 0.93 879
8 0.37845044 0.08 161
9 0.31220032 0.12 65
10 0.44111407 0.06 364
11 0.16608410 0.33 576
12 0.07957298 0.68 576
13 0.15500196 0.36 212
14 0.18324332 0.29 478
15 0.88779589 0.00 564
16 0.06898257 0.74 576
17 0.17550925 0.31 212

When it comes to volatility spillover using BEKK, it always taking my brain out of it. The joint statistics is significant but most of the (almost all) points are statistically insignificant but since joint test comes up with Max break at 478, I created the dummy using the following:
set x 478 478 = 1. Does it sound okay? I do not understand why it displays three values for 576 number observation? Am I using the correct flux procedure? Looking forward for your reply,
Prashant

[TomDoan]

1. A "structural break" in this case is not a one-off outlier---it's a permanent change.
2. You're doing a BEKK model. The "variance constant" is an n x n SYMMETRIC matrix, which is parameterized by an n x n lower triangular matrix. You can't change one variance by itself.
3. The three existing variance constant parameters are #3, #4 and #5 in the output. None of those are showing any clear signs of a structural break.

[prashantj]

Thanks for your reply. Since there are no structural break , does it mean parameter are stable and model is good. My mgarch statistics are not turning out to be significant.

[TomDoan]

Thanks for your reply. Since there are no structural break , does it mean parameter are stable and model is good.

That would be overstating things. It means that there is no evidence of a structural break.

My mgarch statistics are not turning out to be significant.

I'm not sure what you mean. Are you talking about in the fluctuations test, or are you talking about the coefficients of the BEKK model? It would seem rather odd if the A's and B's (at least the diagonal elements) are insignificant.

[prashantj]

I am sorry but this is the output I am getting. I did multiply returns with 100. It says no convergence. Any suggestions? please

output.jpg (263.48 KiB) Viewed 15926 times