Page 1 of 1
Code for tests
Posted: Sun Mar 04, 2018 7:45 am
by humyra
Dear Tom,
What is the code for a WALD test for joint hypothesis is equal to zero. For example,if I want to test the hypotheis in VAR-BEKK-GARCH model that a12 = b 12 = 0, what would be the code for the WALD test.
Also, is RATS able to identify structural breaks in financial time series using Bai and Perron (1998, 2003) test as well as the Lee and Strazicich (2003, 2004) test?
Thanks,
Humyra
Re: Code for tests
Posted: Sun Mar 04, 2018 4:15 pm
by TomDoan
humyra wrote:Dear Tom,
What is the code for a WALD test for joint hypothesis is equal to zero. For example,if I want to test the hypotheis in VAR-BEKK-GARCH model that a12 = b 12 = 0, what would be the code for the WALD test.
The easiest way to do that is with the
Regression Tests wizard. Pick the Exclusion Restrictions and select the two parameters that you want to test.
humyra wrote:
Also, is RATS able to identify structural breaks in financial time series using Bai and Perron (1998, 2003) test as well as the Lee and Strazicich (2003, 2004) test?
First, there are no breaks in a
series. Breaks are in a model.
Bai-Perron is a test for (multiple) breaks in a linear regression.
Lee-Strazicich isn't even a test for breaks, it's a test for unit roots allowing for breaks. So neither applies to finding structural breaks in a GARCH model. The only relatively simple test for breaks in a GARCH model is the
Nyblom fluctuation test.
Re: Code for tests
Posted: Sun Mar 11, 2018 9:41 am
by humyra
Thank you Tom.
One more question, if in an asymmetric VAR-BEKK-GARCH (1,1), A12 is negative, B12 is positive, and D12 is positive too, what explanation might one give? I know you have mentioned that its not easy to interpret these but I'm actually stuck as to what financial interpretation to give. Does this make sense: The bank has a negative effect on current volatility in the financial market due to first lag term shock. High volatility in the bank's stock today will lead to low volatility in the financial market the next trading day. Furthermore, due to a positive B12, this lower volatility will persist and and helping in stabilizing the financial market. How do I incorporate the positive D1 here?
Re: Code for tests
Posted: Mon Mar 12, 2018 11:23 am
by TomDoan
You can't interpret individual coefficients in a BEKK (all coefficients only appear as interactions in the corresponding VECH) and you shouldn't try to nor should you be expected to. Read the section on BEKK in
https://arxiv.org/pdf/math/0702815.pdf
Re: Code for tests
Posted: Tue Jul 17, 2018 2:21 am
by sanjeev
Hi Tom,
I want to know the code to conduct causality tests in mean and variance in a multivariate GARCH_BEKK model.Could you please help?
Thanks.
Regards.
Re: Code for tests
Posted: Tue Jul 17, 2018 9:42 am
by TomDoan
Spillover/causality tests in GARCH models are covered in detail in the 2nd edition of the GARCH e-course. This is an example. Note that the causality in mean tests take exactly the same form as they do for a regular VAR. These are generally done most easily by estimating the GARCH model then using the Regression Tests--Exclusion Restrictions operation.
Code: Select all
*
* GARCH5_4.RPF
* Multivariate GARCH, BEKK Estimation, Tests for "spillovers"
* Data set from Enders, Applied Econometric Time Series, 3rd edition
*
open data "exrates(daily).xls"
calendar(d) 2000:1:3
data(format=xls,org=columns) 2000:01:03 2008:12:23 $
aust euro pound sw ca
*
set reuro = 100.0*log(euro/euro{1})
set rpound = 100.0*log(pound/pound{1})
set rsw = 100.0*log(sw/sw{1})
*
dec vect[string] longlabel(3)
compute longlabel=||"Euro","UK Pound","Swiss Franc"||
*
system(model=mvmean)
variables reuro rpound rsw
lags 1
det constant
end(system)
*
garch(model=mvmean,mv=bekk,robusterrors,pmethod=simplex,piters=20,$
method=bfgs,iters=500,rvectors=rd,hmatrices=hh)
*
test(zeros,title="Test of Exogeneity in Mean of All Variables")
# 2 3 5 7 9 10
*
test(zeros,title="Test of Exogeneity in Mean of Euro")
# 2 3
*
test(zeros,title="Test of Exogeneity in Mean of Pound")
# 5 7
*
test(zeros,title="Test of Exogeneity in Mean of Swiss Franc")
# 9 10
*
test(zeros,title="Wald Test of Diagonal BEKK")
# 20 21 22 24 25 26 29 30 31 33 34 35
*
test(zeros,title="Block Exclusion Test, Euro Variance")
# 22 25 31 34
*
test(zeros,title="Block Exclusion Test, Pound Variance")
# 20 26 29 35
*
test(zeros,title="Block Exclusion Test, Swiss Franc Variance")
# 21 24 30 33
Re: Code for tests
Posted: Fri Sep 28, 2018 2:08 am
by sanjeev
Hi,
Can we interpret the coefficients in a BEKK Model the way we do in any other regression model? Precisely, can we discuss the signs of the coefficients in relation to economic theory?
Please help!
Thanks.
Regards.
Re: Code for tests
Posted: Fri Sep 28, 2018 7:48 am
by TomDoan
sanjeev wrote:Hi,
Can we interpret the coefficients in a BEKK Model the way we do in any other regression model? Precisely, can we discuss the signs of the coefficients in relation to economic theory?
Please help!
Thanks.
Regards.
No. The signs aren't even statistically identified. BEKK coefficients aren't structural. You can't/don't interpret them.