Bai Perron Sequential L+1 vs L breakpoints test

For technical questions regarding estimation of single equations, systems, VARs, Factor analysis and State Space Models in EViews. General econometric questions and advice should go in the Econometric Discussions forum.

Moderators: EViews Gareth, EViews Moderator

seydlitzD
Posts: 1
Joined: Tue Nov 30, 2021 12:35 pm

Bai Perron Sequential L+1 vs L breakpoints test

Postby seydlitzD » Wed Dec 01, 2021 1:42 am

Hello,

I have been using EViews11 Student Version Lite for some time, and I am delighted with the functionality and intuitiveness of the user interface.

EViews is also convincing because in GAUSS (using the GAUSS code provided by Bai and Perron; JAE Data Archive) when the time series is long (in my example T is ca 1400), the results (break only in intercept, HAC Andrews Automatic and Quadratic-Spectral, Sequential L+1 vs. L, maxbreaks=10, trimming=0.05, sign.level=0.05, allow errors to differ across breaks) are different from those in EViews (I have adapted the GAUSS code exactly to the settings in EViews, but tried out different values for the settings to compare outcomes with different settings). Has such a problem (difference between GAUSS and EViews results when T is big) been reported before? In the readme file of the GAUSS code, it is reported that problems can occur if T>500. This is not the case with EViews, is it?

Long story short: In EViews, there is no restriction on the Sample Size (T=1400 in my application) for the functionality of the sequential L+1 vs L breakpoint test regardless of the chosen specification (e.g., trimming =0.05). Right?

Thank you so much for your reply.

Best
Dave

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: Bai Perron Sequential L+1 vs L breakpoints test

Postby EViews Glenn » Wed Dec 01, 2021 10:22 am

There are no known issues with the EViews code at larger sample sizes. That's not to say that there aren't, but that present, I don't know of anything that could be an issue.

To be honest, I don't remember the details of the Gauss code, nor why there were issues for larger sample sizes.

[edit: To remind myself of the issues, I went back to double check size issues in EViews. Obviously, there are speed and size issues associated with doing global breakpoint testing at very large sample sizes. I suspect these are the reasons that the Gauss code cautioned for larger samples. EViews is pretty efficient in its memory and speed, but given these issues, EViews limits the number of periods to 10,000.)

Ed77
Posts: 1
Joined: Mon Nov 29, 2021 12:25 am

Re: Bai Perron Sequential L+1 vs L breakpoints test

Postby Ed77 » Sun Mar 13, 2022 5:55 pm

Hello,
I am applying the sequential breakpoint test with unknown breakpoint dates for nominal weekly exchange rate data (multivariate linear regression) with n=1084.

I have two questions:
I can see in the literature that if I ensure no serial correlation or heteroskedasticity (would this be the 'allow error distributions to varies across breaks tick box?') then I can change the trimming to 0.05 from the usual 0.15. Have I interpreted the literature correctly?

I am unable to find in the literature a recommendation to change the maximum breaks from 5 to, say 10. All the papers I have found refer to 5 as the max number of breaks. With weekly data, I anticipate there will be more than 5 break points. Can anyone point to where I can find support to change the max number of breaks or is there any precedent to do so? Given eviews allows it, I guess that it is able to be applied...but I need to demonstrate this in my thesis.

Thank you


Return to “Estimation”

Who is online

Users browsing this forum: Cscottmcwall-ea, Google [Bot] and 25 guests