xtbreak

[Oscar Javier Cardenas]

Hi. I´m using xtbreak command on a panel data of i = 1, ...32 and t = 2005q1 ... 2024q4. I´m running this command: xtbreak estimate Δu Δy, csa(Δu Δy) breaks(2) trim(0.1) and it gves:

Estimation of break points
Number of obs = 2528
Number of Groups = 32
Obs per group = 79
SSR = 647.72
Trimming = 0.10
--------------------------------------------------------------------------------
# Index Date [95% Conf. Interval]
--------------------------------------------------------------------------------
1 20 2010q1 0910q2 3109q4
2 60 2020q1 2017q1 2023q1
--------------------------------------------------------------------------------
Cross-section averages:
with breaks: Δu Δy

The problem is the confidence interval. It gives dates not incluyded in the data set. What could it be? What can I do?

[Nick Cox]

Perhaps the confidence interval is what you asked for, a statistically candid indication of the uncertainty.

You can show us a graph of your data. I can't see that you used the syntax you report. but some equivalent of this might help:

xtline foo, overlay legend(off)

where naturally foo should be replaced by your variable name.

Bottom line: it is not much of a break.

(Please explain where xtbreak comes from: FAQ Advice #12.)

[JanDitzen]

Hi Oscar,

Nick makes an important point. The confidence intervals tell you the uncertainty of the estimate. The confidence intervals are a function of the difference in the estimates and - roughly speaking - the SSR of the regression. If the difference is small or the SSR large, the intervals become large.

I would check how large the differences in the coefficients is, i.e. run the estimation with the split variables. My expectation would be that the differences in the coefficients are small, hence the break is very small and thus difficult to detect.

On a different note: we recommend not to add the dependent variable as a cross-section average, see p. 530 and Footnote 7 in the Stata Journal article describing xtbreak; or discussion on page 76 of the JAE paper.