Dear all,
I need assistance with interpreting the results from Z-A (1992) test on a variable with structural break point. Here are some results:
The variable is Gini index (measure of income inequality):
zandrews gini, break(trend) trim(0.01) lagmethod(BIC) graph
Results: //Zivot-Andrews unit root test for gini
//Allowing for break in trend
//Lag selection via BIC: lags of D.gini included = 0
//Minimum t-statistic -3.449 at 2006 (obs 27)
//Critical values: 1%: -4.93 5%: -4.42 10%: -4.11
The break point is at 2006 and since the calculated t-stat (-3.449) is below the critical values, i suppose that the variable is non-stationary, so I took the 1st difference and here's the result:
zandrews d.gini, break(trend) trim(0.01) lagmethod(BIC) graph
Results: //Zivot-Andrews unit root test for D.gini
//Allowing for break in trend
//Lag selection via BIC: lags of D.D.gini included = 0
//Minimum t-statistic -6.478 at 2011 (obs 32)
//Critical values: 1%: -4.93 5%: -4.42 10%: -4.11
The t-stat is now -6.478 which is well above the critical values evidencing stationarity but the break point has now shifted to 2011 from 2006. So, which break point is applicable for the analysis?
Kindly advise....thank you.
I also want to read a proper answer to this question!
The year selected by ZA procedure are the years in which the t statistic is minimized.
In my short experience, I add dummies variables based on historical events related to the study's objetive or theme. Sometimes they match with the dates given by ZA test.
The null hypothesis for Zivot-Andrews Unit Root Test of "unit root with structural break in the intercept" of a variable can only be rejected when Zivot-Andrew test statistic value is less than the critical values of (1%, 5% & 10%) levels. OR when the absolute value of this t-statistic is greater then critical values. Coming to your question properly, 2011 should be the correct breakpoint since that is the level of stationarity of the variable in question. Thus, the final analysis must account for unit root and structural break since they have been established from the pre-estimation test (Zivot-Andrews Test).
Hi Ike/Nip...and thanks for these contributions.
But what if I use the break point corresponding to the test statistic indicated by the Gregory-Hansen (1996) co-integration test?...I mean the test statistic by why the null hypothesis of no co-integration is rejected. Will that be appropriate to construct a dummy variable with?
I think the statistically significant break date from the G-H test is more intuitive to use in formulating the dummy variable to capture the structural break when formulating the error correction model since co-integration is established.
For instance, from the Zivot-Andrews test, I have different break points across the three variables (1987, 1995, 2001, 2007 and 2010) but from the G-H test when the 3rd model that captures change in the intercept and slope is used, I obtained 2003 as the date corresponding to a statistically significant test statistic (from Zt). So, I decided to go with that....I just hope that it is academically/econometrically in order.
I still crave more tutoring/guidance on structural break modeling.
Thanks.
Yes, a breakpoint period that gives statistical significant Zt is better in constructing dummy to capture structural break. Also CUSUM stability test can show further evidence of the breakpoint.
The answer given by NIP NIP has been my observation. The dates most times givwn by the ZA test do correspond to dummy dates
HOW DO I KNOW WHETHER THERE IS A SINGLR OR MULTIPLE STRUCTURAL BREAKS IN THE DATA
It is there in STATA "ZAndrews" and in Eviews as a plug-in downloadable from Eviews
How can we estestimate the Zivot andand Andrew unit root test on E-views? I have downloaded it, for now, I can only estimate each variable at levels, what of first difference of variables? Pls, I will need a step-by-step procedure. Many thanks.
Dear Solomon plz creat first difference and test ZA unit root test for it.
Hello Niaz What if after testing for first difference and the result is still not stationary at that point, what will be your conclusion
Thank you, Dr. Niaz. One has to do the differencing manually before applying the ZA test. Many thanks. God bless.
Bosede Ngozi Adeleye the post-zivot break dates are more reliable but, in explaining them you should also take into account historical events that occured in thoses dates (original dates in the level of Andrew-Zivot )
Hi good people. What would be the best model to use in studying artificial intelligence usage to address deforestation
the null says the variable has a unit root
One is to reject null if t-value statistic is lower than tabulated critical value
The issue here, am I to use absolute value irrespective of the sign (+/-) or not ?
While the purpose of some regresssion models is to estimate long and short run relationship in time series analysis, do we need to interpret the historical events in discussion of unit root test, for exampke if we use Zivot Andrew unit root test?
Ashar Awan, the historical events help you in applying the dummies to take care of the break
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series