The RATS Software Forum

Hi,
i am currentlu using Lee and strazicich unit root test of structural breaks. i am not convince of the result of the analysis.
i specified lags=12 and the output is pasted below:
Lee-Strazicich Unit Root Test, Series LBENGDP
Regression Run From 1973:01 to 2007:01
Observations 35
Trend Break Model with 2 breaks
With 12 chosen from 12

Variable Coefficient T-Stat
S{1} -2.8730 -6.5671
Constant 0.0880 6.7146
D(1978:01) -0.0036 -0.1525
DT(1978:01) 0.0595 4.6575
D(1987:01) 0.1059 3.3913
DT(1987:01) -0.1702 -6.2590
does this implies that all the lags are significant?
again, i decided to estimate the model again choosing the lag=8, and the 8 were significant as well and the estimates of the break dates were different. which do i decide on? morealso, how do i know if the t-statistics are significant?
Lee-Strazicich Unit Root Test, Series LBENGDP
Regression Run From 1969:01 to 2007:01
Observations 39
Trend Break Model with 2 breaks
With 8 chosen from 8

Variable Coefficient T-Stat
S{1} -1.4906 -5.0888
Constant 0.0632 5.4491
D(1977:01) -0.0932 -3.1502
DT(1977:01) 0.0486 4.0681
D(1991:01) -0.0173 -0.5916
DT(1991:01) -0.0439 -3.2409

asemota omos wrote:Hi,
i am currentlu using Lee and strazicich unit root test of structural breaks. i am not convince of the result of the analysis.
i specified lags=12 and the output is pasted below:
Lee-Strazicich Unit Root Test, Series LBENGDP
Regression Run From 1973:01 to 2007:01
Observations 35
Trend Break Model with 2 breaks
With 12 chosen from 12

Variable Coefficient T-Stat
S{1} -2.8730 -6.5671
Constant 0.0880 6.7146
D(1978:01) -0.0036 -0.1525
DT(1978:01) 0.0595 4.6575
D(1987:01) 0.1059 3.3913
DT(1987:01) -0.1702 -6.2590
does this implies that all the lags are significant?
again, i decided to estimate the model again choosing the lag=8, and the 8 were significant as well and the estimates of the break dates were different. which do i decide on? morealso, how do i know if the t-statistics are significant?
Lee-Strazicich Unit Root Test, Series LBENGDP
Regression Run From 1969:01 to 2007:01
Observations 39
Trend Break Model with 2 breaks
With 8 chosen from 8

Variable Coefficient T-Stat
S{1} -1.4906 -5.0888
Constant 0.0632 5.4491
D(1977:01) -0.0932 -3.1502
DT(1977:01) 0.0486 4.0681
D(1991:01) -0.0173 -0.5916
DT(1991:01) -0.0439 -3.2409

Yes. That means at least that the final lag is significant, whether you use 8 or 12. While it's not that surprising that the break point differs slightly between the two, the fact that the DT coefficient on the second break is quite different isn't promising. There's also that difficulty with reducing the residuals to white noise. You might want to think about whether this is the correct approach for this data series.

hi tom,
thanks for your response. i would try estimating the model using the crash model.
moreso, i am considering using baiperron procedure :
@BaiPerron( options ) depvar start end
# list of regressors
cosidering the fact that am considering a structural change in a univariate model, what are the regressors using the above procedure.

I don't think you're understanding the point. You start with the model, and look for breaks mainly as a specification test. The point, for instance, that Perron made regarding the unit root tests was that a series with a broken stationary trend could falsely lead to acceptance of a unit root. It's not necessarily that a broken stationary trend is the "correct" model - it's that a unit root process isn't.

thanks Tom. got the gist.