bai perron - differing results

[chiade]

Hi Tom,

I was running linear regressions using bai-perron to find the breaks using nfix options as well as leaving all breaks with all coefficients, including trend and constant.

@baiperron(maxbreaks=2,tests) bsbc 1997:11:28 2011:6:24
# wtic trend1 trend constant HH

The results differ if the regressors are placed differently i.e. # trend1 trend constant HH
wtic. So i got negative and different RSS, resulting in differing BIC. Why is that so?

Some quesions over the state-space model i.e. DLM.
-I got the t-stats for one of the time-varying coefficient which is less than 1.0 as seen from the graph with formula derived from the XSTATES(t)(i)/sqrt(VSTATES(t)(i,i). However, the t-stats for its variance is significant. Why is this so? The other coeffieicnts have mostly t-stats of >2.0 with the resp. variances also significant. Is it alright to ignore the t-stats but look only at the siginificance of the variances to conclude whether the variable is necessary in model?

also, how do i generate the aic/bic from CATS for the individual equations to be compared to the DLM?

many thanks for your clarifications.

Rgds,
Des

[TomDoan]

Hi Tom,

I was running linear regressions using bai-perron to find the breaks using nfix options as well as leaving all breaks with all coefficients, including trend and constant.

@baiperron(maxbreaks=2,tests) bsbc 1997:11:28 2011:6:24
# wtic trend1 trend constant HH

The results differ if the regressors are placed differently i.e. # trend1 trend constant HH
wtic. So i got negative and different RSS, resulting in differing BIC. Why is that so?

What is TREND1? Is it possible that it's collinear with TREND over part of the range?

Some quesions over the state-space model i.e. DLM.
-I got the t-stats for one of the time-varying coefficient which is less than 1.0 as seen from the graph with formula derived from the XSTATES(t)(i)/sqrt(VSTATES(t)(i,i). However, the t-stats for its variance is significant. Why is this so? The other coeffieicnts have mostly t-stats of >2.0 with the resp. variances also significant. Is it alright to ignore the t-stats but look only at the siginificance of the variances to conclude whether the variable is necessary in model?

One is a variance, one is a mean. Because a variance is significant doesn't mean the associated mean has to be. In fact, if the variance for coefficient drift is high, it's not at all unreasonable that the coefficient itself would be insignificant since it can change quite a bit and is subject to considerable uncertainty.

also, how do i generate the aic/bic from CATS for the individual equations to be compared to the DLM?

There really isn't a single equation AIC/BIC for a cointegrated model because the cointegrating vectors are estimated jointly. You can compute a single equation log likelihood (not in CATS; you would have to generate the error correction series and estimate the regression in RATS), but there isn't an obvious choice for the count of estimated parameters that you can assign to an individual equation.

[chiade]

Hi,

Let me rephrase my question again as i made a mistake i.e. trend1 was inadvertently included.

I was running linear regressions using bai-perron to find the breaks using nfix options as well as leaving all breaks with all coefficients, including trend and constant.

@baiperron(maxbreaks=2,tests) bsbc 1997:11:28 2011:6:24
# wtic trend constant HH . BIC=NA

The results differ if the regressors are placed differently i.e.
1) # trend constant HH wtic. BIC = 0.69
2) # constant trend witc hh. BIC=1.23

So i got negative and different RSS, resulting in differing BIC. Why is that so? Half of the regressions I ran encountered this problem whereas the other half doesn't, esp when I fixed wtic and hh.

[TomDoan]

You'll have to attach your program and data.