Changing trimming size causes quite different results

[kenforex]

Hi Prof. Jan Ditzen,

Hope you are well and appreciate for reading my issue.

I am a PhD in UK and really into your xtbreak method. I try to use this to estimate breaks for 345 UK Local Authority's panel data over 226-month time period, where it includes the dependent variable 'hpresa' as the demean log return of the house price divided by annual cpi, and other 3 independent regressors.

I used the xtbreak with command 'xtbreak hpresa hpresa_lag1 salesresa_lag1 employresb_lag1 higheduresb_lag1, nofixedeffects breakconstant vce(hac) csd trimming(0.15)' in terms of pooled ols with no fixed effects, and it gives me a solid result with 5 breaks.

However, when I turn the trimming size to 0.1, it becomes 9 breaks. It seems that changing the trimming size would change the result of breaks. How should I determine the trimming size for breaks?

Please see my results in attached pictures.

Best.

[JanDitzen]

Hi, thank you very much for your interest in xtbreak.

If you choose a smaller trimming, you can estimate a larger number of breaks. With the default trimming of 0.15 (15%) you can estimate at most 5 breaks. If there are more breaks in your data, xtbreak will at most find 5 breaks, most likely the most distinct ones. Looking at your outputs my interpretation would be that your have a lot of breaks in your dataset.

Do results change if you a) omit options csd/vce; b) data is in 1st differences; c) omit options nofe and breakconstant.

Best,

Jan Ditzen