Problem (?) with the Chow breakpoint test
Posted: Thu Sep 24, 2009 7:47 am
Hi,
I'm trying to determine if there is a breakpoint in several series; for this I thought I'd use the Chow test. I've run into an odd problem, however: the Chow test gives me a breakpoint at every point in the series, even though it looks as follows:
What's more, the probability it comes up with is in the order of 1.07261076424792e-46 no matter where I do the test.
I regress the series AHW on the series "time" which is just 1,2,3,4... etc, using
equation eq1.ls ahw c time
I then open up the resulting equation and run the Chow test on any point (100, 200, 500, 900, 1000, etc, my series are undated); the result is always the same, until I reach the last few observations (at 1275, the p-value is 0.163638; there are 1279 observations in total).
This all is somewhat distressing. What am I doing wrong? Or, is that just the nature of the test? Looking at the graph, it seems like a v. good fit to me...
Thanks for the help!
I'm trying to determine if there is a breakpoint in several series; for this I thought I'd use the Chow test. I've run into an odd problem, however: the Chow test gives me a breakpoint at every point in the series, even though it looks as follows:
What's more, the probability it comes up with is in the order of 1.07261076424792e-46 no matter where I do the test.
I regress the series AHW on the series "time" which is just 1,2,3,4... etc, using
equation eq1.ls ahw c time
I then open up the resulting equation and run the Chow test on any point (100, 200, 500, 900, 1000, etc, my series are undated); the result is always the same, until I reach the last few observations (at 1275, the p-value is 0.163638; there are 1279 observations in total).
This all is somewhat distressing. What am I doing wrong? Or, is that just the nature of the test? Looking at the graph, it seems like a v. good fit to me...
Thanks for the help!