In Quandt-Andrews unknown break point test, there are 3 common critical values to decide the significance of obtained break point, viz., MaxF, ExpF and AveF. Would MaxF is enough to decide whether the obtained breakpoint is significant? How about the rest?
First: maxF is better than the other two -- it has more power, as Andrews demonstrated, and as I have personally assessed via simulations and practical examples. Second, the asymptotic critical values given by Andrews are not very good in finite samples, but you can easily compute accurate critical values by simulatons. Third. An assumption of the maxF test is that the regression coefficients change but the variance remains the same -- this does not seem suitable for structural change, I have written a paper showing how to construct a test which does not make this assumption and shown how it provides a superior alternative to the max F test in most cases. This paper is linked below
http://ssrn.com/abstract=2471479
All are valid but I think MaxF has most popularity in published work. Maybe treat the other two as "confirmatory data analysis"?
Hi, all three statistics may necessarily not show a break-date significant. In general, the avgF shows relatively less p-value.
First: maxF is better than the other two -- it has more power, as Andrews demonstrated, and as I have personally assessed via simulations and practical examples. Second, the asymptotic critical values given by Andrews are not very good in finite samples, but you can easily compute accurate critical values by simulatons. Third. An assumption of the maxF test is that the regression coefficients change but the variance remains the same -- this does not seem suitable for structural change, I have written a paper showing how to construct a test which does not make this assumption and shown how it provides a superior alternative to the max F test in most cases. This paper is linked below
http://ssrn.com/abstract=2471479
The Quandt–Andrews test is a modified version of Chow test that allows for dominant endogenous breakpoints in the sample for an estimated equation. The maximum (max F), average (ave F) and exponential (exp F) test statistics can be used in this test. The null hypothesis of no break is rejected if these test statistics are large; however Hansen (1997) derives an algorithm to compute approximate asymptotic p-values of these tests. Multiple breakpoint testing is available in EViews 8. For more details, you should consult: Bai and Perron, 2003 “Computation and Analysis of Multiple Structural Change Models,” Journal of Applied Econometrics, 6, 72–78.
Which structural breaks has another method to detect some unknown breakpoints? If yes,what are them? And which is better method?
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series