0

How does one decide on number of lags to use in this test? Can one just justify the number based on previous papers?

To clarify, I am forecasting conditional variance, and have generated my log returns. Having estimated an OLS, I now need to to test if there are ARCH effects.

Should I test at lags 1, 5, 10? Or is 1 lag enough in case to draw conclusions?

Having run lags, 1, 5, 10, F-statistic was very high, but decreases with increasing lags. P value is significant through no matter what lag is used.

Albe
  • 109
  • 1
  • 10
  • Albe, I have answered a highly related question [here](http://stats.stackexchange.com/questions/231987/lag-length-selection-in-arch-lm-test), hopefully you will find it helpful. – Richard Hardy Apr 03 '17 at 08:13
  • Thank you, I am guessing with a case like mine (see last sentence in original question), 1 lag is more than justifiable? – Albe Apr 03 '17 at 14:43
  • Since you get significance in any case, you could report the statistic for all of these lags. However, there is no general answer for which lag is "the best", unless you want to test some specific hypothesis that has subject-matter importance. Meanwhile, if you only need to justify using a GARCH model, all of lags 1, 5 or 10 are relevant (if you were going to use an ARCH(1) model, then perhaps lag 1 would be more relevant). – Richard Hardy Apr 03 '17 at 16:04

0 Answers0