Timeline for Why do autocorrelation and heteroskedasticity under-report the sample variance of OLS estimates?

9 events

when toggle format	what		by	license	comment
Jan 6, 2020 at 20:45	vote	accept	CuriousMind
Apr 13, 2017 at 12:44	history	edited	CommunityBot		replaced http://stats.stackexchange.com/ with https://stats.stackexchange.com/
Mar 30, 2015 at 17:53	vote	accept	CuriousMind
Mar 30, 2015 at 17:53
Mar 29, 2015 at 10:11	comment	added	Christoph Hanck		@gung, I share the concerns of Mico that, most of the time, the default standard errors are too small, leading to inflated t-statistics and therefore overly liberal tests. This is not a general result, though. If, for instance, $E(\epsilon_i^2) = bx^{−2}_i$, $b>0$, we may also have that standard errors are too small, leading to conservative tests. So it can go either way, I'd say.
Mar 29, 2015 at 8:34	comment	added	Mico		Are you sure about your answer regarding what heteroskedasticity in the errors does -- specifically, that OLS-based SEs are larger than they should be if heteroskedasticity is present? At the very least, you should be more specific about what forms of heteroskedasticity you're allowing.
Mar 29, 2015 at 5:07	history	edited	gung - Reinstate Monica	CC BY-SA 3.0	added 1 character in body
Mar 29, 2015 at 5:02	history	edited	gung - Reinstate Monica	CC BY-SA 3.0	added 1 character in body
Mar 29, 2015 at 4:56	history	edited	gung - Reinstate Monica	CC BY-SA 3.0	added 167 characters in body
Mar 29, 2015 at 4:44	history	answered	gung - Reinstate Monica	CC BY-SA 3.0