7

could you please help me to understand when I should use either white or newey-west standard erros?

I do work with time-series data. I look at both contemporaneous and intertemporal realtionships.

My understanding is that NW s.e. are appropriate when I look at an intertemporal realtionship and one variable is constracted using overalapping observations.

example I have y(t) and x(t) for t=1,...,T.

I construct y3(t) as y(t) + y(t+1) + y(t+2) for t=2:T-2, and then I regress y3(t+1) on x(t).

For this kind of regressions, it is common to use NW s.e., in particular if x is highly persistent.

Can you please help me to understand more in general how to choose between White and NW s.e. (or other s.e. appropriate for time-series analysis)?

Thanks a lot!

frame
  • 91
  • 1
  • 2

1 Answers1

2

White standard errors correct for heteroscedasticity only, whereas Newey West standard errors are robust to both Heteorscedasticity and Autocorrelation (HAC) .

ChinG
  • 1,633
  • 14
  • 20
  • 2
    "White standard errors correct for heteroscedasticity only" ... Well yes, they don't correct for autocorrelation, but even for heteroscedasticity they are not always reliable. There is a useful discussion in Angrist & Pischke's Mostly Harmless Econometrics (http://www.mostlyharmlesseconometrics.com/) pp 294-307 identifying two limitations: bias when the sample size is small, and a higher sampling variance than conventional standard errors. – Adam Bailey Feb 11 '16 at 09:46