Robust Standard Errors Calculation

[MG250]

Hi,

I was wondering if on the exam we are required to know how to calculate standard errors assuming both homoskedasticity and heteroskedasticity for linear regression. Is there a formula for the latter? If not, I assume we just need to understand when to use robust standard errors (when heteroskedasticity occurs).

Thanks,
Mike

 

[ShaktiRathore]

Hi,
Yes there is formula for the heteroskedasticity robust standard errors: http://www3.grips.ac.jp/~yamanota/Lecture_Note_9_Heteroskedasticity.
When there is heteroskedasticity you need to use the heteroskedasticity robust standard errors. Is there is no heteroskedasticity or homoskedasticity then you can use the formula under homoskedasticity,
SE(slope)=sqrt[(var(e)/var(X))*(1/n-1)]
SE(intercept)=sqrt[(var(e)/var(X))*(1/n-1)*(ΣX^2/n))]
thanks

 

[MG250]

Thanks!

Is this an exam requirement? I didn't see it related to an LO in the readings. Robust Standard errors come up when discussing this LO:

LO: Evaluate the implications of homoskedasticity and heteroskedasticity.

 

[ShaktiRathore]

hi,
Yes you dont need to memorise the heteroskedasticity robust standard errors.I think you just need to know whats the implications of heteroskedasticity on standard errors like they become biased so you need to account for heteroskedasticity and correct them.
thanks

 

[MG250]

Great. Thanks for your help!