Estimating the variance of the OLS estimator:
Summary
- Setup:
- a linear regression model \(y=Xβ+ε\) with a random sample of size \(n\)
- the Gauss-Markov assumptions, \(E\left( ε \right|X)=0\) and \(Var\left( ε \right|X)=σ^2I\)
- \(b={\left( X'X \right)}^{-1}X'y\) is the OLS estimator of \(β\) (both are \(k×1\) )
- \(e=y-Xb\) are the OLS residuals.
- \(s^2=e'e/(n-k)\)
- \(Var\left( b \right|X)=σ^2{\left( X'X \right)}^{-1}\)
- We define the estimated (conditional) variance of \(b\) as
\[\widehat{Var}\left( b \right|X)=s^2{\left( X'X \right)}^{-1}\]
- Result: \(\widehat{Var}\left( b \right|X)\) is an unbiased estimator of \(Var\left( b \right|X)\) .