If we assumed that $y \sim N(X\beta,S)$ where

S= $\sigma^2\begin{bmatrix} 1 & \rho & \rho &...\\ \rho & 1 & \rho &...\\ \rho & \rho & 1 &...\\ \rho & \rho & \rho & ...\\ ... & ... & ... & ...\\ \end{bmatrix}$

Then the OLS approximation of $\beta$ would be the same as in the independent case:

$\hat{\beta} = (X'X)^{-1}X'y$.

However, the distribution of $\hat{\beta}$ seems to become harder to determine.

The variance of $\hat{\beta}$ would be $Var((X'X)^{-1}X'y)$ = $(X'X)^{-1}X'SX(X'X)^{-1}$. How can I further simplify this?


(On a side note, this is a case of correlated errors -residuals are always correlated).

You are considering a very specific form of correlation, equicorrelation. Then note that

$$S=\sigma^2[(1-\rho)I+\rho\mathbf i \mathbf i']$$

where $I$ is the identity matrix and $\mathbf i = (1,...,1)'$.

It follows that

$$\text{Var}\left(\hat \beta_{OLS}\mid X\right) = \sigma^2(X'X)^{-1}X'\big[(1-\rho)I_p+\rho\mathbf i \mathbf i'\big]X(X'X)^{-1}$$

$$=(1-\rho)\sigma^2(X'X)^{-1}X'X(X'X)^{-1} + \rho\sigma^2(X'X)^{-1}X'\mathbf i \mathbf i'X(X'X)^{-1}$$

$$=(1-\rho)\sigma^2(X'X)^{-1} + \rho\sigma^2\Big[(X'X)^{-1}X'\mathbf i\Big] \Big[(X'X)^{-1}X'\mathbf i\Big]'$$

Further, on can show that, if the regressor matrix includes a constant term, then

$$\Big[(X'X)^{-1}X'\mathbf i\Big] = (1, \mathbf 0)'$$

so we arrive at

$$\text{Var}\left(\hat \beta_{OLS}\mid X\right) = (1-\rho)\sigma^2(X'X)^{-1} + \rho\sigma^2\cdot \left[ \begin{matrix} 1 & \mathbf 0 \\ \mathbf 0 & \mathbf 0 \end{matrix} \right]$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.