so the formula for calculating the error variance in the ols model is: $σ^2=\frac{\hat{u}^T\hat{u}}{T-k}$. That would be easy to calculate in a practical example, where I have all residuals numerically. Now I unfortunately have an exercise task where only the following is given the (demeaned) linear regression model: $y_t = β_1 x_{t1} + β_2 x_{t2} + u_t$ and the summarized cross products: $\sum_{t=1}^{T} x^2_{t2}$, $\sum_{t=1}^{T} x_{t1} y_t$, $\sum_{t=1}^{T} x_{t2} y_t$, $\sum_{t=1}^{T} x^2_{t1}$, $\sum_{t=1}^{T} y^2_t$, $\sum_{t=1}^{T} x_{t1} x_{t2}$ and T is also given.

Now my first step for calculationg $σ^2$ was: $$σ^2=\frac{\hat{u}^T\hat{u}}{T-k} = \frac{(y-X\hat{β})^T * (y-X\hat{β})}{T-k} = \frac{y^Ty-2\hat {\beta^T}X^Ty+\hat{\beta}X^TX\hat{\beta}}{T-k}$$ Is my approach correct? How do I proceed? I think you can set anything to zero in the last fraction, but I don't quite understand what and why.


Using the definition of inner product $$y^Ty =\overbrace{\sum_{t=1}^{T} y^2_t}^{\textsf{given}} $$ $$\hat {\beta^T}X^Ty = \hat{\beta}_1\overbrace{\sum_{t=1}^{T} x_{t1} y_t}^{\textsf{given}} + \hat{\beta}_2\overbrace{\sum_{t=1}^{T} x_{t2} y_t}^{\textsf{given}} $$

$$\hat{\beta}X^TX\hat{\beta} =\hat{\beta}_1\overbrace{\sum_{t=1}^{T} x_{t1} ^2 }^{\textsf{given}} + \hat{\beta}_2\overbrace{\sum_{t=1}^{T} x_{t2} ^2}^{\textsf{given}} + 2\hat{\beta}_1\hat{\beta}_2\overbrace{\sum_{t=1}^{T} x_{t1} x_{t2}}^{\textsf{given}} $$

and you should compute $\hat{\beta}_1,\hat{\beta}_2$ using OLS.


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.