2
$\begingroup$

Consider the Correlated Random Effects model $y_{it} = \alpha + x_{it}\beta + \bar x \gamma + w_i + \epsilon_{it} $ where $x_{it}$ is a scalar explanatory variable.

The correlated random effects GLS estimator $ \hat \beta_{CRE} $ is the OLS estimator of $\beta$ in the quasi-demeaned regression

$\tilde y_{it} = \delta + \tilde x_{it} \beta + \bar x_i \rho + u_{it} $,

where $\tilde y_{it} = y_{it} - \theta \bar y_i , \tilde x_{it} = x_{it} - \theta \bar x_i $ and $\theta = 1 - (\sigma^2_\epsilon/ (\sigma^2_\epsilon + T\sigma^2_w))^{1/2} $

Question: I need to show that the residuals from the regression of $x_{it} - \bar x_i $ on a constant and $\bar x_i $ is just $x_{it} - \bar x_i $ itself.

Attempt: Regress $x_{it} - \bar x_i = \alpha + \bar x_{i} + \tilde r_{it} $ rearrange to get the residuals, $\tilde r_{it} = (x_{it} - \bar x_i) - (\alpha + \bar x_{i})$ I'm not sure how to proceed.

$\endgroup$
2
$\begingroup$

By stnadard OLS regression results, in the simple regression

$$y_t = \alpha + \beta z_t + v_t, \;\;\;t=1,...,T$$

we have that

$$\hat \beta = \frac {\sum_{i=1}^T (z_t - \bar z)(y_t-\bar y)}{\sum_{i=1}^T (z_t - \bar z)^2}$$

and

$$\hat \alpha = \bar y - \hat \beta \bar z$$

So the residuals are

$$\hat v_t = y_t -(\hat \alpha +\hat \beta z_t)$$

Then, substitute for your particular $y_t$ and $z_t$ in all the above.

$\endgroup$
  • $\begingroup$ Just a quick question: Since my explanatory variable is $ \bar x_i $ (averaged over time), what is the value of the average of $\bar x_i$ Do I average over time again? (in which case the average of $\bar x_i$ would be itself). Or do i average over individuals i? (so that the average of $\bar x_i$ is $\bar x$?) I need this for the numerator of $ \hat \beta $ $\endgroup$ – Kai_M Nov 24 '15 at 11:21
  • $\begingroup$ @user44394 Wait a minute, I must be missing something here. If $\bar x_i$ is the time average per cross-section, then you cannot really regress each cross section separately "on a constant and $\bar x_i$", because both are constant,and so perfectly colinear, and we cannot obtain different estimates for each one. $\endgroup$ – Alecos Papadopoulos Nov 24 '15 at 16:09
  • $\begingroup$ I completely agree. I think there could be a typo in my assignment sheet. This question is one part of a longer question which asks me to use the Frisch-Waugh theorem to show that $ \hat \beta_{CRE} = \hat \beta_{FE}$. I found this paper: econ.msu.edu/faculty/wooldridge/docs/cre1_r4.pdf , and I am basically trying to prove proposition 2.1 (bottom page 12) but for balanced panels. The first part of the proof obtains the residuals from the regression of $x_{it} - \theta \bar x_i$ on $(1-\theta) \bar x_i$. So I guess there could be a typo in my assignment sheet here? $\endgroup$ – Kai_M Nov 24 '15 at 16:39

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.