5
$\begingroup$

I intend to use Pesaran's (2006) common correlated effects pooled (CCEP) estimator. However, I'm not yet very familiar with advanced econometrics and advanced use of eviews. More specifically I want to estimate this model: \begin{equation} y_{it} = \alpha_{i} + \beta_{1}x_{1,it} + \beta_{2}x_{2,it}+\gamma_{i}F_{t}+\epsilon_{it} \end{equation} in which $F_{t}$ is an unobserved common factor and $\gamma_{i}$ is a country-specific factor loading. We were taught that $F_{t}$ can be proxied by: \begin{equation} F_{t}=\frac{(\bar{y_{t}}-\bar{\alpha}-\beta_{1} \bar{x}_{{1,t}} -\beta_{2} \bar{x}_{{2,t}}-\bar{\epsilon}_{{t}})}{\bar{\gamma}}, \end{equation} in which $\bar{y_{t}} = \frac{1}{N}\sum_{i=1}^{N} y_{it}$, and $\bar{\gamma_{}}=\frac{1}{N}\sum_{i=1}^{N} \gamma_{i}$, with $N$ the number of cross-sections.

Substituting the second equation into the first yields: \begin{equation} y_{it} = \alpha_{i}-\frac{\gamma_{i}}{\bar{\gamma}}+\beta_{1} x_{1,it} +\beta_{2} x_{2,it} +\frac{\gamma_{i}}{\bar{\gamma}}\bar{y_{t}} - \beta_{1} \frac{\gamma_{i}}{\bar{\gamma}} \bar{x}_{1,t} -\beta_{2}\frac{\gamma_{i}}{\bar{\gamma}}\bar{x}_{2,t}+\epsilon_{it}-\frac{\gamma_{i}}{\bar{\gamma}}\bar{\epsilon_{t}} \end{equation} or with $\alpha_{i}-\frac{\gamma_{i}}{\bar{\gamma}}=\alpha'_{i}$ and $\frac{\gamma_{i}}{\bar{\gamma}}=\gamma'_{i}$: \begin{equation} y_{it} = \alpha'_{i} +\beta_{1} x_{1,it} +\beta_{2} x_{2,it} +\gamma'_{i}\bar{y_{t}} - \beta_{1} \gamma'_{i} \bar{x}_{1,t} -\beta_{2}\gamma'_{i}\bar{x}_{2,t} +\epsilon_{it}-\gamma'_{i}\bar{\epsilon_{t}}. \end{equation}

To estimate this in eviews, I had the following idea

The cross-sectional averages $\bar{y_{t}}$, $\bar{x}_{{1,t}}$, and $\bar{x}_{{2,t}}$ can be easily calculated from the dataset. I would use cross-sectional fixed effects to estimate all $\alpha'_{i}$. Next, I would need $N$ terms to estimate all $N$ $\gamma'_{i}$. To do this, I would include these $N$ terms: $\gamma'_{A}\bar{y}_{t}dum_{A} + \gamma'_{B}\bar{y}_{t}dum_{B} + ... + \gamma'_{N}\bar{y}_{t}dum_{N}$, in which each capital letter denotes one of the $N$ cross-sections and the dummy variable takes the value of $1$ once for each cross-section. Then, for each averaged explanatory variable, $\bar{x}_{1t}$ and $\bar{x}_{2t}$, I would include these $2 \times N$ terms: $\beta_{1}\gamma'_{A}\bar{x}_{1,t} + \beta_{1}\gamma'_{B}\bar{x}_{1,t}+ ... + \beta_{1}\gamma'_{N}\bar{x}_{1,t}$ and $\beta_{2}\gamma'_{A}\bar{x}_{2,t} + \beta_{2}\gamma'_{B}\bar{x}_{2,t}+ ... + \beta_{2}\gamma'_{N}\bar{x}_{2,t}$.

So, to sum up, my suggested input for eviews (to estimate with cross-sectional fixed effects) is the following: y = c(1)*x1 + c(2)*x2 + c(3)*y_avg*dumA + c(4)*y_avg*dumB + c(5)*y_avg*dumC + ... + c(1)*c(3)*x1_avg + c(1)*c(4)*x1_avg + c(1)*c(5)*x1_avg + ... + c(2)*c(3)*x2_avg + c(2)*c(4)*x2_avg + c(2)*c(5)*x2_avg + .....

In this equation:

  • c(1) = $\beta_{1}$;
  • c(2) = $\beta_{2}$;
  • c(3) = $\gamma'_{A}$;
  • c(4) = $\gamma'_{B}$;
  • c(5) = $\gamma'_{C}$.

These are my questions regarding this estimation:

  • First of all, confirmation of the correctness of my derivation would be welcome;
  • Would the estimation in eviews I suggest do the trick?
  • If so, should I include an intercept in the fixed-effects estimation?
  • If not, is there an alternative procedure to implement the CCEP estimator in eviews?
  • The estimated error terms should be $\epsilon-\gamma'_{i}\bar{\epsilon}$, is this structure automatically obtained? Or should this be imposed one way or another?
  • The same question for $\alpha'_{i}$: it should equal $\alpha_{i}-\frac{\gamma_{i}}{\bar{\gamma}}$. Should this condition be imposed, or is it automatically fulfilled when inputting my suggested input in eviews;
  • Other suggestions regarding the use of CCEP estimator in eviews are certainly welcome.

Any help, also partial answers, is appreciated!

$\endgroup$
  • $\begingroup$ Since this question has received no answers... Have you considered posting this on the official EViews user forum? Hint: that's a very good resource for users of EViews. $\endgroup$ – Graeme Walsh Feb 28 '16 at 19:43
  • 1
    $\begingroup$ Indeed, yesterday I decided that the question may actually be more fitted to the eviews forum than to econimics.se. This is the link to the post: forums.eviews.com/viewtopic.php?f=7&t=14075. If any answer appears there, I'll bring it to this site as well. Thanks for the hint though! $\endgroup$ – Wecon Feb 29 '16 at 8:57
3
$\begingroup$

The technique described in the question is almost correct. Consider a panel data set consisting of three cross-sections ($a$, $b$, and $c$) and three time-periods ($1$, $2$, and $3$). Let y denote the column vector with the observations of the dependent variable, x the column vector with observations of the first explanatory variable, and z the column vector with observations of the second explanatory variable. They take these forms respectively:

$\textbf{y} = \begin{bmatrix} y_{a1} \\ y_{a2} \\ y_{a3} \\ y_{b1} \\ y_{b2} \\ y_{b3} \\ y_{c1} \\ y_{c2} \\ y_{c3} \end{bmatrix}$ ; $\textbf{x} = \begin{bmatrix} x_{a1} \\ x_{a2} \\ x_{a3} \\ x_{b1} \\ x_{b2} \\ x_{b3} \\ x_{c1} \\ x_{c2} \\ x_{c3} \end{bmatrix}$; $\textbf{z} = \begin{bmatrix} z_{a1} \\ z_{a2} \\ z_{a3} \\ z_{b1} \\ z_{b2} \\ z_{b3} \\ z_{c1} \\ z_{c2} \\ z_{c3} \end{bmatrix}$.

Let $\bar{y}_{i} = \frac{1}{3}(y_{ai} + y_{bi}+y_{ci})$, with $i = 1, 2, 3$, and equivalently for $x$ and $z$: $\bar{x}_{i} = \frac{1}{3}(x_{ai} + x_{bi}+x_{ci})$ and $\bar{z}_{i} = \frac{1}{3}(z_{ai} + z_{bi}+z_{ci})$, both for $i = 1, 2, 3$.

This is the correct way to get the CCEP estimator allowing for one common factor, as in the model described in the question: \begin{equation*} \begin{split} \begin{bmatrix} y_{a1} \\ y_{a2} \\ y_{a3} \\ y_{b1} \\ y_{b2} \\ y_{b3} \\ y_{c1} \\ y_{c2} \\ y_{c3} \end{bmatrix} = \beta_{1} \begin{bmatrix} x_{a1} \\ x_{a2} \\ x_{a3} \\ x_{b1} \\ x_{b2} \\ x_{b3} \\ x_{c1} \\ x_{c2} \\ x_{c3} \end{bmatrix}+\beta_{2}\begin{bmatrix} z_{a1} \\ z_{a2} \\ z_{a3} \\ z_{b1} \\ z_{b2} \\ z_{b3} \\ z_{c1} \\ z_{c2} \\ z_{c3} \end{bmatrix}+ \gamma_{a} \begin{bmatrix} \bar{y}_{1} \\ \bar{y}_{2} \\ \bar{y}_{3} \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \end{bmatrix} +\gamma_{b} \begin{bmatrix} 0 \\ 0 \\ 0 \\ \bar{y}_{1} \\ \bar{y}_{2} \\ \bar{y}_{3} \\ 0 \\ 0 \\ 0 \end{bmatrix} + \gamma_{c} \begin{bmatrix} 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ \bar{y}_{1} \\ \bar{y}_{2} \\ \bar{y}_{3} \end{bmatrix} \\ - \beta_{1}\gamma_{a} \begin{bmatrix} \bar{x}_{1} \\ \bar{x}_{2} \\ \bar{x}_{3} \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \end{bmatrix} - \beta_{1}\gamma_{b} \begin{bmatrix} 0 \\ 0 \\ 0 \\ \bar{x}_{1} \\ \bar{x}_{2} \\ \bar{x}_{3} \\ 0 \\ 0 \\ 0 \end{bmatrix} -\beta_{1} \gamma_{c} \begin{bmatrix} 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ \bar{x}_{1} \\ \bar{x}_{2} \\ \bar{x}_{3} \end{bmatrix} -\beta_{2} \gamma_{a} \begin{bmatrix} \bar{z}_{1} \\ \bar{z}_{2} \\ \bar{z}_{3} \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \end{bmatrix} -\beta_{2} \gamma_{b} \begin{bmatrix} 0 \\ 0 \\ 0 \\ \bar{z}_{1} \\ \bar{z}_{2} \\ \bar{z}_{3} \\ 0 \\ 0 \\ 0 \end{bmatrix} -\beta_{2} \gamma_{c} \begin{bmatrix} 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ 0 \\ \bar{z}_{1} \\ \bar{z}_{2} \\ \bar{z}_{3} \end{bmatrix} \\ + \phi_{a} + \phi_{b} + \phi_{c} + \mu_{it} \end{split} \end{equation*} Here, $\phi_{a},\ \phi_{b},$ and $\phi_{c}$ are cross-sectional fixed effects and $\mu_{it}$ a well-behaved error-term that does not require any restrictions.

An interesting remark is that if the six $\beta$s that appear together with the $\gamma$s are not restricted to be equal to each other and equal to the first two $\beta$s, more than one common factor is allowed.

$\endgroup$

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.