1
$\begingroup$

I am reading Yatchew's (2003) Semiparametric Regression for the Applied Econometrician.

On page 2 and 3, he considers a partial linear model of the form: $$y_{i} = z_{i} \beta + f(x_{i}) + \varepsilon_{i}$$ where he assumes that $x_{i}$ is uniformly distributed over the unit interval, and $\mathbb{E}[z_{i} | x_{i}] = g(x_{i})$. Yatchew shows that if the first derivatives of $f$ and $g$ are bounded then we can estimate $\beta$ using OLS on the first-difference. In particular he shows that, if $z_{i} = g(x_{i})+u_{i}$, then: $$y_{i} - y_{i-1} \approx (u_{i} - u_{i-1})\beta + (\varepsilon_{i} - \varepsilon_{i-1})$$ So that we can estimate $\beta$ by $$\hat{\beta}_{diff} = \frac{\frac{1}{n}\sum_{i=2}^{n} (y_{i}-y_{i-1})(z_{i}-z_{i-1})}{\frac{1}{n}\sum_{i=2}^{n} (z_{i} -z_{i-1})^{2}} \approx \beta + \frac{\frac{1}{n}\sum_{i=2}^{n} (\varepsilon_{i} - \varepsilon_{i-1})(u_{i} - u_{i-1})}{\frac{1}{n}\sum_{i=2}^{n} (u_{i} - u_{i-1})^{2}}\\$$ Now, Yatchew (2003) claims that the denominator in the fraction on the right converges to $2\sigma_{u}^{2}$ (and I agree), and that the numerator has mean $0$ (I agree) and variance $6 \sigma_{\varepsilon}^{2} \sigma_{u}^{2}$.

How does how does he get that the variance is $6 \sigma_{\varepsilon}^{2} \sigma_{u}^{2}$? I have tried to derive this for a long time, but consistently get that the variance is $4 \sigma_{\varepsilon}^{2} \sigma_{u}^{2}$. Can anyone show it?

This result is needed to derive the asymptotic distribution of the estimator $\hat{\beta}_{diff}$.

Any help is much appreciated.

$\endgroup$
1
$\begingroup$

\begin{align*} &Var \left( \frac{1}{n}\sum_{i=2}^{n} (u_{i} - u_{i-1})(\varepsilon_{i} - \varepsilon_{i-1}) \right)\\ \\ &= Var \left( \frac{1}{n} \sum_{i=2}^{n} (u_{i}\varepsilon_{i} - u_{i}\varepsilon_{i-1} - u_{i-1}\varepsilon_{i} + u_{i-1}\varepsilon_{i-1}) \right)\\ \\ & =\frac{1}{n^{2}} Var \left( u_{n}\varepsilon_{n}+2 \sum_{i=2}^{n-2} u_{i}\varepsilon_{i} + u_{1}\varepsilon_{1} - \sum_{i=2}^{n} u_{i}\varepsilon_{i-1} - \sum_{i=2}^{n} u_{i-1}\varepsilon_{i} \right)\\ \\ & =\frac{1}{n^{2}} \left( \sigma_{u}^{2}\sigma_{\varepsilon}^{2} +4 \sum_{i=2}^{n-2} \sigma_{u}^{2}\sigma_{\varepsilon}^{2} + \sigma_{u}^{2}\sigma_{\varepsilon}^{2} + \sum_{i=2}^{n} \sigma_{u}^{2}\sigma_{\varepsilon}^{2} + \sum_{i=2}^{n} \sigma_{u}^{2}\sigma_{\varepsilon}^{2} \right)\\ \\ & =\frac{1}{n^{2}} \left( \sigma_{u}^{2}\sigma_{\varepsilon}^{2} +4 (n-2) \sigma_{u}^{2}\sigma_{\varepsilon}^{2} + \sigma_{u}^{2}\sigma_{\varepsilon}^{2} + (n-1) \sigma_{u}^{2}\sigma_{\varepsilon}^{2} + (n-1)\sigma_{u}^{2}\sigma_{\varepsilon}^{2} \right)\\ \\ & =\frac{1}{n^{2}} (6n-8) \sigma_{u}^{2}\sigma_{\varepsilon}^{2} \\ \\ &\approx \frac{6 \sigma_{u}^{2}\sigma_{\varepsilon}^{2}}{n} \\ \end{align*} where the last line is for large $n$.

$\endgroup$

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.