Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I have the following multiple regression model with autocorrelated error terms:



where $u_t$'s are independent $N(0,\sigma^2)$ disturbance terms and $|\rho|<1$. Also, although the $\varepsilon_t$'s are correlated over time, they still have mean 0 and constant variance: $E(\varepsilon_t)=0, \sigma^2(\varepsilon_t)=\frac{\sigma^2}{1-\rho^2}$ with the $\sigma^2$ in the variance, the variance of the $u_t$ disturbance terms.

I need to derive and simplify the term for $Cov(\varepsilon_t,\varepsilon_{t-2})$

This is what I have so far:

$Cov(\varepsilon_t,\varepsilon_{t-2})=E[(\varepsilon_t-\mu_t)(\varepsilon_{t+2}-\mu_{t+2})]=E(\varepsilon_t\varepsilon_{t-2})=E[(\rho\varepsilon_{t-1}+u_t)(\rho\varepsilon_{t-3}+u_{t-2})]=E(\rho^2\varepsilon_{t-1}\varepsilon_{t-3})+E(\rho\varepsilon_{t-1}u_t)+E(\rho\varepsilon_{t-3}u_{t-2})+E(u_tu_{t-2})=\rho^2E(\varepsilon_{t-1}\varepsilon_{t-3})+\rho E(\varepsilon_{t-1}u_t)+\rho E(\varepsilon_{t-3}u_{t-2})+0$

I'm not sure where to go from here. With the $E(\varepsilon_{t-1}\varepsilon_{t-3})$ term, is this not a similar situation as in the beginning, with $E(\varepsilon_t\varepsilon_{t-2})$?

I believe that I should end up with $Cov(\varepsilon_t,\varepsilon_{t-2})=\rho^2(\frac{\sigma^2}{1-\rho^2})$. Am I taking the right approach? Any guidance would be greatly appreciated!

share|cite|improve this question
up vote 2 down vote accepted

Applying the formula $\text{cov} \left( \varepsilon_t, \varepsilon_{t - k} \right) = E \left[ \varepsilon_t \varepsilon_{t - k} \right] - E \left[ \varepsilon_t \right] E \left[ \varepsilon_{t - k} \right]$ for any $k > 0$ and using the fact that $E \left[ \varepsilon_t \right] = E \left[ \varepsilon_{t - k} \right] = 0$, we deduce \begin{eqnarray*} \text{cov} \left( \varepsilon_t, \varepsilon_{t - 2} \right) & = & E \left[ \left( \rho \varepsilon_{t - 1} + u_t \right) \varepsilon_{t - 2} \right]\\ & = & \rho E \left[ \varepsilon_{t - 1} \varepsilon_{t - 2} \right] + E \left[ u_t \varepsilon_{t - 2} \right]\\ & = & \rho E \left[ \varepsilon_{t - 1} \varepsilon_{t - 2} \right] \end{eqnarray*} where we used the fact that $u_t$ is independent of $\varepsilon_t$ (innovation process). We need $E \left[ \varepsilon_{t - 1} \varepsilon_{t - 2} \right]$. We know by stationarity that it is equal to $E \left[ \varepsilon_t \varepsilon_{t - 1} \right] = \rho E \left[ \varepsilon_{t - 1}^2 \right] + E \left[ u_t \varepsilon_{t - 1} \right] = \rho \frac{\sigma^2}{1 - \rho^2}$.

Thus the answer is $\text{cov} \left( \varepsilon_t, \varepsilon_{t - 2} \right) = \rho^2 \frac{\sigma^2}{1 - \rho^2}$.

share|cite|improve this answer
@Jess. Do you understand (or know how to prove) why $u_t$ is independent of $\varepsilon_t$? – Learner Mar 13 '13 at 11:54
Yes. I don't know why I didn't see that initially. Thank you so much for the help! It makes everything a lot clearer. – Jess Mar 14 '13 at 20:10

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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