Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. 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

In the AR(1) model $y_{t}=\beta_{0}+\beta_{1}y_{t-1}+u_{t}$, assuming $E(u_{t-1}|y_{t-1},y_{t-2}...)=0$, how does the law of iterated expectations ensure that the errors must be uncorrelated: $E(u_{t}u_{s}|x_{t},x_{s})=0$?

share|improve this question
Is this a homework problem? In any event, the statement that errors are uncorrelated is an assumption of the AR(1) model, not a (genuine) derivable result. – Arthur Small Jan 14 '13 at 20:36
No not homework - reading through Woolridge intro - page 385. It is stated that given AR(1) and the assumption that the error term has zero mean given all past values of y, the errors are uncorrelated. – B_Miner Jan 14 '13 at 20:42
up vote 5 down vote accepted

Consider $s \neq t$. The law of iterated expectations tells us that

$$\begin{equation*} \mathbb{E}[u_su_t \mid y_{s-1}, y_{t-1}] = \mathbb{E}_{u_s}\left[\mathbb{E}[u_su_t \mid u_s, y_{s-1}, y_{t-1}]\right] = \mathbb{E}_{u_s}\left[u_s\mathbb{E}[u_t \mid u_s, y_{s-1}, y_{t-1}]\right]. \end{equation*} $$

Suppose that $s < t$, with the opposite case following by symmetry.

The AR(1) process has the property that, for $y_t = \beta_0 + \beta_1 y_{t-1} + u_t$, $\mathbb{E}[u_t \mid y_{t-1}, y_{t-2}, \dots] = \mathbb{E}[u_t \mid y_{t-1}] = 0$. In words, if I know what happened last period, then nothing else in the past can help me guess what happens today.

Rewriting our AR(1) equation reveals that $u_s = y_s - \beta_0 - \beta_1 y_{s-1}$; there is a one-to-one transformation betweent $u_s$ and $y_s$. As a result, conditioning on $u_s$ is equivalent to conditioning on $y_s$.

Using this relationship and our fact about the AR(1) process, we find that $$\begin{equation*} \mathbb{E}[u_t \mid u_s, y_{s-1}, y_{t-1}] = \mathbb{E}[u_t \mid y_s, y_{s-1}, y_{t-1}] = \mathbb{E}[u_t \mid y_{t-1}] = 0. \end{equation*} $$

Plugging this in to the LIE result above gives that $$\begin{equation*} \mathbb{E}_{u_s}\left[u_s\mathbb{E}[u_t \mid u_s, y_{s-1}, y_{t-1}]\right] = \mathbb{E}_{u_s}\left[u_s \times 0\right] = 0. \end{equation*} $$

share|improve this answer
In the first part, the conditioning is on $y_{s-1},y_{t-1}$ correct? And when you condition on on $u_{s}$ that amounts to making $u_{s}$ a constant? – B_Miner Jan 25 '13 at 0:47
Yes to both points. – Charlie Jan 25 '13 at 15:44

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.