Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I'm taking this example from a paper I'm reading. I'm having trouble understanding the logic, and I'm hoping someone can help me.

Let $ v_{t}=a \ x_{t}+u_{t} $

where $a$ is some constant and $u$ is a mean zero random variable such that $E[u_{t}x_{t}]=0$ for all $t$.

If $E[x_{t-1} \, x_{t}]=0$ and $E[u_{t} \, x_{t-1}]=0$ then $E[x_{t-1} \, v_{t}] = 0$ This part I can clearly see.

Now Suppose $x_t= p\, x_{t-1} +e_t$ where $e$ is a mean zero white noise term and $|p|<1$.

Now we have $E[x_{t-1} \, v_{t}] = a \, p\, {\rm Var}(x_t) \neq 0$ even if $E[u_{t}\, x_{t-1}]=0$ (Still good)

But, then it goes on to say that if: $ E[x_{t-1}\, u_{t} | x_t]=0$, then we get that $E[x_{t-1}\, v_{t} | x_t] = 0$ and I do not see where that comes from, we still have the same problem as before.

$E[x_{t-1}\, v_{t} | x_t] = E[x_{t-1} (a \, x_{t} + u_{t})|x_t] = a \, x_t x_{t-1} \neq 0$

Am I missing something?

share|improve this question
    
$E[x_{t-1}v_{t}] = ap\operatorname{var}(x_t)$ seems wrong. I get $$E[x_{t-1}v_{t}] =E[x_{t-1}(ax_t+u_t)]= E[x_{t-1}(a(px_{t-1}+e_t)+u_t)]=apE[x_{t-1}^2]$$ (assuming $E[x_{t-1}e_t]=0$). That will generally not be $\operatorname{var}(x_t)$. –  joriki Oct 16 '12 at 5:43
    
@joriki Which equals $ap\mathrm{var}(x_t)$ if the process $(x_t)_t$ is stationary. This hypothesis seems to be missing. As a matter of fact, I suggest that the OP rewrites the question to make clear the exact list of hypotheses which hold when one asks for $E(x_{t-1}v_t\mid x_t)$. –  Did Oct 17 '12 at 7:04
    
Edited to add $|p|<1$, which I believe makes the process stationary. All other assumptions/hypotheses given to me are in the question. –  Greg Oct 17 '12 at 17:19
    
Are you sure you got the indexes right? Are you conditioning $x_{t-1} u_t$ on the "future" ? –  leonbloy Oct 17 '12 at 17:46

1 Answer 1

up vote 1 down vote accepted

When you are given $x_t$, you must treat it as a constant. Then

$$E[x_{t−1}(a x_t+u_t)|x_t] = a x_t E[x_{t-1} | x_{t}] + E[x_{t-1} u_t | x_t]$$

The second term is zero by hypothesis. But the other gives $a\, p \, x_t^2$ (assuming a process that is not only stationary but also time-reversible - which is the case e.g. is the white noise is gaussian). So, it seems there is some error

share|improve this answer

Your Answer

 
discard

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.