3
$\begingroup$

Suppose $ X_t = (\alpha + \beta{t})S_{t} + e_{t}$.

Define $C_{12} = 1 - B^{12}$ as the backshift operator.

Assume $ S_{t}=S_{t-12}$ for all $t$.

Assume that $e_{t}$ is white noise.

Is $C*X_{t}$ stationary? If not, give a different backshift operator that makes the series stationary.

When I solved this I got: $E[CX_{t}] = 12*\beta*S_{t}$ which is a constant.

Also I got that

$$var(CX_{t})= var((\alpha + \beta{t})S_{t} + e_{t} - (\alpha*S_{t-12} + \beta*t*S_{t} - S_{t}*12*\beta + e_{t-12})) $$ $$ = var(e_t +12 \beta S_t - e_{t-12}) = cov(e_{t},e_{t+12})$$

I use $S_{t}$ since $S_{t} = S_{t-12}$ for all $t$.

Apparently this is wrong as the series is nonstationary using this backshift operator.

What am I doing wrong? How is this series nonstationary and how do I find the correct backshift operator that makes it stationary?

$\endgroup$
10
  • $\begingroup$ Your variance looks like it has a typo, or rather two typos. $\endgroup$
    – Glen_b
    Mar 11, 2013 at 4:17
  • $\begingroup$ What do you mean by $var(e_{t},e_{t+12})$? $\endgroup$
    – Glen_b
    Mar 11, 2013 at 4:54
  • $\begingroup$ It's the covariance between the white noise at time t and the white noise at time t+12. And this obviously is not dependent on t only on the lag between t and t+12. I don't how I can make myself more clear $\endgroup$
    – phil12
    Mar 11, 2013 at 4:56
  • $\begingroup$ How do you get $var((\alpha + \beta{t})S_{t} + e_{t} - (\alpha*S_{t+12} + \beta*t*S_{t} + S_{t}*12*\beta + e_{t+12})) = cov(e_{t},e_{t+12})$? $\endgroup$
    – Glen_b
    Mar 11, 2013 at 4:57
  • $\begingroup$ After differencing, you get $var(e_{t} -S_{t}12\beta - e_{t+12})$ and this is constant for all t. I realize my mistake My notation was bad. $\endgroup$
    – phil12
    Mar 11, 2013 at 5:00

1 Answer 1

1
$\begingroup$

Edited to reflect the changed question:

$$C_{12}X_{t} = 12 \beta S_{t-12} + e_t - e_{t-12}$$

The distribution of $e_t - e_{t-12}$ is not changing with time, but note that the mean is changing with time, since $S_t$ isn't constant.

To figure out what operator would make the original series stationary, consider what you'd do to $12 \beta S_{t-12} + e_t - e_{t-12}$ to make it stationary

$\endgroup$
8
  • $\begingroup$ What convinces you that the solution is correct? $\endgroup$
    – Glen_b
    Mar 11, 2013 at 4:28
  • $\begingroup$ "I lost points" only indicates that the person allocating points thought it was non stationary. Do you have some convincing explanation as to why you should agree? (NB: Are you sure you transcribed everything correctly? Please double check, especially your variance.) $\endgroup$
    – Glen_b
    Mar 11, 2013 at 4:31
  • $\begingroup$ et is white noise. I think my explanation makes sense. The variance simply depends on the lag and the mean is constant since St = St-1 $\endgroup$
    – phil12
    Mar 11, 2013 at 4:34
  • $\begingroup$ That's really weird. Thanks for clarifying my understanding of my problem. All the information I gave you was correct. $\endgroup$
    – phil12
    Mar 11, 2013 at 21:57
  • $\begingroup$ It's always possible I've made an error too (I do that sometimes), but in this case I doubt it. Mistakes creep in to even the best-run courses. Can you get a worked solution? $\endgroup$
    – Glen_b
    Mar 11, 2013 at 21:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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