1
$\begingroup$

I am attempting to use a significant autocorrelation (where it lies outside a 95% interval around 0) indicating periodicity of a signal and use it as predictive variable in a regression.

If, for example, there's a significant autocorrelation at frequency 3, does this mean that I can use the time series as predictor of itself by placing together the last signal with a 3 period lag? Or as I have read autocorrelation does not indicate where in time the significant periodicity occurs just the how frequently it occurs. However if I have 3 periods lag, I have 3 choices - is this correct?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

If you want to identify autoregressive model for your time series then you can use sample partial autocorrelation function (SPACF).

If it shows that SPACF dies after one lag you can try to make forecasts with AR(1) model.

Suppose then that your series shows seasonality with cycle of k periods. If your SPACF shows that there are significant peaks at lags 1 and k then you can try AR-model with lags at 1 and k.

Improve this answer
$\endgroup$
1
  • 2
    $\begingroup$ The fact is that I want to use the term for a regression against the return in a multimodel ensemble however not as an AR term for Arima. And I was wondering if the term could be used in this way. Thanks for the hint $\endgroup$
    – Barnaby
    Commented Jan 20, 2014 at 13:57

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.