0
$\begingroup$

I am studying Time series.

I am trying to identify if the pattern in the plot of residuals vs order to identify the type of autocorrelaiton:

enter image description here

I think that it has a negative autocorrelation because almost every negative error is followed by a positive error and vice versa. However, analyzing some examples on the internet I am now confused.

Can anyone help me on this?

$\endgroup$
  • $\begingroup$ please post the actual data ... and your model $\endgroup$ – IrishStat Mar 19 '17 at 12:23
  • $\begingroup$ I only need to identify the pattern to check if autocorrelation is positive negative or no autocorrelation by analysing the plot of the residuals vs time. I am confused because it does not have a cyclic pattern and I think that is negative. But I saw that to be negative all negative erros should follow a positive error and sometimes it does not happen. $\endgroup$ – user290335 Mar 19 '17 at 12:27
2
$\begingroup$

That seems pretty random to me... Deducting autocorrelation from this simple kind of graph is just terrible, just calculate it and put it on a ACF (and maybe PACF also) graph. There's much to say about autocorrelation and time series models, so I won't write any longer and I'll just let you keep studying them.

$\endgroup$
  • $\begingroup$ very well said ! $\endgroup$ – IrishStat Mar 19 '17 at 13:14
0
$\begingroup$

Seems random, but I think is a bad idea analyzing autocorrelation just with a graph (is a kind of... Subjective)

You can try with formal tests (runs, ACF, or Durbin Watson)

You are describing a runs test (viewing the sign of the errors), but as I said, that is subjective IF you just numbered the signs. At runs, you reject randomness if you have a number Of runs too small or too large. (But how much is that? You need a p-value or a rejection region)

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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