1
$\begingroup$

I am trying to do time series forecasting on different data sets.

I created a lag plot on each set and received an unusual, 'symmetric' but grouped pattern:symmetric lag plot

I would say that this implies a lack of autocorrelation but the separated groups confuse me.

What does this tell about the data?

EDIT: Here is a plot of the same data set itself: time series data
(source: ibin.co)

$\endgroup$
2
  • $\begingroup$ It is strange, yes. A plot of the time series, and some description of what the time series is, might help us. $\endgroup$ Oct 8, 2018 at 18:15
  • $\begingroup$ Be careful when inferring no-trend from scatter-plot point-clouds and non-transparent points. It's impossible to discern the actual frequency of observations that fall in a specific cloud. Use a hexbin or transparent point-fills instead. Or estimate the actual trendline. $\endgroup$
    – AdamO
    Oct 9, 2018 at 15:22

1 Answer 1

3
$\begingroup$

It tells you that the $y(t)$ are grouped around -5, 0 and 5.

You can read this information on the horizontal projection only. The facts that there 9 groups of dots is an unintended but inevitable secondary result.

As the central point seams to be less represented, the variable $y(t)$ and $y(t+1)$ are correlated, but maybe not in a linear way.

To go further (if the data is not simulated), you should first ask if the values $\pm10$, $\pm15$ and $\pm20$ have a meaning or if they are outliers. You can then introduce the variable $y^*(t) = 5* round(\frac {y(t)}{5})$ which is equal to -5, 0 and 5 which ever is the nearest. Then study the randomness $\epsilon(t) = y(t)-y^*(t)$. You can also study the lag correlation between $y^*(t)$ and $y^*(t+1)$ with a probability transition matrix (instead of a graph).

$\endgroup$
1
  • $\begingroup$ This is great, thank you! I also would like to read upon this, however, most of the material I found (e.g. about lag plots) discusses more common cases. Would you suggest keywords (or even sources) to find more information? $\endgroup$
    – nocibambi
    Oct 9, 2018 at 8:41

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.