Reverse-engineer interval of rolling average without underlying data?

27 Views Asked by At

Is it possible to analyse a set of numbers and determine the likelihood that they are a rolling average, and if so, what it's duration might be?

My source data is from a website which provides estimated audience sizes for advertisers. These figures change daily - my theory is they are based on a rolling average of X previous days and it would be very useful to figure out how long a period that is.

I have been collecting data for several months, and have daily figures (sequences of 150+ days) for hundreds of topics (plenty of which follow a similar day-to-day trend when you plot them alongside each other).

But I have none of the original data points (whatsoever), and no information about the methodology the site is using.

If there are any techniques I can use to spot patterns, could someone point me in the right direction?

(One thing I notice is where the predicted audience size is low, I see a lot of figures like 997, 998, 999 or 664, 665… Would analysing the frequency of these help at all?)