1

I have some time series data that I am trying to model - basically historical elasticity reads with daily sales data. I really am just interested in finding the 'mean' value - not necessarily predict future elasticity - at least not as of now. Unfortunately, I only have 2 year of so data and daily reads are too noisy.

What are the issues with using overlapping data, for example, rolling 30day average where I calculate a value every day? Obviously this would understate variance, but I don't think variance really matters for what I'm doing. And how unrealistic would it be to use this lower variance to build a confidence interval for prediction? I guess technically, any 30 day elasticity should have such a low variance I would think...

Would this also cause issues when reading ACF and PACF plots to determine the model to use?

I looked at this but none of the links work: Time series regression with overlapping data

Richard Hardy
  • 67,272
confused
  • 3,253
  • Unfortunately, I only have 2 year of so data and daily reads are too noisy: this is pretty fundamental. Going for overlapping data will not remove or reduce the problem. If you get seemingly more precise results from the latter than the former, you might have just tricked yourself. – Richard Hardy Apr 05 '22 at 05:41
  • It was very easy to find the papers behind the broken links by using Google Scholar. I have updated the links now. – Richard Hardy Apr 05 '22 at 05:44

0 Answers0