I have finance data that I need to forecast out for 7 years. My data is generally debits and credits, and those are split into a number of sub-series which share common traits (e.g. similar seasonality and/or trend).

My question is, is it appropriate to

  1. split the total series into component parts;
  2. forecast each part in the most appropriate manner;
  3. re-aggregate the parts into a complete projection?

For example, I'd like to treat the following sub-series/parts differently:


  • Wages/Pension (annual percent increase, based upon contracts)
  • Insurance (annual percent increase based upon prior increases)
  • Supplies (annual percent increase based upon different prior increases)

Then, add the individual sub-series back into a single forecast. Obviously, there would be similar but different sub-series/parts for revenue, which would eventually be combined for a total forecast.

  • 2
    $\begingroup$ The short answer is yes. The tricky part is getting prediction intervals on the re-aggregated forecasts as your components are probably correlated. $\endgroup$ – Rob Hyndman Aug 9 '12 at 0:06
  • $\begingroup$ Did you look at Singular Spectrum Analysis? (R: cran.r-project.org/web/packages/Rssa/index.html). It can be used for decomposition and forecasting of time series. $\endgroup$ – LE Rogerson Nov 7 '16 at 13:31

Your Answer

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

Browse other questions tagged or ask your own question.