0

Suppose I have some time series. I will call it "baseline" time series. Then, I have indicator variables saying that on some dates there was some event, like sunshine. Incidentally, time series is much higher during this period of time compared to baseline.

During this time series elevation, not only the bias changes, but also variance (autocovariance-0) becomes much larger. How can I remove such a period and convert it into the baseline?

Right now my only thought is just to use adjacent baseline values, find their mean, and draw a line instead of time series during the elevation.

SWIM S.
  • 1,006
  • 9
  • 17
  • you don't want to convert it ... you want to normalize/adjust for this exogenous effect via an ARMAX model. – IrishStat Jun 21 '18 at 00:06
  • @IrishStat, but what if I want to use the forecast for the baseline, and some simpler model, like say sarima or holt-winters? Im talking about "in principle". You think it is wrong in principle to do that? This stems from the fact that I dont think there will be any more elevations into the future. – SWIM S. Jun 21 '18 at 00:13
  • What you are suggesting is fundamentally ok .. but if there are trends or pulses to deal with or arima structure to deal with issues might arise . I would have to see the data. Why don't you post it and show precisely what you are attempting to do – IrishStat Jun 21 '18 at 01:36
  • 1
    Please review https://stats.stackexchange.com/questions/348248/how-to-adjust-for-a-temporary-12-month-level-shift-in-time-series/348336#348336 as it deals with a similar question regarding "level shifts" – IrishStat Jun 21 '18 at 08:53

0 Answers0