One sided filter, online filtering, caual filtering

Question

This is actually a repost of a question of mine from some weeks ago. I got good hints but I couldn't inf the perfect solution yet. I am looking for a Filter which just uses historic data for "smoothing". I tried several filters from the robfilter-package and the ets (exponential smoothing) filter from the forecast package, but I am not happy with the results and the computation time takes extremely long for the former mentioned package. I would actually need something with the power of loess or hodrick-prescott which just uses past data. I considered the dfa function from the signalextraction package but my time series is with by now 55 weeks too short. I would be glad if somebody of you could give me a hint! It is 4.40 am here and I got more and more frustrated because the long computation times in combination with bad results are not very motivating in the long run ;-).

Answer 1

ok - I ended up using the Kalman Filter. With the right configuration this works just fine. Thanks for your input!