I'm in the process of starting to collect new time series data (monthly observations) which I would like to perform seasonal adjustment on. While I appreciate that there is no mathematically sound way to do so without a "critical mass" of previous observations, in this case an alternative might be conceivable:

The time series I am collecting is basically a reproduction of an existing time series and expected to behave very similar to the original one. So I was thinking to simply append my new observations to the unadjusted original series and adjust the combined series (using the same adjustment method as for the original time series). The idea is to jump start the adjustment process using the other time series and slowly move away from it when my own time series accumulates more observations.

One of my main concerns is that the new time series may e.g. be offset from the original one by a constant factor resulting in a break of the time series which may be smoothed away by the adjustment procedure or similar.

In general, do you think this approach is feasible? Is there any literature on this that I missed during my search?



An alternative possibility is to fit the original time series, and use the parameters from that fit to create informative priors for your new series. As you add data to the new series, the influence of the prior should weaken.

Your approach sounds reasonable, though. To avoid the offset problem, you could try normalizing the original data and the new data (after 1 cycle) so that they both have an average value of zero (for example). There are a few assumptions in there that might cause problems, but it's a reasonable beginning, in my opinion.

| cite | improve this answer | |

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.