I have a time series with 5 years of monthly data, but the business I'm working with needs only quarterly forecasts. So in theory I could just aggregate my data up to a quarterly series and then model it.

But it seems like a waste of information.

Is there any benefit to modeling it at the month level anyway?

  • 5
    $\begingroup$ The last time I had this option, I learned so much about the process (an economic time series of a type of large financial transaction) using monthly instead of quarterly data that the client asked me not to use the monthly data because they were too revealing of proprietary information! This suggests the answers to your question depend on what goals you want the modeling to achieve. $\endgroup$ – whuber Oct 10 '18 at 17:49

I suggest that you use a quarterly periodicity because at the end of the day you're going to have to generate a quarterly forecast.

If you where to use a monthly periodicity in your model, to obtain the quarterly forecast you're going to have to make a 3-step ahead forecast. In contrast, when modeling with quarterly data, a single step ahead forecast will suffice. The single step ahead forecast is much more desirable as it will have lower variance (i.e. tighter prediction intervals). This is exacerbated if the series that you're modeling follows a unit root.

To me, this observation will greatly outweigh any potential information gains.

Further, not all additional information is good. Imagine you had hourly data instead of monthly data. There is a lot of variation between hours that is not really going to help you determine what is going to happen at the end of the quarter.


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.