How can I forecast with multiple time series sampled at different frequencies?

Question

I am attempting to build a time series model that can predict order volumes for a single firm in the trucking industry.

I have time series data on the firm's order volumes. This data is monthly data.

I expect also to have access to time series data on say 5 macroeconomic variables that I believe are in some way correlated with the order volumes I am seeking to predict.  For the most part, this will be quarterly data.

My question is: does all of the above data need to be standardized in terms of the time interval, or can I build the models with a combination of monthly and quarterly data?  And if I do need to standardize, how do I do this?

(Happy to do this in either R or Python.)

T. A. Wheeler · Answer

I believe you will need to use the same sampling interval for both sets of data. A simple way to reconcile the two frequencies would be to compute a quarterly sum or average (whichever makes sense in this application) for the monthly order volume data. However, I assume that you want to predict monthly order volumes, so that would render this approach unfeasible. Plus, it means throwing away information, which is rarely desirable, if you can avoid it.

Another possibility might be to estimate monthly values from the quarterly data by doing a time series model for each of the macroeconomic variables, and then use interpolation to get the monthly value estimates. Here is a link that discusses how to do that in Python.
https://machinelearningmastery.com/resample-interpolate-time-series-data-python/

How can I forecast with multiple time series sampled at different frequencies?

One Answer

Add your own answers!

Ask a Question