Forecasting high frequency variable with low frequency predictor

Question

Newbie here to forecasting and I have a very basic question. I have 2 distinct time-series data.

• Time Series A is Weekly (High frequency)
• Time Series B is Monthly (Low frequency)

I need to forecast A using B.

I looked at Armesto & colleagues' paper on forecasting with mixed frequencies and am also looking at the midasr user guide. Both talk about forecasting a low frequency variable with a high frequency one.

But can the same approach be used for forecasting weekly data using monthly time-series data? Unfortunately, the data that is available to me is in that format.

From what I understood, I'm pretty sure it can be done, but was hoping someone can confirm. Also, is there anything major I need to watch out for while running this forecast?

Answer 1

There are two quick and dirty solutions. First would be to disaggregate series B to weekly values (R package tempdisagg is great for that) and then do a usual model. Second aggregate series A to monthly frequency, do a forecast and then use disaggregation on the forecast.

The more theoretical approach would be casting problem to a state space model. There are a lot of literature on state space model approach when the dependent variable is observed at lower frequency. It usually assumes that the low frequency variable is really a high frequency variable observed at low frequency periods. You can make the same assumption and then reverse the methodology. Unfortunately I have not seen something similar being done, but I did not look hard enough.

Concerning midasr, I can say that it was designed to work when the dependent variable is observed at the lowest frequency. The reverse situation was not seriously considered.