# Building more granular data into forecasting

I'm new to time-series forecasting, but have a question about more advanced techinques.

I have an aggregate global measure - the average of about 20 countries trends in each year. However, since I only have 8 data points for this global measure (8 years worth), trying to forecast next years figure produces huge confidence bands.

I was wondering whether there is any technique that could make better use of my unaggregated country data in forecasting the global measure? By feeding each individual countries' trend into a model, maybe I can produce a stronger forecast?

I've read through Rob Hyndman's textbook, but couldn't find anything of this type.

I'd really appreciate it if somebody could point me in the right direct, or at least tell me if I'm wasting my time.

Thanks, Ben

• Do you know how the global measure is computed from the individual data sets? – Arun Jose Sep 29 '16 at 11:27
• The global measure is an average across the 20 countries in each year – Ben Davis Sep 29 '16 at 11:30
• My best guess at this point would be in that case to predict individual country data into the next year. Assuming you can do so, average the resulting forecasts. You can compare this to the global forecast to get an idea if you are on the right track. – Arun Jose Sep 29 '16 at 11:32
• Thanks Arun. That is something I'm considering. One issue might be that you lose the error associated with each forecast. I guess you could take the error of the average over the 20 countries' forecasts. However, this is measuring something different entirely (i.e. variance in absolute values of the countries' forecasts). Cheers, Ben – Ben Davis Sep 29 '16 at 11:39
• You said you read "Hyndman's textbook", did you mean FPP? Because he has a chapter about basically this (9.4, "Forecasting hierarchical or grouped time series"). Also, you can absolutely compute out-of-sample forecast error measures no matter how you came up with the forecast, and compare the actual performance of your forecast of the aggregate vs aggregate of the forecasts to see what works best. – Chris Haug Sep 29 '16 at 12:19