# Testing prediction time series against real data

Say I have a series of forecasts and observations like this:

      EntityF   EntityO
2004  120       125
2006  166       173
2008  150       167
2010  152       -


And assume that the (i) entity is the same and (ii) the forecasting methodology is constant.

I'd like to

1. Produce a meaningful metric of the forecasting error.
2. Be able to predict the current forecast (2010) error based on 1.
-

1. You could use Mean Absolute Error (mean of $|F-O|$) or Mean Squared Error (mean of $(F-O)^2$)