(Out-of-Sample) Forecasting of Unbalanced Panel Data

[Ferdinand Mol]

Dear all,

I’m estimating various models (FE, RE, xtabond, etc.) with an unbalanced panel data set, which ranges from July 2008 to July 2019 with weekly observations. My goal is to ultimately forecast the dependent variable 1 year (52 periods) after the last observation of each individual. As the panel data is unbalanced and therefore the forecasting dates differ for the individuals, I’m unsure how exactly I could achieve my goal.

As far as I've discovered up till now, these are my options:

1. Use ‘predict’, which only predicts in-sample.
2. Transform data set to balanced panel, and apply ‘forecast’ for multiple periods. (As the ‘forecast’ option requires strongly balanced panels)

Both these option are not ideal for me; therefore I’m wondering if there’s a method to forecast for unbalanced panel data models, in such a way that each individual has a X number of out-of-sample forecasts? Any help or links regarding this subject would be much appreciated!