prediction using unknown independent variable So suppose I'm predicting "A" using a linear regression model : A = a + bx1 + cx2
I'm positive that X1 and X2 have a linear relationship with A
The problem is: I have historical data on my independent variable x1 and x2 but do not know what their exact value will be a month to three months from now (my predicting time frame). Is it OK to predict x1 and x2 and use that as a proxy for what would be at time t? Does it make sense to use them as predictors? what would be a good way to incorporate those predictors in my forecast in case I don't know the actual forward looking values.
What does the literature say about such situation. 
 A: Sure, just forecast your $x_1$ and $x_2$ and feed them into your model to forecast $A$. That's how, for instance, we use weather forecasts to predict tomorrow's sales of barbecue meat - weather is an important predictor, but we only have weather forecasts, not tomorrow's actual weather.
Of course, this will add another source of uncertainty in your forecasts of $A$. Keep that in mind. If you are interested in predictive densities or prediction intervals for $A$, "standard" formulas won't help you, so I'd recommend that you run simulations.
You could also use a holdout sample and compare what happens if


*

*you forecast $A$ using the actual future values of $x_1$ and $x_2$

*you forecast $A$ using the forecasted future values of $x_1$ and $x_2$


The degradation in forecast accuracy between the two scenarios tells you how much of a problem it is that you need to forecast $x_1$ and $x_2$ - and whether you need to work on forecasting them better. (Or controlling them.)
Finally, benchmark your forecasts also against a simple third option:


*model and forecast $A$ without using $x_1$ and $x_2$ at all

