Forecasting using ARIMA model and exogeneous variables

Question

I have sales data for 7 years on a daily basis with breaks for the weekends.

I have a few questions

1. Should I aggregate it to a weekly or monthly level?
2. I would like to forecast for 'h' steps ahead. How should I make the test and training set?
3. To forecast do I need the future estimates of the exogenous variables as well?
4. What if I do not have the future estimates?

I will be using R to do the analysis.

Answer 1

Responding to your questons....

1. Does the data vary widely on a daily level? Do holidays impact the data? If so, use daily data. And the answer is usually "yes" to those two questions.
2. Maybe you should and maybe you shouldn't do that. There are two schools of thought here. Those in the cross validated camp who ignore outliers and those that use all of the data and rely upon the model to identify changes in the parameters(Chow test), variance (Tsay test), seasonality(ie seasonal pulse) and trend.
3. Yes.
4. Build a model and forecast for each of your causals.