forecasting sales in my company and i'm using ARIMA with
regression. Sometimes, even if the model is well adjusted to the data,
when i try to get forecast for 3 months some of the values are below
0... what can i do to set a limit on that variable?
Most likely the model is not as well adjusted as you think. When
forecasting an item such as sales it is usually better to forecast the
percent sales growth (or equivalently the first different in the log
of sales). R2 will be much lower but the forecast standard error is
likely to more realistic. The Elements of Forecasting book by Frank
Diebold gives guidance on this type of forecasting -- i.e. whether
the forecasting model should be in levels or (percent) changes.