How does your regression model account for the change in accounting rules whereby the cost of a promotional campaign is accounted for by a corresponding reduction in sales?
My guess is that your regression is blind to any changes in laws, regulations, rules, etc that impact how revenues are determined. Which proves my point as to how useless regressions are: they always assume the future is the same as the past. And it rarely is. So even your "near-term" use of regression may be in serious doubt. But we'll let the numbers decide. Heck, you may even get lucky.