# Marketing Mix Modeling revisit

Objective: Obtain incremental sales during promotional period

Suggested Method:

1. Replace sales values during promotion periods with NA.
2. Use stl() to decompose sales into trend, season, remainder
3. Forecast (trend - season) through NAs
4. Add seasonal component back onto the forecasted trend. This is the base.
5. The difference between the base and the actual sales is the incremental dollars due to the promotion.

Problem: Most of the actual sales, at least in my data, are lower than the base sales.

Question:I am looking for some kind of alternative method or error in my current method as I find it hard to believe that promotions have negative incremental dollars.

This is a revisit the conversation I was reading here.

• Have you looked at plots of the data? Maybe they really are lower, it's hard to see how stl() could forecast a spike in demand higher than the actual spike if all the spikes have been removed from the history. – jbowman May 2 '18 at 17:59