Why Defining “Base” as Just Trend + Seasonality is a Bad Idea
In one of our recent MMM projects, we noticed that earlier vendor had modeled Base = Trend + Seasonality.
Few months ago I wrote why the practice of including distribution in Base in MMM is wrong (link in comments).
📌 Base in MMM
Statistically, Base is the intercept and it is interpreted as the ‘value of Y (dependent variable) when all your independent variables are made zero’.
In marketing, this is interpreted as the sales one would get organically because the marketing / media variables are set to zero.
Now apart from this definition and construct of Base, any other construct leads to a lot of problems.
📌 Seasonality:
A seasonal pattern is a fluctuation which occurs at regular time intervals. These time intervals are predictable. Seasonality does not always refer to changing seasons.
It could be anything which happens at predictable and regular intervals.
📌 Trends:
Trend on the other hand is a slow moving direction (downward or upwards) in the data.
With definitions out of the way, let me point out a few problems of Base = Trends + Seasonality
📌 Conflating statistical patterns as business drivers
When you reduce base to just Trend + Seasonality, you are essentially implying that everything structural about the business is just a time pattern !!
📌 MMM stops being Incrementality Discernment Engine
The biggest incrementality test one could ever do is MMM. This is made possible because of the Base.
Base answers the question – ‘how much sales came organically’ and the rest of the marketing levers answer the question – ‘how much extra incremental sales it resulted in over and above the base’
But once you define base as trend + seasonality, you no longer have a incrementality engine.
📌 Over attribution to Marketing Variables
If base is under-specified, Marketing variables are forced to explain what base should have captured. There is a yin – yang effect in MMM.
If your base captures most of the effects, the marketing levers explains very little and vice versa.
In case on based under specification, you end up attributing more to marketing levers than it is actually the case.
This will lead to over confidence in your marketing variables and hence over deployment of marketing dollars.
📌 Inaccurate MMM
If you specify Base as Seasonality + Trend, your whole MMM itself becomes wrong. As a cascading series of events, your model starts having poorer metrics.
Also if you ever want to do two stage modeling, i.e. what is affecting your base as a result of marketing investments, you won’t be able to do.
Because as per Base = Trends + Seasonality, there is no room for marketing/ media explanations !
In my decade plus of experience, I have seen reasonable seasonality only with CPG and tax filling companies, even then in those case it is wrong to model Base = Trends + Seasonality.
The base in MMM is like your foundation. Get it wrong and you get the whole MMM wrong.
