Author : Charles Xavier, founder M720
In my senior year at university, I worked on a dissertation that focused on measuring the impact of free trade agreements on trade among Southeast Asian countries.
When designing the econometric model, I was confronted with a statistical problem (endogeneity) that had to be corrected by a usual technique (instrumental variables). I was perfectly well-versed in the theory, but I could not find good instrumental variables to correct my problem.
Having noticed my concern as I was pacing in the university library, my research supervisor asked me about the nature of my worries, and after listening to me, he then replied: “in econometrics, there is theory and there is art”.
This sentence, which did not fully resonate then, has taken on its full meaning over the past fifteen years spent developing econometric models applied to marketing (marketing mix modeling or “MMM”). Indeed, like many scientific techniques, the understanding of econometric techniques improves accordingly with the multiplication of the implementation of models.
This process of learning through experience will lead to a better understanding of the various statistical problems usually encountered and consequently help correct them in the most appropriate way.
If one takes the example of the consumer goods sector (FMCG) where certain products sold are often weather-sensitive, they will then see that the seasonality of sales is generally “colinear” with average temperatures, meaning that the seasonality is accounted for by the rise in temperatures, as shown in the graph below :
In econometric theory, we are confronted with a problem of “multicolinearity”, meaning that two phenomena (seasonality and temperature) will explain the sales of the product in a similar way. It becomes necessary to resolve this problem so as to measure separately on the one hand the effect of seasonality and on the other the effects linked to temperature variations.
One of the first solutions is to rework the temperature variable by creating variables that will take into consideration :
- the fact that a week might be hotter or colder within seasonal norms, hence shall we measure the incremental effect in the event of a strong variation in temperature,
- a threshold effect (15 ° C, 20 ° C, etc.) above which incremental sales can be generated.
By substituting the original temperature variable with these specific variables, we shall solve the statistical problem of “multicolinearity” and thus be able to isolate both the effects of seasonality and the effects of weather on the sales of the product under consideration.
This form of experiential learning will also generate a better transcription of the marketing reality into econometric models.
This might surely be one of the most difficult points when conceiving an “MMM” model, integrate deftly business expertise within the variables.
The two most meaningful examples are the evaluations of the “Out Of Home” media and the promotional effects (particularly in PGC).
An OOH campaign, unlike other offline media, straddles two weeks, from wednesday to tuesday in France. Consequently, it becomes necessary to go over the data once again in order to consider that the first week includes a weekend (therefore a moment of consumption) and not the second one.
Regarding the measurement of promotions, one must take into account the phenomenon of “storage / destocking” when creating the variables. Indeed, within the context of strong promotions (for example a “BOGOF” type mechanism), consumers will increase their consumption during the period of promotion (“storage”), which can sometimes lead to a drop in sales during the weeks following the promotions. (“destocking”).
When uttering the aforementioned sentence “in econometrics, there is theory and there is art”, my research advisor actually wished to highlight the fact that theoretical knowledge alone does not solve all the problems associated with the creation of an econometric model, as much of it can only be learnt through experience.
Such an observation is all the more true regarding marketing mix modeling. At a time when the measurement of marketing performance is bound to evolve towards more complexity in this digital world (i.e. the end of third-party cookies), many advertisers are likely to turn towards this acknowledged technique to measure the performance of their actions.
Such an approach, whether as part of a service or an internalization process, requires specific assistance from “MMM coachs” who must display the ability to provide them with technical and business counseling.
Useful links :
Econometrics : https://en.wikipedia.org/wiki/Econometrics
Marketing mix modeling : https://en.wikipedia.org/wiki/Marketing_mix_modeling
Photo credit : free for commercial use
