As stated, I am publishing a model of ENS’s dependence on external factors, aimed at assessing the relative effectiveness of marketing.
It should be noted that I am not a developer, but only transferred the logic of econometric programs to Python. I’ve also never published anything on GitHub, so the repository may be unfamiliar with standardization.
The operating principles of the model are described in the form of comments in the code itself.
The program outputs the following metrics:
A high value of external influence means that success/failure over a certain period of time should not be associated with marketing. It is simply a derivative of market movements.
The high importance of marketing means that at a certain point in time, good decisions were made regarding local and global strategies.
The influence of autoregressive/other factors is a rather useless metric if taken out of context. Its high indicators can be described both as “yesterday’s success” and, for example, by a seasonal component, the analysis of which I provided in my research, but the data on which must be constantly updated and applied, including visual and contextual analysis, which is difficult to programmatically.
Such models are used everywhere in classical business, because it is simply impossible to draw a line between the effectiveness of one’s own efforts and external circumstances. It would be very unwise to attribute any sales growth to a particular marketing strategy and continue to invest money in its implementation without understanding that this growth is caused only by macroeconomic/sociodemographic factors or the market.
But the repertoire of classic companies also includes internal marketing metrics that make the analysis more accurate. In this case, the inaccuracy of the model can be explained by the author’s lack of such metrics.
The model is intended to be used by any entity involved in ENS marketing. But it is also worth noting its main drawback - it does not demarcate the efforts of different entities. If more than one organization is involved in marketing, you simply will not be able to determine which of them influenced the number of registrations.
Some variables in the model (data shift, critical z-score, number of neighbors) were entered manually, based on experiments conducted as part of the study. Their evaluation programmatically can lead to distorted results, since expert assessment is also required here. I admit that they will not be relevant forever and I will try to make changes whenever possible.
Open to comments.