Build Time series models and Panel models for MMM (Marketing Mix Modeling) and beyond.
Design both simple models based on the marketing mix standard, allowing for the broadest use of statistics tools, and complex models of arbitrary architecture with multiple levels that provide the most comprehensive description of reality
FMCG
Pharma
Beauty
Retail
Banking
Insurance
Digital Services
Home Appliance
Gadgets
...and much more
Not just an optimizer
We are convinced that a good econometric framework must possess two key features:
FERMATRICA is designed to meet these requirements to the fullest extent.
From modeling to reporting
Build models that go far beyond the complexity of a simple time series model. Combine additive and multiplicative effects, construct panel models, add auxiliary models, and link model chains together. Use local and global optimization algorithms, and specify transformation parameters manually if necessary.
Conduct statistical analysis of the model during and after construction, perform retro-analysis of the model on historical data, and optimize the split and budget for a specified future period. Special variable transformations tailored to business realities are at your disposal. The results of your work can be exported to PPTX slides in a semi-finished state.
Set the main model parameters in a human-friendly XLSX config file, using the programming environment only for writing the basic flow, outputting results, and customizing modeling and reporting. Display model statistics, graphs, and tables in the IDE console or embed them in a Jupyter Notebook using the extensive library of reporting functions.
Data Scientist friendly
FERMATRICA is a set of open-source libraries. To get started, it is usually sufficient to have programming skills in Python and appropriate mathematical training. The developers make efforts to ensure that the framework primarily depends on well-known and long-term supported libraries.
FERMATRICA itself is in a state of continuous development, focusing on both bug finding and fixing, as well as the addition of new features.
With extensive out-of-the-box capabilities, FERMATRICA also offers broad customization options for the modeler without requiring changes to the framework’s source code. If that is not enough, you can always fork the project and modify the code as needed.
Learm more about FERMATRICA features as econometrics framework and research how it works under the hood
Research two optimization contours in FERMATRICA, the algorithms used, and the transformation functions. Build time series and panel regression models.
Learn more
Create fit-predict and factor decomposition plots, and return on investment curves, as well as optimize the marketing split and budget to meet KPI.
Learn more
Write your own transformation functions, modify reporting pre-requisites and fork the whole project if some global changes are required.
Learn moreThe foundation of the project is the extensive experience gained by the Business Data Science teams at OKKAM in building marketing mix models over more than 10 years of work, dating back to the early 2010s. Initially, the solutions were relatively simple, combining linear time series models with a set of nonlinear transformations. As tasks became more complex and expertise grew, the tools also evolved. Progress was made in two main directions:
By 2021-2022, it became clear that the research teams needed a fundamentally new tool to consolidate the knowledge of various teams into a single universal framework. The new framework needed to embody two essential qualities: to be as powerful as possible and as user-friendly as necessary. This marked the beginning of the econometric framework FERMATRICA.
The code for the first version of FERMATRICA was written from scratch between 2022 and the first half of 2023. The new framework natively supported time series models and panel models, providing algorithmic optimization of parameters through global and local algorithms for more complex types of models. The reporting system enabled both retro-analysis of historical data and construction of budget option forecasts for future periods, allowing comparisons between them to select the optimal option.
By the end of 2023, FERMATRICA v010 had become the primary tool for building marketing mix models at OKKAM.
Work on version v020 began immediately after the final adjustments for the first version were completed. Among the most important innovations are further simplification of the framework’s package interface for the user-econometrician and comprehensive algorithmic optimization of the split based on a specified budget size and target KPI in reporting. As a result, the reporting system reached its completed form.
The source code for this version was released in September 2024. Efforts are underway on the following versions - v021, which will address inevitable bugs, and v030, which will introduce new functionality that we will detail after its release.
