Dynamic Bayes Modeling

Home » Marketscience Studio » Dynamic Bayes Modeling

The core modeling module provides a comprehensive user interface (UI) for specifying the model setup which then integrates with the model database to perform any specified transformation of variables and passes the required data-set to the proprietary model estimation algorithm that is housed with-in the Ox metrics analytics package. Results from Ox are then passed back for viewing in the UI.

Underlying the power of the Marketscience Studio is the unique estimation algorithm developed using the State Space Functions Pack (SSF Pack) written and supported by Professor Siem Jan Koopman. TSC and MSC have partnered exclusively with Professor Koopman to develop the SSF pack to accommodate the unique needs of modeling Consumer Demand as an evolution of the traditional Marketing Mix Modeling approach.

The SSF pack was the original and is still the leading algorithm for estimating non-constant parameters in a timeseries model. Our development now adds the Bayesian Pooling capability to enable more rigorous general to specific modeling within large panel data sets. Specifically for marketing this helps us identify common geographical and/or product level marketing response effects.

  • 1

    Starting with a fully random panel estimation

  • 2

    Then pooling and finding common segments of response

  • 3

    Resulting in the most parsimonious (simplified) model possible whilst still maximizing insights

Model results viewing (in ox)….

In addition to the estimation approach our UI incorporates all of the major data transformations necessary to cover all potential marketing response functions

  • PDL structures

  • Adstocks

  • Non-linear adstocks

Data Visualization & Analysis

Dynamic Bayes Modeling

Full System Decomposition

Simulation & Optimization

Forecasting & Reporting