yuggo / sow

SOW

This is a first implementation of several share-of-wallet models (one three-parameter and one two-parameter model) based on

Yuxin Chen, Joel H. Steckel (2012) Modeling Credit Card Share of Wallet: Solving the Incomplete Information Problem. Journal of Marketing Research: October 2012, Vol. 49, No. 5, pp. 655-669.

The models allow estimating category specific share-of-wallets and modelling the respective true purchasing behaviour of customers within the category with observations from this respective category/ retailer only. Further, the models allow for incorporating observed heterogeneity through a hierarchical setup.

The first step of the model, in which the parameters are modelled with a Metropolis-Hastings simulation, is implemented in Python. The hierarchical regression of model parameters based on customer or category covariates is based on JAGS in implemented in R using rjags.

---

Usage

chen_utils.py includes all necessary functions for modelling the parameters of the initial model (Likelihood, Metropolis algorithm and others)

user_simulation_utils.py includes several helper functions for modelling the true and observed purchases of fictive users

chen_model_intro.ipynb provides a walkthrough through the model.

metropolis_simulation.ipynb simulates several users based on predefined covariates and simulates the parameters of interest using the Metropolis-Hastings algorithm.

This trajectory of simulations is then used in a hierarchical model in jags_chen_simulation.R to model the relationship between our covariates of interest and the simulated true parameters during the Metropolis-Hastings modelling.

---

Requirements

See requirements.txtfor Python 3. Further, you will need R and JAGS.