| Date: | March 15, 2015 |
|---|---|
| Version: | 1.0.0 |
| Authors: | Quentin ANDRE
quentin.andre@insead.edu |
| Web site: | https://github.com/QuentinAndre/Guadagni-Little-1983-Python |
| Copyright: | This document has been placed in the public domain. |
| License: | Guadagni-Little-1983-Python is released under the MIT license. |
Guadagni-Little-1983-Python provides a Python implementation of the mixed/conditional logit model of product choice outlined by Guadagni and Little in their 1983 article "A Logit Model of Brand Choice Calibrated on Scanner Data".
- Dataset Recreation.py: A script to create from scratch a purchase history dataset of N consumers over T time periods for K options. All the parameters in the script can be changed to generate a different dataset. As in the original paper, the consumers are heterogenous in their loyalty to the brand and to the different sizes offered. The script outputs four files:
- GuadagniLittle1983.csv, which contains the simulated scanner data
- TrueBetas.csv, which contains the true parameters used to generate the data.
- BrandShares.png, which plots the evolution of market shares for the brands over time.
- SizeShares.png, which plots the evolution of market shares for the sizes over time.
- Mixed Logit Estimation.py: A script to recover the parameters used to generate the data. As in the original paper, the the utility of the first option is constrained to be 1 to allow identification of the (K-1) brand intercepts and of the J utility components for the attributes (which are common to all brands).
- Recreation and Estimation.ipynb: An IPython notebook combining the two aforementioned files, and describing in details the different steps of the data creation and data estimation process
- Recreation and Estimation.html: An HTML export of the aforementioned notebook.
- This code has been tested in Python 3.4, using the Anaconda distribution:
- Using git:
- Download the master branch as a zip:
Guadagni, P. M., & Little, J. D. (1983). A logit model of brand choice calibrated on scanner data. Marketing science, 2(3), 203-238.