Case Study: Modeling Vehicle Choice and Simulating Market Share

Abstract

We present a new method and workflow for estimating market shares of future products based on pre-introduction data, such as syndicated studies conducted before product launch. Utilizing Bayesian networks, alongside BayesiaLab and Bayesia Market Simulator, this approach enables reliable market share simulations that are both practical and economical. This method is demonstrated with a case study on the U.S. introduction of the Porsche Panamera for Model Year 2010, highlighting its applicability even in niche markets with limited observations.

Objective

This tutorial is designed for marketing practitioners exploring the application of Bayesian networks. By illustrating the capabilities of BayesiaLab through a real-world case study, this guide provides insights for analysts in various fields. The detailed steps cover data preparation, network learning, and market share simulation, applicable to a broad range of research tasks beyond marketing.

Introduction

Market share is a critical performance metric, essential for defining sales volume expectations in product planning. Accurate predictions of future market shares require understanding consumer behavior despite the inherent complexity and uncertainty in their decision-making processes. Using choice modeling, particularly Bayesian networks, allows for incorporating both observable data and tacit knowledge to simulate consumer choices.

Bayesian Networks for Choice Modeling

Bayesian networks offer a localized and systematic method for structuring probabilistic information, making them ideal for representing consumer choice dynamics. Unlike traditional utility-based choice models:

1. Non-Restrictive Dimensions: Bayesian networks preserve trade-offs between attributes, e.g., fuel economy vs. price, without collapsing into single scalar values.
2. Nonparametric Flexibility: They do not require predefined functional forms, allowing for the discovery of nonlinear relationships.
3. Inherent Probabilistic Framework: Bayesian networks handle uncertainty natively without the need for additional error terms.

Case Study: Porsche Panamera

Background

The Porsche Panamera, introduced as a four-door luxury sports sedan, entered a competitive segment featuring established contenders like the Mercedes-Benz S-Class and BMW 7-Series. Beyond these competitors, it also faced potential cannibalization from Porsche's own Cayenne SUV.

Study Objective

The case study aims to predict the Panamera's market share using Revealed Preference (RP) data from the 2009 New Vehicle Experience Survey (NVES) without conducting new research.

Methodology

Data Preparation

1. Variable Selection: Selected ~50 variables, focusing on vehicle attributes, consumer demographics, and attitudes.
2. Discretization: Continuous variables, such as price, were discretized into meaningful intervals using algorithms like K-Means.

Network Learning

1. Forbidden Arcs: Restricted learning to relationships between product and market variables to avoid encoding existing product configurations.
2. Unsupervised Learning: Utilized EQ learning to discover associations between variables.

Simulation

1. Baseline Scenario: Defined existing product scenarios using NVES data.
2. Bayesia Market Simulator: Simulated market shares by importing Bayesian networks and testing new product scenarios, such as variations of the Panamera.

Results

The simulation accurately predicted the Panamera's market share, aligning closely with its observed performance post-launch. Additional insights included:

• Identification of substitution and cannibalization effects.
• Scenario testing for product and market changes.

Limitations

While effective for simulating variations of existing configurations, this approach is less suited for entirely new technologies, such as electric vehicles. Expert judgment is essential for defining adequate product and market scenarios.

Outlook

Future enhancements include:

• Expert Knowledge Integration: Leveraging elicited expertise to augment Bayesian networks.
• Satisfaction-Driven Models: Incorporating consumer ratings to evaluate the impact of product improvements.
• Broader Applications: Extending simulations to brand-level or segment-level market dynamics.

Summary

BayesiaLab and Bayesia Market Simulator offer a robust, scientifically grounded framework for choice modeling and market share simulation. By transforming existing consumer data into actionable insights, these tools empower practitioners to make data-driven decisions with precision and efficiency.