Simpler is better: Predicting consumer vehicle purchases in the short run

Area of Science:

• Environmental policy and economics
• Automotive market analysis
• Predictive modeling for regulatory impact assessment

Background:

• Forecasting vehicle purchases is crucial for assessing the impact of environmental regulations, such as greenhouse gas emissions standards set by the US Environmental Protection Agency (EPA).
• Accurate prediction models are needed to understand how changes in fuel economy and vehicle prices influence consumer choices and market dynamics.

Purpose of the Study:

• To compare the predictive accuracy of a simple market share model against a complex nested logit model for forecasting vehicle purchases.
• To evaluate the impact of incorporating vehicle price and fuel economy data into predictive models for regulatory analysis.

Main Methods:

• Utilized historical vehicle market share data from model years 2008, 2010, and 2016.
• Developed and compared a simple model using past market share to predict future market share against a nested logit model.
• Assessed model performance using various goodness-of-prediction measures.

Main Results:

• The simple market share model consistently outperformed the nested logit model across all prediction measures and years.
• Inclusion of vehicle price and fuel economy data led to increased bias in the forecasted market shares.
• The simple model's accuracy held even during the economic disruption of the Great Recession in 2010.

Conclusions:

• A parsimonious approach using past market share offers superior predictive power for vehicle purchases compared to complex logit models in regulatory impact assessments.
• Factors beyond direct cost pass-through, such as unobserved quality improvements or preference shifts, likely explain the bias introduced by price and fuel economy data.
• Observed market share fluctuations during periods of economic stress can provide valuable benchmarks for policy change evaluations.