Consumer Choice Modeling David S. Bunch Graduate School of Management, UC Davis with Sonia Yeh, Chris Yang, Kalai Ramea (ITS Davis) 1
Motivation for Focusing on Consumer Choice Modeling Ongoing general motivation: Developing improved models for analyzing policies related to climate change. Focus of this presentation: Energy Systems Models These are so-called Energy-Economy-Engineering-Environment (E4) models (e.g., CA-TIMES). Frequently focus on technical feasibility over a long time horizon. Include high level of detail on technology performance and costs, but, They fall short in producing realistic consumer response to alternative future market scenarios. 2
Specific Case: CA-TIMES 3
Examples of Technology Choice Options 4
Characteristics of CA-TIMES As shown: High level of detail on technologies and costs. Choices for entire energy system (over a long planning horizon) are based on minimizing NPV total system monetary costs while meeting projections of future demands for energy services. TIMES models are used as black boxes by most researchers. Solution obtained using Linear Programming with perfect foresight. Again: Consumer choice = minimizing a combination of fixed and variable monetary costs. Our focus: Bringing realism to the personal vehicle market.
Point of Contrast: MA 3 T Model (Greene and co-workers) MA 3 T (Market Allocation of Advanced Automotive Technologies), nested multinomial logit model developed by Oak Ridge National Laboratory 1458 consumer groups (Regions, driving behavior, risk attitude, charging infrastructure) Nationwide Model (9 regions in the US) 6
Point of Contrast: MA 3 T Model (cont.-) Focused exclusively on personal vehicle market => Narrower scope than CA-TIMES But: Includes many additional behavioral factors that affect consumer choice (not just costs) Has established connections to existing discrete choice modeling literature => Nonlinear models yielding choice probabilities for competing vehicle technologies. 7
What we have been doing We have developed methodology to import behavioral content from, e.g., MA 3 T, into E4 models like CA-TIMES. [Some of you may seen various versions of this work ] But: We recently completed a major research report (July 2015) that lays out results based on derivations from first principles (economic theory) that establishes the approach. Connects the dots between E4 and discrete choice models. [Paul sent out a notification to STEPS sponsors. Also, ITS Seminar (heavy on the theory): Video available online.] Finally: Previously presented work was based on prototype models (vehicle choice only). Today s results: Based on actual modifications of CA-TIMES.
Overview of what comes next First, review one feature/issue: The need to switch (for now) from previous CA-TIMES vehicle technology assumptions to MA 3 T vehicle technology assumptions. [We will call these old and new technologies.] Then: Review some results that illustrates deficiencies in E4 models we seek to address. Next: Illustrate how the new methodology yields the desired effects. Finally: Review what we are working on at the moment.
Old Technologies Conventional Hydrogen EVs Internal Combustion Gasoline Diesel E85 Hydrogen ICE 100-mile Hybrid Gasoline Diesel E85 Fuel Cell Vehicle 200-mile Gasoline Plugins 10-mile 30-mile 40-mile 60-mile Natural Gas / LPG Natural Gas Vehicle E85 Plugins 10-mile 30-mile 40-mile 60-mile Natural Gas Bi-fuel* vehicle LPG Vehicle Diesel Plugins 10-mile 30-mile 40-mile 60-mile LPG Bi-fuel* Vehicle *Bi-fuel= mix of gasoline and NG/LPG 10
New Technologies Conventional Hydrogen EVs Internal Combustion Gasoline Diesel Hydrogen ICE 100-mile Hybrid Gasoline Diesel Fuel Cell Vehicle 150-mile Gasoline Plugins 10-mile 20-mile 40-mile Fuel Cell Plugins 10-mile 20-mile 40-mile 250-mile 11
CA-TIMES (BAU Case, Old Technologies) New vehicle sales percentages for Passenger Cars. CAFE standards and ZEV mandate have been removed. To illustrate the knife-edge consumer behavior issue, we have also eliminated user-specified growth constraints, specialized discount rates, and market share restrictions. 12
CA-TIMES (BAU Case, Old Technologies) New vehicle sales percentages for Light-Duty Trucks. Note: Less knife-edginess in later years because biomass-based ethanol is a scarce resource that is shared across multiple sectors. For us, this is an important feature for future work 13
CA-TIMES (BAU Case, Old Technologies) Combined Cars + LDT New Vehicle Sales Shares. Note the effect of relative sizes of sub-markets on results. 14
CA-TIMES (BAU Case, Old Technologies) Effect of CAFE/ZEV/user constraints. Unmodified = with CAFE/ZEV/user constraints. 15
Comparison of CA-TIMES Results: Old versus New Technologies BAU with no CAFE/ZEV or user constraints. 16
CA-TIMES: Comparing BAU with GHG Scenario Results obtained with no CAFE/ZEV or user constraints. 17
Moving on: Old versus New Technologies GHG with no CAFE/ZEV/user constraints Again: The Consumer Choice Model is based SOLELY on NPV of capital investment in technology and lifetime fuel costs. 18
CA-TIMES: Effect of Interventions on GHG Results Effect of CAFE/ZEV/user constraints Unmodified = With CAFE/ZEV/user constraints 19
Bringing in Factors from MA 3 T MA 3 T (Market Allocation of Advanced Automotive Technologies), nested multinomial logit model developed by Oak Ridge National Laboratory 1458 consumer groups (Regions, driving behavior, risk attitude, charging infrastructure) Nationwide Model (9 regions in the US) July 6-7, 2015 wholesem 20
Addition of Consumer Segmentation Settlement Type Risk Attitude Urban Suburban Rural Early Adopter Early Majority Late Majority Driving Behavior Recharging Infrastructure Low Annual VMT (8656 miles) Medium Annual VMT (16068 miles) High Annual VMT (28288 miles) Home + Work Home + No Work No Home + Work No Home + No Work (+ public recharging infrastructure common to all) July 6-7, 2015 wholesem 21
National Reference Case Infrastructure Growth Curves 22
Addition of Other Factors Affecting Consumer Decisions Starting point: Vehicle Investment Cost & Fuel Cost In addition: Infrastructure-Related Costs Refueling Inconvenience [Liquid/gaseous stations] [Diesel, natural gas, hydrogen] Costs associated with lost time due lower density of refueling stations. Range Anxiety Cost [Limited range of EVs] Issue: Random daily VMT distributions => Some events with insufficient range. Requires, e.g., replacement option at some cost. 23
Addition of Other Factors Affecting Consumer Decisions cont- Dynamic Effects Associated with New Technologies New Technology Risk Premium Cost regarding the perceived riskiness of new vehicle technologies Can be negative for early adopters. Make/Model Availability Costs As new vehicle technologies penetrate the market, manufacturers offer more makes and models, adding diversity across a range of unaddressed attributes. These change dynamically based on sales patterns. 24
Addition of Other Factors Affecting Consumer Decisions cont- Finally: There is random variation in consumer preferences from a range of other unobserved effects. 25
A Program Note... At this stage: I could show you what happens as we add in these effects one at a time. But: No time for that Quick remarks: What we have now is CA-TIMES-COCHIN Mixes together cars and LDTs into the same personal vehicle market Important caveat: Dynamic effects are driven by National Market (not California!) 26
Recall the BAU Case 27
Recall the BAU Case 28
What about the GHG scenario? 29
GHG Scenario: Ad Hoc User Constraints versus Consumer Choice 30
Summary and Work in Progress We have now successfully integrated behavioral factors from the MA 3 T model into CA-TIMES Demand heterogeneity, Disutility or generalized costs Random error distribution added as costs to introduce nestedlogit structure We are now working on: Endogenous station availability determination Better representation of spatiality Consideration of National versus CA market effects Policy analysis such as carbon cap, infrastructure investment, vehicle subsidies Revisiting, updating, investigating input data on technology options Looking to more investigation of behavior factors themselves