endogenous preferences joan walker, uc berkeley with akshay vij and feras el zarwi december 2015...

Endogenous Preferences

Joan Walker, UC Berkeley

with Akshay Vij and Feras El Zarwi

December 2015 Daniel McFadden Honoris Causa Workshop @ Université de Cergy-Pontoise

2

Outline

• Introduction– Endogenous Preferences ?

• Demonstration of modeling endogenous preferences– Modality Styles in Travel Demand Models

• Conclusion

3

Endogenous Preferences ?!?

• Neoclassical assumptionPreferences (as indicated by taste parameters & choice sets) are characteristics of the decision-maker that are exogenous to the choice situation and stable over time.

• Alternatively, “The issue has never been whether or not tastes change. Rather the question is whether or not economists should concern themselves with such changes.” – Albert and Hahnel (1990)

“… economic theory proceeds largely to take wants as fixed... The economist has little to say about the formation of wants; this is the province of the psychologist. … The legitimacy of any justification … must rest … on the light that is shed ….” – Friedman (1962)

4

“Where do preferences come from? Do they come from the sky?” – Albert and Hahnel (1990)

• “… we have to acknowledge and make use of the fact that preferences are partly the product of peoples’ environment.” – von Weizsacker (1971)

• “Economists have traditionally been suspicious of changing tastes, and a profession’s intellectual tastes change slowly.” – Pollak (1978)

• “… different methods of eliciting preference often give rise to systematically different ordering” – Tversky and Thaler (1990)

• “If advertising increases the utility consumers received from goods that are advertised, [how] should the effect on consumer welfare be measured…?” – Becker (1996)

• “If legal rules have inevitable effects on preferences, it is hard to see how a government might even attempt to take preferences ‘as given’” – Sunstein (1993)

• “The idea that values can change leads to thinking about how public policy might alter values and thereby change responses to public policies.” – Aaron (1994)

• “If preferences are affected by the policies or institutional arrangements we study, we can neither accurately predict nor coherently evaluate the likely consequences of new policies or institutions without taking account of preference endogeneity.” – Bowles (1998)

• “In short, preferences are endogenous to the environment, but formal frameworks for examining this interaction are scarce in modern economics.” Palacios-Huerta & Santos (2004)

5

MODALITY STYLES IN TRAVEL DEMAND MODELING

DEMONSTRATION OF MODELING ENDOGNEOUS PREFERENCES

6

7

Models of Travel Behavior

Traditional Models• Trip-based decision– Consider all transportation

alternatives– Evaluate time and cost (and other) – Make rational decision

• Limited heterogeneity

Modality style Model• Higher-level decision– Lifestyles built around particular

travel modes

Vij, Carrel, Walker (2013)

Latent Modality Style Formulation

Travel-Related Behaviors (B)

Individual Characteristics (S)

Transportation and Land Use Attributes (Z)

Modality Style (m)

Latent Modality Style Formulation

Travel-Related Behaviors (B)

Individual Characteristics (S)

Transportation and Land Use Attributes (Z)

Modality Style (m)

CLASS-MEMBERSHIP PROBABILITY

that the individual has modality style m conditional on characteristics of the individual S

CLASS-SPECIFIC CHOICE PROBABILITY

that individual chooses behaviors B conditional on alternative attributes Z and modality style of the individual m

MARGINAL CHOICE PROBABILITY

unconditional on modality style m

Latent Modality Style Formulation

Travel-Related Behaviors (B)

Individual Characteristics (S)

Transportation and Land Use Attributes (Z)

Modality Style (m)

CLASS-MEMBERSHIP PROBABILITY

that the individual has modality style m conditional on characteristics of the individual S

CLASS-SPECIFIC CHOICE PROBABILITY

that individual chooses behaviors B conditional on alternative attributes Z and modality style of the individual m

MARGINAL CHOICE PROBABILITY

unconditional on modality style m

Latent Modality Style Formulation

Travel-Related Behaviors (B)

Individual Characteristics (S)

Transportation and Land Use Attributes (Z)

Modality Style (m)

CLASS-MEMBERSHIP PROBABILITY

that the individual has modality style m conditional on characteristics of the individual S and expected benefit of each modality style E

CLASS-SPECIFIC CHOICE PROBABILITY

that individual chooses behaviors B conditional on alternative attributes Z and modality style of the individual m

MARGINAL CHOICE PROBABILITY

unconditional on modality style m

Latent Modality Style Formulation

Travel-Related Behaviors (B)

Individual Characteristics (S)

Transportation and Land Use Attributes (Z)

Expected benefit of each modal style (E)

Vij and Walker (2014)

13

Latent Modality Style Formulation

• Latent Modality Style Segments; each segment (m=1, … , M) has its own people and behavior– Set of transportation alternatives considered– Willingness to pay and attitudes– Demographic distributions

• Data mining of travel diary data determines– Number of segments M– Behavior of each segment for m=1, … , M– Demographics of each segment

14

1. Inveterate Drivers 2. Car Commuters 3. Moms in Cars

4. Transit Takers 5. Multimodals 6. Empty Nesters

Example of Estimated Modality Style Segments (San Francisco Bay Area 2000 Travel and Activity Diary)

Vij (2013)

15

Perc

enta

ge o

f the

Pop

ulati

on

Vij, Gorripaty, Walker (2015)

Evidence of Changing Modality Styles (San Francisco Bay Area 2000 to 2012)

16

Perc

enta

ge o

f the

Pop

ulati

on

Vij, Gorripaty, Walker (2015)

Evidence of Changing Modality Styles (San Francisco Bay Area 2000 to 2012)

17

Impact of Changing Modality Styles on Forecasts(San Francisco Bay Area to 2024)

Scenarios to project beyond 2012Modality trends revert back to 2000 levels by 2024Modality style distribution remains constant at 2012 levelsModality trends observed from 2000 to 2012 continue, but at half the rate

Vij, Gorripaty, Walker (2015)

18

-2 months +3 months +10 months +20 months0

5

10

15

20

25

30

35

40

45

Predicting Modality Trend via Integration with HMM(Santiago, Chile 2006-2008 Journey to Work Panel)

Introduction of Transantiago (shock)

time = 0 months

Modality Styles

Transit 1: Bus

Transit 2: Bus/Metro

Transit 3: Metro/Car

The DriversPerc

enta

ge o

f the

Pop

ulati

on

with

in e

ach

Mod

ality

Sty

le

El Zarwi, Vij, Walker (2015)

19

-2 months +3 months +10 months +20 months0

5

10

15

20

25

30

35

40

45

Predicting Modality Trend via Integration with HMM(Santiago, Chile 2006-2008 Journey to Work Panel)

Introduction of Transantiago (shock)

time = 0 months

Modality Styles

Transit 1: Bus

Transit 2: Bus/Metro

Transit 3: Metro/Car

The DriversPerc

enta

ge o

f the

Pop

ulati

on

with

in e

ach

Mod

ality

Sty

le

El Zarwi, Vij, Walker (2015)

20

-2 months +3 months +10 months +20 months0

5

10

15

20

25

30

35

40

45

Predicting Modality Trend via Integration with HMM(Santiago, Chile 2006-2008 Journey to Work Panel)

Introduction of Transantiago (shock)

time = 0 months

Modality Styles

Transit 1: Bus

Transit 2: Bus/Metro

Transit 3: Metro/Car

The DriversPerc

enta

ge o

f the

Pop

ulati

on

with

in e

ach

Mod

ality

Sty

le

El Zarwi, Vij, Walker (2015)

Conclusion

• Philosophical Question Existence and importance of endogenous preferences?

• Methodological Question How to address in econometric models?

• Modality Style Demonstration– Endogenous and dynamic preferences explicitly modeled

Latent Class Choice Model with feedback of Expected Max Utility;Predict trends via integration with HMM

– Evidence of endogenous and dynamic preferences that are vital for application