Reference Effect: Contrasting Reference-dependent Models

34
Y. Masatlioglu and N. Uler University of Michigan

description

Reference Effect: Contrasting Reference-dependent Models. Y. Masatlioglu and N. Uler University of Michigan. Do Reference Points Influence Economic Outcomes?. Standard Neoclassical Theory: - PowerPoint PPT Presentation

Transcript of Reference Effect: Contrasting Reference-dependent Models

Page 1: Reference Effect: Contrasting Reference-dependent Models

Y. Masatlioglu and N. Uler

University of Michigan

Page 2: Reference Effect: Contrasting Reference-dependent Models

Do Reference Points Influence Economic Outcomes?

Standard Neoclassical Theory: If transactions costs are small enough, reference points

should not influence rational consumers.

In practice, defaults make an enormous difference: Organ donation Car insurance Car purchase options Consent to receive e-mail marketing Savings Asset allocation

Moscow 2009 2

Page 3: Reference Effect: Contrasting Reference-dependent Models

Reference-Dependent Behavior

Endowment Effect The gap between WTA and WTP

Status Quo Bias People often stick with their default options

Reference Effect Reference points alter one's choices even when agents do not stick to the reference point itself

3

Page 4: Reference Effect: Contrasting Reference-dependent Models

The Reference Effect

4

r

y

x

Good 1

Goo

d 2 STATUS QUO BIAS

x is more likely to bechosen from the set {x, y} when x is the endowment,

REFERENCE EFFECTx is more likely to bechosen from the set {x, y, r} when there is an endowment, r.

Page 5: Reference Effect: Contrasting Reference-dependent Models

Reference-Dependent Models

• Behavioral Models (Positive Approach)Tversky and Kahneman (1991)

• Rational Models (Normative Approach)Masatlioglu and Ok (2005, 2007, 2009)

5

Page 6: Reference Effect: Contrasting Reference-dependent Models

Tversky and Kahneman (1991)(Loss Aversion Model (LA))

• Reference dependence

• Loss aversion

• Diminishing sensitivity

)]()([)( iii

iii ruxugxrU

][][ agag ii

6

ig concave for a > 0, convex for a < 0

Page 7: Reference Effect: Contrasting Reference-dependent Models

Shortcomings of the LA model

• Non-convex Indifference Curves

• Unusual Cyclical Choices

• Accommodates not only Status Quo Bias but also Status Quo Aversion

Moscow 2009 8

Page 8: Reference Effect: Contrasting Reference-dependent Models

Unusual Preference Cycles Munro and Sugden (2003) show that one can find three alternatives,

x, y and z such that

)()( yUzU yy

)()( zUxU zz

9

)()( xUyU xx

xyzx xyz

Moscow 2009

Page 9: Reference Effect: Contrasting Reference-dependent Models

Loss Aversion Model predicts Status Quo Aversion EXAMPLE: Consider an environment in which an agent needs to choose

among pairs (x,m), where x and m stand for the units of a private good and money, respectively.

0)(20

)(020

)(8.0

8.0

aifaaifa

agaifaaifa

ag mx

*)(*)(),(*)*,( mmgxxgmxU mxmx

10

U(0,0)(1,100) >U(0,0)(0,103)

Choose between “one mug and $100” and “no mug and $103” when there is no reference:

U(1,100)(1,100) <U(1,100)(0,103)

Now, status quo is “one mug and $100” what would you do?

STATUS QUO AVERSION !!!

Moscow 2009

Page 10: Reference Effect: Contrasting Reference-dependent Models

Constant Loss Aversion (CLA)

11

We say Constant Loss Aversion if there is no diminishing sensitivity or constant sensitivity (gi is linear).

While Loss Aversion Model (i) permits non-convex indifference curves and intransitive preferences(ii) accommodates not only status quo bias but also status quo aversion,

While Constant Loss Aversion does not suffer from these implications, it is much more restrictive than the LA model.

Moscow 2009

Page 11: Reference Effect: Contrasting Reference-dependent Models

The Multi-criteria Choice ModelMasatlioglu and Ok (2005)

If y is chosen when x is the status quo point, then, y must be chosen when y is itself the status quo.

12

Axiom of Status Quo Bias

Moscow 2009

Page 12: Reference Effect: Contrasting Reference-dependent Models

Movie

Sports

r1

r2 r

Acceptable alternatives

when r is status quo

Assume that r is the status quo

u2

u1

U = w1u1+w2u2

Masatlioglu and Ok, 2005 (The SQB Model)

Moscow 2009

Page 13: Reference Effect: Contrasting Reference-dependent Models

Loss Aversion vs. Status Quo Bias

Loss Aversion Model

Status Quo Bias Model

)(xrU

14

Q(r)Bxmax

)(xU

Bxmax

Moscow 2009

Page 14: Reference Effect: Contrasting Reference-dependent Models

The Generalized SQB ModelMasatlioglu and Ok (2009)

)()()()(

rUyUrQyrQr

15

Q(r)xmaxS

)(xU

such that Mental Constraint

Moscow 2009

Page 15: Reference Effect: Contrasting Reference-dependent Models

Moscow 2009

The Procedural RD Model Masatlioglu and Ok (2007)

)()()( rQyQrQy iii

)()()()(

rUyUrQyrQr

i

i

16

otherwiseSS

(r)2Qx(r)1Q{r} if(r)1Qx

max

)(xU

such that

Another generalization of SQB model with two mental constraints, Q1and Q2

Page 16: Reference Effect: Contrasting Reference-dependent Models

Some Facts about Experiment

17

Conducted at C.E.S.S. (NYU) A total of 99 subjects Money and Chocolate Each subject answered 16 questionsOn average, earned $14 including the $7 show-up

fee and also some chocolates

Page 17: Reference Effect: Contrasting Reference-dependent Models

Experimental Design

18

r1

r5 r2r3r7

r6 r4

y

x

r0 Money

Bel

gium

Cho

cola

te

Page 18: Reference Effect: Contrasting Reference-dependent Models

A Screen Shot from the Experiment

19

Page 19: Reference Effect: Contrasting Reference-dependent Models

Experimental Design: Theoretical Predictions

r1 r2

Classical Choice Theory

y y Yes

y x No

x x No

x y No

20

r1

r2

y

x

r0 Good 1

Goo

d 2

There are four possible cases:

Assume that y is preferred to x when there is no reference point.

Moscow 2009

Page 20: Reference Effect: Contrasting Reference-dependent Models

Loss Aversion Model

21

x

r2

y

r1

LA model favors x

Good 1

Good 2

r1 r2 LA CLA

y y Yes Yes

y x Yes Nox x Yes Yesx y No No

Page 21: Reference Effect: Contrasting Reference-dependent Models

Moscow 2009

SQB ModelAcceptable Alternatives Choice

r1 r1 x x

r2 r2 x y y

22

r1

x

y

r2

SQB model favors y

u2

u1

u2

u1

r1 r2 SQB

y y Yes

y x Nox x Yesx y Yes

Good 1

Good 2

Page 22: Reference Effect: Contrasting Reference-dependent Models

Generalized SQB Model

r1 r2 SQB GSQB

y y Yes Yes

y x No Yesx x Yes Yesx y Yes Yes

23

Q(r1)

r2 Q(r2)

r1

x

y Q(r) ∩ B Choice

r1 r1 x y y

r2 r2 x x

GSQB model is indecisive

Good 1

Good 2

Page 23: Reference Effect: Contrasting Reference-dependent Models

Procedural R-D ModelElimination

ChoiceQ1(r) ∩ B Q2(r) ∩ B

r1 r1 r1 x y y

r2 r2 x - x

24

x

Q1(r1)

Q2(r1)

r2

y

Q1(r2)

Q2(r2)

r1

PRD model is also indecisive

r1 r2 SQB PRD

y y Yes Yes

y x No Yesx x Yes Yesx y Yes Yes

Good 1

Good 2

Page 24: Reference Effect: Contrasting Reference-dependent Models

Theoretical Predictions for r1 r2

Tversky - Kahneman Masatlioglu - Ok

CLA LA SQB GSQB PRD

r1→ r2 - Favors x Favors y Indecisive Indecisive

25

r1

r2

y

x

r0 Money

Belgium Chocolate

Page 25: Reference Effect: Contrasting Reference-dependent Models

Theoretical Predictions

Tversky - Kahneman Masatlioglu - Ok

CLA LA SQB GSQB PRD

Group 1

r0→ r5 - I - - I

r0→ r6 - I - - I

r0→ r7 - I - - I

Group 2

r1→ r2 - x y I I

r3→ r4 - x y I y

r5→ r6 - x - - I

r5→ r7 - y - - I

26

r1

r5 r2r3r7

r6 r4

y

x

r0

Page 26: Reference Effect: Contrasting Reference-dependent Models

Moscow 2009

Reference Effect: Preference Reversals

No reversal 269 %73Reversal 97 %27# Obs 366

r1→ r2 r3→ r4 r5→ r6 r5→ r7

r1 r2 r3 r4 r5 r6 r5 r6

No reversal 40 33 42 40 32 30 25 27Reversal 14 21 14 16 5 7 11 9# subjects 54 56 37 36

27

Page 27: Reference Effect: Contrasting Reference-dependent Models

Comparison of the Models

Tversky - Kahneman Masatlioglu - Ok Classical TheoryCLA LA SQB GSQB PRD

Total 65 89 68 75 93 48

28

Explanatory Power of the Models (in percentages)

Page 28: Reference Effect: Contrasting Reference-dependent Models

Selten’s Measure of Predictive Success

29

Page 29: Reference Effect: Contrasting Reference-dependent Models

Conclusion

• Existence of the reference effect• An experiment that distinguish between the

models of reference-dependence. • R-D Models make different predictions

regarding the reference effect. • We find that both the PRD and LA models

explain approximately 90 percent of the data.

30

Page 30: Reference Effect: Contrasting Reference-dependent Models

Moscow 2009 31

THANK YOU

Page 31: Reference Effect: Contrasting Reference-dependent Models

Moscow 2009 34

Page 32: Reference Effect: Contrasting Reference-dependent Models

A Sample Question from the Experiment

38

Page 33: Reference Effect: Contrasting Reference-dependent Models

Is SQB Axiom too strong?

)},,({)},,({yzyczxyxcy

)},,({ xzxcz

41

Take x = CUNY, y = OSU and z = UM

OSU = c({CUNY, OSU},CUNY)

UM = c({OSU, UM},OSU)

But CUNY = c({CUNY, UM},CUNY)

Moscow 2009

Page 34: Reference Effect: Contrasting Reference-dependent Models

Relaxing SQB Axiom?

42

Weak Status Quo Bias

andimplies

implies

Moscow 2009