Download as pdf or txt
Download as pdf or txt
You are on page 1of 94

Department of Economics

The perceptual and psychophysical


foundations of decisions under uncertainty

Miguel Barretto Garcia


Zurich Center for Neuroeconomics
University of Zurich

Washington University at St. Louis


8 April 2022
Department of Economics

Decisions are stochastic

>
EUM

Mosteller & Nogee (1951)


Department of Economics

Decisions are biased

t+1

t+n

Preference Reversals
(e.g., Tversky and Thaler, 1990)
Department of Economics

Decisions are biased

t+1

t+n
Heuristic Decisions
(e.g., Gigerezner, 2008)

Preference Reversals
(e.g., Tversky and Thaler, 1990)
Department of Economics

This Talk In A Nutshell:


Perceptual processes are crucial to our understanding observed
stochastic choice behaviour and choice biases in economic choice
Department of Economics

Perceptual and Psychophysical Perspective

dlPFC
Parietal
ACC
Hippocampus
> vmPFC Visual

Choice Process

Deciding Processing Encoding


Conflict Storing
Executive
Department of Economics

Perceptual and Psychophysical Perspective

Which between the two


clusters contain more dots?
Department of Economics

Perceptual and Psychophysical Perspective

Capacity Constraints

1. Limited time
2. Size perception
3. Background contrasts

Which between the two


clusters contain more dots?
Department of Economics

Bias and noisiness in representation is a direct consequence


of the brain’s capacity constraints

Contrast Effect and processing in V2


Department of Economics

Risk aversion as perceptual bias:


Humans don’t (seem to) like uncertainty

Ambiguity aversion as range-dependent:


People don’t like missing information
Department of Economics

Risk aversion as perceptual bias:


Humans don’t (seem to) like uncertainty
Department of Economics

Risk Aversion
Expected Utility Maximization
§ Concave mapping from monetary amounts to utility

>
representations,

§ “Undervaluation” of utilities at larger payoffs, i.e.


“diminishing marginal utility of wealth”
Department of Economics

Risk Aversion
Expected Utility Maximization
§ Concave mapping from monetary amounts to utility

>
representations,

§ “Undervaluation” of utilities at larger payoffs, i.e.


“diminishing marginal utility of wealth”

Mosteller & Nogee (1951) Issues


§ Random utility models (RUM) assume noise, but not
where noise comes from (Webb, 2018)

§ Persistence of risk aversion in small-stakes gambles,


i.e. Rabin’s calibration paradox (Rabin, 2000; Rabin &
Thaler, 2001)
Department of Economics

Risk Aversion
Expected Utility Maximization
§ Concave mapping from monetary amounts to utility

>
representations,

§ “Undervaluation” of utilities at larger payoffs, i.e.


“diminishing marginal utility of wealth”

Mosteller & Nogee (1951) Issues


§ Random utility models (RUM) assume noise, but not
where noise comes from (Webb, 2018)

§ Persistence of risk aversion in small-stakes gambles,


i.e. Rabin’s calibration paradox (Rabin, 2000; Rabin &
Thaler, 2001)
Department of Economics

Rabin & Thaler (2001)


Department of Economics

Dominant Account
Risk Aversion undervaluation of utilities due to concavity

Observed Undervaluation
>
Department of Economics

Perceptual Account
Risk Aversion underestimation of magnitudes
due to cognitive limitations

Observed Underestimation
>
Department of Economics

Perceptual Noisy & Logarithmic


Account representation of magnitude

Humans Monkeys

(Dehaene, 1998)
(Nieder & Marten, 2007)
Department of Economics

Noisy & logarithmic representation of magnitude

(Harvey et al., 2013; Harvey & Dumoulin, 2017)


Department of Economics

Regressive bias
Recent work from perceptual decision-making
links precision of mental representation directly
to inherent biases that reflect prior beliefs (e.g.,
small magnitudes occur more often)

(Petzschner et al., 2016)


Department of Economics

Regressive bias
Recent work from perceptual decision-making
links precision of mental representation directly
to inherent biases that reflect prior beliefs (e.g.,
small magnitudes occur more often)

(Petzschner et al., 2016)


Department of Economics

Can this regressive bias account for risk aversion?

(Piazza et al., 2004)

> (Mosteller & Nogee, 1951)


Department of Economics

Khaw, Li, & Woodford (2020)


Concave mapping from
magnitude to perceived magnitude

Monetary payoffs during risky choice is


encoded logarithmically and noisily
Department of Economics

Khaw, Li, & Woodford (2020)


Concave mapping from
magnitude to perceived magnitude

Monetary payoffs during risky choice is


encoded logarithmically and noisily

Key Predictions
1. Consistency (variability) of risky decisions
depends on the logarithmic ratio of the
payoffs

2. Precision of payoff representation is


inversely related to the degree of risk
aversion
Department of Economics

Khaw, Li, & Woodford (2020)


Concave mapping from
magnitude to perceived magnitude

Monetary payoffs during risky choice is


encoded logarithmically and noisily

Key Predictions
1. Consistency (variability) of risky decisions
depends on the logarithmic ratio of the
payoffs

2. Precision of payoff representation is


inversely related to the degree of risk
aversion
Department of Economics

Khaw, Li, & Woodford (2020)


Concave mapping from
magnitude to perceived magnitude

Monetary payoffs during risky choice is


encoded logarithmically and noisily

Key Predictions
1. Consistency (variability) of risky decisions

probability, f(𝛽 !" )


depends on the logarithmic ratio of the

Risk-neutral
payoffs

2. Precision of payoff representation is


inversely related to the degree of risk
aversion

Precision (1⁄𝜈)
Department of Economics

Questions

1. Do decision makers employ similar magnitude representations for risky


decisions and perceptual decisions? How well does this mechanism generalize
across different choice domains and stimulus contexts?

2. Can we employ neural data to quantify the precision of these representations


and predict risky decision making in a fully independent context?
Department of Economics

The Experiment
To directly relate individual differences (n = 64) in precision of magnitude representations
during numerosity estimation and apparent risk aversion during risk choice
Department of Economics

The Experiment
To directly relate individual differences (n = 64) in precision of magnitude representations
during numerosity estimation and apparent risk aversion during risk choice

Numerosity Task
(fMRI; 216 trials)
Department of Economics

The Experiment
To directly relate individual differences (n = 64) in precision of magnitude representations
during numerosity estimation and apparent risk aversion during risk choice

Gambling Task
(Behavior; 480 trials)
Numerosity Task
(fMRI; 216 trials)

Symbolic Numbers
Coin Clouds
Department of Economics

1. Are the same noisy logarithmic representations used in all three


choice domains?
Department of Economics

1. Are the same noisy logarithmic representations used in all three


choice domains? Numerosity

Consistency of choice is a
logarithmic ratio of the
payoffs
Department of Economics

1. Are the same noisy logarithmic representations used in all three


choice domains? Numerosity Risk

Consistency of choice is a
logarithmic ratio of the
payoffs
Department of Economics

1. Are the same noisy logarithmic representations used in all three


choice domains? Numerosity Risk

Consistency of choice is a
logarithmic ratio of the
payoffs
Department of Economics

1. Are the same noisy logarithmic representations used in all three


choice domains? Numerosity Risk

Consistency of choice is a
logarithmic ratio of the
payoffs

Evidence that a general


mechanism is at play

Page 35
Department of Economics

2. Can we predict risk aversion in the risky choice task by


the performance in the perceptual numerosity task?

Non-symbolic Symbolic
Department of Economics

3. Can we predict risk aversion in the risky choice task, using the
precision of neural representations in the perceptual numerosity task?
Department of Economics

3. Can we predict risk aversion in the risky choice task, using the
precision of neural representations in the perceptual numerosity task?
Step 2: Decode numerical posterior from unseen data – P(s|d)

(van Bergen et al., 2015)


17/06/2020
Department of Economics

3. Can we predict risk aversion in the risky choice task, using the
precision of neural representations in the perceptual numerosity task?
Department of Economics

3. Can we predict risk aversion in the risky choice task, using the
precision of neural representations in the perceptual numerosity task?
Department of Economics

Summary 1
§ We provide empirical evidence that small-stakes risk aversion might (at
least partly) result from a perceptual process, rather than (only) a valuation
process

§ We show that individual differences in risk aversion can be predicted by


both behavior and neural data in a purely numerosity decision-making task

§ These results provide evidence that risk aversion is linked to the precision
of mental/neural magnitude representations that are used for choices in
different domains and contexts
Department of Economics

Risk aversion as perceptual bias:


Humans don’t (seem to) like uncertainty

Ambiguity aversion as range-dependent:


People don’t like missing information
Department of Economics

Ambiguity aversion as range-dependent


Humans will rely on whatever available information
A AMBIGUITYA AMBIGUITY Attention Reference NoiseRefe
Attention

#C
100 #
100C
+ #C #
+
' '
50% %

? ?
" "
" " &
50% ( (
− #$ − #
Decision
#0$ 0$ Decision
# ./0 1 234 567 8 /0./0
9234 /67
1 234: 567 8
+, = +;<=
, = ;<
Department of Economics

EUM and standard cognitive choice models


Assume a utility / value function for each of the gambles
A AMBIGUITYA AMBIGUITY Attention Reference NoiseRefe
Attention

#C
100 #
100C
+ #C #
+
' '
50% %

? ?
" "
" " &
50% ( (
− #$ − #
Decision
#0$ 0$ Decision
# ./0 1 234 567 8 /0./0
9234 /67
1 234: 567 8
+, = +;<=
, = ;<
Department of Economics

EUM and standard cognitive choice models


Assume a utility / value function for each of the gambles
A AMBIGUITYA AMBIGUITY Attention Reference NoiseRefe
Attention

#C
100 #
100C
+ #C #
+
' '
50% %
𝑈(𝑅𝑖𝑠𝑘)
? ?
" "
" " &
50% ( (
− #$ − #
Decision
#0$ 0$ Decision
# ./0 1 234 567 8 /0./0
9234 /67
1 234: 567 8
+, = +;<=
, = ;<
Department of Economics

EUM and standard cognitive choice models


Assume a utility / value function for each of the gambles
A AMBIGUITYA AMBIGUITY Attention Reference NoiseRefe
Attention

#C
100 #
100C
+ #C #
+
' '
50% %
𝑈(𝐴𝑚𝑏) 𝑈(𝑅𝑖𝑠𝑘)
? ?
" "
" " &
50% ( (
− #$ − #
Decision
#0$ 0$ Decision
# ./0 1 234 567 8 /0./0
9234 /67
1 234: 567 8
+, = +;<=
, = ;<
Department of Economics

Second-order probabilities

Camerer and Weber (1992)


Department of Economics

Lessons from Psychophysics and Attention


§ DMs directly attend to and compare stimulus features

§ Possible gains and losses


§ They are not computing higher-order probabilities instantaneously
Department of Economics

Lessons from Psychophysics and Attention


§ DMs directly attend to and compare stimulus features

§ Possible gains and losses


§ They are not computing higher-order probabilities instantaneously

§ DMs perception of magnitudes are noisy and (log)-compressed

§ Range sensitivity: the wider the range between values, the less
sensitive the DM is to deciding

§ Magnitude sensitivity: the larger the overall value, the less sensitive
the DM in deciding
Department of Economics

Basic Building Blocks – Ambiguity


1. DM evaluates the possible gains, g, and losses, l, w.r.t. a reference

A AMBIGUITY Attentio
Possible Losses 𝑙 = 𝑠 − 𝑣& #C +
'
Possible Gains 𝑔 = 𝑣' − 𝑠

?
"
"
(

Decision
#$ ./0 1 2
+, =
Department of Economics

Basic Building Blocks – Ambiguity


1.1 Attention: DM is biased in focusing their attention on winning more
with 𝑣' or losing by receiving only 𝑣&

Possible Losses 1−𝛽 𝑙 A AMBIGUITY Attentio

#C +
Possible Gains 𝛽𝑔 '

?
"
"
Attention Parameter (
𝛽 ∈ 0,1 −
Decision
#$ ./0 1 2
+, =
Department of Economics

Basic Building Blocks – Ambiguity


1.2 Comparison Distortion: DM evaluates G and L imperfectly or noisily

Possible Losses 1 − 𝛽 𝑙( A AMBIGUITY Attentio

𝛽𝑔(
#C +
Possible Gains
'

?
Attention Parameter "
𝛽 ∈ 0,1 "
(
Sensitivity Parameter −
𝛼 ∈ 0,1
Decision
#$ ./0 1 2
+, =
Department of Economics

Basic Building Blocks – Ambiguity


1.3 Gain-Loss Ratio: DM will deviate from S and consider choosing U by
weighing the relative gains and losses.
A AMBIGUITY Attentio

#C +
exp 1 − 𝛽 𝑙( '
𝑚) =
exp 𝛽𝑔(
?
"
"
(

Decision
#$ ./0 1 2
+, =
Department of Economics

Basic Building Blocks – Ambiguity


2. Bias: DM has a bias that is modulated by the payoff values

+, + -.,
𝑚* = 𝑣' 𝑣& A AMBIGUITY Attentio

(Cobb-Douglas form)
#C +
'

?
Attention Parameter "
𝛽 ∈ 0,1 "
(

Bias Parameter
Decision
< 0, biased towards 𝑆 #$ ./0 1 2
𝜃 = ) = 0, unbiased +, =
> 0, biased away from 𝑆
Department of Economics

Basic Building Blocks – Ambiguity


3. Normalization: DM’s overall perception of the comparison process is
constrained with normalization
A AMBIGUITY Attentio
Range Sensitivity #C +
𝑟 = 𝑣' − 𝑣& '

?
Magnitude Sensitivity "
𝑧 = 𝑣' + 𝑣&
"
(

Decision
#$ ./0 1 2
+, =
Department of Economics

Basic Building Blocks – Ambiguity


3.1 Compressed Normalization: DM’s perception of range and magnitude
is constrained via compression

𝑟 ln 𝑟

1
ln 𝑥 𝜂=
𝛿ln 𝑟 ln 𝑧

𝑧 ln 𝑧

Sensitivity Parameter
𝛿 ∈ 0,1
Department of Economics

Basic Building Blocks – Ambiguity


exp 1 − 𝛽 𝑙 (
Gain-Loss Ratio 𝑚) =
exp 𝛽𝑔(

Bias +, + -.,
𝑚* = 𝑣' 𝑣&

1
Compressed Normalization 𝜂=
𝛿 ln 𝑟 ln 𝑧

Next Step: Estimate the probability of choosing the safe gamble


Department of Economics

Derivation
/
𝑚)
Pr 𝑆 > 𝐴 = / / Luce Form (Logit)
𝑚) + 𝑚*
Department of Economics

Derivation
/
𝑚)
Pr 𝑆 > 𝐴 = / / Luce Form (Logit)
𝑚) + 𝑚*

Gain-Loss Ratio
.-
1 , -., exp 1 − 𝛽 𝑙(
= 1 + exp −𝜃 ln 𝑣' 𝑣& + ln
𝛿 ln 𝑟 ln 𝑧 exp 𝛽𝑔(

Compressed Bias
Normalization
Department of Economics

1
Range Sensitivity Model Pr 𝑆 > 𝐴 =
1 + exp −𝑑
we simplify the equation as: Gain-Loss Ratio
$#
exp 1 − 𝛽 𝑠 − 𝑣#
𝑑 = 𝑎! + 𝛼" ln
exp 𝛽 𝑣% − 𝑠 $#

Distortion Parameter 𝛼' = 1 − 𝛿

Choice Consistency 1
𝛼" =
(Slope) 𝛿 ln 𝑟 ln 𝑧

𝜃 ln(𝑚& )
Bias (Intercept) 𝑎! = −
𝛿 ln 𝑟 ln 𝑧
" ?
Department of Economics

Basic Building Blocks – Risk #$


Decision
./
+, =
4. Odds: Probability is represented as odds

𝑝
B RISK Att
𝑝0 =
1−𝑝
“High” odds #C
A

1−𝑝
𝑝1 =
𝑝
“Low” odds "
1−A

#$ Decision
?
./0 >8 1
+< =
" ?
Department of Economics

Basic Building Blocks – Risk #$


Decision
./
+, =
4.1 Distorted Odds: Odds are imperfectly encoded

B RISK Att
#C
𝑝 2
2 “High” odds
𝑝0 =
1−𝑝
A
2
2 1−𝑝
𝑝1 =
𝑝
“Low” odds "
1−A

Odds Distortion Parameter

#$
< 1, 𝑢𝑛𝑑𝑒𝑟𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒 𝑜𝑑𝑑𝑠 Decision
𝛾 = ) = 1, 𝑛𝑜 𝑑𝑖𝑠𝑡𝑜𝑟𝑡𝑖𝑜𝑛 ?
> 1, 𝑜𝑣𝑒𝑟𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒 𝑜𝑑𝑑𝑠 ./0 >8 1
+< =
Department of Economics

Basic Building Blocks – Risk


Distorted odds enters as an “attentional weight” that shifts our attention
towards high and low payoffs

2
exp 1 − 𝛽 𝑝1 𝑙 (
Gain-Loss Ratio 𝑚) = 2
exp 𝛽𝑝0 𝑔(

" "
+,4 + -., 4#
Bias 𝑚3 = 𝑣' ! 𝑣&
Department of Economics

1
Range Sensitivity Model Pr 𝑆 > 𝑅 =
1 + exp −𝑑
we simplify the equation as: Gain-Loss Ratio
) $#
exp 1 − 𝛽 𝑝( 𝑠 − 𝑣#
𝑑 = 𝑎! + 𝛼" ln ) $#
exp 𝛽𝑝* 𝑣% − 𝑠

Distortion Parameter 𝛼' = 1 − 𝛿

Choice Consistency 1
𝛼" =
(Slope) 𝛿 ln 𝑟 ln 𝑧

𝜃 ln(𝑚+ )
Bias (Intercept) 𝑎! = −
𝛿 ln 𝑟 ln 𝑧
Department of Economics

Parameter and Model Recoveries


Department of Economics

Parameter and Model Recoveries


Department of Economics

Predictions: Parameters
Ambiguity / Risk
The higher the noise, 𝛿,
the less sensitive or more random the behaviour
Department of Economics

Predictions: Parameters
Ambiguity / Risk
The higher the noise, 𝛿,
the less sensitive or more random the behaviour

Ambiguity / Risk
The larger the bias, 𝜃,
the more it shifts the DM to choose S
Department of Economics

Predictions: Parameters
Ambiguity / Risk
The higher the noise, 𝛿,
the less sensitive or more random the behaviour

Ambiguity / Risk
The larger the bias, 𝜃,
the more it shifts the DM to choose S

Ambiguity / Risk
The larger the 𝛽 (attention towards 𝑣' ),
the more it shifts the DM towards A or R
Department of Economics

Predictions: Parameters

Risk
The larger the 𝛾,
the more sensitive the DM in choosing
between S and R
%

?
"
" &
(
− #$
Department of Economics
Decision
#$
Experiment 1 +, =
./0 1 234 567 8 /0 9234 /67 :
;<=

A AMBIGUITY B Attention
RISKReference Noise Attention Reference Noise
C
#C #
#C + #C
+ A
' A '
% %
?
"
" " &
"
&
1−A
( (
− #$
− #$ 1−A

#$
Decision
#$ Decision
./0 1 234 567 8 /0 9234 /67 : ? 234 ? 234
+, = ./0 >8 1 567 8 /0 >: 9 /67
;<= +< =
;<=
§ varied R (9 levels)
B andRISK
Z (low and high) Attention Reference Noise
C
# # + C
§ fixed expected values to match
A
ambiguity and risky trialsA
'
%
" "
&
1−A
(
− #$ 1−A
$ Decision
Department of Economics

Predictions: Range
Ambiguity

1 The less sensitive the slope, 𝛼- ,


𝛼" =
𝛿 ln 𝑟 ln 𝑧 as range increases
Department of Economics

Predictions: Range
Ambiguity

1 The less sensitive the slope, 𝛼- ,


𝛼" =
𝛿 ln 𝑟 ln 𝑧 as range increases
S
𝜃 ln(𝑚& ) The lower the bias for s, 𝛼5 ,
𝑎! = −
𝛿 ln 𝑟 ln 𝑧 the larger the shift towards A
A
Department of Economics

Predictions: Range
Ambiguity

1 The less sensitive the slope, 𝛼- ,


𝛼" =
𝛿 ln 𝑟 ln 𝑧 as range increases

Pr(S > A)
S
𝜃 ln(𝑚& ) The lower the bias for s, 𝛼5 ,
𝑎! = −
𝛿 ln 𝑟 ln 𝑧 the larger the shift towards A
S - EV
A
Department of Economics

Predictions: Range
Ambiguity

1 The less sensitive the slope, 𝛼- ,


𝛼" =
𝛿 ln 𝑟 ln 𝑧 as range increases

Pr(S > A)
S
𝜃 ln(𝑚& ) The lower the bias for s, 𝛼5 ,
𝑎! = −
𝛿 ln 𝑟 ln 𝑧 the larger the shift towards A
S - EV
A

DM will choose S on average the wider the range


Department of Economics

Predictions: Range
Risk

1 The less sensitive the slope, 𝛼- ,


𝛼" =
𝛿 ln 𝑟 ln 𝑧 as range increases
Department of Economics

Predictions: Range
Risk

1 The less sensitive the slope, 𝛼- ,


𝛼" =
𝛿 ln 𝑟 ln 𝑧 as range increases 𝑝
1−𝑝
First Half S
% %
,-.$ , "/- .&
𝜃 ln 𝑚+ 𝜃 ln 𝑣% 𝑣#
𝑎! = − =
𝛿 ln 𝑟 ln 𝑧 𝛿 ln 𝑟 ln 𝑧 R
R

Pr(S > A) S - EV
Department of Economics

Predictions: Range
Risk

1 The less sensitive the slope, 𝛼- ,


𝛼" =
𝛿 ln 𝑟 ln 𝑧 as range increases 𝑝
1−𝑝
First Half
% %
,-. , "/- .&
𝜃 ln 𝑚+ 𝜃 ln 𝑣% $ 𝑣#
𝑎! = − =
𝛿 ln 𝑟 ln 𝑧 𝛿 ln 𝑟 ln 𝑧 R

Pr(S > A)
S
Second Half % %
,-. , "/- .&
𝜃 ln 𝑚+ 𝜃 ln 𝑣% $ 𝑣#
𝑎! = − =
𝛿 ln 𝑟 ln 𝑧 𝛿 ln 𝑟 ln 𝑧
R S - EV
Department of Economics

Results
RSM can differentially fit across different conditions of ambiguous and risky trials
Department of Economics

Results
The data is consistent with the model’s prediction that choosing the ambiguous gamble
depends on the range
Department of Economics

Interim Summary
§ We provide evidence that our range sensitivity model (RSM) can predict
behavior in both risky and ambiguous decisions

§ Ambiguity aversion appears to be an epiphenomenon of decision makers


relying on available range information
?
"
" &
(
− #$
Department of Economics Decision
#$ ./0 1 234 567 8 /0 9234 /67 :
+, =
Experiment 2: Ellsberg-type choice ;<=

A B
AMBIGUITY RISK Attention Reference Noise
Attention Reference Noise
#C #C #C + #C
+ A
A ' '
% %
?"
" "
" & &
1−A
(
(
− #$
− #$ 1−A
Decision
#$ #$ Decision
./0 1 234 567 8 /0 9234 /67 :
+, = >8
./0 ;<=
? 234
1 567 8 /0 >:
? 234
9 /67 :
+< =
;<=

B RISK Attention Reference Noise


§ matched the expected value Cfor both ambiguity and risk
# + #C
A
A
§ varied the range (15 levels) '
%
" EV, 6 levels)
§ overall magnitude (or
"
&
1−A
(
− #$ 1−A

#$ Decision
8 ? 234 8 : ? 234 :
Department of Economics

Experiment 2: Ellsberg-type choice


Prediction:
𝛽

§ The only relevant information that varies is odds information


𝑝
§ 𝜸: DM will behave according to the odds 1−𝑝

R
Department of Economics

Experiment 2: Ellsberg-type choice


Prediction:
𝛽

§ The only relevant information that varies is odds information


A
§ 𝜸: DM will behave according to the odds 𝜃 ln(𝑚+& )
𝑎! = −
𝛿 ln 𝑟+& ln 𝑧+&
§ 𝜽: Magnitude will bias the DM to be indifferent,
or choose R or A R
(shifts toward choosing R)
Department of Economics

Results

§ Behaviour varies according to the odds

§ Magnitude modulates the bias whether the DM R or A


Department of Economics

Model Comparison
Our model (RSM) performed better versus classical preference-based / cognitive
models

Experiment 1 Experiment 2
Department of Economics

Model Comparison
Our model (RSM) performed better versus classical preference-based / cognitive
models

Experiment 3 Experiment 4
Department of Economics

Model Comparison
Our model (RSM) performed better versus classical preference-based / cognitive
models
Different Payoff
Low Payoff is Zero
Presentation

Gain/Loss Frame Varying Probabilities Mixed Gambles


Department of Economics

Summary
§ We provide evidence that our range sensitivity model (RSM) can predict
behaviour in Ellsberg-type decisions

§ Ambiguity aversion appears to be an epiphenomenon of decision makers


relying on available probability and magnitude information (i.e., source
uncertainty)

§ RSM not only predicts behaviour, but appears to be a relatively superior


model as opposed to classical preference-based / cognitive models
Department of Economics

Future Directions
§ To further test and refine these psychophysical models to neural data
Department of Economics

Future Directions
§ To further test and refine these psychophysical models to neural data

§ To further unpack more perceptual / psychophysical processes and how


that can explain further economic choice biases

§ Intertemporal choice
§ Preference reversals
§ Information biases
Department of Economics

Future Directions
§ To further test and refine these psychophysical models to neural data

§ To further unpack more perceptual / psychophysical processes and how


that can explain further economic choice biases

§ Intertemporal choice
§ Preference reversals
§ Information biases

§ Efficient coding: To relate perceptual processes to memory and


attentional systems in the brain

§ Micro-foundations of choice: To develop neurobiologically-founded


computational models of decision making and test its assumptions to
electrophysiological data
Department of Economics

Thanks to The Team

Gilles de Hollander Marcus Grueschow Marius Moisa Sebastian Weissengruber


(UZH) (UZH) (UZH)

Rafael Polanía Michael Woodford Christian Ruff


(ETH Zurich) (Columbia) (UZH)
Department of Economics

Thanks to the ZNE

You might also like