Behavioral Time Discounting

Behavioral Economics

Dr. Si Chen
Outline

1. Discounted Utility Models

and Exponential Discounted Utility Model

2. Quasi-Linear Discounted Utility Model

3. Applications of Present Bias

Self-Control Problem
Sophisticates vs. Naifs
Commitment Devices

2025 Dr. Si Chen Slide 2

Temporal Choices I
• Decisions that have a time dimension are central to many economic problems.
• the amount and type of savings;
• timing of retirement;
• investment in liquid versus illiquid assets;
• purchase of an annual, or pay-as-you-go health club membership.
• There is a long history of economic thought devoted to analyzing the complex
issues associated with temporal choices.

2025 Dr. Si Chen Slide 3

Temporal Choices II
• The basic idea in most modern theories of time preference is the following:
• Delaying a reward can increase its size, e.g., due to the accrual of interest
payments.
• However, delayed outcomes are less valuable to impatient individuals.
• This trade-off determines the optimal temporal choice of individuals.

2025 Dr. Si Chen Slide 4

 Discounted Utility Models
and Exponential Discounted Utility Model

2025 Dr. Si Chen Slide 5

Temporal Consumption Profile
• In temporal choices, the basic unit of analysis is preferences over sets of
outcome–time pairs, or sequences of consumption levels.
• We formally define a temporal consumption profile as following:
• Suppose that t ∈ Γ = {0, 1, 2, ..., T }, where 0 < T ≤ ∞.
• Denote by ct ∈ C ⊂ R, the real-valued consumption of an individual at time t.
• Let c0 = (c1 , c2 , ..., cT ) denote the temporal consumption profile of the
individual at time t = 0.
• Denote by C ⊂ RT ++1 , the set of all bounded consumption profiles.

–worksheet–

2025 Dr. Si Chen Slide 6

Discounted Utility Models
• One popular utility representation of preferences over C is the discounted utility
model (DU).
• It requires the existence of a utility function U : C → R such that,

T
X
U (c0 ) = D(t)u(ct ); c0 ∈ C.
t=0

• The discount function, D, is often assumed to be decreasing in order to reflect

impatience, i.e., individuals prefer to receive the same reward earlier rather than
later. Hence, D(τ ′ ) > D(τ ) for τ ′ < τ .

2025 Dr. Si Chen Slide 7

The Exponential Discounted Utility Model (EDU) I
Recall
The Exponential Discounted Utility Model

T
X
U (c0 ) = δ t u(ct )
t=0

where 0 ≤ δ ≤ 1.

• The expected discounted utility model is a special case of discounted utility

models with
D(t) = δ t ,

where δ is known as the discount factor.

• An amount is “discounted less” (or is more salient) if the discount factor applied
to such an amount is higher.

2025 Dr. Si Chen Slide 8

Discount Rate

• θ > 0 captures the common assumption of impatience in the discounted utility

models.
• What is the discount rate of the EDU?
–worksheet–

2025 Dr. Si Chen Slide 9

Constant Discounting
One important assumption of EDU is constant discounting:
The discount rate between any two consecutive periods is θ, irrespective of the
the time–location of the periods.

2025 Dr. Si Chen Slide 10

Time Consistency of Consumption Plans
• The EDU model allows one to compare distinct consumption profiles.
• For instance, at time 0, under EDU, the consumption profile ca is preferred to the
consumption profile cb if and only if

T
X T
X
δ t u(cat ) ≥ δ t u(cbt ).
t=0 t=0

• Now suppose that we allow the individual to reoptimize at some future date
τ > 0. For any t > τ .
• If the optimal plan for time t, τ ≤ t ≤ T, made at two different times, 0 and
τ > 0, is identical, then such a plan is said to be time consistent.
• Under EDU, a consumption profil chosen at time t is always time consistent.
–worksheet–

2025 Dr. Si Chen Slide 11

 Quasi-Linear Discounted Utility Model

2025 Dr. Si Chen Slide 12

The Common Difference effect

Thaler’s (1981) apples example:

It is reasonable to assume that a person will prefer one apple today to two
apples tomorrow. The same person will, however, in all likelihood, prefer two
apples in 51 days to one apple in 50 days.
• Such a pattern of preference reversals is not consistent with the assumption of
constant δ in EDU.

2025 Dr. Si Chen Slide 13

Quasi-Hyperbolic Model
Quasi-Hyperbolic Discounting (Present-Biased Preferences)

T
X
U (c0 , c1 , ..., cT ) = u(c0 ) + β δ t u(ct ),
t=1

where 0 ≤ δ ≤ 1 and 0 ≤ β ≤ 1.

• δ plays the same role that it plays in the EDU model.

• The innovation is the presence of the term β ∈ (0, 1) that shrinks the value of
future lifetime utility relative to current utility, thereby creating an additional
present bias for current consumption.

2025 Dr. Si Chen Slide 14

Quasi-Hyperbolic Model and the Common Difference Effect

T
X
U (c0 , c1 , ..., cT ) = u(c0 ) + β δ t u(ct ),
t=1

• When making a choice between two dates t > 0 and t′ > t, the parameter β
washes out, and choices are made as under EDU.
• However, when these choices are brought forward by t periods, i.e., between
two dates 0 and t′ − t, then the present bias for consumption kicks in through
the parameter β ,
• potentially reversing the earlier choice. Hence, this gives rise to a common
difference effect.

–worksheet–

2025 Dr. Si Chen Slide 15

 Applications of Present Bias

2025 Dr. Si Chen Slide 16

Multiple Selves
• In applications of present-biased preferences, it is typical to consider a decision
maker as a succession of multiple selves, one for each time period.
• And it is assumed that all selves exhibit quasi-hyperbolic discounting.

We introduce a subscript to denote the perspective of the time t self.

T
X
Ut (ct , ct + 1, ..., cT ) = u(ct ) + β δ k u(ck ),
t=k

2025 Dr. Si Chen Slide 17

 Self-Control Problem

2025 Dr. Si Chen Slide 18

Self–Control Problem
“I am going to the gym tomorrow.”
“I’m gonna start writing the term paper next week.”

An example from the DellaVigna and Malmendier (2006) study described

earlier is the finding that 80% of monthly members of the health clubs ended
up paying more per visit on average than they would have paid if they had
chosen a different option of paying for ten-visit blocks.

2025 Dr. Si Chen Slide 19

Benchmark: Exponential Discounting
PT k u(c )
Ut = u(ct ) + k=t δ k

Does self–control problem occur if an individual has exponential discounting?

Example: do I go to the gym on day 1?

T = 3 : three periods {0,1,2};
c : effort cost occurs when the activity takes place at day 1;
v : gains from the activity occurs later at day 2.

Day 0 view: Day 1 view:

Do I plan to go to the gym on day 1? Do I actually go to the gym today?
No gym: U0 (no gym) = 0. No gym: U1 (no gym) = 0.
Gym: U0 (gym) = −δc + δ 2 v Gym: U1 (gym) = −c + δv
Hence, on day 0, I plan to go to gym on Hence, on Day 1, I actually go to gym iff
day 1 iff U0 (gym) > 0 U1 (gym) > 0
2
−δc + δ v > 0 −c + δv > 0
δv > c δv > c
2025 Dr. Si Chen Slide 20
Self–Control Problem I
PT k u(c )
Ut = u(ct ) + β k=t δ k
T =3 : three periods {0,1,2};
c : effort cost occurs when the activity takes place at day 1;
v : gains from the activity occurs later at day 2.

Day 0 view: Do I plan to go to the gym on day 1?

No gym: U0 (no gym) = 0. 
Gym: U0 (gym) = β −δc + δ 2 v
Hence, on day 0, I plan to go to gym on day 1 iff

U0 (gym) > 0
β −δc + δ 2 v > 0
 

δv > c

2025 Dr. Si Chen Slide 21

Self–Control Problem II
k u(c )
P
Ut = u(ct ) + β k∈{1,...,T } δ k

c : effort cost occurs when the activity takes place

v : gains from the activity occurs later
T =2 : three periods: 0,1,2.

Day 1 view: Do I actually go to the gym today?

No gym: U1 (no gym) = 0.
Gym: U1 (gym) = −c + β [δv]
Hence, on day 1, I actually go to gym iff

U1 (gym) > 0
−c + β [δv] > 0
βδv > c

2025 Dr. Si Chen Slide 22

Self–Control Problem III
Day 0 vs. Day 1
Gym iff δv > c βδv > c
c c
v <δ v < βδ
Rearrange

Recall that β ≤ 1.
A person will have self–control problem if she has present bias, i.e., β < 1 and
c
βδ < < δ.
v
Note that without present bias, i.e., if β = 1, the person has no self–control
problem.

2025 Dr. Si Chen Slide 23

 Sophisticates vs. Naifs

2025 Dr. Si Chen Slide 24

Sophistication
Let β̂ be a decision maker’s perceived β.
• Time consistent agents (TCs): β̂ = β = 1;
• Sophisticates: β̂ = β < 1
• Naifs: β̂ = 1

2025 Dr. Si Chen Slide 25

A Case Study: Immediate Costs & Delayed Reward
First, we simplify by δ = 1.
Imagine, you need to write a term paper on one of the following four
weekends.
On each weekend, a different movie is on:
• Week 1: A mediocre movie
• Week 2: A good movie
• Week 3: An excellent movie
• Week 4: Best of all, a Marvel Studio movie.

Numerical example:
• Rewards: realize in week 5, v = 30;
• Costs: c = (3, 5, 8, 13).

2025 Dr. Si Chen Slide 26

Naives vs Sophisticates: A Model I
Strategy
• A strategy is a complete plan of action:
s = (s1 , s2 , s3 , s4 ) with sk ∈ {Y, N }.

Utility (assuming δ = 1)
(
−ck + βv if k = t
Ut (go to movie in week k) =
β(−ck + v) if k > t

2025 Dr. Si Chen Slide 27

Naives vs Sophisticates: A Model II
Equilibrium Concept:
• Intuitively, an equilibrium strategy needs to be optimal in each period, given the
expectations about future behavior.
• The expectations must be correct.
We call this a perception-perfect strategy.

2025 Dr. Si Chen Slide 28

Naifs vs Sophisticates: A Model III
Perception-Perfect Strategy for TCs
A strategy sT C for a time consistent agent is perception perfect if sTt C = Y only if
Ut (t) ≥ Ut (k), ∀k ≥ t.

Back to the example: when would a TC (β = 1) work on the paper?

• Does she work in week 1?


−3 + v if k =1

−5 + v if k =2
U1 (go to movie in week k) =
−8 + v
 if k =3

−13 + v if k =4


• Does she work in week 2?

...
Equilibrium sT C = {Y, Y, Y, Y }.
2025 Dr. Si Chen Slide 29
Naifs vs Sophisticates: A Model IV
Perception-Perfect Strategy for Naifs
A strategy sN for a naive agent with β < 1 is perception perfect if
sN
t = Y only if Ut (t) ≥ Ut (k), ∀k ≥ t.

Back to the example, when would a naive (assume β = .5) work on the paper?

• Does she work in week 1? • Does she work in week 2?

 

−3 + .5v if k = 1 −5 + .5v
 if k = 2

.5(−5 + v) if k = 2 U2 (k) = .5(−8 + v) if k = 3
U1 (k) = 
.5(−8 + v)
 if k = 3

.5(−13 + v) if k = 4

.5(−13 + v) if k = 4

• Does she ...

Equilibrium sN = {N, N, N, Y }.

2025 Dr. Si Chen Slide 30

Naifs vs Sophisticates: A Model V
Perception-Perfect Strategy for Sophisticates
A strategy sS for a sophisticated agent with β < 1 is perception perfect if
′ ′
sS S
t = Y only if Ut (t) ≥ Ut (k ), for k = min{k > t with sk = Y }.

What is sophisticated about this?

• Sophisticate does it today, if utility of doing it today is higher than on date k’
when she would do it otherwise, given her present-biased preference.
• This means, she is aware of her self–control problem.
• And she can anticipate her future behavior.

2025 Dr. Si Chen Slide 31

Naifs vs Sophisticates: A Model VI
Perception-Perfect Strategy for Sophisticates
A strategy sS for a sophisticated agent with β < 1 is perception perfect if
′ ′
sS S
t = Y only if Ut (t) ≥ Ut (k ), for k = min{k > t with sk = Y }.

Back to the example, when would a sophisticate (assume β = .5) work on the
paper? We start backwards, since she knows what her future self would do.
• If time comes to week 4 does she • If time comes to week 2 does she
work? k′ = 4
work? U4 (4) = −13 + .5 × 30. Yes. (
• If time comes to week 3 does she −5 + .5v if k = 2
U2 (k) = Yes.
work? k′ = 4 .5(−13 + v) if k = 4
(
−8 + .5v if k = 3 • In week 1: k′ = 2
U3 (k) = No.
.5(−13 + v) if k = 4 (
−3 + .5v if k = 1
U1 (k) = No.
.5(−5 + v) if k = 2
Equilibrium sS = {N, Y, N, Y }.
2025 Dr. Si Chen Slide 32
Summary of the Example
In our numerical example:

TCs do it in week 1, because the costs are the lowest.

—Prediction of the standard model.

Naifs do it in week 4.
—Come week 2, they think they are going to do it in week 3...

Sophisticates do it on week 2.
—They know that they are not going to do it in week 3. Hence, they prefer
doing it in week 2 rather than week 4.

Have you understood the mechanism behind the results?

2025 Dr. Si Chen Slide 33

Immediate Costs vs. Immediate Reward
The example before : immediate costs & delayed reward
What if : delayed costs & immediate reward

Recall:
• Individuals with present–biased preferences discount future utility.
• Sophisticates are aware of it, while the naifs are not.

Exercise in problem set 2.

2025 Dr. Si Chen Slide 34

Present Bias: General Results I
Theorem
The Naifs
• When costs are immediate, naifs do it no earlier than TCs.
• When rewards are immediate, naifs do it no later than TCs.

Interpretation:
• Tendency towards less immediate cost/more immediate benefit.

2025 Dr. Si Chen Slide 35

Present Bias: General Results II
Theorem
The Sophisticates
• Sophisticates do it no later than naifs.

Interpretation:
• Sophistication can mitigate or exacerbate self–control problem.
• They consume earlier when rewards are immediate (compared to TCs and naifs).

2025 Dr. Si Chen Slide 36

Other Important Applications of Present Bias
Not only procrastination, but also:
• consumption vs. (pension) saving
• health behavior like quit smoking, exercising etc...
• ...

2025 Dr. Si Chen Slide 37

Welfare
Does self-control problem reduce welfare of the individuals?

2025 Dr. Si Chen Slide 38

Summary
• Present-biased preferences can cause significant departures from what standard
model predicts.
• Immediate costs/rewards determine behavioral tendency of the naifs.
• Behavior of Sophisticates is more complex:
• Second behavioral force: anticipation of future actions.

2025 Dr. Si Chen Slide 39

 Commitment Devices

2025 Dr. Si Chen Slide 40

Commitment Devices
• Deposit contracts/commitment contracts that involve people voluntarily
depositing money into accounts that they can access again only if they
accomplish a goal.
• Self-imposed deadlines

• They only work when the individual is aware of their self–control problem.

2025 Dr. Si Chen Slide 41

Commitment Contracts I
Kaur et al. (2010)
Background:
• In an Indian data entry firm, workers use data entry software to type information
from scanned images into fields on their computer screen.
• Workers receive piece-rate payment and are paid weekly.
• Goal: help workers to improve their productivity.

2025 Dr. Si Chen Slide 42

Commitment Contracts II
The commitment contract:
• It allows workers to set their own production target, which can be zero.
• Under it, they receive a piece rate of w as before if they meet the target, but only
receive w/2 if they fall short of it.

2025 Dr. Si Chen Slide 43

Commitment Contracts III
Findings:
• Workers choose positive targets 35 percent of the time, with targets being set at
a non trivial level.
• Simply being offered the choice of a commitment contract increases production
and wages by 2.3 percent on average relative to the control contract.

2025 Dr. Si Chen Slide 44

Self-Imposed Deadlines I
(Ariely and Wertenbroch, 2002)
MBA class at MIT: 99 Students Part of the course requirements: three term
papers & final project.
“Commitment device”: deadlines imposed on the term papers; 1% of the final
grade for each day of delay.
Treatments:
• Free Choice: Students can freely choose if they want to impose a deadline.
• No Choice: Exogenous, evenly spaced deadline impose.

What would you do if you are in the “Free Choice” treatment?

If there are enough sophisticates in the sample, what do we expect to see in
“Free Choice”?
If there are enough naifs in the sample, what do we expect to see in “No
Choice”?

2025 Dr. Si Chen Slide 45

Self-Imposed Deadlines II
(Ariely and Wertenbroch, 2002)
Implications from the toy model with present bias:
• Only sophisticates will commit.
• TCs don’t need commitment
• Naifs think they don’t need commitment.

Empirical strategy:
• Self-imposed commitment is evidence of sophistication.
• Project completion of naifs can be enhanced by deadlines. If exogenously
imposed deadlines improve completion, this is consistent with naiveté.

2025 Dr. Si Chen Slide 46

Self-Imposed Deadlines III
(Ariely and Wertenbroch, 2002)
Distribution of Deadlines:
Only 27% of all students choose the final-week deadline.
Typical students choose deadline two or three weeks before final lecture.
—evidence of self–control problem

Performance in the Treatments:

Free Choice: Average grade 85.56
No Choice: Average grade 88.76
t-test, p < 0.01.
—evidence of naiveté.

2025 Dr. Si Chen Slide 47

Transaction Costs
• Reducing the transaction costs, e.g., nudging: default setting
For example, setting a default of non-enrollment in a 401(k) plan leads to a participation rate of
approximately 40% after one year of tenure as an employee. By contrast, setting a default of
enrollment (and letting employees opt out if they prefer), leads to a participation rate of
approximately 90% after one year of tenure. Results like this have been reproduced in dozens of
studies (for a meta-analysis see ?).

• Other ways: e.g., limit the choice set, etc.

2025 Dr. Si Chen Slide 48

Summary

1. Discounted Utility Models

and Exponential Discounted Utility Model

2. Quasi-Linear Discounted Utility Model

3. Applications of Present Bias

Self-Control Problem
Sophisticates vs. Naifs
Commitment Devices

2025 Dr. Si Chen Slide 49

References I
Dan Ariely and Klaus Wertenbroch. Procrastination, deadlines, and
performance: Self-control by precommitment. Psychological science, 13(3):
219–224, 2002.
Supreet Kaur, Michael Kremer, and Sendhil Mullainathan. Self-control and the
development of work arrangements. American Economic Review, 100(2):
624–28, 2010.

2025 Dr. Si Chen Slide 50