Exponential Discounting Hyperbolic Discounting Tutorial

23005 Behavioural Economics

(Spring 2020)

Lecture IV

Andrea Giovannetti

[Link]@[Link]

Office Hour: Fri (10:30-11:30)

Zoom: 8712136666

University of Technology in Sydney

August 19, 2020

1 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Battleplan:

1. The Standard Exponential Discounting Theory

2. Hyperbolic Discounting I1

3. Tutorial with Luz: Hyperbolic Discounting in Practice

References: Cartwright textbook ”Behavioral Economics”, Chapter 5; Dhami, ”Foundations of Behavioural Economic

Analysis”, Chapter 9, 10 (technical). Angner textbook, pp. 158-161 + examples.

2 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

The psychology of decision making entails three trade-offs:

Returns ! Risks

Today ! Tomorrow

Self ! Others

In previous classes we discussed paradoxes of EUT and introduced a generalized

approach which better accommodates the psychological mechanisms involved in the

first dimension

We now move to time.

In survey we discussed preferences over time and found paradoxical behaviors:

Q.4 - how many times per years should (vs does visit museums in 2018? [2.14 vs 1.14]

Q.11 - Suppose Willy offers you either $100 in cash right now (vs in 52 weeks) or $x in 2 weeks (vs. 54

weeks). What is the x for which you would be indifferent? [$280 vs $123]

We referred to the anomaly stemming out from Q.4 as time inconsistency and to the one

from Q.11 as impatience

Three questions:

1. How does the standard theory models time preferences?

2. How do we relate the observed behaviour to standard theory? Anomalies?

3. How can we narrow the shortcomings of the standard model?

3 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Let’s get back to the cake problem we introduced in L1

Max buys a cake on Sunday (t = 0). Max wants to consume it along the week (Sunday included). He wants to

slice the cake x0 , x1 , ..., x6 in order to experience the greatest prospected enjoyment of the cake.

(For simplicity this time we assume the cake is crammed with preservatives and it doesn’t expires... so no risk

involved ! no probabilities)

Any insightful math representation of the above story should comply with the following

generic structure:

U today stemming from the idea of consuming the cake during the week =

= U of the idea consuming a share x of cake now

+ U of the idea consuming the remainders later, given sacrifice of postponement

We first look at how the standard theory attempts to capture the story, its shortcomings

and then we introduce an insightful deviation

4 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

(Standard) Exponential Discounted Utility (D.U.) Model: Assumptions

2 6
DU0 (x0 , x1 , ..., x6 ) = U0 (x0 ) + · U1 (x1 ) + · U2 (x2 ) + ... + · U6 (x6 )

(We call the above D.U. instead of E.U. because there are no random variables involved)

1. U0 is the agent’s utility today. Uk , ... is the instantaneous utility at time k

2. At time 0 consumption levels x0 , ...x6 are planned. Slices are as 7 different goods.

At time 0, the agent seeks to maximize the utility generated at t = 0 of the idea of

consuming the cake along the whole week.

3. A.1: Time-consistency: once the agent starts moving along the consumption orbit, he

is time-consistent with her own initial plan.

4. A.2: Consumption independence: mathematically this assumption allows us to write x

as:

x = x1 + x2 + ... + x6

That is, good x can be split and allocated across periods

5. A.3: Stationary preferences: Ut = Ut+k .

In this lecture we challenge assumptions (A.1)

In L6 we will challenge assumptions (A.2), (A.3)

5 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

: parameter capturing i’s sacrifice in delaying consumption beyond time t:

1 ! 1 2 3
... 6
exponential decay

From the perspective of time 0, both present consumption and the idea of future

consumption is enjoyable, yet postponed consumption yields lower pleasure at t = 0.

The smaller the , the less we enjoy today the idea of postponed consumption

Everything equal, pleasure decays at exponential rate as we move away from t.

And:

Everything equal, suppose preferences are stationary: if we increment both current

t = 0 and future t = 1 consumption by y units, the utility increases at rate ( U10 )/U00 =

< 1, that is everything equal, consumption today brings more utility. (more on this

in slide 10! )

Why is something that happens in the future worth (but less) than something that

happens in the present?

Impatience: Consumption today is better than consumption tomorrow

Uncertainty about the future: I could get hit by a truck on my way home

However, future may bring additional pleasure today: I may have memory Utility: In

addition to consumption, I might also derive utility from memories or expectations.

6 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Example: A reasonable application

Today (t = 0), Netflix is advertising a special rate of $5.99. The offer will expire in

a few minutes. Rupert is pondering whether to join Netflix at special rate today for

watching a long special on George Romero production which Netflix is about to offer

at time T > 0.

Assume a daily = 0.7. Assume instantaneous utility: Ut (watch special at t) = $40.

I If the special is offered tomorrow: $5.99 + · $40 > 0

! Rupert will subscribe.

I If the special is offered the day after tomorrow, $5.99 + 2 · $40 > 0.

! Rupert Rupert will subscribe.

I ...

I If the special is offered six days from now, $5.99 + 6 · $40 < 0.

! Rupert will not subscribe.

The above representation looks ”reasonable”:

1. If preferences are stationary (i.e. Ut = Ut+k ), and

2. If ”impatience” has no reason to alter the consumption orbit as we move from

t = 0 onwards2 .

... then the exponential discount model produces credible results (i.e. in line with our

own feelings and with experimental results)

2
If you think about it, at t = 0 (i.e. when Rupert decides whether to subscribe) D.U. is at the lowest possible value: D.U.

can only increase when he faces the same problem in following periods. Hence, if he decides to subscribe at t, he will be
happy with his own decision. But what if he decides not to subscribe at t = 0? Will he regret? 7 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

But there are actually a lot of assumptions in place here

1. Time consistency: optimal decisions taken at one point in time remain optimal with the

passage of time (66= Q.4

Q.4, 6= Q.11 in survey).

Example: Suppose i must choose between $100 in 5 years or $125 in 6 years. He will be willing to

wait for the larger amount if: 5 U (100) < 6 U (125) which reduces to U (100) < U (125). Now

suppose 5 years have passed and he can choose again. The choice is now between $100 right now or $125

in 1 year. He will be willing to wait for the larger amount if: U (100) < U (125) which is the same condition

as above.

I Consistency implies that there’s no need to check how the agent make decisions

along the time because the problem solved at t = 0 is replicated across periods

2. Integration: An individual evaluates new alternatives by integrating them with all existing

consumption plans

I If i finds out he will receive a lump money prize at t + 10, he will change his consumption plans for all

periods t, t + 1, ...t + 10, .... This implies that a prospect xt+k is not evaluated in isolation but against

how it will change consumption across the whole life-span!

8 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Further assumptions made in Discounted Utility models3

3. Utility Independence: All that matters is maximizing the sum of discounted utilities.

Decision makers are assumed to have no preference for the distribution of utilities.

4. Consumption Independence: Utility in period t + k is independent of consumption in any

other period. An outcome’s utility is unaffected by outcomes in prior or future periods

I Imagine a world with one consumption good: x = skiing classes. An ED maximizer wants to consume a ski

class (beginner level) in t + 4 and a ski class (pro level) at t + 5. If he does not attend the beginners class at

t + 4, at the end of t + 4 in DU model he still expects to enjoy the advanced class consistently with the plan

as decided at t!

5. Stationary Inter-temporal Choice: It is assumed the utility function is stationary across

periods, that is: functional form of Ut = functional form of Ut+1 , Ut+2 etc.

I But we know tastes often do change over time in predictable ways, or in particular

situations (emotions)
I For the consumption of certain goods time can alter the utility function

! We relax (4) and (5) in L6

This is the parent-class of models which contain ED models: the above assumptions are required to break down the

inter-temporal optimization in a series of summable instantaneous utilities!

9 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Verbose Appendix: Machinery of Inter-temporal substitution in ED models

We said that the standard procedure to optimize inter-temporal problems across T periods is to interpret the good C
PT

as the sum of T different goods such that: t=0 ct = C

Therefore, a standard (static) inter-temporal optimization problem solved at t = 0 can be written such as:

maxx0 ,x1 ,...,xT DU0 (x0 , x1 , ..., xT ) = U0 (x0 ) + U1 (x1 ) +

2
U2 (x2 ) + ... +
T
UT (xT )

That is to say we can perfectly treat consumption of C in each period t = 0, 1, ... as consumption of a specific good

ct which enters exclusively in Ut .

Therefore, the computation of the cost of substitution (across periods) of a portion x of a good C across two periods

echoes the definition of marginal rate of substitution (between goods).

Suppose x ! 0, so that we can study variations by means of derivative calculus

Suppose I want to find the relative cost of eating x in day t = 4 instead of day t = 3. First, compute the benefit of

consumption of c3 :

@DU0 3
· @U3
=

@x3 @x3

Now we compute the benefits of increasing c4 by the same amount x:

@DU0
=+
· @U4

@x4 @x4

Eventually, we derive the relative cost of the postponement by taking the ratio:

! !
3 4

@DU0 @DU0
/ =
· @U3
/
· @U4

@x3 @x4 @x3 @x4

Under stationarity, U3 = U4 (same functional form), hence the derivatives have same magnitude and cancel out.

We end up with:

@DU0 @DU0
/ = 1/ > 1

@x3 @x4
10 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

How does ED machinery work in practice?

Translation: Economists see how people make choices and calibrate in order to have

predicted behaviour as close as possible to observed choices

”Calibrations, like sausages, cease to inspire respect in proportion as we know how they

are made” (attributed to [Link])

Issues signaled by many experimental / empirical studies:

1. In every moment of time, we tend to formulate our plans according to a simple

binary perspective: (”now” VERSUS ”future”).

... ! Which is different from (”now” AND ⇥”now” AND 2 ⇥”now”....)

2. We may be very good in making plans with respect to the ”future” component of

our problem, yet because of (1) we continuously postpone ”future”

generating a long-term inconsistency with our initial plans.

11 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Estimation of

Frederick, Loewenstein, and O’Donoghue (2002) plotted estimations of from published

papers from 1980 2000 in one single graph.

What regularity does emerge from this picture?

12 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Estimation of

Frederick, Loewenstein, and O’Donoghue (2002) plotted estimations of from published

papers from 1980 2000 in one single graph.

What regularity does emerge from this picture?

That there is no regularity at all. No consensus in the literature on what value or range

should pick. 12 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Field estimations of vary of several orders of magnitude

The idea here is that exponential discounting tries to fit in one single parameter at least

two diverse psychological attitudes toward time-decisions.

I Short-term Impatience (in ED theory: ”now” is better than ”later”)

Examples: I prefer $2 today vs $4 tomorrow. I am happy to pay high interest

rates for easily accessible credit card finance against low interest rate mortgages.

Or worst, I use pawn shops instead of bank finance.

I Long-term Patience (in ED theory: ”later” still brings positive utility )

Examples: I do under-performing long-term investments. I commit to expensive

close funds instead of buying cheap ETFs. I mis-price the life-value of my tuition

fees. I overpay my real estate investments.

Problem!! Under the lenses of Exponential Discounting, people are generally very

impatient in the short-term and inconsistently too patient in the long run.

Q.11 from Survey:

Suppose Willy could give you either $100 in cash right now Suppose Willy could give you either $100 in cash in 52
or $x in cash in 2 weeks. What is the x for which you weeks or $x in cash in 54 weeks. What is the x for

would be indifferent between the two options? which you would be indifferent between the two options?

13 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Empirical Evidence (Issue 1 - ”now” vs ”future”)

Thaler (1981) proposed a mental experiment designed to isolate the above tensions.

He awarded participants with a monetary prize. He then announced the prize delivery

was delayed. He asked students’ minimum compensation value for the delayed delivery

He manipulated the delay in collection time and amount of prize in order to compute the

empirical discount factors (standardizing on yearly basis)

14 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Issue 2 - Consistency

what Rupert plans at t what Rupert does at t + k

ExpDisc ”I plan the cake consumption for the upcoming week. ”No reason to deviate from the plan computed at t

I understand I am eager to eat sooner than later because the problem I see now w.r. to the next k periods

and my eagerness is constant as I move forward” is a replica of the one I’ve seen and solved at t ”

[Link] ”If I shift x from today to tomorrow I will feel” ”If I shift x from today to tomorrow I feel
”worst than if shifting x from t + k to t + k + 1” worst than shifting x from t + k + 1 to t + k + 2”

In real life Rupert tends to revise his plans with respect to the near future. This process

of continuous revision generates an inconsistency with our initial plan

In our survey Q.4: the gap between ”now” and ”future” is given by the visit to museums

we ”do” against those we ”prescribe”

Idea: Discounting is a sensible process. However, the way our brain does it is

fundamentally different from the E.D.

I In E.D. theory: Rupert is impatient ! << 1 ! the idea of future consumption

bears low utility now! Rupert consumes more now ! when ”future” comes

Rupert complies with plans

Let us modify our workhorse model in order to account for the above...
15 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Hyperbolic4 Discounting.

4
Fancy for ”curved”. We actually study quasi-hyperbolic functions (Laibson, David (1997). ”Golden Eggs and Hyperbolic

Discounting”. Quarterly Journal of Economics.), a simplified version of hyperbolic discounting which retains the spirit of the

more general formulation and is as tractable as the exponential discounting.

16 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Hyperbolic Discounting

Back to the cake example: we change our interpretation of how we look at future

consumption

For simplicity, let assume again preferences are stationary: Ut = Ut+1 = ... = U

Again, I want to plan at t = 0 the consumption of the cake along the entire week

Hence, I need to set shares x1 , ..., x6 to maximize the enjoyment I get at t = 0 about the

idea of consuming the cake along the entire week.

The Hyperbolic Discounting model reads:

2 6

DU0 (x0 , x1 , ..., x6 ) = U (x0 ) + · U (x1 ) + · U (x2 ) + ... + · U (x6 )

 1 is the short-term discount factor: it scales down all future dates utility.

is the usual discount factor.

17 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Mathematically simple, yet it bears philosophically and pragmatically profound

implications:

1. ”now” vs ”future” is stressed-out: utility from future period consumption can be

heavily depressed w.r. to utility from current consumption.

I If we increment both today versus tomorrow consumption by y units, the

utility increases at rate:

1U 0

0U 0

I However, when we compare day after tomorrow vs tomorrow:

t+2 U 0

t+1 U 0

2. Planning within future dates is unaffected: as in DU, if we move y units of

consumption from t + k to t + k + 1 the utility decreases at rate as before.

(1) + (2) pave the way to inconsistent inter-temporal decision-making

18 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

From period t = 0, looking forward to get a prize

Assume U ($x) = $x like Thaler (1981). A $1000 is offered ”at some point in time”.

We plot: U ($x) = $x, U ($x) = $x, 2 U ($x) = 2 $x, ...

and: U ($x) = $x, U ($x) = $x, 2 U ($x) = 2 $x, ...

The HD formulation matches Thaler’s findings (i.e. strong dislike for near in time

postponement, almost no dislike for postponement from future to further in future dates).
19 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

Moving across periods

Behavioral inconsistency caused by HD raises a fundamental question:

! Is the agent aware of the inconsistency of decision-making process?

To be more precise:

Assume = 1, > 0 and t = 0, 1, 2.

Under HD, at t = 0 the agent discounts both t = 1 and t = 2 relative to the present, but

she does not discount t = 2 relative to t = 1:

DU0 = U0 + (U1 + U2 )

When the agent moves to t = 1, she begins discounting t = 2 relative to t = 1:

DU1 = U1 + U2

at t = 0, does the agent foresee the above inconsistency?

The agent’s awareness (or lack thereof) determines the agent’s way of solving the

inter-temporal optimization problem

We will study two alternative psychological approaches to inconsistency of decision-

making

20 / 35


Exponential Discounting Hyperbolic Discounting Tutorial

The Naive solution approach

Daniel wakes up at t = 0 persuaded of being a great street artist. Unfortunately he gets

caught and fined after his first wall tag.

Daniel can then select one of the following payment schemes (payment date, fine):

(0, $1), (1, $5) or (2, $20)

Daniel’s utility is linear in money U ($x) = x

Unfortunately our Australian Banksy is extremely time-inconsistent: = 1, = 0.1

The psychology of the Naive Agent: meet Daniel

We will assume Daniel does not understand his own behaviour is time-inconsistent

We say that Daniel is a naive agent.

For every t: (1) He recomputes his optimal plan according to the

”now” vs ”future” perspective.

(2) he believes he will stick to the updated plan in (1) once he

moves to next period.

21 / 35


Naive agent: he will compute all available decision orbits

Period after period and every period he reoptimizes the problem

Hence, in period o he computes the black curve and finds that paying the fine at t=1 is optimal

Then he moves in period 1 (red curve) and finds that postponing is still better

Hence he will end in period 2 (that is a point, no choices available other than paying)

The sophisticated knows she is inconsistent. Therefore when she is in period 0 she knows that if she
moves to period 1 she will postpone again and end at period 2. Therefore, she removes option NPN
from the available decision set and only consider:

PNN and NNP

Exponential Discounting Hyperbolic Discounting Tutorial

At t = 0 Daniel thinks: ”If I pay at t = 0, my disutility is $1”

”If I pay at t = 1, my disutility is · $5 = $0.5”
”If I pay at t = 2 my disutility is · $20 = $2”

Hence, at t = 0 Daniel decides to pay the fine at t = 1 (he delays payment of one period)

At t = 1 Daniel thinks: ”If I pay at t = 1, my disutility is $5”

”If I pay at t = 2, my disutility is · $20 = $2”

Again, at t = 1 Daniel will postpone the fee payment to t = 2 and ends up paying $20.

22 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

Alternative solution method: Sophistication

Let us assume that it is a friend of Daniel, call her Chi, who gets fined.

The psychology of the Sophisticated Agent: meet Chi

She is exactly as Daniel apart from one thing: in every period, we assume she is
aware (2) does not hold.
In other words, she understand that if faced with the temptation of procrastinating,
she will definitely do so!
We say that Chi is a sophisticated agent

Chi knows that if she were to follow her impulse, she would delay once.
She also knows that if she delays once, she will delay once more and end up paying the
largest fine.
Being aware of this, she knows that she either pays at t = 0 or pays at t = 2.
As: undiscounted cost of paying now ($1) < discounted cost of paying at t = 2 ($2)
! Chi decides to pay the fine straight away.
23 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

Question: Where does Chi and Daniel solution strategy differ?

Hyperbolic Discounting - Solution book

Naive Agent: (forward reasoning)
1. Start from date 0
2. Solve for the optimal action given all the possible subsequent histories
3. Move to next step
4. Iterate

Sophisticated Agent: (backward reasoning)

1. Start from last date
2. Look for the optimal action
3. Go back 1 step and solve for optimal action accounting for the payoff obtained in (2)
4. Iterate
24 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

Question: Where does Chi and Daniel solution strategy differ?

Because of (2), Daniel trusts his commitment to future plans. A forward-looking solution notion is adopted. Instead,
Chi uses backward-induction to predict her upcoming deviations

Hyperbolic Discounting - Solution book

Naive Agent: (forward reasoning)
1. Start from date 0
2. Solve for the optimal action given all the possible subsequent histories
3. Move to next step
4. Iterate

Sophisticated Agent: (backward reasoning)

1. Start from last date
2. Look for the optimal action
3. Go back 1 step and solve for optimal action accounting for the payoff obtained in (2)
4. Iterate
24 / 35
 Exponential Discounting Hyperbolic Discounting Tutorial

Important: Sophisticated doesn’t mean Optimal!!!

In the previous example agents selected the optimal action in order to minimize disutility
(i.e. U (f ine) = U ( $x) < 0) and we showed that the sophisticated player achieved a
better result than the naive one.
Consider instead the case of maximization of positive utility. Assume Chi and Daniel buy
a frozen pizza each such that the utility of consumption is positive and increasing in the
time the pizza is heated. For both agents:
U(eat after 10 min) = 1, U(eat after 20 min) = 5, U(eat after 30 min) = 20
We assume again = 1 and = 0.1.
Naive Daniel Sophisticated Chi

at t=10min: DU(30m) = 2 > U(10m) = 1 > DU(20m) = 0.5 (eat at 30m) at t=20m: DU(30m) = 2 < U(20m) = 5 (eat at 20m)
at t=20min: U(20m) = 5 > DU(30m) = 2 (eat at 20m) at t=10m: U(10m) = 1 > DU(20m) = 0.5 (eat at 10m)
Net gain: 5 Net gain: 1

Daniel eats the pizza when 20m have passed. Although this is inferior w.r. to waiting
30m, it is still better than what Chi does:
Chi knows that if she fasts for 20m, she will not resist and eat the pizza. Therefore, when
she hits 10m, because she knows that if she does not eat the pizza when 10m are
passed she will certainty eat it after 20m, she decides to eat the pizza after 10m.
Sophistication in this case yields lower utility: Chi is pessimistic on her impatience
General rule: Sophisticated agents take action no later than naive agents.
25 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

The mechanism
Suboptimality of sophisticated behaviour with utility from action arises from two tensions:
(1) Chi understands she will not be able to wait for the best time to eat the pizza
(2) Furthermore, she craves due to = 0.1.
Then (1) and (2) imply she will take action at an earlier stage than the naive player

On the other hand, in the case of disutility (e.g. the fines), (1) + (2) allow Chi to reduce
the costs of procrastination
Rule:
If deviation in future time makes more expensive to deviate today, then sophistication
avoids or reduces deviation along the whole sequence
If deviation in future increases the benefit from acting today, then sophistication
increases the likelihood of deviation by depressing the value of waiting

26 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

Main references cited in this lecture:

Thaler, Richard. ”Some empirical evidence on dynamic inconsistency.” Economics
letters 8.3 (1981): 201-207.
Malhotra, Deepak, George Loewenstein, and Ted O’donoghue. ”Time discounting and
time preference: A critical review.” Journal of Economic Literature 40.2 (2002): 351-401.

27 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

Solving Hyperbolic Discounting Problems

28 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

Hyperbolic Discounting - Solution strategy

Sophisticated Agent: (backward reasoning)
1. Start from last date - 1
2. Look for the optimal action
3. Go back 1 step and solve for optimal action accounting for the payoff obtained in (2)
4. Iterate

Naive Agent: (forward reasoning)

1. Start from date 0
2. Solve for the optimal action given all the possible subsequent histories
3. Move to next step
4. Iterate

29 / 35
Exponential Discounting Hyperbolic Discounting Tutorial
Example - Carriers Courageous
In ancient Greece, hemerodromes were special armored units of highly trained soldiers
capable of running sick distances (>18 hours / day!)
Philippides from Athens is the most famous hemerodrome in history. He derived his
fame for accomplishing two missions:
1. He was sent from Marathon to Sparta (⇡ Sydney - Canberra) to request military
support when Persians disembarked in Marathon (run 240km / 2 days)
2. He brought the news of Greek victory against Persian empire from Marathon back
to Athens (run 40km / few hours)
3. He died just after accomplishing (2) as a result of extreme exhaustion.

30 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

E.1: Marathon-Sparta
Suppose Phil splits the trail in three stages t = 1, 2, 3 with a final stage t = 4 in which he
does not run but incurs in the delayed health costs of running in previous periods
In t = 1 Phil decides either to run or walk. In periods t = (2, 3) Phil decides either to
continue running (if he runs in the previous period), or slow down and walk, or simply
continue to walk if he never start running. Each time he decides to run he will pay c in
round 4.
Therefore, the following actions are admissible: W W W , RW W , RRW and RRR.

Phil’s payoffs: b1 = 5, b2 = 5/2, b3 = 5/2.

Preferences: = 1/2, = 1
marginal cost c = 4.

Example: If Phil runs on t = 1 he shortens time with immediate benefit b1 = 5, incurring

in cost c = 4 at period 4. If he runs both t = 1, 2 he gets a benefit of b1 = 5 at t = 1, a
benefit of b2 = 2.5 at t = 2 and a final health cost of 2c = 8 once he stops in Sparta.

31 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

We assume Phil is a cold-blood sophisticated hemerodrome.

Q.1 How many rounds will the sophisticated Phil run?

32 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

We assume Phil is a cold-blood sophisticated hemerodrome.

Q.1 How many rounds will the sophisticated Phil run?

Suggestion: write down the table of marginal benefits of running for 1, (1,2) and (1,2,3)
periods
t= 1 2 3
c 4 4 4
bt 5 5/2 5/2
t=1 3 3/4 3/4
t=2 - 1/2 3/4
t=3 - - 1/2

32 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

E1: Solution
Because Phil is sophisticated, at t = 0 he will solve the problem backward from stage 3.

He anticipates that no matter his utility when he will enter stage 3, his period 3 self believes that running a further
stage will bring him a net utility of 1/2 (= immediate benefit of b3 = 2.5 - discounted cost of · 4 = 2). Therefore,
his period-3 self will run on t = 3.

By backward reasoning, he anticipates that his period-2 self will foresee his t = 3 self running. From the perspective
of period 2 self, running at period 3 brings a loss of 3/4. Therefore, his period 2 self elaborates the following
prospect:
D.U (running at t = 2) = U (running at t = 2) + · U (running at t = 3)

= (5/2 + · ( 4)) + · (5/2 4) = 1/2 3/4 = 1/4

and compares it with the only alternative prospect available to his period t = 2 self:

D.U (non running at t = 2) = U (not running at t = 2) + · U (not running at t = 3) = 0 + 0

Because his 2 period self knows that running at t = 2 will also imply running at t = 3 and that this prospect is inferior
to not running at all, his period 2 self does not run in period 2.

The period 1 self anticipates that the period 2 self will not run. Therefore, he compares the following prospect:

D.U (running at t = 1) = U (running at t = 1) + 0 + 0 = 5 4=3

Against:
D.U (not running at t = 1) = U (not running at t = 1) + 0 + 0 = 0

Therefore, the sophisticated Phil will run only one period!

33 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

E.2: Marathon-Athens

Athens defeats the Persian empire thanks to a smart manoeuvre by Miltiades. Phil is in
charge of bringing the marry news back to Athens.
Again, we assume the journey is split in three stages. We also retain the geography and
preferences as set in the previous problem (i.e.
c = 4, b1 = 5, b2 = 5/2, b3 = 5/2, = 1/2, = 1)
However, this time Phil is euphoric and he will run like a naive soldier.

Q.2 How many rounds will the naive Phil run?

Notice that the payoff structure is not altered by player’s type.

34 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

E.2: Marathon-Athens

Athens defeats the Persian empire thanks to a smart manoeuvre by Miltiades. Phil is in
charge of bringing the marry news back to Athens.
Again, we assume the journey is split in three stages. We also retain the geography and
preferences as set in the previous problem (i.e.
c = 4, b1 = 5, b2 = 5/2, b3 = 5/2, = 1/2, = 1)
However, this time Phil is euphoric and he will run like a naive soldier.

Q.2 How many rounds will the naive Phil run?

Again, we write down the table of marginal benefits of running for 1, (1,2) and (1,2,3)
t= 1 2 3
c 4 4 4
bt 5 5/2 5/2
periods:
t=1 3 3/4 3/4
t=2 - 1/2 3/4
t=3 - - 1/2

Notice that the payoff structure is not altered by player’s type.

34 / 35
Exponential Discounting Hyperbolic Discounting Tutorial

E.2: Solution

Because Phil is naive, he will solve the problem from t = 1 onwards.

At period 1 he compares the prospect of running at t = 1 (D.U = 5 · 4) against not running (D.U. = 0). So he
decides to run only on the first round.

At period 2 he compares the incremental benefit of running at t = 2 (D.U = 2.5 · 4) against not running (D.U.
=0). So he decides to run also the second round and then walk.

At period 3 he compares the incremental benefit of running at t = 3 (D.U = 2.5 · 4) against not running (D.U.
=0). So he decides to run also the third round and enters in Athens celebrated like a god. Unfortunately, the
3 · c = 12 costs suddenly realize during the gala dinner (i.e. at t = 4). This cause unbearable fatigue to our hero.

And this is how, according to the historical consensus, Philippides from Athens passed from human life to myth.

35 / 35