CHAPTER • 8

Alternative Intertemporal Choice Models

8.1 Time preference 285
Consumption reallocation 285
Intertemporal arbitrage 286
Concave utility 286
Uncertainty 287
Inflation 287
Expectations of changes in utility 288
Anticipatory utility 288
Visceral influences 288
What is time preference? 289

8.2 Time inconsistent preferences 291

Nature 291
Reverse time-inconsistency 292
Issues and approaches 293

8.3 Hyperbolic discounting 293

Nature of hyperbolic discounting 293
Implications of hyperbolic discounting 295
Advantages of hyperbolic discounting 299
Criticisms of the hyperbolic discounting approach 300

8.4 Modifying the instantaneous utility function 301

Habit-formation models 302
Prospect theory models 302
Anticipatory utility models 304
Visceral influence models 304

8.5 More radical models 305

Projection bias models 305
Mental accounting models 306
Multiple-self models 306
Dual-self models 307
The procedural approach 309

283
 8.6 Empirical evidence 310
Behavioral studies 311
Evolutionary biology 313
Neuroeconomics 314

8.7 Policy implications 315

Individuals 315
Firms 320
Government 321

8.8 Summary 324

8.9 Review questions 325
8.10 Applications 326
Case 8.1 Price plans for gym memberships 326
Case 8.2 The savings problem 327
Case 8.3 The desire for rising consumption profiles 332

284
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

Why would we place the alarm clock on the other side of the room from the bed? Well,
we don’t need to do that any more. Now we can buy Clocky, the alarm clock that jumps
off the bedside table and runs away when we try to snooze it, sounding its alarm. A
comment on YouTube said that it would be even better if the clock could be designed
to run and hide under the bed, to make it even harder to get at.
So the question now becomes: why would we buy something that makes life
harder for us, and that causes us to curse in the morning when we can’t turn it off?
Clocky is all about commitment. When we go to bed at night we want to make sure we
wake up on time in the morning. But when we wake up we suffer from a preference
reversal, and we want to stay in bed. Our morning self is a different person from our
evening self, and this pattern is repeated day after day. The result is that our evening
selves learn about the weaknesses of our morning selves and try to overcome these
by making a commitment that our morning self is unable to avoid. This pattern of
preference reversal over time is a very common one in various facets of our lives; we
are always yielding to temptation and procrastinating, and at the same time trying to
take measures to prevent our future selves from doing so.

8.1 Time preference

The most obvious characteristic of Table 7.1 is the very wide disparity between different
measures of the discount rate, even within the same study as well as between studies.
Thus it is not simply differences in methodology in terms of experimental design that
account for these variations. The primary reason for the variability is the existence of
confounding factors in the measurement of time preference. This raises the fundamental
issue of what constitutes time preference. It is necessary to understand this concept and
the factors involved if we are to address and explain the anomalies related to the DUM.
We shall see in the second section of the chapter that these anomalies are frequently
related to self-control problems. The third, fourth and fifth sections of the chapter then
discuss various alternative models to the DUM, while the sixth section examines the
relevant empirical evidence from behavioral, evolutionary and neuroeconomic studies.
The seventh section concludes with the discussion of various policy implications related
to the models and the evidence.
First, we need to examine the various confounding factors involved in the
measurement of time preference.

Consumption reallocation

Most studies use monetary rewards as payoffs rather than consumption. When discount
rates are calculated it is normally assumed that rewards and losses are consumed
immediately at the same point in time that they are received, and that they do not
affect the pattern of consumption at other time periods. For example, if a reward of
$100 is to be received in one year, it is assumed that this amount will be consumed
immediately rather than causing a stream of higher consumption over a prolonged time
period after one year. Furthermore, it is assumed that this reward is not anticipated,

285
PT • I I I I NT ERT EMPO R A L C H O I C E

in terms of increasing consumption in time periods before it is received. It is obvious

from empirical observation that both aspects are unrealistic. Ideally the calculation of
discount rates should take into account the effects of rewards and losses on the whole
lifetime pattern of consumption.

Intertemporal arbitrage

When rewards are tradable, like money, intertemporal choices may not reflect time
preference directly, but may be caused by intertemporal arbitrage. For example, if a
person prefers $100 now to $150 in five years time, this may be because they can invest
$100 now at the market rate of interest and make it worth more than $150 in five years.
When financial markets are efficient it can be argued that discount rates will converge
on the market rate of interest, rather than being a direct reflection of time preference.
Of course market interest rates are affected by time preference, but they are also
influenced by many other factors, such as default risk, uncertainty, liquidity and so on.
However, there is a heavy weight of empirical evidence that financial markets
cannot explain intertemporal choices, since discount rates generally are much higher
than market interest rates. For example, people with substantial savings earning 4% a
year interest should not prefer $100 today over $120 in one year if financial markets
are efficient, yet many do. Such choices and preferences imply either that people are
ignorant about the operations of the markets, or that they are unable to use the markets
properly for some reason. It appears therefore that discount rates do take into account
time preference, and other factors, in ways that the market interest rate does not.

Concave utility

As can be seen from the table in the case study at the end of the last chapter, the
majority of empirical studies involve monetary rewards, and base the calculation of
discount rates on the monetary amounts. As was mentioned briefly in the last chapter
it is misleading to calculate discount rates on this basis, since it is implicitly assumed
that utility increases linearly with monetary amounts. This assumption is in direct
contradiction to both the standard model and the principles of prospect theory. An
example will illustrate the effects of relaxing this assumption and instead incorporating
a concave utility function, as proposed by prospect theory. Say that the average
response of a group of subjects is that they are indifferent between $100 now and $150
in five years. The imputed discount rate (assuming no consumption reallocation) is
8.1% a year on the basis of the monetary amounts. However it may be that the $150
has only 30% more utility than $100 (this is also ignoring the effect of inflation, which is
considered later). Using utility as the basis for discounting, the imputed discount rate is
only 5.2%. This shows that utility discount rates are lower than monetary discount rates
when utility functions are concave.
In terms of empirical findings, Chapman (1996) attempted to allow for the
concavity of the utility function by estimating a utility function from the monetary
amounts, and found that the discount rates calculated from the utility function were
indeed substantially lower than the monetary discount rates. However, it is difficult
empirically to conduct a reliable study to estimate utility discount rates, bearing in mind
the problem of consumption reallocation discussed earlier. For example, although it
may be possible to estimate that $150 has only 30% more utility than $100, what is

286
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

relevant are the streams of utility that flow from the reward of $150. A consumer may
respond by spending an extra $30 a year over the next five years, and a person’s utility
function may not be concave over such small amounts.

Uncertainty

Future rewards and costs are almost invariably associated with uncertainty in practice.
Thus in field studies it is particularly difficult to avoid this confound, regardless of
whether rewards and costs are expressed in monetary terms or in other ways. For
example, even if we can be sure that a particular electrical appliance will save us a
certain amount of electricity in the future (which is unlikely in itself), we cannot be sure
what will be the future price of electricity.
In experimental studies it might appear that investigators could avoid this confound
by assuring subjects that delayed rewards will be delivered with certainty. This is indeed
the common practice in such studies, but whether respondents can accept this situation
from a subjective point of view is questionable. It may be that there is an unconscious
psychological mechanism that automatically relates delay to uncertainty. One reason
for such a mechanism is that, even if rewards are certain in monetary or other terms,
our valuation of these may change over time, since utilities change as tastes change
in ways that are not entirely predictable. For example, in the future we come to value
money less and health more than we do today. Therefore, there will always be an
element of uncertainty relating to future tastes and utilities.
Furthermore, some experimental studies have compounded this problem by
introducing ambiguity into the situation. For example, a study by Benzion, Rapoport
and Yagil (1989) asked respondents to imagine that they had earned money, but
when they arrived to receive payment they were told that the ‘financially solid’ public
institution that had promised to pay them was ‘temporarily short of funds’. They were
then asked to specify future amounts of money that would make them indifferent to
receiving the amount of money that they had been promised immediately, with varying
amounts of delay. The methodological problem here is that there is an inconsistency
in terminology: financially solid institutions should not become temporarily short of
funds. The latter expression introduces an element of uncertainty into the subject’s
consideration.
One finding that seems beyond doubt here is that discount rates are significantly
affected by uncertainty. This is established by studies that introduce objective
uncertainty. For example, in a study by Keren and Roelofsma (1995), one group of
subjects was asked to choose between 100 florins immediately and 110 florins in one
month, while another group was asked to choose between a 50% chance of 100 florins
immediately and a 50% chance of 110 florins in one month. With the first group 82%
preferred the smaller immediate option, but, when rewards were uncertain, only 39%
of the second group preferred the smaller immediate reward. Thus a much higher
discount rate is implied when rewards are certain compared with the uncertain situation.

Inflation

Most studies ignore the effect of inflation in the calculation of discount rates, assuming
that the utility of $100 now is the same as the utility of $100 in ten years at the times
they are received. In practice people are likely to discount future monetary rewards

287
PT • I I I I NT ERT EMPO R A L C H O I C E

according to their experiences with and expectations of inflation. Furthermore, there

is another element of uncertainty here, as the effect of future inflation on purchasing
power becomes more uncertain as the duration of delay increases.

Expectations of changes in utility

It was stated above that our tastes often change in unpredictable ways. However, some
of these changes are partially predictable. Again, we have seen earlier that people
often prefer rising consumption profiles, and this preference is anticipated. The factors
underlying this phenomenon are examined in Case 8.3, along with the implications. At
this point we can observe that it has two effects on preferences and discounting, which
operate in opposite directions. The more obvious effect is that, if we expect to have
higher consumption levels in the future, the marginal utility of $100 of consumption
now is greater than the marginal utility of $100 of consumption in five years, because of
the effect of the law of diminishing marginal utility. This effect exerts an upward bias
on discount rates.
However, there is another effect at work here. People may wish to defer
consumption to later periods in order to have a rising consumption over time, but they
may lack the self-control to save sufficient income earned now to provide for this future
consumption. In such a situation people may welcome some sort of commitment device
that allows them to have more money in the future without the opportunity to spend
it earlier, in the same sort of way that they may commit to paying into a pension fund.
We have already seen something of this effect in the case of the teachers who preferred
to be paid 12 times a year rather than 10. In this situation people may prefer to receive
the money in the future rather than immediately. This effect exerts a downward bias
on discount rates.

Anticipatory utility

This is another phenomenon that has been discussed earlier, in Chapter 3. For example,
people may wish to defer consumption of a restaurant dinner, since the anticipation of
the future utility may increase total utility. The modeling of this factor is discussed in a
later section, but at this stage we can observe that the effect is to exert a downward bias
on discount rates and can also cause reverse time-inconsistency of preferences.

Visceral influences

Again this factor was discussed in Chapter 3, and is modeled in a later section. The
prospect of an immediate reward (the ‘actual presence of the immediate object of desire’
in Rae’s terms) may stimulate visceral factors that temporarily increase the attraction
of the reward. However, like uncertainty, it is difficult to unravel these influences from
time preference. It is argued by Frederick, Loewenstein and O’Donoghue (2002) that
if the visceral factors increase the attractiveness of the immediate reward without
affecting its enjoyment (decision utility rather than experienced utility), then ‘they are
probably best viewed as a legitimate determinant of time preference’ (footnote 33,
p. 383). On the other hand, if visceral factors do affect experienced utility, then ‘they
might best be regarded as a confounding factor’.

288
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

What is time preference?

The presence of the various confounding factors described above raises the fundamental
issue concerning the definition of time preference. In particular, the issue involves
whether time preference is a unitary construct. This has aroused much debate in the
psychology literature. There is a general consensus that psychological constructs or
traits must satisfy the following three criteria:
1 Constancy
Constructs tend only to be useful when they remain constant within the same
person over time. For example, many studies have shown that people’s scores on
intelligence tests change little over time.
2 Generality
Constructs or traits should be able to predict a wide range of behaviors, rather
than just a single, narrow aspect of behavior. Intelligence is again a good example
of such a trait. Impulsiveness is another example. Impulsive people make rash
decisions without much (or any) thought, like purchase decisions, which they
frequently regret afterwards.
3 Correlation between different measures
Valid constructs can be measured in different ways that correlate highly with
one another. In the case of intelligence it is difficult to devise tests that measure
a cognitive skill where test results are not correlated. Tests of personality
characteristics such as impulsiveness are similar. People who rate as impulsive on
one question are likely to rate as impulsive on other questions.
The construct of time preference does not satisfy these criteria well, unlike the so-
called ‘Big Five’ personality traits of openness, conscientiousness, extraversion,
agreeableness and neuroticism. In terms of constancy, there is some evidence that the
ability of children to delay gratification is significantly correlated with other variables
much later in life, such as academic achievement and self-esteem (Mischel, Shoda and
Peake, 1988; Shoda, Mischel and Peake, 1990; Ayduk et al., 2000). However, this only
constitutes evidence of construct validity to the extent that these other variables are
expressions of time preference, which some people would question.
On the other hand, there is a considerable body of evidence that discount rates for
different rewards and costs are at best only weakly correlated. Fuchs (1982) found no
correlation between experimental studies involving hypothetical monetary rewards and
real-world behaviors involving time preference, such as seat belt use, smoking, credit
card debt, and the frequency of exercise and dental checkups. Chapman and Elstein
(1995) and Chapman, Nelson and Hier (1999) have found only weak correlations
between discount rates for money and for health. Furthermore, Chapman and Elstein
(1995) found no correlation between discount rates for losses and rates for gains. The
main evidence of correlations involves addictive behavior: smokers tend to invest less
in human capital, having flatter, as opposed to rising, income profiles (Munasinghe
and Sicherman, 2000); heroin addicts tend to have higher discount rates for monetary
rewards (Kirby, Petry and Bickel, 1999).
It should be noted here that low correlations between different aspects of behavior
involving time preference do not necessarily provide definitive evidence that time
preference is not a unitary construct. For example, people may show a low discount
rate for monetary rewards, implying that they value future revenues highly, but show a

289
PT • I I I I NT ERT EMPO R A L C H O I C E

high discount rate for health-related factors, because they do not exercise. There are
many possible explanations why people might engage in such seemingly inconsistent
behavior: (1) they may have a strong aversion to exercise; (2) they are too busy earning
income for their future to exercise; (3) they may value monetary rewards more than
their health, maybe because they can bequeath them to their children; and (4) they do
not believe that exercise is necessary for good health.
Furthermore, one should add that high correlations would not provide definitive
evidence for a unitary construct either. As is the case with other social phenomena,
time preference and time-related behaviors may themselves be correlated with other
factors, such as intelligence and social class. These factors would have to be identified
and controlled in order for a study to provide proper evidence.
What can be concluded regarding time preference from such diverse and incomplete
findings? It is suggested by Loewenstein et al. (2001) and Frederick, Loewenstein and
O’Donoghue (2002) that a useful approach may be to revert to a pre-DUM model
and ‘unpack’ time preference into its more fundamental psychological components:
impulsivity, compulsivity and inhibition. The advantage of doing this is that each of
these characteristics can be measured reliably, and each explains and predicts different
aspects of behavior. These terms and their influences are now discussed.
Impulsivity refers to the extent to which people act in a spontaneous, unplanned
manner. Impulsive people may act without making conscious decisions, and their
behavior tends to reveal a high discount rate for many types of activity, such as credit
card use. Compulsive people tend to make plans and stick to them. Thus they may
exercise regularly, get medical check-ups regularly, always use a seatbelt, and pay bills
on time. Generally such repetitive behaviors imply low discount rates for the relevant
activities. Inhibition involves the ability to inhibit ‘knee-jerk’ responses or impulsive
behavior that may follow visceral stimuli. People who are inhibited may be criticized
for not following their instincts, for example in terms of sexual behavior, but they may
also be praised for their willpower, when it comes to refraining from eating junk food.
The ability to resist visceral influences implies a low discount rate for the relevant
behaviors, with more importance being attached to the future effects of such behaviors.
The relevance, and indeed the very meaning of the term ‘willpower’, is discussed in the
next section.
A final observation regarding time preference is that, as with aspects of behavior
discussed in other chapters, research in neuroscience is making progress in terms of
locating specific areas in the brain which influence or determine the three psychological
factors described above (Damasio, 1994; LeDoux, 1996). The fact that different brain
areas appear to be involved in affecting these factors would seem to be sufficient
justification for unpacking the concept of time preference. Further progress in the
neuroscientific field may shed more light on the interrelationships between these
psychological characteristics, and their relationships with other aspects of behavior.
This issue is discussed further in section 8.6, related to empirical evidence.
Now that the concept of time preference has been discussed in some detail, and
its component factors examined, we can turn our attention to the consideration of the
most important anomaly related to the DUM, time-inconsistent preferences, which are
a manifestation of self-control problems.

290
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

8.2 Time-inconsistent preferences

Nature

We have seen that intertemporal decision-making often relates to situations where

people are faced with choosing whether to undertake an activity that involves a cost
sooner and a benefit later. Investment goods have this characteristic, for example
investing in one’s education. Other types of decision involve a benefit sooner and a
cost later, as is the case with leisure goods, for example watching TV or eating junk
food. Frequently, the cost involves an opportunity cost, for example saving for one’s
retirement means forgoing consumption in the nearer future in order to consume more
in the further future. In many of these situations the opportunity cost relates to an
activity regarded as unpleasant in the near future, like dieting, undertaking an exercise
regime or giving up smoking. Thus the decision-maker may frame the choice as being
between gaining a benefit (the pleasure of eating junk food, relaxing on the couch or
inhaling tobacco) sooner versus gaining a different benefit (improved health) later.
Since the long-term benefits in these situations are normally considered to exceed the
short-term ones, the basic trade-off can be regarded as ‘smaller-sooner’ (SS) versus
‘larger-later’ (LL). The essence of this kind of self-control problem, which we can refer
to as ‘temptation’, is that when the SS benefit is still some way in the future people
make the decision to forgo it in favor of the LL benefit; however, as time moves on and
the proximity of the SS benefit becomes more immediate, the SS benefit also becomes
more appealing, and people ‘yield to temptation’ and reverse their previous decision.
When payoffs are negative, involving costs, the SS versus LL trade-off relates to
a different kind of self-control problem: procrastination. In this context we can think
of procrastination as meaning unanticipated delay. An example is starting a project
that involves a deadline; if it is started earlier the cost is relatively small, but if it is
delayed then the cost is larger in terms of the greater effort and stress involved. In this
case, when the SS cost is some way in the future people prefer SS to LL; however, as
time moves on and the SS becomes immediate, people tend to switch to preferring
LL and put off starting the project. Of course, this situation still involves temptation;
we may be tempted to do other more pleasurable activities than start the project. We
shall see shortly that both types of self-control problem can be described by the same
process of hyperbolic discounting. This is not to say that hyperbolic discounting is the
most appropriate explanation of these phenomena, as we shall also see. There is also a
more detailed discussion of the nature of self-control problems in the section on policy
implications.
In practice, both field studies and experimental studies have shown that such
inconsistent time preferences are common. Casual observation and personal
experience also indicate the common nature of such situations. For example, we may
decide now that at the restaurant this evening we will resist the temptation to have a
tasty dessert, because that would not be good for our future health. Yet later, when the
dessert trolley comes round (better restaurants know the power of visceral influences,
and therefore display the tempting goods under our noses), we yield to temptation and
indulge ourselves.
Experimental studies have shown the phenomenon of time-inconsistent preferences
in a more precise and quantitative manner. For example, Ainslie and Haslam (1992)
reported that ‘a majority of subjects say they would prefer to have a prize of a $100

291
PT • I I I I NT ERT EMPO R A L C H O I C E

certified check available immediately over a $200 certified check that could not be
cashed before 2 years; the same people do not prefer a $100 certified check that could
be cashed in 6 years to a $200 certified check that could be cashed in 8 years’ (p. 69).
It is important to see how this would affect a subject’s behavior. When presented with
the second choice they will prefer to wait for the larger amount in eight years, but
sometime later, within the next six years, their preference will switch to the smaller
amount being received sooner. This implies that the discount rate used by the subjects
is greater over the short time horizon than over the long time horizon. Similar results
have been found with choices involving a wide range of goods apart from both real and
hypothetical cash rewards, for example health, food and access to video games.
Studies using matching tasks as opposed to choice tasks have confirmed these
findings, as discussed by Thaler (1981) and Benzion, Rapoport and Yagil (1989).
Subjects were asked to state the money amounts where they would be indifferent
between $x at time t and $y immediately, with both x and t being variables. This permits
the direct computation of discount rates for different time periods, and it has been
repeatedly found that the discount rate is a decreasing function of t. Of course, as
we saw in the previous chapter, this kind of methodology lends itself to the confound
influence of transactions costs, since there is no front-end delay with the payment of $y.

Reverse time-inconsistency

The term reverse time-inconsistency (RTI) was introduced by Loewenstein (1987)

to describe situations where people initially preferred the SS reward, but then later
changed their preference to the LL reward, thus switching preferences in the opposite
direction to that more normally observed. He proposed that for ‘vivid and fleeting’
consumption, like Halloween candles or expensive wines, the explanation lay in the
effect of anticipatory utility, with people choosing to delay consumption in order to
prolong savoring the moment.
There will be a discussion of explanations of both RTI and ‘normal’ time-
inconsistency later in the chapter, but it should be noted here that further studies have
observed RTI in different situations from that in the Loewenstein study. Sayman and
Önculer (2009) observed RTI in five different experiments, involving both a two-level
loyalty reward program and hypothetical monetary rewards. In the two-level reward
program customers of a café could receive 1 free croissant when they had bought 10
croissants (SS) or 2 free croissants when they had bought 15 (LL). Of the 47 subjects
participating, 19 (40%) showed RTI switching from the SS to LL after they had bought
10 croissants. Only nine (19%) of the subjects showed standard time-inconsistency,
with the remaining subjects showing time-consistent preferences, either sticking with
SS or sticking with LL.
In the other experiments in this study, the monetary rewards varied from $7 to
$25. In all the experiments the delays were in the range between a few days up to
four weeks. Other studies observing RTI have used larger rewards, equivalent to about
$1400, and delays up to two years (Ahlbrecht and Weber, 1997; Holcomb and Nelson,
1992; Scholten and Read, 2006). Sayman and Önculer (2009) find that RTI is relatively
more likely to be observed when the delay to and between the two rewards is short.

292
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

Issues and approaches

When the DUM was examined in the last chapter it was seen that the model assumed
both constant discounting (at any period the same discount rate was used to discount
outcomes in all future periods) and stationary discounting (the same discount rate was
used in all future periods as in the current period). Under these assumptions people
would have preferences that were time-consistent. There is then a two-fold challenge
for behavioral economists:
1 Determining the psychological processes underlying and explaining the anomaly
of time-inconsistent preferences.
2 Developing alternative models that explain and predict observations better.
Over the last two or three decades there have been many attempts to address these
issues, and these are discussed in the next three sections. We shall see that some
approaches have focused more on the first issue than the second, and some more
on the second issue than the first. We can classify the approaches into three general
categories: hyperbolic discounting; modifying the instantaneous utility function; and
more radical models. The first of these has received the most attention by researchers.

8.3 Hyperbolic discounting

Nature of hyperbolic discounting

The first economist to discuss alternatives to the constant discounting approach of

the DUM was Strotz (1955), who did not see any normative value in the approach.
Strotz also realized the implication of relaxing the assumption of constant discount
rates in terms of the existence of time-inconsistent preferences. Although he did
not propose any specific mathematical form of discount function as an alternative,
he did draw attention to the case of declining discount rates. This really forms the
basis of hyperbolic discounting: people tend to be more impatient in the short run,
using a higher discount rate, and become more patient over longer periods of time.
This phenomenon is referred to as present bias. The first formal models involving
hyperbolic discounting were constructed by Chung and Herrnstein (1967) and Phelps
and Pollak (1968). This work has later been developed, in particular by Ainslie (1975,
1986, 1991, 1992) and by Laibson (1996, 1997, 1998).
In order to clarify the terminology and facilitate the understanding of the
mathematics of the various models, let us begin by distinguishing between the terms
discount rate, discount factor, per-period discount factor and discount function. In
the conventional discounting model of the DUM, the discount rate ␳ is constant and
corresponds to the interest rate at which future utilities are discounted. The discount
factor is the proportion by which each period’s utility is multiplied in order to calculate
the present value of the utility. In the constant discounting model the discount factor is
given by 1/(1 + ␳)t. The per-period discount factor ␦ represents the proportion by which
each discount factor is multiplied to compute the discount factor for the following
period. In the DUM this is again constant, being given by 1/(1 + ␳). The discount
function describes the relationship between the discount factor and time, showing the
total effect of discounting over a range of time periods. The discount function in the

293
PT • I I I I NT ERT EMPO R A L C H O I C E

Figure 8.1 Shapes of discount functions

1.0
0.9
0.8
0.7
0.6 Exponential
0.5 Quasi-hyperbolic
Hyperbolic
0.4
0.3
0.2
0.1
0
1 2 3 4 5 6 7 8 9 10
Year

DUM can therefore be described as D(t) = ␦t, and this is referred to as an exponential
discount function. Thus if a person has a utility function u(x0, x1, x2… x0), the utilities in
the periods 0, 1, 2,… t are discounted by 1, ␦, ␦2,… ␦t. It should be noted that the time
variable is treated as being discrete in this case, only assuming whole numbers.
These concepts are best illustrated by an example and a graph. In the DUM we
will assume that the value of ␳ = 0.1. It follows that ␦ = 0.9091. After ten time periods
the discount factor will have the value of 0.909110 = 0.3856. Thus a utility of 100 units
expected in ten years will have a discounted present utility of 38.56 units. A graph of
this discount function is shown in Figure 8.1.
The original hyperbolic discount function introduced by Chung and Herrnstein
(1967) was based on experimental studies with animals, and took the form D(t) = 1/t.
Herrnstein (1981) also developed another special case of hyperbolic function, where
D(t) = (1 + ␣t)–1.
Phelps and Pollak (1968) used a modified version of this function, referred to as a
quasi-hyperbolic function. This is described below:
D(t) = 1 if t = 0
␤␦t if t > 0
In general ␤ < 1, implying that the discount factor between the current period and
the next is lower than the discount factor in later periods. Thus it can be said that
␤ measures the degree of present bias. In the limiting case where ␤ = 1 the quasi-
hyperbolic function reduces to the exponential function of the DUM. It should also be
noted that this model, often referred to as the (␤,␦) model, can also accommodate RTI,
by allowing ␤ > 1. According to the (␤,␦) model, in contrast to the DUM, the utilities
in the periods 0, 1, 2,… t are discounted by 1, ␤␦, ␤␦2,… ␤␦t.
Phelps and Pollak (1968) originally introduced this functional form in order to study
intergenerational altruism, and the function was first applied to individual decision-making

294
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

by Elster (1979). The pure, generalized hyperbolic function was originally introduced by
Harvey (1986) and has been further developed by Prelec (1989) and Loewenstein and
Prelec (1992). It has also been discussed in various contributions by Ainslie and Laibson
mentioned earlier. This is a continuous function, taking the form D(t) = (1 + ␣t)–␤/␣.
The ␣-coefficient determines how much the function departs from constant discounting.
The limiting case, as ␣ goes to zero, is the exponential discount function D(t) = e–␤t, in its
continuous form. In Figure 8.1 it is assumed that ␣ = 100,000 and ␤ = 3500, again to give
a function that intersects the others half-way through the ten-year period.

Implications of hyperbolic discounting

The primary implication of hyperbolic discounting is that time preferences will be

inconsistent. We have already seen that there is a large body of empirical evidence
that supports the theory of dynamic inconsistency in preferences. It is instructive to
illustrate this effect of hyperbolic discounting with a simple example at this stage,
and for this purpose we can use the values in the study of Ainslie and Haslam (1992),
mentioned earlier. Let us assume that subjects have ␤ = 0.6 and ␦ = 0.9, and they are
faced with the choice between receiving $100 in six years’ time (the SS reward) or $200
in eight years time (the LL reward). We can now write:
V0 ($100 in 6 years) = 0.6(0.9)6 (100) = $31.9
V0 ($200 in 8 years) = 0.6(0.9)8 (200) = $51.7
Thus at the present time the $200 in eight years is more appealing. However, in six
years time the situation has changed so that:
V6 ($100 now) = $100
V6 ($200 in 2 years) = 0.6(0.9)2 (200) = $97.2

Figure 8.2 Exponential discounting and consistent time preferences

Discounted SS
utility
LL
250

200

150

100

50

0
0 2 4 6 8 10
Time (years in the future)

295
PT • I I I I NT ERT EMPO R A L C H O I C E

Figure 8.3 Hyperbolic discounting and inconsistent time preferences

Discounted SS
utility
LL
250

200

150

100

50

0
0 2 4 6 8 10
Time (years in the future)

At this point the value of the immediate receipt of $100 exceeds the value of the
$200 in two years, and the subjects’ preference has reversed. This example illustrates
a ‘temptation’ situation. A comparison of exponential and hyperbolic discounting is
shown in Figures 8.2 and 8.3. This is useful in illustrating the differences between naïve
and sophisticated consumers, which is explained shortly.
We can also see how the (␤,␦) model describes a procrastination situation, by
considering the mirror image of the problem above and changing both payoffs into
negative ones, so that the first payoff is –$100 in six years and the second one is –$200
in eight years. At the current time the smaller discounted cost of –$31.9 of the SS is
preferable to the larger discounted cost of –$51.7 of the LL. However, in six years time,
subjects will prefer to switch to LL, with its lower discounted cost of –$97.2 compared
with the immediate cost of –$100 of the SS.
The reason for the effectiveness of the (␤,␦) model lies in its assumption of a higher
discount rate between the current period and the next (for normal time-inconsistency),
but a constant discount rate thereafter. The per-period discount rate between now and
the next period is (1 – ␤␦)/␤␦, whereas the per-period discount rate between any two
future periods is (1 – ␦)/␦, a smaller value.
The manner in which such present bias and inconsistency affects behavior depends
on the degree of self-awareness of subjects, in terms of how aware they are that their
preferences will change over time. There are two extreme situations: people may be
completely ‘naïve’, believing that their future preferences will be identical to their
current ones. This would imply that people do not learn at all from past experience
of changing preferences. It was assumed in the above numerical example that subjects
were of this type. Naïve agents think that they will use a constant discount rate in
the future, but will actually discount hyperbolically. If we refer to the person’s belief
regarding the value of their ␤ as b, then ␤ < b = 1. Thus these consumers believe
their preferences in the future are shown in Figure 8.2, seeming that LL will always be
preferable to SS, and they will therefore not anticipate any forthcoming conflict.

296
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

The opposite extreme is where people are completely ‘sophisticated’, and can
predict accurately how their preferences will change over time. In this case ␤ = b < 1.
In this case consumers realize that their preferences in the future are shown in Figure
8.3, and anticipate a conflict when the preference reversal occurs. Thus sophisticated
individuals have a key advantage over naïve ones, in that they can pre-commit to certain
courses of action that prevent them from yielding to a preference reversal later. We
saw an example of such a commitment in the introduction to the chapter, concerning
alarm clocks. In reality it appears that most people lie somewhere in the middle of
the naïve–sophisticated spectrum, meaning that ␤ < b < 1, although there is limited
empirical evidence regarding this aspect.
A good illustration of the differences between the behavior of exponential
discounters, naïve and sophisticated hyperbolic discounters, which incorporates
commitment, is given by Ho, Lim and Camerer (2006). They give a hypothetical
numerical example relating to the situation of buying and consuming potato chips.
They use a three-period model as follows:
1 Purchase decision: this involves a choice between a small (one serving) bag or a
large (two servings) bag which involves a quantity discount.
2 Consumption decision: this involves a choice between consuming one serving or two,
and an instantaneous utility related to consumption. If the smaller bag is purchased
in the first period, only one serving can be consumed, but purchase of the larger bag
offers the choice between consuming a single serving and leaving the other to a later
period, or consuming both servings in the same period. Thus buying a smaller bag acts
as a commitment in this case, as far as eating less and improving health is concerned.
3 Health outcome: this is adverse because chips are bad for you, but is much worse if
two servings are consumed rather than one.
Under these conditions, and using reasonable parameters for discounting and
outcomes, the authors conclude that each group of discounters may behave differently:
1 Exponential discounters: these may buy a large bag to benefit from the quantity
discount, but only consume a single serving in the second period to avoid the worst
health outcome.
2 Naïve hyperbolic discounters: these may buy a large bag, believing that they will
behave like the exponential discounters and only consume one serving in the next
period. However, in the second period they discount hyperbolically, applying a
high discount factor to the adverse health effects in the third period, and they end
up consuming both servings.
3 Sophisticated hyperbolic discounters: these may choose the small bag as a self-control
or commitment device, knowing that in the next period they would be unable to
resist the temptation of consuming both servings if they bought the large bag.
A numerical example will illustrate this situation. We assume for simplicity that ␦ = 1
(so that exponential discounters do not discount future outcomes at all) and that ␤ =
0.5. Let c = number of servings consumed in any time period, and p = the price per
serving. We assume p = 1.5 for a small bag, and p = 1 for a large bag, to reflect the bulk
discount. We model the utility flows as follows:
1 At the purchase decision (t0) there is negative utility given by U0 = –cp, which is the
cost of purchasing the chips.
2 At consumption decision (t1) there is instantaneous utility given by U1 = 1 +5c

297
PT • I I I I NT ERT EMPO R A L C H O I C E

3 Health outcome occurs with a one-period lag after consumption and is negative, given
by U2 = 3 – 6c
It is now possible to calculate utilities for the three different types of consumer, rational,
naïve and sophisticated, in the relevant time periods, for each consumption decision, in
order to see how each type of consumer will behave. Table 8.1 shows the situation for
rational consumers.

Table 8.1 Rational consumers (exponential discounters)

t S L(1) L(2)
0 –1.5 –2 –2
1 6 6 11
2 –3 –3 + 6 = 3 –9
3 –3
PV0 1.5 4 0
PV1 N/A 6 2

There are three possible decisions:

1 S refers to purchasing a small bag, limiting consumption to one serving in the following
time period.
2 L(1) refers to the purchase decision to buy a large bag and then consume a single
serving in the next time period. It is assumed that the remaining serving is consumed
in time period 2, so in this time period there is a health outcome of –3 related to
consumption in the previous period and a positive consumption utility of 6, yielding a
net utility of 3. In time period 3 there is the negative health outcome of –3 related to
consumption in period 2.
3 L(2) refers to the decision to buy a large bag, and then consume both servings in the
next time period.
It can be seen from Table 8.1 that the rational consumer who discounts exponentially
will maximize PV0 at the purchase decision by choosing L(1), buying a large bag and
planning to consume one serving in the next period, with the remaining serving being
consumed in t2. When t1 is reached, the consumer will still maximize utility by sticking
to his plan and consuming a single serving in each of the next two time periods.
Let us now consider the situation for naïve consumers. This is shown in Table 8.2.

Table 8.2 Naïve consumers

t S L(1) L(2)
0 –1.5 –2 –2
1 3 (6) 3 (6) 5.5 (11)
2 –1.5 1.5 –4.5
3 –1.5
PV0 0 1 –1
PV1 N/A 6 6.5

298
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

Expected utilities at t0 are shown first; some of these utilities are revised at t1, because of
present bias, and are shown in brackets. As with rational consumers, naïve consumers
maximize expected utility at t0 by choosing L(1), purchasing the large bag, planning to
consume a single serving in each of the following two time periods (PV0 = 1). However,
once they reach t1 these consumers will maximize expected utility by switching to L(2),
which is perceived to have PV1 = 6.5 compared with PV1 = 6 for the L(1) plan.
Sophisticated consumers have complete knowledge of the change in utilities in
the future, and therefore realize that they will change preferences at t1. Since at t0 L(2)
is the worst of all the options, with PV0 = –1, sophisticated consumers will therefore
commit to S to prevent the switch to L(2) in the future.
If the utilities of actual behavior are measured from the viewpoint of the first
period then the exponential discounters end up with the most utility in this example
(PV0 = 4), because they benefit from the discount. Sophisticated consumers end up
with PV. The naïve hyperbolic discounters end up worst off (PV0 = –1), because
of the adverse health effects that they discount heavily compared to the benefits of
consumption when they get to the second period. However, we should not conclude
that sophisticated hyperbolic discounters will always end up better off than naïve ones.
If they anticipate that they will eventually succumb to temptation, they are more likely
to succumb earlier (in an ‘unravelling’ effect) than naïve discounters.
This model therefore explains consumption patterns for both investment goods
and leisure goods. Investment goods have an immediate cost but a delayed benefit,
whereas leisure goods (like eating junk food or watching TV) have an immediate
benefit but a delayed cost. Let us call the immediate payoff x1 at time t = 1 and the
delayed payoff x2 at time t = 2. Therefore the investment good has x1 < 0 and x2 > 0,
while the leisure good has x1 > 0 and x2 < 0. We can now consider and compare three
different concepts from an ex ante viewpoint when t = 0: (1) how much (␤,␦) consumers
want to consume, (2) how much such consumers actually consume, and (3) how much
naïve consumers expect to consume.
1 Consumers want to consume at t = 0 if ␤␦x1 + ␤␦x2x2 > 0, or x1 + ␦x2 > 0.
2 Consumers actually consume at t = 1 if x1 + ␤␦x2 > 0.
Investment goods have x2 > 0, and since this is multiplied by ␤, which is smaller
than 1, the result is that these agents underconsume investment goods. Similar
reasoning indicates that they overconsume leisure goods (x2 < 0).
3 Naïve consumers expect to consume if x1 + b␦x2 > 0.
Since b < ␤, these consumers overestimate their consumption of investment goods
and underestimate their consumption of leisure goods.

Advantages of hyperbolic discounting

There are two main advantages of the (␤,␦) model, described below.
1 Analytical tractability
The model maintains most of the analytical tractability of the exponential model.
It is again a discrete function, and after period 1 the per-period discount factor is
␦, the same as the exponential function. In Figure 8.1 it is assumed that ␤ = 0.7
and ␦ = 0.98, since this produces a function that crosses the exponential function
half-way through the ten-year period for ease of comparison.

299
PT • I I I I NT ERT EMPO R A L C H O I C E

2 Congruence with reality

The model fits empirical findings well, mimicking the qualitative property of the
hyperbolic discount function. This can be seen in Figure 8.1 where hyperbolic,
quasi-hyperbolic and exponential functions are compared. DellaVigna (2009)
summarizes several studies that support the (␤,␦) model, explaining anomalies in
the DUM. These studies relate to: excessive preference for membership contracts
in health clubs (DellaVigna and Malmendier, 2006); positive effects of deadlines on
homework grades and preference for deadlines (Ariely and Wertenbroch, 2002);
preference for pre-teaser rather than post-teaser interest rates in credit-card take-up
(Ausubel, 1999); liquid credit card debt combined with simultaneous illiquid wealth
accumulation (Laibson, Repetto and Tobacman, 2009); demand for illiquid savings
as commitment devices (Ashraf, Karlan and Yin, 2006); demand and default for
payday loans (Skiba and Tobacman, 2008); and default settings in retirement savings
plans (Madrian and Shea, 2001; Cronqvist and Thaler, 2004). The findings of these
studies are discussed in more detail in the section related to empirical evidence.

Criticisms of the hyperbolic discounting approach

It is fair to say that hyperbolic discounting has entered into the mainstream of
behavioral economics, largely due to its well-documented empirical superiority over
the exponential model and its analytical convenience. However, it has not been without
its critics. It should be noted that these critics are largely not defenders of the traditional
DUM, but are proponents of newer and more radical models.
Let us examine criticisms from defenders of the DUM first. These fall mainly into
three categories, all of which have been mentioned earlier:
1 Failure to use front-end delay. This results in a confound with transactions costs
explained in the last chapter.
2 Use of hypothetical rewards. This may lead to unreliable results due to a lack of
incentives compared with the use of monetary rewards.
3 Failure to provide information relating to the annual interest rates implied in the
different options. Most studies simply give the options in terms of choice or
matching tasks without such information. Coller and Williams (1999) found that
discount rates were significantly lower when annual interest rate information is
provided.
The first two criticisms have largely been countered over the last decade by numerous
field studies involving hyperbolic discounting, many of which were summarized earlier
in the section. These studies have often involved situations with delayed SS rewards or
costs and these rewards and costs have been real rather than hypothetical.
As far as the third criticism is concerned, it has been sometimes suggested that
there are legal requirements in many countries that require the provision of such
information, and that this makes the provision of interest rate information a realistic
condition. This may be true in different types of lending/borrowing situations, but the
majority of intertemporal choice decisions are not of this type. When we are debating
whether to eat a dessert, or join a health club, or tidy the garage, these are not situations
where interest rate information is realistically going to enter the decision process.
Furthermore, as was discussed at the end of the last chapter, studies that have
incorporated the desired control elements, such as that by Harrison, Lau and Williams

300
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

(2002), have not eliminated the nonconstant discount rate phenomenon. While rates
did not vary much between one and three years, rates were higher for the six-month
time frame. Many hyperbolic discounting studies show higher rates for periods less
than a year.
We will now move on to consider criticisms from proponents of newer discounting
models. One more recent theory that has been proposed as being superior to hyperbolic
discounting is subadditive discounting. The concept of subadditivity has been discussed
in Chapter 5, and when applied to discounting it implies that people are less patient
(i.e. have higher discount rates) per unit of time over shorter intervals regardless of
when they occur. Thus the theory suggests that people would have a higher discount
rate for a daily period than for a monthly period, even if the daily period were at a
relatively distant point in the future. Read and Roelofsma (2003) conclude that
subadditive discounting is superior to hyperbolic discounting in terms of explaining
empirical results for both choice and matching tasks.
However, the main criticism that has been aimed at hyperbolic discounting, which
applies also to subadditive discounting, is that it lacks a psychological foundation. It
is basically a descriptive theory rather than an explanatory one. Although one can say
that the (␤,␦) model describes ‘temptation’ and procrastination self-control problems,
this begs the question why people should use such a discounting method, particularly
since it does not appear to result in optimizing welfare, and therefore is maladaptive in
evolutionary terms. For example, a study of procrastination by students has shown that
those with time-inconsistent preferences performed worse than those with consistent
preferences, and this was true even with sophisticated students who had full awareness
of their time inconsistency (Wong, 2008).
Even though the phenomenon has been widely observed among animals as well as
humans, and has been studied by researchers in many disciplines, its strongest proponents
have failed until recently to provide a good psychological foundation for it. Most
researchers have entirely ignored this aspect, and of course this is quite legitimate from
the standpoint of standard economic theory, which is only concerned with behavior, not
psychological processes. However, given that there are alternative theories to explain the
same behavior, which have different implications, it is worthwhile to examine psychological
foundations. In the last decade some researchers have been conscious of this failure. For
example, Ainslie (2001), a psychiatrist, has considered the possible evolutionary origins
of such a psychological mechanism, in terms of how it might have increased inclusive
fitness compared with the more intuitively appealing exponential discounting approach.
Ultimately he admits that he has no idea how a hyperbolic discounting mechanism could
have evolved. However, more recently there have been attempts to explain this (Robson,
2002; Robson and Samuelson, 2007, 2009). These explanations are discussed in the
section on empirical evidence, in terms of evolutionary biology.

8.4 Modifying the instantaneous utility function

In the last chapter we described various anomalies observed in the DUM, and in
the first section of this chapter many of these are viewed as confounds as far as
measuring time preference and discount rates are concerned. It can be argued that
it is more appropriate to treat these confounding factors as additional aspects of
the instantaneous utility function. Models that attempt to achieve this are now discussed.

301
PT • I I I I NT ERT EMPO R A L C H O I C E

Habit-formation models

There is a long tradition of habit-formation models in economics, going back to

Duesenberry (1949). His hypothesis that the level of consumption depended on a past
peak of consumption was based on the idea that current utility depends not only on
current consumption but on past consumption:
Ut = f(Ct, Ct–1, Ct–2,…) (8.1)
In most models all the values of past consumption are combined together into a
composite variable, Zt, that may be exponentially weighted to give more importance to
more recent periods, and that is increasing with past consumption. The utility function
then becomes:
Ut = f(Ct, Zt) (8.2)
Thus the more a person has consumed in the past the more utility current consumption
will yield, causing the person to consume more now. It should be noted that this ignores
the possibility of shifting reference points, discussed shortly, which would have the
opposite effect. This model can be applied in macroeconomic terms to all consumers
and all goods, or on a microeconomic basis to particular goods. For example, Becker
and Murphy (1988) have used a habit-formation model to examine the effects of past
and future prices on the current consumption of addictive goods. These models have
also often been used to explain the equity premium puzzle discussed in Case 6.1.
We will examine their relevance again in Case 8.2, which concerns the relationships
between savings, consumption and growth.
As mentioned above, shifting reference points have important implications for
habit formation. Another important factor related to habit formation is diminishing
marginal sensitivity. Since both of these factors are elements of prospect theory, a
more complete discussion of habit formation necessitates an examination of prospect
theory models. We will also see in a later section that there are some important
policy implications related to habit formation, especially as far as food intake and diet
are concerned.

Prospect theory models

Perhaps the single most salient characteristic of prospect theory is its use of reference
points, as explained in Chapter 5. When a person’s reference point for current
consumption is past consumption, a reference-point model is identical with a habit-
formation model. However, this is only a special case of the more general reference-
point model. Reference points can also be dependent on expectations of the future, or on
social comparisons, as we have seen. The importance of social comparisons is considered
in Case 8.3, in connection with the preference for rising consumption profiles.
Other important features of prospect theory that affect the instantaneous utility
function are loss-aversion and diminishing marginal sensitivity. Loewenstein and Prelec
(1992) have used a utility function incorporating such features in order to explain the
anomalies of the ‘magnitude effect’, the ‘sign effect’ and the ‘delay-speedup’ asymmetry
discussed in the last chapter. The factor that is particularly important in their analysis is
the concept of the elasticity of the utility function. The elasticity concept ‘captures the
insight that people are responsive to both differences and ratios of reward amounts’

302
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

(Frederick, Loewenstein and O’Donoghue, 2002). Thus a person may be indifferent

between receiving $10 now and $20 in a year, but prefer $200 in a year to $100 now. It
may appear that a lower discount rate is being used for the larger amounts; however,
in reality the person may have a constant time preference in the two options but may
be more responsive to the difference between $100 and $200 than to the difference
between $10 and $20.
Similarly, the Loewenstein–Prelec model can explain both the ‘sign effect’ and
the ‘delay-speedup’ asymmetry in terms of loss-aversion, with people discounting gains
more than losses. In the case of the ‘delay-speedup’ situation it needs to be recognized
that any shift in consumption either forwards or backwards in time is made less
desirable by loss-aversion, since although one gains consumption in one period, one
also loses consumption in another period (when it was originally expected). Thus the
gains from accelerating consumption are not as great as the losses from delaying it.
Reference-point models incorporating loss-aversion have also been applied to draw
conclusions regarding the Permanent Income Hypothesis (Friedman, 1957). According
to the conventional log-linear version of the PIH model, changes in future income,
while affecting the level of consumption, will not affect the rate of consumption
growth; consumption in all future periods will be increased or decreased by the same
proportion in every period. However, as shown by Bowman, Minehart and Rabin
(1999), a decrease in future incomes may not reduce current consumption very much
because of loss-aversion, thus causing future consumption to fall by a greater amount
as people are forced to adjust. Consumption growth may therefore respond more to
future income decreases than to future income increases, and two studies by Shea
(1995a and 1995b) lend some support to this hypothesis.
The characteristic of diminishing marginal sensitivity also has implications for
habit formation models, especially when combined with shifting reference points.
This concept has different effects over different time frames, and for different types
of consumption, so we will start by considering a fundamental example, food. Of all
self-control problems the tendency to overeat and the corresponding health problems
related to obesity are among the most serious. We have already seen the nature of
some of these problems in Chapter 3, particularly related to the studies by Wansink
and colleagues (2005, 2006, 2009). The central theme in these studies is that there are
many anchoring effects that determine how much we eat, like size of plate, number
and sex of eating companions, and whether we are watching TV at the same time.
The role of diminishing marginal sensitivity essentially relates to satiation. In the short
term, which in the case of food can be over the normal course of a meal, or even over
a few days, we tend to become satiated as we eat more food in general or more food of
a particular type (Epstein et al., 2009). The same study found that people react in the
same way to sexual stimuli of a particular type as far as diminishing marginal sensitivity
is concerned, and of course it is well known that reactions to recreational drugs are
similar. However, over the longer term, the diminished sensitivity can lead to a shifting
of the reference point to a higher level of consumption. Hence drug-takers need to
increase their dosage over time to get the same ‘fix’ or high. The physiology underlying
this phenomenon, in terms of chemical receptors and the function of neurotransmitters
is fairly well known in this case, but varies somewhat from case to case. With food intake
much research still needs to be done in terms of how diminishing marginal sensitivity
and shifting reference points operate, and also how they vary from person to person.
Some policy implications of these aspects are discussed in a later section.

303
PT • I I I I NT ERT EMPO R A L C H O I C E

Anticipatory utility models

We have seen in Chapter 3 that people derive utility from anticipation of future
consumption as well as from current and past consumption. This effect can be expressed
in a manner parallel to the habit-formation model and (8.1):
Ut = f(Ct, Ct+1, Ct+2,…) (8.3)
This phenomenon works in the opposite direction to the normal direction of time
preference, for both gains and losses. Thus we may prefer to delay consumption
of certain products where anticipatory rewards are important, and accelerate bad
outcomes rather than have them hanging over us (Loewenstein, 1987). We have seen
that such preferences associated with anticipatory utility can explain reverse time-
inconsistency. There are certain exceptions to this phenomenon: waiting for a good
outcome can be frustrating, while one may prefer to delay the possibility of a bad
outcome to avoid spoiling the weekend (for example, when students take a test). It
should be noted that if there is uncertainty regarding the outcomes this adds another
dimension to the situation, as the emotions of hope and anxiety enter the picture. Such
visceral influences are discussed shortly. If a normal discounting approach is taken,
without modifying the utility function, the result is that different discount rates will be
calculated with different goods, as is often observed in empirical studies. In this case
the reality is that the utilities have not been modified to allow for anticipation and that
if this is done the discount rates may actually be constant.

Visceral influence models

The nature of visceral influences has been discussed earlier, both in this chapter and
in Chapter 3. In particular, we have seen that the temporal proximity of an outcome
may increase its desirability. This may cause a higher rate of discount to be computed
for near-future outcomes when factors like anger, hunger, lust and sleeplessness are
involved, thus seeming to support the hyperbolic discounting approach. However, it
may be more appropriate to modify the instantaneous utility function to allow for a
momentary increase in utility in certain circumstances (Loewenstein, 1996; 2000).
It should be noted that temporal proximity of the outcome is only one of these
circumstances. Other cues may also be important, for example spatial proximity, or the
presence of associated sights, sounds or smells.
The influence of visceral factors is more complicated than just the effect on the
instantaneous utility function. It has been found, for example, that when people are
under their influence they tend to overestimate how long their effect will last, while
when people are not under their influence they tend to underestimate the magnitude
of their effect in the future. People also tend to perceive immediate emotions as being
more intense than previous emotions, a phenomenon referred to as ‘immediacy bias’
(Van Boven, White and Huber, 2009). This bias may be caused by salience or by greater
availability of information regarding the present emotion. When people are reminded
that information about emotions naturally decays from memory, this tends to reduce
immediacy bias.
Another effect of visceral factors leads to a paradox, in that people may not want
to want certain things, even expressing their preference for not doing something while
they are actually doing it, like taking drugs. This phenomenon relates to the difference

304
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

between wanting and liking, so it may be more meaningful to state that people may
not want to like certain things. Obviously self-control factors are relevant here, and the
concepts of temptation and willpower. These will be further discussed in later sections,
in the context of multiple-self models and policy implications.

8.5 More radical models

In Chapter 5 we saw that it was possible to distinguish between conventional

modifications to expected utility theory and non-conventional alternatives. However,
even these non-conventional alternatives, like prospect theory, are really best viewed
as extensions of the standard model rather than outright rejections of it. Similar
considerations apply to the DUM, but it is more difficult here to draw the line between
the conventional and the non-conventional. So far the alternative models to the DUM
that we have examined have involved modifying either the discount function or the
instantaneous utility function. Even here, though, some of the modifications involve
non-conventional factors related to prospect theory. The models examined in this
section involve more radical differences from the DUM, although in some cases they
are still often considered conventional models, like the dual-self model of Fudenberg
and Levine (2006). The reason for this lack of clarity, or blurring of distinctions, is that
intertemporal models are more complex than static models of preference, consisting of
more components. This has led to a hybridization of models, in that some components
of a model may be ‘conventional’ while other components may be ‘non-conventional’.
This issue is best explained in more detail as we examine the individual models.

Projection bias models

This is an example of a phenomenon that was discussed earlier. People’s tastes change
over time, and there is a general tendency to underestimate the magnitude of these
changes (Kahneman and Snell, 1992). The presence of visceral influences, discussed
above, is only one of the factors that can cause this; habit formation and the shifting of
reference points are two other important factors that can cause the same phenomenon.
This bias is contrary to the assumption of rational expectations in the standard model,
which implies that people can forecast changes in their tastes accurately. It has been
modelled by Loewenstein, O’Donoghue and Rabin (2003), who review extensive
evidence for the phenomenon. In the case of habit-formation the utility function in
(8.2) may be appropriate as the instantaneous utility function at time t: Ut = f(Ct, Zt).
This can be expressed more simply as Ut(Ct, Zt), where Zt again represents a composite
variable reflecting past consumption. At time t +1 an individual’s true instantaneous
utility function may be Ut+1(Ct+1, Zt+1), and their expectation of this function at time
t may be Ũt+1(Ĉt+1, Zt+1  Zt). This represents expected utility in time t +1 of expected
consumption in time t +1 and past consumption up to that period, given the current
level of past consumption at time t. According to the projection bias model:
Ut(Ct, Zt) < Ũt+1(Ĉt+1, Zt+1  Zt) < Ut+1(Ct+1, Zt+1)
This can be modeled more precisely using a weighted function to indicate how
accurately people forecast future utilities:

305
PT • I I I I NT ERT EMPO R A L C H O I C E

Ũt+1(Ĉt+1, Zt+1 ⱍ Zt) = ␣ [Ut(Ct, Zt)] + (1 – ␣)[Ut+1(Ct+1, Zt+1)] (8.4)

where 0 ⱕ ␣ ⱕ 1
The higher the value of ␣, the greater the degree of projection bias, meaning that there
is a greater tendency to underestimate future utilities.
This phenomenon, which resembles myopia, has important policy implications,
for individuals, firms and governments. For example, people may consume more of a
good now, underestimating the effect that this will have on future utility, and therefore
underestimating future consumption, or the desire for future consumption, of the
good. In this situation people may discount future consumption too highly as far as
the maximization of welfare in terms of experienced utility is concerned. The policy
implications are discussed in the next section.

Mental accounting models

The features of mental accounting were discussed extensively in Chapter 6, and

it was seen that these involved the aspects of framing and editing, fungibility and
choice bracketing. We now need to examine the effects of these behavioral aspects on
intertemporal choice and discounting.
One of the main implications of lack of budget fungibility is that different discount
rates are applied to different goods. We have seen that small purchases may be classified
as ‘petty cash’, with the result that people may be more inclined to spend on these,
using a higher discount rate. Goods involving a larger expenditure, such as durables,
may be evaluated more carefully, using a lower discount rate.
There are also a number of implications of choice bracketing which contradict
the predictions of the DUM. We have seen that people often prefer to prepay for
various expenses to avoid the ‘pain of paying’ later for something they have already
consumed (Prelec and Loewenstein, 1998), whereas the DUM predicts a preference
for paying later. Furthermore, they may prefer to receive payment for work after rather
than before performing it, again in contradiction to the DUM. We have also discussed
the preference for payment decoupling, which may lead to fixed-fee pricing with zero
marginal costs, as for example with many health club membership schemes. This is
another contradiction of the DUM, which predicts a dislike for up-front fees. This
situation is examined in more detail in Case 8.1.
Another anomaly of the DUM observed earlier, that relates to choice bracketing,
is the preference for spread of consumption. Loewenstein and Prelec (1993) found
that people tend to prefer to spread the ‘treats’ of dining at a fancy French restaurant,
although in this case there was also a preference for an improving sequence of
outcomes, as predicted by prospect theory. The preference for spread is a separate
phenomenon, and appears to be related to anticipatory utility.

Multiple-self models

The term multiple-self has an element of ambiguity that needs to be clarified from the
outset. First the concept can be applied to the situation where the ‘self’ is a dynamic
and ever-changing entity over time. This is most clear when we compare our current
self, in terms of attitudes, values and beliefs, with our self in some period well in the
past; sometimes we have difficulty understanding our past selves and may indeed be
embarrassed by them. However, the term multiple-self can also be applied to situations

306
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

at a particular point in time, when there appear to be conflicts between our short-term
‘self’ and our long-term ‘self’. Models of this situation are often referred to as dual-self
models, for example that of Fudenberg and Levine (2006) discussed later.
All these models are inspired by the observation that self-control problems are
commonplace, and often involve forms of commitment described in the discussion of
hyperbolic discounting. Indeed the term ‘self-control’ can be considered meaningless
unless there is more than one ‘self’. It begs the questions:
1 Who is doing the controlling, if not the self?
2 Why is there a need for self-control, if there is just a single self seeking to maximize
some kind of preference function?
Furthermore, there is significant neuroscientific evidence that we have two separate
systems that are involved in intertemporal decision-making. This is discussed in the
next section.
There are a variety of multiple-self models. Some models involve a near-sighted
or myopic self and a far-sighted self who are in conflict and alternately take control of
behavior (Winston, 1980; Schelling, 1984; Ainslie and Haslam, 1992). These models
are criticized by Frederick, Loewenstein and O’Donoghue (2002) on the grounds that
they fail to explain why either type of self gains control, and they do not capture the
fundamental asymmetry between the far-sighted and near-sighted selves. Far-sighted
selves can make commitments to control the behavior of near-sighted selves, but not
vice versa. An example is where the far-sighted self leaves the alarm clock across the
room from the bed, so that the near-sighted self cannot just slam it off quickly in the
morning and go back to sleep.
As noted by Frederick, Loewenstein and O’Donoghue (2002), ‘few of these
multiple-self models have been expressed formally, and even fewer have been used to
derive testable implications that go much beyond the intuitions that inspired them in
the first place’ (p. 376). However, as they also point out, this is not so much a failure of
the models themselves as an indication of the complexity of the underlying phenomena.
Certainly the models do help to explain the existence of various self-control strategies,
and can also, with the aid of game theory, provide a much-needed psychological
foundation for hyperbolic discounting.
A recent psychological foundation for hyperbolic discounting is provided by
another multiple-self model involving the theory of psychological connectedness
(Bartels and Rips, 2010). In general, we are more psychologically connected to our
proximate selves in the future, because key personality characteristics and preferences
are less likely to change. We are likely to be less psychologically connected to our more
distant selves. This theory, supported by experimental evidence, not only accounts for
hyperbolic discounting, but it also accounts for another anomaly in the DUM: why
people tend to discount larger rewards more than smaller ones. Larger rewards are
more likely to change them psychologically and reduce the psychological connectedness
with the current self.

Dual-self models

A variation of the above model has been proposed by Thaler and Shefrin (1981), along
the lines of principal-agent theory. The far-sighted self is the principal or ‘planner’,
while there is a sequence of near-sighted selves who constitute the agent or ‘doer’. Thus
the model captures the asymmetry aspect. The far-sighted planner is concerned with

307
PT • I I I I NT ERT EMPO R A L C H O I C E

future utilities, whereas the near-sighted doer is only concerned with the instantaneous
utility function at a particular point in time. The planner is at least partially aware of
the conflicts that will occur in the future, for example when the dessert trolley comes
around, and can adopt commitment strategies to control the behavior of the doer, by
perhaps only going to restaurants that do not serve tempting desserts. This type of
principal–agent model involves aspects of game theory discussed in the next chapter.
A further type of dual-self model is also based on game theory, in this case relating
to social interaction and the choice between cooperation and defection (Elster, 1985b).
This is fundamentally a prisoner’s dilemma situation on a repeated basis. Self-control
requires the continued cooperation of a series of instantaneous selves. As we will again
see in the next chapter, there is a tendency for an unraveling sequence of defection,
whereby sequential selves repeatedly give in to temptation. There is also a self-
signaling effect here, which has been discussed earlier, whereby giving in to temptation
signals the next self that they lack the self-control to commit themselves to avoiding
temptation in the future, thus destroying confidence and ‘willpower’.
The most well developed model of this type is arguably the dual-self model of
Fudenberg and Levine (2006). This actually rejects hyperbolic discounting and for this
reason may be termed a conventional model. However, it also incorporates elements
of prospect theory and mental accounting which are unconventional. The Fudenberg–
Levine (FL) model posits a patient long-run self and a sequence of myopic short-run
selves. These selves are involved in playing a game in a sequence of stages. The long-
run and short-run selves share the same preferences over the outcomes in each stage,
but they regard the future differently. The short-run self is impatient and has ‘baseline
preferences’ only for the current stage, while the long-run self also has preferences for
future stages. Each stage consists of two phases. In the first phase the long-run self can
choose a self-control action that influences the utility of the short-run self. This means
that, at some cost in utility for both selves, the long-run self can choose preferences
other than the baseline preferences. In the second phase, once its preferences have
been determined, the short-run self takes the final decision. This whole process is
illustrated in Case 8.1 related to joining a gym and exercising.
Fudenberg and Levine emphasize the advantages of their model over hyperbolic
discounting, in that it produces a single equilibrium for behavior, rather than the
multiple equilibria that are associated with hyperbolic discounting and the multiple-
self model. While being analytically simpler and making more precise predictions, they
claim that it can explain empirical facts just as well.
There are a number of predictions or implications of this model. For example, the
authors find that self-control costs lead to longer delays. They also develop a banking-
savings model where it is predicted that people will use self-control in limiting the
amount of pocket cash that they have available to spend later in a nightclub scenario. It
is notable that this aspect of the model incorporates mental accounting concepts, in that
bank cash is regarded differently from pocket cash in terms of the marginal propensity
to consume. The concept of a reference point is also used, in that the amount of pocket
cash is used as the reference point for spending, not one’s total wealth. It is important
that the constraint on spending here is not liquidity, since in principle one could write
a check or use a credit card in the night club. However, these are ‘nonanonymous’
accounts, meaning that spending from them will result in an identifiable transaction
later, which may cause self-recrimination – or recrimination from one’s partner. Cash,
on the other hand, is an anonymous account, as we discussed in the chapter on mental

308
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

accounting. Fudenberg and Levine also explain Rabin’s ‘risk paradox’ (2000) in a
similar way: people are averse to taking small risks which involve pocket cash, but do
not have similar risk-aversion for large gambles that involve bank cash.
One final implication of the model is important. It proposes that the costs of self-
control are nonlinear, meaning that an increase in self-control involves an increasing
marginal cost. The underlying principle here is that self-control is an exhaustible
resource, and that therefore the law of diminishing returns applies, as discussed in
more detail in Chapter 3. The consequence is that increasing cognitive load reduces
self-control. The empirical evidence regarding the various aspects of the FL model is
discussed in the next section.
Dual-self models have also been proposed by psychologists and neuroscientists,
and there is an element of consilience here, in that findings in different disciplines using
different approaches have tended to come to similar conclusions. Brocas and Carrillo
(2008b) approach the subject from the neurological basis of brain modularity, where
there is ample evidence (examined in the next section) that the brain not only consists
of different systems, but also that these systems are in conflict in various ways. They
propose three main sources of conflict: asymmetric information, temporal horizon, and
incentive salience. So far we have concentrated mainly on the second of these sources,
although the first is relevant in the ‘planner-doer’ model. Brocas and Carrillo propose
that decreasing impatience and hyperbolic discounting emerge as a result of these two
conflicts. However, there is another ‘dual-self’ aspect related to incentive salience that
has been referred to in Chapter 3 related to wanting and liking. Robinson and Berridge
(2003) and Berridge (2001) show that there is one system that mediates the feeling
of pleasure and pain (the ‘liking’ system) and a different system that mediates the
motivation or incentive to seek pleasure and avoid pain (the ‘wanting’ system). The
evidence relating to this is discussed later in relation to neurological studies.

The procedural approach

In the discussion of hyperbolic discounting it was noted that its most important failing
was the lack of a psychological foundation. Rubinstein (2003) both disputes the
empirical evidence for hyperbolic discounting, and provides an alternative framework
for decision-making which he claims does have a legitimate psychological foundation.
Like various other models we have seen, Rubinstein’s approach is based on a heuristic
process. This proposes that the decision-maker uses a procedure that applies similarity
relations, involving a money dimension and a time dimension, in a series of three steps.
The objects of choice in intertemporal situations can be described as being in the
form (x, t), where $x is received with a delay of t units of time. Thus a decision-maker
may have to compare two choices: A = (x, t) and B = (y, s). According to Rubinstein
(2003) many decision-makers go through the following three steps:
1 Search for dominance
If x > y and t < s then A dominates B, since it is preferable in both dimensions (a
larger reward is received sooner).
2 Search for similarities in the two dimensions
If the decision-maker finds similarity in one dimension only, he determines his
preference using the other dimension only. For example, if x is similar to y, but t > s,
then B is preferred to A, since the rewards seem similar, but B involves less delay.

309
PT • I I I I NT ERT EMPO R A L C H O I C E

3 Use of a different criterion

If the first two stages do not give a result, some different criterion must be used
(Rubinstein is not specific regarding the nature of this).
Rubinstein conducts three experiments to test how the procedural approach explains
behavior compared with hyperbolic discounting. It is worth describing the first
experiment here, since it will aid an understanding of the different models and how
they compare. This experiment was performed in 2002 with a total of 456 students on
a between-subjects basis involving choice tasks and money. Different students were
asked to answer either question 1 or 2 below:
Q1 Choose between the following two options:
a) Receiving $467.00 on 17 June 2004
b) Receiving $607.07 on 17 June 2005
Q2 Choose between the following two options:
a) Receiving $467.00 on 16 June 2005
b) Receiving $467.39 on 17 June 2005 (p. 1211)
Fifty-five per cent of the subjects chose delay in question 1, while only 46% chose
delay in question 2. According to the procedural approach the 39 cents difference in
question 2 is too small to be meaningful, causing subjects to tend to prefer the shorter
delay, even by only one day. In question 1, however, neither amount nor time period
are similar, so subjects have to resort to the third step of the approach. Rubinstein
comments that the results contradict any hyperbolic discounting approach. The one-
day discount rate implied by the preference for earlier delivery in question 2 is lower
than the one-year discount rate implied by the preference for delay in question 1.
While this example does not provide conclusive evidence in favor of the procedural
approach, when combined with the results of the other two experiments, it does
pose significant questions regarding the hyperbolic discounting model. It can also be
observed that Rubinstein’s results contradict the subadditive model of discounting,
which predicts a greater one-day discount rate.

8.6 Empirical evidence

There are three main sources of empirical evidence regarding inter-temporal

decision-making models: (1) behavioral studies; (2) evolutionary biology; and (3)
neuroeconomic studies. Before examining these various sources we should bear in mind
certain important principles in terms of how this empirical evidence can be interpreted.
First, it is necessary to realize that some theories described so far are not mutually
exclusive, but may instead be complementary. This applies to hyperbolic discounting in
particular, which can be combined for example with multiple-self models. On the other
hand, while ‘planner-doer’ models may be compatible with visceral factor models,
these are both mutually exclusive with hyperbolic discounting. A second and related
principle involves reductionism: different theories apply at different levels. Hyperbolic
discounting is a purely economic ‘explanation’; underlying it may be a psychological
explanation related to psychological connectedness. Similarly we shall see that dual-self
models may be reliant on psychological evidence related to ‘hot’ and ‘cold’ decision-
making processes, which in turn may be reliant on neurological studies relating to brain

310
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

systems. A third point involves confounds: some studies provide evidence that does not
distinguish between different theories, meaning that the evidence can be explained by
different theories, even though they are in principle contradictory. Studies that provide
evidence that can distinguish between theories are particularly important.

Behavioral studies

We have already mentioned a number of studies that provide evidence for hyperbolic
discounting. Some of this research has also estimated parameter values for the (,)
model. DellaVigna and Malmendier (2006) study health club membership, and find
that a model involving  = 0.70 and  = 0.9995 fits their data. The Laibson, Repetto
and Tobacman (2009) study estimated  = 0.70 and  = 0.96. Paserman (2008) uses job
search data to estimate  = 0.40 and  = 0.99 for low-wage workers and  = 0.89 and
 = 0.99 for high-wage workers, assuming sophistication. Skiba and Tobacman (2008)
allow for partial naïveté, estimating  = 0.53, b = 0.90 and  = 0.45. It is notable that
borrowers of pay-day loans have a very high discount rate, otherwise the enormous
interest rates on such loans would not be acceptable.
The problem is that these studies were typically designed to test the theory of
nonconstant hyperbolic discounting versus constant exponential discounting; they
were not designed to test hyperbolic discounting against other behavioral theories, like
visceral influence models that modify the instantaneous utility function, and therefore
cannot discriminate between the two theories. Thus when we observe a preference
reversal, like switching from an LL reward to a SS reward, we need to ask if this
is because of a change in the discount factor being used or because of a change in
perceived utility of the reward. Alternatively, perceived probabilities or attitudes to risk
may have changed. This means that evidence of preference reversal is not necessarily
proof of nonconstant discounting (Gerber and Rohde, 2010).
Examples of research that discriminates between different theories are the studies
by Mischel (1974) and Mischel, Shoda and Rodriguez (1992). Children were placed in
a room by themselves and taught that they could summon the experimenter by ringing
a bell. They would then be shown a superior and inferior prize and told that they
would receive the superior prize if they could wait successfully for the experimenter
to return. One main finding was that the children found it harder to wait for a delayed
reward if they were made to wait in the presence of either one of the immediate
or delayed reward objects. This finding is particularly important since it provides
evidence for the visceral factor theory as against the nonconstant discounting theory.
According to the latter children should be more willing to wait in the presence of the
superior delayed reward. This result does not indicate that the hyperbolic discounting
theory is false in general, merely that it may not be the best explanation of behavior
in certain situations.
A number of behavioral studies also provide evidence of increasing costs of self-
control, meaning that our facility for self-control is a limited resource, and becomes
increasingly costly to utilize as cognitive load increases. An example of this phenomenon
is described in the paper by Fudenberg and Levine (2006), which reports an experiment
by Shiv and Fedorikhin (1999), where subjects were asked to memorize either a two- or a
seven-digit number, and then walk to a table with a choice of two desserts, chocolate cake
and fruit salad. In one treatment the actual desserts were on the table, whereas in a second
treatment the desserts were represented by photographs. It was hypothesized that:

311
PT • I I I I NT ERT EMPO R A L C H O I C E

1 Subjects would face a self-control problem regarding the cake, in the sense that
it would have a higher emotional or visceral appeal, but be less desirable from a
‘cognitive’ viewpoint.
2 Subjects’ reactions were more likely to be determined by emotional reactions
when cognitive resources were constrained by the need to remember the longer
number.
3 The ‘cognitive overload effect’ would be greater when subjects were faced with
actual desserts rather than with their pictures.
All three hypotheses were supported by the experimental results. When faced with real
desserts, subjects who were asked to remember the longer number chose the cake 63%
of the time, while subjects given the two-digit number only chose the cake 41% of the
time, a statistically significant difference. However, when faced with pictures of the
desserts, the choices were 45% and 42% respectively, an insignificant difference. These
results can also be explained by the dual-self model, in that the long-run self, faced with
increasing cognitive costs, is less well able to exert self-control. The FL paper also notes
that the increasing marginal cost of self-control implied by the Shiv and Fedorikhin
study contravenes one of the axioms proposed by Gul and Pesendorfer (2001) in
relation to self-control, specifically the axiom relating to set-betweenness. This axiom
was discussed in the context of EUT in Chapter 5, where we saw that evidence did not
support it in that context either.
More recent studies by psychologists have generally supported the theory of
increasing costs of self-control, for example Gailliot et al. (2007), Vohs et al. (2008),
Burger, Charness and Lynham (2011), Fedorikhin and Patrick (2010), Usta and Häubl
(2010) and Bucciol, Houser and Piovesan (2011). Bucciol, Houser and Piovesan (2011)
find that exposure to temptation reduces the productivity of young children, aged 6 to
13, but not older children. This is in keeping with the findings of the Mischel studies
described earlier.
However, the empirical evidence regarding the costs of self-control and its effects
is sometimes surprising. Burger, Charness and Lynham (2011) study procrastination
by students, and report two main findings. First, unlike previous studies, they find that
the imposition of interim deadlines for a fairly long-term project (five weeks) does not
improve performance in terms of completing a given task. Second, they find that in
the short term over a two-day period, exposure to temptation reduces productivity in
the first day, but actually increases the probability of completing a task over the whole
two-day period. The authors suggest that this may be due either to the self-signaling
effect of exerting willpower, or to the suffering on the first day creating a commitment
to persevere and ‘see things through’ on the second day.
Fedorikhin and Patrick (2010) find that although positive mood generally facilitates
resistance to temptation as far as healthy food choices are concerned, any emotional
arousal accompanying this mood can reduce this resistance by increasing the cognitive
load. For example, the study indicates that watching an exciting video clip while in a
positive mood is more likely to be associated with choosing M&Ms as an unhealthy
snack rather than grapes as a more healthy option. Thus it is not just the kind of mood,
but also the intensity of feeling, that affects choice.
Another finding that is perhaps counter-intuitive comes from a recent study
by Usta and Häubl (2010). One might expect the delegation of decision-making to
others, such as physicians or financial advisers, to reduce the cognitive load and stress

312
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

factor. However, the study finds that such delegation, while reducing actual decision-
making effort, depletes self-regulatory resources and impairs subsequent ability for
self-control. Even recalling past episodes of such delegation has the same effect. The
authors propose that the reason for this is that delegation of decision-making poses
a threat to self-esteem, in terms of viewing oneself as a free agent. In support of this
reasoning, the study finds that the depletion of self-regulatory resources does not occur
when subjects are given the opportunity to affirm their sense of free agency.
The Gailliot et al. (2007) study takes a reductionist approach, explaining the
phenomenon of increasing costs of self-control in physiological terms: self-control relies
on glucose (as do brain processes in general), and this is a limited energy source. The study
showed that acts of self-control, like coping with thoughts of death or stifling prejudice
during an interracial interaction, reduced glucose levels, and this impaired self-control on
subsequent tasks. Furthermore, consuming a glucose drink eliminated these impairments.
This kind of reductionist approach, coupled with neuroeconomic studies, is an
important aid in discriminating between different theories related to intertemporal
decision-making, and in particular achieving an understanding of time-inconsistent
preferences. As can be seen from the studies surveyed above, much further research is
necessary in this area in order to clarify a number of current issues.

Evolutionary biology

At first sight it might seem that basic neo-Darwinian principles in evolutionary biology
would favor the evolution of a brain that is a cohesive entity. This would not necessarily
rule out the possibility of brain modularity with different systems for different functions,
but one might suppose that these systems would not be in conflict, as in multiple-self
and dual-self models. However, there is now extensive evidence that such conflicts
occur in various areas such as memory, information processing and motivation. The
approach of Brocas and Carrillo (2008a,b), and the model of the brain they propose
to account for the three conflicting areas of asymmetric information, temporal horizon
and incentive salience, in many ways parallels the tradition in economics of modeling
the firm as a nexus of agents with conflicting objectives. Yet this raises the fundamental
issue: why would the process of natural selection favor the evolution of a brain with
such inbuilt conflicting systems?
Different scientists from different disciplines have approached this problem from
different angles, exhibiting a remarkable degree of consilience. The ethologist and
evolutionary biologist Dawkins (1976) argues that selection occurs primarily at the
level of the gene, and this is bound to result in conflicts at the level of the individual.
Related to this is the fact that most of the genetic material within our bodies is actually
‘foreign’, for example the bacteria in the gut, without which we could not survive. The
evolutionary psychologists Cosmides and Tooby (1992) argue that many of our internal
conflicts are a consequence of our evolutionary past, and are no longer adaptations,
in the same way as our taste for sugary, fatty foods is no longer adaptive for humans
living in developed societies. The economist and dietitian de Vany (2011) argues along
similar lines, in particular proposing that the brain’s demand for glucose is often in
conflict with the tendency of the pancreas to release more insulin which reduces the
glucose level in the blood, storing it in muscle and fat tissues rather than letting it
go to the brain. He attributes the conflict to the modern over-availability of simple
carbohydrate, high glucose foods that send our insulin soaring and then crashing.

313
PT • I I I I NT ERT EMPO R A L C H O I C E

De Vany credits his ideas in turn to a work by Peters et al. (2004), which presents
a neuroeconomic model of competition for energy resources. Wang and Mariman
(2008) focus on the physiological consequences of this competition, proposing that the
modern tendency towards insulin resistance, leading to type-2 diabetes, obesity and
associated health problems, is a result of the brain’s strategy to protect its supply of
glucose. However, the result of this internal competition need not necessarily be bad.
Livnat and Pippenger (2006) show that competition between different sub-systems
may actually lead to improved biological outcomes, just as in a free market economy
competition tends to improve welfare.
Moving on to more specific aspects of evolutionary biology, it might at first appear
that natural selection would favor constant exponential discounting over hyperbolic
discounting and decreasing impatience. After all, how could time-inconsistent
preferences serve as an adaptation in terms of improving our prospects of survival
and reproduction? This is an issue where much research is still needed to clarify the
situation. Robson and Samuelson (2009) propose that the existence of aggregate
uncertainty can result in nonconstant discounting and present bias with decreasing
impatience; they claim this may be a more basic phenomenon than preference
reversals. Aggregate uncertainty refers to systematic uncertainty where all individuals
are faced with the same uncertainty, like the possibility of an earthquake or a flood.
Unsystematic or idiosyncratic uncertainty is individual-specific, like the possibility of
being attacked or robbed. Robson and Samuelson argue that evolution has caused
us to have the belief that idiosyncratic uncertainty is controllable, since we can often
take steps to avoid these kinds of danger. Therefore we tend to have a greater fear of
uncontrollable aggregate uncertainty. However, they conclude that although this can
account for present bias, it should not lead to preference reversals.

Neuroeconomics

Studies in the 1990s by Damasio (1994), LeDoux (1996) and Bechara et al. (1999)
tended to support the psychological theories of dual systems of decision-making as
far as temporal horizon is concerned. Damasio showed that patients with damage in
the ventromedial prefrontal cortex had an impaired ability to engage in long-term
planning while appearing emotionally flat, meaning that events that would normally
make people happy or sad did not seem to register with them emotionally. LeDoux
and Bechara both provided evidence that the amygdala played a crucial role in the
expression of impulsive and emotional behavior. Bechara (2005) has also gone on to
distinguish between an impulsive system (mainly the ventral striatum and amygdala)
that processes information about immediate rewards, and a reflective system (mainly
the ventromedial and dorsolateral prefrontal cortex and anterior cingulate) which
processes information about future rewards. This in many ways parallels the ‘planner-
doer’ model of Thaler and Shefrin (1981) in terms of principal and agent, and is also
similar in this way to the model proposed by Brocas and Carrillo (2008b).
McClure et al. (2004) take the analysis one step further. Using the fMRI
technique, they found that decisions relating to immediately available rewards involve
the preferential activation of parts of the limbic system associated with the midbrain
dopamine system, including the paralimbic cortex. In contrast, with intertemporal
choices generally, regions of the lateral prefrontal cortex and posterior parietal cortex
are engaged uniformly. The study also found that:

314
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

The relative engagement of the two systems is directly associated with the
subjects’ choices, with greater relative fronto-parietal activity when subjects
choose longer term options (p. 503).
The authors hypothesize that:
short-run impatience is therefore driven by the limbic system, which responds
preferentially to immediate rewards and is less sensitive to the value of future
rewards, whereas long-run patience is mediated by the lateral pre-frontal
cortex and associated structures, which are able to evaluate trade-offs between
abstract rewards, including rewards in the more distant future (p. 504).
McClure et al. (2004) conclude that the interaction between short-sighted and far-
sighted systems provides neuroscientific support for hyperbolic discounting.
The evidence in this study is also supported by comparisons with advanced
primates, who have substantially smaller prefrontal cortexes than humans, and with
subjects with prefrontal brain damage. In both cases individuals are heavily influenced
by the availability of immediate rewards, and are unable to delay gratification or
plan ahead.
However, the conclusions of McClure et al. (2004) have been recently challenged
by Glimcher, Kable and Louie (2007) and Glimcher (2009). They argue that there is
no compelling evidence for concluding that the brain is divided into emotional and
rational areas, and that in monkeys research has shown that activity in the posterior
parietal cortex predicts preferences for both immediate and delayed rewards, with no
suggestion of the existence of dual systems. It should be noted that Glimcher does not
doubt the notion that emotions affect decision-making; he suggests, like Bechara, that
the amygdala may be involved in this respect. Obviously, once again, more research is
needed at the neurological level to clarify the situation regarding the role of different
brain areas in intertemporal decision-making. This is a very important priority in terms
of the direction of future research.

8.7 Policy implications

There are a number of normative aspects as far as policy implications are concerned
that arise from the models of intertemporal choice discussed here. These relate to
individuals, firms and governments.

Individuals

We have seen that the main implication of the various models presented in this
chapter is that people have self-control problems, meaning dynamic conflicts over
time causing preference reversals. These problems in particular relate to temptation,
where we switch from preferring a larger benefit later to a smaller benefit sooner, and
procrastination, where we switch from preferring a smaller cost sooner to a larger
cost later (Tice and Baumeister, 1997). In both cases we come to regret our decisions
later in retrospect, because we are not optimizing our behavior. 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

315
PT • I I I I NT ERT EMPO R A L C H O I C E

they would have paid if they had chosen a different option of paying for ten-visit blocks,
because they overestimated their gym usage. Furthermore, the study found evidence
that these members did not learn from their mistakes, by changing their membership
plan later once they realized their low usage; instead they continued with their monthly
plans. They were also slow to cancel their monthly memberships, with an average delay
of over two months between last gym visit and cancellation, losing even more money.
This situation is discussed in more detail in Case 8.1.
Failure to optimize behavior also arises in situations where people tend to
underestimate usage. Miravete (2003) examined people’s choice of telephone calling
plan when South Central Bell changed their tariff structure. The new structure involved
either paying a flat monthly rate, or paying a fixed rate plus call charges. Again the study
finds that many people chose the wrong option, but in this case people tended to be
quicker to learn from their mistakes. Whereas 40% of people initially chose the wrong
option, paying too much, two months later this proportion was reduced to 33%. Not
all studies of situations where naïve consumers underestimate usage present such an
optimistic picture. Heidhues and Köszegi (2010) find that naïve consumers overborrow
on both credit cards and subprime mortgages, where the baseline repayment terms are
cheap. However, there are large penalties for delaying repayment, and the study finds
that these consumers end up paying the penalties, thus suffering large welfare losses.
It is important to realize that the issues above related to present bias and
optimization do not arise with the constant exponential discounting of the DUM. With
the constant discounting model, as stated earlier, there is no conflict between short
term and long term: one path of action will at all times seem preferable. In the example
where one yields to the temptation of having the dessert one had previously not
intended to have, according to the DUM one will at all times either want the dessert or
not want it; there is no scope for the exertion of willpower. If the discounted benefits of
eating the dessert exceed the discounted costs one will indulge, and if they do not then
one will not indulge. The important implication here, discussed extensively by Ainslie
(2001) in his book Breakdown of Will, is that in the DUM the concepts of temptation
and willpower are redundant. Of course, present bias is not the only factor related to
non-optimization in some cases; overconfidence and self-serving bias are also relevant.
The role of self-awareness is always important in good decision-making, and it is
particularly important in the context of self-control situations. Psychological conflict
arises because the subject will normally remember that their preference in the past was
different, and, if the subject is sufficiently self-aware, they will also realize that in the
future they will come to regret their action if they indulge, because from that future
standpoint, discounted costs exceeded discounted benefits.
The main implication of self-awareness is that people will make commitments to
prevent them from taking later actions that fall into the category of ‘vices’. We have
already seen some examples of this, relating to buying an alarm clock that is difficult
to turn off, and buying a small bag of chips rather than a large one. Plentiful evidence
of such commitments provides a ‘smoking gun’ as far as hyperbolic discounting, time-
inconsistent preferences and self-awareness are concerned. Ancient references to such
commitment devices include the story of Ulysses ordering his shipmates to tie him
to the mast so that he could listen to the song of the sirens without being lured onto
the rocks and shipwrecked. Burnham and Phelan (2001) provide some off-the-wall
modern examples: smearing one’s brownie with mayonnaise when given lunch on a
plane, and posting one’s internet cable to oneself. Both actions are designed to prevent

316
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

later indulgence when preferences have changed, but obviously they are only possible
if a certain degree of self-awareness is present. Other common forms of commitment
involve the use of whole life insurance policies and illiquid savings accounts, paying
health club memberships for a year where there is no refund or cancellation option,
leaving one’s credit card at home when going shopping, or using a debit card instead of
a credit card (King and King, 2011).
A good example of commitment from sophisticated subjects concerns an
experiment conducted by Ariely and Wertenbroch (2002). This involved executive
education students at MIT, who had to write three papers in a semester for a particular
class. One group was given evenly spaced deadlines throughout the semester for the
three papers, while the other group was allowed to select their own deadlines. The
penalty for an overdue paper was the same in each case. Although it was possible for
the second group to have made all their papers due at the end of the semester, in the
experiment many did in fact commit to spacing out their deadlines. It was also notable
that those who did have evenly spaced deadlines, whether externally or internally
imposed, performed better than those who did not. Thus it appears that the more
sophisticated subjects, who foresaw self-control problems, made the commitment
involving evenly spaced deadlines, and as a result improved their welfare.
As already stated, people generally lie somewhere in the middle of the spectrum
of self-awareness, and O’Donoghue and Rabin (2001) introduced a model of partial
self-awareness that accounts for various aspects of observed behavior, although it
was specifically designed to account for the phenomenon of procrastination. In this
model people are aware that they will have self-control problems in the future, but they
underestimate the magnitude of the problems. The authors observe that, when people
choose from a menu of options involving costs and benefits at different times in the
future, they may now eschew an option involving immediate action and relatively small
benefits in favor of an option involving action and greater benefits in the long term.
However, later on, they may forsake the latter option in favor of another option that
involves action and even greater benefits still further in the future. Thus preferences
may constantly shift, with actions continuously being delayed. For example, we may
decide not to tidy the garage this week, because we plan to redecorate it next month.
Next month we may decide that redecoration of the garage is not as important as fencing
the garden. While this model cannot explain all types of procrastination, for example
where a given task is continuously delayed, it does have important policy implications.
The same authors have investigated other causes of procrastination (O’Donoghue
and Rabin, 2008), and find that procrastination is more likely in multi-stage projects
when the costs of completing the different stages are more unequal, and in particular
when the later stages are more costly. Furthermore, if the cost structure is endogenous,
people are prone to choose cost structures with lower costs early on, but this is likely to
cause them to start but not finish projects. Commitment to finish rather than just start
projects is an important policy implication in this case.
Another interesting example of commitment has been researched by Frank and
Hutchens (1993), and involves the preference for rising wage profiles. This preference
has been discussed earlier, and is also the subject of Case 8.3. It is argued that the
main pressure for rising wage and consumption profiles is a social one, and that if
the workers involved can commit to restricting their wages in the short term in return
for the promise of higher wages later in their careers, they can improve their welfare.
Although the authors admit that their evidence is insubstantial quantitatively, it is highly

317
PT • I I I I NT ERT EMPO R A L C H O I C E

suggestive. To our knowledge there has been no extensive study of the phenomenon
discussed by Frank and Hutchens, and it is probable that the primary reason for this is
the lack of available data for a wide range of professions.
All of the examples of commitment described to this point involve deliberate
actions. There have also been a number of studies, particularly in the psychology
literature, where commitment does not involve such action, but instead involves
the automatic involvement of the emotions. Although economists such as
Becker and Akerlof had broached this subject earlier, Hirshleifer (1987) and Frank
(1988) were the first economists to develop a formal theory of the emotions as
commitment, based on game theory. Since that time there has been considerable input
from various fields, including political scientists, psychologists and neuroscientists.
Until the work of Hirshleifer and Frank there had been much puzzlement regarding
the usefulness of emotions such as anger, envy and hatred, all of which are commonly
viewed as ‘negative’ emotions, in contrast to the ‘positive’ emotions such as love, joy
and pride. Negative emotions often caused people to take actions that were self-
destructive, and therefore they seemed to serve no Darwinian purpose, meaning that
it was difficult to see why they had evolved. The main contribution of Hirshleifer and
Frank was to propose a theory that such emotions served as a credible commitment,
deterring others from taking advantage of us. Thus, although such emotions might
cause short-term harm (and long-term harm in individuals where they are excessive),
in most individuals they serve to promote our long-term welfare.
A number of misunderstandings tend to occur regarding the Hirshleifer–Frank
model. First of all, there is a difference between the capacity to feel an emotion and
the actual feeling of the emotion. As Elster (1998) has observed, an irascible person
may rarely feel anger, because others may take care not to provoke the person’s anger.
To correct another misunderstanding, the theory does not mean that the capacity for
anger is always ‘good’ for us, any more than the capacity for pain is always good for us;
it simply means that on average it has served people well in the past, or in biological
terms it has improved our inclusive fitness. The theory also does not mean that we
should always display anger when it is provoked; sometimes it is better to keep calm,
particularly in view of social conventions.
Finally, it should be noted that emotions would not be necessary as commitment
devices if the DUM and constant discounting were applicable. According to the DUM,
if someone offends us, a ‘rational’ calculation of self-interest will tell us how to react,
in terms of whether to punish the offender, how and when; there is no conflict between
short-term and long-term interests.
Hyperbolic discounting models may describe such situations well in many cases,
but as far as lending insight into their nature and normative aspects, multiple-self
models have a significant advantage. This is because they can highlight the essential
asymmetry involved between the current or myopic self and the meta-self that is at
least partially aware of future changes in preferences. As already stated, the meta-self
can make strategic commitments to constrain the later behavior of the myopic self, but
the reverse cannot happen. For example the person who finds it difficult to rise in the
morning may place the alarm clock on the other side of the room to ensure that the
myopic self does not simply slam it off immediately it rings the next morning, but the
myopic self cannot reply in any strategic manner to such a commitment.
Commitments can be either external or internal. External commitments, once
made, become less controllable by the individual, and therefore are most effective

318
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

because they do not depend so much on the person’s willpower. Putting money into
an illiquid life insurance policy is an example. Some people even using blogging as
an external commitment, since public announcements of intention involve a loss of
face if the person then reneges on their commitment. The disadvantage of external
commitments is that they lack flexibility if a person’s circumstances change. Internal
commitments involve making personal or private rules, for example saving 10% of
one’s salary every month, but these are more vulnerable to temptation.
For individuals, therefore, the key to maximizing experienced utility over the long
run may be the use of appropriate strategic commitment devices that constrain the
future desires of the myopic self. It is also important, as we have seen, that the meta-
self is aware of the changing preferences that inevitably occur. Of course, it will never
be possible for any meta-self to have rational expectations to the extent that all future
preferences will be accurately predicted; however, the more able people are to learn
from past experience regarding such changes, the more likely it is that they will be
successful in anticipating future changes and conflicts, and taking appropriate action.
A further implication of the fact that agents have only incomplete information
about future preferences is that abstinence may be a better policy than moderate
consumption, as illustrated in Case 5.1. In situations involving the possibility of
addiction, such as gambling, smoking or drinking, an abstinence rule, though a second-
best rule, can act as a commitment device against inefficient learning that would lead
to future excesses (Carrillo, 2004).
However, there is a final twist to this situation that we have also touched on earlier,
and this relates to self-signaling. When the self-control problem is repeated, as is often
the case, a yielding to temptation (or ‘defection’) in the first round can lead to a loss
of self-confidence, thus making defection more likely in the next round and so on.
Thus a far-seeing self may envisage the likely succession of failures if too harsh a rule
is made initially, and decide instead to adopt a less strict policy as far as commitment
is concerned.
It is therefore difficult to draw definite conclusions regarding how individuals
should make commitments in self-control situations. The main general conclusion is
that those agents who know themselves, and can predict their future selves, best are
also best able to maximize their own welfare in terms of experienced utility.
One other aspect of self-control problems that has important policy implications
concerns the effects of diminishing marginal sensitivity and shifting reference points.
For controlling food intake there are a number of relevant factors in particular:
1 There is a multi-sensory system in operation, involving visual, olfactory and gustatory
senses. The interdependence of seeing, smelling and tasting foods is not well-known at
present, but it may be possible that if just one of these senses suffers from diminishing
sensitivity in eating a meal, this may override the other senses. On the other hand
diminishing sensitivity may only occur when all three senses are affected.
2 The phenomenon of diminishing marginal sensitivity is highly specific to different
foods (Epstein et al., 2009). People who eat meals with more varied foods, or whose
intake in general is more varied, tend to eat more in total. This is especially true of
sugary and fatty foods, where a large number of convenient snacks are available. When
we eat a monotonous diet we eat less as diminishing sensitivity sets in much earlier.
3 People vary as to how much they can consume before diminishing sensitivity
occurs. There is evidence that obese people are not affected until larger intakes

319
PT • I I I I NT ERT EMPO R A L C H O I C E

are consumed (Epstein et al., 1997; Temple et al., 2007). This may be because
obese people have shifted their reference points to higher levels, or it may be an
independent genetic factor in operation.
4 Diminishing sensitivity is delayed when people perform other activities while
eating, like watching TV or reading a book (Epstein et al., 1997, 2005; Temple et
al., 2007). This may be because these environmental distractors affect memory.
People whose memories are impaired, such as amnesiacs, can eat one meal
immediately after another.
There is a need for further research relating to these factors affecting diminishing
marginal sensitivity, but some of the policy implications for people suffering from self-
control problems with food intake are obvious.

Firms

Many policy implications for firms are the flip-side of those for individuals, since firms
may be able to exploit the weaknesses of naïve consumers in particular, increasing
their profits at the expense of consumer welfare. However, the picture is not quite so
simple as this, since firms’ policies aimed at the exploitation of naïve consumers may
actually in turn be exploited by sophisticated consumers and benefit such consumers,
as we shall see.
In general there has not been as much research relating to policy implications for
firms as there has been for individual agents or governments. However, DellaVigna
and Malmendier (2004) have shown that dynamic inconsistencies and other anomalies
regarding timing of rewards and payment have important implications for contract
design in the case of both investment goods and leisure goods. Investment goods
are defined as those where there are immediate costs, in terms of money and effort,
and delayed benefits, for example health club memberships. Leisure goods involve
immediate benefits and delayed costs, such as credit card financing. DellaVigna and
Malmendier (2004) construct sophisticated models of consumer and firm behavior, the
mathematics of which are omitted here, and investigate various possibilities according
to the degree of naiveté of consumers and different market conditions. They find that
empirical evidence from various industries confirms the predictions of the models. The
authors summarize their findings in terms of three main implications:
1 Firms should price investment goods below marginal cost.
Naïve consumers tend to overestimate usage of such goods, and therefore
overestimate the value of the discount on marginal cost. For example, in the health
club industry it is common to have a zero marginal cost for users, who mainly pay
annual or monthly fees (DellaVigna and Malmendier, 2003). Price discrimination
cannot explain such a practice, since more frequent users with less elastic demand
would be charged a higher per-usage price, and this type of strategy by firms is not
commonly observed. As far as sophisticated consumers are concerned, they can
use the high initial cost as a form of commitment. This situation is examined in
more detail in Case 8.1.
2 Firms should price leisure goods above marginal cost.
In this case naïve consumers underestimate future usage, for example credit card
financing. They are therefore attracted by offers that have favorable initial terms.
Sophisticated consumers can take advantage of this by paying off their outstanding

320
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

balances each month, and therefore borrowing for up to six weeks free of charge,
since many card companies do not charge annual fees. Mobile phone companies
have similar charging schemes, with free minutes per month, but high charges for
excess time. Naïve consumers may be attracted by the free minutes, but tend to
underestimate their phone usage and may therefore end up paying high monthly
bills. Mail order firms use similar attractive offers with free books or CDs, but
high charges for additional items. DellaVigna and Malmendier (2004) note a
slightly different strategy in the gambling industry. In this case hotels, notably in
Las Vegas, charge attractive low rates for accommodation and dining, since naïve
gamblers underestimate their gambling activity and losses. Thus for the hotels the
gambling activity subsidizes their core business. Again, sophisticated consumers
can take advantage of this strategy by staying and dining in the hotels, but take in
shows or play golf rather than gamble.
3 Firms should charge back-loaded fees and introduce switching costs for all goods.
It is common for credit card companies to have introductory or ‘teaser’ offers,
like zero interest rate charges on balance transfers for limited periods like six
months. After the initial period the interest rates usually rise very significantly,
typically to about 10% above prime or base rates. As we have already seen in the
study by Heidhues and Köszegi (2010), such a strategy is profitable since naïve
consumers underestimate the amount of their borrowing after the teaser period is
over (Ausubel, 1999). Switching costs relate to the costs, both in money and effort,
of either switching to a new provider or cancelling the agreement. For example
health clubs typically offer automatic renewal, and only allow members to cancel
their memberships in person or by letter, rather than by email or by phone. The
result is that users tend to remain members longer than otherwise. DellaVigna and
Malmendier (2004) found that there was an average period of over two months
between a member’s last usage and the cancellation of their membership.
DellaVigna and Malmendier also draw conclusions regarding the welfare effects of these
policy implications. They observe that, for sophisticated consumers, market interactions
need not reduce their welfare. In fact they may gain if they are in effect being subsidized
by naïve consumers, as is the case with credit card financing. In addition, market
mechanisms encourage firms to create commitment devices that allow sophisticated
consumers to increase their long-run welfare, for example by investing in life insurance
policies. However, for naïve consumers who have non-rational expectations, two adverse
welfare effects are noted. First, there is an overall reduction in efficiency in terms of
net surplus to consumers and producers. Second, in monopoly there is a redistribution
of surplus from consumers to producers, who are able to take advantage of the lack of
consumer awareness to increase their profits. In perfect competition the second effect is
eliminated, but this situation rarely arises in reality.
These adverse effects on the welfare of naïve consumers also have implications for
government policy. This general aspect is now discussed.

Governments

The models of intertemporal choice presented in this chapter have significant

implications for a number of areas of government policy. The following aspects are now
discussed: (1) incomplete self-knowledge, (2) addiction, (3) savings, (4) investment, (5)
social security, (6) social projects and (7) environmental policy.

321
PT • I I I I NT ERT EMPO R A L C H O I C E

1 Incomplete self-knowledge
DellaVigna and Malmendier (2004) have shown that naïve consumers are not able
to maximize their welfare, even if there is perfect competition. There is a role for
a paternalistic government to intervene in such situations, if it can obtain more
information regarding the future preferences of these consumers than the consumers
themselves. Heidhues and Köszegi (2010) recommend the prohibition of large
penalties for deferring small amounts of repayment, and recent regulations in the
US credit card and mortgage markets have indeed moved in this direction. While
this may be possible in some circumstances, the information requirements for policy
intervention are large, and even then intervention may not be a complete remedy. It
may well be that, as DellaVigna and Malmendier recommend, the best policy is to
educate naïve consumers as far as possible regarding their lack of self-awareness.
2 Addiction
Addiction is, of course, one specific area where incomplete self-knowledge is
relevant. However, there are certain tax implications in this case that have been
ignored by the main body of public finance literature. Of particular interest here is
a study by Gruber and Köszegi (2001), which arrives at two main conclusions. First,
there is evidence that smokers are forward-looking in their smoking decisions, in
that announced but not yet effective tax increases lead to both increased sales
and reduced consumption. This apparently contradictory behavior implies that
smokers are rational to the extent of buying more while the price is lower, while
at the same time reducing consumption in anticipation of the price rise. Second,
given the empirical evidence regarding time-inconsistent preferences, there is a
justification for basing taxes not only on the external costs imposed by smokers
but also on the internalities that they impose on themselves. Internal costs for
smoking are far greater than external ones, with the study estimating that a pack
of cigarettes costs $30.45 in terms of lost life expectancy. The authors estimate that
optimal internality taxes are probably at least $1 per pack in the US.
3 Savings
Naïve individuals tend to overestimate their ability to save for the future and do
not take advantage of available commitment devices to help them save more.
Laibson (1997) has argued that as new and liquid financial instruments have
proliferated since the 1980s, due to the deregulation of banking systems in
various countries, this problem has been aggravated. While deregulation may
have increased competition and efficiency, one undesirable result has been that
many automatic commitment devices in the form of illiquid savings instruments
have disappeared. The policy implications are further complicated by changes in
mandatory retirement laws in many countries. Diamond and Köszegi (2003) use a
multiple-self model to argue that recent changes may cause people to save more
in order to retire earlier. Governments, however, often want people to retire later
in order to reduce the financial burden on public finances. Trying to encourage
people both to save more and to retire later is a major problem facing many
governments. This area of government policy is discussed further in Case 8.2.
4 Investment
We have seen that time-inconsistent preferences can lead to procrastination.
Entrepreneurs who are sophisticated in terms of being aware of this tendency
may make the commitment of foregoing free information in order to avoid

322
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8

procrastination and invest now. Brocas and Carrillo (2004) argue that this
phenomenon may lead to excessive investment in the economy and entry mistakes.
They further argue that government intervention, forcing investors to acquire
information before making investment decisions, may reduce interest rates and
lead to an overall improvement in welfare in the economy as a whole.
5 Social security
People receiving social security benefits, in cash or in other forms, tend to receive
these benefits on either a weekly or monthly basis. Cash benefits are usually paid
weekly. However, the food stamp program in the US operates on a monthly
basis. Providing benefits in the form of food stamps is in itself an automatic
commitment mechanism, since it prevents recipients from using the benefits to
buy goods regarded as undesirable by the government. However, given a self-
control problem and a lack of other forms of commitment, monthly provision
may lead to excessive consumption in the first part of the month. Shapiro (2005)
provides evidence that caloric intake declines by 10% to 15% over the food stamp
month, providing further evidence for time-inconsistent preferences and against
the permanent income hypothesis. It may therefore be that this program would
improve welfare if it operated on a weekly basis, although these improvements
would have to be balanced against the increased transaction costs for both
recipients and government.
6 Social projects
Such projects relate to major infrastructure investments, like building roads,
schools, hospitals, power stations and railways. Governments must determine an
appropriate official discount rate to apply to costs and benefits in order to make
optimal investment decisions. Evans and Sezer (2004) observe that countries have
used very different approaches in this area. For example, Germany bases its 3%
real rate on financial market data, while France has applied an 8% real rate based
on the marginal product of capital. In 2003 the UK switched from a 6% real rate,
based mainly on the cost of capital, to a 3.5% real rate based entirely on social time
preference, which Evans and Sezer argue is the appropriate rate to use. For public
investment decisions it might initially seem that time-inconsistent preferences
and self-control problems may not apply, but governments may also be inclined
to place short-run electoral benefits before longer-run budgetary considerations.
In this case they might use high official discount rates. Evans and Sezer examined
official discount rates and estimated social time-preference rates (STPRs) in six
major countries: Australia, France, Germany, Japan, the UK and the USA. The
only major country where the official discount rate is less than the estimated
STPR is Germany, where the 3% official discount rate is less than the estimated
STPR of 4.1%. France appears to have a large discrepancy, applying the rate of
8% whereas the estimated STPR is only 3.5%. Such a policy could lead to severe
public underinvestment, to the cost of future generations.
7 Environmental policy
This is another area where bad decisions can harm future generations. Once again
procrastination is a major problem. There are on the one hand good reasons for
waiting until reliable evidence is available before making major global decisions
that could impose large and immediate costs. However, some authors have
argued that time-inconsistent preferences have been one of the main problems

323
PT • I I I I NT ERT EMPO R A L C H O I C E

of enforcing and enlarging the scope of the Kyoto protocol (Winkler, 2006).
Additional problems are involved here, relating to the ‘tragedy of the commons’
situation, discussed in Chapter 9, and the degree of uncertainty. It is argued by
Newell and Pizer (2003) that this uncertainty also relates to appropriate future
interest rates used for discounting purposes, and that such rates should be much
lower than the present rate, being only a half of this rate in a hundred years. These
authors estimate that such a procedure would nearly double the net present value
of the benefits of environmental protection policies.
8 Recreational drugs
Historically US laws have been very strict regarding sales of such drugs, imposing
severe penalties, but not so strict regarding possession. This policy may have a
perverse effect on behavior, as noted by Fudenberg and Levine (2006). Severe
penalties have the effect of increasing the fixed cost of making a transaction, causing
consumers to buy larger quantities in each transaction. Such stockpiling is likely
to lead to greater consumption, as we saw in the hypothetical example involving
potato chips. Fudenberg and Levine, along with other economists, have therefore
recommended legalization of such ‘temptation’ goods, combined with a high excise
tax, similar to the policies used in many countries in relation to cigarettes. Such a
policy provides a greater incentive to reduce consumption of harmful products.
A recent approach to government policy in general has been outlined by Thaler and
Sunstein (2008), in the term ‘nudge’, which is the title of their book. This was briefly
discussed in Chapter 6, but the implications are particularly important for intertemporal
decisions. The authors describe their approach as libertarian paternalism, where
people are not forced into making decisions of a certain type, but are ‘nudged’ by
the framing of the decision situation into making decisions that later, in retrospect,
they believe were in their best interests. They suggest that this is especially important
in the areas of health care and saving for retirement. In particular they are critical
of recent legislative changes in US health care on a number of grounds, such as the
randomization of default settings for choices and the offering of too many choices,
which they suggest is confusing to many people. Furthermore, the authors suggest
that the ‘nudge’ approach can be applied to a wide variety of social policies, such as
environmental policy, involving fuel economy and electricity usage, smoking, littering,
teenage pregnancy and filing tax returns. However, not all behavioral economists are
convinced by the ‘nudge’ approach. Some are of the opinion that people should be free
to make their own mistakes. This is essentially a normative issue.

8.8 Summary
• There are many confounding factors involved in the measurement of time
preference.
• Time preference does not appear to be a unitary psychological construct, since it
does not satisfy the three main criteria of constancy, generality and correlation
between different measures.
• It may be more useful to decompose time preference into three main elements:
impulsivity, compulsivity and inhibition.
• Hyperbolic discounting involves applying higher discount rates to time periods
nearer in the future, and lower ones to periods further away.

324
 A LT ERNAT
ERNATIVE
IVV E INTERTEM
INN TE R PORAL C HOIC E M ODELS CH • 8

• Hyperbolic discounting has become widely applied in behavioral economics because

it describes empirical data much better than the exponential discounting of the
DUM. In particular it explains time-inconsistent preferences, which are an anomaly
in the DUM.
• The existence of time-inconsistent preferences is the cause of many self-control
problems. This is particularly true when visceral factors are present.
• The main strategy used by individuals confronted with self-control problems is to
make commitments. This is only possible for self-aware or sophisticated agents.
External commitments involve less willpower than internal ones.
• The concepts of temptation and willpower are redundant in the DUM, since a
particular path of action will always dominate preferences.
• Emotions are an important form of commitment. These are internal, but to a large
extent uncontrollable. The expression of emotion is the controllable element.
• The main criticism of the hyperbolic discounting approach is that it is purely
descriptive, and does not consider underlying psychological factors.
• Alternative behavioral models involve modifying the instantaneous utility function.
Such models also can explain time-inconsistent preferences.
• More radical models involve the concepts of projection bias, mental accounting
models, multiple-self models, dual-self models and the procedural approach.
• Models should not be seen as being mutually exclusive; many can be complementary
to each other, and may operate at different levels of explanation in a reductionist
manner.
• There are a number of policy implications of these behavioral models, for individuals,
firms and governments. These relate particularly to self-control problems and
commitment.

8.9 Review questions

1 Describe the various confounding factors involved in the concept of time preference.
2 Explain the meaning of temptation and procrastination in terms of how they relate
to inconsistent time preferences.
3 Explain how the (,) model of quasi-hyperbolic discounting can explain preference
reversals, using a numerical example.
4 Discuss the advantages and disadvantages of hyperbolic discounting.
5 Compare and contrast hyperbolic discounting and visceral factor models as
approaches to explaining preference reversals.
6 Explain the contribution of prospect theory models to understanding intertemporal
preferences.
7 Explain the nature of multiple-self models and how they address preference reversals.
8 Explain the meaning of the term commitment in connection with self-control
problems, and discuss its role in addressing these problems.
9 Discuss the role of neuroeconomic evidence in understanding and developing
models of intertemporal decision-making.
10 Explain what is meant by the term ‘nudge’ as far as government policy is concerned,
giving examples.

325
PT • I I I I NT ERT EMPO R A L C H O I C E

8.10 Applications

The applications considered here all involve time-inconsistent preferences, self-control

problems and the making of effective commitments. The first two cases involve the
common problems of exercising and saving. The last problem, related to the preference
for rising consumption profiles, also involves social preferences.

Case 8.1 Price plans for gym memberships

We have already mentioned various policy implications following from studies by
DellaVigna and Malmendier (2004, 2006). One particular situation these authors
examined was the optimal pricing structure for firms facing consumers with hyperbolic
preferences for gym memberships. They developed a three-stage model as follows:
Period 1
The firm offers the consumer a membership plan with a membership fee F and a per-
use fee p. The consumer either accepts or rejects the contract.
Period 2
If the consumer accepts the contract, he or she pays F and then makes the decision
whether to exercise (E) or not (N). If the consumer chooses E, he or she incurs a cost
c which relates to the personal effort of exercising, and also pays the firm the usage
fee p. If the consumer chooses N, then there is no cost c or usage fee p.
Period 3
If the consumer chooses E then there is the delayed health benefit b; this is obviously
not received if the consumer chooses N.
The firm incurs a setup cost of K whenever a consumer accepts the contract, and a unit cost
a if the consumer chooses E. The consumer is also assumed to be a hyperbolic discounter
with parameters , b and , as explained in the section on hyperbolic discounting. For
simplicity, the firm is assumed to be time consistent with a discount factor .
For the naïve hyperbolic discounter choosing to exercise, the decision process can be
described as follows:
Period 1
The utility from choosing E is  (b − p − c), and the payoff from N is 0. Therefore
the consumer chooses E if c  b − p.
Period 2
Choosing E only gives a utility of b − p − c, so the consumer actually chooses E
only if c  b − p, a smaller amount than in period 1.
Thus we can see that the naïve hyperbolic discounter, by misinterpreting his or her
own future discounting process, overestimates the net utility of E when buying the
membership. Such consumers choose to exercise less often than they planned to when
buying the membership.
The sophisticated consumers, on the other hand, are under no illusions regarding their
propensity to exercise and correctly predict their choice of E.


326
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8


Assuming profit maximization, DellaVigna and Malmendier predict that, for time-
consistent consumers (with  = 1), the firm simply sets p* (the optimal per-use fee)
equal to the marginal cost a. However, for hyperbolic discounters with  < 1, the
optimal pricing contract involves setting the per-use fee below marginal cost (p* < a),
and the membership fee F* above the optimal level for time-consistent consumers.
There are two reasons for this result:
1 Sophisticated consumers like the lower per-use fee because it serves as a
commitment device for increasing the probability of exercising. They know that they
will be tempted to skip going to the gym unless the per-use fee is low.
2 The higher membership fee allows the firm to exploit the overconfidence of naïve
consumers. They will be willing to pay the higher membership fee because they
overestimate their frequency of usage and the resulting benefits.
DellaVigna and Malmendier also present empirical evidence in support of their model.
They showed that firms in the health club industry typically charged high membership
fees and low, often zero, per-use fees. More specifically, they found that the average
membership fee was about $300 per year. Most gyms also have the option of paying
no membership fee but paying a higher per-use fee (about $15 per visit) instead. The
study found that the average gym member goes to the gym so rarely that their actual
per-use cost works out at about $19 per visit. These consumers would be better off not
buying the membership and just paying on a per-use basis. Therefore this forecasting
mistake allows us to conclude that many gym members behave like they are naïve
hyperbolic discounters.

Questions
1 Compare and contrast the purchasing decision in the health club situation with the
purchasing situation modeled earlier in the chapter relating to buying potato chips.
2 If naïve consumers learn to become more sophisticated, how is this likely to affect
their buying behavior and firms’ strategy in the health club industry?
3 Explain the implications if a health club were to abandon a fixed fee structure and
just charge a relatively low per-use fee of $10.

Case 8.2 The savings problem

Over the last 20 years household savings rates in many of the rich OECD countries have
fallen sharply. The so-called Anglo-Saxon countries – US, Canada, UK, Australia and
New Zealand – have the lowest rates of household saving. Americans save on average
less than 1% of their after-tax income today compared with 7% at the beginning of the
1990s. In Australia and New Zealand personal saving rates are negative, as people
borrow in order to consume more than they earn. The general pattern can be seen in
Figure 8.4.

327
PT • I I I I NT ERT EMPO R A L C H O I C E


Figure 8.4 Trend in household savings rates

Net household saving rates

as % of disposable household income
Italy Japan
Germany United States
France Australia
30

25

20

15

10

5
+
0
-
5
1980 85 90 95 2000 04*
Source: OECD *Estimate

Other countries with rapidly ageing populations, especially Japan and Italy, have also
seen their personal saving rates plummet, though from a higher level. The Japanese today
save 5% of their household income, compared with 15% in the early 1990s. Only a few
of the rich countries, notably France and Germany, have avoided this pattern of reduced
saving. Germans saved around 11% of their after-tax income in 2004, up slightly from
the mid-1980s.
In the US the overall trend in saving masks sub-trends in the components of saving. Evidence
suggests that while saving of high-income earners has proved stable, middle income saving
has collapsed, and low-income earners are increasingly dissaving (Bunting, 2009).
This general trend in the rich countries raises a number of issues:
1 What is the appropriate way to measure a country’s savings?
2 Are rich countries saving enough?
3 What kinds of government policy are effective in encouraging saving?
All of these issues involve certain aspects of behavioral economics, although some
of the aspects are not directly related to intertemporal choice. We will focus on those
aspects that are related to intertermporal choice, observing differences between the
standard model and its behavioral alternatives.

328
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8


The appropriate way to measure savings
The most fundamental point here is that, as far as countries are concerned, it is the total
amount of savings by households, firms and governments that is important. Thus saving
by firms in the form of retained profit, and budget surpluses by governments can in
principle make up for any deficit by households. However, there appears to be at least
some interrelationship between these different categories. A theory called ‘Ricardian
equivalence’ holds that increases in public saving are cancelled out by falls in private saving
as individuals anticipate future tax cuts. An OECD study (Pelgrin and de Serres, 2003) of
16 rich countries between 1970 and 2002 has found that, on average, about half of any
improvement in public finances is offset by lower private saving in the short term, and
about two-thirds in the long term. However, in the US, one of the most extreme cases of low
national saving, the offset was smallest. This raises policy issues discussed later.
As far as the household saving rate is concerned, this is calculated by subtracting
consumption spending from after-tax income. One measurement problem is that the
definitions of both income and spending that statisticians use in the national accounts
often bear little resemblance to what people think of as saving and spending. Realized
capital gains, for instance, are not included in income, even though the taxes paid on
capital gains are deducted from income. There is an aspect of mental accounting that
is relevant here. As seen in Chapter 6, people tend to classify income and wealth into
different accounts and their marginal propensities to spend and save from these different
accounts are also very different. For example, people tend to have a high MPC with current
income, but a much lower one for various categories of wealth, like capital gains. We shall
see that this lack of fungibility has important implications for government policy.

Adequacy of saving
There are both macro and microeconomic aspects of this issue, and both have become
the subject of highly controversial debate amongst economists and policy-makers in
recent years. The macroeconomic aspects relate to the function of saving in the economy
as a whole, and in particular its role in funding investment and stimulating growth. We
are not so much concerned with this issue here, although many economists would
say that, with a current net national savings rate of only 2%, the US economy would
definitely benefit from a boost in saving as far as economic growth is concerned. As a
result of fiscal stimulus and multiple bailouts, the budget deficit for 2010 is estimated
at 9% of GDP, a historical high. Investment tends to be low, and the sustainability of
overseas borrowing is questionable.
The main issue from a behavioral point of view concerns the microeconomic aspects
of saving: are individuals saving enough? In the last decade at least four studies have
suggested that people in the US are not saving enough, while at least another four
studies have suggested that they are saving enough. The reason for the disagreement is
that different studies are based on different assumptions regarding expected earnings,
attitudes to saving, retirement age, desirable levels of consumption during retirement,
government policy and other crucial factors that affect savings adequacy.


329
PT • I I I I NT ERT EMPO R A L C H O I C E


In order to address the issue of savings adequacy we must consider the three main
motives for individual saving:
1 Precautionary – people want to insure against a sudden drop in income.
2 Consumption smoothing – people often wish to consume more than their income
when they are both young and old, and therefore save most in their middle age.
3 Bequest motive – people want to leave assets to their children.
Therefore, the issue whether people are setting aside enough from their current income
depends on assumptions regarding what those people will want to consume or bequeath in
future, what wealth they have already accumulated, and what returns on those assets will be.
In the 1990s many economists argued that in the US individual saving was insufficient, notably
Bernheim (1993). However, more recent studies have argued the opposite case, for example
Engen, Gale and Uccello (1999) and Scholz, Seshadri and Khitatrakun (2003). The last of
these studies concluded that 80% of US households had accumulated adequate saving.
However, the main weakness of these more optimistic studies lies in the assumptions made.
First, they include individuals’ equity in their house as part of their financial assets. Again
the fungibility issue is relevant here. While there is some evidence in both the US and the
UK that increases in property values have fuelled increased consumption, people still do not
treat such wealth in the same way as other forms of wealth. Not only are such unrealized
paper gains subject to reversal, but there is also an endowment effect here; many old
people are reluctant to sell their house to finance their retirement consumption. If only half
an individual’s house equity is included, the most optimistic study suggests that just under
60% of US households have adequate savings.
A second important assumption in the studies mentioned is that future state pension
benefits will be paid as promised. Given the budgetary pressures posed by the baby-
boomers in many countries, a reduction in benefits is quite probable, particularly in the US.
For poorer Americans, any cut in promised pension benefits would significantly reduce the
adequacy of their current saving. Projected payments from social security exceed the value
of all other financial assets for the bottom one-third of the income distribution.
In the UK, where the government’s level of pension provision is set to replace a much smaller
proportion of earnings than in the US, the situation is similar. A recent report by Britain’s Pension
Commission argued that, given downward trends in the occupational pensions provided by
employers and the erosion of state pensions, 60% of workers over 35 are not saving enough.
A third assumption concerns the rate of return on savings. In recent years, the biggest
difference between high-saving and low-saving OECD countries has been the return on assets.
A recent report from the McKinsey Global Institute (Farrell, Ghai and Shavers, 2005) observes
that between 1975 and 2003 asset appreciation was responsible for almost 30% of the
increase in the value of household financial assets in the US, whereas in Japan high saving
rates made up for negative returns on assets. Based on current rates of return and saving
patterns in big industrial economies, the McKinsey study is not optimistic regarding the
adequacy of global wealth accumulation. There is currently much uncertainty regarding
future rates of asset appreciation.


330
 A LT ERNATIVE INTERTEM PORAL C HOIC E M ODELS CH • 8


Implications for government policy
How can governments increase the amount households save? Tighter monetary policies would
certainly help. In the US in particular policy has been loose by most standards for many years,
encouraging borrowing at the expense of saving. Most governments also use tax-incentives to
some extent. The simplest incentive would be to switch from an income-tax structure, where
tax is deducted twice (once from company profit and again when people receive investment
income), to a consumption-based structure. However, governments tend to limit such a switch
because it is regressive in nature, shifting the tax burden from rich to poor.
Some government policies have the effect of reducing saving rather than encouraging it.
For example, in the US eligibility for welfare assistance such as food stamps is phased out
if a couple has assets over $3000. In the UK, the means-tested pension credit, designed
to help pensioners, has the perverse result of making saving for workers on low incomes
an unattractive proposition: for every pound of savings income they can incur marginal tax
rates of at least 40%. However, the new pension system which came into operation in 2011
should address some issues, by incorporating a default of participation in the scheme, with
a 4% level of income contribution (Independent Public Service Pension Commission, 2011).
One major alternative tax incentive has been to shelter retirement accounts, in effect
subsidizing them. In the US the subsidy on retirement-saving accounts is 27% of the value,
amounting to 1% of GDP in terms of foregone tax revenue. There is a debate regarding the
effectiveness of this policy, with some economists arguing that it merely displaces saving
from one form to another, without increasing overall saving. However, a study by Venti and
Wise (1987) concluded that ‘the vast majority of IRA (Individual Retirement Account) saving
represents new saving, not accompanied by a reduction in other saving’ (p. 38). These
results were confirmed using a different methodology by Feenberg and Skinner (1989).
In summary, there are three main aspects of behavioral economics that have important
policy implications in terms of the adequacy of saving:
1 Fungibility
Different forms of saving and wealth are not treated as being fungible or substitutable.
This is demonstrated by the evidence from the Venti and Wise study and the Feenberg
and Skinner study. Governments can make use of this lack of fungibility to encourage
more saving.
2 Self-control and commitment
IRAs, like other retirement accounts, are illiquid, since they involve a 10% tax surcharge
if money is withdrawn before the investor reaches 59½ years old. Venti and Wise (1987)
commented: ‘Some persons of course may consider the illiquidity of IRAs an advantage:
it many help to insure behavior that would not otherwise be followed. It may be a means
of self-control’. As stated earlier, the general trend in global financial markets towards
greater liquidity may have discouraged saving by removing such commitment devices.
Therefore governments can encourage more saving by creating additional commitment
devices in the form of illiquid savings accounts with tax incentives, such as Individual
Savings Accounts (ISAs) in the UK.


331
PT • I I I I NT ERT EMPO R A L C H O I C E


3 Framing
The desire to save, particularly for retirement, can be much influenced by the way in which
the options in retirement plans are framed, as noted in Chapter 5. Poorer people, for
example, are more likely to be enrolled in private retirement plans if that is the employer’s
default option than if workers have to elect to enrol. A study by Madrian and Shea (2001)
indicated that shifting to automatic enrollment raised participation among poorer workers
from just over 10% to 80%. UK pension policy has now moved in this direction. This kind
of ‘nudge’ policy is very much endorsed by Thaler and Sunstein (2008).

Questions
1 Why have different studies come to different conclusions regarding the adequacy of
saving?
2 Explain why the putting of money in a retirement account might not reduce other
forms of saving.
3 Explain why fungibility is an issue as far as increasing saving is concerned.
4 In what circumstances is illiquidity of assets a desirable characteristic?

Case 8.3 The desire for rising consumption profiles

Frank and Hutchens (1993) have investigated the factors that may cause wage profiles
to rise in ways that are not explained by increases in productivity. In particular they
examined the cases of airline pilots and intercity bus drivers, both of whom have
relatively constant productivity over most of their careers, but who have average annual
earnings at the end of their careers 600% higher and 50% higher, respectively, than at
the start of their careers. The authors rejected four existing explanations of the rising
wage profiles, relating to investment in firm-specific capital, binding contracts, risk-
aversion, and adverse selection, before proposing a theory relating to commitment.
The workers involved had to commit to accepting lower earnings than was justified
by their productivity in the early years of their careers. It was further argued that such
commitment is more likely in circumstances where much of the social activity of the
workers involved is with fellow workers, and they showed that this is indeed the case
with the two groups of workers they studied. Although the evidence in the study was by
no means conclusive, being limited to only two groups of workers, it is highly suggestive.

Questions
1 What are the behavioral factors underlying a preference for rising wage and consumption
profiles?
2 Explain why pilots and bus drivers have relatively constant productivity over their career.
3 Explain why investment in firm-specific capital cannot satisfactorily account for rising
wage profiles as far as airline pilots and intercity bus drivers are concerned.
4 Explain the role of commitment in causing rising wage profiles, and why social activity
with fellow workers is important as far as the likelihood of commitment is concerned.

332