1
Organizational Economics:
A behavioral approach
Colin Camerer, Caltech
camerer@hss.caltech.edu
2/17/2016 9:26 PM
Chapter 1: Thinking like a behavioral economist
This book is about basic theories and facts underlying how companies organize
workers, structure their companies, and choose business strategies, to be more
productive. An important part of this process is how top managers and boards govern
those companies to accomplish various goals.
The ideas in this book are a new blend of economics and psychology, and a little
sociology.
The central approach is economic. Conventional economic theories of individual
behavior rely on a preferences-beliefs-constraint system. That is, people know what they
want (have complete and stable preferences) and make the best choices given their beliefs
about uncertainties and constraints on time and money. (In technical language, when we
say “preferences are complete”, we mean that each choice can be assigned a numerical
“utility” and people are making the best choice, which is mathematically the same as
maximizing numerical utility.)
Extended to organizations, maximization means that workers pick jobs which
provide them the most combination of happiness, financial reward, and other job aspects
they value (responsibility, independence, fun, etc.). Managers anticipate what workers
will do given an incentive scheme, and sort workers into the jobs they like best and make
best use of their skills, and create companies which provide the most economic wealth to
shareholders. Sometimes these good organizations are created by design. More typically,
they come about through trial-and-error, or imitation-- a good design succeeds and is
copied by other firms.
Psychology enters as a way of modifying the economic theories to respect the fact
that people have natural limits on their abilities to make difficult calculations, to resist
temptation, and to act self-interestedly when interacting with friends and enemies. In
2
forming beliefs, people are often optimistic about the future, underestimate how long it
will take to solve a hard problem, and are overconfident about their skills and good
qualities compared to other people. People also care emotionally about people other than
themselves (e.g., they feel envy, guilt, moral obligation, and care about how they are
perceived by others), which affects how they behave at work. Judgments about how
valuable, numerous, or likely things are, are also constrained by psychophysical
properties and the nature of brain mechanisms.
Sociology reminds us that people are “socialized”: What we want is often
influenced by what others have or want; how we behave is influenced by informal norms
of proper behavior, as well as by formal laws and rules; and people are linked to others in
social networks which influence what they know and whose opinions and needs they care
most about.
“Behavioral economics” is a relatively new approach to economics which takes
into account these ideas from psychology and sociology to modify economic theories.
The idea is to use facts and ideas from these neighboring social sciences to improve
economics while respecting the two stylistic principles which made economics
successful—namely (i) use mathematics to express ideas clearly and generate insight and
surprising predictions that are tested with data, and (ii) to explain phenomena (and give
advice) in naturally-occurring situations.
This book is intended to fill a gap between economics books which focus solely
on the conventional economic model of worker motivation (e.g., Brickley, Smith and
Zimmerman, 2001) and psychology books on organizational design which focus on
worker motivation and satisfaction, and do not take the economic model very seriously.
The goal of behavioral economics is to take the best ideas and methodologies from both
approaches, to create a behavioral economics model that is precise and designed to
explain patterns, but takes psychological regularity into account in a mathematical and
empirical way.
1. Why organizational rules and incentives matter
The essence of organization is the idea that if you take the same group of people
with the same amount of money, skill, and time, the way their work is organized can
make a substantial difference. Organizational economics studies different organizational
practices with an eye toward finding what works best. Some of the most dramatic
examples come from countries, rather than companies.1
Many studies of economic development suggest that economic and political
“institutions”— the rules of how economic activity and politics work— have a huge
In “Eat the Rich” (1998, Picador Press), P.J O’Rourke summarizes the argument that human resources
alone do not determine a country’s wealth. He says: “Why do some places prosper and thrive while others
suck? It’s not a matter of brains. No part of the earth (with the possible exception of Brentwood) is dumber
than Beverly Hills, and the residents are wading in gravy. In Russia, meanwhile, where chess is a spectator
sport, they’re boiling stones for soup” (p. 1). O’Rourke later (pp. 233-234) summarizes his view of the six
pillars of good economies: Hard work, education, responsibility, property rights, the rule of law, and
democratic government.
1
3
influence on a country’s economic welfare (CITES). Many poor countries in Africa, for
example, have stores of some of the most valuable natural resources on earth, such as oil
and diamonds. (Nigeria produces 3% of the world’s oil and is the 7th largest OPEC
producer; Sierra Leone produces 0.3% of diamonds.) The countries are poor, it appears,
because the value from selling these natural resources does not trickle down to the poor,
because of governmental corruption, and because companies and bureaucracies in those
countries are not organized to align worker talents with jobs in a way that motivates
them. Inefficient organizations in such countries often lead to “brain drain” in which the
most talented and ambitious workers leave for other countries, if they can, where their
skills are better used and appreciated.
The last couple of decades of thinking suggest that three parts of the
organizational design are crucial: Decision rights; incentives, and evaluation. All three
have to fit together for the organization to be productive. In economic language, they are
complements; in business language, they must produce synergies.
Decision rights refer to the assignment of who has the authority to make
decisions, when the authority has not been clearly spelled out in advance (often called
“residual decision rights”). The authority to make decisions shows up most sharply when
there is a dispute— when one side wants A and another wants B, and there is no rule for
adjudicating the dispute that was agreed-upon, what actually happens? In many legal and
political systems, for example, there is a clear system of decision rights (typically with
checks and balances). In the US, for example, the President has the clear authority to
nominate Federal judges. But they must be confirmed by the “advise [advice] and
consent” of the United States Senate.
In companies, there is usually an important distinction between nominal authority
and real authority (sometimes called “formal” and “informal”). For example, in many
families you would think the parents make decisions. But a screaming toddler often has
the real authority because whenever the toddler screams loudly the parents do what she
wants. Another example comes from universities. In universities the faculty department
chair typically comes and goes with a short term (say, 3 years) while a senior
administrator (typically a woman) is in her job for 10 years or more. So in practice, while
the faculty department chair has formal authority, the senior administrator knows how
things work, how to get things done, and authority is often ceded to her. Generally, we
care about informal authority more than formal authority. But the conflicts that arise
when the two mismatch are of interest too, as they can contribute to crises and
inefficiencies.
Incentives are the way in which evaluations determine pay and anything else that
workers value— promotions, “voice” in decisions, job titles, larger offices, parking spots
budgets, and so forth. Keep in mind that money is a powerful incentive because
everybody wants more of it, and it has a crisp numerical measure. But other incentives
matter too. Monetary pay may even erode or “crowd out” other kinds of incentives, like
pride in a job well done, or the fun of being part of a team. As a result, choosing the right
balance of pay and other attributes workers value is tricky.
Evaluation refers to the system by which performance is evaluated. In wellfunctioning firms there is usually a formal component and an informal component. A
4
formal component might be an annual review or numerical measurement of how much a
worker produced. The informal component is typically opinions, in the form of written
letters or a meeting at which opinions are expressed.
Good organizations understand that these three components have to work
together. The idea is to give the right to make decisions to people who have an incentive
to make good decisions, and evaluate them so that incentives are paid out for good
decisions and penalties doled out for bad ones.
The last paragraph sounds so bland and obvious that you might wonder why you
need a whole book on this topic. The answer is that getting the assignment of decision
rights, incentives, and evaluation— three different features-- to all fit together at the same
time is not so easy. It is a little like juggling three balls: A juggler has to constantly be
passing two balls between her hands rapidly, while the third ball is in the air.
Furthermore, changes in personnel, technology, law and cultural norms mean the
organization has to constantly adjust the mix of decision rights, incentives and evaluation.
The best mixture might also be different for different types of workers and cultures.
For example, a typical mistake in balancing decisions, incentives and evaluation
occurs in multitasking, when a job requires somebody to perform more than one task. For
professors, a typical example is teaching and research; for salespeople, it might be direct
selling (for a commission) and training other salespeople. Incentives which reward one
activity, but not another, can lead to people overproducing the incentivized task and not
doing enough of the task they aren’t paid for, even though the organization would prefer
the worker to do some of both. For example, suppose a fast-food restaurant asks people
working at the counter, directly serving customers, to clean up when there are no
customers to serve. If serving customers is more fun than cleaning up, workers prefer to
linger at the counter or appear to be busy. They might implore friends to come and
“visit”, essentially posing as customers so they can avoid the unpleasant duty of cleaning
up. (Generally, the solution is to link incentives to the task which can be measured and
monitor the other for an adequate level of quality.)
Another fact about human nature that makes organization difficult is that people
are different. Some people like responsibility in their job and like having decision rights.
Others get stressed by tough decisions and prefer to let other people decide. Therefore,
decision rights and evaluation have to be tailored to individuals to some extent.
Different people are also motivated by different incentives. In this book, the
default meaning of “incentive” is money. However, do not make the mistake of thinking
that money is the only incentive people care about, or is even the strongest. In wartime,
people risk their lives for combat pay, but also for national pride, the camaraderie of
engaging in a history-making enterprise, out of a principled sense of justice, or to protect
the lives of the fellow soldiers they have come to treat like siblings. In some cultures,
shame is such a powerful incentive that people commit suicide rather than face shame, or
kill their own relatives in “honor killings” if they have brought shame upon the family for
actions that other cultures would completely forgive (such as being involuntarily
assaulted, or marrying for love). These examples show how powerful emotions can be as
motivators.
5
The importance of making workers emotionally happy, not just rich— though
riches probably contribute to happiness—is recognized in starting companies.
Entrepreneurs often describe their goal as providing a fulfilling place to work, which
creates a wonderful product that makes people’s lives better. These entrepreneurs see
financial earnings as the market’s way of rewarding them for making a great product, and
for taking a large risk to do so.
Because people are different, the right combination of decision rights, incentives,
and evaluation will also depend on the people who work in the organization. This raises
an important challenge of sorting—finding the right people for the right jobs.
Organizational meltdowns
One way to appreciate why organization matters is to study cases where
organizations failed, even when they are composed of talented, energetic people. (We
will see many throughout this book, and we will study successes as well.) These stories
remind us of the difficulty of coordinating decision rights, incentives, and evaluation,
especially when people care about different goals.
For example, many adventure books describe how physically fit, mentally tough
adventurers band together to accomplish some goal. Many of these stories end badly
because of social mayhem, despite the participants enduring incredibly physical hardship.
In Shooting the Boh (1992), Tracey Johnston describes a white-water rafting trip
to a very difficult river in Borneo which white men had never rafted before. Their
expedition endures terrible challenges in the form of tropical jungle diseases and the
physical challenge of rafting a steep river surrounded by huge rocks on both sides
through many of its swiftest rapids. In Running the Amazon, Joe Kane describes a
kayaking trip from the very headwaters of the Amazon river, on a mountaintop in Peru,
through the entire Amazon to the Atlantic Ocean. (His book ends with one word—
“salt”— which is what he tastes on his fingers as he dips them in the ocean, after
paddling in the fresh water river for months.)
Both adventures end on bitter notes. (Keep in mind that these are the successful
adventures-- everybody survived, and the trips were worth writing books about.
Presumably the unsuccessful expeditions faced even bigger organizational challenges.)
Johnston’s Borneo group bickers over physical hardship. Stress is caused by the
fact that ill-prepared travelers end up borrowing routine items from others, like dry socks,
which become extremely valuable in the wet rainforest. The heroic river guide, who
single-handedly helps the group overcome many rafting and logistical obstacles (e.g.,
finding food) ends up romantically involved with a beautiful ex-model who is on the trip.
Others come to think that the intrepid guide favors the ex-model at their expense, which
causes jealousy and tension.
Kane’s Amazon trip includes an Olympic-level paddler, and a businessman who
raised money to have part of the adventure filmed. The trip goes slower than planned and
the group begins to run out of money. The Olympian is frustrated by the slow paddling
pace of the others. The businessman, who is the slowest paddler, holds the group back but
feels no guilt because his fundraising (for the planned film) financed much of the trip.
Only Kane and one other paddler finish the entire trip.
6
Adventure trips like these are special organizations in two respects: First, they are
temporary or “instant” organizations created once, to accomplish a special challenge.
This means that people often go into the trip not knowing much about the goals and skills
of others. It also means that the threat of not doing business with somebody in the future
if they don’t live up to expectations, which is often a strong incentive in long-lived
businesses, has little power.
Second, the expeditions require people to travel together, and are physically
demanding. These trips are like a chain whose strength depends on its “weakest link”. If
one person complains a lot, eats too much food, sleeps too late, or paddles too slowly, the
entire group is demoralized, hungry, or slowed down.
Organizations like these are especially vulnerable to design mistakes. One design
mistake reflected in these adventures is pure optimism: The adventurers almost always
underestimate how difficult their tasks were. This is common in business and government
as well: Large building projects, for example, usually take twice as long as planned and
cost twice as much.2 Another design mistake is failing to recognize how differently
people are motivated. For example, the Borneo trip’s guide is certainly motivated by pay
and pride. But on this particular trip, he was also motivated by personal feelings toward
one of the travelers. Either the guide could not put these feelings aside— and no
evaluation system was in place to discipline him—or, more likely he felt he could fulfill
his obligation to the group while still enjoying his romantic time with one of the people in
the group.
In simple economic terms, the issue with the river guide was the extent of his
working hours—during downtimes, is he still “on the clock” and obligated to deal with
the group’s endless suffering, or is he “off the clock” and entitled to rest and relaxation of
his own. The group thought they had the decision rights to allocate the guide’s spare
time, and the guide thought he had the decision rights. This story also shows how
psychology enters: Different people often have perceptions of situations which are “selfserving”, and lead them to blame others rather themselves, so that both sides end up
blaming the other.
Fraud: Burning down the house
A useful way to appreciate the importance of balancing decisions, incentives and
evaluation is in studying large-scale frauds and scandals, which often have a huge
financial cost to firms (and entire industries, through reputational spillovers) far beyond
the gains to the people who perpetuated them. 3
Let’s see a couple of examples and then think about what lessons can be learned
from them.
2
[FINISH CITE LOVALLO KAHNEMAN HARVARD BUSINESS REVIEW
EXAMPLES]
3
The website at http://www.erisk.com/Learning/CaseStudies.asp has some good case studies, two of which
are excerpted here.
7
Daiwa: How to lose a billion dollars…slowly. Toshihide Iguichi went to work
for Daiwa Bank in 1977. Daiwa was a large bank with $200 billion in assets and $8
billion in reserves (as of 1996) and a growing trading operation. For seven years, Iguichi
worked in the back office of Daiwa’s growing New York government bond trading
operation, keeping track of the enormous amount of paperwork involved in reconciling
accounts of a trading operation. Iguichi became a trader in 1984—authorized to buy and
sell government bonds for customers, and for the Bank’s own account—but Iguichi also
kept his job supervising the “custody” department (which kept track of who owned the
bonds). Daiwa’s custody account was administered via a “sub-custody” account at
Bankers Trust. Daiwa and its customers kept track of activity in their accounts through
reports from Bankers Trust. The reports were filtered through Iguichi.
Early in his trading career, Iguichi lost a few hundred thousand dollars trading
bonds. He decided to sell other bonds in the Bankers Trust account to cover his losses,
but falsify the reports so nobody who saw the reports realized the bonds had been sold.
As he lost more and more money trading, Iguichi falsified more and more Bankers Trust
reports to disguise transactions he had made to cover the trading losses. The false
reporting persisted because Daiwa’s internal auditors never checked Iguichi’s false
reports against Bankers Trust’s own records. Presumably they trusted Iguichi because he
had worked in back-office operations for many years before beginning to trade.
Iguichi’s fraud lasted not one year, or two…it lasted 11 years. During that time,
Iguichi sold $377 million of customers’ securities, and $733 million of Daiwa’s
securities—a total of $1.1 billion. To cover these fraudulent trades, he forged 30,000
trading slips, an average of 10 each day.
In fact, Iguichi was never actually caught: He confessed. On July 13, 1995, he
sent a 30-page letter to the president of Daiwa in Japan explaining what he’d done. He
said he confessed because “after 11 years of fruitless efforts to recover losses, my life
was simply filled with guilt, fear and deception”.4 He got tired of covering up his bad
trades, and of waiting to be found out.
The aftermath of Iguichi’s massive, long-lasting fraud is just as remarkable.
Daiwa waited a month to tell the Japanese Ministry of Finance about the losses, on
August 8. Despite a clear legal requirement to report the fraud to U.S. regulators, Daiwa
actually waited another month to tell the U.S. Federal Reserve Board. (The finance
minister, Masayoshi Takemura, later denied that his Ministry had failed in its reporting
duties, made a veiled apology to the U.S. Treasury Secretary, then later denied
apologizing.) During September, Daiwa higher-ups took Iguichi’s advice to keep the
losses secret until “appropriate measures” could be taken to ensure stability after the
news broke. Iguichi was told to pretend he was on vacation when he was actually hiding
out in the apartment of a Daiwa manager in New York trying to reconstruct the history of
his long-lasting fraud.
Daiwa finally told the Feds what had happened, in mid-September. Unhappy
about being left in the dark, FBI agents interviewed Iguichi on September 23. Shortly
4
http://www.time.com/time/magazine/1997/int/970210/interview.i_didnt_set.html interesting interview
8
after that, Iguichi was arrested and the bank itself was indicted on 24 counts including
conspiracy, fraud, and failure to disclose federal crimes.
It then become apparent that Daiwa’s New York trading operation had a long
history of rogue operation which was invisible to Daiwa’s top managers back in Japan.
The bank had operated an unauthorized trading area for seven years, from 1986 to 1993;
they even disguised as a trading room temporarily as a storage room when regulators
stopped by. Iguichi also said that from 1984 to 1987, other New York traders had created
major losses which were concealed from regulators by shifting the losses to overseas
affiliates.
In November, 1995, the Fed ordered Daiwa to quit doing business in the U.S. A
month later, it sold its U.S. assets and 15 offices to Sumitomo Bank for $3.3 billion. In
February, 1996, Daiwa agreed to a $340 million fine to settle the indicted charges—a
record for a criminal case. Many senior executives resigned or took early retirement.
Iguichi himself was sentenced in December 1996 to four years in prison and a $2.6
million fine. The final sting was an order from a Japanese court requiring 11 current and
former board members to pay the bank $775 million in damages, as compensation to the
bank’s shareholders and as a penalty for the managers’ failure to report the incident
promptly.
An important cost from this kind of fraud is an “externality”-- the loss of
reputation of people and institutions associated in people’s minds with Daiwa. Even if the
fraud was the work of a single rogue (and unskilled) trader, if the capital markets think it
is the tip of an iceberg, and other frauds are going on undetected, that can harm capital
market credibility. Indeed, before the scandal, Japanese banks were charged a .1%
premium above the London Interbank Offer Rate (LIBOR, an international benchmark
for interest rates). Afterwards the premium went up to .25%. Daiwa’s corporate mistake
therefore tainted other Japanese corporations and cost them huge sums in extra borrowing
costs. One analyst said the problem was that the Japanese Ministry of Finance was
“severely compromised” as a regulator, making it “difficult to be reassured that this is not
going to happen to someone else.”5 The markets therefore perceived Daiwa’s fraud as
something that could be going on in other banks, that would not be sharply regulated by
the Finance Ministry, which led them to charge an extra premium just in case.
What can we learn from the Daiwa case? The basic one is that Daiwa did not have
decision rights, incentives, and evaluation balanced properly. Iguichi had the right to
make trades independently, but was also in charge of the back-office reporting of account
gains and losses, which provide numbers to evaluate the performance of his trades and
those of others. Iguichi was in charge of supervising himself.
Keep in mind that people who are in charge of their own evaluations do not
always cheat. The sorting process, assigning traders to small offices where they run their
own back-office operations, could work fine if you can find enough traders who are both
talented traders and honest back-office worker bees. But as one source wrote, “Bankers
who hire money hungry geniuses should not always express surprise and amazement
5
http://www.asiaweek.com/asiaweek/95/1027/biz2.html
9
when some of them turn around with brilliant, creative, and illegal means of making
money”.6
Furthermore, Daiwa’s actions upon learning of Iguichi’s fraud made matters
worse, and perhaps reflected a culture of tolerating or covering up scandal which licensed
Iguichi to start covering up his own losses in the first place. Daiwa managers, and the
Finance Minster, seemed incapable of taking quick action or admitting fault, until a lot of
diplomatic and economic damage was done.
A behavioral economics question is: What motivated Iguichi to commit his
crimes? (Understanding his motivation might help us prevent situations like these or
know how to sort for employees who are both talented and unlikely to commit fraud.) In
a 1997 interview conducted in prison, Iguichi said his early actions did not feel like a
crime. “To me,” he said, “it was only a violation of internal rules.” He then explains that
his actions were driven by a deep desire to erase trading losses—a pattern psychologists
call the “break-even effect”, or in studies of pathological gambling it is called “going on
tilt”.7 As Iguichi described it: “I think all traders have a tendency to fall into the same
trap. You always have a way of recovering the loss. As long as that possibility is there,
you either admit your loss and lose face and your job, or you wait a little – a month or
two months, however long it takes”.
In fact, most traders do not fall into this trap for as long as Iguichi did. The ability
to resist “chasing losses” by taking bigger risks, or covering up the losses with a backoffice fraud, and making peace with losses, may be one of the traits that distinguish
successful traders from those with perennial losing records, like Iguichi.
Barings: TBA
6
The quotation is from a speech by the financial thriller writer Linda Davies on the Psychology of Risk,
Speculation and Fraud, at a conference on EMU in Amsterdam.
7
E.g., Johnson and Thaler (19??) Management Science. FINISH
10
Figure: Nick Leeson (seated at right), the man who destroyed Barings Bank,
talking to the press after his release from prison in Singapore in July, 1999.
Most of these frauds have three ingredients: Poorly-designed rules which enable
the fraud to go on without being revealed; high stakes which make the fraud worthwhile;
and a moral willingness or psychological motive to commit the fraud, and sometimes
embarrassment or shame which prevents the fraud from revealing his crime, and leads to
desperate attempts to cover it up. Sometimes the guilty party blames others for not
discovering it, or just sees the fraud as winning at a game with poorly-designed rules.
An interesting question is: Are these frauds common or rare?8 No monitoring,
enforcement, or regulatory mechanism is perfect. Assume that the marginal cost of
monitoring rises dramatically with the probability p of detecting a fraud; i.e., the marginal
cost of raising the probability c’(p) as p 1. (Mathematically, that is the same as
saying that is expensive to track down the most wily and clever fraud — then there will
8
CFC FINISH Joe Jett case
http://www.smeal.psu.edu/faculty/huddart/Courses/BA521/Jett/JettNYTimes97.shtml
(note starbuck quote) http://money.cnn.com/1999/12/29/investing/century_greed/
11
be some frauds even if firms are optimizing.)9 An interesting question is how widespread
such frauds are and whether firms are taking all the right steps to prevent them. It is
difficult to show that firms are systematically misaligning decisions, incentives and
evaluation in these case studies without a full model showing the costs and benefits of
mistakes.
An analogy which can help us think about the optimal amount of fraud is
“shrinkage”—the polite term for employee theft and customer shoplifting—in retail
stores. A typical shrinkage rate is about .5% of inventory. In some businesses, like
groceries, profit margins are only 1% of sales, so shrinkage may do a lot of damage to the
bottom line. But reducing shrinkage to zero is just too costly. The lowest rates among
retail chains are only on the order of .1%.
2. Building blocks of economic analysis
Before proceeding, let’s spend some time on the most basic concepts of
economics. These form a foundation for eventually thinking about workers in companies.
Marginal analysis
In economics, we start with the presumption that people will perform an activity
until the marginal benefit of continuing is equal to the marginal cost of continuing. This
way of thinking is loosely called “marginal analysis”.
Marginal analysis is also important at the market level: If prices are determined
by the intersection of supply and demand, then prices are not determined by the person
who values a good most, or the seller who can make it most cheaply: Instead, prices are
determined by the marginal buyer and seller.
A famous example is water and diamonds. Suppose there was a massive shortage
of fresh water on the earth. Because water is necessary for life, the price would skyrocket.
(In fact, after natural disasters like earthquakes and floods, water and other valuable
staples, and useful items like batteries, candles and shovels, often do rise sharply in
price.) So why is water so cheap? The answer is that water is plentiful—sometimes you
can even collect your own in a bucket for free, when it falls from the sky. Developed
economies can purify and distribute clean water on a massive enough scale to spread the
fixed costs across a large market. Thus, the marginal cost of supplying one more gallon
of water is quite low. Since the water business is usually competitive (and often regulated
as a public utility), prices are low too.10 So even though water is necessary for life, it is
not very valuable “at the margin”.
9
This is why economists think policies which are intended to reduce the number of violations of some rule
to zero, are noneconomic pipe dreams. The rhetoric of zero-tolerance may prove politically useful, or may
motivate people to get closer to that goal, but it is not the result of optimization if the marginal cost of
detection skyrockets when you try to catch every last user or terrorist.
10
Keep in mind that prices depend on cost and how competitive a market is. Suppose there was a water
monopolist who controlled all the water and could charge whatever they would like. Because water is such
a necessity (we say its price elasticity—the percentage change in demand relative to the percentage change
in price—is low) the monopoly price would be high. (Generally, when a good is price-inelastic, a
monopoly’s profit-maximizing price is high). Thus, water is cheap in developed economies because it can
12
An opposite case is diamonds. While diamonds may be a girl’s best friend, no girl
ever died if she went without wearing them for a few days, as she would if she went
without water. So why are diamonds so expensive? Part of the answer is that the supply
of diamonds is a classic monopoly. A large portion of the world’s diamonds are
controlled by the DeBeers corporation, which carefully controls the supply to keep prices
high. Diamonds are also rare and laborious to mine, compared to the cost of pumping and
purifying water.
Constrained utility maximization
The central modeling tools in the economics of consumer behavior are the
concept of utility, budget constraint, and (combining the two) maximization of utility
subject to constraint. These ideas form the mathematical underpinning of a “demand
curve”—a function that tells you the number of units of demand for a product D(P) if the
product’s price is P.
A product or activity’s “utility” is a number which reflects how much people
seem to anticipate liking the product when they make choices. (Some nuance to this
definition will be offered below.) There is much confusion about the concept of utility.
Here’s how you should think about it: Suppose you sat down and watched a bride and
groom and a wedding planner, planning a wedding. Suppose each wedding “menu” has a
certain number and type of flowers, a venue where the wedding is held, and a food menu
for dinner. As the wedding planner leafs through a thick book listing these (flowers,
venue, menu) “bundles”, the bride says “I like this one more than that one” many, many
times. We call this list of statements of what she likes a “preference relation”, often
written mathematically in the notation A ≥ B (read “A is weakly preferred to B”, or A ~
B if A and B are equally preferred). Suppose that her preferences are “complete”, so that
for any pair of (flowers, venue, menu) triples she can decide that she likes one, or that she
is truly indifferent between them. Suppose further, that her preferences are transitive—if
she likes A more than B, and B more than C, she also likes A more than C.
If the bride’s preference relation is complete and transitive, then we can keep
track of what she likes by assigning numbers to each triple of (flowers, venue, menu).
Suppose there are N such triples. Just assign N points to the one she likes most, N-1
points to the one she likes second-most, and so forth. These point assignments are
“utilities” which are revealed by her expressed preferences. The utility numbers just
equate the statement “I would rather have A=(roses, Peninsula Hotel, seafood) than
B=(tulips, Malibu beachfront, filet mignon)” with the statement “A has a higher utility
number than B”, or “u(A)>u(B)” (where u(x) is the function which assigns a number to
x).
Notice that we are not insisting that the bride actually consciously has numbers in
mind when she expresses preferences (although people may do so in unusual cases). But
if she can always make up her mind between two bundles, and does so transitively, she is
acting just as if she is would act by consulting a list of numbers, and picking the bundle
with the higher number.
be mass-produced cheaply, and also because the supply side of the market is usually competitive (or
regulated to create low prices).
13
The important part of this utility-maximization process for economics is the
concept of tradeoff: Assuming the bride has a limited budget, she must decide whether
she is willing to spend less on flowers in order to have more hors d’ouevres served at the
reception. We call the amount of one good which is worth giving up to get more of
another the marginal rates of substitution between the two goods.
In most choices people actually face, there are two other components which are
missing from the simple sketch above: Risk, and time.
Usually we are not sure exactly what the consequences will be when we make a
choice—the choice is risky. For example, if the wedding is planned for a Malibu
beachfront, there is a small chance that it will rain and the wedding must be moved
inside. So the bundle (tulips, Malibu beachfront, filet mignon) is really a probabilistic
series of different realized bundles.
In traditional economics we usually assume that people value these probabilistic
“gambles” by taking a probability-weighted sum of the utilities of the possible
consequences. That is, if a gamble G has n different possible outcomes, denoted xi, and
each has probability pi then u(G)= i= 1n pi u(xi). This is called the “expected utility” (EU)
rule. It is mathematically equivalent to a person’s evaluations of different gambles
obeying a series of “axioms” or general principles. The key axioms which imply the EU
form are “completeness” of preference and transitivity, and cancellation. The
completeness axiom says that people can always decide whether one gamble is better
than another, or they are indifferent between two different gambles. The cancellation
axiom says that if two gambles have a common component—if they are drawn as a series
of branches in a “decision tree”, then a common branch with the same probability of the
same outcome in both gambles can be cancelled out when comparing the gambles.
An important variant of the EU form is when the probabilities of outcomes are not
given by historical evidence (like records of car accidents insurance companies use to
establish auto insurance premiums) or a theory which makes a sharp prediction about
probabilities. More typically, events have a “subjective” or “personal” probability, such
as a person’s perception of the chance they will be promoted, or a company’s guess about
whether a rising executive can be a capable CEO someday. Then the EU form is replaced
by a subjective probability p(S) over a “state of nature” or random event S. The SEU
form for the overall value of a gamble G that yields outcome u(x(S)) in state S is
u(G)=S p(S)u(x(S)). Moving from known risks of outcomes xi, in the EU form, to
unknown risks over states S, allows us to model differences in opinion and phenomena
like optimism or overconfidence (i.e., a worker might think p(Success) is greater for
themselves than other people think it is).
The EU and SEU forms will play a substantial role in thinking about employment
contracts where a worker’s income can be high or low because of random factors out of
their control (so that valuing different jobs for their income potential is like choosing a
gamble).
Most practical choices also yield a stream of goods or activities which unfold over
time. This raises a question of how people weight the utility of good and activities they
get at different points. The standard theory is called “discounted utility” (DU). The idea is
14
that people weight utilities which they receive one time period from now by a discount
factor, δ. Utilities received t periods in the future are weighted by δt. Therefore, the
discounted utility of a stream of outcomes received at points t=0, 1, 2,… T in the future is
u(x0,x1,x2, … xT)= t=0T δt u(xt). This exponential form is logically equivalent to valuing
outcome streams in a way that is “dynamically consistent”. Dynamic consistency means
that if people value one future outcome stream I which begins at a future time T over
another stream J today, then when time T arrives they still prefer I over J (provided there
is no new information which might change what they prefer).
Behavioral economics: Limits on rationality, temptation and self-interest
The simple model of utility maximization above, on which almost all modern
economic theories rest, has proved to be a very powerful tool for explaining many
features of our economy, for giving people advice, and for engineering new rules and
systems. However, because the economic model is a mathematical simplification of a
complicated cognitive process, a more careful study of human psychology complicates
the simple picture of utility maximization (and the extensions to including risk and time
delays).
“Behavioral economics” is an approach to economics which tries to use empirical
regularities and ideas in psychology (and to some extent other fields) to improve
economics mathematically, to make it more empirically accurate and more useful
(Camerer, Loewenstein and Rabin, 2004). The main theme in behavioral economics is
that some people are limited in computational ability, willpower, and greed, in some
economics domains. Highly experienced experts, or unusual types of people, may have
lower rationality limits or may have learned to behave in a highly rational way from
experience. It is important to keep in mind, therefore, that the rational model is
sometimes a useful special case, and that heterogeneity across workers and managers is
important in determining outcomes. For example, even if most people do not know how
to value very rare risks, like floods or car accidents, highly-trained actuaries can do so.
Therefore, an insurance company might set accurate premiums for insurance because
actuaries “do the math”, even if their customers cannot (and may even think their
premiums are not an accurate guess about their own personal risks).
Here I will just give a quick sketch of some of the basic ideas that will be put to
use later in this book. Table 1.1 shows a comparison of rational choice principles on
which most economic theory is based, and some alternative models which have become
common in behavioral economics.11
For example, an alternative to the EU form for valuing risky gambles with known
probabilities pi is “prospect theory”. In prospect theory people value outcomes relative to
some point of reference, r. Experiments and studies with field data from many different
parts of the economy (see Camerer, 2000) suggest that in the vicinity of the reference
point, people dislike perceived losses (outcomes below x) roughly twice as much as they
like perceive gains (outcomes above x). Furthermore, in prospect theory probabilities are
thought to have a decision weight w(p) which is not linear—specifically, low
probabilities may be overweighted (w(p)> p for small p) and high probabilities are
11
Note that there is still much healthy debate about which of many alternatives explains data best, and the
advantage of different principles for producing theory and new predictions.
15
underweighted (w(p)<p for high p). A typical specification which modifies the EU rule
and incorporates this properties is the following: U(G)=v(G+) +v(G-), where v(G-)=xi<r
w(pi)(λu(xi--r) and v(G+)=xi>r w(pi)u(xi-r). The coefficient λ measures the strength of loss
disutility relative to gain utility; it can be used to help understand phenomena like why
workers dislike having their wages cut, and why firms are reluctant to do so even in a
recession when the demand for labor falls. A commonly-used weighting function is
w(p)=1/exp[(ln(1/p)) γ]. Note that when γ=1 the function reduces to linear weighting of
probability, w(p)=p.
Table 1.1: Some rational-choice principles and behavioral economics alternatives
Rational (or simplifying)
Behavioral alternative model
Representative citation
Framing, reference-
Kahneman-Tversky 1979
dependence
Koszegi-Rabin 2005
Contingent weighting
Slovic-Lichtenstein, 1971,
assumption
Complete preferences
description-invariance
procedure-invariance
Grether-Plott 1979
context-independence
separable u(x)
Comparative utility
Tversky-Simonson ?
Regret, disappointment
Loomes-Sugden ?, Gul 1991
Prospect theory
Kahneman-Tversky 1979
Nonadditive decision weight
Schmeidler 1989
Hyperbolic β-δ discounting
Laibson 1997
Inequality-aversion, fairness
Rabin, 1993, Fehr-Schmidt
Choice over
Risk
Ambiguity
Time
Self-interest
1999
Bayesian judgment
Equilibrium
Overconfidence
Odean 1998
Encoding bias
Rabin-Schrag 1999
Learning
Erev-Roth 1998
Camerer-Ho 1999
Quantal response, cognitive
McKelvey-Palfrey, 1995,
hierarchy
Camerer-Ho-Chong 2004
16
The field of behavioral economics is now so well-developed that reviewing all the
details of the ideas compiled in Table 1 will take up too much time at this point. Also,
many of the ideas have not been carefully applied in organizational economics. So they
will be referred to one by one as they become useful in the discussions later in this book.
Some day, our understanding of the limits on rationality which are expressed by
the behavioral economics models in Table 1 will be supplemented by a detailed
understanding of the brain processes which create human behavior. (The subfield that
strives to do this is called “neuroeconomics”.12)
Property rights
A “property right” is the right to use or sell property. Two components of
property rights are “use rights”—the right to use the property— and selling rights (or
“alienable” rights). In America, you have a use right to your kidney but not a selling right
(it’s illegal for you to sell it, though you can donate it). If you rent an apartment, you
have a use right but not a selling right.13
Some property rights are enforced by social convention. For example, in Chicago
it snows a lot in the winter. Many people do not have garages and park their cars on the
street. As a result, people often spend laborious hours shoveling snow away from their
parked car after a big snowfall, so that they can drive the car to work. It is common after
digging out your car to put a chair or traffic cone into “your” empty space (sometimes
with a homemade “do not park here” sign). These markers are understood by Chicagoans
to establish a property right to the empty space. The idea is that when a person arrives
home from work, “their” space is still available.
The system incentivizes people to dig spaces out. Those people who do not mark
their spaces with a chair or cone are “donating” the space to whomever needs it later. The
social norm works amazingly well, despite the fact that it is not backed by legal
enforcement. If you removed a cone and parked in a space that somebody had claimed for
themselves, the police could not issue you a ticket for parking illegally or “parking space
theft”. (However, the police would probably berate you for “stealing” the space and
encourage you to move your car to another spot.) The system is also enforced by
vandalism— if you park in somebody’s marked spot, there is a chance your car would be
scratched by a key or even have its tires slashed.
In developed economies, clear property rights are taken for granted and are
enforced by the “rule of law”-- a system of clearly specified legal rules, policing, access
to legal representation, and fair judges. Poor enforcement of property rights can lead to
underinvestment, because firms are afraid their machinery will be confiscated or stolen.
Poor enforcement can also lower consumer demands for valuable goods which other
people can steal.
The damage done to an economy by weak enforcement of property rights can be
seen most clearly in less developed economies. In developed economies, enforcement of
property rights often lapses temporarily during dramatic occasions such as riots and
floods, when police are overwhelmed or busy.
12
13
See Camerer, Loewenstein and Prelec (2004a,b).
Can you think of an object that has a selling right but not a use right?
17
Some small-scale societies practice what some anthropologists call “tolerated
theft”. Tolerated theft is a norm of sharing community property, in which the property
“owner” is not obliged to offer, but the “owner” cannot turn down a request to share. For
example, suppose you spend a couple of hours picking a large basket of berries in the
heat, and head down the road toward our village. If I see you and grab a handful of
berries out of your basket, and you don’t stop me, then you’ve tolerated theft. Tolerated
theft means that your private property isn’t just yours. This practice, in turn, reduces the
private incentive to gather anything valuable, if it takes work to do so (like hunting,
fishing, or gathering).
Not surprisingly, small-scale societies with tolerated theft are usually poor. In
some of these societies, one of the most valuable goods a man can own is a pair of pants.
Pants with large pockets are particularly treasured. Why? In a society with tolerated theft,
pants with pockets are like a safe deposit box—they are a place in which goods can be
hidden and protected from theft. Pants are an economic institution.
Poaching on property rights in a way that is akin to tolerated theft can occur at
much higher economic scales too. In 2004, Nigeria was the world’s seventh-largest oil
exporter. There, Alhaji Mujahid Dokubo-Asari openly admits that his homegrown militia
of up to 2,000 armed volunteers siphon off hundreds of barrels of crude oil from pipelines
operated by Royal Dutch Shell, refine it, and sell it cheaply to locals (LATimes, 2004).
Dukobo-Asari claims to control 3 local governments. He says he is fighting to “end the
disenfranchisement of his Ijaw people in Nigeria” following allegations of ballot fraud in
the national election.
Dukobo-Asari’s group do not see themselves as stealing somebody else’s oil.
They say they are simply reclaiming some of the oil that should belong to the people, but
which was given away by a corrupt elite who he accuses of “eating the money” from the
oil. “As far as we are concerned,” Dokubo-Asari says, “it is the Nigerian state that is
stealing oil”. He says he is being pursued by the state for criticizing the government about
electoral ballot fraud. The government says Dokubo-Asari is wanted for gangland killings
involving another outlaw, Ateke Tom; but Dokubo-Asari says Tom was hired by the
government to kill him. A Shell-commissioned report concluded that thieves annually
took $1.5-4 billion of crude oil illegally (though less interested estimates are much
lower). When he is asked how much crude oil his militia siphons from the pipelines,
Dokubo-Asari says “As much as we can. It’s free.”14
Economic transition in formerly-Communist countries in Eastern Europe, and in
the former U.S.S.R., has been erratic and spotty. One problem is that poor property rights
enforcement gives firms an incentive to invest heavily in private security (as a substitute
for publicly-supplied security in the form of policing and legal enforcement of contract;
e.g., Grief and Kandel, 1995). This gives a competitive advantage to firms that are Mafiaconnected; as a result, many observers are shocked by the deep influence of various
organized crime mafias on legitimate Russian business.
14
Questions to think about: Who is right and who is wrong in this situation? As you think about your
answer, suppose you were a judge or a UN Committee chair charged with adjudicating this situation. What
information would you need to know to make a judgment?
18
Another problem in economic transition is “deprivatization”—when governments
change the rules after a privatization. After the collapse of the USSR, about 70,000 staterun companies were privatized (i.e., sold by the state to a private owner or syndicate).
Some or the new owners were investors from Europe and the United States. For example,
the LBO firm Kohlberg Kravis Roberts teamed up with an enterprise called the USRussia Investment fund and bought a majority interest in the Lomonosov Porcelain
Factory in St. Petersburg in 1998 (Business Week, 1999). But in October 1999, a Russian
court stripped the American owners of their majority interest by declaring the company’s
initial privatization several years earlier had been illegal. The problem is that judgments
of illegality often hang on the tinest of legal threads or judicial whims—like a missing
piece of paper in a document file, or a failure to meet a requirement by an obscure
deadline. The newly-deprivatized company was rumored to be sold to its former Sovietera managers who were due to be “downsized” when the Americans took over.
Figure 1.1 below shows the powerful effect of property right protection on
economic growth during transition (Hoff and Stiglitz, 2004). The y-axis shows the ratio
of GDP in the year 2000 to 1989 GDP. The x-axis is the percentage of survey
respondents in different countries who disagreed with the statement “I am confident that
the legal system will uphold my contract and property rights in business disputes”. A
high percentage (right hand side) means a lot of respondents do not think their property
rights are protected. (This measure is also correlated with other measures of the “rule of
law”, such as a 0-10 index constructed by the Wall Street Journal.)
Each circle in Figure 1.1 is a different post-Communist society. In all those
countries which grew in GDP (the y-axis ratio is above 1, Poland, Slovenia, Hungary, and
Slovakia) less than 40% of respondents were insecure about their property rights. In the
countries with the most insecurity (highest disagreement percentage, which is furthest to
the right, such as Ukraine, Macedonia, and Russia) GDP actually shrank over the 11
years by half or more.
19
Figure 1.1: Growth and property rights in security in 20 transition economies (Hoff and
Stiglitz, 2004)
Complications with property rights can also arise when there are too many layers
of property rights which are not clearly prioritized. A nightmare version of this problem
was seen after the devastating 1994 earthquake that struck Kobe, Japan. Landlords,
landowners, tenants and subtenants all had various legal rights over decisions about
property they owned or rented. The Japanese decided that everyone with any rights over
property had to agree to a rebuilding place before construction could begin. But the
thicket of property rights made the process very slow. One city block had 303 different
renters, lessees, and subletters. Two years after the quake, despite a $30 billion rebuilding
fund, only 4 of 100 “construction teams” had actually begun doing any new building
(Wall Street Journal, 1996).
_______________________________________________________________________
20
Disputed property rights and law: The Barry Bonds 73rd Home Run Ball15
On October 7th 2001, Barry Bonds [pictured below] hit his 73rd home run of the year into
the right field arcade of PacBell Park in San Francisco, setting the Major League Baseball (MLB)
record for most home runs in a single season. The MLB single season home run record is
arguably the most prized record in sports; as a result, the record breaking ball was highly sought
after by sports enthusiasts (and was guessed to be worth a million dollars).
Left: San Francisco Giant Barry Bonds. Right: Patrick Hayashi, photographed by Josh Keppler
after picking up the disputed Barry Bonds 73rd home run ball.
As the baseball soared over the right field fence, dozens of fans scrambled and fought for
a chance to catch the ball. The first fan to get his glove on the ball was Alex Popov. But Popov
was quickly “buried face down on the ground under several layers of people…[and was] grabbed,
hit and kicked ”16.
Another fan, Patrick Hayashi [pictured], was also “involuntarily forced to the
ground…While on the ground he saw the loose ball. He picked it up, rose to his feet, and put it in
his pocket”. Hayashi motioned to a photographer, Josh Keppel, to point the camera to him, which
Keppel eventually did when someone else in the crowd asked him to. Then Hayashi “held the ball
in the air for others to see. Someone made a motion for the ball and Mr. Hayashi put it back in his
glove.”
Who owned this valuable ball? This case illustrates how property rights can be under
dispute, even in a developed economy with clear legal rules. As the Superior Court verdict noted,
“While there is a degree of ambiguity built into the term possession, that ambiguity exists for a
purpose. Courts are often called upon to resolve conflicting claims of possession in the contexts
of commercial disputes…Because each industry has different customs and practices, a single
definition of possession cannot be applied to different industries without creating havoc”. Indeed,
different legal rules of possession have emerged differently in different businesses. For example,
in hunting and fishing, “the hunter acquires possession upon the act of wounding the animal not
the eventual capture”.
On Oct 24th, Popov sued Hayashi for ownership of the ball, claiming that it had been
stolen from him. Popov and Hayashi both thought they were the rightful owners and would
prevail in the case.
15
Ian Krajbich wrote the background material for this section.
This quote, and all others in this description (unless noted otherwise), is from the December 18, 2000
Superior Court verdict at http://www.courttv.com/trials/baseball/docs/BondsVerdict.PDF . See also
http://www.courttv.com/trials/baseball/ballsold_ctv.html
16
21
According to the law, prior to the time the ball was hit, it was owned by Major League
Baseball. At the time it was hit, the ball became ‘intentionally abandoned property’. The first
person who came in possession of the ball owned it.
But who possessed the ball? Popov was the first to touch the home run ball, but he lost it
in the crowd. The Court defined possession like so: “The central tenant [sic17] of Gray’s Rule is
that the actor must retain control of the ball after incidental contact with people and things” [note
emphasis] and thereby concluded that Popov had not possessed the ball.
Hayashi argued that Popov did not possess the ball because he did not have “complete
control” of the ball. As a law professor testified, “the first person to pick up a loose ball and
secure it becomes its possessor.” That person was Hayashi.
But wait. The contact between Popov and the fans nearby was clearly not “incidental”. It
is not as if Popov bumped into a fan and dropped the ball, which would then be legally possessed
by whomever picked it up. The fans mobbed Popov. A major purpose of law is to protect its
citizens and to discourage reckless and dangerous behavior. Denying Popov possession of the
ball because of the illegal actions of the fans who mobbed him would set a dangerous precedent
by encouraging violent behavior at ballgames.
As the Court ruled: “The reason we do not know whether Mr. Popov would have retained
control of the ball is not because of incidental contact. It is because he was attacked. His efforts
to establish possession were interrupted by the collective assault of a band of wrongdoers …Mr.
Popov should have had the opportunity to try to complete his catch unimpeded by unlawful
activity. To hold otherwise would be to allow the result in this case to be dictated by violence.
That will not happen.”
The judge acknowledged that Popov’s chance to control and possess the ball was
undermined by the unruly San Francisco mob. At the same time, the judge noted that awarding
the ball to Popov would be unfair to Hayashi because Popov could not prove that he would have
caught the ball even if the mob weren’t surrounding him. Popov’s fumble established a “prepossessory interest” in the ball. But, in the court’s words, that interest “does not establish a full
right to possession that is protected from a subsequent legitimate claim” (like Hayashi’s).
Taking into account both interests—neither should win, but neither should lose--, the
judge ruled that both men had equal rights to the ball and should evenly split the proceeds from
selling it. When the ball was actually auctioned off, it sold for only $450,000, roughly half of its
originally estimated value.
After the sale, Hayashi and Popov were both disappointed with the sale price. Popov
hinted that his legal debt was in the hundreds of thousands of dollars, because he had paid his
lawyer by the hour. (Whoops.18) Hayashi had cut a deal with his lawyers for a contingency fee
based on a percentage of the ball’s sale price, so he came out ahead compared to Popov.
In 1960 Ronald Coase (later awarded the Nobel Prize in 1991) wrote a remarkable paper
suggesting that if two parties can easily bargain to an agreement, which party has a property right
doesn’t matter for social efficiency—because the party who values the property right most will
bargain to get it.19 The Barry Bonds disputed ball-possession case should be a perfect example of
the Coase theorem in action: Hayashi had the ball in his possession. If it really was more valuable
for Popov—either he wanted it more, or thought he could sell it for more-- then Popov should
have paid Hayashi to turn it over. But if Hayashi wanted it more, or thought it was more valuable,
The Court meant “tenet”.
Suppose you hire someone to paint your house and you pay by the hour. Your neighbor hires somebody
and pays a lump sum for the whole paint job. Whose house do you think gets painted more rapidly? More
carefully?
19
Of course, the property right assignment matters for who gets the most from the deal, because if the
property right is assigned to the lower-value owner, then that person can get paid by the higher-value party.
If the higher-value party gets the property right, then they don’t have to pay.
17
18
22
there would be no price Popv could offer. Either way, the person who valued the ball most should
have ended up with it— the only question is whether was whether Popov would have to pay.
Looking back, neither Hayashi nor Popov valued the ball just to keep it in a glass case.
They wanted to sell it. They should have agreed to sell it and split the money. They would have
saved themselves enormous lawyer’s fees, which probably chewed up half or more of the ball’s
$450,000 value at auction.
This case illustrates why the Coase theorem can fail. The crucial assumption in the Coase
theorem is that people can cheaply reach an agreement by bargaining. Then it doesn’t matter who
has the property right (“ownership”) to begin with; after bargaining, the person who values it
most will end up with it.
The problem, of course, was that Popov and Hayashi each thought they had the morally
rightful claim to the ball, and also thought they would prevail in court and get all the money. The
Coase theorem can fail if bargaining fails, because both sides think they are more likely to win
than they really are, so they are willing to tolerate a costly delay in bargaining or risk an
expensive litigation process. Psychologists call this kind of collectively-inconsistent belief “selfserving bias”.
______________________________________________________________
Specialization, gains from trade, comparative advantage
People are good at different kinds of work. In an ideal economic system, people
specialize in what they are best at and enjoy. (Those two things can be different of
course.) Specialization is the engine of economic gain because it is easy to show that if
everyone can specialize in what they are best at (as judged by the tastes of consumers),
and trade occurs freely, the system makes the largest quantity of products everyone
wants. Specialization is essentially the same as a division of labor in which tasks are
assigned to those people who are best able to do them.
When specialization is allowed to flourish, economic systems benefit when
people have diverse skills as workers, and tastes as consumers, because the value of
dividing labor is higher. An example that illustrates this principle is the television show,
“The Amazing Race”. In the show, pairs of people are required to complete a series of
physical and adventure tasks— like riding a board across a 15-foot waterfall, or finding
an envelope with a clue in a dark cave filled with bats-- while racing around the world.
The teams which complete the tasks fastest win prizes. Some tasks require brainpower,
others require brawn, and still others require bravery. Generally, there are two big
challenges for the teams: (1) Coordinating their activity, so they don’t waste time
bickering or arguing about what to do; and (2) having at least one person in the team who
is particularly good at each task.
One season’s show featured a pair of identical twins. Is being in a team with your
twin a good or bad idea, if you want to win “The Amazing Race”? For coordinating with
your teammate, working with your twin is probably good— twins share all their genes,
and probably think and act alike on many dimensions. On the other hand, a pair of twins
will have less diversity in abilities, and are poorly equipped to have at least one teammate
who is a specialist in each type of task. In fact, the twins—two athletic men— placed in
about the middle of the competition, perhaps because their skills were too similar so the
benefits of a division of labor and specialization were too small.
23
The benefits of specialization can be illustrated by a joke: In Heaven, all the chefs
are French, the car mechanics are German, the bankers are Swiss, the police are British,
and the lovers are Italian. In Hell, the chefs are British, the car mechanics are French, the
bankers are Italian, the police are German, and the lovers are Swiss.
The advantage of specialization is driven by comparative advantage. A worker’s
absolute advantage is her productive capacity in a certain activity. A worker’s
comparative advantage is her relative absolute advantage—that is, her absolute
advantage relative to other workers. The point is that even workers who are not good at
any activity in absolute terms have some comparative advantage because there is always
an activity they are relatively less bad at. For example, a father making sushi is faster
than his 10-year old at making proper sushi rice, cutting nori seaweed, buying sushigrade fish, making sushi rolls, and setting the table for dinner. But the 10-year old is not
so bad at setting the table, and is hopeless at everything else. So the 10-year old’s
comparative advantage is in setting the table.
It is useful to think about comparative advantage in terms of opportunity cost. An
activity’s opportunity cost is the value of the opportunity that is lost by doing that
activity versus another. If a friend gives you Lakers playoff tickets as a treat, does going
to the game cost you anything? At first blush, the answer is “No”—the tickets were
free.20 But suppose that you could have sold the tickets for $500 to a ticket agent, or
“scalped them” in the active market that assembles before each game just outside Staples
Center in Los Angeles. Then by going to the game you gave up the opportunity to have
$500 instead. The tickets “cost” you $500 in an economic sense (though not in
accounting sense).
In behavioral economics, we often find that people underweigh opportunity costs
(Thaler, 1985). The ticket example is a classic one, in fact— many people would not pay
$500 to go to the game, but would go and forego the opportunity to earn $500. It seems
that people don’t see giving up $500 they could easily have as psychologically equivalent
to pulling $500 out of an ATM. Thinking about opportunity costs does not seem to be
natural: It takes the ability to reason counterfactually, to figure out something that did not
happen (but could easily have), and give the counterfactual the same cognitive status as
what did happen.
An example that illustrates the link between opportunity cost and comparative
advantage is the fictional small town in which there is just one lawyer, who is also the
fastest typist among the 100 people in town who can type at all. What should the lawyer
do with her scarce workday? If she types, the most typing will be produced. But if she
only types, the opportunity cost is high because the fast-typing lawyer is the only lawyer;
no lawyering will get done. The correct answer is that the lawyer should practice only
You’re thinking like an economist if you immediately thought that the tickets might
have been costly because you could owe the friend a favor in the future (i.e., the tickets
create a debt which has a future cost), or you got them because you were being repaid
from a previous debt (so that accepting the tickets forecloses the opportunity to collect on
the debt in another form).
20
24
law and have somebody else type. The best activity for each worker to perform is the one
that has the lowest opportunity cost.
Partial and general equilibrium: Unintended consequences
A very important concept in economics is “partial” versus “general” equilibrium.
A general equilibrium analysis takes account of all of the likely reactions and
counterreactions of all agents in the economy. A partial equilibrium analysis assumes—
usually incorrectly—that some agents do not react after a change, or are not behaving
optimally.
Why would you do a partial equilibrium analysis rather than a general one? One
answer is, “For simplicity”. Computing what happens in general equilibrium, when all
agents are optimizing, is often enormously difficult. And ignoring the actions of some
agents is often a reasonable approximation to the result of a perfect analysis that includes
actions by all agents (just as ignoring Pluto’s gravitational pull on the Earth is not a bad
approximation).
For example, suppose you are interested in the effects of a serious earthquake in
Los Angeles on the world economy. The most obvious economic effect is commercial
and residential property damage in Los Angeles. A partial equilibrium analysis would
analyze only the immediate effect on businesses and people if there is an earthquake. But
an earthquake might also cause people to relocate away from Los Angeles. A general
equilibrium analysis will include this relocation effect and takes into account its impact
on the cities Los Angelenos move too. Suppose Los Angelenos move to Las Vegas,
which causes a spike in housing prices. Suppose higher property taxes that are collected
by the city are spent on infrastructure, like expanding the airport (which is always busy).
That lowers the cost of traveling to Vegas to gamble, which lowers demand for Native
American casinos in northern California. So one effect of an earthquake in Los Angeles
could be a recession in Native American casinos which are hundreds of miles away.
General equilibrium analysis is like the zany machines designed by Rube
Goldberg (see Figure 1.2 below). At each step of the analysis you ask “And then
what?”—given the temporary result of the analysis, do any agents have an incentive to
change what they will do? The analysis ends when no agents have an incentive to change;
the economy is then said to be “in equilibrium”.
25
Figure 1.2: A Rube Goldberg machine
The distinction between partial and general equilibrium is important because a general
equilibrium analysis will often reveal surprising effects or “unintended consequences”
of a policy which a partial equilibrium analysis misses. (We will see some of these
unintended consequences below in studying the effects of changes in incentive systems—
sometimes they create cheating, or produce too much of one kind of worker effort at the
expense of another.)
A common unintended consequence is that a change in the economy which is bad
for many people actually benefits some. For example, the end of the Japanese stock
market bubble in the late 1990’s led to a severe recession which led to a rise in burglaries
in big cities Tokyo and Osaka. Where’s the silver lining in this gloomy cloud? Ask
Makoto Iida, the founder of a leading security systems company Secom. Iida said “I feel
very good now” because sales of his $80/month intruder-detector systems rose 20% a
year through the mid-90’s as the crime rate rose.
Incidence of taxes is another example of how the general equilibrium effects of an
economic change might be quite different than they appear at first blush. When a tax is
applied, who ends up actually paying the tax depends on supply, demand and
competitiveness. For example, suppose you impose a profit tax on corporations. Suppose
profits were determined by competitive forces before the profit tax, so that shareholders
who supplied capital were earning a fair rate of return on investing in corporations. Then
after the tax is imposed the net (post-tax) corporate profit will be lower; and by inference,
it will be less than shareholders will accept because the new profit rate will be lower than
the higher pre-tax rate. To make up the effect of the tax to shareholders, firms will raise
price (or cut costs). Raising prices will usually cut demand, to an extent which depends
on the price elasticity. If profit taxes are imposed across the board (on all firms), then
prices will generally increase. So consumers may end up paying higher prices because of
a tax on corporate profit.
26
A few years ago, a luxury tax was imposed on the sales of fancy cars and boats
that only wealthy people buy. The tax was 10% of the price of cars and boats above some
threshold (which implicitly defined the threshold of “luxury”). A small political debate
ensued after people realized that the tax would reduce demand for boats, which were
often built in small shipbuilding towns like New Bedford, Massachusetts which were
struggling economically even before the proposed luxury tax. So what appears to be a tax
on the rich had an unintended consequence of raising unemployment of some un-rich
craftsmen.
Unintended consequences I: Gang members in Iraq?
Under heavy pressure to produce enough soldiers to fight the war in Iraq, in 2004,
the Pentagon created a “Moral Waiver” program whose stated goal was to “better define
relationships between pre-Service behaviors and subsequent Service success”. In
practice, the changes appear to have relaxed previous restrictions on backgrounds of
people recruited into the Armed Forces. The Baltimore Sun reported (February 2006) that
from 2004 to 2005 the number of recruits with “serious criminal misconduct” in their
background (including aggravated assault, vehicular manslaughter and—ironically—
making terrorist threats) rose by half.
Even worse, the Chicago Sun-Time reported in 2006 that there was some
evidence of gang members being recruited.
At Camp Cedar II, about 185 miles southeast of Baghdad, a guard shack was
recently defaced with "GDN" for Gangster Disciple Nation [a Chicago gang],
along with the gang's six-pointed star and the word "Chitown," a soldier who
photographed it said.
...A law enforcement source in Chicago said police see some evidence of soldiers
working with gangs here. Police recently stopped a vehicle and found 10 military
flak jackets inside. A gang member in the vehicle told investigators his brother
was a Marine and sent the jackets home, the source said. (from Sun-Times)
"We're lowering our standards," [Defense Department gang detective Scott]
Barfield said.
"A friend of mine is a recruiter," he said. "They are being told less than five
tattoos is not an issue. More than five, you do a waiver saying it's not gangrelated. You'll see soldiers with a six-pointed star with GD [Gangster Disciples]
on the right forearm."[....]
27
(Picture from Sun-Times article).
Why would gang members be eager to join the Army? Take a guess. As the
Chicago Tribune reported (“FBI Probes Military Gangs”May 3 2006):
Of particular concern are reports that the Folk Nation, consisting of more than a
dozen gangs in the Chicago area, is placing young members in the military in an
effort to gather information about weapons and tactics, said FBI Special Agent
Andrea Simmons, who is based in El Paso, Texas.
"Our understanding is that they find members without a criminal history so that
they can join, and once they get out, they will have a new set of skills that they
can apply to criminal enterprises," Simmons said. "This could be a concern for
any law enforcement agency that has to deal with gangs on a daily basis."
Amazingly, Army investigators deny that there are many real gang members.
According to the Tribune, “nearly every one of the cases that we have looked into, it is a
young man or woman who thought that the symbol looked cool," said Christopher Grey,
spokesman for the Army's Criminal Investigation Command. "We have found some
people even get gang tattoos not really knowing what they are, or at least that they have
not had any gang affiliation the past."
_________________________________________________________________
THINK ABOUT THIS: Who do you believe about this threat, the FBI’s Andrea
Simmons, or the Army’s Christopher Grey? Why?
_________________________________________________________________
Unintended consequences are particularly important because of a phenomenon
called a “Lorenz effect”. Lorenz was interested in weather systems. These systems are
often chaotic, which means that even deterministic equations produce behavior that looks
random. Chaotic systems often exhibit an important property called “extreme sensitivity
28
to initial conditions” or “path-dependence”. This means that a small initial effect can
have a large long-run consequence.
Extreme sensitivity to initial conditions is an important idea for two different
reasons. First, it means that if you are trying to create a system that evolves to a good
long-run solution, starting in the right initial condition can be crucial. Second, it means
that it is often fruitless to search for a “big” reason why two systems would be vastly
different, because the reason might be a small difference in initial conditions. (The
“continental divide” game described below in section 4 is a good example.)
Here is an actual example. The Vietnamese dominate the nail salon business in
California. By one count reported in 2005, about 6500 of almost 8000 salons were owned
by people with Vietnamese last names (Central City Extra, 2005). Why?
Nobody is quite sure of the starting point. One story is that the film actress Tippie
Hedren, star of “The Birds”, visited a resettlement center in the 1970’s for Vietnamese
refugees and had a pleasant manicure. To help the Vietnamese she offered courses and
urged the state to give the test in Vietnamese. Nobody is sure if the story is true or
whether Hedren played a large, small, or no role. A more likely mixture of ingredients is
mentioned by Philip Nguyen, CEO of Southeast Asian Community Center, says:
Most if not all Vietnamese are refugees and they want to work desperately for
their living, to get themselves out of public assistance, and to support their loved
ones still in Vietnam. Only a very small percentage can speak fluent English.
They will jump into any businesses which do not require much English, have
short training and or training in Vietnamese and low capital for opening the
business: Nail salons fit in very well with their expectation.
Starting in the 1996, the written test manicurists must pass for state licensing was
offered in Vietnamese. Poor English skills become a barrier to entry in other service
professions, and a huge draw for the nail salon business. A complementary input is
capital. Vietnamese have a practice called “hui”, known more generally as a “ROSCA”
or rotating credit and savings association. In a ROSCA a group of people agree to meet
regularly, typically every week or month. Each person brings a fixed sum of cash to each
meeting. At the meeting, exactly one person collects everyone’s cash in that meeting
(after collecting, the winner cannot collect again until everyone has gotten cash). This is
an informal forced-savings mechanism which is equivalent to making a regular payment
into a bank and then withdrawing a large sum to start a business.21
21
FINISH
http://www.santersghostwriting.com/tippi.htm
http://www.billingsgazette.com/index.php?id=1&display=rednews/2002/06/16/bu
ild/business/14-nailsalon.inc
more on ethnic specialization…
http://www.thejournalnews.com/apps/pbcs.dll/article?AID=/20050930/BUSINES
S01/509300305/1066/BUSINESS01
29
An initial start in nail salons can lead to “increasing returns” and industry-wide
spillovers, which favor the Vietnamese salons over others, in a variety of ways; for
example, Vietnamese people may be able to convey their knowledge to one another by
sharing tips about business, because they all speak the same language or are more
inclined to share tips because of ethnic ties.
An important force that may cement one group’s presence in an industry is
learning-by-doing. In almost every human activity, people get better and better at the
activity as they practice. In fact, the “learning curve” is so regular that learning often
follows a simple power law— the time it takes to perform an activity is t(q)=qa, where q
is the accumulated output.22 So if the Vietnamese women have a headstart, their costs
can be lower, and quality higher, because they have more experience (higher q) than later
entrants. Put simply, they are further down the learning curve.
Supply and demand
A central principle in price theory is that supply and demand adjust to “clear” the
market— to find a quantity of trade and a price at which everybody who wants to buy
and sell at that price can do so.
An important feature of this adjustment process is that when the price of
something rises, supply will often appear in creative ways; and when prices fall, supply
shrinks (businesses fail and people switch jobs).
Economists have great faith in the ability of markets to create supply, perhaps
rapidly, when prices rise to make supplying more goods profitable. A surprising example
of the emergence of supply is in massively multiplayer online role playing games
(MMORPGs), a new genre of computer games where a player assumes the role of a
particular character and interacts with other online players, either attacking them to steal
in-game items and money or working with them to complete quests or defeat computercontrolled enemies.23 The main goal in most of these games is to build up “experience
points” (gained from completing quests and killing enemies) which then make your
character more powerful and give you access to new moves, items, quests and enemies.
Many of these games are addictive and time-consuming. A major motivation is to
build up one’s character within the game in the hopes of eventually rising to elite status.
Since building up a character’s resources is so time-consuming, some players have
sought to economize by paying real money for “pre-played” characters that have been
significantly built up by someone else. Responding to this demand for “pre-played”
characters, a videogames “sweatshop” industry has emerged. Companies hire foreign
workers, whose time is literally cheaper, to play online videogames and advance
characters to a point where they can be sold to gamers for a handsome profit. The first
known example of this type of business was a company named BlackSnow. “BlackSnow
hired workers in Tijuana to earn gold by ‘farming” in Ultima Online. [BlackSnow] sold
that in-game tender online for a handsome real-world profit while only paying [their]
22
FINISH GET SOME MORE ON THIS. E.G. CIGAR ROLLING DATA FROM PSYCH BOOK
The discussion of outsourcing MMORPG’s was contributed by Ian Krajbich. FINISH CITE FROM LA
TIMES
23
30
employees pennies on the dollar.” Industry insiders have estimated that this new industry
has grossed roughly $500 million.
Innovators in the industry have created scripts which do most of the work, so that
all the workers have to do is “fend off the occasional player itching for a fight or game
master who’s hunting for these automated farming programs.” Several game makers have
tried to fight this new industry. The creators of Lineage II did so when they banned
certain Chinese IP addresses from playing the game. The experience-for-hire industry has
responded by becoming trickier about how they go about their business. For example, the
best firms in the business are reputed to play on every game server available to spread out
their activities. They also use multiple accounts with different IPs, credit cards and
computers, to hide the accumulation of experience points.
Another example comes from a story about how many “extra” unmarried males
are expected to exist in China, because of a longstanding preference for male children
over females (perhaps combined with one-child policy instituted to fight population
growth). One estimate is that by 2020 there will be 30 million extra males.
The Chinese government has reacted by trying to both limit the ability to abort
girl babies, and incentives for having more girls (such as a 100 yuan/month girl bonus in
rural areas plus free educational fees). As ABC News reports24:
Today, as urban areas become more economically prosperous, parents can afford
to find out the sex of the fetus and can opt for sex-selection abortions. So the
government in China's largest province has banned the use of ultrasound
machines unless warranted by medical reasons. And the ban on abortion drugs
goes into effect this year.
But Chinese couples determined to have a son can easily get around the new laws.
A black market has sprung up of people with ultrasound machines in the trunks of
cars who will reveal the sex of the fetus for a price.
Managing markets and firms: Centralized and decentralized planning
Questions about ideal organizational design within a firm often revolve
around the same questions which are posed more generally in economics,
about whether centralized government planning or decentralized planning
are better methods of allocating capital and resources. To make the
argument simple, let’s talk about bosses and workers. Roughly speaking,
centralization is better if bosses have information that workers do not
have, if workers do not value control and autonomy too much, and if
centralization enables bosses to coordinate activity so that workers
are all doing the same thing (like an army) or if the bosses can
allocate different workers to tasks better than the workers could do
themselves. Decentralization is preferable when the opposite factors
24
ABC News: China fears lopsided sex ratio could spark crisis, Jan 12, 2007.
http://abcnews.go.com/International/story?id=2790469&page=1
31
hold, if workers know things the bosses do not, if workers like
autonomy, and if workers can coordinate and self-organize without the
boss’s interference.
Generally, economists are skeptical about the virtues of centralized planning.
Below are five stories about centralization and decentralization. One highly stylized
example illustrates the disadvantage of centralization and two historical examples
illustrate big disadvantages of centralization leading to death and poverty. The fourth
example illustrates a beautiful system which is mostly decentralized. The fifth example
illustrates a disadvantage of centralization.
Bad centralization of decisions: In communist Russia people could always buy
bread. State-owned factories were ordered to produce a certain amount of bread and sell
it at a low price. As an economist you should think: If bread is cheap, who determines
who gets the bread? Do people get the bread they most want?
The answer to how bread is allocated, which is common when goods are
deliberately underpriced, is lining up (queueing) and favoritism. Russians would line up
for hours to buy cheap bread. Relatives of bakers would have bread put aside they could
get after hours or without queuing. Economists hate queues because people waste time
standing in line, and their valuable time spent lining up could be better used as labor to
make something. Two disadvantages of this system are also that the quality of bread
might be uniform and poor. People who like fancy breads, and will pay more for them,
cannot buy them. Typically, state-owned agencies have no direct incentive to produce
good bread. The advantage of this system (perhaps its only social advantage) is that there
is a steady supply of bread. Mao-tse Tung spoke of the “iron rice bowl”, his way of
saying that in Maoist China people could always be guaranteed (hence “iron”) of getting
a bowl of rice. Supplying a staple regularly at a low price is often a hallmark of a
centralized regime, but also an indicator that people who want to buy a wider variety of
goods are not likely to get them from a system focused on such a simple goal.
Compare this system to a decentralized one, in which the government allows
anybody to bake bread and charge whatever price they like. In theory, firms which make
good bread can demand higher prices; their higher prices also signal other firms about
how to compete. If people switch from bread to a substitute, like pasta, then bread makers
go out of business and pasta companies spring up.
Bad centralization of expeditions: A beautiful example of the difference
between centralized planning and decentralized market activity is a study by Jonathan M.
Karpoff (2001) on expeditions to the North Pole and the Northwest Passage, between
1818 and 1909. There were three major goals of these expeditions: (1) discovery and
navigation of the Northwest Passage (2) discovery of the lost Franklin expedition (3)
discovery of the North Pole. Some of these expeditions were motivated by profit, as there
were rewards out for some of these discoveries: For example, the British government
posted a 15,000 pound award for the discovery of a Northwest Passage, since it would
have provided a potential alternate trade route to the Orient that would not be exposed to
attacks from the Spanish and Portugese.
Some expeditions were organized by governments and funded by public funds.
Others were privately financed. While it is harder to determine which expeditions were
more successful (because scientific discovery can be hard to quanity), it is easier to
32
measure components of failure. Table 1.2 compares three measures of failure: The
frequency of deaths of crew members (as a percentage of the total crew); the number of
ships on the expedition, and the number that were lost; and the percentage or crew
members who came down with scurvy (a debilitating and ultimately fatal disease due to a
vitamin C deficiency). In all cases the public expeditions did worse: More crew members
died, a higher number and percentage of ships were lost, and almost half got scurvy
(ouch).
Why did the public expeditions do worse? Karpoff suggests there are two main
reasons. First, the public expeditions often had worse leaders. One can imagine that a
political process influenced the choice of leaders, and those who emerged the victors
were not necessarily the most brave or best-qualified. Second, the public expedition
leaders adapted slowly to new information and were reluctant to change course.25 Their
slow responses might have been due to a desire to save political face (if changing plans is
seen as admitting a mistake) or just an inability to process new information in an
objective way.
Table 1.2: Failure rates in public and private expeditions to the North Pole and Northwest
Passage (Karpoff, 2001)
Public (n=35)
Private (n=56)
Crew deaths (%)
8.0
6.2
Number of ships (number
lost)
1.63 (.53)
1.15 (.24)
% with scurvy
46.7%
13.2%
Bad centralization in Communist China: The Great Leap ‘Backward’: In
1958, China’s Communist government launched the Great Leap Forward (GLF)
movement. The goal of the movement was to use collectivization and central planning to
make agriculture more efficient, and then divert resources towards surpassing Great
Britain and the United States in industrial production. The Leap proved to be backward,
not forward. National grain output plummeted by 15% in 1959 and another 16% by 1961,
causing a horrible national famine. Estimates of premature deaths during the GLF famine
range from 16.5 to 30 million, making it the worst famine in recorded world history. Not
surprisingly, given the Chinese government’s penchant for secrecy and making farfetched claims to save face, China’s officially blamed the famine on bad weather.
However, recent research in economic history has shown that the famine can more likely
be attributed to poor organization and decision making.
25
The modern equivalent is somebody like Michael Brown, the head of FEMA who was widely blamed for
reacting slowly to Hurricane Katrina, which devastated New Orleans in 2005. Brown was a political
appointee whose previous job had been legal counsel to the Arabian Horse Association. He had no
background in emergency management, a field in which experience surely counts. Among many wellpublicized mistakes, “Brownie” (as President George Bush called him) said he was not worried about the
hurricane because he got up in the morning and read in the papers that New Orleans ‘dodged a bullet’; but
no newspaper headlines used that phrase, and many said the opposite.
33
The Great Leap Forward was expected to be successful based on the “false
premise ingrained in the dominant Soviet economic ideology that collectivization would
transform Chinese agriculture from small household farming into large-scale mechanized
production, achieving a great leap in productivity” (Li and Yang, 2005). With increased
agricultural productivity, the government could then divert more resources to the
accelerated industrialization campaign, in the form of surplus workers and increased tax
revenue, and “impose excessive burdens of grain procurement on the rural population”
(i.e., force the farmers to sell more grain to industrial workers and consume less
themselves).
The main organizational change of the GLF was in combining small cooperatives
into 26,500 “people’s communes”, each one comprised of thousands of households.
Political propaganda about the benefits of collectivization led to wild, falsified estimates
of grain yield, such as “Grain output in 1958 was forecasted to grow to 525 million
metric tons from just 195 in 1957!” The actual 1958 output was around 200 million
metric tons, only a tiny improvement, and output then fell from 1958 to 1959.
Accepting the wishful thinking behind these baseless agricultural predictions,
China’s government reduced the agricultural labor force by 38 million peasants between
1957 and 1958, and reduced the amount of cropland by 13%.
Table 1.3: How to cripple an agricultural economy: Statistics during China’s
“Great Leap Forward”
Start (1957)
End (1959)
Forecasted output
195 MM tons
525 MM tons (1958)
Grain output
195 MM tons
170 MM tons
Grain imports
2.1 MM tons
3.9 MM tons
Grain procured from
communes
46 MM tons
64 MM tons
Grain kept by rural areas
273 kg/capita
193 kg/capita
Rural per-capita calorie
consumption
2100 calories/day
1500 calories/day (1960)
% of provinces where
workers can not exit
communes
20% (1955)
60% (1957)
Table 1.3 summarizes agricultural statistics between 1957 and 1959. Total grain
output actually fell, presumably due to inefficiencies in the large-scale commune system.
Such communes create a classic “free riding” problem or prisoner’s dilemma— why
should a farmer work hard when he or she cannot enjoy the additional output, as they
would from a home garden? Smaller-scale cooperatives appear to overcome the free
riding problem by scolding slow workers, kicking them out of the cooperative, and
relying on family and ethnic ties to promote cooperation.
34
While grain output fell, imports grew only a tiny amount. Most importantly, the
amount of grain procured from the communes to feed industrial workers and others
(perhaps Communist party officials) rose from 46 to 64 million tons. As a result, there
was a sharp drop in grain retained to feed farmers. Per-day caloric consumption fell from
2100 calories to 1500. (Healthy typical Americans consume about 2200 calories for men
and 1500 for women.) This drop in calorie intake has been shown to adversely affect a
human’s physical capacity to do manual labor, and therefore must have adversely
affected labor productivity.
Why did central planning fail? “Diversion of resources reduces agricultural output
directly. Excessive procurement, when combined with an actual reduction in productivity
caused by collectivization, significantly reduces food available for consumption in rural
areas, leading to a severe nutritional deficiency among rural workers. The resulting
reduction in physiological capacity to carry out manual labor would in turn reduce the
quality of labor input in growing next year’s crops, leading to an additional decline in
production.” In plainer English, as farm productivity fell, and more grain was moved
from communes to industrial areas, hungry workers became too weak to produce more
food, so productivity fell further, in a literal death spiral.
Using an econometric model, Li and Yang found that the most important
determining factor was the diversion of resources from agriculture, accounting for 33%
of the decline in output from 1958 to 1961. “Excessive procurement of grain, which
decimated the physical strength of the peasantry” was responsible for 28% of the collapse
in output. Bad weather played a role too, but accounted for only 13% of the decline.
Another contributing factor to the famine is the “deprivation of peasants’ rights to
exit from the commune.” Some researchers have argued that “the threat of withdrawal
from an agricultural collective by harder-working members helps discipline would-be
shirkers. The removal of exit rights destroyed this self-enforcing discipline, reduced work
incentives, and hence contributed to the fall in grain output.” The fraction of provinces
with no exit rights increased from 20% in 1955 to 60% in 1957.
Imagine what would have happened if a system of property rights for workers’
output, labor mobility, prices, and free trade were suddenly installed in 1959. The most
productive workers would exit the communes to keep profits from their grain and would
produce more. High prices would signal to other countries that they can profit from
selling grain to China, producing a flood of imports (assuming trade is free).
Instead of this scenario, the ironically-named Great Leap Forward was a perfect
storm of bad organizational design. Wild optimism led leaders to believe productivity
could soar, so more grain could be taken from the communes. The best workers could not
leave the communes; discouraged and hungry, their private incentives to work hard were
muted. (Suppose you were pretty hungry and spent all day peeling potatoes, only to see
them shipped away to someone else. Wouldn’t you be hungry and angry?) Furthermore,
the centralized structure of the Communist party meant that information about the
disaster was slow to reach the top party officials, or was withheld along the way for fear
of disapproval. As Li and Yang note, “the GLF crisis could have been far less devastating
had local officials not faced strong political incentives to implement apparently poorly
35
conceived policies and to conceal unfavorable information on local economic
performance.”
Good decentralization: Claiming races for horses:
The public likes competitive races (bet more when odds are close). How to
arrange them?
Obvious, not-so-great idea: Somebody at the racetrack knows all the horses and
asks around, and puts them in races together.
Good idea: Announce a race for $20,000 purse and $20,000 claiming price. If
horses are entered they can be claimed (bought) by anybody for $20k. Horses worth more
than $20k won't enter. Bad ones won't either (too expensive to race in a hopeless race).
Bad decentralization in a nonprofit foundation: The William and Melinda
Gates Foundation (founded by Microsoft mega-billionaire Bill Gates) has $66 billion,
about half supplied by famous investor Warren Buffet. Their assets represent about 10%
of all the private foundation funds in the US (Los Angeles Times Jan 8 2007, “Money
clashes with mission”) and is twice as large as the next-largest foundation.
The Gates Foundation has awarded $1.2 million in grants to the nonprofit Solid
Ground, a Seattle Foundation that helps victims of “predatory lending”. Predatory lenders
are firms that encourage poor-credit borrowers to apply for loans and, allegedly,
sometimes falsify loan applications or encourage deception to get loans for borrowers
they would not otherwise qualify for. These borrowers are then burdened by large
payments they did not expect and some have had their lives turned upside down and net
worth drained.
One lender accused of these practices, Ameriquest, is owned by ACC Capital
Holdings Corp., which in January 2006 settled a class-action suit with 49 states and DC
for $295 million (without admitting wrongdoing). The company’s CEO Aseem Mital
said, "Doing the right thing for the people we serve has always been one of our core
values. We regret those occasions when our associates have not met this ideal."
Since the Gates Foundation seems committed to helping borrowers avoid these
challenges (their “preferences are revealed” by their grants to the Solid Ground
nonprofit), you would think they have no interest in profiting from so-called predatory
lenders. But at the end of 2006, the Gates Foundation had more than $2 million invested
in Ameriquest.
Why is the Gates Foundation investing in nonprofits who help people avoid the
headaches caused by companies like Ameriquest, and investing in Ameriquest at the
same time? According to the LA Times:
The Gates Foundation did not respond to written questions about specific
investments and whether it planned to change its investment policies. It maintains
a strict firewall between those who invest its endowment and those who make its
grants.
36
The “firewall” is a deliberate communication gap between the division of the
Gates Foundation that maximizes its endowment through stock and bond investment, and
the division that spends money. This is a deliberate decentralization. It is designed
presumably to prevent conflicts of interest (which are carefully avoided by nonprofits)
like making grants to a nonprofit to promote awareness of AIDS, while also investing in a
textbook publisher that profits from AIDS-related publications.26
The deliberate decentralization means that the organization is helping avoid bad
practice B while the organization also profits from bad practice B. Other philanthropists
decry this practice. The LA Times reported:
[Senior vice president of the Rockefeller Philanthropy advisors, a nonprofit
foundation group that counsels foundation, Douglas] Bauer said he had this
message for the Gates Foundation and others that invest with a blind eye to the
consequences: Good returns from bad companies can cancel out the beneficial
effects of grants.
"In the extreme example, it would be a wash," he said. "Trustees should say to
themselves: 'As we think about the work that we are trying to do with this
foundation, which ultimately should contribute to the public good, we should
be making sure that everything we do is aligned to the mission, including the
investments we make.' "
Another nonprofit foundation had an interesting reason for avoiding centralization
of grants and investment. The LA Times reported:
If the Hewlett Foundation created a large staff to monitor the social impact of its
investments, said Eric Brown, the foundation's communications director, "The
L.A. Times might call us and ask why we have so much overhead."
Prediction markets: TO BE ADDED
Orange juice Roll study
3. Models of human nature TBA
A crucial ingredient for organizing people is a working hypothesis about human
nature. Do people only care about themselves? Or do they care about doing right, or
To be fair, nonprofits are carefully scrutinized for conflicts of interest which the Gates Foundation’s
“firewall” is designed to prevent. Since nonprofits are not taxed, one way to avoid taxes is to set up a
nonprofit which coordinates giving grants to activity A while holding stock in firms that benefit when more
activity A is done. A good way to avoid the appearance of conflict is to clearly separate the grant-making
and investment divisions of the nonprofit. The problem, of course, is that you may get the opposite result:
While the IRS is worried about the two parts of the nonprofit working together, it might be that the two
parts cancel out each other’s humanitarian efforts.
26
37
helping their friends, or something beyond narrow self-interest? Are there different types
of people in these categories; and if so, how can we identify them and match them with
the right kinds of jobs?
Job satisfaction: One model of human nature is that a happy worker is a
productive worker. Unfortunately, many studies measuring both job satisfaction and
productivity suggest the correlation between these two variables is zero.
To an economist, this makes sense in one way, and makes no sense in another.
On the one hand, if being paid more than you are worth produces satisfaction,
then a satisfied worker is an overpaid worker, which is bad for the firm. Thus,
maximizing job satisfaction should not be the goal of managers.
On the other hand, if wages reflecting compensating differentials—or psychic
income from jobs that are fun or awful—then if workers are happy at their work, you can
pay them less (since they earn “psychic income” from enjoying their work). So creating
satisfying jobs reduces the firm’s total wages paid (“the wage bill” in labor economics
terms).
Perhaps the reason the correlations between satisfaction and productivity are zero
is that these forces conflict. An autoworker on an assembly may not be happy but is
productive. A happy worker at a retail store, on the other hand, might be spending too
much time schmoozing and isn’t productive.
In any case, the job satisfaction model is not very helpful unless the job
satisfaction leads to productivity, to low turnover of employees, or to a willingness to
accept less wage and benefits (compensating differentials).
Organizational citizenship: Group psychology. Identity. Essentialism. Nike
tattoos. Gangs etc.
Social preferences and heterogeneity: Economists tend to use pure self-interest as a default
assumption, and it is often a good approximation. The belief in self-interest stems from a skepticism that
even when people talk a lot about fairness, helping others, doing what is right, and so on, when push
comes to shove their own self-interest will win out—or worse, the rhetorical arguments about fairness
are a way to cloak their self-interest in high-minded garb. As Stigler (1981) wrote, cynically, “when
self-interest and ethical values with wide verbal allegiance are in conflict, much of the time, most if the
time in fact, self-interest theory…will win.”27
Just below we will consider models in which some players are purely self-interested but others
have social preferences (perhaps accompanied by emotional feelings like envy, guilt, shame and so
forth). Despite skepticism like Stigler’s, there is a long history of models that attempt to formalize when
people trade off their own payoffs for payoffs of others (e.g., Edgeworth, 1881; “equity theory” in social
psychology; and Loewenstein, Bazerman and Thompson, 1989).
27
Stigler, G. (1981). Economics or ethics? Tanner Lectures on Human Values. S. McMurrin.
Cambridge, Cambridge University Press.
38
Sensible models of this type face a difficult challenge: Sometimes people sacrifice to increase
payoffs of others, and sometimes they sacrifice to lower the payoffs of others. The challenge is to
endogenize when the weights placed on payoffs of others switch from positive to negative. A
breakthrough paper is Rabin’s (1993), based on psychological game theory, which includes beliefs as a
source of utility. In Rabin’s approach, players form a judgment of kindness or meanness of another
player, based on whether the other player’s action gives the belief-forming player less or more than a
reference point (which can depend on history, culture, etc.). Players prefer to reciprocate in opposite
directions, acting kindly toward others who are kind, and acting meanly toward others who are mean. As
a result, in a coordination game like “chicken”, there is an equilibrium in which both players expect to
treat each other well, and they actually do (since doing so gives higher utility, but less money). But there
is another equilibrium in which players expect each other to act meanly, and they also do. Rabin’s model
shows the thin line between love and hate. Falk and Fischbacher (2005) and Dufwenberg and
Kirchsteiger (2004) extend it to extensive-form games, which is conceptually challenging.
A different approach is to assume that players have an unobserved type (depending on their
social preferences), and their utilities depend on their types and how types are perceived (e.g., Levine,
1998 and Rotemberg, 2004). These models are more technically challenging but can explain some
stylized facts.
Simpler models put aside judgments of kindness based on intentions, and just assume that people
care about both money and inequity, either measured by absolute payoff deviations (Fehr and Schmidt,
1999) or by the deviation between earnings shares and equal shares (Bolton and Ockenfels, 2000).
Charness and Rabin (2002) introduce a “Rawlsitarian” model in which people care about their own
payoff, the minimum payoff (Rawlsian) and the total payoff (utilitarian). In all these models, selfinterest emerges as a special case when the weight on one’s own payoff swamps the weights on other
terms.
These models are not an attempt to invent a special utility function for each game. They are
precisely the opposite. The challenge is to show that the same general utility function, up to parameter
values, can explain a wide variety of data that vary across games and institutional changes (e.g.,
Fischbacher, Fong and Fehr, 2003).
Experimental evidence suggests two important modifications to the view that self-interest is
reliable and basic:
First, not everybody is self-interested all the time. In fact, in experiments only a
minority of people will act purely self-interestedly when they know it will harm
somebody else, even when they can get away with it.
Second, people are different (or at least, in any given situation people will not all
act in the same way). Heterogeneity matters because it can interact with economic
structures in interesting ways.
The existence of limits on greed, and heterogeneity, is nicely illustrated by a story
in Dawes and Thaler (1988):
39
Several models have emerged of “social utility”. These models try to add
variables that represent emotions or other forces to the basic model of self-interested used
in game theory and labor economics. The idea is to calibrate parameters which represent
weights on the variables and use them to see whether the same basic model of social
utility can explain patterns in many different games.
One model was created by Fehr and Schmidt (1999). Call the vector of payoffs
created in a game X≡(x1, x2,…xn ), where xi is the payoff to player i (and there are n
players). They assume the utility of player i for the payoff vector X is
(1)
ui(X)=xi – α/(n-1)k=1n x [xk–xi]0 –β/(n-1)k=1n [xi –xk]0
where the function [x] 0 denotes the maximum of x and 0 (i.e., if x<0 then the function is
set to zero). The first term (after the own-payoff xi) represents “envy”, how a player’s
utility is reduced by knowing that other players are getting more than she is. The second
term represents “guilt”, how much a player’s utility is reduced by knowing that other
players are getting less than she is. The parameters α and β represent the “strength” of
envy and guilt relative to self-interest.
Experimental evidence from various types of games suggests that α and β are both
positive and α > β (i.e., envy is stronger than guilt; in fact β=0 is often a good
approximation and reduces the theory to one with one free parameter). In fact, some
observations suggest that people like being ahead of others, so that β<0 and people get
pleasure from having higher income status than others.
One problem with the Fehr-Schmidt formulation in (1) is that it does not
incorporate a conception of intention…
FINISH MORE FROM BOOK
ADD ROBERTO
TO FINISH suggests there are three types of people: Conditional cooperators;
purely self-interested agents; and “saints” who always cooperate, regardless of whether
others do. But rationalization…point is that there are differences and possibly structural
influences.
40
“Trust, but verify”- Sam Walton.
“Traits or states”: Good people, or good times?
“Love thy neighbor but keep your guard up/You can’t trust nobody when they hard up”—
hip hop lyric.28
Attribution theory is a branch of social psychology that deals with how people
make natural attributions of cause and effect. There are two main findings which will be
important later as we think about performance evaluation:
1. Attributions are often self-serving;
2. Attributions overattribute cause to traits of people rather than situational
influences (in Western populations).
Self-serving attribution means that if something goes wrong, you blame bad luck,
and if something goes right, you take credit. One study of corporate annual reports
showed exactly this type of difference in the way companies described their annual
performance (Bettman and Weitz? Admin Sci Q). There is some evidence [check this]
that men are more likely to make self-serving attributions than women.
The tendency to blame or credit a person, rather than a situation, was called the
“fundamental attribution error” by Nisbett and Ross. In formal terms, you can think of a
person’s unobserved “type” and a situation’s “type” as combining in an equation to
determine performance. For example, some jobs are hard and some are easy; some
workers are skilled and some are not. The overattribution to personal types means that
people in easy jobs will be thought to be more skilled than they are, and people in hard
jobs will be thought to be less skilled. It is as if the brain runs a regression with
performance on the left hand side, and variables for personal skill and situational
difficulty on the right hand side, and derives too high of a coefficient on personal skill.
Of course, it is difficult to establish whether a behavior is truly caused by a person
or by some situational factor that cannot be controlled. For example, the overattribution
to personal skill predicts that after bad seasons, sports coaches will often be fired, on the
grounds that they caused the poor season. But of course, a bad season probably is some
evidence that a coach is unskilled, so it is difficult to tell whether coaches are being fired
unfairly. Because it is difficult to measure attribution mistakes in naturally-occurring
settings, the best evidence comes from experiments in which situational factors can be
manipulated, and see whether subjects appreciate the importance of the situational
change.
A classic experiment exhibits what I call the “Anchorman effect”. In a group of
students, two students are picked at random. One is told to ask some questions and the
When I was Googling to find information on this topic I typed in “love thy neighbor but keep your
guard”. The top-ranked link contained the phrase “Love thy neighbor but keep your eye on ‘em”. Clicking
on that link led to a page with a link for “Potent Love Psychic $10”. Clicking through to that link got you to
an “insight guide”, such as “Mystic Phaedre” who could tell you “Is it true love? Are they ready to
commit? Will you change careers?” for only $1 per minute. Isn’t this empirical confirmation of the need to
keep your guard up?
28
41
other student has to answer them. All the others watch. After the questions-and-answers,
the students rate the person who asked, and the person who answered, on various
personality traits, including intelligence. Typically, the person who answered questions is
rated as less intelligent. Why? The answering person probably did not know some of the
answers and made mistakes. The person asking the questions, however, is in a situation
where he or she cannot possibly make the same mistakes. The students rating them
should adjust for this difference in the roles they have, of course, but typically do not. I
call this the anchorman effect because news reporters on TV are comfortable on camera,
and usually ask questions (typically written down for them, or on a teleprompter). The
interview subjects are less practiced on camera, often nervous, and don’t know the
questions in advance.
Recently, Nisbett and colleagues have been studying whether overattribution to
individual traits is universal across cultures. It appears not to be; subjects from Asian
cultures often are quicker to blame a situation for success or failure rather than a person,
so the overattribution to people seems to be not “fundamental” at all.
One study of oil company executives showed self-serving overattribution to
personal skill that illustrates these patterns perfectly. Bertrand and Mullainathan (FINISH
CITE) reasoned that oil prices are largely exogeneous—or situational, in the psychology
jargon— because any single oil company CEO has little to do with raising or lowering oil
prices (which are largely set by worldwide demand, OPEC decisions and supply shocks).
But if the board of directors, who sets the CEO’s compensation, do not realize the
influence of the oil price variable on profits, they will give the CEO large bonuses when
oil prices go up (and profits go up too). That is exactly what happens. But what happens
when oil prices go down? Then self-serving attribution kicks in, and the CEO’s
compensation does not suffer.
The scandals at Abu Ghraib (in which American soldiers were photographed
humiliating prisoners in various ways) seems to be an example. Any social psychologist
would have warned that taking lightly-trained soldiers (many from the National Guard,
who never expected to serve overseas), putting them in a hot, uncomfortable foreign
country, and having them guard people who are culturally different is likely to lead to at
least some trouble. But the Army’s response, to a large extent, was to treat the problem as
if it was caused by a few rogue soldiers.
FINISH Example of Asian vs white student attributions of skill and effort in educational
achievement (New York Times)
TO FINISH Money as symbolic (exchange of $ with Mrs. Wong, rejecting ultimatum
offers, “fine is a price” etc).
Class notes from insurance
Data from Ian on auction, coin auction, etc
42
TO FINISH
4. Game theory and incentives
A “game” is a term social scientists use for a mathematical description of a
strategic interaction. The elements of a game are: Players, strategies, information, the
order of moves (moving early or late is sometimes a strategy), outcomes, and utilities.
Outcomes are determined by the strategies players choose, and the information they have.
While we will often talk about financial outcomes as a canonical case (especially
in thinking about companies), an outcome is usually a physical event in the world—
getting promoted, bankruptcy, earning profits, an apology and money in a legal
settlement, getting custody of children, and so forth. We assume that people have
numerical utilities over outcomes. Usually we only need to assume that people can rank
which outcomes they like best (i.e., utility is “ordinal”—1st, 2nd, 3rd, and so on).
Game theory is a promising mathematical language to partly unify natural and
social sciences because games can be played at many levels of analysis—from genes to
nations. Here are some examples: Tennis players deciding whether to serve to the left or
right side of the court; the only bakery in town offering a discounted price on pastries just
before it closes; employees deciding how hard to work when the boss is away; an Arab
rug seller deciding how quickly to lower his price when haggling with a tourist; rival
drug firms investing in a race to reach patent; an e-commerce auction company learning
which features to add to its website by trial-and-error; real estate developers guessing
when a downtrodden urban neighborhood will spring back to life; San Francisco
commuters deciding which way to work will be quickest when the Bay Bridge is closed;
Lamelara men in Indonesia deciding whether to join the day's whale hunt, and how to
divide the whale if they catch one; airline workers hustling to get a plane away from the
gate on time; MBA's deciding what their degree will signal to prospective employers (and
whether quitting after the first year of their two-year program to join a dot-com startup
signals guts or stupidity); a man framing a memento from when he first met his wife, as a
gift on their first official date a year later (they're happily married now!); and people
bidding for art or oil leases, or for knickknacks on eBay.
There are many books on game theory. The goal in the next few pages is to give
the briefest sketch, and some notation, to equip you to grasp the essentials of game theory
as it can help us understand organizational life. If you do not have some other
background in game theory, and are serious about deeply understanding the theories and
examples which will be described later, you should definitely read other books.29
First, some notation. Usually we talk about a finite number (n) of players, indexed
by i. Player i’s strategy is denoted si. A vector of strategies, one for each player is
29
A good introductory book (low on math) is Dixit and Skeath (1999). More
mathematical books include Rasmusen (1994) and Osborne and Rubinstein (1995). Gintis (1999)
includes fresh material on evolutionary theory and experimental data, and tons of problems. The
heavy tomes which are used in graduate classes at places like Caltech include Fudenberg and
Tirole (1991).
43
denoted S={s1, s2, ….sn}. The part of this vector which removes player i's strategy (i.e.,
contains every other player's strategy) is denoted s-i. The utility of player i's payoff from
playing si when others are playing s-i is denoted ui(si,s-i).
An example: The “p beauty contest” game
Let’s start with an example. It is a simple game that has been played in many
experiments for money, so we can compare different theories with some data. In this
game each of N (>2) players in a group chooses any number in the interval [0,100] at the
same time, without talking beforehand. The numbers are collected and averaged. A target
number is defined, which is p times the average. Suppose p=2/3 for concreteness.
(Shortly we’ll see what might happen for different values of p.) The player whose
number choice is closest to 2/3 of the average, in absolute value, wins a fixed prize of
$20.30
This game is sometimes called the “p-beauty contest”, because of a famous
passage in John Maynard Keynes’s book The General Theory of Employment, Interest
and Money. Keynes wrote:
Professional investment may be likened to those newspaper competitions in
which the competitors have to pick out the six prettiest faces from a hundred
photographs, the prize being awarded to the competitor whose choice most nearly
corresponds to the average preferences of the competitors as a whole; so that each
competitor has to pick, not those faces which he himself finds prettiest, but those
which he thinks likeliest to catch the fancy of the other competitors, all of whom
are looking at the problem from the same point of view. It is not a case of
choosing those which, to the best of one's judgment, are really the prettiest, nor
even those which average opinion genuinely thinks the prettiest. We have
reached the third degree where we devote our intelligences to anticipating what
average opinion expects the average opinion to be. And there are some, I believe,
who practise the fourth, fifth and higher degrees.
Think of numbers in the game as the time at which you decide to sell in a rising
stock market. If you sell too early (e.g., at time 0) then you miss the chance to profit as
stocks are rising. But if you sell too late, by choosing a selling date number higher than
other people, then you’ll be too late—if everybody else sold before you, the price will
have crashed and you’re selling too low. So the trick is to sell just before the tide turns
and everyone else sells. This is equivalent to choosing a number which is below what
other people pick, but not too far below. And of course, they are trying to do the same
thing.
In our notation, their number choices are 0<si<100 for i{1,2…N}. The target is t
N
t=pk=1 si/N. Defining the winning number as si*=argmini |si-t| (that is, the strategy
which minimizes the distance between the strategy and the target wins). Then the utility
function is ui(si,s-i)=$20 if si=si* and $0 if si≠si* . We can write this utility function more
30
Ties are broken randomly, or the players can share the $20; in theory it does not matter which
tie rule is used.
44
compactly using an “identity function” I(a,b), which equals 1 if a=b and equals 0 if a a≠b.
Then ui(si,s-i)=($20)(I(si,si*)).
Before we proceed, think about how you would play this game with a group of
your classmates. Actually write down a number.
[Pause for your thinking.]
[Write down a number!]
I will describe two different analyses of this game. You can judge which might be
more plausible as a guess about how people behave, and what the different types of
analyses are good for.
One analysis, called a “cognitive hierarchy theory”, works like so.31 Different
players perform different numbers of steps (k) of strategic thinking. Players doing more
steps have figured out what the lower-step players will do. The overall behavior of a
group depends on what each of the k-step players does, and what percentages of such
players there are (call those percentages f(k)).
To start the process off, assume some people aren’t really sure what to do. They
don’t spend much time wondering what others will do and choose a lucky number or a
random number. Suppose, to make the theory precise, they are equally likely to choose
any number in the [0,100] interval. We call these folks “0-step thinkers”.
Other “1-step” players think they are playing against people who are doing 0steps of thinking. (Alternatively, you can think of them as just having no clue what others
will do, sometimes called a “diffuse prior” or “ignorance prior belief” in Bayesian
decision theory.) The 1-step players use their belief to calculate expected values for
different strategies and pick the strategy with the highest expected value. We call this
“best-responding”.
Why stop there? Some players are probably thinking even more strategically. We
can imagine that there are 2-step players who believe they are playing against 0- and 1step thinkers. The 2-step players have to make a guess about what percentage of players
are doing 0-steps and what percentage are doing 1-step. In our CH specification, we
assume that they somehow guess the relative percentages correctly, but ignore the fact
that other players are doing two steps of thinking (or more).
Continuing on, 3-step players believe they are playing 0-, 1-, and 2-step players.
They figure out what those players do, guess the relative proportions correctly, and
choose a strategic best response given their guesses.
Now let’s translate this model into notation. First define the probability that
strategy si) is chosen by a player doing k steps of thinking as Pk(si). To make things
simple assume that only integer choices are allowed so the strategies are 0, 1,… 100.
(There are 101 such strategies.) The notation is also written out as if there are just two
players; when there are many players the notation becomes a bit more complicated.
The 0-step players are simple. They have Pk(si)=1/101 (all numbers are equally
likely to be picked).
31
The form of this theory described here is from Camerer, Ho and Chong (2004). The basic idea
was proposed earlier by Nagel (1995), for the “p-beauty contest” game, and by Stahl and Wilson
(1995). See also Crawford, Costa-Gomes and Broseta (2001) for an elegant study of decision
rules using information about what payoffs subjects are looking at on a computer screen (cf.
Camerer et al, 1993; Johnson et al, 2002).
45
For higher-step players we need to compute their beliefs about the distribution of
others’ steps, the expected payoffs that result, and their strategy probabilities. These kstep players form the belief about the percentage of h-step types of gk(h)=f(h)/∑n=0k-1f(n).
For example, the 1-step player forms beliefs g1(0)=f(0)/f(0)=1 (i.e., the 1-step player
believes everyone is doing 0 steps of thinking. The 2-step player forms beliefs
g2(0)=f(0)/(f(0)+f(1)) and g2(1)=f(1)/(f(0)+f(1)). (The beliefs g2(n)=0 for n>2.)
A 3-step player forms beliefs about the percentage of 0-, 1- and 2-step players
g3(0)=f(0)/(f(0)+f(1)+f(2)), g3(1)=f(1)/(f(0)+f(1)+f(2)), and g3(2)=f(2)/(f(0)+f(1)+f(2)).
Given these beliefs, the k-step player (for k>1) computes an expected payoff for
strategy si of
Ek(si)= ∑j=0100 π(si, sj) (∑n=0k-1gk(n)Pn(sj)).
The term ∑n=0k-1gk(n)Pn(sj) is the average probability that the strategy sj will be
chosen by the other players, weighted by the fraction of players of different steps. The
Ek(si) summation is then taken over all the possible strategies sj. Finally, the k-step player
chooses the strategy which has the highest expected payoff. That is,
Pk(si)= 1 if si= argmaxs Ek(s)
To make the theory precise, we need to figure out what the distribution f(k) is.
Because the thinking steps are thought to be discrete, we need a distribution that only
gives percentage for integer values of k (or we need to take a familiar distribution, like a
Gaussian bell-curve, and discretize it somehow). It is also useful, for doing statistical
estimation and for proving theorems, to have a distribution f(k) whose shape can be fully
characterized by only one parameter (or two at most). Finally, doing more and more
thinking steps is hard cognitive work. A 3-step player has to figure out what 0-step
players do, what 1-step players think 0-step players do (to deduce what 1-step players
will do), guess the relative percentage of 0- and 1-step players, and then combine all
these pieces to choose a best response. So it is plausible that the relative percentage of
players who do more and more steps of thinking goes down as k goes up. For example,
suppose that the fraction of players who stop at k-1 steps, rather than going on to do just
one more step, f(k-1)/f(k), is proportional to k. That is, as k goes up more and more
players stop at k-1 steps rather than doing one more step.
It turns out the proportionality assumption f(k-1)/f(k)  k implies a unique
distribution, the Poisson distribution f(k|)=e-(k)/k! The Poisson distribution is cool! It
has a French name that is fun to pronounce (“poisson” means fish in French), drops off
very sharply as k goes up (because of the proportionality assumption that is
mathematically equivalent), and has the constant e in it. Its shape is completely
determined once you specify the parameter , which is the mean and the variance of the
Poisson distribution.
But what’s ? Our approach is to let data tell us.32 That is, we collected a lot of
experimental data sets, derive predicted distributions of strategies for different values of
32
You could try to derive it mathematically from an analysis in which players compare
the costs of doing more thinking with the benefits of making a better guess, but that is hard to do
without knowing more about what cognitive “costs” really mean.
46
, and see which value of  gives the best fit to the data. Generally we find that =1.5
usually gives a good fit.
Figure ? shows what the CH theory predicts when =1.5 (see Camerer and Fehr,
Science 06). The theory predicts that every number has a chance of being chosen
(because of the 0-step players randomizing), but that most people will pick numbers from
about 20-33.
Standard game theory takes a very different view of the p-beauty contest game. In
standard theory the central concept is the idea of an “equilibrium” pattern of play. When
47
players are “in equilibrium”, they have either figured out by introspection, or learned
from experience or some other mechanism, to guess accurately what other players will
do, and they choose best responses. That is, if players are in equilibrium they will not be
surprised when the choices of other players are revealed.
Here is a mathematical definition of equilibrium. The strategies (si*, s-i*) are
equilibrium strategies if ui(si*,s-i*)>ui(si,s-i*) for all players i, and all opponent strategies
s-i. In plain English, if a strategy si* gives the highest utility (or equally-high utility—note
that the inequality is weak, not strict) compared to all other strategies si, assuming the
other players are choosing their equilibrium strategies s-i*, then all the strategies si* are in
equilibrium.
The idea of an equilibrium as I’ve defined it was first proposed by John Nash. He
won the Nobel Prize in 1994 and his amazing life was portrayed in the Oscar-winning
movie “A Beautiful Mind”. (The movie was based on a book by Sylvia Nasar, which
goes into much more interesting detail about game theory than the movie did.)
I think the right way to think of an equilibrium is as the ending point of some
adaptive process. Indeed, in his thesis Nash talked about an equilibrium arising after
“mass action” by a population of players adjusted by trial and error, until the players
could not profit by changing their moves. But if an equilibrium comes about from
adaptation (or possibly communication, imitation, or evolution of a cultural or biological
type).
In any case, the concept of equilibrium is valuable for at least two reasons. One is
that in finite games, an equilibrium always exists. That means that there is always some
kind of prediction a modeller can make. (Indeed, as we will see below there are often
multiple equilibria, so then the question is which equilibrium is more likely to occur.)
Second, in experiments and naturally-occurring processes most kinds of learning and
adjustment from experience will move people in the direction of the equilibrium. So even
if people are not in equilibrium right away, they will tend to move in that direction,
sometimes quite rapidly.
What does Nash equilibrium predict in the p-beauty contest game for p=2/3? The
answer is simple, but surprising. If other players are expected to choose numbers with an
average of A, then a player should choose si =pA. But if this condition holds, si <A so the
players who are choosing A are choosing too high. They are not choosing their best
strategy; they are out of equilibrium. In equilibrium, since the game is symmetric,
everyone should be choosing the same strategy, so that si=psi, or si=0. Furthermore, the
Nash equilibrium is zero for every value of p<1 (even .9999). The value p=1 is quite
different however—now everyone wants to pick exactly what everyone chooses so there
are multiple equilibria. If everyone thinks the average will be 14, they’ll pick 14. If they
think it will be 62 they will pick 62.
What happens in the beauty contest game with p=2/3? Figure ? above shows the
predictions of the CH model (with =1.5), the Nash prediction of 0, and some data from
an experiment with Singaporean engineering students (see Camerer, Ho and Weigelt,
1998). While the CH model does not fit perfectly, it capture the key properties of the
data—very few people choose numbers close to zero, and most choose numbers between
10-50 or so.
Table ? shows data from many different subject pools playing with p=2/3 for
modest stakes, and the values of  which fit each data set best. There is definitely some
48
apparent variation across groups, but there is also a surprising amount of regularity
(compared to the extreme Nash prediction of 0).
When the game is played repeatedly, and subjects learn the winner each time,
they do tend to converge to the Nash equilibrium of 0 rather rapidly. So the Nash
prediction is useful as a guess about the limiting point of an adjustment process. But
“behavioral game theory” (Camerer, 2003) fills in the rest of what we would like to
know: What happens the first time people face a new game? And also, how rapidly do
they learn (an important topic which is beyond the scope of this book)?
Finally, note that there is an important link between the CH approach and
equilibrium. In the p-beauty contest game (and other games which are “dominancesolvable”), think about what happens if  grows very large. As ∞, the prediction of the
CH model approaches the Nash prediction of 0. So in this game, you can see the Nash
prediction as a kind of extreme version of limited strategic thinking, in which there are no
limits. Mathematically, for this class of games, the Nash prediction is a special case of the
more general CH theory.
Basics of game theory
Now that we have a game in mind, and two kinds of theories, based on limited
strategic thinking and equilibrium, we can back up and define some of the key building
blocks of game theory.
Let’s start with the principle(s) of dominance. A strategy di is a strictly
dominant strategy if it is a strict best response to any feasible strategy that the others
might play. That is, ui(di,s-i)> ui(si,s-i) for all si ≠ di and for all s-i. (Sometimes a strategy
di only dominates one or more other strategies si; then the phrase “for all si ≠ di” in the
last sentence is replace by “for some si ≠ di”.)
Sometimes one strategy is better than another for some choices by other players,
but is not always strictly better and is never worse. Such a strategy di is called weakly
dominant. Strategy di weakly dominates si if ui(di,s-i)> ui(si,s-i) for some s-i and ui(di,s-i)>
ui(si,s-i) for all s-i.
Strict dominance is a useful idea because there is no reason why a player should
not choose a strictly dominant strategy, if the utilities fully reflect what the player most
wants.33 Choosing a (weakly) dominated strategy is not necessarily a mistake (as
choosing a strictly dominated strategy is), but it could be a mistake, depending on what
other players do.
Consider the simple normal-form game below. In a normal-form (a/k/a strategicform, or matrix) game players are presumed to move simultaneously so there is no need
to express the order of their moves in a graphical tree (or extensive-form). Each cell
shows a pair of payoffs. The left payoff is dollars for the row player and the right is for
the column player. The payoffs are utilities for consequences. That is, in the original
game the consequences may be money, pride, reproduction by genes, territory in
33
Later we will talk about what happens when the payoffs are in some observable amounts, like
dollars, and players care about how large the other player’s payoff is, because of psychological
forces like envy or altruism. Then we need a theory of “social utility” that specifies how different
dollar payoffs to all players create utility.
49
wars, company profits, pleasure or pain. A key assumption is that players can express
their satisfaction with these outcomes on a numerical utility scale. The scale must at least
be ordinal-- i.e., they would rather have an outcome with utility 2 than with utility 1-- and
when expected utility calculations are made the scale must be cardinal (i.e., getting 2 is as
good as a coin flip between 3 and 1).
Table ?: A dominance-solvable game
T
B
L
30, 20
20, 20
R
10, 18
20, 18
Suppose you were the row player, deciding between T and B. What would you
choose? If you choose B you earn 20 for sure. If you choose T, you are basically
gambling that the other player is more likely to choose L. To figure that out, you should
think about what the other player’s payoffs are.
For the column player, strategy R is strictly dominated by L (because L always
gives 20 and R always gives 18). “Deleting” strategy R—a mathematical way of betting
heavily that a rational column player will never choose R—then makes the effective
payoff from T higher than B, which means T strictly dominates B after R is “deleted” (or
assumed to be impossible). So if you are willing to bet that the column player is rational
and wants to earn the most money, you should choose T. The unique Nash equilibrium is
therefore (T,L).
What does the cognitive hierarchy theory predict? If =1.5, then 11% of the
players are 0-step column players who choose R half the time. All other k-step players
(with k>0) choose L. For row players, 0- and 1-step thinkers randomize equally between
T and B. Row players doing 2 or more steps of thinking will figure out that half the 0step column players pick R and the 1-step column players all pick L. Their belief about
what will happen is therefore a (.5)(.22/.55)+(1)(.33/.55)=.80 chance that L will be
played. (The fraction .22/.55 is the percentage of 0 players if there are only 0 and 1-step
player and =1.5.) So if they are neutral toward risk (i.e., they maximize expected value),
T will give an expected value of (.80)(30)+(.20)(10)=26, which is better than 20.
The overall percentage of players predicted to choose different strategies by the
Nash equilibrium and the CH model are shown Table ? below, along with some data from
my class in the winter of 2005. In this case, the Nash equilibrium is not far off, because
most column subjects guessed correctly that hardly anybody would play R. But even in
this sophisticated group—about one quarter of them had taken a game theory course—the
CH model fits a little better than Nash overall.
Table ?: A dominance-solvable game with predictions and data
T
B
Nash
L
R
Nash
CH
30, 20
20, 20
1.00
10, 18
20, 18
.00
1.00
.00
.73
.27
Data
06
.81
.19
Data
07
.86
.14
50
CH
Data 06
Data 07
.89
.95
.95
.11
.05
.05
Mixed Strategies
A mixed strategy for player i is a probability distribution over all the strategies si,
Consider the game below, a variation of “matching pennies” (or rock-paper-scissors).
In this game, the row player wants the strategy pair to “match” on the diagonal (either
(T,L) or (B,R)) and the column player wants to mismatch (either (B,L) or (T,R)). If X=1
the game is symmetric. Notice that there are no equilibria in which players choose a
“pure” strategy (i.e., play one strategy with probability one, or play it every time if they
are playing repeatedly). How can there be an equilibrium then?
The answer is that strategies are mixtures of “pure” strategies. The easiest way to
solve for equilibrium mixtures is to just write down a system of utilities and maximize
them. Suppose the row player’s probabilities of playing T and B are p and 1-p and the
column player’s probabilities are q and 1-q (as shown in the table). Denote these strategy
mixtures by p and q, respectively. Dropping out the zero terms, the row player’s expected
utility is urow(p,q)=X(pq)+1(1-p)(1-q)=(X+1)pq+1-p-q. The column player’s expected
utility is ucolumn(q,p)=1(p)(1-q)+1(1-p)(q)=p+q-2pq.
The row player wants to choose a value of p (her strategy mixture) to maximize
her expected utility. We can find that optimal value by differentiating urow(p,q) with
respect to p and setting the derivative equal to zero. Doing so gives (X+1)q*-1=0 or
q*=1/(X+1). Differentiating ucolumn(q,p) with respect to q gives 1-2p*=0 or p*=1/2.
This is odd, isn’t it? This is an equilibrium because both players are bestresponding given their beliefs. If the column player’s mixture of L and R uses
probabilities (1/X+1) and X/(X+1), then the row player can expect to get X with
probability 1/(X+1) if she plays T, and 1 with probability X/(X+1) if she plays B. Since
these are the same, she is indifferent to whether she plays T or B. Because she is
indifferent, she should tolerate choosing any probabilistic mixture of the two.
Similarly, the column player’s expected payoffs for L and R are the same, so she is
willing to mix between them since neither strategy yields a higher expected payoff than
the other.
The weird part is that the row player is supposed to mix between T and B equally,
for any value of X, but X affects the column player’s mixtures. Intuitively, as X goes up
the column player is more and more likely to choose R, as if she wants to keep the row
player from getting the high X payoff. But if she is more likely to choose R, the row
player wants to choose B. The only pair of mixtures that are mutual best responses are
(1/2,1/2) for row and (1/(X+1), X/(X+1)) for column.
Table ?: Asymmetric matching pennies (x>1)
T (p)
B (1-p)
L (q)
X,0
0,1
R (1-q)
0,1
1,0
51
Mixed-strategy equilibrium is a curious concept. Introducing mixed
strategies makes the space of payoffs convex (i.e., for any two points in the space, all
points in between are in the space too), which is necessary to guarantee existence of a
Nash equilibrium (in finite games). Guaranteed existence is a beautiful thing and is part
of what makes game theory productive: For any (finite) game you write down, you can be
sure to find an equilibrium. This means that a policy analyst or scientist trying to predict
what will happen will always have something concrete to say.
However, the behavioral interpretation of mixing strategies is dubious. By
definition, a player only desires to mix when she is indifferent among pure strategies,
which means she does not (strictly) desire to mix with particular probabilities; she just
doesn't care what she does. Furthermore, one player's equilibrium mixture probabilities
depend only on the player's payoffs which is odd. A modern interpretation of mixedstrategy equilibrium (called ``purification") is that one player might appear to be mixing
but is actually choosing a pure strategy conditional on some hunch variable they privately
observe. Mathematically, this works the same way-- as long as each player's belief about
the other players' choice matches the predicted probabilities, the mixed equilibrium is a
mutual best-response point.
Mixing makes more intuitive sense when there is hidden information (as we’ll
discuss further below). For example, a liquour store in the South that I heard about had a
sign prominently displayed: “A vicious dog guards this store three nights a week. Try to
guess which ones.” If a would-be burglar couldn’t outguess the store owner, and a 3/7
chance of being attacked by a vicious dog is enough to deter a burglar, then the clever
owner is able to save the costs of having a watchdog on hand while deterring the burglar.
Mixing also makes some sense when a game is played over and over and the
player’s histories are visible to other players. An example is serving to the left or right
side of the court in tennis or kicking left or right in a soccer overtime shootout (where a
single kicker faces a goalie and no other players are involved). Some field studies of
these sports have shown that it appears that players do randomize fairly well by not
exhibiting any predictable patterns (Walker and Wooders, etc., Levitt, Palacio-Huertas).
Prisoners' dilemma
The table below shows a famous game called the prisoners’ dilemma (PD) in a
specific numerical form. In this game D (for “defect”) is a dominant strategy. But if both
players choose D they each earn 1. If they could both commit to cooperate (C) they
would each earn 2. The dilemma is that if players behave individually irrationally (and
care only about their own payoffs) then their collective payoff is Pareto-inferior (i.e., they
earn (1,1) together rather than (2,2)).
C
D
C
2,2
3,0
D
0,3
1,1
In the general form, the PD looks like this: The payoffs are often labeled “t” for
temptation and “s” for sucker. A PD is defined by the algebraic constraints c>d (both
52
players would be better off cooperating if they could); t>c, d>s (so that D dominates C);
and (t+s)/2 < c (so that alternating between (C,D) and (D,C) in a repeated PD is not a
better strategy than choosing (C,C)).
C
D
C
c,c
t, s
D
s, t
d,d
Empirically, when the PD is played for money payoffs, there is a high rate
of cooperation (typically 50%) even if the game is only played once. The rate of
cooperation does respond to the variables you would think it responds to—if you raise the
relative payoffs t-c and d-s, there is more defection; if you raise the gains from
cooperation, c-d, there is more cooperation. However, by far the strongest variable
influencing cooperation rates is preplay communication about the game. If players talk
and agree to cooperate, they appear to feel obliged to actually cooperate as they promised
to.
Coordination
Many interesting games have multiple equilibria which have different payoff
consequences. These games capture the problem of “coordination”.
An important game in this category is “battle of the sexes” (BOS), shown in
Table? Two people, Pat and Chris can choose a big fish or a little fish. (This version is
modeled on anthropological situation in which two people in a small-scale society are
trying to decide how to divide two differently-valued goods.) Both prefers to divide the
two fish between themselves, so that one gets the big fish and one gets the little fish, but
each prefers to get the big fish. If the players mismatch, and both choose the same-sized
fish, they get zero. (If the payoffs from both sides grabbing for the little fish are between
1 and 3, then this a game of “chicken”.34).
The BOS game is not dominance-solvable. Neither strategy is dominant (or
dominated) for either player because there is no one strategy that is always best. Put
differently, each strategy might be best depending on what you think
the other person will do.
The BOS is a perfect game to illustrate “mixed motives”: Both players want to
agree on one of the two divisions on the diagonal of the table (since either division if
preferred to getting zero) so they have a common interest in choosing one of the
diagonals. But each prefers a different division, so even if they can agree to choose one
diagonal—either by talking, or through historical precedent or some other mechanism—
they directly disagree about which of the two diagonal outcomes is better (for them).
Table: Battle of the sexes (BOS) game
Chris’s choice
34
MORE HERE
Big fish
Little fish
3,1
Pat’s choice
Big fish
0,0
53
Little fish
0,0
1,3
Another interesting coordination game is called “stag hunt”, or the “assurance
game” (see Table ?). In this game players can play it safe and earn 1 for sure, by hunting
for rabbit. Or they can join a stag hunt, and earn a larger amount x (with x>1) if the other
person also hunts stag. However, hunting stag is risky because if the other players hunts
rabbit the stag-hunter gets 0. So hunting stag is a bet that the other player is likely to hunt
stag too. It is called an “assurance game” because hunting stag requires an assurance that
another player will help out.
Below is a picture of a “whale hunt” which is a naturally-occurring analogue to
stag hunt. Whale hunters in Indonesia, for example, go out in a large boat to hunt for
sperm whale35 and other “big game” (mostly manta and devil rays). The alternative is
fishing with a line and net in a one- or two-person boat. Large boats (average crew size
11) were able to go out when whale kills were common, coordinating activity, but boat
owners had trouble fielding sailors to join them when whale kills declined.
Figure: A naturally-occurring whale hunt game in Indonesia.
Michael Alvard and David Nolin, “Rousseau’s whale hunt: Coordination among big-game hunters”
Current Anthropology, 43, August-Oct 2002, 533-549
http://anthropology.tamu.edu/faculty/alvard/downloads/Whale_paper.pdf#search='whale%20gintis').
35
54
The stag hunt game is arguably a better model of organizational decisions than the
prisoners’ dilemma. A PDAuthority in organizations often gives bosses the power to
demand that players choose “stag” (i.e., go along with a risky corporate plan) or be fired.
MORE HERE
Table: Stag hunt game (x>1)
Chris’s choice
stag
rabbit
Stag
x,x
1,1
Pat’s choice
rabbit
0,1
1,1
Betting games: DETAILS TBA
1 and 2 are each privately informed about a "set of states"
1 will know "It's A or B" (A B) or "It's C or D".
2 will know "It's A" or "It's B or C" or "It's D".
A
B
C
D
1's payoffs
+32 -28
+20 -16
2's payoffs
-32
+28 -20
+16
e.g. A="earnings will be great"
B= "earnings will be good"
C= "earnings will be bad"
D="earnings will be terrible"
1 knows whether earnings will be (good or great) or (bad or terrible)
2 knows whether earnings will be surprising-- either great, (good or bad), or terrible.
When should you bet?
Game theory in action:
The point of the Groucho Marx theorem is to humble you. If a deal looks too
good, you should always ask— Am I missing something here? Similarly, if you have
created a plan that you think can fool somebody else, who is experienced and on the
lookout for trickery, you should think very hard about whether you are smarter than the
other guy or gal.
A gruesome example of this is the sorry tale of University of Pennsylvania
economist Rafael Robb. Robb, 56, lived in a home with his wife legally separated for 10
years (she had filed for divorce in 1993 but the suit was dropped in 2004). He came home
one afternoon in December, 2006 and, he told police, found his wife brutally murdered,
the presumed victim of a burglary attempt (since a window was smashed in).
Police were immediately suspicious because in cases like these, the spouse is
always the initial and primary suspect. Glass from the window did not appear to be have
55
been stepped on and crushed, as it would if a murdered came through the window from
the outside.
It later came out that Robb’s wife was planning to divorce him after their long
separation. Robb was arrested on January 8, 2007.
Figure ?: Professor Rafael Robb: Guilty of hubris and murder or neither?
A possibly ironic touch comes at the very end of the AP story (“Penn Professor
Charged in Wife’s Slaying”, jan 8, 2007):
Penn officials said earlier that they had arranged for someone else to teach Robb's
graduate seminar in game theory this semester.
Rafael Robb was a game theorist! He almost surely teaches the “Groucho Marx
Theorem” in his class, but, perhaps, did not believe it in practice— if he did commit the
crime, he thought he could outwit seasoned homicide cops who have seen hundreds of
crime scenes.
Game theory and organizational contracting
Before shaking hands: Hidden information
Until the 1970’s or so, game theory was only capable of addressing strategic
thinking when both players have the same information, which is quite unrealistic in most
organizational contexts. Workers generally know how hard they have been working and
usually have an incentive to hide what they know from employers. Other times the
opposite is true—employers may know something the employees don’t because they
have more experience, like how dangerous high-rise construction work is or how
frequently biotech projects lead to valuable products.
As is usually true in scientific modeling, game theorists were well aware that
people often have different information—and people with less infomraiton may know
that there is something they don’t know, but don’t know what that something is-- but they
didn’t know how to model such situations. Then in 1967 John Harsanyi showed how, for
which he shared the Noble Prize in 1994.
56
Harsanyi suggested studying these situations by assuming that each player has
some private information, summarized by their “type”. The key step which makes the
modeling workable is that the chance that each player is of a particular type is
“commonly known”—which means both players know the chances, both know that
others know the chances, and so on. This, too, is unrealistic but Harsanyi’s framework
provided a useful start.
The ability to mathematically model situations with differences in information
rapidly drew attention to the influence of two different kinds of informational
asymmetries in organizational life. One is called “adverse selection” or “hidden
information”. The term “adverse selection” comes from the insurance business. Consider
car insurance. Insurance companies can ask a few basic questions before deciding which
drivers to insure—such as age, gender, accident history, zip code, how many miles the
driver usually drives, and the type of car the driver owns. But these variables do not
perfectly predict whether a person is a safe driver who is likely to get in accidents and file
insurance claims. Remember that in a competitive market, insurance companies will offer
insurance at a price which equals the expected cost of accidents (plus a premium to pay
for administrative overhead, and profit margin). Drivers will choose the policies with the
lowest prices.
Now suppose the drivers themselves know something about how safe a driver
they are, which the insurance company doesn’t know. Then, in theory, the drivers who
realize they are unsafe drivers are more likely to buy insurance, if the price is less than
what they know the true expected cost to be.36 Thus, the self-selection of buyers who buy
insurance goes against the company’s interest—we say they are being “adversely selected
against”.
The way in which the kinds of people who select to undertake an action depends
on their hidden information happens throughout life and is central to modern economic
thinking. (Figure ? below shows how it might apply to savvy fish.) In labor markets,
selection is sometimes called “sorting” or “assortative matching”. For example, when
companies switch to a system where pay is more closely linked to measured output or
performance, performance often goes up. Part of the effect is probably due to the fact that
people work harder because they now earn more because of their extra effort. But part of
the effect could also be due to the fact that the higher bonus draws very hard-working
employees into the firm. (At the national level, this is familiar in the stereotype of hardworking immigrants who often literally risk their lives to migrate to a country where hard
work can earn them more money.)
36
We are sneaking in an important assumption here—that drivers who are reckless and those who are quite
safe have the same “risk tastes” in evaluating bad outcomes like accidents. If safe drivers really dislike risk,
and drivers don’t mind, then the safest drivers might buy insurance which solves the adverse selection
problem for insurance companies. This is called “propitious selection” or— a term I prefer and just made
up—beneficial selection. This consideration complicates the analysis a lot because it means we must
consider a driver’s “type” to have more than one dimension, and types who are high on one dimension (e.g.
reckless) might want insurance while those are high on another (e.g., risk-aversion) might want insurance
too. If the types are negatively correlated (reckless drivers are gamblers) then it is not clear what
equilibrium outcomes of insurance markets will result.
57
Of course, companies are often well aware of the adverse selection problem.
There are basically two techniques which help companies minimize adverse selection by
helping companies: Measure, or screen.
Measurement means trying to uncover the hidden information which potential
buyers have about themselves. Insurance companies use “experience rating”, which
means charging more for people who have had more accidents, or refusing to sell
insurance at any price for very high-risk groups. A related technique is to offer lower
rates to people who are members of a group, since the insurance company can gather
reliable statistics if the group is large, and simply being a member of a group may be
correlated with driving safety. For example, young drivers are notoriously accident-prone
and insurance companies dislike insuring them. Student with good grades in school can
often get discounts on their insurance rates, because insurance companies think (probably
based on large samples of accident rates) that students who are conscientious and sharp
enough to get good grades are also likely to be safer drivers than average.
Screening refers to creating a set of different insurance policies so that drivers
with different self-perceived ability choose different contracts. That way, the contracts
“screen out” the bad drivers from low-cost policies (which are priced to make profits only
if expected accident costs are low). A good example is deductibles. A deductible is the
amount that is deducted from the accident cost before the insurance company pays the
claim; or put differently, it is the amount the driver “copays” if there is an accident.
Drivers who believe they are really safe drivers will not mind a policy with a
large deductible ($1000 is typical for a large deductible) since they will not expect to pay
that amount very often. Bad drivers will prefer a lower deductible-- or no deductible at
all—because they think there is a high chance they will have an accident and have to pay
the deductible.
The difference in insurance premiums for different deductible policies is
enormous. No-deductible policies are wildly expensive. [IK put some numbers here].
58
This is consistent with the theory that deductibles are helping the insurance company
screen drivers by ability: The safe ones accept high deductibles and pay much lower
premiums; the riskier ones are happy with lower deductibles even though their premiums
are much higher.
A famous adverse selection problem was introduced by George Akerlof and then
studied by Max Bazerman and some other psychologists in experiments.
Imagine a company that is worth some amount V, uniformly distributed between
[0,100] where the value represents the per-share stock price. The current owners of the
company know its value. You are trying to acquire the company. From the point of view
of economic efficiency, suppose you can manage the company better (or there is some
synergy between this company and your own), so that whatever the value of V is, it is
worth 1.5V to you.
To make it simple, suppose you must make a take-it-or-leave-it offer P. The
owners will accept it if your offer P is above the value of the company, and turn it down
otherwise. Suppose your objective is to maximize the expected difference between the
price you pay and the value of the company to you.
What should you offer? Think carefully and write down an offer.
[Pause for thinking]
If you offer P then there is a P/100 chance that your offer will be accepted,
because it is accepted if P<V and V is uniformly distributed. If it is accepted, then—here
is the crux move—the conditional expected value is uniform between 0 and P. That is,
suppose you offer 60. Then if its value is between 0 and 60 your offer will be accepted.
The average of all these possible values is P/2. The value to you is 1.5 times as large, or
1.5P/2=.75P. But you paid P. Adding all the pieces together, your expected value is
(P/100)[.75P-P]+(1-(P/100))[0]= -P2/400. The higher bid, the more money you can
expect to lose!
The best bid you can make is zero (ust say no). This remarkable problem has
tricked hundreds of groups of people, from high school students to top business
executives. The tricky part is realizing that whether your offer is accepted or not
immediately gives you some information. If your offer is accepted it tells you that the
firm’s value is low rather than high. Ideally, when making an offer you should anticipate
this effect and bid accordingly.
Even when this game is played over and over, it is difficult for people to learn to
bid zero. One problem is that if you feedback is random. If you bid 60, for example, then
even though your expected value is -15 if the offer is accepted (that’s 1.5(30)-60), one
third of the time you make money (if V>40) and two thirds of the time you lose money.
Furthermore, even though people sometimes bid less after they lose, by bidding less they
typically have their offers rejected so the rate of useful feedback slows down. So natural
learning forces take a long time to teach people, from trial-and-error, that bidding a
positive sum is a losing strategy in the long run.
Winner’s curse MORE HERE
Insurance and market failure: A canonical example of hidden information is in
health insurance. Insurers routinely adjust rates or often refuse insurance at any price to
people in various professions or who prescribe to various medication (from LA Times,
59
January 8, 2007, “Health insurers deny policies in some jobs”). To an economist, it is
surprising that people with various observable jobs or prescription drug use would be
denied coverage at any price. Why?
One explanation of this practice comes from a simple model. Suppose people
have an observable condition that correlates with risk p of filing a claim of size X in a
year. Suppose, to simplify the example, the risks across people are uniformly distributed
from 0 to 1 and people are risk neutral (i.e., they will buy coverage if their expected cost
is above the price). The expected cost to an insurer of such a person is .5X per year. If the
company offers coverage at the price .5X, then only people with risks above .5 will buy
the insurance. Since risks are distributed from 0 to 1, the expected cost will be the cost of
those people who take the policy, which is those with risks from .5 to 1, or .75X. Then
the insurer must price at .75X to break even. But then people with risks less than .75 will
prefer to forego coverage, and only people with risks above .75 will buy insurance. But
the average risks of such people is (.75+1)/2=.875. So the insurer who offers a price of
.75 is losing money. For any price, the company is attracting only the highest-risk people
and cannot make money.
This stylized example is often used as an illustration of why insurance markets
can “fail”— i.e., why people who want to buy insurance at a fair price given their risks
cannot do so. But the model leaves out many details. For example, can’t insurance
companies figure out something about a person’s risk from their occupation and health?
In fact, companies try to do so but their measures seem draconian. For example, insurers
routinely deny coverage at any price to people in certain jobs or who are taking certain
prescription medications (see Table ? below). Some are very dangerous but others are
not.
Here are some occupations that could render workers ineligible for health insurance
under the policies of some insurance companies. (LA Times Jan 8 2007)
Air traffic control
Building, moving
Chemical/rubber manufacturing
Circus or carnival work
Concrete or asphalt work
Crop dusting
Firefighting
Furniture and fixtures manufacturing
Lumber work, including wood chopping, timber cutting and working in a sawmill
Migrant labor
Oil well or refinery work
Police work
Roofing
Sandblasting
Sports, semi-pro or professional
Stockyard work, with or without butchering
Stables, all employees
Stunt work
60
Telecom installation
Transportation and aviation
Tree climbing
Tunnel work
War reporting
Window work at heights exceeding three stories
Source: Health plans' broker guidelines
Eight of the 20 top-selling prescription drugs, ranked by their 2005 U.S. sales (which are
often grounds for uninsurability):
Lipitor (cholesterol)
Zocor (cholesterol)
Nexium (heartburn, ulcers)
Prevacid (heartburn, ulcers)
Advair (asthma)
Zoloft (depression)
Singulair (asthma)
Protonix (heartburn, ulcers)
ALSO ON THE LIST
Accutane (acne)
Allegra (allergies)
Celebrex (arthritis)
Celexa (depression)
Concerta (attention deficit)
Famvir (herpes)
Imdur (angina)
Imitrex (headaches, migraines)
Lamisil (fungal infections)
Parolodel (menstrual disorders)
Prozac (depression)
Ritalin (attention deficit)
Tagamet (heartburn, ulcers)
Tapazole (hyperthyroid)
Topamax (epilepsy, migraines)
Source: Health plans' broker guidelines
After shaking hands: Hidden actions
Another problem of limited information is called “hidden action” or “moral
hazard”. Hidden action means that one party has an incentive to take an action to benefit
themselves at the expense of the other. An extreme form is flat-out fraud and cheating.
Oliver Williamson uses the term “opportunism”, which he defines as “self-interest
seeking with guile”.
Hidden action often shows up in ordinary transactions. In many African countries,
bags of rice are sold by weight. Stones are sometimes added to the bottom to increase the
61
weight of the bag, and covered up with rice so the stones are not visible. Savvy buyers
know to dig their hands into the rice bag looking for stones before they buy.
Hidden action is common in one-time transactions, especially if one side has paid
in advance for a product or service. Once I hired a contractor to paint my house. We
signed a contract and I paid a deposit. The verbal agreement was for 2-4 painters to come
the next day and finish the job over a couple of days. The next morning only one painter
came; the job took a total of 5 days. In situations like this, the seller of the service has the
buyer over the barrel— once the house is partly painted, the buyer typically does not
want to cancel the contract and fire the sellers.
In professional service firms (such as law) where clients pay for billable hours of
service performed by professionals, overbilling can be common. In fact, it is possible that
an equilibrium emerges in which clients expect to be overbilled, but also expect to haggle
for a reduction in their total legal bill when a case or project is finished. If so, the firm
may tolerate overbilling because if they billed honestly the client might still expect a
reduction.
Cardboard
Sometimes markets flourish in one place, rather than another, because
geographical differences or other forces influence the extent of hidden action. For
example, a lot of used cardboard boxes are collected and sold in Los Angeles, to be used
again or recycled. Why Los Angeles? Cardboard is not more plentiful in Los Angeles
than in another large city, like Houston or New York.
The reason why the Los Angeles market flourishes has to do with the nuances of
how the market operates. It takes too much time to count the sheets of cardboard in a
large stack; it is easier to sell cardboard by weight. But selling cardboard by weight
creates the potential for hidden action. Cardboard is porous and becomes much heavier
when it absorbs water. This property of cardboard makes it hard for buyers to tell whether
they are overpaying because the cardboard they are buying is wet, or because sellers
watered the cardboard to increase its weight and sell it for more money. But Los Angeles
is basically a desert city; it only rains about 10 days a year. The used cardboard market
flourishes in Los Angeles because the air is dry so that variations in cardboard weight
from water are low. This fact reassures buyers who are paying by the pound and makes it
relatively easy to see if cardboard has been artificially watered to increase its weight.
Pizza barter
Pizza stores that have delivery to your home or office have Pizza delivery
Pizza delivery guy http://tipthepizzaguy.com/stories/story17.htm wild stories!! He would
take pizzas that couldn’t be delivered (which as a matter of policy the drivers were
allowed to return, and eat or give away) and keep them in his car. Occasinally somebody
would flag him down and ask if he had a spare pizza; he’d sell them the old undelivered
pizza and pocket the cash.
62
Some case studies:
The same driver would offer to pick up anything else somebody wanted (“beer,
cigarettes[sic], McDonalds fries, tampons, whatever…”) for a 100% delivery surcharge.
Moral hazard: Employee bartered cheap pizza (due to employee discount) for services.


I used to pay for my oil changes with pizza. I think we paid 25% of retail
for pizzas, so an XL one topper carryout was like $3. Equivalant to the oil
change guy, $20. Price of an oil change, $20. I got it for $3.
I attempted to pay for car repair wit pensive to pay for with pizzas. Pony
up the $. ;-)
h pizzas, but found myself being this guy's "pizza $@&#!" for two months. Car
repair is too ex
The same driver would offer to pick up anything else somebody wanted (“beer,
cigarettes[sic], McDonalds fries, tampons, whatever…”) for a 100% delivery surcharge.
Moral hazard: Employee bartered cheap pizza (due to employee discount) for services.


I used to pay for my oil changes with pizza. I think we paid 25% of
retail for pizzas, so an XL one topper carryout was like $3.
Equivalant to the oil change guy, $20. Price of an oil change, $20. I
got it for $3.
I attempted to pay for car repair with pizzas, but found myself being
this guy's "pizza $@&#!" for two months. Car repair is too
expensive to pay for with pizzas. Pony up the $. ;-)
(Near-) death from above: Air traffic control
One of the most remarkable studies of moral hazard is a study of air traffic
controllers, by Staten and Umbeck (1982).
Air traffic controllers work under great stress, to guide planes safely into major
airports. Because of job burn-out, the Federal Aviation Administration (FAA) has an
elaborate set of rules for detecting worker disability and remedying it. Like all
government employees, controllers are covered by the Department of Labor’s Office of
Workers’ Compensastion Programs (OWCP), starting in 1974. The Federal Employees’
Compensation Act (FECA) provides disabled workers with either a fixed award or a
63
percentage of salary paid for the duration of an injury. If the OWCP verifies that an
injury is disabling, and job related, controllers with dependents can earn 75% of their
base salary tax-free (which can higher than their after-tax pay for highly-paid controllers
with many years on the job).
Starting in September 1974, several amendments to the FECA act increased the
incentive for burned-out controllers to collect disability. One amendment was to allow
testimony of clinical psychologists, rather than M.D.’s, as primary medical evidence of a
disability. Another amendment allowed controllers to pick the private doctor of their
choice to supply medical testimony, rather than being examined by their airport’s flight
surgeon. These two amendments allowed controllers to “shop” for a doctor willing to
attest to their disability. These amendments came at a time when the OWCP,
coincidentally, abolished its investigative staff (who previously scrutinized dubious
claims) and assigned investigation to busy claims examiners with heavy caseloads,
without increasing the number of examiners proportionally to the increased caseload.
Furthermore, in May 1972 Congress authorized funds for a “Second Career Training
Program”, to provide up to two years of pay for job training, as well as paying 100% of
the base rate of pay, for controllers who had worked for more than five years.
A controller who files for disability with the intention of being deemed medically
unfit to continue is said to “punch out” of the job. The 1974 changes in the OWCP, the
reduction in claims investigation, and the Second Career incentive created a “perfecdt
storm” situation that is ripe for moral hazard by controllers who want to punch out and
get fully paid for two years while training for a new job.
Table ?
Disorder
Respiratory
Muscles
Ear,nose,throat
Abdominal
Eye
Bones and joints
Cardiovascular
Neuropsychiatric
Pre-Second Career
Program Change
incidence
1.9
.5
6.7
16.7
5.6
2.3
22.1
10.9
Post-Second Career
Program Change
incidence
1.5
.4
8.0
20.4
7.4
4.6
32.5
27.2
Percentage change
-21
-20
+19
+22
+32
+39
+47
+150
Table ? shows the increase in medical incidence (per thousand person years of service)
between the periods before and after the Second Career provision. Notice that reports in
all categories rose (except for rare muscle and respiratory cases). The increase was by far
the largest in the “neuropsychiatric” category, which almost tripled.
The number of controllers who were disqualified from duty for medical reasons,
who were not otherwise eligible for early retirement (older than 50, or with 25 or more
years on the job) also rose sharply during the mid-1970’s, from 143 in 1973 (before the
OWCP changes) to about 400 a year after 1975. This increase, and in the medical
incidence increases shown in Table ?, are rough clues that at least some controllers were
64
appealing for medical disability in higher numbers when the incentive to do so rose,
because it was easier to shop for a doctor to testify to disability, and because of the
generous Second Career provision (two years of full pay during retraining).
The most striking evidence, however, comes from specific incidents. In
evaluating claims of psychological burnout (as opposed to medical claims, which are
easier to verify), OWCP claims examiners “have been instructed to look for specific
stressful incidents on the job that were symptoms of or contributed to a disability”. In
fact, there are two prominent categories of “specific stressful incidents” the FAA
carefully tracks—“system errors”, and the ominously-named “near mid-air collisions”
(NAMCs). At air terminals, the minimum physical separation between incoming planes is
three horizontal miles, or 1,000 vertical feet. (Keep in mind that jets going 600 miles an
hour cover three miles of separation in 18 seconds, and descend before landing at about
1,000 feet in about a minute.) A NMAC occurs when aircraft are forced to take evasive
action, or come less than 500 feet apart (a distance they can cover in half a second).
Staten and Umback did a statistical regression of the number of system errors in
each quarter of the year, for each of 10 FAA regions, from 1973 to 1976, a period that
covers the time before and after the OWCP amendments. The biggest determinant of the
number of system errors is the amount of air traffic—there were .012 errors per aircraft,
or about one for every 80 planes.
Amazingly, the number of quarterly errors per region went up by 3.12 after the
OWCP amendments (a number which is quite significantly different than you would
expect by chance). It appears that some controllers were deliberately letting planes get
close together, creating system errors that could be verified by OWCP claims examiners,
in order to get out of their jobs and get two years of full pay and job retraining!
The controllers were not too reckless, however. A regression of the number of
near-miss NAMCs did not show an increase after the OWCP amendments. Furthermore,
before the OWCP amendments, about 40% of the system errors were reported by the
controllers themselves, and a third of those were also reported as NAMC’s. After the
amendments, the percentage of system errors the controllers reported themselves rose to
58%, but only a sixth were also reported as NAMCs. The data suggest the controllers
were deliberately letting planes get close together, but not close enough to create
dangerous near-misses. (If anything, they were more careful to avoid NAMC’s while
creating more system errors.)
As with any statistical analysis, it is important to disentangle causality from
correlation. It is possible that the OWCP amendments were passed because air traffic
control was getting more difficult, and more system errors would have happened anyway.
(The analysis tries to control for this by including the amount of traffic, however.) A final
piece of evidence suggests, again, that the increase in system errors was a deliberate
response to the increased incentive to get medical disability. Remember that controllers
needed five years on the job to become eligible for the generous Second Career full-pay
retraining benefits. In 1974, before the changes, 20% of the controllers had less than five
years experience, but accounted for 40% of the controllers with reported system errors (a
ratio of about 2); the corresponding ratio of error-prone controllers with 5-10 years of
experience, compared to the number of controllers with that much experience, was about
.8. These data suggest, not surprisingly, that inexperienced controllers were creating more
errors. But by 1976, the ratio of error-prone controllers among inexperienced
65
controllers—those ineligible for the Second Career benefit—was about 1.2, and the ratio
for eligible controllers (those with 5-10 years experience) rose to 1.8: That is, the more
experienced controllers were making more than their share of errors.
This amazing story shows how how hard it might be to detect a terrible response
to a change in incentives. System errors are relatively rare. The pattern detailed above
only came to light when two clever economists, armed with a theory that led them to be
suspicious that maybe increased incentives would change behavior, put together a lot of
data and saw the pattern. Furthermore, this true-life tale shows how carefully incentive
systems must be balanced. Changing the OWCP rules may have been a good policy to
combat air traffic controller burnout, but if the OWCP had anticipated an increase in
dubious claims, and had increased the number of investigators when the doctor-shopping
amendments and Second Career benefits were passed, the number of system errors might
not have risen so sharply, or at all.
What does behavioral economics have to say about the frequency of people
exploiting their hidden action? The evidence described in section ? on human nature
suggests that while self-interest is common, reciprocity springing from moral obligation
is common too. The effects observed in experiments make it hard to believe that selfinterest seeking “with guile” is so prevalent in the business world.
Hidden action and hidden information are thought to be the central issues in
businesses like insurance. In principle, insurance is a simple business because no physical
product is produced: The insurance company simply creates a contract spelling out the
situations in which it will or will not pay an insurance claim. The goal of the contract is
simply to shift risk from the insured to the insurer, while maintaining the insured party’s
incentive to take care, and protecting the insurer from the risks of too many insured
parties with “bad” hidden information buying its insurance. In theory, insurance markets
“fail”— that is, people want insurance but can’t buy it, or cannot buy as much as they
want—because of hidden information and hidden action.
At least one example suggests, however, that these two forces may not be the
whole story underlying underprovision of insurance. The example is the market for
insurance of jockeys who ride thoroughbred racehorses. Jockeys usually weigh about 100
pounds, and horses weigh 1500 pounds, racing in close quarters at speeds of around 40
miles per hour. (You can imagine the danger by imagining what it is like to cling to the
top of a small Mini-Cooper car, in a race around turns at 40 miles an hour with other cars
only a foot or two away from you, and sometimes bumping into you.)
Riding horses is extremely dangerous—even the best riders are seriously injured a
couple of times in their careers, and some are killed or disabled. (Ron Turcotte, most
famous for riding Secretariat, ended up in a wheelchair after a terrible spill during a race.)
Yet jockeys complain that they can rarely get as much insurance as they would like. For
example, in YEAR, 15 jockeys protested the low ceiling on their accident insurance,
$1000 per incident, by refusing to ride for a couple of days at Churchill Downs (the home
of the Kentucky Derby). The track, incidentally, took the side of the jockeys by banning
the 15 boycotters from riding at their track for the rest of the racing season that year (and
also found another carrier who would extend the cap to $500,000 for a modest additional
premium of $100-200). CITE
66
What makes this example important is that, in theory, hidden information and
hidden action which makes underwriting insurance hazardous for insurance companies
can be largely avoided by “experience rating”—having a large sample of clear evidence
of how accident-prone an insured person is— and monitoring how carefully they take
steps to ensure their own safety when they are covered by insurance. But in horse racing,
every single time a jockey rides in a race, the entire race is filmed and available for
scrutiny. So except for some other activities that might a jockey unsafe (such as losing
too much weight, or a drinking or drug problem), there really is no hidden information or
hidden action in insuring jockeys for accidents during a race. In theory, insurance
companies can see for themselves whether a jockey rides recklessly, and can judge from
a film when they are riding even more recklessly if they are heavily insured. So it is a
puzzle, from the point of view of standard theory, why insurers are unwilling to provide
more insurance to jockeys who say they are willing to pay for it.
IAN KK Post it: “Why does insurance market fail? Adverse selection and moral hazard.”
Article title: They Have the Run of Churchill Downs
Article gist: 15 jockeys were barred from racing at Churchill Downs for the rest of the
season after they tried to protest their accident-insurance by sitting out a couple days. The
insurance was a group program that has a ceiling of $100,000 per incident. Churchill has
supposedly found a carrier that will write a $500,000 individual supplementary policy for
a $100-200 premium, and “several riders have signed up for the coverage.”
______________________
Preventing damage from hidden information and action
The most important force that reins in cheating is reputation from repeated
interactions. If one person cheats another, the cheated party usually finds out and quits
trading with the cheater. Sometimes the cheated party can sue for fraud or publicize what
happened. The online auction site EBay developed a novel system of “testimonials”:
Buyers who bought goods from a seller can post online comments about the quality of
service from that seller. New buyers can see the comments; if they’re bad that hurts the
seller’s business a lot. [give specific example here FINISH]
If you think people are usually guileless in trying to cheat others, you’d be
surprised how well the EBay system works. People routinely send checks in the mail and
receive goods from people far away who they’ve never met, and may never buy from
again. At the same time, the system does not work perfectly, and the amount of fraud
from online internet transactions has increased dramatically in recent years. EBay
eventually introduced an escrow system in which the buyer can deposit money with
EBay, which is transferred to the seller only after the buyer receives their goods and
reports their satisfaction.
Holdup
A special kind of hidden action is called the "holdup problem". The potential for
holdup occurs when one party makes a large investment which is only valuable in a
trading relationship with another party. After the investment is made, the investing party
67
is at risk of being “held up” by the other party renegotiating the deal or, more subtly,
being stingy in negotiations as quality and price are adjusted over time. To the extent that
the investment is irreversible (or “sunk”), and specific to the relationship, a profitmaximizing investing firm has no choice but to accept the unpleasant result of new
negotiations.
For example, it is common after getting married for men to gain a little weight,
and for women to cut their hair shorter or spend less time on their personal appearance.
(These kinds of changes don’t appear to be permanent changes in tastes due to marriage,
since after divorcing, it is common for men and women to go back to their single ways.)
The most charitable way to interpret these common changes is that people work harder at
their appearance before they get married, and slack off a little afterwards. Marriage raises
the marginal costs of exiting the relationship, and newlyweds take advantage of this fact
by investing less than they did earlier.
A common economic example is suppliers building a plant near a large customer,
to economize on shipping costs. A famous example in economic history is Fisher Body,
who were asked to build a plant near General Motors (GM) factories, to build automobile
bodies for their large customer GM that would be cheaper to ship to their factories. As
the story goes, Fisher Body balked at making this investment, which led GM to vertically
integrate by simply buying Fisher Body, to eliminate the post-investment haggling.
Another example comes from journalism. Magazines which are produced every
week or month usually do not own the printing presses that physically produce their
magazine. [check this] . But daily newspapers usually do own their own printing presses.
Why are these ownership structures different?
The difference is probably not just due to the fact that the daily papers use the
printing press more often. Magazine publishers could own their own presses as well, and
simply rent them out when they are not being used to print the magazine.
The holdup problem suggests a possible explanation. Once a daily newspaper’s
content is delivered to the printing press (or “put to bed”), the newspaper is extremely
vulnerable. If the paper’s publication is delayed, or a printing mistake is made, it is a
small disaster since the paper is expected to come out daily, and on time. If a magazine is
published late, or a printing error is made, the damage is lower for a magazine because
the news they report is less timely. So if a newspaper owns the printing press facility,
they can get more control (e.g. higher priority, and the ability to fix a mistake rapidly). A
magazine, in contrast, doesn’t benefit as much from added control over last-minute slips.
Assuming there is some cost to integrating—typically, that the employees are less
motivated because they are paid a fixed wage, rather than having their earnings depend
on market competition— the benefits to a newspaper to owning their presses can be
higher than the costs, but the benefits to a magazine are not worth the added costs.
It is an open question how common holdup is in business relationships, and why it
occurs. There are not many empirical studies. FINISH here.
Other mechanisms for limiting hidden action FINISH
Monitoring (e.g. "spying" on retail employees, LA times article)
Baker et al paper on tracking truckers with GPS). F{TBA}
68
Reputational incentives
Word of mouth
Lawsuits (e.g., class actions)
Escrow
"Bonding" or collateral (e.g., maid services are bonded)
long-lived institutions (e.g., family names, Krazy Karl's Karpet Barn, political
parties)
Stock market value after airplane crashes, product recalls etc.
But…markets may under/overreact (e.g., all Arthur Anderson employees get tainted)
Bad workers can get good recommendations to "outplace" (academic hiring-- phone vs
letter)
"Capital T" endgame/lame duck problem (e.g., President Clinton final-hour pardons)
FINISH
________________________________________________________________________
Monitoring versus Honor37
Versus
Virtually all corporations older than a month or two have some sort of corporate culture.
What about academic institutions? Most do – they range from the emphasis on sports at the
University of North Carolina, to the pledge not to drink, have sex or even dance that students at
Wheaton College in Illinois make, to the cherished liberalism and activism at the University of
California in Berkeley.
37
This material was drafted by Galen Loram.
69
The California Institute of Technology, a small university in Southern California with a
focus on math and science, has at the core of its culture an honor code. This code simply states:
“No member of the Caltech community shall take unfair advantage of any other member of the
Caltech community.”
According to Rutgers University Professor Donald McCabe with the Center for
Academic Integrity, the recognized expert on the topic, 75% of college students admit to cheating
at some point during their college years. But at Caltech students, faculty and staff estimate that
only 2-4% cheat. What is responsible for the 20-fold discrepancy between Caltech and other
universities?
Many schools have some sort of honor code but most are taken lightly. Students often do
not know about the existence of a code, or its details or “teeth” (penalties for violating it). The
Caltech honor code seems to work extraordinarily well. So what differentiates an effective honor
code from one that hardly makes it into the guide for incoming freshman?
One reason is that the honor code is visible in practice at many places, and makes life
easier. Students and professors leave their doors open, and often unlocked. Virtually all exams are
take-home exams (sometimes “closed book”), typically with a maximum time limit that students
can spend taking the exam stated. It is also assumed that people will read, understand and follow
detailed policies on exam rules and homework collaboration– or clarify misunderstandings.
A recent study by the Center for Academic Integrity found that 82% of its members –
colleges nationwide – felt that if a school did not already have an effective honor code, it was
either ‘very difficult’ or ‘not possible’ to create one. This suggests a strong founder effect: when
the university was founded tens or hundreds of years ago, what the founders decided to
implement in the way of an honor code has a marked impact on the community years later.
Cultures do not arise overnight, and one which requires a vast amount of trust on the part of all
concerned and rests on peer enforcement is fragile at best; thus while in the process of attempting
to create such a culture a few ‘bad apples’ could ruin the barrel.
A key feature of the Caltech honor code is peer enforcement in two forms:
First, if a student sees an honor system violation it is his or her responsibility to report it.
Failing to report an honor system violation is, in and of itself, an honor system violation. This
meta-norm works best if there is a culture in which new members of the community are
indoctrinated into the system, under the assumption that the code generally works so that
violations are rare and will be noticed. (If people thought the code was not followed, then
violations would congest the system and violators would not worry as much about being caught
and brought to justice.) Perception is incredibly important: if students, faculty or staff do not
believe that the code is followed – whether or not it actually is - they will not feel obligated to
follow it themselves. Thus a strong sense of community—upholding the code-- is necessary to
maintain the code.
Second, adjudication of honor code violations is conducted entirely by students, with
minimal faculty oversight.This saves faculty time, and also means the students know they are
being judged by peers. For the students who participate in the Board of Control (BoC) process, it
provides a way to contribute to campus life and learn to cope with emotional, nuanced issues of
morality which helps prepare them for tackling these questions later in their scientific, corporate
or public life.
While the benefits of the honor code come in a variety of forms, one of the most
important is reduced monitoring costs. Unproctored exams liberate professors to focus on their
research instead of wasting time futility looking for cheaters. Conversely, students benefit from
the chance to take exams in a situation of their own choosing – blasting Nine Inch Nails on their
stereo, sitting next to a tranquil stream, or sitting a desk in the classroom where they learned the
material. Another benefit comes in the form of gift exchange. Students feel that they are trusted
and thus feel obliged to return the favor in the form of not cheating. Graduate TAs are freed from
deeply probing for similarities, instead only looking at something if it catches their eye.
70
The honor code has another huge advantage at Caltech: Most students at Caltech study
science and engineering, and many will pursue lives working in universities or corporations.
These students will work in environments where they have enormous personal responsibility and
are relatively unsupervised (e.g., managing a large R&D lab). The pressure to cut corners, take
unfair credit, downplay results which discredit a pet theory, or even fabricate data can be severe.
The honor code enables scientists-to-be to “practice” and exercise their ability to do the right
thing in a relatively benign environment, as a warm-up to the outside scientific world where the
moral issues are often subtler and the stakes higher.
Surprisingly, the vast majority of the honor system violations are blatant and occur when
someone is under extreme stress and “cracks” (perhaps a close family member is diagnosed with
terminal cancer or a similarly grave tragedy occurs).
The other main source of violations is ignorance. For example, a student from a foreign
country may not understand the idea of plagiarism and might submit a paper that heavily relied on
a couple of sources and failed to cite them. The increased atmosphere of trust also allows for
other benefits, people can accidentally leave their bag around campus with hundreds of dollars
inside and come back hours later to find it still sitting there unperturbed. The unspoken threat of
social ostracism of a cheater tends to be enough to keep those who would still consider cheating
in line.
There are drawbacks too, of course. Doubtless, without the close scrutiny a higher
percentage of those who do cheat manage to ‘slip through the cracks’ and not get caught. For
those 15-20 cases of suspecting cheating each year that do get reported the chair and secretary
determine whether or not the issue at hand is an honor code issue, and whether enough evidence
is present to warrant the attention of the Board of Control.
Examples of something that would not warrant the attention of the Board would be
someone being rude to someone else or a case of suspected copying when the two exams had
nothing in common. The BoC then views the evidence, interviews any potential defendants and
witnesses and then makes three decisions – whether or not an honor system violation has
occurred, how to nullify any unfair advantage gained, and how to protect the community from
such a thing happening again. These are recommendations to the deans of students, and are
accepted about 95% of the time. Because there are so few cases each case is treated meticulously,
in a time consuming process. An average case of cheating or “overcollaboration” that goes to the
BoC takes about 60 person-hours to resolve. The fact that cases are taken so seriously also
upholds faith in the system—a student who loses a BoC case can at least take consolation in the
fact that the system worked hard at adjudicating the case, even if the outcome is unpleasant for
him or her.
_______________________________________________________________
5. Summary
The central idea in organizational economics is the triangle of (residual) decision
rights, incentives, and evaluation. Incentives refer to the contract and motivations of
workers. Evaluation is a system that values performance to award incentives. Decision
right refer to the latitude workers have to decide what to do when their contract does no
clearly specify what they should do.
A good organization balances incentives, and evaluation, so that workers with
residual decision rights will make the right decisions.
One way to see the importance of balancing these three elements is to study when
organizations or countries melt down and create horrible outcomes. One example is
China’s “Great Leap Forward” from 1957-61. The Chinese government collectivized
71
farms from small units into large ones (1000+ workers) and falsely forecasted a big boost
in productivity. Their optimistic forecast led to a central plan in which farms would
contribute a lot of their grain to industrial workers in cities (“excessive procurement”). In
fact, output gradually fell because the large scale of the communes reduced the workers’
incentives to work hard. Imports rose only slightly. As a result, workers got to keep less
of the grain they produced and consumed fewer calories per worker. Poorly fed workers
produced less, creating a death spiral of output. To keep the system from failing, the
percentage of workers who could voluntarily leave the communes was ratcheted up
sharply. The result was a massive famine, the largest in recorded history. The Great Leap
Backward is a perfect storm of how centralized planning (combined with poor feedback
to central planners, or a reluctance to admit a mistake) can lead to disaster.
Some basic economic concepts are useful in understanding organizations. A
property right is a combination of the right to use, and to sell, a piece of property. (The
property might be a physical good, like a car or machine, or an intangible good like a
slice of time—such as a landing slot at an airport—or a band of radio spectrum.) Many
studies of economic development across countries show that enforcement of basic
property right is highly correlates with economic prosperity and growth. Property rights
are the broader equivalent of “decision rights” in firms.
A “partial equilibrium” analysis in economics holds behavior of some agents
fixed, while changing behavior of other agents is analyzed. When all agents are assumed
to respond to change the analysis is a “general equilibrium” one. Partial equilibrium
reasoning either assumes that the agents whose behavior is fixed are unimportant, or the
economic analyst simply cannot solve mathematically what will happen when all agents
change their behavior. A byproduct of a bad partial equilibrium analysis is “unintended
consequences”. There are many examples of such behavior. ONE HERE
General equilibrium analysis often goes under the useful name of “supply and
demand”. MORE
Models of human nature
Basics of Game theory
Hidden information and hidden action
72
Partial References
Wall Street Journal (1996). Building blocked, A1. December 12 (byline J. Sapsford).
Business Week. (1999). That Russian company you bought? Maybe you didn’t. p. 70,
December 13 (byline M. Coker).
Los Angeles Times (2004). Militiamen ‘reclaim’ oil for Nigerians. July 19, p. C3 (byline
Michael Peel).
Hoff, Karla and Joseph Stiglitz. (2004). After the big bang? Obstacles to the emergence
of the rule of law in post-Communist societies. American Economic Review, June, 94,
753-763.
Staten, Michael E. and John Umbeck. 1982. American Economic Review, 72(5), 10231037.
Dawes, Robyn M. and Richard Thaler. Anomalies: Cooperation. Journal of Economic
Perspectives, Summer 1988, vol 2, 187-197.
Jonathan M. Karpoff (“Public versus Private Initiative in the Arctic Exploration” Journal
of Political Economy, 109(1):38-78, 2001)
From Fisman and Miguel,
73
Joseph Jett and Kidder Peabody
"If I am operating under their system to generate profits, with
their approval," Jett asked, "and they turn around and change
the rules so the reasons you had for doing this no longer exist -and, by the way, we are taking all the money back we paid you - is that fair?"
Lynch, however, argues that exploiting a dysfunctional system
can indeed constitute fraud. "It is like receiving a hundred
paychecks each week when you should get one," he said. "It's
not a defense to say, 'I assumed the company meant to pay me
a hundred times.' "
That defense, however, did sway many of the 500 graduate
business students at New York University who gathered one
evening last fall to hear Jett unabashedly discuss his case. So,
was his story an object lesson to a future generation of
financiers eager to avoid the same fate?
Not really. William Starbuck, the professor who invited Jett, said
"the overwhelming majority" felt that Kidder and GE "had
implicitly defined a game by setting up their accounting system."
In the students' view, Starbuck said, "it is the job of employees
to play the game, not to decide whether they are good games or
bad games."
Afterword
74
New York Times April 6, 1997. “Joseph Jett: Scapegoat or scoundrel?”