If the logic is right, why do these markets still exist?
The last lecture ended with a puzzle. Akerlof's logic says that a used-car market with hidden quality should shrink until only the worst cars are left. The same logic threatens insurance and hiring. Yet in every country people buy used cars, buy insurance and get jobs. So something must be closing the gap.
Go back to Amara and the two cars. The seller of the good car has a problem too. She knows her car is good, but saying so is useless, because the seller of the bad car can say exactly the same words. Words cost nothing, so they prove nothing.
Every solution in this lecture answers one question: how can the hidden thing be made visible, or hiding it be made expensive? The answer depends on who acts. The side that knows can send a signal. The side that does not know can design a test. Or the government can change the rules. For each one we ask how it works and when it fails.
| Who acts | Tool | Examples |
|---|---|---|
| The side that knows | Signalling | warranties, reputation, qualifications |
| The side that does not know | Screening | inspections, medical checks, menus of contracts, trial periods |
| The government | Rules | required disclosure, minimum standards, licensing, a requirement to join, direct provision |
Signals: proof that costs something
A signal is an action by the informed side that shows its hidden quality. The key word is "shows". A claim does not show anything. An action shows something only if it would be too expensive for the other type to copy.
A signal is credible when it is cheap enough for the high-quality type to send, but too expensive for the low-quality type to copy.
An invented example. Two dealers sell cars that look the same. A 12-month warranty means the dealer pays for any repairs in the first year. For a good car, the expected repair bill in that year is $200. For a lemon, it is $1 500. Buyers will pay $1 000 more for a car with a warranty.
The dealer with good cars gains on each car by offering the warranty. So she offers it.
The dealer with lemons would gain $1 000 but pay about $1 500 in repairs, a loss of $500 a car. So he does not offer it.
Now the warranty tells buyers something true. Only good cars carry one, so the extra $1 000 is a fair price for the knowledge. It works because it costs the two dealers different amounts. If a lemon's expected repair bill were only $900, both would offer warranties, and the signal would mean nothing.
A seller writes "Excellent condition, never had a problem" in her advert. Another offers a 12-month warranty. Both are trying to tell the buyer the same thing. Why does only one of them help Amara?
Show the answer
The advert costs nothing to write, whether the car is good or bad. So the seller of a lemon writes it too, and Amara learns nothing. Economists call a free claim like this "cheap talk". The warranty is different. It costs the seller money if the car is bad, so a lemon seller will not offer it. A signal is believable only when sending it would hurt someone who is lying.
Reputation works in the same way, over time. A firm that knowingly sells a bad product risks losing its future customers. That future loss is the cost of cheating. A stranger who will never meet the buyer again bears no such cost, so her promises are worth less.
Education as a signal
In 1973 the American economist Michael Spence published a model of the job market that made this idea famous. An employer cannot see how able an applicant is. Spence asked what would happen even if a degree taught nothing useful at all. His answer was that it could still be valuable, as a signal. Studying costs less able students more time and effort. So a long course of study is cheaper for able people to complete. An employer who sees the qualification can then infer ability.
Explore the logic below, with invented numbers. A good job pays $40 000 more in total. Each year of study costs an able applicant $8 000 in fees, effort and lost earnings, and a less able applicant $16 000. The employer chooses how many years to ask for.
Change the number of years the employer asks for. For each applicant, compare the cost of studying with the $40 000 of extra pay. Find the range where only able applicants study. Then find the cheapest point in that range.
An invented job market. The good job pays $40 000 more. A year of study costs an able applicant $8 000 and a less able applicant $16 000. An applicant studies if the cost is below $40 000. A cost of exactly $40 000 marks the edge of the range.
| Years asked for | Cost: able | Cost: less able | Who studies | Does the signal work? |
|---|---|---|---|---|
| 1 | $8 000 | $16 000 | both | no: everyone qualifies |
| 2.5 | $20 000 | $40 000 | able only | yes, at the lowest cost |
| 3 | $24 000 | $48 000 | able only | yes |
| 5 | $40 000 | $80 000 | able only (at the edge) | yes, at the highest cost |
| 6 | $48 000 | $96 000 | nobody | no: nobody qualifies |
What did you see? At 1 year, both applicants studied, so the qualification said nothing. At 6 years, nobody studied. Between 2.5 and 5 years, only able applicants found study worth it. There the signal worked.
That is it. That is the whole idea of a signal. It works only in a range of costs, and the range exists because the cost is different for the two types. Notice also what the signal costs. Even at the cheapest working point, each able applicant spends $20 000. If the study taught nothing, that money buys only sorting. Nothing new is produced.
Is education a signal or a skill?
Spence's model does not say that education teaches nothing. It says that a qualification would have value even if it taught nothing. Most economists think education does both. The human capital view says that education raises productivity, because students learn skills. The signalling view says that part of its value is that it reveals ability the student already had.
The balance matters for policy. If education is mostly skill, more education makes the economy more productive, and it may have positive externalities. That supports public spending on it. If it is mostly signal, the private return is larger than the social return. A graduate earns more partly by being sorted ahead of someone else, not by producing more.
Suppose degrees were pure signals. A government pays for twice as many people to get degrees. What happens to the total output of the economy, and to the value of a degree?
Show the answer
Output does not rise, because a pure signal teaches no skills. The same people do the same work. But a degree now separates able from less able applicants less well, because many more people hold one. Employers may start to ask for a higher degree instead, and signalling costs everyone more. Real degrees also teach skills, so the effect is less extreme. But this is why the two views lead to different policies.
Screening: the side that does not know designs a test
Screening means the uninformed side designs a check or a choice that makes the informed side reveal what it knows.
The simplest screens are checks. Amara can pay a mechanic to inspect the car. An insurer can ask for a medical examination. An employer can offer a trial period. Each one buys information directly.
A cleverer screen is a menu. The insurer offers two policies, and the customer's choice reveals the risk. Predict which customer chooses which.
An invented example. Car insurance is required by law, so each driver must choose one policy. Policy A costs $850 and pays for every repair. Policy B costs $300, but the driver pays the first $1 500 of any repair (the excess). An accident costs $2 000 to repair. Mei has a 10% chance of an accident this year. Tomás has a 40% chance. Who chooses Policy B?
Each driver compares the price plus the excess they expect to pay.
| Driver | Policy A | Policy B: $300 + chance × $1 500 | Chooses |
|---|---|---|---|
| Mei (10%) | $850 | $300 + $150 = $450 | B |
| Tomás (40%) | $850 | $300 + $600 = $900 | A |
The low-risk driver chooses the high excess. The insurer learns each driver's risk from the choice.
The menu sorted the drivers without a single question about their driving. Look at what it cost, though. Mei now carries most of the risk of her own accidents. She cannot buy full cover at a fair price, because full cover is priced for Tomás. Screening reveals the hidden type, but the low-risk side pays for the sorting by accepting less cover.
An excess also helps with moral hazard, the hidden-action problem. A driver who pays the first $1 500 of every repair still has a reason to drive carefully. In the RAND Health Insurance Experiment from the last lecture, people who paid 95% of their bills spent about 30% less on care than people whose care was free. A bonus for staying claim-free and pay linked to results work in the same way.
Rules: when the government closes the gap
Signals and screens are private solutions, and some gaps are too wide for them. Then the government can change the rules.
Required disclosure forces the informed side to reveal information: allergens on food, the fees on a loan, the energy rating of a fridge. Minimum standards make some information unnecessary, because a car that fails its safety test cannot be sold. Licensing allows only people who have passed a test to work as doctors or electricians. A patient who cannot judge medical advice then knows that a minimum level has been met.
The policy. Diners cannot see a restaurant's kitchen. In November 1997, a local television news series in Los Angeles showed unclean restaurant kitchens. In December 1997, Los Angeles County passed a rule that every restaurant must show its hygiene inspection result in its window, as a grade card: A, B or C. An A needed at least 90 points out of 100. The cards appeared from January 1998.
The prediction. The card puts the hidden information at the door, where the diner decides. Demand should move towards clean restaurants, so restaurants have a reason to clean up.
What happened. The economists Ginger Jin and Phillip Leslie studied the results. Inspection scores rose by about 5% on average. The share of restaurants scoring an A rose from about a quarter to more than half. Revenue at A restaurants rose by about 5.7%, while revenue at C restaurants fell by about 1%. Most important, hospital admissions for food-related illness in Los Angeles County fell by about 13% in 1998, while they rose by about 3% in the rest of California. The fall lasted through 1999 and 2000. (Jin and Leslie, Quarterly Journal of Economics, 2003; Simon and colleagues, Journal of Environmental Health, 2005; Stanford Graduate School of Business summary, 2005.)
The verdict. The disclosure worked. Jin and Leslie found that the fall in illness was not fully explained by diners switching restaurants, so the restaurants themselves improved. The information was simple, a single letter, and it appeared at the moment of choice. One warning: scores bunched just above 90, which suggests some inspectors were generous. But gains also appeared in parts of the inspection that are hard to fake, such as ventilation and toilets.
Disclosure does not always work this well. In 2008 New York City required chain restaurants to show the calories of every item. A study of Starbucks coffee shops by Bryan Bollinger, Phillip Leslie and Alan Sorensen found that calories per purchase fell from 247 to 232, about 6%. Calories from food fell by 14%, but drink choices hardly changed. (National Bureau of Economic Research, 2010.) So the information helped, but modestly. A number is harder to use than a letter. Disclosure works best when the information is simple and comes at the moment of choice.
Requiring everyone to join
The last lecture's Harvard case was a death spiral: the healthiest members left, and the plan collapsed. Disclosure cannot fix that, because the problem is that low-risk people choose to leave. Two rules attack it directly.
Use the invented insurance pool from the last lecture. Fifty low-risk people each cost $200 a year to cover and would pay up to $300. Fifty high-risk people each cost $1 000 and would pay up to $1 200.
With no rule, the pool unravels until only high-risk people buy, at $1 000. Each gains , so the total gain is . The low-risk people are uninsured.
With a requirement to join, everyone is in the pool, and the insurer can charge everyone the average cost, $600. High-risk people gain each, a total of $30 000. Low-risk people pay $600 for cover worth $300 to them, so each loses $300, a total of −$15 000. The total gain is .
So the requirement adds $5 000 of gains: exactly the $5 000 that adverse selection destroyed in the last lecture. But low-risk people now pay more than their cover is worth to them. Each pays $400 more than they cost, and that money pays for high-risk cover.
The other rule is risk-based pricing: each person pays their own expected cost, $200 or $1 000, and both groups buy. The total gain is again . But the insurer must screen to learn each risk, and high-risk people, often people who are already ill, face high prices. A requirement to join with one price shares the cost instead, by asking the healthy to pay for the sick.
Some governments provide the service directly. A patient often cannot judge whether a treatment is needed, and a doctor paid per treatment may recommend more. A tax-funded health service with salaried hospital doctors removes that reason to over-treat. But salaried staff have weaker reasons to work quickly, and waiting lists can grow. Every solution moves the problem somewhere. None removes it completely.
Three classic mistakes
It seems natural that safe customers want the "safer" policy with a low excess. It is the other way round. A high excess costs most to people who claim often. So low-risk customers choose the high excess and the low premium, and high-risk customers pay more for a low excess. That is why the menu works as a screen.
A promise, an advert or a confident manner can be copied by the low-quality type at no cost, so it reveals nothing. A signal is credible only if it costs the low-quality type more than it is worth to them. Always ask: who would find this too expensive to copy?
Information helps only if buyers understand it and use it. Small print is disclosure that nobody reads, and calorie labels changed choices only a little. Licensing guarantees only a minimum level. Always say how well a solution works, and what it costs.
Work each one on paper first. Open a hint only when you are stuck.
-
Is each of these a signal, a screen, or a government rule? Give the reason. (a) A phone maker offers a three-year guarantee. (b) A bank asks loan applicants for six months of payslips. (c) The law requires an energy rating on every new fridge. (d) Only licensed electricians may wire a new house.
Hint 1
Ask who acts: the side that knows, the side that does not know, or the government.
Hint 2
The phone maker knows its quality. The bank does not know the applicant's income.
Full answer
(a) A signal: the informed side, the maker, acts, and a long guarantee would be costly for a maker of unreliable phones. (b) A screen: the uninformed side, the bank, designs a check that reveals the hidden fact. (c) A government rule, required disclosure: the law forces the seller to reveal efficiency. (d) A government rule, licensing: it guarantees a minimum level of skill that a homeowner cannot judge.
-
An invented market. Under a two-year warranty, the expected repair cost of a good laptop is $40, and of a poor laptop $260. Buyers pay $150 more for a laptop with a warranty. (a) Which sellers offer the warranty? (b) Does the warranty work as a signal? (c) What changes if buyers pay $300 more?
Hint 1
For each seller, compare the extra price with the expected repair cost.
Hint 2
Good seller: $150 − $40. Poor seller: $150 − $260.
Full answer
(a) A good seller gains , so offers it. A poor seller would lose , so does not. (b) Yes. Only good laptops carry the warranty, so buyers can trust it. (c) At $300, a poor seller gains , so poor sellers offer it too. Now every laptop has a warranty, and it no longer tells buyers which are good. The extra price must lie between $40 and $260 for the signal to work.
-
An invented job market. A good job pays $30 000 more in total. A year of study costs an able applicant $6 000 and a less able applicant $10 000. (a) For which numbers of years does a qualification separate the two types? (b) What is the cheapest requirement that works, and what does it cost each able applicant?
Hint 1
Able applicants study if . Less able applicants study if .
Hint 2
The signal works when the first is true and the second is false.
Full answer
(a) Able applicants study when . Less able applicants study when . So the signal separates them when . At exactly 3 years, a less able applicant would gain nothing from studying, so does not bother. (b) The cheapest working requirement is 3 years. It costs each able applicant . If the study taught nothing, this is the price of sorting. Check at 3 years: able $18 000 < $30 000 ✓; less able $30 000, no gain ✓.
-
Use the insurance pool from Worked Example 3. A government requires everyone to join, at one price of $600. A critic says: "This rule makes society better off." A second critic says: "This rule is unfair to healthy people." Using numbers, explain how both can be right. Then suggest one change that would reduce the unfairness, and say what it costs.
Hint 1
Find the total gain with and without the rule, then the gain or loss for each group separately.
Hint 2
Without the rule: $10 000, all to high-risk people. With it: $15 000 in total, but low-risk people lose $300 each.
Full answer
The first critic is right about the total. The rule raises total gains from $10 000 to $15 000, because the 50 low-risk people are now insured, and each costs $200 to cover but values cover at $300. The second critic is right about the shares. Each low-risk person pays $600 for cover worth $300 to them, so each is $300 worse off than with no insurance. Each high-risk person pays $600 instead of $1 000, so each is $400 better off than before. The healthy are paying for the sick. One change: let the price differ by risk, or give low-risk people a partial rebate paid for from taxes. Risk-based prices remove the transfer, but high-risk people then pay $1 000, and the insurer must screen to find out who is who. A tax-funded rebate keeps one price but moves the cost onto taxpayers. Efficiency and fairness pull in different directions here, and choosing between them is a judgement, not a calculation.
Look around you with this lecture in mind. Warranties, brand names, exam certificates, medical licences, food labels, compulsory car insurance, grades in restaurant windows. They look like a random collection. They are not. They exist to solve one problem: one side knows something the other cannot see.
In 2001 Michael Spence shared the Nobel Memorial Prize in Economic Sciences with George Akerlof and Joseph Stiglitz for their work on markets with asymmetric information. Akerlof showed how hidden information breaks markets, Spence how the informed side can signal, and Stiglitz how the uninformed side can screen.
Several solutions here are government rules, and a rule can fail too. A licence can be used by a profession to keep competitors out. When an intervention makes the outcome worse, economists call it government failure. That is the subject of the next lectures.
- Solutions to information failure either make the hidden thing visible or make hiding it expensive, and they come from the informed side, the uninformed side, or the government.
- A signal is credible only if the low-quality type would find it too expensive to copy, which is why a warranty works and a promise does not.
- Screening lets the uninformed side design a choice that reveals the hidden type, as when low-risk drivers choose a high excess and a low premium.
- Required disclosure works best when the information is simple and appears at the moment of choice, as Los Angeles' restaurant grade cards showed.
- A requirement to join an insurance pool stops a death spiral and raises total gains, but it makes low-risk people pay for high-risk people.
- Every solution has a cost, such as wasted signalling, less cover for the low-risk, or barriers to entry, so each must be judged against what it fixes.