Credence Goods Marco A. Schwarz 1 1 University of Innsbruck June 2, - - PowerPoint PPT Presentation

1/ 88

Credence Goods

Marco A. Schwarz1

1University of Innsbruck

June 2, 4, and 5, 2020

These slides are partially based on slides by Rudi Kerschbamer. Thanks!

2/ 88

Outline

1

Credence Goods: Introduction and Theory Characteristics of Credence Goods Basic Model Relaxing Assumptions

2

Credence Goods: Experiments

3

Commissions and Kickbacks

4

Literature

3/ 88

What are Credence Goods?

Seminal paper (Darby and Karni, 1973):

“an individual renting specialized knowledge can evaluate only the results and not the procedure”

“Goods and services where an expert knows more about the quality a consumer needs than the consumer himself are called credence goods.” (Dulleck and Kerschbamer, 2006) Other definition (not here): The the attributes of the good remain unobservable even after consumption. Examples of credence goods: health sector, repairs, financial advice, cab rides in an unknown city, construction, ...

4/ 88

Introduction: Motivation

Problems (in a vertically differentiated market):

Overtreatment: Expert provides more expensive treatment than necessary. Undertreatment: Expert provides insufficient treatment. Overcharging: Expert charges more expensive treatment than provided.

Inefficiencies in real-life markets:

Healthcare:

US: Up to 10% of the 3.3 trillion US$ of yearly health expenditures are estimated to be due to fraud (FBI, 2011). 28% of dentists’ recommendations involve overtreatment (Gottschalk et al., 2020).

Fraud in repair services:

Cars (Taylor, 1995; Schneider, 2012; Rasch and Waibel, 2018) Computers (Kerschbamer et al., 2016, 2019; Bindra et al., 2020) Phones (Hall et al., 2019)

Taxi rides (Balafoutas et al., 2013, 2017)

5/ 88

Basic Model: Players and Actions

(based on Dulleck and Kerschbamer, 2006) Customer(s) (“he”)

knows he has problem and knows he has

a major problem with probability h or a minor problem with probability 1 − h

a minor problem can be successfully treated with a minor or a major treatment; a major problem requires a major treatment decides whether to commit to an expert recommendation (visit) and pay price charged (in)sufficient treatment gives net payoff v − p (−p)

Expert(s) (“she”)

sets prices ¯ p and ¯ p for major (¯ t) and minor treatment (¯ t) has costs of ¯ c and ¯ c < ¯ c for providing major and minor treatment performs (costless) diagnosis and charges for recommended treatment → profit = price-cost margin (no visit: zero profit)

6/ 88

Basic Model: Assumptions

Customers and experts are homogeneous. Once the customer decided to visit an expert, he commits to be treated (and to pay) that expert and the expert commits to treat the customer and charge one of the posted prices. The diagnosis is costless (or at least observable and verifiable) and perfect. The customer does not know which treatment he received. The expert is not liable for providing an insufficient treatment. It is a one-time interaction. We will relax some of these assumptions later.

7/ 88

Basic Model: Two Versions of Payoffs

Version 1 (customer gets v if the problem is successfully treated): customer gets ¯ t ¯ t no visit customer ¯ t v − ¯ c v − ¯ c needs ¯ t −¯ c v − ¯ c Version 2 (customer suffers if the problem is not successfully treated): customer gets ¯ t ¯ t no visit customer ¯ t −¯ c −¯ c −¯ l needs ¯ t −¯ l − ¯ c −¯ c −¯ l

8/ 88

Basic Model: Game Tree

No visit ¯ p ¯ p ¯ t ¯ p ¯ p t Visit P Major problem (h) ¯ p ¯ p ¯ t ¯ p ¯ p t Visit No visit P Minor problem (1 − h) Nature Expert sets prices Customer h 1 − h h 1 − h Expert provides Expert provides Expert Expert Expert Expert

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• − ¯

p ¯ p−¯ c

• −

¯ p ¯ p−¯ c

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• v− ¯

p ¯ p−¯ c

• v−

¯ p ¯ p−¯ c

9/ 88

Basic Model: Game Tree

Efficient treatment and adequate charge

No visit ¯ p ¯ p ¯ t ¯ p ¯ p t Visit P Major problem (h) ¯ p ¯ p ¯ t ¯ p ¯ p t Visit No visit P Minor problem (1 − h) Nature Expert sets prices Customer h 1 − h h 1 − h Expert provides Expert provides Expert Expert Expert Expert

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• − ¯

p ¯ p−¯ c

• −

¯ p ¯ p−¯ c

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• v− ¯

p ¯ p−¯ c

• v−

¯ p ¯ p−¯ c

10/ 88

Basic Model: Game Tree

Undertreatment

No visit ¯ p ¯ p ¯ t ¯ p ¯ p t Visit P Major problem (h) ¯ p ¯ p ¯ t ¯ p ¯ p t Visit No visit P Minor problem (1 − h) Nature Expert sets prices Customer h 1 − h h 1 − h Expert provides Expert provides Expert Expert Expert Expert

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• − ¯

p ¯ p−¯ c

• −

¯ p ¯ p−¯ c

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• v− ¯

p ¯ p−¯ c

• v−

¯ p ¯ p−¯ c

11/ 88

Basic Model: Game Tree

Overtreatment

No visit ¯ p ¯ p ¯ t ¯ p ¯ p t Visit P Major problem (h) ¯ p ¯ p ¯ t ¯ p ¯ p t Visit No visit P Minor problem (1 − h) Nature Expert sets prices Customer h 1 − h h 1 − h Expert provides Expert provides Expert Expert Expert Expert

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• − ¯

p ¯ p−¯ c

• −

¯ p ¯ p−¯ c

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• v− ¯

p ¯ p−¯ c

• v−

¯ p ¯ p−¯ c

12/ 88

Basic Model: Game Tree

Overcharging

No visit ¯ p ¯ p ¯ t ¯ p ¯ p t Visit P Major problem (h) ¯ p ¯ p ¯ t ¯ p ¯ p t Visit No visit P Minor problem (1 − h) Nature Expert sets prices Customer h 1 − h h 1 − h Expert provides Expert provides Expert Expert Expert Expert

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• − ¯

p ¯ p−¯ c

• −

¯ p ¯ p−¯ c

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• v− ¯

p ¯ p−¯ c

• v−

¯ p ¯ p−¯ c

13/ 88

Basic Model: Game Tree

Equilibrium (depending on ¯ c)

No visit ¯ p ¯ p ¯ t ¯ p ¯ p t Visit P Major problem (h) ¯ p ¯ p ¯ t ¯ p ¯ p t Visit No visit P Minor problem (1 − h) Nature Expert sets prices Customer h 1 − h h 1 − h Expert provides Expert provides Expert Expert Expert Expert

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• − ¯

p ¯ p−¯ c

• −

¯ p ¯ p−¯ c

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• v− ¯

p ¯ p−¯ c

• v−

¯ p ¯ p−¯ c

14/ 88

Basic Model: Liability and Verifiablility

Liability

The customer can observe and verify the outcome ex post (but cannot always tell which treatment was provided) and hold the expert accountable. Rules out undertreatment. If the treatment is not verifiable, overcharging dominates

• vertreatment (for the expert).

Verifiability

The customer can observe and verify the treatment ex post (but cannot always tell which treatment was necessary). Rules out overcharging.

15/ 88

Basic Model: Liability

No visit ¯ p ¯ p ¯ t Visit P Major problem (h) ¯ p ¯ p ¯ t ¯ p ¯ p t Visit No visit P Minor problem (1 − h) Nature Expert sets prices Customer h 1 − h h 1 − h Expert provides Expert provides Expert Expert Expert

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p− ¯ c

• v− ¯

p ¯ p−¯ c

• v−

¯ p ¯ p−¯ c

16/ 88

Basic Model: Verifiability

No visit ¯ p ¯ t ¯ p t Visit P Major problem (h) ¯ p ¯ t ¯ p t Visit No visit P Minor problem (1 − h) Nature Expert sets prices Customer h 1 − h h 1 − h Expert provides Expert provides Expert Expert Expert Expert

• v− ¯

p ¯ p− ¯ c

• −

¯ p ¯ p−¯ c

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p−¯ c

17/ 88

Basic Model: Liability and Verifiability

No visit ¯ p ¯ t Visit P Major problem (h) ¯ p ¯ t ¯ p t Visit No visit P Minor problem (1 − h) Nature Expert sets prices Customer h 1 − h h 1 − h Expert provides Expert provides Expert Expert Expert Expert

• v− ¯

p ¯ p− ¯ c

• v− ¯

p ¯ p− ¯ c

• v−

¯ p ¯ p−¯ c

18/ 88

Basic Model: Liability and Verifiability I

Proposition

Under either liability or verifiability or both, market institutions solve the fraudulent expert problem at no cost (Dulleck and Kerschbamer, 2006). Under liability, charging a uniform price solves the problem.

Undertreatment ruled out by assumption, overtreatment and

• vercharging by price structure.

Monopolistic expert charges ¯ p = ¯ p = v. The competitive prices are ¯ p = ¯ p = h¯ c + (1 − h)¯ c.

Under verifiability, equal-markup prices solve the problem.

Overcharging ruled out by assumption, over- and undertreatment no profitable deviation due to price structure. Monopolistic expert charges ¯ p = ¯ p − (¯ c − ¯ c) = v − h(¯ c − ¯ c). The competitive prices are ¯ p = ¯ p − (¯ c − ¯ c) = ¯ c.

19/ 88

Basic Model: Liability and Verifiability II

Under both liability and verifiability, a (weakly) smaller profit margin for the major treatment solves the problem.

Undertreatment and overcharging ruled out by assumption, so the price structure just needs to insure that overtreatment is not a profitable deviation. In a monopolistic market, prices satisfy ¯ p − ¯ c ≤ ¯ p − ¯ c and h ¯ p + (1 − h) ¯ p = v. In a competitive market, prices satisfy ¯ p − ¯ c ≤ ¯ p − ¯ c and h ¯ p + (1 − h) ¯ p = h¯ c + (1 − h)¯ c.

Note that a monopolistic expert benefits from more efficient treatment because she can extract all of the surplus through higher prices; in a competitive market, inefficient expert would leave the market (because they would make negative profits or customers would avoid them).

20/ 88

Basic Model: Imperfect Diagnostic Ability

(Liu et al., 2019)

0.5 1 1

∆c v

q h Po Pe

L

Pu

Figure: Pricing decision.

Expert receives a signal about the problem that is correct with probability q. Treatment is verifiable. Over-/undertreatment can be

• ptimal (high/low h).

Pe

L: Increase in q leads to higher

surplus.

21/ 88

Basic Model: Expert Heterogeneity in (Unobservable) Diagnostic Ability

(Liu et al., 2019)

0.5 1 1

(1−x)∆c (1+x)v−2x∆c ∆c v (1+x)∆c (1−x)v+2x∆c

q h Po, Pe

do

Pe

dd

Pu, Pe

du

Pu, Pe

dd, Pe du

Po, Pe

dd, Pe do

Figure: Pricing decision.

Two types of experts: high (probability x) and low diagnostic ability. Identical profits for both types. Unobservable high-ability type makes lower profit (than when

• bservable).

Inefficiencies possible. Usually, the policy maker would want to make types observable. Pe

do/Pe du: efficient if existent.

Requiring experts to follow their diagnosis might be inefficient!

22/ 88

Basic Model: Heterogeneous Treatment Costs

(Hilger, 2016) Suppose there is a second type of expert with ¯ c2 < ¯ c < ¯ c < ¯ c2. Customers cannot observe the cost function and only know the probability of either expert type. Treatments are verifiable. There is no PBE in which both expert types treat customers efficiently.

The cost differentials are different, hence such an equilibrium must be separating. In such an equilibrium candidate, one expert type would have an incentive to imitate the other type.

Depending on parameters, there exist pooling equilibria in which

• ne or both types over- or undertreat or separating equilibria in

which the second type undertreats and the first type overtreats. Also see Heinzel (2019) for a setting with regulated prices.

23/ 88

Basic Model: Heterogeneous Experts and Reputation I

(Ely and V¨ alim¨ aki, 2003; Ely et al., 2008) Suppose there is a second expert type that always chooses the minor treatment (or always chooses the major treatment). Again, customers cannot observe the expert’s type but only know the probability of either type. Further suppose that customers would not want to be treated by an expert that provides a treatment independent of their problem and that treatments are verifiable. If the probability of the second type is low enough, customers visit the expert.

24/ 88

Basic Model: Heterogeneous Experts and Reputation II

Now assume that the game is infinitely repeated with new customers. Further assume future customers can observe the past treatments of the expert. Then, the first expert type has an incentive to separate herself from the second type by providing the treatment that type would not provide. Because this signalling incentive is independent of the customer’s problem, the customer would not visit the expert in the first place. Thus, reputational concerns can lead to a market breakdown.

25/ 88

Basic Model: Heterogeneous Experts and Reputation III

Balafoutas et al. (2020) show (in a setting with just strategic types) that competition can increase signalling incentives, and thus, inefficiencies.

If experts can choose the price level, but customers cannot choose which expert to visit, experts can charge higher prices if they have a better reputation. If customers can choose which expert to visit, but prices are fixed, experts can attract more customers if they have the best reputation. If customers can choose which expert to visit and experts can choose the price level, future profits are low. Thus, misaligned expert types exploit their customers early and then drop out of the market.

26/ 88

Basic Model: Heterogeneous Customers I

If verifiability holds (but not liability) and customers are heterogeneous with respect to h or v and experts cannot or may not condition their prices on that information (e.g., because that information is private to customers), there may be rationing or over- or undertreatment.

If a monopolistic expert cannot price-discriminate, she might not want to serve customers with a high h or a low v to extract higher profits from the other customers (Dulleck and Kerschbamer, 2006). If a monopolistic expert can price-discriminate, she will rather

• ffer a second tariff to customers with a high h in which she

always provides the major treatment. Customers with a low h do not want to choose this contract because it is too expensive for them. If customers differ with respect to v, the customers with a low v receive a tariff in which they are undertreated. Customers with a high v do not choose this tariff because they have a higher cost of not always receiving v (Dulleck and Kerschbamer, 2006, 2007).

27/ 88

Basic Model: Heterogeneous Customers II

(Jost et al., 2019) Let a share k customers be informed about their problem. They cannot treat themselves, but they also cannot be defrauded. Experts know when they face an informed customer. Under verifiability, a small k and a large k may cause inefficiencies.

If v < ¯ p, an informed customer with a major problem does not want to visit an expert. If k is small, the expert might ignore that. If k is large, the expert might not post equal-markup prices. Thus, uninformed customers would not want to visit. For an intermediate k, the expert caters to both types by posting moderate equal-markup prices. Educating some customers may decrease efficiency.

Under liability, k does not change anything.

28/ 88

Basic Model: Lack of Expert’s Commitment

If the expert cannot commit to ¯ p, she always has an incentive to offer the major treatment to the customer if the expected benefit is larger than the cost difference and the treatment is verifiable (Richardson, 1999). If the expert cannot commit to treating a customer, treatments that do not cover the cost are not feasible. The uniform price solution in the case with liability is not feasible if there is perfect competition or if v is rather low. Fong (2005) analyzes such a situation for the case in which unsolved problems do not provide the same disutility (and a lack of customer’s commitment). The monopolistic expert charges ¯ p = ¯ l and ¯ p = ¯ l and recommends the appropriate treatment. Customers reject the major treatment with a positive probability.

29/ 88

Basic Model: Diagnosis Cost and Lack of Customer’s Commitment I

(Dulleck and Kerschbamer, 2006) The results so far do not depend on a costless diagnosis. If the diagnosis costs d > 0 and is observable and verifiable, the customer would pay this cost. (Monopolistic experts would have to reduce their treatment prices by d.) Let’s now assume customers cannot commit to be treated by an expert:

If d is large enough, the results remain unchanged because it is too expensive to visit a second expert Otherwise, there may be more (and inefficient) equilibria.

30/ 88

Basic Model: Diagnosis Cost and Lack of Customer’s Commitment II

Assume d is small, liability holds, and there are at least four

• experts. Then there is an equilibrium with expert

specialization: At least two experts (the “cheap experts”) post prices ¯ p = ¯ c and ¯ p > ¯ c + d and always recommend the appropriate

• treatment. At least two experts (the “expensive experts”) post

prices ¯ p ≤ ¯ p > ¯ c and always recommend the expensive

• treatment. Customers first visit a cheap expert. Only if that

expert recommends the expensive treatment, they visit an expensive expert.

31/ 88

Basic Model: Diagnosis Cost and Lack of Customer’s Commitment III

Assume d is small, liability holds, ¯ p = ¯ c + ∆ is fixed (with ∆ < ¯ c − ¯ c), and there are at least two experts. Then, there are usually two symmetric overcharging equilibria: If the customer has the major problem, experts always suggest the major treatment. If he hast the minor problem, they suggest the major treatment with positive probability. The customer always accepts the minor treatment, but rejects the major treatment with a positive probability at his first visit. There is one equilibrium in which experts rarely cheat (then customers rarely reject because the recommendation is likely correct) and one in which they often cheat (then customers rarely reject because the other expert would likely cheat as well).

32/ 88

Basic Model: Diagnosis Cost and Lack of Customer’s Commitment IV

S¨ ulzle and Wambach (2005) look at similar conditions and a customer that is partially insured and only bears a fraction λ of the treatment cost.

These two equilibria in mixed strategies exist (in addition to

• ne in pure strategies).

An increase in the coinsurance rate does not necessarily increase fraud. In the equilibrium in which experts rarely cheat, they have to cheat even less to keep the customer indifferent. In the equilibrium in which experts frequently cheat, however, they have to cheat more to keep the customer indifferent.

Mimra et al. (2016b) consider similar conditions and verifiability (but no insurance). They fix prices such that the major treatment yields a large profit margin.

A lower diagnosis cost does not necessarily decrease fraud.

33/ 88

Unobservable Diagnosis and Discounters I

(Dulleck and Kerschbamer, 2009) Additional assumptions (other than the ones in the basic model):

Treatment is verifiable. Customers live two periods and receive 2v if their problem is solved in the first period, otherwise v. Customers cannot commit and have to pay s to visit another seller. Experts and discounters (discounters cannot diagnose). Experts need to invest d > 0 for the correct diagnosis (the investment is unobservable). Experts can offer a warranty payment, if their treatment fails. Here, we focus on parameters such that investing d is efficient (rather than always providing one or the other treatment).

34/ 88

Unobservable Diagnosis and Discounters II

If s is large, experts offer a sufficiently large warranty (such that they have an incentive to invest), invest, and charge a markup on the minor treatment. If s is small, customers would go to a discounter after the diagnosis, so experts keep customers partially uninformed by mixing between investing (and recommending the appropriate treatment) and blindly recommending the minor treatment.

Also see Bester and Dahm (2017).

35/ 88

Outline

1

Credence Goods: Introduction and Theory

2

Credence Goods: Experiments Lab Experiments Field Experiments

3

Commissions and Kickbacks

4

Literature

36/ 88

The economics of credence goods: An experiment

• n the role of liability, verifiability, reputation, and

competition

(Dulleck et al., 2011) 936 subjects, between-subject design 2x2x2x2 design

Liability Verifiability Reputation: Customers can identify experts they interacted with, and they know the prices and their payoff from that transaction. Competition: Four customers can choose between four experts

v = 10, h = 0.5, c = 2, ¯ c = 6, ¯ p ≤ ¯ p ∈ {1, ..., 11}, outside option = 1.6, 16 periods

37/ 88

The economics of credence goods: An experiment

• n the role of liability, verifiability, reputation, and

competition

38/ 88

The economics of credence goods: An experiment

• n the role of liability, verifiability, reputation, and

competition

39/ 88

The economics of credence goods: An experiment

• n the role of liability, verifiability, reputation, and

competition

Too much interaction without liability and verifiability (prediction: 0).

Reason: Customers seem to care mostly about a sufficient treatment: interaction frequency in the next round 57% vs. 32%, slow learning

Verifiability has no effect. Liability improves efficiency (as predicted by theory). Reputation should have no effect, but increases interactions without liability and verifiability. Competition lowers prices (as predicted) and increases interactions (as predicted for neither liability nor verifiability). Adding reputation to competition has no significant effect (a slight increase without liability and verifiability goes in the right direction).

40/ 88

Second opinions in markets for expert services: Experimental evidence

(Mimra et al., 2016b) 420 subjects, 8 periods, between-subject design Customers can visit a second expert at cost k ∈ {7, 13, ∞}. Experts do not know whether it is the customer’s first or second visit. Verifiability and liability Fixed prices such that ¯ p − ¯ c > ¯ p − ¯ c How does a change in k affect outcomes? x overtreatment rate, s search frequency Control treatments: 16 periods, k = 14.5 (too expensive to search), experts can observe whether first or second visit

41/ 88

Second opinions in markets for expert services: Experimental evidence

Figure: Source: Mimra et al. (2016b)

42/ 88

Second opinions in markets for expert services: Experimental evidence

Search reduces the overtreatment rate. Actual reduction in overtreatment is even higher than elicited by strategy method because customers only search if they were recommended the major treatment. A lower search cost induces more search but does not affect the

• vertreatment rate.

k = 14.5 treatment achieves the same reduction. Absolute efficiency is higher with low search cost compared to no search, otherwise no significant differences. Experts reduce overtreatment in the following period, if a customer searched for another expert. Experts overtreat customers more on their second than on their first visit if they can observe this (but less than predicted by theory). No changes over time.

43/ 88

Price competition and reputation in credence goods markets: Experimental evidence

(Mimra et al., 2016a) 320 subjects, 16 periods Neither liability nor verifiability, costs (2, 6) Markets with 4 experts and 4 customers, customers can choose their expert. Customers observe past prices posted and charged as well as their payoff for each of their interactions with every expert. Customers may or may not observe the history other customers had with an expert, depending on experimental condition. Experts may or may not choose prices, depending on experimental condition. If prices are fixed, they are (4, 8) (or (3, 7) as robustness check) for the first 9 periods and 0, 3 thereafter.

44/ 88

Price competition and reputation in credence goods markets: Experimental evidence

Predictions:

With competitive prices, undertreatment, overcharging, and interaction is an equilibrium. With fixed prices, undertreatment, overcharging, and no interaction is an equilibrium. Appropriate treatment for the first 9 periods, then undertreatment, and either always overcharging or undercharging in the beginning, then overcharging, and interaction are equilibria (Reputation equilibria). (Not supported by theory:) Reputation equilibria more likely with public history and fixed prices.

45/ 88

Price competition and reputation in credence goods markets: Experimental evidence

Figure: Source: Mimra et al. (2016a)

46/ 88

Price competition and reputation in credence goods markets: Experimental evidence

Figure: Source: Mimra et al. (2016a)

47/ 88

Price competition and reputation in credence goods markets: Experimental evidence

With competitive prices, prices decline over time. Experts who undertreated try to compensate by lower prices in the next period. Customers usually interact and choose their expert based on prices in the condition with competitive prices and based on successful treatments in the condition with fixed prices. With fixed prices, experts seem to build up a reputation, with competitive prices not so much. Public vs. private history does not affect the treatments, but public history makes overcharging less likely under fixed prices.

48/ 88

Is Reputation Good or Bad? An Experiment

(Grosskopf and Sarin, 2010) 192 subjects?, between-subjects design Horizontally differentiated credence good (= major treatment does not solve minor problem). Expert meets a sequence of six customers. Expert can be good (prefers appropriate treatment) or bad (always prefers minor treatment), customers only know the probability. Customers can observe the past treatments of the expert or not. Good or bad reputation framework

Good: Customer always has a major problem. Good expert computerized. Bad: h = 0.5. Bad expert computerized.

49/ 88

Is Reputation Good or Bad? An Experiment

(Grosskopf and Sarin, 2010) Predictions:

Bad + No reputation: Customers indifferent, good expert should choose the appropriate treatment. Bad + Reputation: Good experts should choose major treatment to signal, hence customers should not visit them. Good + No reputation: Bad expert should choose minor treatment, hence customers should not visit them. Good + Reputation: Bad expert should choose major treatment in the beginning, then mix (and choose minor treatment in the last period); customers should visit in the beginning, then mix.

50/ 88

Is Reputation Good or Bad? An Experiment

(Grosskopf and Sarin, 2010) When theory predicts a market breakdown, there is none. Bad experts often choose the appropriate treatment, even if this cannot be observed, and customers anticipate that and visit them. Their social preferences and the anticipation of those brings them higher profits! Customers visit good experts, even when they should not (in terms of monetary payoffs). Good experts often choose the appropriate treatment then (despite having a signaling incentive).

51/ 88

Agency problems and reputation in expert services: Evidence from auto repair

(Schneider, 2012) 91 undercover garage visits of independent shops in North America (40 by the author himself) Loose battery as reason for garage visit, potentially four additional defects. Garage asked to check regarding additional issues. Low-reputation treatment: “I will move to Chicago in two weeks.” (supported by moving boxes in the car) OR claiming to be on vacation and providing an out-of-province address High-reputation treatment: “I am moving nearby.” Car to be checked for Montreal trip in two weeks. OR providing a nearby address.

52/ 88

Agency problems and reputation in expert services: Evidence from auto repair

In 21 of 40 visits, two or fewer (out of five) defects were detected. Even the loose battery cable was only corrected in 78% of cases. Severe undertreatment. Many unnecessary fixes were suggested. Canadian data: battery cable fixed in 78%, overcharging in 6%,

• vertreatment in 27% of cases.

Inspection cost roughly a third cheaper with high reputation, driven by visits with no repairs. No evidence for fraud depending on reputation.

53/ 88

What drives taxi drivers? A field experiment on fraud in a market for credence goods

(Balafoutas et al., 2013) Three undercover passengers (endowed with GPS satellite loggers) for 174 cab rides in Athens. Would show up at the same taxi stand briefly one after another and take the same route in different roles.

Local Non-local: “Do you know the destination? I’m not familiar with the city.” Foreigner: Same as non-local, but in English. Varying perceived income by clothes and hotel (when applicable).

54/ 88

What drives taxi drivers? A field experiment on fraud in a market for credence goods

(Balafoutas et al., 2013) Local taken on 5% longer routes and overcharged in 5% of cases. Non-local taken on 10% longer routes and overcharged in 9%

• f cases (tariff system is the same everywhere in Greece).

Foreigner taken on 10% longer routes and overcharged in 19%

• f cases.

Richer passenger defrauded more (9% vs. 5% longer routes), mostly driven by locals.

55/ 88

The effect of priming on fraud: Evidence from a natural field experiment

(Bindra and Pearce, 2020) 600 undercover cab rides in and around Vienna Four experimental conditions:

Baseline (all treatments English speaking) Honesty: “Did you hear about that study where researchers found that around 80 % of taxi drivers were shown to behave honestly towards passengers, always taking them on the cheapest route? I read about it on the internet.” Dishonesty: “Did you hear about that study where researchers found that around 20 % of taxi drivers were shown to behave dis-honestly towards passengers, taking them on more expensive routes than necessary? I read about it on the internet.” Uber: “I checked the Uber price on line and it seemed cheap.”

56/ 88

The effect of priming on fraud: Evidence from a natural field experiment

(Bindra and Pearce, 2020) Roughly 50% of the trips involve overcharging. The Honesty and Uber treatments increase fares by around 5% relative to the Baseline treatment. No effect on distance

57/ 88

Reveal it or conceal it: On the value of second

• pinions in a low-entry-barriers credence goods

market

(Bindra et al., 2020) Mystery shoppers visited a total of 103 computer repair shops in Germany Destroyed RAM module. Unique diagnostics beep code occurred that required additional effort to diagnose the problem correctly. Three conditions:

“I bought this computer used, and it does not start.” “Another computer shop has already seen the computer and diagnosed a problem with the hardware. I would rather get a second opinion and that’s why I’m here.” “Another computer shop has already seen the computer and diagnosed a problem with the hardware which is irreparable. I would rather get a second opinion and that’s why I’m here.”

58/ 88

Reveal it or conceal it: On the value of second

• pinions in a low-entry-barriers credence goods

market

Mentioning a second opinion has no significant influence on repair probability and increases the price by roughly 20% (more for the last condition). Gathering two opinions without revealing it and taking the cheaper one saves roughly 20%.

59/ 88

Insurance coverage of customers induces dishonesty of sellers in markets for credence goods

(Kerschbamer et al., 2016) Undercover experimenter to 61 computer repair shops in Austria Computer with destroyed RAM module “I will need a bill for the repair.” vs. “I will need a bill for the repair because I have insurance that covers the repair costs.” Repair costs: 70.17e vs. 128.68e , 70% of the difference explained by additionally billed work time, 30% by unnecessary repairs. Survey two years later: Shops state that higher bills are likely due to second-degree moral hazard (that the customer cares less about minimizing the cost because a third party pays).

60/ 88

Second-degree moral hazard in a real-world credence goods market

(Balafoutas et al., 2017) 400 cab rides in Athens by female and male passengers (non-local natives). Two conditions:

“Can I get a receipt at the end of the ride?” “Can I get a receipt at the end of the ride? I need it in order to have my expenses reimbursed by my employer.”

No effect on overtreatment, but on overcharging.

61/ 88

Second-degree moral hazard in a real-world credence goods market

Figure: Source: Balafoutas et al. (2017)

62/ 88

Credence goods markets and the informational value of new media: A natural field experiment

(Kerschbamer et al., 2019) 71 computer repair shops in Germany loose RAM module Two treatments other than baseline:

Correct: “I informed myself a bit on the internet and I think that the beep is caused by a problem of the RAM modules. Maybe this helps.” Incorrect: “I informed myself a bit on the internet and I think that the beep is caused by a problem of the motherboard. Maybe this helps.”

Repair prices: 38.21e vs. 44.85e vs. 84.50e Second wave: 58 computer repair shops in Berlin (only baseline) Shops with more good reviews on the Internet charge less. Watch out: Shops with more non-recommended good reviews charge more!

63/ 88

Uncovering sophisticated discrimination with the help of credence goods markups: Evidence from a natural field experiment

(Hall et al., 2019) Four male undercover mystery shoppers: two Turks, two Syrians Cellphone repair shops in Antalya Drained battery, destroyed charging port Ordinary good: “Hi! I can’t switch on my mobile anymore and I know that a defective charging port causes the problem. Could you please repair it?” Credence good: “Hi! I can’t switch on my mobile anymore and I don’t know what the problem is. Could you please repair it?” Credence good markup is 40% (on average). No discriminatory markup for ordinary good. Credence good markup for minority is almost twice as large as for majority.

64/ 88

Outline

1

Credence Goods: Introduction and Theory

2

Credence Goods: Experiments

3

Commissions and Kickbacks

4

Literature

65/ 88

Commissions and Kickbacks

Product providers often pay commissions and kickbacks.

Kickbacks are unindisclosed. Commission means either disclosed or is a broader term.

Concerns that commissions and kickbacks may influence advisors (and experts more generally) and result in worse

• utcomes for customers.

Gifts for physicians from pharmaceutical companies. Financial advisors selling more or bad financial products.

66/ 88

Misselling through Agents: Model I

(Inderst and Ottaviani, 2009) Firm wants to sell a good to a customer through an agent. Customer (“he”)

is one of two types θ ∈ {h, l} type h with probability q, does not know his type receives utility uθ from acquiring the product, with ul < 0 < uh

Agent (“she”)

is protected by limited liability effort cost of cs for contacting a customer with probability µ receives a signal s ∼ F(s) about the customer’s type (the higher s, the more likely the customer is of type h, and the posterior probability q(s) has q(0) = 0 and q(1) = 1) advises customer on suitability

67/ 88

Misselling through Agents: Model II

Firm

specifies price p and compensation to the agent has cost kθ > 0 if the customer acquires the product cannot condition the compensation to the agent on the customer’s type receives (post-sale) a signal with probability φ if the customer has been type l and can condition the agent’s compensation

• n that and whether a sale has been made

has to pay an expected penalty ρ if sold to an agent of type l,

• f which a fraction η goes to the customer

68/ 88

Misselling through Agents: Timing

Firm sets prices. Firm specifies compensation for agent (customer does not

• bserve this).

Agent decides whether to exert effort. If she does, a customer arrives with probability µ. Agent observes s. Agent advises customer. Customer decides whether to buy. If the customer is of type l, firm gets signal with probability φ. If the customer is of type l, firm pays expected penalty ρ.

69/ 88

Misselling through Agents: Results

Note that if kl + ρ ≤ kh, the firm’s and the customer’s interests are completely misaligned; thus, the customer would not listen to the agent. Assume kl + ρ > kh. Agent will recommend to buy for s ≥ s∗ (above a suitability standard). Agent optimally gets compensated 0 if the firm receives a bad post-sale signal, w if she does not make a sale, and w + b if she makes a sale without a bad post-sale signal. Increasing w increases s∗, increasing b decreases s∗.

70/ 88

Misselling through Agents: Results

Agent gets an information rent of at least w (by not exerting any effort). This information rent increases in s∗.

If the firm increased b while holding w constant, s∗ would decrease. Thus, in order to increase s∗, the firm has to increase w.

A higher cs/µ increases the information rent.

A higher commission b necessary to induce effort. Then the intuition is the same as above.

A higher φ decreases w and also the marginal cost of increasing s∗.

This directly increases s∗ and makes blindly recommending a purchase less attractive for the agent, so the firm can increase the commission at a lower cost.

A higher cs/µ or η and a lower φ or ρ will induce the firm to choose a lower s∗ if the customer cannot observe the compensation scheme.

71/ 88

Misselling through Agents: Results

A higher cs/µ or η and a lower φ or ρ will induce the firm to choose a lower s∗ if the customer cannot observe the compensation scheme.

A higher cs/µ and lower φ increase the cost of increasing s∗ due to the agency problem. A higher η makes the customer more willing to buy, thus increasing his willingness to pay; if customers have a higher willingness to pay, the firm wants to sell more. A lower ρ makes selling to l customers more costly.

If the firm would directly sell to the customer, s∗ would be higher.

The agency problem makes it more costly to implement a high s∗.

72/ 88

Misselling through Agents: Results

A firm may have an incentive to increase ρ.

If ρ = 0, there might be a market breakdown (when the firm and customers have opposing interests).

A firm without an agency problem would choose the same ρ (and s∗) as the social planner. A firm with an agency problem would choose a lower ρ and thus s∗. The firm has an incentive to make its compensation scheme transparent (because this provides commitment), and this would results in a higher s∗. Results are qualitatively robust to advising cost. Employing two different agents for prospecting and advising would result in a higher s∗.

73/ 88

Competition through Commissions and Kickbacks: Model

(Inderst and Ottaviani, 2012a) Firms A and B, selling products A and B, respectively. Firm A is weakly more cost-efficient, cA ≤ cB. Customer wants to buy one of the two, depending on the state

• f the world θ ∈ {A, B}. Receives vh if state is matched, vl
• therwise, with vh > vl > 0.

Advisor receives a signal that results in a posterior belief q that the state of the world is A. Focus on symmetric and “nice” G(q). Advisor gets wh > wl > w0 if the customer purchases the right, the other, or no product. Focus on informative pure strategy PBE.

74/ 88

Competition through Commissions and Kickbacks: Timing

Firms set their kickbacks fn ≥ 0 for a sale of their product (observable to customer or not; customer has passive beliefs). Firms set prices pn. Advisor recommends A or B. Customer decides. Payoffs are realized.

75/ 88

Competition through Commissions and Kickbacks: Results

The advisor recommends A iff q ≥ q∗ := 1/2 − (fA − fB)/(wh − wl). Prices are equal to the expected utilities conditional on the product being recommended, given the expected q∗. Kickbacks are strategic complements. If firms are symmetric, kickbacks cancel each other out and q∗ = 1/2. If A is more cost-efficient, it has a larger incentive to increase its sales and thus pays more kickbacks; consequently, q∗ < 1/2. The sales share of A in that case becomes larger when

(wh − wl) decreases.

Commissions have the same comparative statics and are weakly lower than kickbacks.

76/ 88

Competition through Commissions and Kickbacks: Results

Disclosure reduces A’s market share in the asymmetric case. A’s market share is inefficiently low with disclosure (if wh − wl is a penalty).

Increasing the commission is less attractive if one’s market share is higher.

With kickbacks, A’s market share is too low iff wh − wl is large (and too low otherwise). Disclosure increases welfare iff wh − wl very small.

77/ 88

How (not) to pay for advice: A framework for consumer financial protection: Model

(Inderst and Ottaviani, 2012c) Firms A and B, selling products A and B, respectively. Customer wants to buy one of the two, depending on the state

• f the world θ ∈ {A, B}. Receives vh if state is matched, vl
• therwise, with vh > vl > 0.

Zero costs, B exogenously given, q0 < 1/2 (customer chooses B without advice) Advisor gets fixed payment T (potentially negative) and kickback t. Advisor gets wh > wl > w0 if the customer purchases the right

• r the other product.

A charges p, price for B is zero. Advisor charges f ≥ 0 for advising.

78/ 88

How (not) to pay for advice: A framework for consumer financial protection: Results

In equilibrium, t = p = 0 and the firm extracts all surplus via T and indirectly f. This equilibrium is efficient because advice is unbiased. Assume customers are naive and believe that they always receive unbiased advice. Then f = 0, advice is biased (potentially even uninformative), and A sells at a high price. Now assume some customers are naive and others are wary. If

• nly one contract is available, the firm will choose t = p = 0 if

there are many wary customers and otherwise exploit naive customers (and not serving wary ones). If the firm can offer two contracts, it offers the contracts for the equilibria with only naive and only wary customers. Welfare (weakly) increases with a cap on kickbacks ¯ t ≥ 0. If T ≥ 0, T = f = 0 because the firm can only extract the surplus by charging positive prices. Prices are higher for naifs.

79/ 88

A Dynamic Theory of Regulatory Capture

(De Chiara and Schwarz, 2020) Now: Government is the customer! Regulators may be influenced (“captured”) by a regulated firm and act in its interest. Bribes are illegal and often kept hidden. Regulators pursuing a career in the private sector after their employment at a regulatory agency (revolving door) is neither uncommon nor illegal in many countries. We highlight the informational advantage of the revolving door.

Firms want to signal their commitment to reward friendly regulators. Public nature of firms’ recruitment decisions helps collusive equilibria.

80/ 88

A Dynamic Theory of Regulatory Capture: Model

Infinitely-lived principal

Maximizes welfare

Infinite stream of short-lived experts.

One regulator per period Each expert maximizes her own payoff Wealth constrained

Infinitely-lived firm

Maximizes profits In each period gets private payoff of G > 0 if production allowed. Production causes unverifiable damages D > G to third parties if the state of the world is θ = U (which happens with probability q, otherwise θ = S, iid draws over time, qD < G).

81/ 88

A Dynamic Theory of Regulatory Capture: Model

Expert observes s ∈ {∅, U}

s = U if θ = U, s = ∅ if θ = S

Expert sends public report r to the principal.

Information can be concealed but not forged. Private concealment cost c ∈ {¯ c, ∞}, with c = ∞ with probability η.

At the beginning the principal commits to a production authorization policy xr ∈ {0, 1}, where x = 1 allows production; and to paying a bonus βr to the expert depending

• n the report at social cost λβr.

The firm can make a wage offer wr to the expert, depending on the report.

82/ 88

A Dynamic Theory of Regulatory Capture: Timing

The timing of the stage game is as follows: The principal commits to the authorization decision as a function of the report xr ∈ {0, 1}. Nature draws the state of the world θt and the concealment cost ct. The principal sets the bonus scheme for the experts, βt(rt) ≥ 0. If the expert accepts to work for the regulatory agency, he

• bserves the signal st.

The expert sends a public report rt to the principal who authorizes or not production according to xr and pays the expert βt. The firm may make an offer wt. The repeated game we analyze involves the infinite repetition of the stage game starting at stage 1. The principal and the firm use the same discount factor δ < 1.

83/ 88

A Dynamic Theory of Regulatory Capture: Results

In the stage game or if θ < ˜ θ, regulatory is not an issue and the principal sets βU = β∅ = 0. Otherwise, regulatory capture is an issue and the principal deters capture by βU = wMax − ¯ c iff λ ≤ ˆ λ. Suppose the firm can a hidden bribe b to the expert for reporting r = ∅. There is no self-enforcing bribery implicit contract which induces capture. With a explicit bribing technology, capture becomes an issue if it is efficient enough, i.e., τ > ˜ τ. Bribing and revolving door are perfect substitutes. Preventing bribes (or commissions and kickbacks) may lead to a use of the revolving door.

84/ 88

A Dynamic Theory of Regulatory Capture: Results

Assume there are experts with a high and with a low ability.

Experts with a high ability are productive in the industry. Experts with a low ability never collect s = U and are not productive in the industry.

If G is large and D relatively small, the principal tolerates

• capture. If λ is small, the principal deters capture with bonuses.

Otherwise, the principal opens the revolving door after a r = U. If the principal uses bonuses, she may also use cooling-off periods to reduce the bonus payments.

85/ 88

Conclusion

Credence goods are omnipresent, yet understudied! Literature overview articles: Dulleck and Kerschbamer (2006) Kerschbamer and Sutter (2017) Balafoutas and Kerschbamer (2020) Inderst and Ottaviani (2012b) Innsbruck Winter School 2021 Thank you for your attention! Questions, ideas, or comments? Email me at marco.schwarz@uibk.ac.at.

86/ 88

Literature I

Balafoutas, L., A. Beck, R. Kerschbamer, and M. Sutter (2013). What drives taxi drivers? A field experiment on fraud in a market for credence goods. Review of Economic Studies 80, 876–891. Balafoutas, L. and R. Kerschbamer (2020). Credence goods in the literature: What the past fifteen years have taught us about fraud, incentives, and the role of institutions. Journal of Behavioral and Experimental Finance, 100285. Balafoutas, L., R. Kerschbamer, M. A. Schwarz, and M. Sutter (2020). The interaction of reputation and competition in markets for credence goods. Working Paper. Balafoutas, L., R. Kerschbamer, and M. Sutter (2017). Second-degree moral hazard in a real-world credence goods market. Economic Journal 127, 1–18. Bester, H. and M. Dahm (2017). Credence goods, costly diagnosis and subjective evaluation. Economic Journal 128, 1367–1394. Bindra, P. C., R. Kerschbamer, D. Neururer, and M. Sutter (2020). Reveal it or conceal it: On the value of second opinions in a low-entry-barriers credence goods market. MPI Collective Goods Discussion Paper (2020/11). Bindra, P. C. and G. Pearce (2020). The effect of priming on fraud: Evidence from a natural field experiment. Working Paper. Darby, M. R. and E. Karni (1973). Free competition and the optimal amount of fraud. Journal of Law and Economics 16, 67–88. De Chiara, A. and M. A. Schwarz (2020). A dynamic theory of regulatory capture. Working Paper. Dulleck, U. and R. Kerschbamer (2006). On doctors, mechanics, and computer specialists: The economics of credence goods. Journal of Economic Literature 44(1), 5–42. Dulleck, U. and R. Kerschbamer (2007). Price discrimination via the choice of distribution channels. In Proceedings of the 2007 Australasian Meeting of the Econometric Society:, pp. 1–35. The University of Queensland. Dulleck, U. and R. Kerschbamer (2009). Experts vs. discounters: Consumer free-riding and experts withholding advice in markets for credence goods. International Journal of Industrial Organization 27, 15 – 23. Dulleck, U., R. Kerschbamer, and M. Sutter (2011). The economics of credence goods: An experiment on the role of liability, verifiability, reputation, and competition. American Economic Review 101, 526–555. Ely, J., D. Fudenberg, and D. K. Levine (2008). When is Reputation Bad? Games and Economic Behavior 63(2), 498–526. Ely, J. C. and J. V¨ alim¨ aki (2003). Bad reputation. Quarterly Journal of Economics 118(3), 785–814.

87/ 88

Literature II

Fong, Y.-f. (2005). When do experts cheat and whom do they target? RAND Journal of Economics 36, 113–130. Gottschalk, F., W. Mimra, and C. Waibel (2020). Health services as credence goods: A field experiment. Economic Journal. Forthcoming. Grosskopf, B. and R. Sarin (2010, December). Is reputation good or bad? an experiment. American Economic Review 100(5), 2187–2204. Hall, J., R. Kerschbamer, D. Neururer, and E. Skoog (2019). Uncovering sophisticated discrimination with the help of credence goods markups: Evidence from a natural field experiment. Technical report, Working Papers in Economics and Statistics. Heinzel, J. (2019). Credence goods markets with heterogeneous experts. Working Paper, Center for International Economics,

• No. 2019-01.

Hilger, N. G. (2016). Why don’t people trust experts? Journal of Law and Economics 59, 293–311. Inderst, R. and M. Ottaviani (2009). Misselling through Agents. American Economic Review 99(3), 883–908. Inderst, R. and M. Ottaviani (2012a). Competition through commissions and kickbacks. The American Economic Review 102, 780–809. Inderst, R. and M. Ottaviani (2012b). Financial advice. Journal of Economic Literature 50, 494–512. Inderst, R. and M. Ottaviani (2012c). How (not) to pay for advice: A framework for consumer financial protection. Journal of Financial Economics 105, 393–411. Jost, P.-J., S. Reik, and A. Ressi (2019). The information paradox in a monopolist’s credence goods market. mimeo. Kerschbamer, R., D. Neururer, and M. Sutter (2016). Insurance coverage of customers induces dishonesty of sellers in markets for credence goods. Proceedings of the National Academy of Science 113, 7454–7458. Kerschbamer, R., D. Neururer, and M. Sutter (2019). Credence goods markets and the informational value of new media: A natural field experiment. MPI Collective Goods Discussion Paper (2019/3). Kerschbamer, R. and M. Sutter (2017). The economics of credence goods—a survey of recent lab and field experiments. CESifo Economic Studies 63(1), 1–23. Liu, F., A. Rasch, M. A. Schwarz, and C. Waibel (2019). The role of diagnostic ability in markets for expert services. Working Paper.

88/ 88

Literature III

Mimra, W., A. Rasch, and C. Waibel (2016a). Price competition and reputation in credence goods markets: Experimental

• evidence. Games and Economic Behavior 100, 337–352.

Mimra, W., A. Rasch, and C. Waibel (2016b). Second opinions in markets for expert services: Experimental evidence. Journal

• f Economic Behavior & Organization 131, 106–125.

Rasch, A. and C. Waibel (2018). What drives fraud in a credence goods market? Evidence from a field study. Oxford Bulletin

• f Economics and Statistics 80, 605–624.

Richardson, H. (1999). The Credence Good Problem and the Organization of Health Care Markets. Working Paper. Schneider, H. S. (2012). Agency problems and reputation in expert services: Evidence from auto repair. Journal of Industrial Economics 60, 406–433. S¨ ulzle, K. and A. Wambach (2005). Insurance in a Market for Credence Goods. Journal of Risk and Insurance 72(1), 159–176. Taylor, C. R. (1995). The economics of breakdowns, checkups, and cures. Journal of Political Economy 103, 53–74.