The Morality of Markets and the Nature of Competition∗

Mathias Dewatripont† Jean Tirole‡

December 18, 2020

Abstract: Does competition undermine ethical behavior when suppliers and stakeholders have social preferences? Under price regulation (health, schools, regulated professions, franchising), a higher price and competitive pressure compromise (boost) ethics when cutting ethical corners increases (reduces) demand. Competitive pressure has no ethical impact when reduced ethics lowers cost and/or with market-determined prices. A more ethical ﬁrm commands a lower market share under price regulation, but not in a “low-morality market” with market prices. Finally, intense competition induces behavioral convergence among ethical types and/or corporate forms, to the better on the price front and the worse on the ethical front.

Keywords: Competition, social preferences, replacement excuse, non-proﬁts, corporate social responsability, strategic complementarities, race to the ethical bottom.

JEL numbers: D21, D43, D6, I11, L21.

∗This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement no. 669217 - ERC MARK- LIM). Jean Tirole acknowledges funding from ANR under grant ANR-17-EURE-0010 (Investissements d’Avenir program), and financial support from the Chaire SCOR “Market risks and Value creation”. We gratefully acknowledge the comments of participants at a Bonn symposium in honor of Martin Hellwig, the TSE conference on “Markets, Morality and Social Responsibility”, the Chicago-LSE “Economics of Social Sector Organizations conference”, as well as a seminar at UniversitéLibre de Bruxelles (ECARES and I3h), and of Tim Besley, AiméBierdel, Paul-Henri Moisson, Charles Pébereau and Joel Sobel. †Universitélibre de Bruxelles (Solvay Brussels School of Economics and Management, ECARES and I3h). ‡Toulouse School of Economics (TSE) and Institute for Advanced Study in Toulouse (IAST). 1 Introduction

Does the market obliterate our moral compass? While 18th century thinkers viewed markets as creating trust among otherwise unrelated individuals (Montesquieu’s “doux com- merce”), today’s public opinion, many social scientists, politicians and religious leaders feel that markets promote unethical behavior. For instance, numerous prominent philoso- phers have warned against the religion of the marketplace, with a variety of viewpoints from the necessity to ban repugnant markets to the stance that a market economy is an unlikely path to a harmonious society.1 Furthermore, recent experimental work demonstrates the power of the “replacement excuse”, the idea that if a supplier refuses to engage in an immoral trade, “someone else will”. This work echoes widespread narratives used by firms and countries selling weapons to dictatorships or bribing officials to win a contract, by banks selling toxic products or providing short-term incentives to talents they want to attract, by employees ingratiating themselves to their superiors in order to be promoted, by doctors overprescribing opioids, or by professional athletes taking illicit drugs to defeat their competitors.2 Indeed, the replacement excuse has impeccable logic to anyone with a consequentialist bent. On the theoretical front, Dufwenberg et al. (2011) and Sobel (2015) identified conditions under which an economy whose agents have other-regarding preferences delivers the same allocation as if these agents were perfectly selfish, i.e., when perfectly competitive markets destroy any velleity of doing good. This paper takes on board the idea that markets may undermine ethical values by creating alternatives and studies how competition impacts ethical behavior when these conditions are not met. The “nature of competition” in the title of the paper will refer to five attributes that define the suppliers’competitive environmment: (a) market structure (number of firms, product substitutability), (b) price flexibility (whether prices are constrained or market determined), (c) whether cutting ethical corners boosts or reduces demand, (d) rivals’ ethical incentives (their ethical preferences and corporate status), and (e) the social responsibility of input suppliers (investors and workers). The situation in which firms lack price flexibility is much more common than one would expect. First, prices may be fixed by a regulatory authority, a private franchisor or a platform. Price inflexibility is a good approximation of many publicly regulated markets (taxis, notaries) and unregulated ones (e.g. apps and franchising environments). Other

1E.g. Anderson (1993), Sandel (2012), Satz (2010), and Walzer (2008). 2Opioid overconsumption illustrates internalities, given the addictive nature of such painkillers. Opi- oids represent both a useful treatment for acute pain (e.g. in case of terminal cancer) but also run the risk of addiction without proven medical beneﬁts in the case of chronic pain (e.g. back pain). Opioid overdoses have been called the worst drug epidemic in the history of the United States (McGreal 2018). The crisis has multiple dimensions, including the role of companies like Purdue Pharma in inducing doctors to prescribe their opioid OxyContin. Our paper focuses on doctors’ decision when facing patient demands for opioid prescriptions (see Schnell 2019 for a recent assessment of policies aimed at keeping opioid prescriptions in check).

1 prominent industries in which providers receive externally-set prices are education -the price is the value of a voucher-, and health- for which the price is the payment allowed by the Social Security system, Medicare or an HMO-. Second, prices in two-sided markets are often constrained by arbitrage to be non-negative; we will therefore often use the vocabulary “constrained prices” instead of “regulated prices”. The framework, developed in Section 2, posits that suppliers play an important role in shaping the morality of markets, in line with Henderson (2020)’s view that managers are key engines for “reimagining capitalism” (we later embody stakeholders’ ethical concerns as well). The framework accordingly assumes that suppliers are driven by both a profit motive and an ethical concern. “Ethics”, though, has a variety of meanings; this paper builds on the familiar social-preferences paradigm. Social preferences imply that suppliers internalize part of their impact on social welfare,3 and that they reflect on what would happen if they did not serve the client, consistent with the evidence on the replacement excuse. We will use the “moral” or “ethics” terminology in reference to the social-preferences paradigm, although it certainly does not capture the richness of ethical thinking. We later analyze the robustness of our insights to a few alternative ethical criteria. We initially assume that suppliers’ moral compass is driven by a disconnect between what is desired by the client and what is good for society. The first possible wedge may be traced to an internality (as is the case when a doctor overprescribes opioids, which is attractive to the client’s “current self” but, being addictive, detrimental to her “long-term self”). Second, the disconnect may stem from an externality (as when doctors deliver fake medical certificates to allow their client not to be vaccinated or to take sick leave, or when a firm bribes an official). A third wedge is associated with shrouded attributes,4 as when a supplier misrepresents the product or exploits the client’s incorrect prior or inattention. We provide numerous examples of these three wedges between client and social demands. The three wedges are unified within a single framework. The paper thus focuses first on the differential in assessments of quality by customers and a social planner. [Later, it will show how results extend or change when cutting ethical corners does not boost demand but reduces the production cost.] A bit more formally, the oligopolists wage price and non-price competition. Supplier i’s demand, Di, depends on her net price, pi − ui, where ui is the gross surplus or quality perceived by consumers, and on the rivals’ net prices. The novel aspect is that because of an internality, an externality or a shrouded attribute, the social welfare associated with the transaction, wi = W (ui), is decreasing over the relevant range. We look at the standard model of oligopolistic competition with unit demands, augmented with social preferences: Besides her material

3That economic agents may behave altruistically has received support in experimental economics and is a common assumption in the literature on social responsibility (see e.g. Besley-Ghatak 2018, Broccardo et al 2020, Green-Roth 2020, Hart-Zingales 1997, Landier-Lovo 2020, Oehmke-Opp 2020). 4We borrow the terminology from Gabaix and Laibson (2006)’s work on imperfect competition subject to such attributes.

2 P payoff, supplier i internalizes a fraction αi of social welfare, wiDi + j6=i wjDj. Thus, with P marginal cost c and price pi, her objective function is (pi − c)Di + αi(wiDi + j6=i wjDj). Like other works building on social preferences, we assume that tort law does not provide a perfect Pigovian correction of the wedge between market and society’s demands (which is the case for the applications envisioned throughout the paper). This may hold for multiple familiar reasons.5 The imperfect Pigovian taxation leaves scope for an internalization of the wedge by suppliers with social preferences. For notational simplicity, we will ignore such taxation. Section 2 obtains some preliminary results. Supplier i’s ethical choice is increasing in the product, ηiLi, of the firm’s elasticity of demand and a generalized Lerner index. The latter is the standard markup-over-price ratio, except that the marginal cost is reduced by the supplier’s internalized “social responsibility index”, Li ≡ [pi − (c − αiSi)]/pi. This social responsibility index, Si ≡ cov(σij, wi −wj), captures supplier i’s competitive impact on overall welfare and is equal to the covariance between her relative substitutability with other suppliers (σij, the fraction of market share gained from supplier j when lowering P price or increasing surplus: j6=i σij = 1) and her relative welfare impact compared with these suppliers. The social responsibility index takes a particularly simple form when the equilibrium is symmetric (Si = 0 since all suppliers make the same ethical choice) or when an increase in supplier i’s market share impacts equally all other suppliers (σij = σik, and P so Si = j6=i[(wi − wj)/(n − 1)]). The two cases are developed in more details in Sections 3 and 4, respectively. Section 2 then establishes two key results. First, fixing prices (say, the prices are constrained), ethical choices are strategic complements: the more unethically I behave, the more unethically you do. This results from two effects. When I attempt to please the consumer, I increase the competitive pressure on you and, under a standard assumption, raise the elasticity of your demand; when prices are fixed, the only way for you to restore your competitiveness is to also behave less ethically. This “elasticity effect” is comple- mented by a “social responsibility effect” that results from the previous discussion: The less ethically I behave, the higher your moral urge to take market share away from me; when prices are fixed, the only way to do so is for you to behave less ethically. Second, when prices are market-determined, the familiar formula “Lerner index = inverse elasticity of demand” generalizes, so that Li = 1/ηi. Under market prices, strategic neutrality therefore obtains: Supplier i’s ethical behavior depends only on her own ethical concerns (αi) and is invariant to market structure (number of firms, product substitutability) or other suppliers’ ethical concerns. When prices are market-determined, suppliers wage good and bad competition. They can lower price or cut ethical corners to

5(a) Limited liability, risk aversion or managerial turnover may prevent the collection of the Pigovian tax; (b) there may be no political will to levy or enforce such a tax (for example due to lobbying); (c) laws may embody loopholes; (d) the behavior may not be veriﬁable (e.g. corruption, doctors’ “judgement”, “interpretation of advertising”); (e) law enforcement is too costly for minor misdemeanors (Kaplow- Shavell 2007); (f) a last reason is the inability to tax externalities, such as the use of child labor, corruption or pollution, exerted in other jurisdictions when suppliers produce abroad.

3 attract consumers. Like in the case of constrained prices, a stronger competition gives more prominence to the possibility of replacement and thereby lowers the incentive to be virtuous; however, more competition also reduces prices and therefore the stake in attracting clients through unethical behavior. The two effects, associated with price and non-price competition, exactly offset each other. Section 3 specializes to the symmetric-oligopoly case and shows that the extent to which ethical concerns help reduce the wedge between equilibrium ethical choices and socially desirable ones depends on market conditions. Unethical behavior under constrained prices is more likely, the higher the stake (the markup), and the more competitive the market, vindicating the view that non-price competition may work against ethical behavior. The latter result’s intuition is that ethical behavior implies a substantial loss in market share in unconcentrated markets. When cutting ethical corners does not boost demand, but reduces the cost of production, ethical behavior is invariant to whether prices are constrained or not, generating a testable difference between the two ways of cutting ethical corners. Intuitively, cutting ethical corners on the supply chain front does not affect demand as long as consumers do not have social preferences or are ill-informed. The ethical choice is then driven by personal ethics, but not by competitive pressure. Section 3 then allows stakeholders (consumers, investors, workers) and not only suppliers (managers, entrepreneurs, doctors, depending on the application) to have social preferences. The following results emerge: (i) All results derived under supplier altruism extend to investor and worker altruism. (ii) In the case of consumer altruism, the invariance result under market prices again applies, but under constrained prices a strengthening of competition makes the market more moral since now a more ethical behavior boosts demand. Finally, Section 3 allows for decreasing marginal utilities of income. An increase in competition or a reduction in the regulated price then has an immiseration effect that encourages suppliers to cut ethical corners. Under constrained prices, the impoverishment and replacement effects both imply that an increase in the number of competitors makes the market less moral. By contrast, the two effects have opposite implications for a decrease in the fee, making the impact of the latter on market morality ambiguous. Under market prices, the invariance result no longer holds; an increase in the number of competitors (or an increase in the substitutability among them) unambiguously makes the market less moral. Section 4 allows for supplier heterogeneity (different moral preferences or different corporate forms) under the assumption of symmetric impacts. It first studies the equilibrium under heterogeneous ethical concerns. Unsurprisingly, with constrained prices, less ethical suppliers command a higher market share than more ethical ones. Under market-determined prices, a supplier with stronger social preferences, while still exhibit- ing a more ethical behavior, also sets a lower price. Intuitively, the more moral type offers a more ethical, but less attractive service, and so must charge lower prices; further-

4 more, she is eager to gain market share as she is more virtuous than the other suppliers, inducing her to further lower the price. Under a linear demand system, a more ethical supplier commands a higher market share than a low-ethics one if and only if the fraction of low-ethics suppliers is large enough (the market is a “low-morality market”). Intense competition makes more ethical firms behave like not-for-profits and ultimately erases behavioral heterogeneity among ethical types. Section 4 then studies competition between for- and not-for-profit entities when prices are market-determined. In conformity with previous results, the low-powered incentives faced by not-for-profits make them behave more ethically. Interestingly, when competition is intense, there is a two-sided mimetism, resulting in a convergence of behavior: Not-for- profits force for-profit firms to levy low markups, but they align their own ethical behavior to the level of for-profits in order to keep market share. This suggests that it is difficult to tell for- and not-for-profits apart when product market competition is intense. The results are consistent with available evidence. Section 5 discusses possible extensions (a) that allow social norms to make ethical behavior context-dependent and (b) in which the exchange institution affects moral preferences. It then studies the robustness of our insights to alternative moral criteria. It considers three alternative moral preferences: a narrow moral mandate (in which the individual cares only about the welfare associated with her own sales), which exhibits the strategic complementary that is characteristic of the replacement excuse; Kant’s deontologism (under which the individual cares about her selected action rather than about its consequences) and Kant’s categorical imperative, for which this strategic complementarity fails. We thus obtain testable differences in the predictions of consequentialist and Kantian moral criteria. Section 6 relates the paper to the existing literature. Section 7 summarizes the main insights and concludes with some alleys for future research. Omitted proofs are relegated to the Appendix.

2 Framework and ﬁrst analysis

2.1 A model of competition under ethical concerns

There are n suppliers, i ∈ {1, . . . , n} and a mass 1 of unit-demand clients. Suppliers provide a service for clients. The cost of providing this service is c per client. Ethical choice. Suppliers compete in a price and a non-price dimensions. The latter involves an ethical choice and is characterized by a perceived gross surplus u ∈ [0, u¯], expressed in monetary terms, for the client and a resulting impact w on social welfare. The “Pareto frontier,” pictured in Figure 1, obeys the following properties (which are satisﬁed by our various applications):

5 Assumption 1 (Pareto frontier). The efficient frontier w = W (u) is a smooth function on [0, u¯] and is single peaked, with a peak at uˆ ∈ (0, u¯) (such that wˆ ≡ W (û) = max W (u)). Furthermore, W is concave on [û, u¯] and W 0(¯u) = −∞.

wˆ Wu()

u uˆ u ethical dilemma region

Figure 1: Pareto frontier.

The relevant region for supplier i is ui ∈ [ˆu, u¯], since raising ui when ui < uˆ both increases demand and is more moral.

Demand functions. Let {pi}i∈{1,...,n} and {ui}i∈{1,...,n} denote the prices charged by the suppliers and the gross consumer surpluses that they deliver (similarly {wi ≡ W (ui)} denote the corresponding welfares). Net prices are denotedp ˆi ≡ pi − ui. The demand Di(ˆp) for supplier i depends on the vector of net prices ˆp = (ˆp1,..., pˆi,..., pˆn). We assume that, in the relevant range of parameters, the market is covered, that is

ΣiDi = 1; for example everyone uses a doctor. As shown later, the results extend to a downward sloping total demand.

We will consider two cases: (i) Constrained prices: pi = p for all i, where p is ﬁxed exogenously.6 (ii) Market-determined prices: unconstrained suppliers set their prices non- cooperatively.

The suppliers are substitutes (∂Di/∂pˆi < 0 < ∂Di/∂pˆj) whenever 0 < Di < 1, and their marginal revenue is decreasing in price ((pi − c)Di(ˆp) is concave in pi). We will let ηi ≡ (−∂Di/∂pˆi)/(Di/pi) denote the price elasticity of demand for supplier i’s services (which we assume is bounded above).

6As discussed in the introduction, the price p may be set by a regulator. The assumption of regulated prices is reasonable for medical markets in a number of countries. But it may also be relevant in economies in which prices are negotiated between, say, an HMO and a hospital or a group of doctors. Other examples of activities with regulated prices include taxis, schools in a voucher system, apps whose prices are capped by platforms (examples include no-surcharge rules, and best-price guarantees), and franchises whose tariﬀs are set by franchisors. Finally, prices may be “constrained” rather than “regulated”, as an important subclass of ﬁxed prices is associated with the zero lower bound on pricing, so ubiquitous in the tech industry (see Section 2.3).

6 Assumption 2 (elasticity of demand). Supplier i’s elasticity of demand decreases when competitive pressure is relaxed: ∂η i < 0. ∂pj

The condition for competition to increase the elasticity of demand (∂ηi/∂pj < 0), 2 namely ∂Di/∂pj ≥ −[Di/(−∂Di/∂pi)] (∂ Di/∂pi∂pj), is closely related to the one stating 2 that the goods are strategic complements, namely ∂Di/∂pj ≥ −(pi − c)(∂ Di/∂pi∂pj). The left-hand side of both conditions, ∂Di/∂pj, is positive and reflects the fact that the increased competition reduces the number of supplier i’s inframarginal consumers. 2 A sufficient condition for either is ∂ Di/∂pi∂pj ≥ 0. Furthermore, the two conditions coincide when supplier i’s price is optimized. Objective functions. Suppliers care about profit, but have social preferences, as reflected 7 in their partial internalization of overall welfare. Let αi ≥ 0 denote supplier i’s (common knowledge) weight on welfare relative to that on profit.8 Supplier i maximizes the sum of profit and internalized social welfare.9

Vi ≡ (pi − c)Di + αi[wiDi + Σj6=iwjDj].

Suppliers’ market behavior will thus depend on the relative weight they put on profit and social welfare. For example, the behavior of a supplier i who is an agent with social preferences αi and receives a fraction ξi of the profit associated with their activity is in- distinguishable from that of a residual claimant for the firm’s profit with social preference parameterα î ≡ αi/ξi. Such reinterpretations must be kept in mind when thinking about the opioid scandal, as Purdue pharma had access to the doctors’ prescription data and could (and did) provide high-powered incentives to its sales representatives.

2.2 Some preliminary results and the deﬁnition of the replacement excuse

Consider, first, the first-order condition with respect to the ethical choice (∂Vi/∂ui = 0). Behaving less ethically (increasing ui) has three effects:

7See Section 5 for alternative moral imperatives. 8Social preferences reflect a “broad mandate”, using Oehmke-Opp (2020)’s terminology. See also Green and Roth (2020), who contrast sophisticated and naive social investors (or “impact” and “value” investors). 9We could add consumer net surplus from consumption (weighted by α), but this would not change the derivations below; for, a small change in i’s policy implies a shift, from or toward alternative suppliers, of marginal consumers who by definition are indifferent between supplier i and their best alternative supplier. So the marginal impact on consumer surplus is 0. The reasoning holds for constrained as well as for competitive prices.

7 ∂Di h ∂Di ∂Dj i 0 (pi − c) + αi wi + Σj6=iwj + αiW (ui)Di = 0. ∂ui ∂ui ∂ui | {z } | {z } | {z } negative ethical profit ethical impact of impact relative to increase gain in market share inframarginal customers A relaxation of supplier i’s ethical standards increases her demand, resulting in an increase in profit proportional to her markup (pi − c). Supplier i also gains market share at the detriment of the other suppliers. The ethical impact of this reallocation of market shares depends on the relative morality of the suppliers, as we will analyze below. Finally, the increase in ui has a negative ethical impact on supplier i’s inframarginal consumers. The latter two effects are proportional to supplier i’s social preferences parameter αi.

Using the fact that ∂Di/∂ui = −∂Di/∂pi, one can rewrite this ﬁrst-order condition in the following simple form: 0 −αiW (ui) = ηiLi, (1) where ηi is the elasticity of demand facing supplier i and Li is the “generalized Lerner index” pi − (c − αiSi) Li ≡ . pi

Si is the “social responsibility index”:

Si ≡ Σj6=iσij(wi − wj), where σij ≡ [∂Dj/∂pi]/[−∂Di/∂pi] measures the fraction of the market share gain by supplier i that comes from supplier j’s customers when supplier i lowers her price by one unit; it satisﬁes Σj6=iσij = 1. The social responsibility index captures supplier i’s competitive impact on overall welfare and is equal to the covariance between her relative substitutability with other suppliers and her relative ethical impact compared with these suppliers. Si is positive if supplier i is on average more ethical than her rivals and negative otherwise. The social responsibility index takes a particularly simple form in the following two cases, which will be studied in Sections 3 and 4, respectively:

(a) Symmetric equilibrium. Suppose that the model is symmetrical: Demand functions

are symmetric (and so σij = σik = 1/(n − 1) for arbitrary but diﬀerent i, j, k) and social preferences are identical (αi = α). Then in a symmetric equilibrium, wi = wj for all j and so

Si = 0. Whether the symmetric price p is constrained or results from equilibrium behavior,

the generalized Lerner index Li is equal to the ordinary Lerner index: p − c L = L ≡ . i p

8 (b) Symmetric impact. When ∂Dj = ∂Dk for all (j, k), as is the case for example for ∂pi ∂pi 1 Σk6=j pˆk the linear demand system Dj ≡ n − σ pˆj − n−1 , but the equilibrium is not necessarily symmetric (say, because social preferences diﬀer), then σij = 1/(n − 1) Σj6=iwj 10 and so, lettingw ¯−i ≡ n−1 denote the average welfare footprint of i’s rivals,

Si = wi − w¯−i.

We can now more precisely deﬁne the replacement excuse. To do so, return to condition

(1) for a given price vector p, and consider how supplier i changes her ethical behavior ui when supplier j behaves less ethically (increases uj). The familiar phrase “If I don’t do it, someone else will”, that captures the essence of the replacement excuse can be understood by looking at the impact of uj on ηi and Li:

• Elasticity eﬀect. From Assumption 2, the elasticity facing supplier i is increasing in uj, and so an increase in uj induces an increase in ui.

• Social responsibility eﬀect. Second, an increase in uj lowers wj and thusw ¯−i, raising the social responsibility index Si. The less ethical the rivals, the higher the moral urge to take market share away from them; when prices are ﬁxed, the only way to do so is to be less ethical oneself.

Both effects make ethical behaviors strategic complements. Section 5 compares our “broad mandate” model with a “narrow mandate” version in which supplier i looks only at the morality of her choices, independently of the ethical behavior of her rivals: Vi = [pi − c + αiwi]Di. The elasticity effect is then the same as for the broad mandate, but the social responsibility effect is nil (Li = (p−c)/p is independent of uj). Ethical behaviors are still strategic complements. The replacement excuse, “if I don’t do it, someone else will”, can be understood as reflecting either the elasticity effect alone or both effects. The familiar phrase explicitly suggests a need for less ethical behavior when facing a strong competitive pressure. More implicitly, the “someone else” is behaving in an unethical manner (the mere loss of market share per se does not seem a strong moral defense). This is a positive argument in favor of the broad mandate, which in our view already commands the moral high ground relative to the narrow mandate on the normative front. But we are agnostic; agents may well reason on the basis of a narrow mandate. Section 5 will check that both mandates deliver similar results, unlike some other moral criteria. When prices are constrained at some level p, ethical behavior is driven by the elasticity and social responsibility effects. Under market prices, the first-order condition with respect to pi is:

h ∂Di i h ∂Di ∂Dj i Di + (pi − c) + αi wi + Σj6=iwj = 0 ∂pi ∂pi ∂pi | {z } | {z } impact on proﬁt impact on ethical outcome

10 For a duopoly, it is always the case that Si = wi − wj.

9 rewriting of this condition yields a generalization of the familiar formula “Lerner index = inverse elasticity of demand”: 1 Li = . (2) ηi Combining (1) and (2) then yields

0 −αiW (ui) = 1. (3) Proposition 1 (strategic interdependence.)

(i) For given prices (and so in particular in the constrained-prices environment), both the elasticity eﬀect and the social responsibility eﬀect make ethical behaviors strategic complements.

(ii) Under market prices, strategic neutrality obtains: Supplier i’s ethical behavior de-

pends only on her own ethical concerns (αi) and is invariant to market structure (number of ﬁrms, product substitutability) or other suppliers’ ethical concerns.

2.3 Three wedges between consumer and social demands

We provide three illustrations of the model (this section can be skipped by readers who are familiar with the three wedges). In the first two, being ethical means “being less nice” to the client (or at least her current self); in the third one, being ethical means “being nicer” to the client. In all cases, there is a wedge between the quality perceived by the customer and that assessed by the social planner; and being more ethical means reducing demand. Example 1: Internalities (painkiller prescriptions). The supplier (a doctor) decides whether to prescribe an opioid to the client (the patient). The fee pi paid by the patient is either regulated or competitive, and is paid for the visit, regardless of what the doctor will prescribe. The painkiller brings known benefit b, but has side effects with cost γ. This cost is observed only by the doctor (who learns who is at risk) and is distributed according to distribution G(γ) and density g(γ). The doctor chooses a threshold γ∗ under which she prescribes the painkiller. Assume that the patient knows γ∗; one may have in mind that patients know the doctor’s reputation for being easy (pill mill doctor) or tough on prescriptions. Welfare is b − γ, but clients have hyperbolic preferences with coefficient β < 1: They long for quick relief and value the prescription at b − βγ. And so ∗ ∗ Z γi Z γi ui = (b − βγ)g(γ)dγ while wi = (b − γ)g(γ)dγ. 0 0

∗ R b/β The maximum gross surplus corresponds to γi = b/β and is equal tou ¯ ≡ 0 (b − R b/β ∗ βγ)g(γ)dγ, yieldingw ¯ ≡ 0 (b − γ)g(γ)dγ. The welfare optimum corresponds to γi = b and so Z b Z b wˆ ≡ (b − γ)g(γ)dγ andu ˆ ≡ (b − βγ)g(γ)dγ. 0 0

10 ∗ More generally, the function wi = W (ui) is obtained by substituting γ , and satisfies Assumption 1.11 Instances of overconsumption due to imperfect self-control or biases in predicting one’s future behavior are many outside the health domain (excessive indebtedness, gambling, videogaming, impulsive clicking on privacy consent forms. . . ). Example 2: Externalities (vaccines, overprescription of branded drugs). This example replaces the internality of Example 1 by an externality. Patients have heterogenous prob- abilities x of getting sick in the absence of vaccination, in which case they suffer damage d and contaminate an expected number e of other people. Patients are selfish and value being vaccinated at E[b − γ], where b = xd is the benefit and γ is a cost of vaccination. The social planner attaches value E[(1+e)b−γ]. It is easy to check12 that this externality example is mathematically akin to the internality example, Example 1.13 Underconsumption, unlike overconsumption, raises the question of how the supplier can provide a quantity that exceeds the client’s desired consumption: a doctor cannot physically vaccinate a patient who refuses to be inoculated. A first interpretation of the underconsumption model goes as follows: when the state mandates children to be vaccinated in order to be able to go to school or public sport facilities, parents may try to obtain a complacent (fake) vaccination certificate from the doctor (underconsumption of vaccines). Similarly, in some countries, occupational physicians may routinely deliver fake medical certificates allowing employees to take paid sick leave (underprovision of work). In both examples, unethical supplier behavior is associated with a fraudulent report to a third party. A second interpretation applies when no law or rule mandates a level of consumption in excess to that desired by the client. Ethical/unethical behavior then relates to the intensity with which the doctor puts pressure on the patient, say to be vaccinated; it may range from attempts at persuasion to outright refusal to keep seeing a patient who refuses the vaccination. Overconsumption occurs in the case of antibiotics. Another case in point is the overconsumption of branded drugs when generics are available, imposing an externality on the social security system. A fraction of French patients has always viewed generics as inferior products. Until the mid-90s French doctors faced no cost of prescribing branded drugs instead of generics (and pharmacists’ compensation was proportional to the price of the drug!). Lo and behold, doctors pandered to their patients and generics’ market share was about 2%. A reform introduced incentives for doctors to prescribe generics, and also gave

11One has d b − γ∗ . du W 00(u) = < 0. dγ∗ b − βγ∗ dγ∗

12 ∗ R ∞ R ∞ One has u ≡ b∗ (b − γ)dF (b) and W ≡ b∗ [(1 + e)b − γ]dF (b). A supplier with ethical concerns will ∗ ∗ γ 0 ∗ (1+e)b −γ 00 eγ db∗ choose b in 1+e , γ . Then W (u ) = b∗−γ < 0 and W (u) = (b∗−γ)2 du < 0. 13Underconsumption of vaccines may also be driven by a misperception of their side-eﬀects. For example, a triple jab of the measles vaccine was falsely accused in The Lancet of causing autism, which led to a substantial drop in MMR vaccination. Such misperceptions way be captured as an underestimation of the net value of the vaccine, independently of contagion considerations.

11 pharmacists the ability to replace a branded drug by an equivalent generic. The share of generics’ prescriptions improved, especially with general practitioners (whose patients are more loyal than for specialists, in conformity with the theory developed below). But the low percentage (36%)14 of generics among prescriptions reimbursed by the social security system suggests that there is still substantial pandering. Example 3: Shrouded attributes. Tobacco companies’ advertising failed to warn against the harmful effects of smoking. More generally, firms typically emphasize positive attributes of their goods and services and rarely their flaws. To be certain, consumer protection agencies’ and courts’ mission is to combat inappropriate statements or frauds. But there is a thin line between outright misrepresentation and fraud on the one hand, and omis- sion, opaque language and the exploitation of consumer inattention on the other hand. Unexpected obsolescence, vague recommended usage or the downplaying of side effects may not be illegal or else hard to regulate given their ubiquity and the limited means of the agencies. Like internalities and externalities, shrouded attributes leave scope for moral judgment. One way to formalize this within our model goes as follows: Suppose that the good actually delivers gross surplusu ˆ to the consumers. Supplier i can inflate this surplus and claim it is ui ∈ [û, u¯]. Consumers take the announcement at face value (see below for a more rational version) and plan around the announced value, leading to later incon- venience (complementary investments miscalibration, misleading claims made to down- 0 0 00 stream users. . . ) cost Γ(ui −uˆ), with Γ(0) = 0, Γ (0) = 0, Γ (ui −uˆ) > 0 and Γ (ui −uˆ) > 0 0 for ui > uˆ, and Γ (¯u − uˆ) = +∞ for some maximum exaggeration levelu ¯ − uˆ. Then W (u) ≡ uˆ − Γ(u − uˆ) satisfies the general assumptions. Supplier i’s objective function is

(pi − c)Di(ˆp) + αi[W (ui)Di(ˆp) + Σj6=iW (uj)Dj(ˆp)]. Finally, note that a more rational, asymmetric-information, version of the model would have consumers not know about the misreporting function. For example, with some probability they believe that misreporting is infeasible; the “irrational version” is just the limit of the “rational version” as this probability goes to 1. The three wedges together; the case of apps. The tech industry illustrates both the possibility of constrained prices and the three factors of wedge between consumer demand and social welfare. First, note that many applications (such as search, videos, GPS navigation software, social networks. . . ) are available free of charge to the consumer (pi = 0). As emphasized in the industrial organization literature, such services have negative oppor- tunity costs (the proﬁt from targeted advertising and data collection far outweighs their tiny physical marginal cost), so that their optimal price is constrained by the zero lower bound linked to the threat of arbitrage by bots. Apps also exhibit the “wedge trilogy”. First, we all click impatiently, perhaps overem- phasizing our immediate gratiﬁcation over the long-term loss of privacy (internality). Some apps indeed design their privacy policy and consent procedure so as to exploit our

14Generics penetration in the US, UK and Germany exceeds 80%.

12 impatience. Second, our e-mails and postings on social networks reveal information about others, violating their privacy (externality) and apps may exploit these “data externalities”.15 Third, the false sense of privacy experienced by most consumers is an illustration of shrouded attributes; for instance, most users of the “Do not track” privacy setting do not know that big tech companies do not abide by the spirit of the request16. More broadly, tech companies work on reassuring consumers more than on really protecting their privacy.

3 Symmetric oligopoly

This section assumes that the demand functions are symmetrical and that Di is invariant to permutations of the net prices {pˆj}j6=i. It investigates the impact of competition on equilibrium ethics.

3.1 Constrained prices

Consider an arbitrary constrained price p ≥ c.

Assumption 3 Suppose that the price is constrained at some level, pi = p ≥ c for all i. Let η(p − u∗) ≡ − ∂Di(ˆp) np denote the elasticity of demand in a symmetric equilibrium ∂pi ∗ (ui = u for all i) and L(p) ≡ (p − c)/p denote the ordinary Lerner index. Then, for all u∗ ∈ [ˆu, u¯],

∂(η(p − u∗)L(p)) (i) > 0, ∂p ∂η(p − u∗)/∂u∗ W 00(u∗) (ii) ≤ . η(p − u∗) W 0(u∗)

Assumption 3, which will ensure the uniqueness of equilibrium and the monotonicity of equilibrium behavior u∗ with respect to symmetric constrained price p, is reasonable. dL Note first that dp > 0. So for Assumption 3(i) to be satisfied, it suffices that the elasticity of demand does not decrease too fast with p. For example, the elasticity of demand is proportional to p (and so is increasing in p) when consumers’ demand obeys a discrete choice model, in which idiosyncratic taste shocks obey a symmetric distribution.17 An- other illustration relates to the free-entry outcome: intuitively, a higher price leads to more entry, which exacerbates competition.18 As for part (ii), it is satisfied for example

15See Choi et al (2019), Acemoglu et al (2020) and Bergemann et al (2020). 16Web sites are legally entitled to decide what they think is right. 17 A consumer h selects to buy from supplier i if ui + εih − p > maxj6=i{uj + εjh − p}, where ε·h are idiosyncratic taste shocks obeying a symmetric distribution. Then, in a symmetric equilibrium, ∗ Di = Pr(ui − u ≥ maxj6=i{εjh − εih}) is independent of p and [−∂Di/∂pi]/[Di/p] is proportional to p. 18 p(p−c) ¯ This is indeed the case for the Lerner-Salop model, for which both η = tV¯ (where V is the p−c reservation utility or ﬁxed entry cost) and L = p are increasing in p.

13 for CES and Hotelling-Lerner-Salop demand functions, as well as in the discrete choice model (for which ∂η/∂u∗ = 0). Our assumptions guarantee that there is a unique symmetric equilibrium; if they are relaxed, there may be multiple symmetric equilibria (see Supplementary Appendix A). We saw in Section 2.2 that the ﬁrst-order condition19 in a symmetric equilibrium writes: −αW 0(u∗) = η(p − u∗)L(p) (4)

p−c ∗ where, recall, L(p) ≡ p is the ordinary Lerner index, and η(p−u ) denotes the elasticity of individual demands when net prices are all equal to p − u∗. We will assume that the second-order condition is satisﬁed; a suﬃcient condition for this is that the demand be linear or concave.20

Proposition 2 (impact of competition on ethics under constrained prices and a covered market). There exists a unique21 symmetric equilibrium under constrained prices. The symmetric-equilibrium level of ethics is given by

−αW 0(u∗) = η(p − u∗)L(p).

(i) Suppose that an exogenous parameter (e.g. a substitutability parameter22 or the number of ﬁrms) moves the elasticity η. Then the equilibrium level of ethics is decreasing (u∗ is increasing) in the intensity of competition (η), from the socially desirable level (u∗ =u ˆ) when the clients are captive (η = 0) to the ethic-free outcome (u∗ =u ¯) under perfect competition (η = ∞).

(ii) The equilibrium level of ethics is decreasing (u∗ is increasing) in the constrained price (p), from the socially desirable level uˆ for p = c to the ethics-free outcome u¯ as p tends to ∞.23

(iii) It is increasing (u∗ is decreasing) with α ∈ [0, +∞) from the ethic-free outcome to the socially desirable level. 19The solution is interior since W 0(¯u) = −∞ and the elasticity is bounded above. 20 ∗ ∗ The second-order condition writes, using W (ui) = w at ui = u :

2 00 0 ∂Di ∂ Di α W Di − 2W + [p − c] 2 < 0. ∂pi ∂pi

21Assumption 1 implies that the continuous function −αW 0(u∗) − η(p − u∗)L(p) is negative atu ˆ and ∗ 00 ∗ 0 ∗ h ∂η/∂u W (u ) i ∗ positive atu ¯. Its derivative when it takes value 0 is αW (u ) η − W 0(u∗) . So if [∂η/∂u ]/η < W 00/W 0, W 0(u∗) < 0 implies that the function equals 0 at exactly one value of u∗. 22 1 Σj6=ipˆj Suppose for instance a linear demand system: Di = n − σ[ˆpi − n−1 ]. Then, at a symmetric ∗ p−c equilibrium, η(p − u )L(p) = [σnp][ p ] = σn(p − c). So the substitutability parameter σ and the number of ﬁrms n are alternative measures of how competitive the industry is. To have the whole range from monopoly to perfect competition, one can for instance use the Hotelling model with co-located outside options (as in Supplementary Appendix C.1). 23Of course, for very large p, the assumption that the market is covered becomes much less plausible. More on this shortly.

14 3.2 Some simple extensions for constrained prices

Free entry. Suppose that n is endogenous and determined by a free entry condition p−c ¯ n = V − e, where e ≷ 0 is a parameter of entry facilitation (e.g. an entry subsidy) and V¯ − e the entry cost or reservation utility of suppliers (see Supplementary Appendix C.2 for a more careful discussion of the implicit assumptions behind free entry analysis). Under constrained prices, a facilitation of entry generates an increase in the number of competitors and makes the market less moral.

Robin Hood strategy. Striking, both in the gilded age and today’s superstar-ﬁrms environment, are the very generous charitable donations made by entrepreneurs who otherwise have opposed any regulation of their monopoly position, or possibly have exploited a rather unethical line of business. In some respects, the model can shed light on this: If money allows an entrepreneur to do philanthropy beneﬁting her favorite cause (universities, museums, poor people. . . ), then her valuation for money increases; this is formally equivalent to a decrease in α.24

Patient imperfect information Serious controversy arose regarding the aggressive advertising of OxyContin by Purdue. More generally, direct advertising to consumers is viewed as putting pressure on doctors to prescribe drugs. Suppose that a fraction of patients are not aware of the availability of the drug (and consequently do not know the doctor’s reputation in the matter). Those patients go to the doctor nonetheless to be treated; under regulated prices, each doctor receives market share 1/n in this subgroup of patients. Other patients are as described above. If the doctors can discriminate between the informed (who do insist on being prescribed the drug) and the uninformed (who do not), then u∗ =u ˆ for the uninformed and −αW 0(u∗) = η(p − u∗)L(p) for the informed. If z is the fraction of informed clients, the lack of consumer information generates a social gain equal to (1 − z)[W (ˆu) − W (u∗)]. In this model direct-to-consumer advertising should be banned as it augments the pressure on suppliers to please the client.25

24We here assume that the philanthropic approach is planned, and not just an afterthought. Besides, our model sheds light only on the positive side. When it comes to normative issues (should the government subsidize philanthropy?), much more thought is required. Conventional wisdom has it that the government should correct market failures and have a monopoly on “philanthropy”. Are subsidies for philanthropy motivated by a state failure (for example, the political process in a majoritarian regime might not deliver top universities or great philharmonics or enough research that delivers services for the citizens in the long run, or may not care enough for the poor, who do not vote much)? Is this meant to empower citizens, who may have a demand for giving (but in this case analyzing this demand is important, as part of this demand is associated with a desire for signaling, largely a zero-sum game)? 25Of course, the model abstracts from potential benefits of direct-to-consumer advertising, such as informing the patient that a treatment exists, inducing her to see a doctor (this effect is most likely for recently approved drugs). Another potential benefit of direct-to-consumer advertising is that a prudent pharmaceutical company might warn patients against the possibility that their doctor does mischief.

15 This corollary of our analysis would be worth of empirical investigation. Casual empiri- cism,26 based on a few legislations (Table 1), indeed suggests that consumer advertising is much more tightly regulated for prescription drugs, for which overconsumption can be very costly, with some exceptions in the case of underconsumption (vaccines, tobacco addiction reduction). By contrast, regulations on advertising are much more lenient for non-prescription drugs. This table suggests a link between consumer advertising and the presence of internalities/externalities. Empirical work could further tease out the role of the nature of competition. Besides making some consumers aware of the existence of the drug, advertising also tends to make individual demand more elastic.

Canada Australia China France Germany Sweden USA Yes, subject to Yes, subject Yes if no risk for Advertising of non-prescription medicine FDA regulation on to prior Yes Yes Yes for adults Yes public health content approval

No except for Yes but limited to vaccines and No except for Advertising of prescription-only medicines name price and No No No Yes tobacco addiction vaccines quantity reduction

Yes, can Yes but no Yes if no Yes if no Yes but not Yes, can discuss mention Disease awareness campaign reference to reference to a reference to focusing on Yes treatment options treatment medicine medicine a medicine treatments options Table 1: drug advertising [Built using data from https://iclg.com/compare/pharmaceutical-advertising]

Outside option So far, clients had no choice but picking one of the suppliers: Outside options were never attractive. In contrast, let us introduce an outside option yielding utility u0 and welfare w0. This outside option might stand for not consulting a doctor, ﬁnding illegal drugs to replace OxyContin, etc. When performing comparative statics below, we will posit that w0 covaries negatively with u0 (although not necessarily according to the function W , as the outside option might have a diﬀerent nature). Supplementary Appendix C.1 adds an outside option (in a symmetric oligopoly context) and generalizes condition (4). The equilibrium level of ethics now depends not only on the “within elasticity” (the term ηL in (4)), but also the “across elasticity” of the total demand curve.27 Suppliers’ ethics goes down when the outside option becomes more ethical.

26A proper analysis would take into account the drug’s price determination (which varies substantially from country to country) and the ﬁnancial incentives for drug sales representatives. ∗ 27 0 ∗ p−c+α(w −w0) The formula is −αW (u ) = ηL + εK where ε is the across elasticity and K = p is again ∗ a generalized Lerner index (K > L if and only if w > w0).

16 Setting the regulated fee The constrained fee, when it results from a public regulation, may reflect a variety of political-economy considerations. Supplementary Appendix C.2 studies the purely economic tradeoffs involved in price setting. In the absence of viability constraints, setting the fee equal to marginal cost (p = c) would kill two birds with one stone: The clients would pay a low price and the suppliers, having no stake would pick ui =u ˆ. Supplemen- tary Appendix C.2 looks at the socially optimal price under free entry. Suppliers face a fixed entry cost and as earlier internalize a fraction α of welfare, which now includes the consumers’ benefit from product diversity (which was constant when the number of suppliers was fixed) on top of the ethical level of consumption. The key result is that the suppliers’ social preferences lead to a lower regulated fee. The intuition is that the social planner can harness social preferences by reducing the suppliers’ financial stake at the expense of a smaller number of suppliers, implying a lower product diversity.

3.3 Market-determined prices: the invariance result

Suppose now that suppliers choose their prices as well as the gross surplus they offer to their clients. As shown in Section 2.2, ethical behavior is then given by −αW 0(u∗) = 1. This condition states that the marginal cost of being unethical (−αW 0(u∗)) is equal to the marginal benefit: Indeed, an increase in ui can be offset by an equal increase in pi, leaving 28 demand Di constant. Supplementary Appendix B shows that the global second-order condition is satisfied.

Proposition 3 (invariance of ethical behavior under market-determined prices). The level of ethics under market-determined prices is independent of the extent of competition and is given by u∗ = u†, where −αW 0(u†) = 1. (5)

Price caps and price ﬂoors

Regulated prices may be price caps (pi ≤ p) rather than rigid price instructions. Such caps are binding if and only if ηL < 1. Conversely, the regulated price may take the form of a minimum price (pi ≥ p); such a ﬂoor is binding whenever ηL > 1. Combining Propositions 2 and 3 yields:

Corollary 1 (ethics in regulated and unregulated markets). When compared with unfet- tered market competition, ethics is always weakly higher (lower) when prices are capped (subject to a ﬂoor).

28Another way to think about (5) is through its comparison with Spence (1975). Supplier i can be viewed as oﬀering a product of quality ui (note that marginal and inframarginal consumers have the same ∗ valuation for quality) at cost c − α[W (ui) − w ]. When prices are endogenous, the supplier equates the 0 marginal cost of quality, −αW (ui), and the marginal beneﬁt, 1.

17 Price floors may be set by a cartel or a government protecting an industry through regulation or duties. And, as we earlier noted, another case in which the constraint on pricing takes the form of a price floor arises in two-sided markets; one side of the market (usually the consumers) would optimally face a negative price but arbitrage leads the platform or the app to make it available for free to the consumers (pi = 0). The cost c is then negative (this is the advertising profit made on the other side, or else the value of collected data). In that case, the price floor, by limiting price competition, raises the stake, and the theoretical prediction is that it leads to a less ethical behavior than would prevail if arbitrage could be prevented.29

3.4 Cost-side beneﬁts of unethical behavior

While the focus of the paper has so far been on the demand-side incentives for unethical behavior, the beneﬁts for the supplier may also arise on the cost side, as in the case of the use of polluting technologies or of child labor to save on wages. So we look at externality situations in which the delivery of the same products can be achieved through a costlier production process that is better for the workers or the environment. Should we expect the analysis to be the same for cost and demand beneﬁts? Interestingly, this is not the case.

Using the same notation, the demand now depends only on the prices (ˆpi = pi), but 30 the production cost is c − ui where ui ∈ [û, u¯]. The corresponding welfare is W (ui), which satisfies Assumption 1. Supplier i’s objective function is now, in a symmetric configuration:

[pi − (c − ui)]Di(ˆp) + α[W (ui)Di(ˆp) + Σj6=iW (uj)Dj(ˆp)].

0 † Maximizing with respect to ui yields −αW (ui) = 1, and so letting −αW (u ) ≡ 1,

† ui = u for all i.

Ethical behavior is invariant to competitive pressure regardless of whether prices are constrained or market determined. Because the ethical choice does not aﬀect demand, it is not impacted by the degree of competition and does not exhibit the pattern associated with the replacement excuse. The diﬀerence between the two environments in the case of constrained prices could be the object of empirical testing.

29A one-sided market example with a price floor occurs when introductory pricing leads to free availability. 30 The invariance result derived below extends to a general cost function C(Di, ui) with C increasing in its first argument and decreasing in its second one. Indeed, the first-order condition with respect to 0 ∂C ∗ ui yields −αW (ui)Di = − (Di, ui). In a symmetric equilibrium Di = 1/n and so u is independent ∂ui of the demand function (or of the price if the latter is regulated).

18 3.5 Stakeholder altruism

So far, we have focused on corporate altruism and considered four cases: (a) constrained vs. market prices and (b) demand-side vs. cost-side beneﬁts from cutting ethical corners. We now add social preferences for the three most prominent stakeholders, and show that implications are diﬀerent for consumers on one side and for investors and workers on the other.

Investor and worker altruism We view investor social preferences as aﬀecting the cost of capital. For instance, suppose that production requires an amount of equipment that is proportional to the quantity ∗ produced. We can then formalize the marginal costs of production as c − a[W (ui) − w ] (in symmetric equilibrium): investors are willing to obtain a lower return when their money funds an ethical production. Workers’ social preferences can be accounted for in a similar manner: to the extent that they are willing to receive a lower wage to operate in a more ethical environment, the impact of their social preferences is on the cost side; and so the parameter a of investor social preferences should more broadly be thought of as the sum of the parameters for investors and workers. This implies that investor and worker altruisms work exactly like corporate altruism: equations (4), for constrained prices and demand beneﬁts, and (5), for the three other cases, still apply, with α simply being replaced by a (or appearing additively with a in the presence of both types of altruism).

Consumer altruism While introducing investor or worker altruism does not change anything to the results, what can we say about consumer altruism? Here, we should focus on cost-side benefits from cutting ethical corners which now reduce demand rather than increase it; for, con- ceptually consumer demand may increase or decrease with unethical policies, but not both simultaneously. And indeed, one observes both cases: one in which consumers directly benefit (or feel they benefit) from socially unpalatable supplier choices, due to externalities, internalities or shrouded attributes; and the other, in which consumers have no personal stake in corporate behavior along the global value chain but want to improve workers’ rights and income and to reduce the carbon footprint.31 Put differently, the two cases correspond to two different corporate decisions.

32 Each supplier i can reduce their marginal cost, c − ui, by exerting an externality (increasing ui). Consumers however internalize a fraction a of the impact of their purchase decision on welfare. Letp ˆi ≡ pi − aW (ui). Supplier i faces demand function Di(ˆp) in a

31Such consumer social responsibility requires information and therefore certiﬁcation of sustainability standards (fair trade, rainforest alliance, ethical trading initiative). 32In case of an internality, the current incarnation cares about a distorted view of her own welfare. Under shrouded attributes, the client is imperfectly informed or unaware of a misrepresentation by the supplier.

19 symmetric equilibrium, max [pi − (c − ui)]Di(ˆp) yields: 1 Constrained fee: −aW 0(u ) = (6) i ηL 0 Market fee: −aW (ui) = 1 (7)

Under consumer altruism, competition allows the stakeholders to better express their social preferences. Like under corporate altruism, competition forces the suppliers to pander to the stakeholders’ demand, but the diﬀerence is that the consumers here want to promote the common good. Proposition 2 is therefore turned on its head. On the other hand, the invariance result still holds under market prices.

3.6 Summary: the general result

The following proposition (whose proof in Main Appendix A veriﬁes that the insights are not changed when the various ingredients are assembled) states the general result:

Proposition 4 (main result) Let α, a and a denote the social-preferences parameters of suppliers, investors/workers and consumers respectively (where a = 0 in the case of demand-side beneﬁts)

(i) Under market prices, the invariance result holds whether cutting ethical corners raises demand or lowers cost. The ethical concerns of suppliers, consumers and investors/workers add up:

[α + a + a][−W 0(u∗)] = 1. (8)

(ii) Under constrained prices, the morality of the market depends on the nature of competition

• if cutting ethical corners increases demand, a more intense competition or a higher stake (an increase in ηL) results in a lower level of market morality (strictly so when suppliers and/or investors/workers have social preferences: α + a > 0) • if cutting ethical corners lowers cost, the invariance result holds, except when consumers prefer a more moral mode of production, implying a higher cost (a > 0), in which case a more intense competition makes the market more moral.

3.7 Income eﬀects

Returning to suppliers’ social responsibility, conditions (4) and (5), properly re-interpreted, have implications for Shleifer (2004)’s argument that ethical behavior is a normal good,

20 and therefore likely to improve as societies become richer. Suppose that a supplier’s value of money is an increasing and concave function Φ. Then Vi = Φ((pi −c)Di)+α[W (ui)Di + 33 Σj6=iW (uj)Dj]. The analysis is then the same as earlier, except that α is replaced in conditions (4) and (5) by α αˆ ≡ . 0 p−c Φ n The denominator ofα ˆ captures Shleifer’s impoverishment eﬀect: an increase in the number of competitors n exerts an upward shock on the marginal utility of money Φ0, increases the suppliers’ relative weight 1/αˆ on money and ceteris paribus makes them behave less ethically. Similarly, a decrease in the price p makes the suppliers behave unethically.

Constrained prices. An increase in the number of competitors n lowers the individual suppliers’ demand, impoverishes them, and reduces their effective altruismα ˆ. The impoverishment effect of an increase in n operates in the same direction as the facilitation of supplier switching. So overall an increase the number of competitors makes the market less moral. By contrast, an increase in the price p now has an ambiguous effect (and so the same holds, mutatis mutandis, for a decrease in p). The replacement effect makes suppliers less ethical; yet, suppliers are richer and because ethical behavior is a normal good, the increased stake also makes the suppliers more ethical. The overall impact on market morality is ambiguous. So it is not necessarily the case that, ceteris paribus, rich doctors should be expected to behave better than their poorer counterparts in other parts of the world.34

Market prices. Under market prices, market morality is given by −αWˆ 0(u∗) = 1. The invariance result no longer holds: an increase in the stake (p − c) and in the number of competitors (n) again have two opposite eﬀects: a higher stake makes the market more moral. An increase in competition –here an increase in n– makes it less moral. Note, though, that the equilibrium price depends on the number of competitors while individual market shares remain equal to 1/n. An increase in n lowers p and so the overall eﬀect is that an increase in competition (number of competitors, substitutability among them) lowers market morality.

Free entry. Under free entry, a facilitation of entry (an increase in the parameter e) makes the market less moral under both constrained and market prices. We summarize these insights in:

33 ∗ 0 ∂Di 0 Because in equilibrium W (ui) = w , the FOC with respect to ui is −Φ ·(pi −c) +αW (ui)Di = 0. ∂pi Under market prices, the optimal price pi solves Di + (pi − c)(∂Di/∂pi) = 0. 34There is another caveat. A society’s average other-regarding preference (α) may evolve to reflect changes in the benefit of prosocial behavior. Thus, while the effects analyzed in the text seem robust, they may not be the only relevant effects.

21 Proposition 5 (immiseration) An increase in the number of competitors n or a decrease in the constrained price p ceteris paribus induce suppliers to put less weight on ethics under decreasing marginal utility (the impoverishment eﬀect).

(i) Under constrained prices, the impoverishment and replacement eﬀects both imply that an increase in the number of competitors makes the market less moral. By contrast, they have opposite implications for a decrease in the price, making the impact of the latter on market morality ambiguous.

(ii) Under market prices, the invariance result no longer holds. An increase in the number of competitors (or an increase in the substitutability among them) makes the market less moral.

(iii) Under free entry, a facilitation of entry makes the market less moral for both constrained and market prices.

4 Supplier heterogeneity

We now allow suppliers to diﬀer in their ethical value, corporate form, or talent. We assume away income eﬀects and ignore stakeholders’ ethical concerns for expositional simplicity. To handle asymmetric environments, we make (in this section only)

1 h Σj6=ipˆj i Assumption 4 The demand system is (symmetrically) linear: Di = n − σ pˆi − n−1 .

As noted earlier, a linear demand system satisﬁes Assumption 3. An important property of this linear demand system is that ﬁrm i’s change of behavior impacts other suppliers symmetrically. For example, we later provide a strongly asymmetric demand system for which Proposition 6 does not hold.

Assumption 5 (financial viability) Suppliers must be financially viable: pi ≥ c for demand-side benefits of cutting ethical edges, and pi ≥ c − ui for cost-side benefits.

To understand the rationale for Assumption 5, consider an otherwise symmetrical duopoly situation in which one supplier is more ethical than her rival and prices are market-determined (a special case of the model of Section 4.1 below). Cutting ethical corners boosts demand, say. Supplier 1 is selfish (α1 = 0), and therefore selects u1 =u ¯; supplier 2 is a saint (α2 = +∞), and therefore selects u2 =u ˆ and is willing to set any price that will take market share from firm 1. In equilibrium, the ethical firm corners the market (D2 = 1) and sets p2 ≤ c − (¯u − uˆ) < c. In words, a very ethical supplier is willing to lose money when facing a much less ethical rival. Assumption 5 deserves a couple of further comments. First, ignoring the issue of access to capital, Assumption 5 is irrelevant when differentials in social preferences are “not

22 too large”; what this exactly means depends on the intensity of competition.35 Second, Assumption 5 is innocuous in the absence of investors who have strong social preferences and are willing to foot the bill for virtuous actions. To be certain, one can think of undertakings that are financed by such investors (NGOs, Gavi. . . ), but the thrust of the debate on market morality is on firms that must at the very least break even (whether for-profits or not-for-profits). Third, this assumption will be relevant only in Section 4.1: In Section 4.2, not-for-profits will break even, and as we will show, for-profits will be profitable, even when ethical. The proofs in this section will first posit an interior solution, and then study if a “corner solution” instead prevails. Intuitively, there can be two kinds of corner solutions: A supplier’s demand may fall to 0; and a supplier’s financial viability may be constraining. At most one of these constraints will be binding, though. Under constrained prices, financial viability will not be a concern when benefits are on the demand side (the price necessarily satisfies p ≥ c), while all suppliers are guaranteed equal market shares when the benefit is on the cost side. Under market prices, the financial viability constraint will become binding when competition is intense, while demands will remain positive (an ethical supplier can always mimic a non-ethical one and just be a bit more ethical). Finally, we will say that there is a race to the ethical bottom if

lim ui =u ¯ for all i. σ→+∞

4.1 Does the market drive out more ethical types?

36 Rank suppliers by the intensity of their social preferences: α1 ≤ α2 ≤ · · · ≤ αn. We do not re-state the result that ethical behavior is independent of the nature of competition if either prices are market determined or cutting ethical corners lowers production cost, although it applies to asymmetric environments as well. The following proposition instead focuses on relative market shares:

Proposition 6 (ethics heterogeneity and market shares) Consider an oligopoly and a linear demand system. Whether cutting ethical corners increases demand or lowers cost:

(i) Under constrained prices (p > c), less ethical ﬁrms command a higher market share:

α1 ≤ α2 ≤ · · · ≤ αn ⇒ D1 ≥ D2 ≥ · · · ≥ Dn.

with equality (by construction) under cost-side beneﬁts.

35 For example, in Section 4.1. applied to a duopoly (α2 > α1), a sufficient condition for the financial 0 constraint not to bind is 1 ≥ 2σα2(w2 − w1), where αi[−W (ui)] = 1 and wi ≡ W (ui). 36We assume that these social preferences are common knowledge. Otherwise, assuming that suppliers are reputation-conscious, the objective function below has to be augmented with an image term as in, e.g., Bénabou-Tirole (2006).

23 The equilibrium exhibits the race-to-the-ethical-bottom property when cutting ethical corners increases demand (ethics is invariant to competitive pressure when cutting ethical corners reduces the production cost).

(ii) Suppose that there are n1 low-ethics suppliers (with social preferences α1) and n2 high-ethics suppliers (with social preferences α2 > α1). Under market-determined prices,

∗ • Fixing n, there exists n1 such that a high-ethics firm commands a higher market share than a low-ethics one if and only if the number of low-ethics firms, n1, ∗ is large enough: n1 > n1. In particular, in the duopoly case, the ethical firm commands a higher market share despite its more ethical choice. • There exists a substitutability threshold σ∗ such that high-ethics suppliers (with

social preferences α2) behave like not-for-profit suppliers (that is, p2 = c for demand-side benefits, and p2 = c − u2 for cost-side benefits), if and only if σ ≥ σ∗; the financial viability constraint is never binding for low-ethics suppliers. • The behavior of the two groups of suppliers converge as σ → ∞. Prices converge to marginal cost, but there is no race to the ethical bottom.

Part (i) or Proposition 6 (proved in Main Appendix B) is intuitive: More ethical firms are less attractive to consumers (demand-side benefit) or face higher production costs (cost-side benefit). So they have a handicap in the marketplace (although they secure the same market share under cost-side benefits). Note, though, that the symmetric-impact property of the linear demand system is important to guarantee the monotonicity of ethical choices in ethical values. It ensures that when wooing the consumers a supplier steals consumers away from other suppliers in a pattern that is not too asymmetric among rivals. To see why such a condition is needed, consider the following strongly-asymmetric- impact example: There are 4 suppliers with α1 < α2 < α3 < α4, and two islands/separate markets, one with suppliers 1 and 2 and the other with suppliers 3 and 4. The islands are identical except for the ethical values of their suppliers. Then supplier 2 behaves more 37 ethically than supplier 3 whenever the difference α3 − α2 is small. Intuitively, supplier 2 (supplier 3) gains market share at the expense of an unethical (ethical) supplier.38

37Letting σ denote the substitutability parameter in each of the two duopoly markets, the ﬁrst-order conditions are 0 −α2W (u2)D2 = σ[p − c + α2(w2 − w1)] and 0 −α3W (u3)D3 = σ[p − c − α3(w4 − w3)] 1 with w2 > w1, w4 > w3, and D2 < 2 < D3. So for α2 = α3, u2 < u3; and u2 < u3 still holds for α3 − α2 positive but small. 38A broader model that accommodates both Assumption 4 and this island model goes as follows: 1 D = − σ(ˆp − Σ σ pˆ ) i n i j6=i ij j where Σj6=iσij = 1 and σij = σji (which guarantees that ΣiDi = 1). Then σij = 1/(n − 1) in the symmetric-impact model. And σ12 = 1 while σ13 = σ14 = 0, etc. in the island model.

24 The last part in (i) -intense competition results in a race to the ethical bottom when cutting ethical corners increases demand- is intuitive. When ethical choices affect cost instead, firms have no mean of influencing demand and are guaranteed equal market shares; they are thus unaffected by their rivals’ ethical choices and freely express their idiosyncratic ethical preferences. Proposition 6(ii) turns Proposition 6(i) on its head. The reason is interesting: Under constrained prices, a less ethical supplier panders more to the consumer than a more ethical one; the former accordingly commands a larger market share. Under market- determined prices, a more ethical supplier must lower price to offset her “quality” dis- advantage. She also may lower price to gain market share in particular at the expense of less ethical suppliers; the ethical impact of such undercutting hinges on the “market’s morality”. With few unethical suppliers, the ethical gain is low and an ethical firm still commands a lower market share than an unethical one. By contrast, in a low-morality market, the ethical firm has a big impact when undercutting and ends up commanding a higher market share. The second bullet in part (ii) of Proposition 6 indicates a switch of regime when the offers are highly substitutable. High-ethics suppliers are then constrained by financial viability. The analysis is then that of Section 4.2 below (which characterizes competition between for-profits and not-for-profits for any degree of substitutability). In particular, the results of Section 4.2, applied to different ethical concerns, will imply convergence of behavior as the goods become highly substitutable. Indeed, the third bullet in part (ii) shows that intense competition leaves little scope for the expression of preference diversity. The one exception is the case of constrained prices and cost-side benefits, for which suppliers have no mean of influencing demand and are guaranteed equal market shares; they are thus unaffected by their rivals’ ethical choices and have a free hand at expressing their idiosyncratic ethical preferences. Discussion. We took the set of suppliers as fixed and looked as whether less-ethical suppliers drive more-ethical ones out of the market. A related, but different question would look at the incentive to join the industry. We analyze free entry equilibria in the symmetric-oligopoly case in Supplementary Appendix C.2, but not in the more interesting, asymmetric one, where composition effects arise. In our view, this important question deserves a separate paper for two reasons. First, when agents’ preferences are heterogenous, their reservation utilities also differ and are determined in general equilibrium (see the large literature on occupational choice, starting with Kihlstrom-Laffont 1979 and Lu- cas 1978). Second, what represents “the morally right thing to do” is not obvious, as philosopher Peter Singer (2015) points out. It may well be that entering an immoral industry in which one can make a difference is more moral than entering an ethical one.39

As shown in Section 2.2, condition (B.1) in the proof of Proposition 6 obtains more broadly, withw ¯−i ≡ Σj6=iσijwj. 39Moisson (2020) shows that the moral pecking order is highly context speciﬁc; a known example of this general point concerns socially responsible investment, for which best-in-class strategy may have a bigger impact than the exclusion of sin stocks. See also Green-Roth (2020).

25 Entry decisions then depend not only on the number of entrants in each occupation, but also on their composition.

4.2 Competition between for- and not-for-proﬁts

The focus thus far has been on for-profit suppliers with social preferences. It is often argued that industries that are highly exposed to ethical choices, such as health and education, are particularly suited to the non-profit paradigm.40 Is this so? Should we expect not-for-profit hospitals or schools to behave differently when in competition with for-profit entities? In this section, we assume market-determined prices so as to make the distinction between a for-profit and a not-for-profit relevant. To identify the impact of the corporate form on ethical behavior, we could assume that non-profit suppliers have the same social preference parameter as for-profit ones, but face different incentives as they are meant to break even. We actually will not need this assumption. Hence we assume that for- profits (type-1 firms) are managed by suppliers with type α1 while not-for-profits (type-2 firms) are run by managers with type α2 > 0 and we do not assume a specific ranking ∗ between α1 and α2. In Section 4.1, we saw that for intense competition (σ ≥ σ ), the most ethical suppliers behave like not-for-profits; in this case, α2 > α1. Alternatively, we could assume that firms with different corporate status attract employees with different social preferences.41 A natural assumption on the empirical side is assortative matching

(α2 > α1: not-for-profits attract more ethical employees), but as we already noted, it is not clear that working for a not-for-profit is necessarily the moral thing to do for someone who wants to have a strong ethical impact. The following result, though, reflects the strength of the zero-profit condition: not-for-profits by definition have no material stake and so, whatever their α2, are solely concerned by the maximization of social welfare. They are however increasingly worried about losing market share to for-profits as competition intensifies, and so their ethical behavior converges in the limit to that of for-profits.42 And so our results hold regardless of the actual sorting process.

Proposition 7 (corporate form). Assume a linear demand system, n1 for-profits with social preferences α1 and n2 = n − n1 not-for-profits with social preferences α2 > 0, and market-determined prices for for-profits. Whether cutting ethical corners increases demand or lowers cost:

(i) Not-for-proﬁts behave more ethically than for-proﬁts.

40Indeed, many health and school systems around the world are structured around not-for-profit entities, when not by state-owned providers. 41See e.g. Besley-Gathak (2005), Prendergast (2007), Brekke-Nyborg (2008), Kosfeld-von Siemens (2011), Lazear et al (2012), Barigozzi-Buranib (2019); for field experiments on sorting and prosociality, see Ashraf et al (2020) and the references therein. 42Of course, there may be no such thing as a pure not-for-profit. Insiders may manage to convert profits into private benefits; private benefits are an inefficient currency, but more to the point, such conversion of profits would reinstate a role for the not-for-profit suppliers’ exact level of altruism.

26 (ii) Two-sided mimetism: When competition (as indexed by σ) is intense, the behaviors of the two corporate forms converge: The for-profits mimick the not-for-profits’ low price, while the latter behave no more ethically than for-profits. Competition is intense along both dimensions, promoting welfare along the price dimension, but not so along the non-price one. There is no race to the ethical bottom unless for-

proﬁts are purely selﬁsh (α1 = 0).

Under intense competition for consumers, suppliers end up charging similar net prices.

The price of not-for-profits is fixed at marginal cost (p2 = c under demand-based benefits and p2 = c−u2 under cost-based benefits), while the gross consumer surplus offered by for- † 0 † profits equals u1 = u1 given by −α1W (u1) = 1. The for-profits must lower their markup toward 0 to not lose all demand, while the not-for-profits must pander at (approximately) † level u1 for the same reason. Competition homogenizes behavior across corporate forms. † Convergence happens toward the “anchors”: (p = c, u = u1) under demand-based benefits † † and (p = c − u1, u = u1) under cost-based benefits. This proposition (proved in Main Appendix C) is consistent with evidence about the hospital sector. As argued in classic work by Weisbrod (1988) and Hansmann (1996), not-for-profits have historically been an important commitment device against excesses associated with the profit motive (see also Besley and Ghatak 2005). In recent decades, though, for-profit hospitals have made inroads in the sector, and, unsurprisingly, have been shown to put more emphasis on profit-related managerial compensation (Ballou- Weisbrod 2003) in comparison with their not-for-profit peers, consistently with part (i) of Proposition 8. In support of part (ii), Arnould et al. (2005) show that more competition from for-profit hospitals leads to a higher importance of the “profit motive” (i.e. net financial income) among not-for-profit ones, both in terms of the structure of managerial compensation and of managerial turnover decisions (and this is understood by donors, who reduce their contribution as a result of this weakening of the not-for-profit mission).

4.3 Intrinsic quality diﬀerences

Are talented doctors inclined to behave more ethically than bad ones? Should they be expected to command a larger market share? To study these questions, we let si denote supplier i’s (known) talent, expressed in monetary equivalent for the client, with s1 ≤ s2 ≤ · · · ≤ sn. Letp ˆi ≡ pi − ui − si the net price set by the supplier. Supplier i’s demand is then Di(ˆp). The following proposition is proved in Supplementary Appendix D:

Proposition 8 (inherent quality diﬀerentials) Suppose a linear demand system, and that suppliers diﬀer in their quality s1 ≤ s2 ≤ · · · ≤ sn.

(i) Under constrained prices, higher-quality suppliers are more ethical (u1 ≥ u2 ≥ · · · ≥ un) and command a higher market share (D1 ≤ D2 ≤ · · · ≤ Dn) when cutting

27 † ethical corners increases demand. They are equally ethical (u1 = ··· = un = u ≡ (W 0)−1(−1/α)) and command a higher market share when beneﬁts are on the cost side.

† (ii) Under market prices, all suppliers are equally ethical (u1 = ··· = un = u ) and higher-intrinsic-quality suppliers charge higher prices and command a higher market share, whether cutting ethical corners increases demand or lowers cost.

5 Alternative moral preferences: A broader perspec- tive on the replacement excuse

In accordance with much theoretical and experimental literature, this paper presumed that suppliers and stakeholders have social preferences. But, as pointed out in the introduction, there is a variety of alternative ethical criteria. Because social preferences are consequentialist, we start this discussion with some common empirical criticisms of consequentialism and show how they are amenable to be taken on board in this framework.

Context-dependent ethics. Experiments show that moral behavior is driven not only by intrinsic and extrinsic incentives, but also by self- and social-image concerns. Because social preferences have been assumed to be common knowledge, there was no loss of generality in ruling out such image concerns. One could more generally allow for privately- known social preferences and add a term to supplier i’s objective function, µE[αi | ui],that would reﬂect both an intensity of image concerns, µ, and inferences about the individual’s empathy, αi, as a function of his action ui. Image concerns would lead to more ethical choices. While the introduction of image concerns would not aﬀect the main insights of this paper, it would open the door to social norms based on “what society deems acceptable”.43

Reverse causality. The paper has taken the stance that ethical preferences are given (up to the above remark on context-depending ethics, which in a sense maintains this approach) and investigated the consequences of competition under this assumption: The exchange institution per se does not make suppliers selﬁsh or unselﬁsh. But one could make the alternative assumption that competition alters moral criteria, that is that α decreases when competition increases. This approach assumes that competition is intrinsically bad for ethics, which the exogenous-preferences approach does not (and indeed shows that this conclusion depends on the environment). One way to partially reconcile the two views is to introduce excuses. Often individuals are pivotal in decisions with moral overtones, but manage to distort reality so as not to feel responsible for otherwise-unethical behavior (as if they were not pivotal). This vio-

43As in, say, B´enabou-Tirole (2006, 2013).

28 lation of consequentialism44 is an apparent violation, rather than a conceptual one: It is precisely because individuals assess moral behavior through the lens of the consequences of their actions that they distort reality. We know from the large empirical literature on “moral wiggle room” that subjects seize on the slightest excuse to not behave morally.45 The presence of competition and therefore of substitution opportunities for the client represents such an excuse. Combining this replacement excuse with the norm-based approach would add an additional eﬀect to our analysis: The suppliers could prevail themselves of the replacement excuse to avoid looking too selﬁsh.46 Returning to the symmetric oligopoly case in which cutting ethical corners increases demand, we investigate the impact of some common alternative formulations of “ethics” –narrow mandate, deontologism, categorical imperative– on our results so as to examine their robustness and derive a few testable implications that might help distinguish ethical concepts.

Narrow mandate. Suppose that supplier i accounts only for the welfare associated with 47 0 ∗ p−c+αW (u∗) her own activity: Vi = [pi − c + αW (ui)]Di. Then −αW (u ) = p η under constrained prices (so u∗ increases with p and with the elasticity of demand), and −αW 0(u∗) = 1 for market-determined ones. The conclusions are therefore qualitatively the same as with a broad mandate. Similarly, the results under heterogeneity can be generalized. Strategic complementarity, which is at the core of the replacement excuse, is satisﬁed by narrow-mandate social preferences. It is precisely this strategic complementarity that is absent in the next two moral criteria. Both the narrow and the broad mandate reﬂect consequentialist preferences48 in that supplier i cares about the consequences, and therefore the scale (Di) of their act. The violation of this property eliminates strategic complementarity, as shown by the deontological preferences and the categorical imperative below.

Deontologism. In Kant’s deontologism, the morality of an action is based on whether the action is in itself right or wrong, rather than on the consequences of this action. Suppose therefore that supplier i values the act per se rather than its consequences. For instance,

44Another apparent violation relates to the role of intentions: suppose that I meant to hurt you, but by clumsiness failed to do so. There is then a signaling aspect to my choice, that hinges on the consequences that would have prevailed, had I been more skillful. 45See e.g., Bartling-Fischbacher (2012), Dana et al (2006), Dana et al (2007), Exley (2016), Hamman et al (2010) for early contributions to the field, and Bartling et al (2020), Bursztyn et al (2020), Foerster-van der Weele (2018a, b), and Saccardo-Serra Garcia (2020) for recent entries. 46They would form motivated beliefs that overestimate the elasticity of demand they face when they behave more ethically. 47Note a conceptual difficulty with a narrow mandate: Behavior depends on the normalization of welfare; for example, for a given elasticity of demand and a constrained fee, a fixed translation of W alters behavior. 48Act consequentialism may be interpreted as either, if one follows the definition of the Stanford Encyclopedia of Philosophy: “Act consequentialism is the claim that an act is morally right if and only if that act maximizes the good, that is, if and only if the total amount of good for all minus the total amount of bad for all is greater than this net amount for any incompatible act available to the agent on that occasion.”

29 supplier i’s payoff could be Vi = (pi − c)Di + αW (ui), where W (ui) could be, as here, the welfare associated with choice ui, or more generally a decreasing function of ui on [û, u¯]. Such preferences are only partly deontological, as they still reflect a profit component. The equilibrium conditions for the constrained-prices and market-prices environments are −αW 0(u∗) = ηL/n in the former, and −αW 0(u∗) = 1/n in the latter. For linear demand, an increase in competition for a given n (an increase in σ) makes the market less moral under constrained prices and has no impact under market prices. An increase in the number of competitors n for a given σ has no impact on market morality under constrained prices (−αW 0(u∗) = Lpσ) and increases market morality under market prices, despite the fact that cutting ethical edges brings benefits on the demand side. Like for the categorical imperative below, implications are very different from those for consequentialist preferences, generating testable implications. In the linear-demand case, and allowing for ethical heterogeneity, the ethical behavior 0 of supplier i when prices are constrained is given by −αiW (ui) = σ(p − c) and is inde- 0 pendent of rivals’ behavior. When prices are market-determined, then −αiW (ui) = Di = (1/n) + σ(ui − u¯−i), whereu ¯−i ≡ (Σj6=iuj)/(n − 1), and so ethical behaviors are strategic substitutes. These two results show that deontologism is at odds with the replacement excuse.

Categorical imperative. Suppose that suppliers follow Kant’s categorical imperative. If the market is covered, then each supplier, behaving as if her choice was to be mimicked by other suppliers, selects the socially optimal u∗ =u ˆ. This holds regardless of the supplier’s exact preferences, as long as her utility increases with overall welfare. In contrast, when the market is not covered and an outside option generates welfare w0 < W (û), then a supplier following the categorical imperative selects u∗ > uˆ so as to reduce the outside option’s market share. Either way, suppliers behave fully ethically (as if α = +∞). There is by construction no scope for the replacement excuse. In conclusion, while it is impossible to cover all moral criteria that have been proposed, we hope that the above will convince the reader that the model is sufficiently tractable to accommodate a range of variations on the moral criterion. We saw that different criteria have different implications for the impact of the nature of competition. The broad and narrow mandates exhibit strategic complementarity and the associated replacement excuse, in contrast with deontologism and the categorical imperative.

6 Related literature

The replacement excuse has received support in experimental work (let alone anecdotal evidence). Falk et al (2020) show that (the perception of) pivotality is key to sustaining moral behavior. Organizations that aim at being ethical should accordingly attribute individual responsibility to their members. Bartling and Ozdemir¨ (2017) demonstrate that the replacement excuse is less prevalent when there is a strong social norm, though.

30 Indirect evidence for the replacement excuse is supplied by the experiments on moral wiggle room (see footnote 45), that show convincingly that subjects cling to the (real or perceived) possibility – however small – that they may not be pivotal to select immoral decisions when they would not have done so, were the pivotality indisputable. On the theory front, Dufwenberg et al. (2011) consider consumers with other-regarding preferences and analyze under which conditions Walrasian equilibria are the same as with selfish agents. If individual consumptions have no influence on the clearing prices or other consumers’ budgets, consumers choose their own consumption as if they were selfish, pro- vided there is separability between the felicity associated with their own consumption and other consumers’ consumption and budget sets. Sobel (2015) obtains a similar result in auction markets with full replacement or in large economies: An observer who only sees behavior cannot distinguish between selfish agents and agents with other-regarding preferences. We depart from perfect competition and we allow for (i) non-price competition and (ii) a price-dependent ethical behavior. The latter invalidates the causality from competition to ethics. The observational equivalence of economies populated with selfish and other-regarding agents no longer obtains, which allows us to study the implications of supplier and stakeholder socially responsible behavior. The paper also has a strong connection with the corporate social responsibility (CSR) literature.49 A prominent view of CSR equates it with “delegated philanthropy”. The firm is a channel for the expression of citizen values; put differently, stakeholders have a demand for corporations to engage in philanthropy on their behalf: consumers will be willing to pay a bit more for their coffee if Starbucks embraces the fair-trade approach. Environmentally conscious investors will accept getting a smaller return from green funds. Workers will take a wage cut when employed by an NGO. As in the first approach, the consumer-product company or the investment institution maximize profit, as they pass through the higher cost or the lower return to the stakeholders. This view is embraced in Aghion et al. (2019), Besley-Ghatak (2007), Besley-Persson (2019), Green- Roth (2020), Landier-Lovo (2020), Moisson (2020) and Oehmke-Opp (2020). Aghion et al. formalizes CSR behaviour as a quality parameter, shows that under some conditions competition induces greener behaviour and tests this hypothesis. Besley-Persson adds a political determination of the tax/subsidy system and allows values to adjust over time to the (endogenous) technological evolution. Moisson puts particular emphasis on image concerns and consequentialism; in the baseline model, investors care about impact investment. The paper studies the evolution of the green premium paid by socially responsible investors as environmental consciousness grows. An alternative view of CSR is “insider-initiated corporate philanthropy”, namely philanthropy that clashes with the profit-maximization hypothesis.50 This is the approach taken in Hart-Zingales (2017). In that paper, shareholders also compare their monetary gains with the ethical impact of their actions. This tradeoff has “bite” when they vote

49See, e.g., the taxonomy in Bénabou-Tirole (2010). 50Even leaving aside the agency literature, there is of course a long tradition of analyses of non-profit- maximization goals: Beckerian discrimination theory, labor-managed firms, etc.

31 at the general assembly or board of directors, since both impacts are non-zero only if their vote is pivotal. By contrast, this leads them to focus solely only on monetary gains when they buy shares (there is no socially responsible investment), since they rationally expect to be pivotal and therefore affect the company’s future actions only with a tiny probability. Broccardo et al (2020) extend the analysis in a model where they endogenize investor divestments and consumer boycotts (which they call “exit” mechanisms) where individual investors and consumers internalize their (nonzero) impact on firm behavior on aggregate social surplus. In their model, under social preference parameters consistent with experimental evidence, divestments and boycotts are insufficient and shareholder engagement through voting (“voice”) is socially preferable. Our paper belongs to these two literatures, as we allow both the supplier and the stakeholders to have social preferences and allow ethical choices to maximize corporate profits or to reduce them. Its unique focus is on the impact of the nature of competition on market morality and on the predictions of heterogeneity in preferences and corporate form for market shares. Our paper is also related to several strands of the industrial organization literature. The result that a more intense competition may deliver poor non-price outcomes may be reminiscent of the work on the value of the banking franchise. In the industrial organization realm, Shapiro (1983) showed that the quality of experience goods may fall when intense competition reduces the payoff to maintaining a reputation for quality. Our framework however links non-price choices to ethical concerns rather than reputation, and has distinct policy implications. More broadly, the IO literature on quality provision refers to “good quality”, namely to quality that is demanded by both consumers and the social planner; the constraint on quality is then its cost. While Spence (1975) has shown that firms with market power may provide too little or too much quality, quality is not both demanded by consumers and frowned upon by the social planner (“bad quality”), unlike in this paper. The paper also relates to the multi-task incentive literature. Holmström-Milgrom (1991) stress that high-powered incentives by a principal may compromise the agent’s provision of non-contractable quality.51 Relatedly, the paper also connects to the literature on not-for-profit firms. This literature emphasizes that the absence of profit motive reduces the incentive to cut on unobservable quality (Hansmann 1980, Glaeser-Shleifer 2001, Besley-Malcomson 2018)52. Our paper is complementary: It assumes by contrast that the “quality” assessed by consumers is observable (directly, or through word of mouth or reputation) but not socially desirable; and it looks at a different set of issues (e.g., the

51Where quality here is viewed from the principal’s standpoint. In Lazear (1989), two workers are engaged in a tournament. The relative performance determines individual pay raises, which is conducive to “sabotaging”. Itoh (1991) studies optimal incentives for team workers who have individual performance measures but help each other. 52For instance, Besley and Malcomson posit that not-for-profits internalize the benefits of various dimensions of quality, although maybe in a paternalistic fashion. Their focus is on the ease of entry by a non-profit facing a for-profit incumbent, and variations thereof, to match the observations on entry in the school and hospitals sectors.

32 convergence of behavior of for- and not-for profit firms as a function of the degree of competition). Finally, Becker (1957) made the opposite point that (perfect) market competition weeds out suppliers who have a preference for discrimination. There is no contradiction here. Becker considered situations in which the immoral behavior raises the cost of business. None of the firms have social preferences, and so there is no replacement excuse. By contrast, we are interested in environments where the immoral behavior raises demand (because of an externality, an internality or shrouded attributes) or lowers cost (use of child labor).53

7 Summary and some alleys for research

As critics of market economies have long emphasized, our institutional context frames our ethical choices. Competition- understood as an increase in the number of competitors or in their substitutability- strengthens incentives to please the consumer, raising potential concerns. This paper considered market environments in which attracting the customer or minimizing cost differs from pursuing social welfare. In the presence of externalities, internalities or shrouded attributes, suppliers may cut ethical corners in an attempt to woo customers. A mitigating force is social preferences, the partial internalization of welfare by corporate, investors and workers, whose altruisms add up and have identical impacts on outcomes (as they all amount to a reduction in marginal cost). The paper showed how market morality is affected by the intensity and instruments of competition, the type of material benefit obtained from cutting ethical corners, the social preferences of management and stakeholders, and income effects. 1) Critics of the market are correct in arguing that competition among suppliers to attract consumers may weaken their ethical concerns. The replacement effect– a refusal to please the consumer will make them look elsewhere- and, when suppliers have decreasing marginal utilities, the immiserating-competition effect- competition impoverishes suppliers and reduces ethics, which is a normal good- make the market less moral when competition becomes more intense. When prices are constrained, the replacement effect is more potent, the higher the number of competitors, the less differentiated they are, and the higher the regulated fee. The immiserating-competition effect is more potent the higher the number of competitors, the less differentiated they are, and the lower the regulated price.

53 Using the notation of this paper, one could frame Becker’s analysis as a competition among n2 ordinary profit-maximizing firms (they solve max(pi − c)Di(pi, p−i)) and n1 “discriminatory firms”: Letting ui ≥ 0 denote the amount of discrimination, the latter raise their marginal cost to c + ui and gets benefit 0 00 B(ui), with B (0) > 1 and B < 0, from doing so. They therefore solve max{pi,ui}[pi − (c + ui) + B(ui)]Di(pi, p−i) subject to the financial viability constraint pi ≥ c+ui. As long as the equilibrium price 0 −1 p1 satisfies p1 ≥ c + (B ) (1), competition has no effect on discrimination. When competition becomes 0 intense, discriminatory firms become de-facto not-for-profit firms, selecting B (u1)D1 = B(u1)σ for a linear demand system.

33 2) These detrimental effects of competition however hold for specific environments and disappear or are reversed in others: a) Competition, to the extent that it reduces prices, also lowers the suppliers’ benefit from attracting consumers in ethically dubious ways. Thus, high administered prices induce less ethical behavior than competitively determined ones, not more. b) When cutting ethical corners lowers cost instead of increasing demand, the intensity of competition plays no role under corporate, investors and workers altruism. c) When consumers care about the production process not involving child labor or pollution, pleasing customers involves being more ethical, not less; competition then allows consumers to better express their social preferences and makes the market more moral. The paper’s second contribution is an analysis of the impact of competition among suppliers that differ in their ethics, either intrinsically or because of their corporate mission: 3) When suppliers differ in their ethical concerns, constrained prices work to the advantage of low-ethics suppliers. By contrast, price competition leads to interesting compositional benefits: More ethical suppliers always charge lower prices and have higher market shares if and only if the number of unethical firms is high enough. Intense competition simultaneously leads more ethical suppliers to behave like not-for-profits and induces a convergence of behavior among suppliers with different social preferences. 4) Not-for-profits behave more ethically than for-profit suppliers. However, when competition between the two corporate forms becomes intense, the difference between the two, both in ethical behavior and in pricing, vanishes: For-profits charge low prices and not-for-profits cut corners and align their ethical choices with those of for-profits. Let us conclude with three issues of interest left open by our paper: Markets and organizations. Criticism of the market economy must envision its counterfac- tuals. In that respect, our study of ethics in markets only begins to analyse the general issue of the morality of markets, raised in the 18th century and very prominent in the political discourse since the financial crisis and the backlash against globalization. The market is only a specific system of incentives. Within organizations, different ways of incentivizing managers and workers lead to different ethics. And who runs the organizations matters. The interaction between ethical behavior and the design of incentives is a rich object of study for future research. Leading by example. Suppliers may have information on the gravity of unethical behavior, i.e. on the magnitude of the externality. By moving first, they may signal to other suppliers that unethical behavior is just unacceptable (for example by withdrawing from the market altogether); and even if the other suppliers are also informed about the size of the externality, society may not be, and the revelation that “there are other ways to run the business” may put stakeholder pressure on these other suppliers.54

54This is reminiscent of both Hermalin (1998)’s theory of leading by example and the informed principal approach in B´enabou-Tirole (2013).

34 Stakeholders’ ethical concerns. Section 3.5 showed how the “delegated philanthropy” view of CSR interacts with the “corporate philanthropy” view of CSR. Introducing heterogeneous ethical concerns on the suppliers’ and stakeholders’ sides would provide a richer picture. One would expect some “assortative matching” with more ethical suppliers en- joying a higher market share with ethical clients. The analysis of the impact of this partial specialization under heterogeneous clienteles on overall ethical behavior would be of interest.

35 References

Acemoglu, D., Makhdoumi, A., Malekian, A. and A. Ozdaglar (2020), “Too Much Data: Prices and Inefficiencies in Data Markets,” mimeo. Aghion, P., Bénabou, R., Martin, R., and A. Roulet (2019), “Environmental Preferences and Technological Choices: Is Market Competition Clean or Dirty?,” mimeo. Alpert, A., Powell, D. and R. Liccardo Pacul (2018), “Supply-Side Drug Policy in the Presence of Substitutes: Evidence from the Introduction of Abuse-Deterrent Opioids,” American Economic Journal: Economic Policy, 10(4): 1–35. Anderson, E. (1993), Value in Ethics and Economics, Harvard University Press. Arnould, R., Bertrand, M., and K.F. Hallock (2005), “Does Managed Care Change the Mission of Nonprofit Hospitals? Evidence from the Managerial Labor Market,” In- dustrial and Labor Relations Review, 58: 494–514. Ashraf, N., Bandiera, O., Davenport, E. and S. Lee (2020), “Losing Prosociality in the Quest For Talent? Sorting, Selection, And Productivity in The Delivery of Public Services,” American Economic Review, 110(5): 1355–1394. Bala, R., and P. Bhardwaj (2010), “Detailing vs. Direct-to-Consumer Advertising,” Management Science 56: 148–160. Ballou, J.P., and B.A. Weisbrod (2003), “Managerial Rewards and the Behavior of For- Profit, Governmental, and Nonprofit Organizations: Evidence from the Hospital In- dustry,” Journal of Public Economics, 87: 1895–920. Barigozzi, F., and N. Buranib (2019), “Competition for Talent when Firms’ Mission Matters,” Games and Economic Behavior, 116: 128–151. Bartling, B., and U. Fischbacher (2012), “Shifting the Blame: On Delegation and Re- sponsibility,” Review of Economic Studies, 79: 67–87. Bartling, B., and Y. Ozdemir¨ (2017), “The Limits to Moral Erosion in Markets: Social Norms and the Replacement Excuse,” mimeo. Bartling, B., Valero, V., Weber, R.A. and L. Yao (2020), “Public Discourse and Socially Responsible Market Behavior,” mimeo. Becker, G. (1957), The Economics of Discrimination, University of Chicago Press. Bénabou, R. and J. Tirole (2006), “Incentives and Prosocial Behavior,” American Eco- nomic Review, 96(5): 1652-1678. Bénabou, R., and J. Tirole (2010), “Individual and Corporate Social Responsibility,” Coase lecture, Economica, 77: 1–19. Bénabou, R., and J. Tirole (2013), “Laws and Norms,” mimeo. Bénabou, R., and J. Tirole (2016), “Bonus Culture: Competitive Pay, Screening and Multitasking,” Journal of Political Economy, 124(2): 305–370.

36 B´enabou, R., Falk, A. and J. Tirole (2019), “Narratives, Imperatives and Moral Persua- sion,” mimeo. Bergemann, D., Bonatti, A. and T. Gan (2020), “The Economics of Social Data,” mimeo. Besley, T., and J. Malcomson (2018), “Competition in Public Service Provision: The Role of Not-for-Proﬁt Providers,” Journal of Public Economics, 162: 158–172. Besley, T., and M. Ghatak (2005), “Competition and Incentives with Motivated Agents,” American Economic Review, 95: 616–636. Besley, T., and M. Ghatak (2007), “Retailing Public Goods: The Economics of Corporate Social Responsibility,” Journal of Public Economics, 91(9): 1645–1663. Besley, T., and M. Ghatak (2018), “Prosocial Motivation and Incentives,” Annual Review of Economics, 10: 411–438. Besley, T., and T. Persson (2019), “Escaping the Climate Trap: Values, Technology and Politics,” mimeo. Bonnefon, J.F., Landier, A., Sastry, P. and D. Thesmar (2019), “Do Investors care about Corporate Externalities? Experimental Evidence,” mimeo. Bowles, S. (1998), “Endogenous Preferences: The Cultural Consequences of Markets and Other Institutions,” Journal of Economic Literature, 36: 75–111. Brekke, K.A., and K. Nyborg (2008), “Attracting Responsible Employees: Green Produc- tion as Labor Market Screening,” Resource and Energy Economics, 30(4): 509–526. Brekke, K.R., and M. Kuhn (2006), “Direct to Consumer Advertising in Pharmaceutical Markets,” Journal of Health Economics, 25: 102–30. Broccardo, E., Hart, O., and L. Zingales (2020), “Exit vs. Voice,” mimeo. Brodback, D., Guenster, N., and S. Pouget (2019), “The Valuation of Corporate Social Responsibility: An Experiment,” mimeo. Bursztyn, L., Haaland, I., Rao, A., and C. Roth (2020), “Disguising Prejudice: Popular Rationales as Excuses for Intolerant Expression,” mimeo. Choi, J.P., Jeon, D.S., and B.C. Kim (2019), “Privacy and personal data collection with information externalities,” Journal of Public Economics, 173: 113–124. Dana, J., Cain, D.M., and R.M. Dawes (2006), “What You Don’t Know Won’t Hurt Me: Costly (But Quiet) Exit in a Dictator Game,” Organizational Behavior and Human Decision Processes, 100(2): 193–201. Dana, J., Weber, R., and J. Kuang (2007) “Exploiting Moral Wriggle Room: Experiments Demonstrating an Illusory Preference for Fairness,” Economic Theory, 33: 67–80. DellaVigna, S., and U. Malmendier (2004), “Contract Design and Self-Control: Theory and Evidence,” The Quarterly Journal of Economics, 119(2): 353–402. Diamond P.A. (1971), “A Model of Price Adjustment,” Journal of Economic Theory, 3(2): 156–168.

37 Dufwenberg, M., Heidhues, P., Kirchsteiger, G., Riedel, F. and J. Sobel (2011), “Other- Regarding Preferences in General Equilibrium,” Review of Economic Studies, 78: 613– 639. Exley, C.L. (2016), “Excusing Selfishness in Charitable Giving: The Role of Risk,” The Review of Economic Studies, 83(2): 587–628. Falk, A., and N. Szech (2013), “Morals and Markets,” Science, 340(6133): 707–711. Falk, A., Neuber, T., and N. Szech (2020), “Diffusion of Being Pivotal and Immoral Outcomes,” Review of Economic Studies, 87(5): 2205–2229. Foerster, M., and J.J. van der Weele (2018a), “Denial and Alarmism in Collective Action Problems,” Tinbergen Institute Discussion Paper, TI 2018-019/I. Foerster, M., and J.J. van der Weele (2018b), “Persuasion, justification and the commu- nication of social impact,” Tinbergen Institute Discussion Paper, TI 2018-067/I. Gabaix, X., and D. Laibson (2006), “Shrouded Attributes, Consumer Myopia, and In- formation Suppression in Competitive Markets,” Quarterly Journal of Economics, 121(2): 505–540. Glaeser, E.L., and A. Shleifer (2001), “Not-for-Profit Entrepreneurs,” Journal of Public Economics, 81(1): 99–115. Green, D., and B. Roth (2020), “The Allocation of Socially Responsible Capital,” mimeo. Grennan, M., Myers, K., Swanson, A. and A. Chatterji (2020), “No Free Lunch? Welfare Analysis of Firms Selling Through Expert Intermediaries,” mimeo. Hamman, J.R., Loewenstein, G., and R. Weber (2010), ”Self-Interest through Delegation: An Additional Rationale for the Principal-Agent Relationship,” American Economic Review, 100(4): 1826–1846. Hansmann, H., (1980), “The Role of Nonprofit Enterprise,” Yale Law Journal, 89: 835– 901. Hansmann, H., (1996), The Ownership of Enterprise, Cambridge MA: Belknap Press. Hart, O., and L. Zingales (2017), “Companies Should Maximize Shareholder Welfare Not Market Value,” Journal of Law, Finance, and Accounting, 2: 247–274. Henderson, R. (2020), Reimagining Capitalism in a World on Fire, New York: Public Affairs. Hermalin, B. (1998), “Toward an Economic Theory of Leadership: Leading by Example,” American Economic Review, 88: 1188–1206. Holmström,B., and P. Milgrom (1991), “Multitask Principal-Agent Analyses: Incen- tive Contracts, Asset Ownership, and Job Design,” Journal of Law, Economics, and Organization, 7: 24–52. Inderst, R., and M. Ottaviani (2012), “Competition through Commissions and Kick- backs,” American Economic Review, 102(2): 780–809.

38 Itoh H. (1991), “Incentives to Help in Multi-Agent Situations,” Econometrica 59(3): 611– 636. Jäger,S., Schoefer, B., and J. Heining (2019), “Workers in the Board Room: The Causal Effects of Shared Corporate Governance,” mimeo. Kaplow, L., and S. Shavell (2007), “Moral Rules, the Moral Sentiments, and Behavior: Toward a Theory of an Optimal Moral System,” Journal of Political Economy, 115: 494–514. Kihlstrom, R., and J.J. Laffont (1979), “A General Equilibrium Entrepreneurial Theory of Firm Formation Based on Risk Aversion,” Journal of Political Economy, 87(4): 719–748. Kosfeld, M., and F. Von Siemens (2011), “Competition, Cooperation and Corporate Culture,” Rand Journal of Economics, 42: 23–43. Landier, A., and S. Lovo (2020), “ESG Investing: How to Optimize Impact?,” mimeo. Lazear, E.P. (1989), “Pay Equality and Industrial Politics,” Journal of Political Economy, 97(3): 561–580. Lazear, E.P., Malmendier, U., and R.A. Weber (2012), “Sorting in Experiments with Application to Social Preferences,” American Economic Journal: Applied Economics, 4(1): 136–163. Lucas, R. (1978), “On the Size Distribution of Business Firms,” Bell Journal of Eco- nomics, 9(2): 508–523. McGreal, C. (2018), American Overdose: The Opioid Tragedy in Three Acts, Public Affairs. Moisson, P.-H. (2020), “Ethics and Impact Investment,” mimeo. O’Donoghue, T., and M. Rabin (2006), “Optimal Sin Taxes,” Journal of Public Economics 90: 1825–1849. Oehmke, M., and M. Opp (2020), “A Theory of Socially Responsible Investment,” mimeo. Prendergast, C. (2007), “The Motivation and Bias of Bureaucrats,” American Economic Review, 97: 180–196. Saccardo, S., and M. Serra-Garcia (2020), “Cognitive Flexibility or Moral Commitment? Evidence of Anticipated Belief Distortion,” available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract id=3676711 Sandel, M. J. (2012), What Money Can’t Buy: The Moral Limits of Markets, Macmillan. Satz, D. (2010), Why Some Things Should Not Be For Sale: The Moral Limits of Markets, Oxford University Press. Shapiro, C. (1983), “Premiums for High Quality Products as Returns to Reputation,” Quarterly Journal of Economics, 98(4): 659–680.

39 Shleifer, A. (2004), “Does Competition Destroy Ethical Behavior?,” American Economic Review, Papers & Proceedings, 94: 414–418. Schneider, F.H., Brun, F., and R.A. Weber (2020), “Sorting and Wage Premiums in Immoral Work,” mimeo. Schnell, M. (2019), “The Opioid Crisis: Tragedy, Treatments and Trade-offs,” Stanford Institute for Economic Policy Research Policy Brief, February. Singer, P. (2015), The Most Good You Can Do, Yale University Press. Sobel, J. (2015), “Do Markets Make People Selfish?,” Mimeo. Spence, A.M. (1975), “Monopoly, Quality, and Regulation,” The Bell Journal of Eco- nomics, 6(2): 417–429. Tirole, J. (2001), “Corporate Governance,” Econometrica, 69(1): 1–35. Walzer, M. (2008), The Free Market and Morality on YouTube. Weisbrod, B. (1988), The Nonprofit Economy, Cambridge MA: Harvard University Press. Weyl, G., and M. Fabinger (2013), “Pass-Through as an Economic Tool: Principles of Incidence under Imperfect Competition,” Journal of Political Economy, 121(3): 528– 583.

40 Main Appendix

A Proof of Proposition 4

We can subsume the overall analysis in a single optimization. Let ui denote the ethical choice of supplier i, with welfare consequence W (ui) (satisfying Assumption 1), x = 1 if cutting ethical corners brings beneﬁts on the demand side and x = 0 if the beneﬁt is on the cost side. Let α, a and a denote the supplier’s, consumers’ and investors’ degrees of internalization of social welfare. Supplier i’s optimization program is:

∗ ∗ ∗ ∗ max pi − [c − (1 − x)ui] + (α + a)[W (ui) − w ] Di(pi − xui − aW (ui), p − xu − aw ).

p−c+(1−x)u∗ In a symmetric equilibrium, let L ≡ p denote as usual the relative price-cost margin, and η = − ∂Di / Di denote the elasticity of demand. The FOC with respect to ∂pi pi the ethical choice is

(1 − x) + [α + a]W 0(u∗) + ηL[x + aW 0(u∗)] = 0.

Under market-determined prices, ηL = 1 and so

[α + a + a][−W 0(u∗)] = 1.

The ethical concerns simply add up. Under constrained prices, all depends on the impact of cutting ethical corners. If cutting corners increases demand (x = 1), then

[α + a + aηL][−W 0(u∗)] = ηL.

If cutting corners lowers cost (x = 0), then

[α + a + aηL][−W 0(u∗)] = 1.

B Proof of Proposition 6

We structure the proof along the following lines: We start with demand-side benefits and then move on to cost-side ones. Each time, we study sequentially constrained and market prices. (a) Demand-side benefits. Suppose that suppliers differ in their ethical concerns; without loss of generality, assume that 0 ≤ α1 ≤ α2 ≤ · · · ≤ αn. Supplier i solves:

max {(pi − c)Di + αi [wiDi + Σj6=iwjDj]} .

41 Σj6=iwj Constrained prices. Let wi = W (ui) andw ¯−i = n−1 . In the constrained-price context, pi ≡ p for all i, and the only decision variable is ui. The maximization of supplier i’s objective function with respect to ui yields ﬁrst-order condition

0 σ[(p − c) + αi(wi − w¯−i)] + αiW (ui)Di = 0. (B.1)

The intuition behind condition (B.1) goes as follows: A unit increase in ui attracts σ new clients, bringing markup (p − c) on each of them. This increase in market share 55 further improves welfare by wi − w¯−i (decreases it if wi < w¯−i). Finally, the decrease in the welfare corresponding to the Di clients of supplier i has an ethical cost for supplier i.

Consider two arbitrary firms i and j. Suppose that αi ≥ αj, but ui > uj. Then wi < wj n and so (wi −w−i)−(wj −w−j) = n−1 (wi −wj) < 0. Furthermorep î < pˆj and the symmetry 0 0 of the demand functions imply that Di > Dj. Finally, [−W (ui)] > [−W (uj)] > 0. And so (B.1) cannot be satisfied for both i and j, a contradiction. This proves (i) for this case. Next, let us check whether the solution to the FOCs satisfies financial viability and positive demand. Because p must exceed c for firms to be financially viable, the financial constraint is not binding. Remembering that more ethical firms have a lower market ∗ share, let i ≡ max{i|Di > 0}. We show that all firms actually have a positive market ∗ ∗ share. Suppose, a contrario, that i < n. Any supplier i > i could set ui∗ − ε for ε small, command a positive market share, make a financial profit and improve overall morality. Hence all suppliers have a strictly positive equilibrium market share. To show the race-to-the-ethical-bottom property, consider (B.1) for i = n. It implies 0 that −αnW (un) ≥ nσ(p − c) (using Dn ≤ 1/n and wn − w−n ≥ 0). Therefore as σ goes 0 to ∞, W (un) goes to −∞, implying u1 ≥ u2 ≥ · · · ≥ un all go to ∞.

Market-determined prices. The ﬁrst-order condition with respect to pi is

−σ[(pi − c) + αi(wi − w¯−i)] + Di = 0. (B.2)

The intuition for (B.2) is similar to that underlying condition (B.1). Furthermore, combining (B.1) and (B.2) yields

0 −αiW (ui) = 1. (B.3)

As we earlier noted, the invariance property extends to heterogenous ethical concerns. While (B.2) relies on a linear demand system, (B.3) actually holds for general demand. Do ﬁrms with higher ethical concerns still command a lower market share? Condition (B.3) implies that their ethical choices will make them unattractive to clients. But this

55Thus, a less ethical supplier shows slightly less eagerness to capture market share than would happen if other suppliers were equally ethical. Note, though, that this effect is small, precisely because that supplier’s ethical concerns are small in the first place; indeed, there is no such effect for an ethics-free supplier. More generally, such concerns are always dominated by the profit incentive and the less ethical supplier captures a larger market share.

42 is not the end of the story. Their lack of attractiveness calls for lower prices. And their ethical concerns also make them eager to capture market shares away from less scrupulous suppliers, who conversely do not want to gain market share for that speciﬁc reason. We study the horserace between these two forces in part (ii) of the next proposition.

Under a linear demand system, the demand faced by a supplier with ethics αi is 1 nj Di = n − σ n−1 (ˆpi − pˆj). Subtracting the ﬁrst-order conditions (B.2) for the two groups yields 2n − 1 h n n i (ˆp − pˆ ) = u − u + (w − w ) α 2 + α 1 . n − 1 2 1 1 2 1 2 1 n − 1 2 n − 1

Condition (B.3) implies that u1 > u2, w1 < w2; and the concavity of W , together with condition (B.3) yields

u1 − u2 + α1(w1 − w2) > 0 > u1 − u2 + α2(w1 − w2).

∗ ∗ Thus there exists n1 < n such thatp ˆ2 < pˆ1 if and only if n1 ≥ n1. In the duopoly case (n1 = n2 = n − 1),p ˆ2 < pˆ1 and so the ethical firm commands a higher market share. Next, let us show that the ethical group’s financial viability constraint is binding for σ ≥ σ∗ for some σ∗. Let  n1  c2 ≡ c − α2 (w2 − w1)  n − 1 n  c ≡ c + α 2 (w − w ) > c  1 1 n − 1 2 1 2 denote the two groups’ “virtual costs” (which reflect the social responsibility indices and are independent of the intensity σ of competition). The FOC with respect to the prices are

σ(fk − ck) = Dk for k ∈ {1, 2}, implying pk ≥ ck. Adding up (with weights nk), one has 1 n (p − c ) + n (p − c ) = . 2 2 2 1 1 1 σ

Furthermore, using the linear expressions for Dk and the invariance result under market 56 prices, indicates that p2 −p1 remains constant as σ moves. So p2 and p1 decrease equally as σ increases. Because p1 ≥ c1 > c > c2, the constraint p2 ≥ c becomes binding for high enough σ, while p1 ≥ c is never binding. When p2 = c, the ethical firms’ financial viability becomes constraining, and so these firms behave like not-for-profits (see Section 4.2).

0 Because u1 is independent of the elasticity of demand (−α1W (u1) = 1) and u2 < u1, there is no race to the ethical bottom. The reason why the property obtained in the constrained prices case does not hold here is that the ﬁnancial stakes in attracting consumers converge to 0 (pi − c −→ 0) as σ goes to ∞.

56 1 p2 − p1 = − 2n−1 [n(u1 − u2) + (n1α2 + n2α1)(w2 − w1)] < 0.

43 (b) Cost-side benefits. When cutting ethical corners lowers cost, supplier i’s objective function is h h Σ w ii V = p − (c − u ) + α W (u ) − j6=i j D . i i i i i n − 1 i As earlier, cost-side benefits imply that ethical behavior is invariant to the nature of competition: 0 −αiW (ui) = 1. This implies that there is no race to the ethical bottom in that case and that ethical suppliers have the same market share as unethical ones under constrained prices. The demand faced by all firms is fixed at 1/n. Because financial viability for a type-αi supplier 0 1 requires that p ≥ c − ui, one has ui = max{W − , c − p}. This proves (i). αi Is this still the case under market-determined prices? Suppose that they are two groups of suppliers: n1 (resp. n2) have social preferences α1 (resp. α2 > α1), with n1 + n2 = n. Let k ∈ {1, 2} denote the two types of suppliers. A supplier with social preferences αk selects pk given by h (n − 1)w + n w i σ p − c + u + α w − k k l l = D k k k k n − 1 k h n (w − w )i ⇔ σ p − c + u + α l k l = D (B.4) k k k n − 1 k

n1(w2−w1) n2 ∗ We claim that u2 + α2 n−1 > u1 + α1 n−1 (w1 − w2) if and only if n1 ≥ n1 for some ∗ ∗ n1, since α1(w2 − w1) < u1 − u2 < α2(w2 − w1). If n1 ≥ n1, f2 ≥ f1 would imply that the LHS of (B.4) is higher for type 2. But D2 ≤ D1 and so the RHS is smaller for type ∗ 2, a contradiction. So for n1 ≥ n1, ethical firms have a larger market share. Similarly, ∗ if n1 < n1 and p2 ≤ p1, the LHS of (B.4) is smaller and the RHS larger for type 2, a contradiction. So unethical firms have a larger market share. We have so far focused on the FOCs. Could the ethical suppliers’ financial viability constraint be binding? Rewrite condition (B.4) lettingp ˜k ≡ pk + uk: h n (w − w )i σ p˜ − c + α l k l = D , k k n − 1 k where the uk, ul, wk, wl are independent of σ from the invariance property. Formally, we are back to the demand-side-benefit, market-price case, and we can apply the analysis derived there. This proves (ii) for this case.

C Proof of Proposition 7

(a) Demand-side beneﬁts. A not-for-proﬁt sets pi = c. Its ethical choice is then obtained by solving

max α2[DiW (ui) + Σj6=iDjW (uj)].

44 So, assume that there are n1 for-profit suppliers and n2 = n − n1 not-for-profit ones. 1 (ni−1)ˆpi+nj pˆj Demands are Di = n −σ1 qî − n−1 when a firm of type i deviate from equilibrium net pricep î to net feeq î. In equilibrium, a for-profit supplier sets p1 ≥ c (firms cannot make a negative profit, although this requirement will not be binding in the solution † obtained below) and u1 = u1 (recall that condition (5) always holds when prices are unconstrained) and receives demand D1, while a not-for-profit one picks p2 = c and u2, yielding demand D2. For a linear demand system, the two first-order conditions are h n i α W 0(u )D + (w − w )σ 1 = 0 (C.5) 2 2 2 2 1 1 n − 1 and h n i α W 0(u )D + (w − w )σ 2 + σ (p − c) = 0, (C.6) 1 1 1 1 2 1 n − 1 1 1 † 0 † where u1 = u1 with −α1W (u1) = 1.

(i) The first result is that, as is intuitive, u2 ≤ u1: Not-for-profits behave more ethically. Suppose, to the contrary, that w1 > w2. Then (C.5) yields u2 < uˆ and (C.6) (together with p1 ≥ c) implies that u1 > uˆ. But a not-for-profit firm’s deviation to ui =u ˆ is both more ethical (it increases wi), and takes market share away from other not-for- profits and for-profits (for whom wj ≤ wˆ) a contradiction. So w2 ≥ w1. Finally, if u1 < uˆ, a for-profit could raise ui tou ˆ, gain market share, raise wi in absolute and relative to all competitors. So u1 > u2 > uˆ overall.

(ii) Next, suppose that competition is intense (σ1 goes to inﬁnity). Then u1 − u2 → 0 † from the ﬁrst condition (u2 cannot go tou ¯ as u2 < u1 = u1). The second condition 57 implies that p1 → c.

(b) Cost-side benefits. As earlier, we assume that there are n1 for-profits and n2 not-for- profits. The novel feature here is that not-for-profits’ price reflects their ethical choice:

The not-for-proﬁts must select p2 = c − u2. So they solve:

h (n2 − 1)W (u2) + n1W (u1)ih 1 h (n2 − 1)(c − u2) + n1p1 ii max α2 W (˜u2) − − σ1 c − u˜2 − {u˜2} n − 1 n n − 1 while for-proﬁts solve:

h h (n1 − 1)W (u1) + n2W (u2)iih 1 h n2(c − u2) + (n1 − 1)p1 i max p˜1−(c−u˜1)+α1 W (˜u1)− −σ1 p˜1− . {p˜1,u˜1} n − 1 n n − 1

The FOCs for not-for-proﬁts and for-proﬁts are n [−W 0(u )]D = σ 1 (w − w ), (C.7) 2 2 1 n − 1 2 1 0 α1[−W (u1)] = 1 (C.8)

57 α1 n2 0 0 0 As p1 − c = W (u2)D2 − W (u1)D1 with D1,D2 ≤ 1 and |W (ui)| ≤ 1/αi. σ n1

45 and h n i σ p − (c − u ) + α 2 (w − w ) = D (C.9) 1 1 n − 1 1 2 1

1 n1 1 n2 where D2 = n − σ1 n−1 [(c − u2) − p1] and D1 = n − σ1 n−1 [p1 − (c − u2)].

Let ∆w ≡ w2 − w1 and ∆u ≡ u1 − u2. From (C.7), ∆w > 0 (and so ∆u > 0). Using (C.7), (C.8) and (C.9),

n1 ∆w D 0 2 = n−1 [−W (u2)] . n2 ∆w D1 (p1 − c + u1) − 0 n−1 [−W (u1)]

Furthermore

D2 > D1 ⇐⇒ p1 > c − u2 ⇐⇒ p1 − c + u1 > ∆u.

So h n1 1 n2 1 i D2 > D1 ⇐⇒ ∆u < 0 + 0 ∆w n − 1 [−W (u2)] n − 1 [−W (u1)] 0 0 knowing that, from the concavity of W ,[−W (u1)]∆u > ∆w > [−W (u2)]∆u. So for n1 = n − 1, D2 > D1. † Next, suppose that competition becomes intense (σ → +∞). Because u2 ≤ u1 = u1, u2 cannot tend tou ¯ and so (C.7) implies convergence in ethical behavior: ∆w, ∆u → 0, † which in turn implies that u2 → u1. Condition (C.9) and the fact that D1 ≤ 1 imply that p1 → c − u1 (marginal cost pricing).

46 Supplementary Appendix

A Multiplicity of equilibria for constrained prices

∂η W 00 Equation (4) is guaranteed to have a unique solution if ∂u∗ < W 0 , which is for instance satisﬁed for CES, Hotelling-Lerner-Salop demand functions and the symmetrical linear demand system.58 But multiple equilibria may co-exist more generally (in which case the “most ethical equilibrium” Pareto dominates the others59). To illustrate the potential multiplicity, assume that the market is covered and that a uniform change in prices leaves all demands invariant (which is a necessary implication of full coverage in discrete choice demand models, since a uniform change in price does not alter the relative attractiveness of products). Finally, suppose that there are only two levels of utility, say u1 and u2 with u1 > u2, with associated welfares w1 < w2 (which violates Assumption 1). For these two levels to be both equilibria, it must be the case that p − c ≥ [p − c − α(w2 − w1)]D(p − u1, p − u2) n ≥ [p − c + α(w2 − w1)]D(p − u2, p − u1)

We now show the following:

Proposition 9 (strategic complementarities). Suppose that n = 2 or/and demand is h P (p−u ) i ˜ 1 j6=i j linear (Di = n − σ (p − ui) − ( n−1 ) ). Then, in the binary utility case, as p increases from c to +∞,

∗ • ﬁrst, the only equilibrium is the ethical one (u = u2);

• then both levels of utility are sustained in symmetric equilibrium;

∗ • ﬁnally, for large prices, the only equilibrium is the unethical one (u = u1).

Proof. As we noted, demands are independent of p. For conciseness let D+ ≡ D(p − 1 − u1, p − u2) > n > D ≡ D(p − u2, p − u1). The conditions for the two potential equilibria can be rewritten as:

α(w − w ) 1 ≥ [1 − 2 1 ]nD+ p − c and α(w − w ) 1 ≥ [1 + 2 1 ]nD− p − c

58Hotelling-Salop-Lerner demand functions are linear, but the impact of supplier i’s policy is not symmetrical. 59To deﬁne Pareto dominance, one must look at the suppliers’ gross utilities, and not the net utilities that are used to derive optimal strategies; indeed the latter are equal to (p − c)/n in all symmetric equilibria.

47 Note that the RHS of the first inequality increases with p ∈ (c, ∞) from −∞ to nD+ > 1, while the RHS of the second inequality decreases from +∞ to nD− < 1. To conclude the proof it suffices to show that when, say, the first inequality is satisfied with equality, the second inequality is strictly satisfied. Suppose, first, that n = 2. Then D+ + D− = 1 (as the market is covered). The result then follows from the property that D+ > 1/2. When n is arbitrary, but demand is linear, the reasoning goes as follows: ˆ u1+u2 + 1 define f as the “average net price” (ˆp ≡ p− 2 ) and rewrite D = n −σ(ˆp−ε)+σ(ˆp+ε) − 1 and D = n −σ(ˆp+ε)+σ(ˆp−ε). Using the facts that the two inequalities are satisfied as ∂D+ ∂D− + − equality for exactly the same p when ε = 0, that ∂ε = 2σ = − ∂ε and that D > D , we obtain the result announced in the text.

B Verifying global second-order condition under market- determined prices

Let us check that the tentative equilibrium is a global optimum for each supplier. Equi- librium behavior requires that there be no (pi, ui) such that

∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ (p − c)Di(p − u , p − u ) < [pi − c + α[W (ui) − W (u )]] Di(pi − ui, p − u ) ≡ V(pi, ui).

The concavity of W and condition (5) imply that

∗ ∗ α[W (ui) − W (u )] ≤ −[ui − u ].

So ∗ ∗ ∗ V(pi, ui) ≤ (pi − c − ui + u )Di(pi − ui, p − u ).

∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ We also know that for allp ˜i,(p − c)Di(p − u , p − u ) ≥ (˜pi − c)Di(˜pi − u , p − u ). ∗ Applying this top ˜i = pi − ui + u yields

∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ (p − c)Di(p − u , p − u ) ≥ (pi − c − ui + u )Di(pi − ui, p − u ), a contradiction.

C Some extensions

C.1 Outside option

We consider the same oligopoly model, but in which outside options are “co-located” with the products. This model oﬀers the convenience of a smooth transition from perfect competition to pure monopoly.60 In this nested discrete choice model, (1) the client

60B´enabou-Tirole (2016). This formulation, unlike Hotelling-style speciﬁcations, allows a distinction between the within-industry elasticity and the elasticity with respect to the outside good, as seen below.

48 chooses among the suppliers (so Σi=1,...,nDi = 1); (2) the client chooses between her preferred within option (i, say) and an outside option co-located with i. The outside option oﬀers utility u0 and welfare w0; it involves an idiosyncratic cost κ ∈ R of using the outside option, distributed according to cumulative distribution X(κ). Without loss of generality, we assume that the price of the outside option is equal to that of the regulated- price products (p0 = p); if this is not the case, one can renormalize the distribution of κ to obtain the expressions below. We assume that X0/(1 − X) is increasing (monotone ∗ hazard rate property). The demand for good i is then [1 − X(u0 − ui)]Di(p − ui, p − u ). ∂D i 0 ∂pi −Xi Let η ≡ − Di denote the “within elasticity” and ε ≡ 1−Xi denote the “across elastic- pi pi ity”. The total elasticity isη ˆ ≡ η + ε. Similarly, the Lerner indices in equilibrium are equal to: p − c p − c + α(w∗ − w ) L ≡ and K ≡ 0 . p p The following proposition shows that unethical and attractive outside options exacer- bate the need to be attractive to capture demand and reduce suppliers’ ethics:61

Proposition 10 (Outside options). In the presence of co-located outside options, the ethical choices are given in a symmetric equilibrium by

α(w∗ − w ) −αW 0(u∗) = ηL + εK =ηL ˆ + 0 ε. p

When demands are linear and the cumulative distribution X(κ) exponential, suppliers are less ethical, the less ethical the outside option (the lower w0 is).

Application: There is detailed empirical evidence that, when the FDA forced Purdue pharma to make OxyContin “abuse-resistant”,62 patients massively shifted towards heroin. As we suggest, a wider availability of illegal drugs (e.g. when fentanyl made illegally in China becomes increasingly available in the US and Europe) has an additional eﬀect: Doc- tors become less ethical (not in preferences, but in behavior as they increase u∗) because

61The proof is obtained by noting that supplier i solves (assuming a symmetric equilibrium, so we can ∗ rewrite the demand function as Di(p − ui, p − u ))

∗ ∗ max Vi = X(u0 − ui)Di(p − ui, p − u )[αw0] + [1 − X(u0 − ui)]Di(p − ui, p − u )[p − c + αW (ui)] {ui} ∗ ∗ ∗ ∗ ∗ + X(u0 − u )[1 − Di(p − ui, p − u )][αw0] + [1 − X(u0 − u )][1 − Di(p − ui, p − u )][αw ], and by taking the first-order condition. In a symmetric equilibrium, the linear demand for supplier i is: ∗ −λκ Di = 1/n−(n−1)σ[(p−ui)−(p−u )], while the exponential cumulative distribution is: X(κ) = 1−e . Given these assumptions, one can show that the first-order condition is: −αW 0(u∗) = [n(n − 1)σ + λ](p − ∗ ∗ c) + αλ(w − w0), which implies that 0 < dw /dw0 < 1. 62The abuse-deterrent version of OxyContin, approved by the FDA in 2010, makes the pill difficult to crush or dissolve, making it less dangerous. Alpert et al (2018) show that the OxyContin reformulation significantly reduced OxyContin misuse, but also led to a large increase in heroin deaths.

49 they feel that keeping patients addicted to prescription opioids limits a (worse) addiction to illegal opioids.63

C.2 Optimal price

(a) Socially optimal price under free entry. Assume that the suppliers have reservation utility V¯ > 0, and look for a free entry equilibrium for an arbitrary administered price p . Adding an entry decision implies that a supplier can impact the overall outcome both through its entry decision and the ethical decision once in the market. We assume that the replacement effect operates fully at the entry stage; this amounts to assuming that if a supplier does not enter, another supplier will. Put differently, if in equilibrium n suppliers find it individually rational to enter, were one of them not to enter, another would replace him. Thus, the individual rationality p−c ∗ ¯ ∗ constraint for the entrants is n + α[W (u (n)) + D(n)] ≥ V + α[W (u (n)) + D(n)] where D(n) measures the diversity benefits, with D0 > 0.64 So let us rewrite the individual rationality constraint as: p − c V = ≥ V.¯ i n This constraint, which is binding, defines a function n(p) ≡ (p − c)/V¯ .65 Letting u∗(p, n) denote the equilibrium level of u, the social planner’s objective function reflects the clients’ welfare (suppliers have no surplus under free entry):

W(p) ≡ W (u∗(p, n(p))) − p + D(n(p)) − nV.¯

For example, in the Lerner-Salop generalization of the Hotelling model (n firms are symmetrically located on a circle of length 1, and consumers are uniformly distributed along the circle and have unit transportation cost t), D(n) = −t/4n. We assume that W(p) is quasi-concave (a sufficient condition for this is that the diversity benefit D(n) be concave, as is the case in the Lerner-Salop model). The social planner maximizes W subject to

−αW 0(u∗(p, n(p))) = η(p, n(p))L(p).

63It would be worth thinking more about the general implications for the legal treatment of various drugs. Another case in point is cannabis, which is viewed both as a drug and a painkiller which could help reduce the opioid epidemic. In our model, if doctors could prescribe it, it could prevent OxyContin addiction for some patients, while now it is “just an outside option”, which is moreover less attractive to dealers than heroin or fentanyl. 64In theory it is possible that each of the n suppliers would prefer an outcome in which he refrains from entering and nobody replaces him. This would yield a partial replacement effect. Our assumption, which simplifies expressions, can be justified as a perfect equilibrium of a game in which firms take turn making an entry decision and do enter as long as the number of firms that have entered yet is smaller than the level, n, given by the free entry condition. 65As usual, we treat n as a continuous variable rather than as an integer, solely for conciseness purposes.

50 Let pR(α) denote the optimal regulated fee. Absent ethical concerns (α = 0), the optimal regulated price pR(0) would satisfy

D0(n(pR(0)) = V.¯

With ethical concerns (α > 0), the optimal regulated price pR(α) further embodies its du∗ ∂u∗ ∂u∗ dn ethical impact. To see how, letting dp ≡ ∂p + ∂n dp and similarly for d(ηL)/dp, one has: du∗ d −αW 00(u∗) = (ηL). dp dp

Next, from Assumption 3(i) and

dW [−W 0] d D0(n(pR(α)) = (ηL) − 1 + = 0, dp αW 00 dp V¯ and so we obtain pR(α) < pR(0).

(b) HMO Consider now a health maintenance organization, which sells coverage to clients and contracts with health-care suppliers. So the price for coverage if u∗ +D(n). Let us assume that the HMO faces the same participation constraint as the social planner. The HMO has no ethical concern and maximizes its proﬁt

π ≡ u∗ − p + D(n) subject to p − c ≥ V¯ n and −αW 0(u∗(p, n)) = η(p, n(p))L(p). The same reasoning shows that, under Assumption 3, the price set by the HMO satisﬁes

pHMO > pR(0).

Proposition 11 The prices set by a social planner under ethical concerns (pR(α)) and no ethical concerns (pR(0)), and by an HMO under ethical concerns (pHMO) satisfy

pR(α) < pR(0) < pHMO.

In particular, a social planner sets a low price and reduces the intensity of competition so as to enlist the suppliers’ ethical concerns, while an HMO panders to the clients and to do so raises the fee and the intensity of competition.

51 C.3 Pigovian policies

Divergences between social welfare and individual goals call for Pigovian corrections when feasible (see the introduction for the various reasons why Pigovian taxation in practice does not exist or is suboptimal). At an abstract level, and assuming that ui is veriﬁable, the government might want to levy a tax t(ui − uˆ) on “excessive attractiveness” (ui − uˆ). For example, under regulated prices, supplier i would solve

∗ max[p − c − t(ui − uˆ) + α[W (ui) − W (u )]]Di. {ui}

Simple computations show that to achieve the most ethical level u∗ =u ˆ, the tax should be set at level t = ηL where L ≡ (p − c)/p. Example. Consider the overconsumption model (Example 1). While we expressed the tax t in terms of utils, it is more natural to express it in terms of quantities (number ∗ of prescriptions). Let τ denote the unit tax on “excess supply” G(ci ) − G(b). Simple computations show that τ = [(1 − β)b]t. Supplier i maximizes

h ∗ ∗ ∗ i ∗ ∗ p − c − τ[G(ci ) − G(b)] + α[W (ci ) − w ] D(p − u(ci ), p − u(c ))

0 ∗ ∗ ∗ 0 ∗ ∗ ∗ with W (ci ) = (b − ci )g(ci ) and u (ci ) = (b − βci )g(ci ). In the case of a constrained fee, ∗ the tax that delivers the optimal consumption ci = b is given by:

τ = ηL(1 − β)b. (C.1)

It is instructive to see how this differs from the classical Pigovian prescription. We know from DellaVigna-Malmendier (2004) that the optimal Pigovian tax when the supply side is competitive and profit-maximizing exactly offsets the internality on the future self:

τ = (1 − β)b.

For a market-determined fee, the ﬁrst-order condition with respect to pi is given by

ηL = 1.

And so, unlike in the constrained-price case, the optimal tax is the same as when the market is competitive and suppliers are selﬁsh, i.e. the DellaVigna-Malmendier level:

τ = (1 − β)b. (C.2)

52 D Proof of Proposition 8

Demand-side beneﬁts

Constrained prices. Supplier i’s objective function is (pi −c)Di +α(wiDi +wjDj). With n 1 Σj6=ipˆj firms and a linear demand system Di = n −σ pî − n−1 , supplier i’s first-order condition is 0 σ[p − c + α(wi − w−i)] = −αW (ui)Di (D.3)

Σj6=iwj where w−i ≡ n−1 . Suppose, a contrario, that si > sj and ui ≥ uj. Then Di > Dj, so from the concavity of W , the RHS of (D.3) is strictly larger for supplier i than for supplier j. Because wi − w−i ≤ wj − w−j, the LHS of (D.3) is weakly smaller for supplier i, a contradiction. The intuition is that the high-quality supplier ceteris paribus faces a larger demand and so bears a bigger share of the moral duty to generate an ethical outcome. In that, it is consistent with Proposition 2, which considers the symmetric case, but shows that a monopoly behaves more ethically than an oligopoly. Market-determined prices. Supplier i solves

max (pi − c)Di + α(ΣwjDj), {pi,ui} where Di = D(ˆp1,..., pˆi,..., pˆn). The ﬁrst-order conditions with respect to pi and ui are, respectively ∂Di Di + [pi − c + α(wi − w−i)] = 0 (D.4) ∂pi and 0 ∂Di αW (ui)Di − [pi − c + α(wi − w−i)] = 0. (D.5) ∂pi

0 † 0 −1 1 Combining (D.4) and (D.5) yields: αW (ui) = −1 ⇔ ui = u ≡ (W ) − α for all i. And so 0 † −αW (u )Di = σ(pi − c). (D.6)

Take two suppliers i and j with si > sj. Suppose, a contrario, that pi ≤ pj and so pî < pˆj. Then Di > Dj. So (D.6) cannot be satisfied for both i and j. The same reasoning shows that while pi > pj,p î < pˆj and so supplier i commands a higher market share. ˆ Cost-side benefits. Supplier i solves, letting fi ≡ fi − si

max[ˆpi + si − (c − ui)]Di(ˆpi, pˆ−i) + α[W (ui)Di(ˆpi, pˆ−i) + Σj6=iW (uj)Dj(ˆpj, pˆ−j)].

† 0 † Under both constrained and market prices, ui = u for all i, where −αW (u ) = 1. So, for constrained prices,p ˆi ≤ pˆj ⇐⇒ si ≥ sj ⇐⇒ Di ≥ Dj.

For market prices, the FOC w.r.t.p ˆi yields

† σ(ˆpi + si − c + u ) = Di.

53 Suppose that si ≥ sj, but Di ≤ Dj. Thenp ˆi ≥ pˆj and sop ˆi + si > pˆj + sj, which contradicts the above FOC.