Efficiency in Dynamic Agency Models

Size: px
Start display at page:

Download "Efficiency in Dynamic Agency Models"

Transcription

1 Efficiency in Dynamic Agency Models Anqi Li First Draft: January 2015 This Draft: December 2017 Abstract We examine a dynamic agency model where the agent s hidden action can affect current and future signals. We show that when players interact for a large number of instances, asymptotic efficiency can be attained if the monitoring technology satisfies two general conditions named measure concentration and informativeness, and if the agent can be penalized by reductions in the instantaneous consumption or the future payoff. We use this result to establish a Folk Theorem for discrete-time agency models with patient players, and to identify signal processes that yield asymptotic efficiency in discretized continuous-time agency models. Key words: incentive contract; concentration inequality. Department of Economics, Washington University in St. Louis. anqili@wustl.edu. I am indebted to my advisor Ilya Segal for inspiration, guidance and support. I also thank Marinho Bertanha, Huiyu Li, George Mailath, and seminar participants at Berkeley, UPenn, SAET 2014 and ASSA 2015 for helpful discussions. 1

2 1 Introduction The question of when and how we can achieve near-efficiency has been the central focus of the existing studies on dynamic agency models with moral hazard. So far, various answers to this question have been given, including: Radner (1985), which establishes a positive result in discrete-time models with patient players; Holmstrom and Milgrom (1987), Hellwig and Schmidt (2002), Sannikov (2008) and Sadzik and Stacchetti (2015), which obtain negative results in (discretized) continuous-time models with Brownian motion signals. These studies, albeit insightful ones, consider specific signal processes and limit the agent s actions to affecting only the concurrent signals. In this paper, we examine a T -instant model where the agent s actions can affect both current and future signals, whereas signals exhibit serial independence for any given action profile. We identify two conditions on the monitoring technology, namely measure concentration and informativeness, that ensure the attainability of near-efficiency when T is large, and examine the implications of this result for the efficiency properties of discrete-time and discretized continuous-time agency models. Our analysis exploits concentration inequalities for independent random variables, which have served as a major tool in computer science, information theory, learning theory, etc., but have somehow been overlooked by the dynamic agency literature. Our focus is on McDiarmid s (1989) concentration inequality, which prescribes a lower bound for the probability that a test statistic with bounded differences is concentrated around its mean (hereafter, concentration bound). Measure concentration says that such test statistic can be constructed, and that its concentration bound holds uniformly over all signal-generating action profiles and converges to one as T grows to infinity. Informativeness goes one step further, stipulating that the mean test statistic is informative of the agent s long-term actions, in that actions generating similar mean test statistic to efficient actions are themselves near-efficient. When both conditions are met, we can attain near-efficiency through a simple test contract when T is large. The contract pays a fixed wage at the first T 1 instances, and penalizes the agent at instant T if the sample test statistic falls short of the mean test statistic under the efficient actions minus a vanishing adjusted term. A key step in the proof is to bound the probability that the agent passes the test by sheer luck, as well as his gain from carefully adjusting action choices as information gets revealed 2

3 over time. Our starting observation is that since the monitoring technology exhibits serial independence, it follows that under any optimal decision rule, signals convey no information about each other and hence satisfy the preconditions of McDiarmid s concentration inequality for any given profile of realized actions. Further, since the concentration bound holds uniformly over all action profiles, it suffices to focus on the high probability event where the sample test statistic is concentrated around its mean when T is large. Meanwhile, informativeness stipulates that in this high probability event, the agent s actions be near-efficient in order to pass the test. This suggests that a big penalty for failure suffice to induce near-efficient actions most of the time, from which all the other results will follow. We give two applications of this result. First, we generalize Radner s (1985) Folk Theorem for discrete-time agency models by allowing signals to depend on past actions. Second, we identify signal processes that yield asymptotic efficiency in discretized continuous-time agency models, such as Poisson and Gamma processes, but not Brownian motion with drift. Compared to Poisson and Gamma processes, Brownian motion tends to drift apart from its mean more, suggesting that we keep the performance threshold slack in order to satisfy measure concentration. But this comprises informativeness, which requires that we keep tightening the performance threshold as players interactions become increasingly frequent. This result sheds light on how the principal should choose between different monitoring technologies in real-world situations. 1.1 Literature Review The current paper provides a unifying analysis of the efficiency properties of discretetime and discretized continuous-time agency models. Existing studies on review contracts, including Radner (1981), Rubinstein and Yaari (1983) and Radner (1985), use pointwise convergence theorem to prove efficiency results in discrete-time agency models. Existing discretized continuous-time agency models include, but are not limited to Hellwig and Schmidt (2002) and Sadzik and Stacchetti (2015), both dealing with signal processes that converge in law to Brownian motion with drift. In order to handle persistence, we exploit universal properties of the test statistic, namely measure concentration and informativeness, that apply to all signal-generating action profiles. Our result holds true even if signals can depend on past actions, and 3

4 it enables us to identify new signal processes that attain asymptotic efficiency in discretized continuous-time models. Concentration inequalities have served as a major tool in computer science, probability theory, information theory, learning theory, etc., but have somehow been overlooked by the dynamic agency literature. See Boucheron et al. (2013) for a textbook treatment of this subject matter. For a non-exclusive list of economics applications, see Kalai (2004) and Deb and Kalai (2015) for studies on large games, Al-Najjar and Smorodinsky (2001) for repeated large games, and Jackson and Golub (2012) for learning in social networks. In order to construct the test contract, the principal needs to know the the mean test statistic and the action cost under the efficient action profile, but nothing else. In a related paper on robust dynamic incentive contracting, Chassang (2013) assumes that the output process is unknown to the principal and proposes a limited-liability contract that approximates the performance of a linear contract. The construction exploits Blackwell s (1956) approachability theorem, whose proof itself involves the use of concentration inequalities. Our goal of attaining asymptotic efficiency is different than that of Chassang (2013), and our monitoring technology is more restrictive than his. Further, our construction does not invoke Blackwell s (1956) regret-minimizing algorithm, and it can be made to satisfy limited liability only in limiting cases. The remainder of this paper is organized as follows: Section 2 outlines the model; Section 3 states the main result; Section 4 investigates several applications; Section 5 concludes. See Section A for omitted proofs, as well as the online appendix for additional results. 2 The Model 2.1 Notations In what follows, T denotes an integer number and v T a vector that consists of an ordered list of T objects (v 1,, v T ). The notation for T is occasionally suppressed, as long as this causes no confusion. The Landau notations O( ) and Θ( ) stand for at most the order of and exactly the order of, respectively. 4

5 2.2 Setup Primitives Consider a series of economies indexed by T = 1, 2,, each involves a risk-neutral neutral principal (she) and a risk-averse agent (he) interacting for T instances. At each instant t = 1,, T, the agent takes a hidden action a t in R +, a signal X t is publicly realized, and the principal pays the agent a real-valued wage w t. Throughout t = 1,, T, the agent derives a utility U(w T ) from consumptions and pays a cost C (a T ) for taking actions, whereas the principal earns an expected revenue V (a T ) and incurs a wage expenditure E (w T ). We maintain a few standard assumptions, including (1) the expected revenue depends only on the agent s actions, (2) the lowest action cost is equal to zero, and (3) the cheapest way to compensate the agent is to pay him a fixed wage. With a slight abuse of notation, we will write E(u) as the minimum wage expenditure that confers a utility u to the agent, whose reservation utility at the outset is normalized to zero. We will assume that E(u) is continuous in u. Monitoring technology Define the monitoring technology by the collection of signal processes generated by all T -instant action profiles. For any given action profile a T, let X t be a signal defined on a probability space (S T, Σ T, P T ( a t )), where probability measure P T ( a T ) can depend on the first t actions a t = (a 1,, a t ). Further, let the ( signal process X T = (X 1,, X T ) be defined ) on the product probability T space S T S T, Σ T Σ T, P T ( a t ), and hence signals are serially t=1 independent once actions have been taken into account. The monitoring technology is players common knowledge, though this assumption will prove to be stronger than necessary. 2.3 Statistics Background McDiarmid s inequality Our analysis makes extensive use of McDiarmid s concentration inequality: Lemma 1 (McDiarmid (1989)). Let Z 1,, Z T be independent random variables taking values in a set S. Further, let ϕ be a test statistic defined on Z 1,, Z T 5

6 satisfying ϕ (z1,, z t, z T ) ϕ (z 1,, z t,, z T ) γt, t, z 1,, z t, z t S. Then for all ɛ > 0, and where P {ϕ Eϕ ɛ} α m (ɛ, ϕ), P { ϕ Eϕ ɛ} 2α m (ɛ, ϕ), ( ) α m (ɛ, ϕ) = exp 2ɛ2. T t=1 γ2 t In words, McDiarmid (1989) says that if a test statistic ϕ defined on T independent (but not necessarily identical) random variables has bounded differences γ 1,, γ T, then the probability that it is ɛ-concentrated (resp. ɛ-semi-concentrated) around the mean is bounded below by 1 2α m (ɛ, ϕ) (resp. 1 α m (ɛ, ϕ)) for all ɛ > 0. The term α m (ɛ, ϕ) is hereafter named McDiarmid s concentration bound, or concentration bound for short. Notice that α m (ɛ, ϕ) depends only on ɛ and the Lipschitz coefficient T t=1 γ2 t of ϕ but nothing else. This property, often referred to as the universality of the concentration bound by the literature, plays a crucial role in the upcoming analysis. Facts Take any δ (0, 1) and r > 0, and let = 1/T. The next two test statistics: ϕ 1,T = 1 δ 1 δ T T δ t 1 Z t t=1 and ϕ 2,T = 1 e r 1 e r serve as the basis for the upcoming analysis. T e rt Z t, Lemma 2. Let Z 1,, Z T be independent random variables taking values in a bounded set S R. Then for all ɛ > 0, t=1 6

7 (i) ( α m (ɛ, ϕ 1,T ) = exp (1 + δ)(1 ) δt ) (1 δ)(1 + δ T ) 2ɛ 2, S 2 where S sup s,s S s s ; (ii) when T is large, ( α m (ɛ, ϕ 2,T ) exp 2T ) ɛ2 g(r), S 2 where g(r) 2(1 e r ) 2 r(1 e 2r ). Proof. Omitted. 2.4 Main Assumptions We make three important assumptions. First, in each economy T, there exists a pure action profile that maximizes the social surplus (hereafter, the target action profile): Assumption 1. In each economy T, there exists a pure action profile a T that solves max σ (R T +) E [V (a T ) σ] E (E [C (a T ) σ]). Under Assumption 1, let VT, C T and S T denote the expected revenue, action cost and social surplus generated by the target action profile, respectively, and let η T = inf σ (R T + ) E [V (a T ) σ] V T be the ratio between the worst-case and target-level revenues. These quantities will prove to be useful soon. Second, the monitoring technology satisfies two basic conditions called measure concentration and informativeness: Assumption 2. There exist series of test statistics { ϕ T : ST T R} T =1 reals {ɛ T } T =1 such that (i) each ϕ T has bounded differences and α m (ɛ T, ϕ T ) is vanishing in T ; and positive 7

8 (ii) there exist vanishing series of positive reals {v T } T =1 and {c T } T =1 such that for each T and a T, V (a T ) VT (1 v T ) and C (a T ) CT (1 c T ) if E [ϕ T a T ] E [ϕ T a T ] 2ɛ T. In words, measure concentration (Part (i) of Assumption 2) ensures the existence of test statistics with vanishing concentration bounds, whereas informativeness (Part (ii) of Assumption 2) requires that actions generating similar mean test statistics to target ones be themselves near-efficient. For notation ease, we shall hereafter write the concentration bound as αt m whenever this causes no confusion. The last assumption imposes some regularities on our problem: Assumption 3. (i) V T, C T, S T Θ(1) as T ; (ii) lim T max {α m (ɛ T, ϕ T ), c T } η T = 0; (iii) the agent faces unlimited liability at instant T. Parts (i) and (ii) of Assumption 3 stipulate that the target-level revenue, action cost and social surplus be non-vanishing in T, and that the worst-case revenue should not decrease too fast as T grows to infinity. Part (iii) of Assumption 3 can be relaxed if the T -instant economy is part of an infinite-horizon model where the agent can be severely penalized by reductions in the future payoff instead (e.g., discrete-time model with patient players, discretized continuous-time model). Due to space limitations, we do not fully describe such model here, and refer interested readers to the online appendix for further details. 2.5 Examples Our framework nests (1) discrete-time agency models with patient players and (2) discretized continuous-time agency models. Below we give two illustrative examples along these lines. In each example, we first describe the model setup and then check the validities of our assumptions. Example 1. The horizon lasts T periods and players share a common discount factor δ (0, 1). In each period t, players observe the realization of the concurrent revenue X t, and their payoffs are given by u(w t ) c(a t ) and X t w t. We maintain the standard assumption that u > 0, u < 0, c > 0 and c > 0, while allowing the distribution 8

9 of X t to depend on the first t actions a t. The last assumption generalizes the main condition of Radner (1981) and Radner (1985), namely revenue can depend only on the concurrent action. In the current setting, the social surplus generated by an arbitrary T -period action profile a T is V (a T ) u 1 (C (a T )), where V (a T ) = 1 δ T 1 δ T t=1 δt 1 E [X t a t ] and C (a T ) = 1 δ T 1 δ T t=1 δt 1 c(a t ). Since the function u 1 (C) is continuous and convex, a sufficient condition for Assumption 1 to hold true is that V is continuous and concave. ) To verify the validity of Assumption 2, take ϕ T = ϕ 1,T and ɛ T Θ (T 12 +ξ, where ξ can be any number in ( 0, 1 2). Notice the following: 1. Since the revenue space S is bounded and independent of T, a straightforward application of Lemma 2 shows that lim T lim δ 1 α m (ɛ T, ϕ 1,T ) = Informativeness follows from the fact that the mean test statistic equals the expected revenue. Formally, since E [ϕ 1,T a T ] = V (a T ) for all a T, it follows that E [ϕ 1,T a T ] E [ϕ 1,T a T ] 2ɛ T holds true if and only if V (a T ) V T 2ɛ T. Under the assumption that a T is efficient, any action profile a T satisfying E [ϕ 1,T a T ] E [ϕ 1,T a T ] 2ɛ T must also satisfy C (a T ) u (u 1 (C T ) 2ɛ T ), because otherwise there is a way of generating more than the full surplus, a contradiction. Putting these results together, we see that v T = 2ɛ T and c T = u (u 1 (C T )) C T 2ɛ T, and that lim T v T = lim T c T = It is easy to construct examples that satisfy Assumption 3, e.g., let E [X t a t ] = a t + ρa t ρ t 1 a 1 for some ρ [0, 1) and u be unbounded below. Example 2. Time evolves continuously over [0, 1] and players share a common interest rate r > 0. The economy indexed by T divides [0, 1] into T subintervals of length = 1/T, with the action, signal and wage over [(t 1), t ] being given by a t, X t and w t, respectively. Throughout [0, 1], the agent s utility is 1 e r T 1 e r t=1 e rt [u(w t ) c(a t )] and the principal s profit 1 e r T 1 e r t=1 e rt [a t w t ]. For now, suppose signals are generated by a Poisson process with arrival rate t s=1 ρt s a t over [(t 1), t ]. In particular, ρ can be any number in [0, 1), with ρ = 0 representing the case of no persistence as studied by Myerson (2010). Let Z t be the jump size over [(t 1), t ] and let X t = Z t 1 Zt 1 ρz t 1 1 Zt 1 1. Notice that X t takes values in S = { ρ, 0, 1 ρ, 1}, and that the mean of X t equals approximately the expected revenue over [(t 1), t ], i.e., E [X t a t ] a t, when T is large. 9

10 A casual inspection reveals the isomorphism between the current example and Example 1, namely, 1. Assumption 1 holds true for the same reason as given in Example 1. ) 2. Take ϕ T = ϕ 2,T and ɛ T Θ (T 12 +ξ, where ξ can be any number in ( 0, 2) 1. Since S is bounded and independent of T, it follows from Lemma 2 (ii) that lim T α m (ɛ T, ϕ 2,T ) = 0. Further, the very fact that the mean statistic equals the expected revenue, i.e., V (a T ) = 1 er T 1 e r t=1 e rt a t = E [ϕ 2,T a T ], implies informativeness, for the same reason as given in Example 1. 3 Main Result In this section, we demonstrate how we can achieve asymptotic efficiency through the use of test contract, which consists of (1) a test statistic ϕ T, (2) a threshold E [ϕ T a T ] ɛ T and (3) a pair of wages w T and w T. The contract pays the agent a fixed wage w T throughout t = 1,, T 1. At t = T, if the sample statistic exceeds the threshold, then the agent passes the test and earns the same wage w T as before. Otherwise he fails the test and earns w T instead. The penalty for failure amounts to U (w T,, w T ) U (w T,, w T ) = λc T, where λ can be any number that is greater than one and independent of T. Under any test contract, the agent s decision rule is a collection of state-continent plans σ T = { σ t,t : A t 1 S t 1 T (A) }, where each σ t,t prescribes the instant-t action choice based on the history of actions and signals. Define F = {ϕ T < E [ϕ T a T ] ɛ T } as the event where the agent fails the test. Any optimal decision rule σ T solves min σ T E [C (a T ) + 1 F λc T σ T ], and the contract induces voluntary participation at the outset if U (w T,, w T ) E [C (a T ) + 1 F λc T σ T ] 0. 10

11 Theorem 1. Take any series of economies that satisfies Assumptions 1-3 for each T, as well as any series of test contracts that satisfies the above described properties. Then for each T N, (i) the probability that the agent fails the test satisfies P (F σ T ) 3λαm T + c T λ 1 + c T ; (ii) the ratio between the expected profit generated by the test contract and the full surplus is bounded below by 1 S T As T, { [ ( )] } 3λα VT m 1 v T (1 v T η T ) T + c T E ((1 + λαt m ) CT ). λ 1 + c T (a) P (F σ T ) O (max {αm T, c T }); (b) the ratio in Part (ii) converges to one. Proof. See the Appendix. Theorem 1 prescribes a lower bound for the test contract s profitability in each economy T and shows that this lower bound converges to the full surplus as T grows to infinity. This result holds true independently of the exact details of the signal process and even if signals can depend on past actions. A key step in the proof is to bound the probability that the agent passes the test by sheer luck, as well as his gain from carefully adjusting action choices as information gets revealed over time. Our starting observation is that since the monitoring technology exhibits serial independence, it follows that under any optimal decision rule, signals convey no information about each other and thus satisfy the preconditions of McDiarmid s concentration inequality for any given profile of realized actions. Further, since the concentration bound holds uniformly over all action profiles, it suffices to focus on the high probability event where the sample test statistic is concentrated around its mean when T is large. Meanwhile, informativeness stipulates that in this high probability event, the agent s actions be near-efficient in order to pass the test. This suggests that a big penalty for failure suffice to induce near-efficient actions most of the time, from which all the other results will follow. 11

12 4 Applications Generalization of existing models Theorem 1 suggests that we can attain asymptotic efficiency in (1) the discrete-time model described in Example 1 as players become arbitrarily patient, as well as (2) the discretized continuous-time model with Poisson signals as described in Example 2. The first result generalizes Radner s (1985) Folk Theorem by allowing revenues to depend on past actions. The second result is alluded to by Myerson (2010), which examines the case of no persistence and binary actions, and uses the assumption that high action must be induced all the time to rule out the use of the test contract. 1 A new model Theorem 1 enables us to identify new signal processes that yield asymptotic efficiency in discretized continuous-time models: Example 2 (Continued). In Example 2, now suppose, instead, that signals follow a Gamma process: over each interval [(t 1), t ], jumps (denoted by Z t ) of size in [x, x + dx] arrive according to a Poisson process with intensity ν t (x)dx, where x 0 and ν t (x) = ( t s=1 ρt s a t ) x 1 exp( λx) for some arbitrary ρ [0, 1) and λ > 0. Let κ = 1 0 exp( λx)dx and X t = κ 1 ( Z t 1 Zt 1 ρz t 1 1 Zt 1 1). Notice that X t takes values in a set S = [ κ 1 ρ, κ 1 ] that is bounded and independent of T, and that the expected value of X t equals approximately the expected revenue over [(t 1), ) t ], i.e., E [X t a t ] a t, when T is large. Take ϕ T = ϕ 2,T and ɛ T Θ (T 12 +ξ, where ξ can be any number in ( 0, 2) 1. Results so far suggest that we satisfy both measure concentration and informativeness, and can thus attain asymptotic efficiency when T is large. A negative result The analysis so far assumes the existence of signals that equates the mean test statistic with the expected revenue. The next example, which is adapted from Sadzik and Stacchetti (2015), shows that this assumption itself is not enough to ensure asymptotic efficiency: Example 2 (Continued). In Example 2, now suppose, instead, that the revenue over [(t 1), t ] is X t, where X t N (a t, σ 2 T ). To see that this revenue process cannot 1 The same assumption appears in other dynamic contracting models with Poisson signals, e.g., Biais et al. (2010). However, these models typically involve addition ingredients that make the comparison between our model less straightforward. 12

13 satisfy both measure concentration and informativeness, notice that if the contrary were true, then we could use the test contract as a building block to construct a near-efficient contract in an infinite-horizon model where the agent is protected by limited liability and is penalized by changes in the future payoff beyond instant T (see the online appendix for further details). But this contradicts with the result of Sadzik and Stacchetti (2015), namely the principal s profit is strictly bounded away from the full surplus in the above described environment. To gain intuition into this negative result, take, for example, ϕ T = ϕ 2,T, and notice that ϕ T N (V (a T ), σ 2 /g(r)) when T is large (see Lemma 2 for the definition of g(r)). Thus for any series of positive reals {ɛ T } T =1, we have v T = 2ɛ T and c T = u (u 1 (CT)) 2ɛ CT T, and hence a sufficient and necessary condition for informativeness to hold true is lim T ɛ T = 0. But then we cannot satisfy measure concentration, because lim T P { ϕ T E [ϕ T a T ] < ɛ T } = 0. What is going on here? As T, the discounted sum of revenues converges in law to a Brownian motion with drift. Since Brownian motion tends to drift apart from its mean, we must leave some slackness in the performance threshold in order to satisfy measure concentration. But this comprises informativeness, which requires that we keep tightening the performance threshold as T grows to infinity. 5 Conclusion We conclude by posing open questions for future research. First, it will be interesting to carry out similar exercises in dynamic games with imperfect monitoring. Second, the result on discretized continuous-time models suggests that the principal prefer Poisson or Gamma signals to Brownian motion if she can choose between the various kinds of monitoring technologies. Likewise, Poisson or Gamma signals should be traded at a premium compared to Brownian motion if monitoring technologies can be bought at a price. The empirical merits of these predictions await a careful test. A Proof of Theorem 1 Part (i): We proceed in three steps: Step 1. Since the monitoring technology exhibits serial independence, it follows 13

14 that under any optimal decision rule σt, past signals affect future signals only through future action choices. Thus, for any given profile of realized actions a T, signals convey no information about each other and hence satisfy the precondition of McDiarmid s concentration inequality. Further, since McDiarmid s concentration bound αt m holds uniformly over all possible a T s, it follows that under σt, the probability that the sample test statistic is ɛ T -concentrated around its mean under the true action profile is bounded below by 1 2α m T. Formally, define G = { ϕ T E [ϕ T a T ] < ɛ T } as the event where the sample statistic is ɛ T -concentrated around its mean under true actions. The above discussion suggests that P (G σ T ) = E [E [1 G a T ] σ T ] 1 2α m T. Step 2. Recall the definition of event F: F = {ϕ T < E [ϕ T a T ] ɛ T }, and define π T = P (F G σ T ) as the probability of the event where the sample test statistic is ɛ T -concentrated around its mean under true actions and yet the agent fails the test. Then, P (F c G σ T ) = 1 P (G c σ T ) P (F G σ T ) 1 2α m T π T. Throughout t = 1,, T, the agent s total loss, which includes the action cost and the penalty for failure, is bounded below by π T λct + E [1 }{{} F c G C (a T ) σt ] +P (G c σt ) }{{}}{{} 0. (a) (b) (c) In particular, (a) and (c) constitute lower bounds for the loss in event F G 14

15 and G c, respectively. Further, (b) can be bounded below as follows: E [1 F c G C (a T ) σt ] = E [E [1 F c G C (a T ) a T ] σt ] E [E [1 F c G CT (1 c T ) a T ] σt ] = P (F c G σt ) CT (1 c T ) (1 2αT m π T ) CT (1 c T ), where the first inequality can be obtained by noticing that in event F c G, the very fact that (1) the test statistic is ɛ T -concentrated around its mean, i.e., ϕ T E [ϕ T a T ] < ɛ T, and that (2) the agent passes the test, i.e., ϕ T E [ϕ T a T ] ɛ T, implies ϕ T E [ϕ T a T ] 2ɛ T and hence C (a T ) CT (1 c T ) by Assumption 2 (ii). Putting these results together, we obtain the following lower bound for the agent s loss under σt : π T λc T + (1 2α m T π T ) C T (1 c T ). (A.1) Step 3. If the agent takes a T instead, then his total loss is bounded above by C T + α m T λc T. (A.2) Since (A.1) (A.2), it follows that π T λc T + (1 2α m T π T ) C T (1 c T ) (1 + λα m T ) C T. Dividing both sides of the above inequality by C T and rearranging, we obtain: π T (λ + 2)αm T + (1 2αm T )c T λ 1 + c T. Thus, P (F σ T ) P (F G σ T ) + P (G c σ T ) π T + 2α m T 3λαm T + c T λ 1 + c T, and this completes the proof of Part (i). 15

16 Part (ii): The agent s expected payoff under σt is bounded below by U (w T,, w T ) (1 + λα m T ) C T, suggesting that the following wage expenditure suffices to make him accept the contract at the outset: E ((1 + λα m T ) C T ). Meanwhile, the expected revenue generated by σ T is bounded below by P (F G c σt ) VT η }{{} T + E [1 F c G V (a T ) σt ]. }{{} (a) (b) In particular, (a) π T + 2αT m, and (b) can be bounded below as follows: E [1 F c G V (a T ) σ T ] = E [E [1 F c G V (a T ) a T ] σ T ] E [E [1 F c G V T (1 v T ) a T ] σ T ] = P (F c G σ T ) V T (1 v T ) (1 2α m T π T ) V T (1 v T ), with the first inequality being implied by the fact that in event F c G, combining (1) ϕ T E [ϕ T a T ] < ɛ T and (2) ϕ T E [ϕ T a T ] ɛ T yields ϕ T E [ϕ T a T ] 2ɛ T and hence V (a T ) V T (1 v T ) by Assumption 2 (ii). Together, these results imply the following upper bound for the expected profit: V T [ ( )] 3λα m 1 v T (1 v T η T ) T + c T E ((1 + λαt m ) CT ), λ 1 + c T and this completes the proof of Part (ii). Parts (a) and (b): Under Assumptions 2 and 3, letting T in the above derivation gives the desired result. 16

17 B Online Appendix This section extends the analysis so far to infinite-horizon models. To avoid repetition, we only consider discretized continuous-time models, and leave it to the reader to verify that the result below carries over to discrete-time models with patient players. Time evolves continuously over [0, + ). Over each interval [(t 1), t ], t = 1, 2,, the agent takes a hidden action a t 0, a signal X t is publicly realized, and the principal pays a non-negative wage w t 0. Throughout [0, + ), the agent s utility is r t=1 e rt (u(w t ) c(a t )), and the principal s expected profit r t=1 e rt (a t w t ). We maintain the standard assumptions that u(0) = 0, u > 0, u < 0 and that c(0) = 0, c > 0 and c > 0. We define the efficient action a by the unique interior solution to max a R+ a u 1 (c(a)). Fix any l > 0, ɛ > 0 and λ > 1 independent of T such that the following condition holds true: l λc (a ) e rs ds < ε 0 l e rs ds. Divide [0, + ) into review cycles of length l, i.e., [(k 1)l, kl], k = 1, 2,. Suppose that actions affect only signals of the concurrent review cycle. Suppose, in addition, that for each review cycle k, there exist series of test statistics {ϕ k,t } T =1 and positive reals {ɛ k,t } T =1 that satisfy Assumption 2. Consider the following generalization of the test contract: at t = (k 1)l, if the relationship survives, then the principal pays a flow wage u 1 (c (a ) + ε) throughout review cycle k; at t = kl, if ϕ k,t exceeds E [ϕ k,t a ] ɛ k,t, then the agent passes the test and the relationship continues; otherwise he fails the test and is fired with probability λc(a ) l ε l 0 e rs ds e rs ds after which both players earn zero reservation utilities in the remainder of their lives., Corollary 1. Suppose the environment satisfies the above descriptions. T, Then as (i) the principal s expected profit converges to a u 1 (c(a )) O(ε); (ii) over each review cycle, the agent s expected payoff equals approximately ε, and the probability that he fails the test is vanishing in T. 17

18 Proof. Suppose the agent adopts the optimal decision rule prescribed by Theorem 1 in each review cycle. Then the penalty he pays for failing the test equals approximately the following term when T is large: ( λc(a ) l ε l 0 e rs ds e rs ds ) ε l l e rs ds = λc(a ) e rs ds. 0 The remainder of the proof follows closely that of Theorem 1 and thus is omitted. References Al-Najjar, N. I. and R. Smorodinsky (2001): Large Nonanonymous Repeated Games, Games and Economic Behavior, 37(1), Blackwell (1956): An Analog of the Minimax Theorem for Vector Payoffs, Pacific Journal of Mathematics, 6, 1-8. Biais, B., T. Mariotti, J. C. Rochet and S. Villeneuve (2010): Large Risks, Limited Liability, and Dynamic Moral Hazard, Econometrica, 78(1), Boucheron, S., G. Lugosi and P. Massart (2013): Concentration Inequalities: A Nonasymptotic Theory of Independence, Oxford, United Kingdom: Oxford Unviersity Press. Chassang, S. (2013): Calibrated Incentive Contracts, Econometrica, 81(5), Deb, J., and E. Kalai (2015): Stability in Large Bayesian Games with Heterogeneous Players, Journal of Economic Theory, 157, Hellwig, M. F., and K. M. Schmidt (2002): Discrete-Time Approximations of the Holmstrom-Milgrom Brownian-Motion Model of Inter-temporal Incentive Provision, Econometrica, 70(6), Holmstrom, B., and P. Milgrom (1987): Aggregation and Linearity in the Provision of Intertemporal Incentives, Econometrica, 55(2),

19 Jackson, M., and B. Golub (2012): How Homophily Affects the Speed of Learning and Best-Response Dynamics, The Quarterly Journal of Economics, 127(3), Kalai, E. (2004): Large Robust Games, Econometrica, 72(6), McDiarmid, C. (1989): On the Methods of Bounded Differences, in Surveys in Combinatorics, ed. by J. Siemons, London Mathematical Society Lecture Note Series 141, Cambridge: Cambridge University Press. Myerson, R. B. (2010): Leadership, Trust, and Power: Dynamic Moral Hazard in High Office, Working Paper, University of Chicago. Radner, R. (1981): Monitoring Cooperative Agreements in a Repeated Principal- Agent Relationship, Econometrica, 49(5), (1985): Repeated Principal-Agent Games with Discounting, Econometrica, 53(5), Rubinstein, A., and M. E. Yaari. (1983): Repeated Insurance Contracts and Moral Hazard, Journal of Economic Theory, 30(1), Sadzik, T., and E. Stacchetti (2015): Agency Models with Frequent Actions: a Quadratic Approximation Method, Econometrica, 83, Sannikov, Y. (2008): A Continuous-Time Version of the Principal-Agent Problem, Review of Economic Studies, 75(3),

Robust Incentive Contract with Disagreement over Performance Evaluation and Compensation

Robust Incentive Contract with Disagreement over Performance Evaluation and Compensation Robust Incentive Contract with Disagreement over Performance Evaluation and Compensation Anqi Li First Draft: August 2013 his Draft: November 2017 Abstract We examine a dynamic agency model where the agent

More information

Contracting with Disagreement about Performance Evaluation and Compensation

Contracting with Disagreement about Performance Evaluation and Compensation Contracting with Disagreement about Performance Evaluation and Compensation Anqi Li Department of Economics Washington University in St. Louis March 2015 Motivation Organizational frictions from disagreements

More information

Lecture Notes - Dynamic Moral Hazard

Lecture Notes - Dynamic Moral Hazard Lecture Notes - Dynamic Moral Hazard Simon Board and Moritz Meyer-ter-Vehn October 27, 2011 1 Marginal Cost of Providing Utility is Martingale (Rogerson 85) 1.1 Setup Two periods, no discounting Actions

More information

Combinatorial Agency of Threshold Functions

Combinatorial Agency of Threshold Functions Combinatorial Agency of Threshold Functions Shaili Jain 1 and David C. Parkes 2 1 Yale University, New Haven, CT shaili.jain@yale.edu 2 Harvard University, Cambridge, MA parkes@eecs.harvard.edu Abstract.

More information

Deceptive Advertising with Rational Buyers

Deceptive Advertising with Rational Buyers Deceptive Advertising with Rational Buyers September 6, 016 ONLINE APPENDIX In this Appendix we present in full additional results and extensions which are only mentioned in the paper. In the exposition

More information

An Introduction to Moral Hazard in Continuous Time

An Introduction to Moral Hazard in Continuous Time An Introduction to Moral Hazard in Continuous Time Columbia University, NY Chairs Days: Insurance, Actuarial Science, Data and Models, June 12th, 2018 Outline 1 2 Intuition and verification 2BSDEs 3 Control

More information

Lecture Notes - Dynamic Moral Hazard

Lecture Notes - Dynamic Moral Hazard Lecture Notes - Dynamic Moral Hazard Simon Board and Moritz Meyer-ter-Vehn October 23, 2012 1 Dynamic Moral Hazard E ects Consumption smoothing Statistical inference More strategies Renegotiation Non-separable

More information

The Revenue Equivalence Theorem 1

The Revenue Equivalence Theorem 1 John Nachbar Washington University May 2, 2017 The Revenue Equivalence Theorem 1 1 Introduction. The Revenue Equivalence Theorem gives conditions under which some very different auctions generate the same

More information

Appendix of Homophily in Peer Groups The Costly Information Case

Appendix of Homophily in Peer Groups The Costly Information Case Appendix of Homophily in Peer Groups The Costly Information Case Mariagiovanna Baccara Leeat Yariv August 19, 2012 1 Introduction In this Appendix we study the information sharing application analyzed

More information

EC476 Contracts and Organizations, Part III: Lecture 2

EC476 Contracts and Organizations, Part III: Lecture 2 EC476 Contracts and Organizations, Part III: Lecture 2 Leonardo Felli 32L.G.06 19 January 2015 Moral Hazard: Consider the contractual relationship between two agents (a principal and an agent) The principal

More information

Impatience vs. Incentives

Impatience vs. Incentives Impatience vs. Incentives Marcus Opp John Zhu University of California, Berkeley (Haas) & University of Pennsylvania, Wharton January 2015 Opp, Zhu (UC, Wharton) Impatience vs. Incentives January 2015

More information

Test Contract. Anqi Li. February Abstract. I examine a dynamic agency model with imperfect public monitoring where

Test Contract. Anqi Li. February Abstract. I examine a dynamic agency model with imperfect public monitoring where est Contract Anqi Li February 204 Abstract I examine a dynamic agency model with imperfect public monitoring where the outcome process depends arbitrarily on past actions and exhibits moderate serial correlation.

More information

Mechanism Design: Basic Concepts

Mechanism Design: Basic Concepts Advanced Microeconomic Theory: Economics 521b Spring 2011 Juuso Välimäki Mechanism Design: Basic Concepts The setup is similar to that of a Bayesian game. The ingredients are: 1. Set of players, i {1,

More information

Online Appendix for Dynamic Procurement under Uncertainty: Optimal Design and Implications for Incomplete Contracts

Online Appendix for Dynamic Procurement under Uncertainty: Optimal Design and Implications for Incomplete Contracts Online Appendix for Dynamic Procurement under Uncertainty: Optimal Design and Implications for Incomplete Contracts By Malin Arve and David Martimort I. Concavity and Implementability Conditions In this

More information

Microeconomic Theory (501b) Problem Set 10. Auctions and Moral Hazard Suggested Solution: Tibor Heumann

Microeconomic Theory (501b) Problem Set 10. Auctions and Moral Hazard Suggested Solution: Tibor Heumann Dirk Bergemann Department of Economics Yale University Microeconomic Theory (50b) Problem Set 0. Auctions and Moral Hazard Suggested Solution: Tibor Heumann 4/5/4 This problem set is due on Tuesday, 4//4..

More information

Optimal Incentive Contract with Costly and Flexible Monitoring

Optimal Incentive Contract with Costly and Flexible Monitoring Optimal Incentive Contract with Costly and Flexible Monitoring Anqi Li 1 Ming Yang 2 1 Department of Economics, Washington University in St. Louis 2 Fuqua School of Business, Duke University May 2016 Motivation

More information

A New Class of Non Existence Examples for the Moral Hazard Problem

A New Class of Non Existence Examples for the Moral Hazard Problem A New Class of Non Existence Examples for the Moral Hazard Problem Sofia Moroni and Jeroen Swinkels April, 23 Abstract We provide a class of counter-examples to existence in a simple moral hazard problem

More information

Bayesian Persuasion Online Appendix

Bayesian Persuasion Online Appendix Bayesian Persuasion Online Appendix Emir Kamenica and Matthew Gentzkow University of Chicago June 2010 1 Persuasion mechanisms In this paper we study a particular game where Sender chooses a signal π whose

More information

Screening. Diego Moreno Universidad Carlos III de Madrid. Diego Moreno () Screening 1 / 1

Screening. Diego Moreno Universidad Carlos III de Madrid. Diego Moreno () Screening 1 / 1 Screening Diego Moreno Universidad Carlos III de Madrid Diego Moreno () Screening 1 / 1 The Agency Problem with Adverse Selection A risk neutral principal wants to o er a menu of contracts to be o ered

More information

Contracts in informed-principal problems with moral hazard

Contracts in informed-principal problems with moral hazard Contracts in informed-principal problems with moral hazard Nicholas C Bedard January 20, 2016 Abstract In many cases, an employer has private information about the potential productivity of a worker, who

More information

Oblivious Equilibrium: A Mean Field Approximation for Large-Scale Dynamic Games

Oblivious Equilibrium: A Mean Field Approximation for Large-Scale Dynamic Games Oblivious Equilibrium: A Mean Field Approximation for Large-Scale Dynamic Games Gabriel Y. Weintraub, Lanier Benkard, and Benjamin Van Roy Stanford University {gweintra,lanierb,bvr}@stanford.edu Abstract

More information

Measuring the informativeness of economic actions and. market prices 1. Philip Bond, University of Washington. September 2014

Measuring the informativeness of economic actions and. market prices 1. Philip Bond, University of Washington. September 2014 Measuring the informativeness of economic actions and market prices 1 Philip Bond, University of Washington September 2014 1 I thank Raj Singh for some very constructive conversations, along with a seminar

More information

NOTES ON CALCULUS OF VARIATIONS. September 13, 2012

NOTES ON CALCULUS OF VARIATIONS. September 13, 2012 NOTES ON CALCULUS OF VARIATIONS JON JOHNSEN September 13, 212 1. The basic problem In Calculus of Variations one is given a fixed C 2 -function F (t, x, u), where F is defined for t [, t 1 ] and x, u R,

More information

Price and Capacity Competition

Price and Capacity Competition Price and Capacity Competition Daron Acemoglu, Kostas Bimpikis, and Asuman Ozdaglar October 9, 2007 Abstract We study the efficiency of oligopoly equilibria in a model where firms compete over capacities

More information

Agency Models with Frequent Actions: A Quadratic Approximation Method

Agency Models with Frequent Actions: A Quadratic Approximation Method Agency Models with Frequent Actions: A Quadratic Approximation Method Tomasz Sadzik, NYU Ennio Stacchetti, NYU April 19, 2012 Abstract The paper analyzes dynamic principal-agent models with short period

More information

What happens when there are many agents? Threre are two problems:

What happens when there are many agents? Threre are two problems: Moral Hazard in Teams What happens when there are many agents? Threre are two problems: i) If many agents produce a joint output x, how does one assign the output? There is a free rider problem here as

More information

Minimum Wages and Excessive E ort Supply

Minimum Wages and Excessive E ort Supply Minimum Wages and Excessive E ort Supply Matthias Kräkel y Anja Schöttner z Abstract It is well-known that, in static models, minimum wages generate positive worker rents and, consequently, ine ciently

More information

Moral Hazard: Hidden Action

Moral Hazard: Hidden Action Moral Hazard: Hidden Action Part of these Notes were taken (almost literally) from Rasmusen, 2007 UIB Course 2013-14 (UIB) MH-Hidden Actions Course 2013-14 1 / 29 A Principal-agent Model. The Production

More information

Online Appendix for. Breakthroughs, Deadlines, and Self-Reported Progress: Contracting for Multistage Projects. American Economic Review, forthcoming

Online Appendix for. Breakthroughs, Deadlines, and Self-Reported Progress: Contracting for Multistage Projects. American Economic Review, forthcoming Online Appendix for Breakthroughs, Deadlines, and Self-Reported Progress: Contracting for Multistage Projects American Economic Review, forthcoming by Brett Green and Curtis R. Taylor Overview This supplemental

More information

The Folk Theorem for Finitely Repeated Games with Mixed Strategies

The Folk Theorem for Finitely Repeated Games with Mixed Strategies The Folk Theorem for Finitely Repeated Games with Mixed Strategies Olivier Gossner February 1994 Revised Version Abstract This paper proves a Folk Theorem for finitely repeated games with mixed strategies.

More information

A New and Robust Subgame Perfect Equilibrium in a model of Triadic Power Relations *

A New and Robust Subgame Perfect Equilibrium in a model of Triadic Power Relations * A New and Robust Subgame Perfect Equilibrium in a model of Triadic Power Relations * by Magnus Hatlebakk ** Department of Economics, University of Bergen Abstract: We present a new subgame perfect equilibrium

More information

NBER WORKING PAPER SERIES PRICE AND CAPACITY COMPETITION. Daron Acemoglu Kostas Bimpikis Asuman Ozdaglar

NBER WORKING PAPER SERIES PRICE AND CAPACITY COMPETITION. Daron Acemoglu Kostas Bimpikis Asuman Ozdaglar NBER WORKING PAPER SERIES PRICE AND CAPACITY COMPETITION Daron Acemoglu Kostas Bimpikis Asuman Ozdaglar Working Paper 12804 http://www.nber.org/papers/w12804 NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts

More information

Asymmetric Information and Search Frictions: A Neutrality Result

Asymmetric Information and Search Frictions: A Neutrality Result Asymmetric Information and Search Frictions: A Neutrality Result Neel Rao University at Buffalo, SUNY August 26, 2016 Abstract This paper integrates asymmetric information between firms into a canonical

More information

University of Warwick, EC9A0 Maths for Economists Lecture Notes 10: Dynamic Programming

University of Warwick, EC9A0 Maths for Economists Lecture Notes 10: Dynamic Programming University of Warwick, EC9A0 Maths for Economists 1 of 63 University of Warwick, EC9A0 Maths for Economists Lecture Notes 10: Dynamic Programming Peter J. Hammond Autumn 2013, revised 2014 University of

More information

Multi-armed bandit models: a tutorial

Multi-armed bandit models: a tutorial Multi-armed bandit models: a tutorial CERMICS seminar, March 30th, 2016 Multi-Armed Bandit model: general setting K arms: for a {1,..., K}, (X a,t ) t N is a stochastic process. (unknown distributions)

More information

Economic Growth: Lecture 8, Overlapping Generations

Economic Growth: Lecture 8, Overlapping Generations 14.452 Economic Growth: Lecture 8, Overlapping Generations Daron Acemoglu MIT November 20, 2018 Daron Acemoglu (MIT) Economic Growth Lecture 8 November 20, 2018 1 / 46 Growth with Overlapping Generations

More information

u( x) = g( y) ds y ( 1 ) U solves u = 0 in U; u = 0 on U. ( 3)

u( x) = g( y) ds y ( 1 ) U solves u = 0 in U; u = 0 on U. ( 3) M ath 5 2 7 Fall 2 0 0 9 L ecture 4 ( S ep. 6, 2 0 0 9 ) Properties and Estimates of Laplace s and Poisson s Equations In our last lecture we derived the formulas for the solutions of Poisson s equation

More information

Efficient Random Assignment with Cardinal and Ordinal Preferences: Online Appendix

Efficient Random Assignment with Cardinal and Ordinal Preferences: Online Appendix Efficient Random Assignment with Cardinal and Ordinal Preferences: Online Appendix James C. D. Fisher December 11, 2018 1 1 Introduction This document collects several results, which supplement those in

More information

Moral Hazard. EC202 Lectures XV & XVI. Francesco Nava. February London School of Economics. Nava (LSE) EC202 Lectures XV & XVI Feb / 19

Moral Hazard. EC202 Lectures XV & XVI. Francesco Nava. February London School of Economics. Nava (LSE) EC202 Lectures XV & XVI Feb / 19 Moral Hazard EC202 Lectures XV & XVI Francesco Nava London School of Economics February 2011 Nava (LSE) EC202 Lectures XV & XVI Feb 2011 1 / 19 Summary Hidden Action Problem aka: 1 Moral Hazard Problem

More information

Online Appendices for Large Matching Markets: Risk, Unraveling, and Conflation

Online Appendices for Large Matching Markets: Risk, Unraveling, and Conflation Online Appendices for Large Matching Markets: Risk, Unraveling, and Conflation Aaron L. Bodoh-Creed - Cornell University A Online Appendix: Strategic Convergence In section 4 we described the matching

More information

Concentration behavior of the penalized least squares estimator

Concentration behavior of the penalized least squares estimator Concentration behavior of the penalized least squares estimator Penalized least squares behavior arxiv:1511.08698v2 [math.st] 19 Oct 2016 Alan Muro and Sara van de Geer {muro,geer}@stat.math.ethz.ch Seminar

More information

Optimal Contract to Induce Continued Effort

Optimal Contract to Induce Continued Effort Optimal Contract to Induce Continued Effort Peng Sun Duke University, psun@duke.edu Feng Tian University of Michigan, ftor@umich.edu We consider a basic model of a risk-neutral principal incentivizing

More information

Ramsey Cass Koopmans Model (1): Setup of the Model and Competitive Equilibrium Path

Ramsey Cass Koopmans Model (1): Setup of the Model and Competitive Equilibrium Path Ramsey Cass Koopmans Model (1): Setup of the Model and Competitive Equilibrium Path Ryoji Ohdoi Dept. of Industrial Engineering and Economics, Tokyo Tech This lecture note is mainly based on Ch. 8 of Acemoglu

More information

The Mathematics of Continuous Time Contract Theory

The Mathematics of Continuous Time Contract Theory The Mathematics of Continuous Time Contract Theory Ecole Polytechnique, France University of Michigan, April 3, 2018 Outline Introduction to moral hazard 1 Introduction to moral hazard 2 3 General formulation

More information

Appendix B for The Evolution of Strategic Sophistication (Intended for Online Publication)

Appendix B for The Evolution of Strategic Sophistication (Intended for Online Publication) Appendix B for The Evolution of Strategic Sophistication (Intended for Online Publication) Nikolaus Robalino and Arthur Robson Appendix B: Proof of Theorem 2 This appendix contains the proof of Theorem

More information

The deterministic Lasso

The deterministic Lasso The deterministic Lasso Sara van de Geer Seminar für Statistik, ETH Zürich Abstract We study high-dimensional generalized linear models and empirical risk minimization using the Lasso An oracle inequality

More information

Bandit models: a tutorial

Bandit models: a tutorial Gdt COS, December 3rd, 2015 Multi-Armed Bandit model: general setting K arms: for a {1,..., K}, (X a,t ) t N is a stochastic process. (unknown distributions) Bandit game: a each round t, an agent chooses

More information

Area I: Contract Theory Question (Econ 206)

Area I: Contract Theory Question (Econ 206) Theory Field Exam Summer 2011 Instructions You must complete two of the four areas (the areas being (I) contract theory, (II) game theory A, (III) game theory B, and (IV) psychology & economics). Be sure

More information

Ph.D. Preliminary Examination MICROECONOMIC THEORY Applied Economics Graduate Program May 2012

Ph.D. Preliminary Examination MICROECONOMIC THEORY Applied Economics Graduate Program May 2012 Ph.D. Preliminary Examination MICROECONOMIC THEORY Applied Economics Graduate Program May 2012 The time limit for this exam is 4 hours. It has four sections. Each section includes two questions. You are

More information

Worst case analysis for a general class of on-line lot-sizing heuristics

Worst case analysis for a general class of on-line lot-sizing heuristics Worst case analysis for a general class of on-line lot-sizing heuristics Wilco van den Heuvel a, Albert P.M. Wagelmans a a Econometric Institute and Erasmus Research Institute of Management, Erasmus University

More information

Mostly calibrated. Yossi Feinberg Nicolas S. Lambert

Mostly calibrated. Yossi Feinberg Nicolas S. Lambert Int J Game Theory (2015) 44:153 163 DOI 10.1007/s00182-014-0423-0 Mostly calibrated Yossi Feinberg Nicolas S. Lambert Accepted: 27 March 2014 / Published online: 16 April 2014 Springer-Verlag Berlin Heidelberg

More information

Are Obstinacy and Threat of Leaving the Bargaining Table Wise Tactics in Negotiations?

Are Obstinacy and Threat of Leaving the Bargaining Table Wise Tactics in Negotiations? Are Obstinacy and Threat of Leaving the Bargaining Table Wise Tactics in Negotiations? Selçuk Özyurt Sabancı University Very early draft. Please do not circulate or cite. Abstract Tactics that bargainers

More information

problem. max Both k (0) and h (0) are given at time 0. (a) Write down the Hamilton-Jacobi-Bellman (HJB) Equation in the dynamic programming

problem. max Both k (0) and h (0) are given at time 0. (a) Write down the Hamilton-Jacobi-Bellman (HJB) Equation in the dynamic programming 1. Endogenous Growth with Human Capital Consider the following endogenous growth model with both physical capital (k (t)) and human capital (h (t)) in continuous time. The representative household solves

More information

1. The General Linear-Quadratic Framework

1. The General Linear-Quadratic Framework ECO 37 Economics of Uncertainty Fall Term 009 Notes for lectures Incentives for Effort - Multi-Dimensional Cases Here we consider moral hazard problems in the principal-agent framewor, restricting the

More information

x ax 1 2 bx2 a bx =0 x = a b. Hence, a consumer s willingness-to-pay as a function of liters on sale, 1 2 a 2 2b, if l> a. (1)

x ax 1 2 bx2 a bx =0 x = a b. Hence, a consumer s willingness-to-pay as a function of liters on sale, 1 2 a 2 2b, if l> a. (1) Answers to Exam Economics 201b First Half 1. (a) Observe, first, that no consumer ever wishes to consume more than 3/2 liters (i.e., 1.5 liters). To see this, observe that, even if the beverage were free,

More information

Monotonic ɛ-equilibria in strongly symmetric games

Monotonic ɛ-equilibria in strongly symmetric games Monotonic ɛ-equilibria in strongly symmetric games Shiran Rachmilevitch April 22, 2016 Abstract ɛ-equilibrium allows for worse actions to be played with higher probability than better actions. I introduce

More information

Implementability, Walrasian Equilibria, and Efficient Matchings

Implementability, Walrasian Equilibria, and Efficient Matchings Implementability, Walrasian Equilibria, and Efficient Matchings Piotr Dworczak and Anthony Lee Zhang Abstract In general screening problems, implementable allocation rules correspond exactly to Walrasian

More information

Ambiguity and Information Processing in a Model of Intermediary Asset Pricing

Ambiguity and Information Processing in a Model of Intermediary Asset Pricing Ambiguity and Information Processing in a Model of Intermediary Asset Pricing Leyla Jianyu Han 1 Kenneth Kasa 2 Yulei Luo 1 1 The University of Hong Kong 2 Simon Fraser University December 15, 218 1 /

More information

CS261: A Second Course in Algorithms Lecture #11: Online Learning and the Multiplicative Weights Algorithm

CS261: A Second Course in Algorithms Lecture #11: Online Learning and the Multiplicative Weights Algorithm CS61: A Second Course in Algorithms Lecture #11: Online Learning and the Multiplicative Weights Algorithm Tim Roughgarden February 9, 016 1 Online Algorithms This lecture begins the third module of the

More information

Price and Capacity Competition

Price and Capacity Competition Price and Capacity Competition Daron Acemoglu a, Kostas Bimpikis b Asuman Ozdaglar c a Department of Economics, MIT, Cambridge, MA b Operations Research Center, MIT, Cambridge, MA c Department of Electrical

More information

This is designed for one 75-minute lecture using Games and Information. October 3, 2006

This is designed for one 75-minute lecture using Games and Information. October 3, 2006 This is designed for one 75-minute lecture using Games and Information. October 3, 2006 1 7 Moral Hazard: Hidden Actions PRINCIPAL-AGENT MODELS The principal (or uninformed player) is the player who has

More information

Advanced Macroeconomics

Advanced Macroeconomics Advanced Macroeconomics The Ramsey Model Marcin Kolasa Warsaw School of Economics Marcin Kolasa (WSE) Ad. Macro - Ramsey model 1 / 30 Introduction Authors: Frank Ramsey (1928), David Cass (1965) and Tjalling

More information

Robustness of Equilibria in Anonymous Local Games

Robustness of Equilibria in Anonymous Local Games Robustness of Equilibria in Anonymous Local Games Willemien Kets October 12, 2010 Abstract This paper studies the robustness of symmetric equilibria in anonymous local games to perturbations of prior beliefs.

More information

Simple Consumption / Savings Problems (based on Ljungqvist & Sargent, Ch 16, 17) Jonathan Heathcote. updated, March The household s problem X

Simple Consumption / Savings Problems (based on Ljungqvist & Sargent, Ch 16, 17) Jonathan Heathcote. updated, March The household s problem X Simple Consumption / Savings Problems (based on Ljungqvist & Sargent, Ch 16, 17) subject to for all t Jonathan Heathcote updated, March 2006 1. The household s problem max E β t u (c t ) t=0 c t + a t+1

More information

Economics 201B Economic Theory (Spring 2017) Bargaining. Topics: the axiomatic approach (OR 15) and the strategic approach (OR 7).

Economics 201B Economic Theory (Spring 2017) Bargaining. Topics: the axiomatic approach (OR 15) and the strategic approach (OR 7). Economics 201B Economic Theory (Spring 2017) Bargaining Topics: the axiomatic approach (OR 15) and the strategic approach (OR 7). The axiomatic approach (OR 15) Nash s (1950) work is the starting point

More information

Existence and monotonicity of solutions to moral hazard problems

Existence and monotonicity of solutions to moral hazard problems Existence and monotonicity of solutions to moral hazard problems G. Carlier Université Paris Dauphine, CEREMADE, UMR CNRS 7534, Place du Maréchal De Lattre De Tassigny 75775 PARIS CEDEX 16 R.-A. Dana Université

More information

1. Introduction. 2. A Simple Model

1. Introduction. 2. A Simple Model . Introduction In the last years, evolutionary-game theory has provided robust solutions to the problem of selection among multiple Nash equilibria in symmetric coordination games (Samuelson, 997). More

More information

Pricing and Capacity Allocation for Shared Services: Technical Online Appendix Vasiliki Kostami Dimitris Kostamis Serhan Ziya. λ 1 + λ 2 K.

Pricing and Capacity Allocation for Shared Services: Technical Online Appendix Vasiliki Kostami Dimitris Kostamis Serhan Ziya. λ 1 + λ 2 K. Appendix Proof of Proposition 1 Pricing and Capacity Allocation for Shared Services: Technical Online Appendix Vasiliki Kostami Dimitris Kostamis Serhan Ziya The prices (p 1, p ) are simultaneously announced.

More information

Coordination and Continuous Choice

Coordination and Continuous Choice Coordination and Continuous Choice Stephen Morris and Ming Yang Princeton University and Duke University December 2016 Abstract We study a coordination game where players choose what information to acquire

More information

Moral Hazard. Felix Munoz-Garcia. Advanced Microeconomics II - Washington State University

Moral Hazard. Felix Munoz-Garcia. Advanced Microeconomics II - Washington State University Moral Hazard Felix Munoz-Garcia Advanced Microeconomics II - Washington State University Moral Hazard Reading materials: Start with Prajit Dutta, Chapter 19. MWG, Chapter 14 Macho-Stadler and Perez-Castrillo,

More information

Introduction. 1 University of Pennsylvania, Wharton Finance Department, Steinberg Hall-Dietrich Hall, 3620

Introduction. 1 University of Pennsylvania, Wharton Finance Department, Steinberg Hall-Dietrich Hall, 3620 May 16, 2006 Philip Bond 1 Are cheap talk and hard evidence both needed in the courtroom? Abstract: In a recent paper, Bull and Watson (2004) present a formal model of verifiability in which cheap messages

More information

Banks, depositors and liquidity shocks: long term vs short term interest rates in a model of adverse selection

Banks, depositors and liquidity shocks: long term vs short term interest rates in a model of adverse selection Banks, depositors and liquidity shocks: long term vs short term interest rates in a model of adverse selection Geethanjali Selvaretnam Abstract This model takes into consideration the fact that depositors

More information

Strategic Properties of Heterogeneous Serial Cost Sharing

Strategic Properties of Heterogeneous Serial Cost Sharing Strategic Properties of Heterogeneous Serial Cost Sharing Eric J. Friedman Department of Economics, Rutgers University New Brunswick, NJ 08903. January 27, 2000 Abstract We show that serial cost sharing

More information

Game Theory and Economics of Contracts Lecture 5 Static Single-agent Moral Hazard Model

Game Theory and Economics of Contracts Lecture 5 Static Single-agent Moral Hazard Model Game Theory and Economics of Contracts Lecture 5 Static Single-agent Moral Hazard Model Yu (Larry) Chen School of Economics, Nanjing University Fall 2015 Principal-Agent Relationship Principal-agent relationship

More information

Entropic Selection of Nash Equilibrium

Entropic Selection of Nash Equilibrium Entropic Selection of Nash Equilibrium Zeynel Harun Alioğulları Mehmet Barlo February, 2012 Abstract This study argues that Nash equilibria with less variations in players best responses are more appealing.

More information

Moral Hazard and Persistence

Moral Hazard and Persistence Moral Hazard and Persistence Hugo Hopenhayn Department of Economics UCLA Arantxa Jarque Department of Economics U. of Alicante PRELIMINARY AND INCOMPLETE Abstract We study a multiperiod principal-agent

More information

Near-Potential Games: Geometry and Dynamics

Near-Potential Games: Geometry and Dynamics Near-Potential Games: Geometry and Dynamics Ozan Candogan, Asuman Ozdaglar and Pablo A. Parrilo September 6, 2011 Abstract Potential games are a special class of games for which many adaptive user dynamics

More information

On Robust Arm-Acquiring Bandit Problems

On Robust Arm-Acquiring Bandit Problems On Robust Arm-Acquiring Bandit Problems Shiqing Yu Faculty Mentor: Xiang Yu July 20, 2014 Abstract In the classical multi-armed bandit problem, at each stage, the player has to choose one from N given

More information

Preliminary Results on Social Learning with Partial Observations

Preliminary Results on Social Learning with Partial Observations Preliminary Results on Social Learning with Partial Observations Ilan Lobel, Daron Acemoglu, Munther Dahleh and Asuman Ozdaglar ABSTRACT We study a model of social learning with partial observations from

More information

6.254 : Game Theory with Engineering Applications Lecture 7: Supermodular Games

6.254 : Game Theory with Engineering Applications Lecture 7: Supermodular Games 6.254 : Game Theory with Engineering Applications Lecture 7: Asu Ozdaglar MIT February 25, 2010 1 Introduction Outline Uniqueness of a Pure Nash Equilibrium for Continuous Games Reading: Rosen J.B., Existence

More information

SUPPLEMENT TO ROBUSTNESS AND SEPARATION IN MULTIDIMENSIONAL SCREENING (Econometrica, Vol. 85, No. 2, March 2017, )

SUPPLEMENT TO ROBUSTNESS AND SEPARATION IN MULTIDIMENSIONAL SCREENING (Econometrica, Vol. 85, No. 2, March 2017, ) Econometrica Supplementary Material SUPPLEMENT TO ROBUSTNESS AND SEPARATION IN MULTIDIMENSIONAL SCREENING Econometrica, Vol. 85, No. 2, March 2017, 453 488) BY GABRIEL CARROLL This supplement contains

More information

UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS

UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS Home exam: ECON 5200 / ECON 9200 Exam period: 5.12.2011, at 09:00 9.12.2011, at 12:30 (unless other individual deadline granted) Grades are given: 3.01.2012.

More information

Moral hazard in teams

Moral hazard in teams Division of the Humanities and Social Sciences Moral hazard in teams KC Border November 2004 These notes are based on the first part of Moral hazard in teams by Bengt Holmström [1], and fills in the gaps

More information

Bayesian Learning in Social Networks

Bayesian Learning in Social Networks Bayesian Learning in Social Networks Asu Ozdaglar Joint work with Daron Acemoglu, Munther Dahleh, Ilan Lobel Department of Electrical Engineering and Computer Science, Department of Economics, Operations

More information

Deterministic Calibration and Nash Equilibrium

Deterministic Calibration and Nash Equilibrium University of Pennsylvania ScholarlyCommons Statistics Papers Wharton Faculty Research 2-2008 Deterministic Calibration and Nash Equilibrium Sham M. Kakade Dean P. Foster University of Pennsylvania Follow

More information

Random Extensive Form Games and its Application to Bargaining

Random Extensive Form Games and its Application to Bargaining Random Extensive Form Games and its Application to Bargaining arxiv:1509.02337v1 [cs.gt] 8 Sep 2015 Itai Arieli, Yakov Babichenko October 9, 2018 Abstract We consider two-player random extensive form games

More information

On Reputation with Imperfect Monitoring

On Reputation with Imperfect Monitoring On Reputation with Imperfect Monitoring M. W. Cripps, G. Mailath, L. Samuelson UCL, Northwestern, Pennsylvania, Yale Theory Workshop Reputation Effects or Equilibrium Robustness Reputation Effects: Kreps,

More information

Organization, Careers and Incentives

Organization, Careers and Incentives Organization, Careers and Incentives Chapter 4 Robert Gary-Bobo March 2018 1 / 31 Introduction Introduction A firm is a pyramid of opportunities (Alfred P. Sloan). Promotions can be used to create incentives.

More information

Mathematical Appendix. Ramsey Pricing

Mathematical Appendix. Ramsey Pricing Mathematical Appendix Ramsey Pricing PROOF OF THEOREM : I maximize social welfare V subject to π > K. The Lagrangian is V + κπ K the associated first-order conditions are that for each I + κ P I C I cn

More information

General idea. Firms can use competition between agents for. We mainly focus on incentives. 1 incentive and. 2 selection purposes 3 / 101

General idea. Firms can use competition between agents for. We mainly focus on incentives. 1 incentive and. 2 selection purposes 3 / 101 3 Tournaments 3.1 Motivation General idea Firms can use competition between agents for 1 incentive and 2 selection purposes We mainly focus on incentives 3 / 101 Main characteristics Agents fulll similar

More information

SURPLUS SHARING WITH A TWO-STAGE MECHANISM. By Todd R. Kaplan and David Wettstein 1. Ben-Gurion University of the Negev, Israel. 1.

SURPLUS SHARING WITH A TWO-STAGE MECHANISM. By Todd R. Kaplan and David Wettstein 1. Ben-Gurion University of the Negev, Israel. 1. INTERNATIONAL ECONOMIC REVIEW Vol. 41, No. 2, May 2000 SURPLUS SHARING WITH A TWO-STAGE MECHANISM By Todd R. Kaplan and David Wettstein 1 Ben-Gurion University of the Negev, Israel In this article we consider

More information

5. Relational Contracts and Career Concerns

5. Relational Contracts and Career Concerns 5. Relational Contracts and Career Concerns Klaus M. Schmidt LMU Munich Contract Theory, Summer 2010 Klaus M. Schmidt (LMU Munich) 5. Relational Contracts and Career Concerns Contract Theory, Summer 2010

More information

Moral Hazard in Teams

Moral Hazard in Teams Moral Hazard in Teams Ram Singh Department of Economics September 23, 2009 Ram Singh (Delhi School of Economics) Moral Hazard September 23, 2009 1 / 30 Outline 1 Moral Hazard in Teams: Model 2 Unobservable

More information

Math 104: Homework 7 solutions

Math 104: Homework 7 solutions Math 04: Homework 7 solutions. (a) The derivative of f () = is f () = 2 which is unbounded as 0. Since f () is continuous on [0, ], it is uniformly continous on this interval by Theorem 9.2. Hence for

More information

Strongly Consistent Self-Confirming Equilibrium

Strongly Consistent Self-Confirming Equilibrium Strongly Consistent Self-Confirming Equilibrium YUICHIRO KAMADA 1 Department of Economics, Harvard University, Cambridge, MA 02138 Abstract Fudenberg and Levine (1993a) introduce the notion of self-confirming

More information

Dynamic Mechanisms without Money

Dynamic Mechanisms without Money Dynamic Mechanisms without Money Yingni Guo, 1 Johannes Hörner 2 1 Northwestern 2 Yale July 11, 2015 Features - Agent sole able to evaluate the (changing) state of the world. 2 / 1 Features - Agent sole

More information

Patience and Ultimatum in Bargaining

Patience and Ultimatum in Bargaining Patience and Ultimatum in Bargaining Björn Segendorff Department of Economics Stockholm School of Economics PO Box 6501 SE-113 83STOCKHOLM SWEDEN SSE/EFI Working Paper Series in Economics and Finance No

More information

Labor Economics, Lecture 11: Partial Equilibrium Sequential Search

Labor Economics, Lecture 11: Partial Equilibrium Sequential Search Labor Economics, 14.661. Lecture 11: Partial Equilibrium Sequential Search Daron Acemoglu MIT December 6, 2011. Daron Acemoglu (MIT) Sequential Search December 6, 2011. 1 / 43 Introduction Introduction

More information

Game Theory Fall 2003

Game Theory Fall 2003 Game Theory Fall 2003 Problem Set 1 [1] In this problem (see FT Ex. 1.1) you are asked to play with arbitrary 2 2 games just to get used to the idea of equilibrium computation. Specifically, consider the

More information

On continuous time contract theory

On continuous time contract theory Ecole Polytechnique, France Journée de rentrée du CMAP, 3 octobre, 218 Outline 1 2 Semimartingale measures on the canonical space Random horizon 2nd order backward SDEs (Static) Principal-Agent Problem

More information