PhD Course Voting and Political Debate Lecture 5 Francesco Squintani University of Warwick email: f.squintani@warwick.ac.uk 1 Strategic Communication (Crawford and Sobel, 1982) Model • An expert privately observes a state t ∈ [0, 1]. • The expert sends a report r ∈ ℜ (a real number) to a decision maker. • After receiving the report, the decision maker chooses an action y ∈ ℜ. • The DM knows that t is uniformly distributed in [0, 1]: the p.d.f. is f (t) = 1 for t ∈ [0, 1]. • The players payoffs are independent of r. (I.e., the report is cheap talk). 2 • Both players care that the action matches the state. But the expert is biased relatively to the DM. • The experts payoff is uE (y, t) = −(y − (t + b))2 . (Her optimal action given the state t is y = t + b). • The DM’s payoff is uDM (y, t) = −(y − t)2 . (Her optimal action given the state t is y = t). 3 Perfect Communication • I first show there cannot be a PBE in which the expert accurately reports the state. (I.e., an invertible strategy ρ cannot be part of a PBE.) • By contradiction, say that ρ is invertible and PBE. • Then, beliefs consistency implies that the DM believes (correctly) that the state is t, when the report is r = ρ(t). • Hence, upon receiving report r = ρ(t), the DM chooses the action y = t. • Anticipating this, when the state is t, the expert deviates from the strategy ρ. • Instead of sending the report r = ρ(t), she sends the report r = ρ(t + b), which induces the action y = t + b. • Hence, ρ is not sequentially rational, and cannot be part of a PBE. 4 Perfect Bayesian Equilibrium • By the above result, every PBE is described by a partition of the state space [0, 1]. • Each partition is composed of K intervals, where the admissible set of K depends on the bias level b. • The K intervals are [t0 , t1 ), [t1 , t2 ), ..., [tK−1 , tK ], where t0 = 0 and tK = 1. • For any k = 1, ..., K − 1, all expert types t ∈ [tk−1 , tk ) send the same report rk . • Hence, upon receiving report rk , the DM knows that t is in [tk−1 , tk ), and chooses y so as to maximize ∫ − tk tk−1 (y − t)2 dt. Pr{t ∈ [tk−1 , tk )} 5 • Taking the first-order condition, [ ] ∫ tk tk−1 + tk 0 = −2 (y − t)dt = −2 y − . 2 tk−1 • So, upon receiving report rk , the DM chooses the action y = (tk−1 + tk )/2. • In a PBE, the expert anticipates that if reporting r = rk , she induces the action y = (tk−1 + tk )/2. • For any k = 1, ..., K − 1, an expert of type tk must be indifferent between sending rk and rk+1 . [ ]2 [ ]2 tk−1 + tk tk + tk+1 I.e., − − (tk + b) = − − (tk + b) , 2 2 (1) or tk+1 − tk = tk − tk−1 + 4b. • If conditions (1) hold, then, for any k = 1, ..., K, every type of expert t ∈ [tk−1 , tk ) prefers to send report r = rk than any other report. 6 • The conditions (1) yield a second-order difference equation, solved with boundary conditions t0 = 0 and tK = 1. • Solving the difference equation, we obtain that the interval [tk−1 , tk ) has length tk − tk−1 = t1 + (k − 1)4b. • By the boundary conditions, the sum of the [tk−1 , tk ) for k = 1, ..., K, equals 1: t1 + (t1 + 4b) + ... + [t1 + (K − 1)4b] = 1, or i.e., Kt1 + 4b[1 + ... + (K − 1)b] = 1, Kt1 + 2bK(K − 1) = 1. (2) • So, if b is small enough that 2bK(K − 1) < 1, then there is a value of t1 > 0 that satisfies condition (2), and hence a PBE with K intervals. • For larger b, the maximum number of intervals K ∗ that can be a PBE is larger. Specifically, for b > 1/4, K ∗ = 1. For 1/4 < b < 1/12, K ∗ = 2, etc. 7 • I conclude by ranking the PBE in ex-ante sense (i.e., before the state t is realized). • The experts ex-ante equilibrium payoff is: EUE = − K ∑ ∫ Pr{t ∈ [tk−1 , tk )} tk tk−1 k=1 [(tk−1 + tk )/2 − (t + b)]2 dt. Pr[tk−1 , tk ) • The decision makers ex-ante equilibrium payoff is: ∫ tk K ∑ [(tk−1 + tk )/2 − t]2 EUDM = − Pr{t ∈ [tk−1 , tk )} dt. Pr[tk−1 , tk ) tk−1 k=1 • Expanding the experts ex-ante equilibrium payoff, K ∫ tk ∑ EUE = − {[(tk−1 + tk )/2 − t]2 + b2 − 2b[(tk−1 + tk )/2 − t]}dt. k=1 tk−1 8 • Because, for any k, ∫ tk [(tk−1 + tk )/2 − t]dt = (tk−1 + tk )/2 − (tk−1 + tk )/2 = 0, tk−1 we obtain: EUE = − K ∫ ∑ k=1 tk [(tk−1 + tk )/2 − t]2 dt − b2 . tk−1 • In sum, the experts and the decision makers ex-ante equilibrium payoff differ only by the constant b2 . • The PBE that maximizes a player’s payoff, maximizes also the other’s. • Further, it can be shown that, for any b, the PBE maximizing ex-ante payoff is the one with K ∗ intervals. • This is because the sum of the loss terms in the ex-ante payoff expression is larger, the coarser is the equilibrium partition. 9 Related Literature • Communication Games Theory: Sobel (1985), Aumann and Hart (2003), Forges (1986), Morgan and Stocken (2008), ... • Political Economy: Gilligan and Krehbiel (1987, 1989), Austen-Smith (1990), Morris (2001), Battaglini (2002), Horner, Morelli and Squintani (2015), ... Political Debate: Penn (2015), Patty (2015), Dewan, Galeotti, Ghiglino and Squintani (2015), Dewan and Squintani (2015), ... • Financial Advice: Benabou and Laroque (1992), Ottaviani and Sorensen (2006a, 2006b), ... • Behavioral Cheap Talk: Crawford (2003), Kartik, Ottaviani and Squintani (2007), Kartik (2009), ... 10 • Experiments: Dickhaut, McCabe and Mukherji (1995), Costa-Gomes, Crawford, Broseta (2001), Wang, Spezio and Camerer (2010), ... • Verifiable or Costly Talk: Milgrom (1981), Milgrom and Roberts (1986), Dewatripont and Tirole (2005), Kamenica and Gentzkow (2011), ... 11 Multi-player Communication (Galeotti, Ghiglino and Squintani) Model • State of nature θ uniformly distributed on [0, 1]. • N = {1, ..., n} players. • Each player i has bias bi ; b = {b1 , ..., bn }, b1 ≤ b2 ≤, ..., ≤ bn . • Each player i receives a private signal si ∈ {0, 1} about θ, with Pr(s = 1|θ) = θ. • The exogenous communication network g ∈ {0, 1} describes “who might talk to whom”: when gij = 1, i can send a message to j . • The communication neighborhood of i is Ni (g) = {j ∈ N : gij = 1}. • The audiences are a partition Ni (g) of Ni (g) : i must send the same message to all players in audience J, for all J ∈ Ni (g). 12 Model: First Stage • A pure communication strategy is mi (si ) = {miJ }J∈Ni (g) , when miJ (si ) = si , i communicates truthfully to audience J . • m = {m1 , ..., mn } communication strategy profile, where mi : {0, 1} → {0, 1}|Ni (g)| . • g and m define the equilibrium communication network c(m|g): cij (m|g) = 1 if and only if gij = 1 and mij (s) = s for every s ∈ {0, 1}. • Clearly, c(m|g) is a subgraph of g. 13 Model: Second Stage • After communication, each player i chooses an action ŷi ∈ [0, 1]. • An action strategy is yi : {0, 1} |Ni−1 (g)| × {0, 1} → [0, 1], where Ni−1 (g) is the set of players communicating with i. • y is the action strategy profile. • The payoffs of i facing y and θ is ui (y|θ) = − ∑ (yj − θ − bi )2 j∈N • Natural extension of Crawford and Sobel’s quadratic utility to a network. 14 Equilibrium and Welfare Concept • Our notion of efficiency is ex-ante Pareto optimality. • The equilibrium concept is pure-strategy Perfect Bayesian Equilibrium. • Given the received messages mNi−1 (g),i , by sequential rationality agent i chooses yi to maximize his expected utility, which yields: yi (si , mNi−1 (g),i ) = bi + E[θ|si , mNi−1 (g),i ] 15 Equilibrium: Private Communication • Given an equilibrium communication network c(m), let kj (m) be the in-degree of j, i.e., the number of agents who are truthful to j. Corollary 1. For any communication network g, a profile (m, y) is an equilibrium if and only if, whenever i is truthful to j, 1 |bi − bj | ≤ 2[kj (m) + 3] • i′ s capability to communicate is independent across players. • i’s capability to communicate with j declines with: – their bias difference and, – how many other players communicate truthfully with j . 16 Intuition Corollary 1 (the ‘equilibrium congestion effect’) • If i is truthful to j , then yj increases if mij = 1, while it decreases if mij = 0. • The effect of i’s message on j ’s action depends on how well informed j is: – it is large if j is poorly informed, – while it is small if j is well informed. • Now, suppose i has an incentive to bias j ’s action upward, i.e., bi > bj , and that i has received signal si = 0. • When j is well informed, if i mis-reports si = 0 and sends message mij = 1, then j ’s action increases only slightly and it gets closer to i’s bliss point bi + E[θ|Ωi ]. • So, i deviates from truthful communication, which cannot be part of an equilibrium. • When j is poorly informed, if i mis-reports si = 0 and sends mij = 1, then j ’s action increases substantially, possibly overshooting i’s bliss point. 17 Equilibrium: General Audiences • Consider a player i with communication neighborhood Ni (g). • The communication technology of i can be represented as a partition of his communication neighborhood, Ni (g). • Player i must send the same message to all j ∈ J , for each J ∈ Ni (g). • With private communication, the partition Ni (g) is composed of singleton sets. • The trivial partition Ni (g) = {Ni (g)} models public communication. 18 Theorem 1. Consider a communication network g and a collection of audiences {Ni (g)}i∈N . The strategy profile (m, y) is an equilibrium if and only if, whenever i is truthful to an audience J ∈ Ni (g) , ∑ b − b ∑ 1 j i . ≤ 2 kj (m) + 3 j∈J 2 (kj (m) + 3) j∈J • Player i will be willing to communicate if the average bias difference with players in the audience J is not too large. • Within the audience J, the relevance of each player j ’s bias difference bj − bi in the above average is inversely related to j ’s in-degree kj (m), i.e., to how informed j is. • We call these insights the ‘equilibrium average bias effect’. 19 Pareto Optimality • Given an equilibrium (m, y), let the proportion of players with in-degree k be denoted by P (k|m). Theorem 2. Consider communication networks g and g′ , with associated equilibria (m, y) and (m′ , y′ ) respectively. The equilibrium (m, y) Pareto dominates the equilibrium (m′ , y′ ) if and only if n−1 ∑ 1 P (k|m) < k+3 k=0 20 n−1 ∑ 1 . P (k|m ) k+3 k=0 ′ • We can therefore rank equilibria based on their distribution of truthful messages Corollary 2 (First Order Stochastic Shifts). If P (k|m) First Order Stochastically Dominates P (k|m′ ) then equilibrium (m, y) Pareto dominates equilibrium (m′ , y′ ) Corollary 3 (Mean Preserving Spread). If P (k|m′ ) is a Mean Preserving Spread of P (k|m) then (m, y) Pareto dominates equilibrium (m′ , y′ ) • Redistributing truthful reports evenly across players leads to a Pareto improvement. • We call this result the ‘welfare congestion effect’. 21 Related Literature • Multiple Senders - One Receiver: Battaglini (2001), Levy and Razin (2007), Ambrus and Takahashi (2008), Meyer, Moreno de Barreda and Nafziger (2014). • One Sender - Two Receivers: Farrell and Gibbons (1989). • Communication in Organizations: Wolinsky (2002), Dessein (2002), Alonso, Dessein and Matouscheck (2008), Migrow (2015). • Organization Design: Sah and Stiglitz (1986), Radner (1993), Bolton and Dewatripont (1994), Garicano (2000), Cremer Garicano and Prat (2007), ... • Network Economics: Bala and Goyal (2000), Jackson and Wolinsky (1996), Ballester, Calvo-Armengol and Zenou (2006), ... 22