The Price of Anarchy Shan Li Justin Azadivar Price of Anarchy-the Idea Motivation: equilibria of noncooperative games typically inefficient Prisoner’s Dilemma ¾ Congestion of the highway ¾ Price of anarchy: quantify inefficiency w.r.t some objective function Example: Braess’s Paradox Setup: traffic (e.g.,many cars or packets) pick s-t paths to minimize travel time Travel time increases with congestion C(x) = x C(x) = 1 t s C(x) = 1 C(x) = x Example: Braess’s Paradox C(x) = x ½ of traffic C(x) = 1 t s C(x) = 1 C(x) = x ½ of traffic At “equilibrium”, travel time = 3/2 for all Example: Braess’s Paradox Initial Network: x 1/2 1/2 1 t s 1 1/2 Travel time = 1.5 1/2 x Example: Braess’s Paradox Augmented Network: x s 1 1 t 0 Travel time = 2 All traffic worse off! x The Definition Price of anarchy: worst-case ratio between “social cost” of equilibrium and of optimum Definition: Price of Anarchy of a game Obj fn value of a Nash equilibrium = ---minimization: ---maximization: Optimal obj fn value >=1 <=1 Pigou’s Example 1 unit of traffic travels from s to t c(x) = x s Cost depends on congestion t c(x) = 1 No congestion effects Note: top edge is a dominant strategy ¾ ¾ ¾ In equilibrium: all traffic on top edge All traffic incurs cost 1 Due to [Pigou 1920] Could we do better? Consider instead : traffic split equally c(x) = x s t c(x) = 1 Flow = 1/2 Flow = 1/2 Cost of Nash flow = 1*1+0*1 = 1 Cost of optimal flow = ½*½ + ½*1 = 3/4 Price of anarchy = 4/3 xd Nonlinear Pigou’s Example Bad Example: xd 1 (d large) 1-e s t 0 1 e Equilibrium has cost 1, min cost 0 Nash flow can cost arbitrarily more than the optimal (min-cost) flow ---even if cost functions are polynomials Resource Allocation Games Based on model of [Kelly 97] Players submit bids, price-based mechanism responds by allocating capacity to players Kelly’s assumption: players are “price-takers”, don’t anticipate how their bid affects price computed by mechanism --standard assumption when many players Thm: selfish behavior allocation optimal Resource Allocation Games [Johari / Tsitsiklis 04]: but what if players are price-anticipating? -- selfish behavior no longer achieves optimal solution Main 4/3 result: price of anarchy is exactly Supply Chain Games Quantifying the efficiency of a decentralized system vs. the performance of a centralized system Inefficiencies arise from decentralizing operations (bullwhip effect, double marginalization), competition, and information asymmetries. Supply Chain Games The Price of Anarchy in Supply Chains: Quantifying the Efficiency of Price-Only Contracts [Gergia Perakis , Guillaume Roels MIT, 2005] Supply Chain Games Price-only Contracts: A buyer and a seller agree only on a constant transaction price Without specifying the amount that will be transferred Supply Chain Games Price-only Contracts: It is well known that it does not provide incentives to the parties to coordinate their inventory/capacity decisions. The level of inventory in the decentralized supply chain is less than the optimal inventory level that would be held if the supply chain was integrated. Supply Chain Games Example of Loss of efficiency due to decentralization ---double marginalization Limited performance of price-only contracts ¾Retailer bears demand risk vs. Supplier ¾Assembly system ¾Stochastic demand vs. Deterministic demand Supply Chain Games Improve coordination in supply chain Buy-back Revenue sharing Quantity flexibility Sales rebate Quantity discount contracts Supply Chain Games Back to price-only contract Important to quantify the loss of efficiency Price of Anarchy Model Framework Consider a supply chain facing the newsvendor problem. The supply chain has to build its inventory Q before observing the demand realization Model Framework We assume: Costs are linear c per-unit purchasing cost p per-unit selling cost Salvage value is zero (What if nonzero salvage cost?) Model Framework We assume (cont.) Demand D is random and has a cumulative distribution function F(x) that is strictly increasing and differentiable, with probability density function f(x). Model Framework We assume (cont.) The demand distribution has an increasing generalized failure rate (IGFR) h(x) = f (x) / F (x) is the hazard rate g ( x ) = xh ( x ) is the generalized failure rate The IGFR assumption is sufficient to guarantee a well-behaved (concave) problem for the contract initiator in a decentralized setting Model Framework Centralized Supply Chain max pE[min{Q, D}] − cQ Q≥0 It is well known that it is newsvendor problem – concave – has unique optimal solution c Q = F (c / p ) Model Framework Decentralized Supply Chain ( Price of Anarchy) Since local optimization is always dominated by global optimization, decentralizing operations usually results in a loss of efficiency Model Framework Defination: The Price of Anarchy (PoA) is the worstcase ratio of the profit achieved with the centralized supply chain over the profit achieved with the decentralized supply chain − cQ c + pE[min{Q c , D}] PoA = max F ∈Φ − cQ d + pE[min{Q d , D}] Where Φ is the set of nonnegative demand distribution with IGFR property Model Framework Model Framework Model Framework Pull Mode: The upstream partner bears the whole risk of the supply chain Push Mode: The downstream partner bears the whole risk of the supply chain Model Framework 1. The leader offers the follower a contract specifying the per-unit whole sale price w 2. The follower accept the contract if his expected profit is above his reservation profit, assumed to be zero. Otw, no transaction between parties. 3. The manufacturer chooses his/her level of inventory at a per-unit cost c Model Framework Pull: 4. Demand D is realized. 5. The retailer orders what is needed to meet demand, at a per-unit cost w. Each unit of satisfied demand generates a revenue p. If the manufacturer does not have enough inventory, the excess demand is lost at no cost. Model Framework Push: 4. The retailer places an order, at a per-unit cost w. 5. Demand D is realized. Each unit of satisfied demand generates a revenue p. If the manufacturer does not have enough inventory, the excess demand is lost at no cost. Serial Push Supply Chain The inventory is held at the retailer’s site. ie. the retailer orders before observing the demand Consider two Stackelberg games. Serial Push Supply Chain Manufacturer is the Leader The incentive compatibility (I.C.) constraint states that the retailer maximizes his expected profit The individual rationality (I.R.) constraint ensures that the retailer earns at least his reservation profit. Serial Push Supply Chain Theorem: In a two-stage push supply chain, when the manufacturer is the initiator of the price-only contract, Serial Push Supply Chain Retailer is the Leader It is easy to see that the retailer proposes a price w=c. Therefore, full coordination is achieved, i.e.PoA = 1. Serial Pull Supply Chain The inventory is held at the manufacturer’s site. ie. the retailer orders after observing the demand Consider two Stackelberg games. Homework 1) 2) Consider the Serial Pull Supply Chain structure, derive the optimization formulation when Retailer is the leader Manufacturer is the leader