The Market Power Function Theoretical results Experimental results Role of Network and Production Capacity in Allocating Market Power Jiangzhuo Chen Matthew Macauley Achla Marathe Network Dynamics and Simulation Science Laboratory Virginia Bioinformatics Institute Virginia Polytechnic Institute and State University Trans-Atlantic INFRADAY University of Maryland, College Park November 2, 2007 The Market Power Function Theoretical results Experimental results Outline 1 The Market Power Function Definitions Networks Market power spectrum 2 Theoretical results Local capacity changes Threshold capacities Global capacity changes 3 Experimental results Set-up Local capacity changes Global capacity changes Elasticity The Market Power Function Theoretical results Experimental results Definitions Networks Market power spectrum Definitions: Market power function ◮ Setting: ς distinct sellers, distributed across a network, all selling a uniform commodity. ◮ Production capacity vector: c = (c1 , . . . , cς ). Definition (Market power) Seller i has Pi (c) units of market power if the total amount of the demand that can be fulfilled ∆(c), decreases by MPi (c) in the absence of supplier i . ◮ The market power function is defined by mp : Rς+ −→ [0, 1]ς , mpi : Rς+ −→ [0, 1], mpi (c) = Pi (c) . ∆(c) ◮ The global market power function is the sum of the individual market functions: MP : Rς+ −→ [0, 1], MP(c) = ς X i =1 mpi (c) = mp(c) · (1, 1, . . . , 1). The Market Power Function Theoretical results Experimental results Definitions Networks Market power spectrum Modeling a market over a network ◮ Let G = (V , E ) be an undirected graph with edge capacities C : E −→ R+ . ◮ There is a set of suppliers (or sellers, generators, etc) S ⊂ V , with capacity vector |S| c : S −→ R+ . |T | ◮ There is a P set of consumers T ⊂ V , with demand vector d : T −→ R+ , and total demand D = i ∈T di . ◮ The goods must be delivered from the sellers to the consumers across the network G . The Market Power Function Theoretical results Experimental results Definitions Networks Market power spectrum Market graphs ◮ Without loss of generality, we may assume that the graph G contains: a super-source s, and edges {s, si } for all si ∈ S, each with capacity ci . a super-sink t, and edges {ti , t} for all ti ∈ T , each with capacity di . ◮ Such a graph is called a market graph. ◮ Sending the goods from the suppliers to the consumers is modeled as an s-t network flow. The maximum that can be delivered is the max-flow, or min-cut of the network. Definition For a market graph G , the submarket graph G (S ′ ) of a subset of suppliers S ′ ⊂ S is constructed from G by setting the capacity of the edge {s, sj } = 0 for all sj 6∈ S ′ . ◮ Thus, G (S ′ ) is the market graph when all suppliers not in S ′ are off-line. ◮ Observe that the market power of seller i is: mpi (c) = mincut(G (S \ {si }), s, t) mincut(G , s, t) − mincut(G (S \ {si }), s, t) =1− . mincut(G , s, t) mincut(G , s, t) The Market Power Function Theoretical results Experimental results Definitions Networks Market power spectrum Resource constraints Example: Consider 3 sellers with capacity vector c = (9, 5, 3) and one consumer with demand 10 (and no network). This can be modeled by the following market graph: s1 9 s 5 3 9 s2 s3 5 t1 10 t 3 ◮ The market powers of the individual sellers are mp1 (c) = 2 , 10 mp2 (c) = 0, mp3 (c) = 0 . ◮ Observe that if sellers 1 and 2 merge into one entity, then together, they will have 7 units (or 70%) of market power. The behavior of market power under merging was the topic of an earlier paper. The Market Power Function Theoretical results Experimental results Definitions Networks Market power spectrum The market power spectrum ◮ Given a market graph G = (V , E ), with capacity vector c, consider two extreme cases: Scale the capacity vector kc, for k → 0. Scale the capacity vector kc, for k → ∞. ◮ As k → 0, then for small enough k, the edge capacities in the network will be irrelevant, and the market will be a purely resource-constrained market. ◮ As k → ∞, then for large enough k, the relative production capacities will be irrelevant, and the market will be a purely network-constrained market. ◮ The range of possible markets between these two extremes is called the market power spectrum. Question: How is the market power function vary across the market power spectrum? Where on the spectrum do certain real-world markets lie? The Market Power Function Theoretical results Experimental results Local capacity changes Threshold capacities Global capacity changes Local capacity changes ◮ Consider how the market power function changes as we change the capacity of a single generator. ◮ Let f ∗ (G , ci ) denote max-flow in the market graph G when capacity of generator i is ci . ∗ := f ∗ (G , c = 0), the max-flow when s drops out. ◮ Let f−i i i ◮ Using the theory of residual flows, we can derive: ( if ci ≤ ci∗ (G ), f ∗ + ci f ∗ (G , ci ) = −i ∗ ∗ f−i + ci (G ) otherwise. where ci∗ (G ) is the maximum possible flow that generator i can provide. The Market Power Function Theoretical results Experimental results Local capacity changes Threshold capacities Global capacity changes Theorem When ci increases, market power of si is non-decreasing; market power of sj , j 6= i , is non-increasing. Remark. The total market power MP(ci ) is non-decreasing in ci when ci ≤ ci∗ (G ), non-increasing when ci ≥ ci∗ (G ). ◮ In our experimental results with the Portland electrical network, the total market power is always non-increasing in ci . However, this need not always be true. Example. s1 1 s 3 3 s2 t1 3 10 t 3 3 s3 3 ◮ If c1 = 1, then mp1 = 0.25, mp2 = 0, mp3 = 0, and MP = 0.25. ◮ If c1 = 2, then mp1 = 0.40, mp2 = 0, mp3 = 0, and MP = 0.40. The Market Power Function Theoretical results Experimental results Local capacity changes Threshold capacities Global capacity changes Threshold capacities ◮ The local threshold capacity of generator i , denoted Tiℓ , is the minimum non-negative value such that if ci > Tiℓ , then ∇ei mpi = 0 ◮ The local threshold capacity of generator i , denoted Tiℓ , is the minimum non-negative value such that if ci > Tiℓ , then ∇ei MPi = 0 ◮ Mathematically, Til = sup{Supp (∇ei mpi ) ∪ {0} }, Tig = sup{Supp (∇ei MP)}, where sup is supremum of a set, and Supp is the support of a function. ◮ Note. ∇ei MP has a non-empty support, but ∇ei mpi may not. ´ ` Prop. If we define Til = sup{Supp ∇ei mpj ∪ {0}}, then for any i 6= j, Tiℓ ≤ Tij ≤ Tig . The Market Power Function Theoretical results Experimental results Local capacity changes Threshold capacities Global capacity changes Global capacity changes ◮ Consider the scenario when the capacity vector is scaled by k > 0. ◮ It is intuitive to conjecture that as k increases, the total market power should decrease, i.e., ∇c MP(c) ≤ 0. However, this is not always the case, as the following example shows. Example. s1 1 s 3 3 s2 t1 3 10 t 3 3 s3 3 ◮ If k = 1, then mp1 = 0.25, mp2 = 0, mp3 = 0, and MP = 0.25. ◮ If k = 2, then mp1 = 0.40, mp2 = 0, mp3 = 0, and MP = 0.40. The Market Power Function Theoretical results Experimental results Set-up Local capacity changes Global capacity changes Elasticity Experimental set-up ◮ Electrical grid of Portland, Oregon. ◮ Over 600 nodes and 700 transmission lines. ◮ 41 generators locations, but 11 have zero production capacity. ◮ Peak demand vector is taken from FERC, and rescaled by a factor of 0.337 so that total supply is 105% of the total demand. ◮ Only six generators have any market power, and the total market power is MP(c) = 0.398. ◮ Two sets of experiments: Scale the individual production capacity ci independently (local change). Scale the production capacity vector c (global change). The Market Power Function Theoretical results Experimental results Set-up Local capacity changes Global capacity changes Elasticity Summary of results Gen. ID Prod-cap (ci ) Deg-cap. mpi Max mpi Max MP Min MP Degree Til g Ti 17 608 688 0.145 0.145 1 0.270 1 451 688 Highest Market Power (mpi ) Generators 19 152 485 601 18 390 372 333 260 241 429 348 456 300 276 0.075 0.061 0.057 0.033 0.027 0.075 0.061 0.057 0.033 0.027 1 1 1 1 1 0.335 0.398 0.287 0.334 0.342 1 1 1 1 1 233 191 176 103 84 429 348 402 300 276 Highest Deg-cap Generators 83 55 356 100 89 32 15 2 4542 2075 2265 1162 0 0 0 0 0 0 0 0 0.669 0.486 0.434 0.402 0 0 0 0 11 3 5 2 0 0 0 0 540 483 466 453 Table: Characteristics of Select Generators (in Kw) The Market Power Function Theoretical results Experimental results Set-up Local capacity changes Global capacity changes Elasticity Remarks about the results For each seller, ci > Tiℓ , i.e., production capacity is greater than its local threshold capacity. 5 of the 6 sellers with market power have Tig equal to their “degree-capacity” (sum of capacities of all incident edges). For 9 of 10 sellers, ci < Tig . This suggests that the market is more resource-constrained than network constrained. For the 4 sellers with the highest degree-capacity, Til = 0. They have no market power, but can eliminate global market power completely by increasing their capacity. Tig > Til . A generator may be able to affect global market power without being able to affect its individual market power, but not vice-versa. The Market Power Function Theoretical results Experimental results Set-up Local capacity changes Global capacity changes Elasticity Individual market power function plots Figure: Individual market power (mpi ), as a function of ci . The Market Power Function Theoretical results Experimental results Set-up Local capacity changes Global capacity changes Elasticity Global market power with respect to local changes Figure: Total market power (MP), as a function of ci . The Market Power Function Theoretical results Experimental results Set-up Local capacity changes Global capacity changes Elasticity Global market power Figure: Total market power as a function of the supply scaling factor. Figure: Individual market power of the top 7 sellers, as a function of the supply scaling factor. The Market Power Function Theoretical results Experimental results Set-up Local capacity changes Global capacity changes Elasticity Remarks about the results For k > 1, MP monotonically decreases with k (this is in general not guaranteed!) ci For k < 1, the values of mpi quickly converge to P . Thus even the slightest cj excess demand in the market will make it predominately resource-constrained. j ∇c MP(kc) increases with k, i.e., MP is convex. A small increase in supply can wipe out a large amount of the market power. The graphs of mpi for sellers 152 and 485 intersect, an explicit example of how network topology can play a role in market power. Overall, the values of ∇ei mpi (kc) for the top seven generators are not significantly different, suggesting that the market is more resource-constrained rather than network-constrained. The Market Power Function Theoretical results Experimental results Set-up Local capacity changes Global capacity changes Elasticity Experiment ◮ Base case: scale demand so that supply = demand. ◮ Scale both supply and demand vector independently. (Recall: the base case in the previous experiments were when supply was 105% of demand, and global market power was 0.398.) Observations. Changing the ratio of supply to demand by 2% makes the value of k required to eliminate market power decrease by 10%. The market power function converges very quickly. In particular, the market power functions are virtually identical for the cases when supply is 130%, 150%, 200%, and 300% of demand. All functions take on a value of 1 for k < 1 but for k = 1.2, the total market power is less than 10% in every case. This implies that demand elasticity can play a significant role in mitigating market power. The Market Power Function Theoretical results Experimental results Set-up Local capacity changes Global capacity changes Elasticity Figure: Market power as a function of the supply scaling factor, for different initial ratios of supply and demand. The Market Power Function Theoretical results Experimental results Set-up Local capacity changes Global capacity changes Elasticity 3-D plot of the previous figure Figure: Market power as a function of supply and demand. The Market Power Function Theoretical results Experimental results Set-up Local capacity changes Global capacity changes Elasticity Acknowledgments Thank you for your attention! Special thanks: Network Dynamics and Simulation Science Laboratory (NDSSL), at the Virginia Bioinformatics Institute, at Virginia Tech. Los Alamos National Laboratory. Contact info: {chenj, mmacaul, amarathe}@vbi.vt.edu. NDSSL: Web: http://ndssl.vbi.vt.edu