Multi-unit multiple bid auctions in balancing markets: An agent-based Q-learning approach

Abstract

There is an ongoing debate on the appropriate auction design for competitive electricity balancing markets. Uniform (UPA) and discriminatory price auctions (DPA), the prevalent designs in use today, are assumed to have dierent properties with regard to prices and eciencies. These properties cannot be thoroughly described using analytical methods due to the complex strategy space in repeated multi-unit multiple bid auctions. Therefore, using an agent-based Q-learning model, we simulate the strategic bidding behaviour in these auctions under a variety of market conditions. We nd that UPAs lead to higher prices in all analysed market settings. This is mainly due to the fact that players engage in bid shading more aggressively. Moreover, small players in UPAs learn to free ride on the price setting of large players and earn higher prots per unit of capacity owned, while they are disadvantaged in DPAs. UPAs also generally feature higher eciencies, but there are exceptions to this observation. If demand is varying and players are provided with additional information about scarcity in the market, market prices increase only in case asymmetric players are present.

Computational Performance of Deep Reinforcement Learning to Find Nash Equilibria

“…Action and state space scope details how states and actions are modelled. Action spaces are either defined as (i) choosing predefined strategies that directly compute bid prices from averages, minima and maxima of the latest observed prices (Spooner et al, 2018;Lussange et al, 2021), or by (ii) picking from a discrete scale of prices (Aliabadi et al, 2017;Calvano et al, 2020;Viehmann et al, 2021;Asker et al, 2021). While (i) is a behavioral restriction and hence Bertrand-to monopoly prize (critic),…”

Machine learning applications for electricity market agent-based models: A systematic literature review

“…Viehmann et al [82] analyse the different markets: Uniform Pricing (UP) and Discriminatory Pricing. UP is where all agents pay the maximum accepted price, and DP pay their accepted bid.…”

“…Market Type Application Algorithm Used 2021 Bose S. [6] Local energy market Peer to peer trading Roth-Erev 2021 Naseri N. [55] Local energy market Tariff design SARIMAX, MDP 2021 Tang C. [78] International/National Tariff design Novel WoLF-PHC 2021 Liu D. [42] International/National Bidding strategies, Peerto-Peer MADDPG 2021 Deng C. [12] International/National Demand response DDPG 2021 Viehmann J. [82] International/National Market investigation Q-Learning 2020 Tomin N. [80] Microgrid Electricity grid control Q-Learning 2020 Liang Y. [39] International/National Bidding strategies DDPG 2020 Kim J.-G. [35] International/National Peer to peer trading Deep Q-Learning 2020 Shafie-Khah M. [73] Local energy market Demand response Q-Learning 2020 Sousa J.C. [76] International/National Bidding strategies Q-Learning 2020 Poplavskaya K. [64] International/National Bidding strategies Q-Learning 2020 Liu Y.…”

Strategy and design of multi‐unit auction in grouped asymmetric setting

“…The equilibrium analysis of the auction model is still the theoretical basis of most auction research. In terms of equilibrium solution, many scholars have made a series of attempts under different assumptions, such as two-bidder (Dobzinski & Nisan, 2015), three-bidder (Perry & Reny, 2005), two-item (Alvarez & Maz on, 2012;Meng & Gunay, 2017), multi-unit (Viehmann et al, 2021), discrete bid (Kastl, 2012;Markakis & Telelis, 2015), multi-attribute (Chetan et al, 2018;Ran et al, 2018) and different auction types (Xu et al, 2021). Most of these studies follow the framework of Wilson (1979), which regards the bidder strategy as a bid on the symmetric surplus supply curve and derives the optimal bidding curve through Euler equation.…”