{"paper_id":"2408.11773","title":"Deviations from the Nash equilibrium and emergence of tacit collusion in a two-player optimal execution game with reinforcement learning","authors":["Fabrizio Lillo","Andrea Macrì"],"paper_date":"2024-08-21","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2408.11773v1","page_url":"https://thequant.space/flowcharts/deviations-from-the-nash-equilibrium-and-emergence-of-tacit/","scores":{"empirical_rigor":6,"math_complexity":7.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper develops a Double Deep Q‑Learning algorithm to train two autonomous agents for optimal execution in a market impact game.","quote":"As our main contribution, we have developed a Double Deep Q-Learning algorithm where two agents are trained using RL.","summary":"The use of reinforcement learning algorithms in financial trading is becoming increasingly prevalent. However, the autonomous nature of these algorithms can lead to unexpected outcomes that deviate from traditional game-theoretical predictions and may even destabilize markets. In this study, we…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Permanent market impact is linear and constant: p(vt/τ)=κ vt/τ.","type":"model","stated":"explicit","quote":"St evolves because of a diffusion part ξt∼N(0,1) multiplied by the price volatility σ and a drift part, termed permanent impact, p(vt/τ), assumed to be linear and constant: p(vt/τ)=κvt/τ."},{"assumption":"Temporary market impact is linear and constant: h(vt/τ)=α vt/τ.","type":"model","stated":"explicit","quote":"The price ˜St received by the agent is equal to the mid-price St but impacted by a temporary impact term also assumed to be linear and constant in time: h(vt/τ)=αvt/τ."},{"assumption":"Agents are risk‑neutral (λ=0).","type":"other","stated":"explicit","quote":"In the following, we are going to consider risk neutral agents (λ = 0) and in this case the optimal trading schedule is the TWAP, i.e. vt = q0/N."},{"assumption":"Trading occurs in discrete time with N equal intervals of length τ = T/N.","type":"execution","stated":"explicit","quote":"time window [0,T], which, in the discrete‑time setting, is divided into N equal time increments of length τ = T/N."},{"assumption":"Mid‑price follows a random walk with drift depending on traded quantity.","type":"market","stated":"explicit","quote":"The main assumption of the model is that the mid‑price (or the efficient price) evolves according to a random walk with a drift depending on the traded quantity."},{"assumption":"Impact parameters (α, κ) are constant over time.","type":"model","stated":"explicit","quote":"impacts are the classic components of market impact."}]},"datasets":{"status":"none_found","items":[],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"synthetic"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":null,"detail":"Linear permanent impact p(vt/τ)=κ vt/τ and linear temporary impact h(vt/τ)=α vt/τ are incorporated into the price dynamics and implementation shortfall calculation.","quote":"St evolves because of a diffusion part ξt∼N(0,1) multiplied by the price volatility σ and a drift part, termed permanent impact, p(vt/τ), assumed to be linear and constant: p(vt/τ)=κvt/τ. The price ˜St received by the agent is equal to the mid‑price St but impacted by a temporary impact term also assumed to be linear a","source":"front_matter","rule":"R3","candidates":{"front_matter":"yes","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"Linear permanent impact p(vt/τ)=κ vt/τ and linear temporary impact h(vt/τ)=α vt/τ are incorporated into the price dynamics and implementation shortfall calculation.","card_quote":"St evolves because of a diffusion part ξt∼N(0,1) multiplied by the price volatility σ and a drift part, termed permanent impact, p(vt/τ), assumed to be linear and constant: p(vt/τ)=κvt/τ. The price ˜St received by the agent is equal to the mid‑price St but impacted by a temporary impact term also assumed to be linear a","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"unclear","findings":[{"finding":"The strategies learned by the RL agents deviate significantly from the Nash equilibrium and exhibit supra‑competitive, collusive‑like behavior that aligns closely with the Pareto‑optimal solution.","kind":"theoretical","quote":"Our findings reveal that the strategies learned by the RL agents deviate significantly from the Nash equilibrium of the market impact game."},{"finding":"The supra‑competitive strategies remain robust when agents are trained in one volatility regime and tested in different volatility regimes.","kind":"performance","quote":"Remarkably, we find that the strategies learned in one volatility regime remain supra‑competitive, even when adopted in various volatility conditions, underscoring the robustness of the training process."},{"finding":"For risk‑neutral agents the Pareto‑optimal (collusive) strategy coincides with the TWAP schedule.","kind":"theoretical","quote":"Finally, we have shown how the Pareto‑optimal strategy, that is indeed the collusion strategy in this setup, is the TWAP for risk‑neutral agents."}],"reported_performance":null,"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":20,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T18:03:45Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}