{"paper_id":"2402.09129","title":"Optimal Automated Market Makers: Differentiable Economics and Strong Duality","authors":["Michael J. Curry","Zhou Fan","David C. Parkes"],"paper_date":"2024-02-14","paper_type":"theoretical","paper_url":"https://arxiv.org/pdf/2402.09129v1","page_url":"https://thequant.space/flowcharts/optimal-automated-market-makers-differentiable-economics-an/","scores":{"empirical_rigor":1.5,"math_complexity":8.5,"quadrant":"Lab Rats"},"proposes":{"status":"available","statement":"The paper proposes that finding an optimal market maker is dual to an optimal transport problem and uses differentiable economics to search for and conjecture optimal multi‑good market‑making mechanisms.","quote":"In this paper, we show that finding an optimal market maker is dual to an optimal transport problem, with specific geometric constraints on the transport plan in the dual. To search through the space of mechanisms and develop our theoretical results and additional conjectures on optimal designs, we make use of differentiable economics [17], which employs the tools of modern machine learning to search through the","summary":"The role of a market maker is to simultaneously offer to buy and sell quantities of goods, often a financial asset such as a share, at specified prices. An automated market maker (AMM) is a mechanism that offers to trade according to some predetermined schedule; the best choice of this schedule…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Traders are additive and quasilinear.","type":"model","stated":"explicit","quote":"Under a deliberately simple model of additive and quasilinear traders, and considering an objective for the market maker that can take into account the possibility of adverse selection (following that of Milionis et al. [32]), we ask how much of the full space of possible mechanisms will a market maker want to use?","support":"full"},{"assumption":"The market maker has an initial belief vector c over the values of d goods.","type":"market","stated":"explicit","quote":"2.1 Mechanism Design Objectives We assume that the market maker has some initial belief vector c ∈X = [0, b]d, for some maximum value b > 0, on the value for each of d ≥1 goods.","support":"full"},{"assumption":"Trader types are drawn from a known distribution with density f(x).","type":"model","stated":"explicit","quote":"The trader’s value is, in mechanism design terms, their type, drawn from some distribution with probability density function, f(x).","support":"full"},{"assumption":"After observing a trader, the market maker updates beliefs via π(c,x) = λc + (1−λ)x.","type":"model","stated":"explicit","quote":"To capture this possibility, following Milionis et al. [32], we adopt an update function, π(c, x) : X × X →X. When the market maker observes a trader with value x, they update their own belief about the value of the goods from c to π(c, x) (note that, as in Milionis et al. [32], we require that π(c, c) = c).","support":"full"},{"assumption":"The market maker can buy or sell up to one unit of each good.","type":"market","stated":"explicit","quote":"We assume that the market maker is willing to buy or sell up to one unit of each good.","support":"full"},{"assumption":"Mechanisms considered are strategyproof and individually rational direct‑revelation mechanisms or menus of lottery bundles.","type":"other","stated":"explicit","quote":"These correspond to either 4 strategyproof and individually rational (defined below) direct-revelation mechanisms with feasible allocation rules, or menus of (lottery) bundles at associated prices.1 An important detail of the market setting is that we also need to handle adverse selection, reflecting that the trader may be better informed than the market maker.","support":"full"}]},"datasets":{"status":"available","items":[{"name":"Uniform distribution on [0,1] for two goods","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"5.2 A new family of optimal mechanisms We consider the case of two kinds of goods, with values uniformly distributed on [0, 1], a market maker with initial valuation of c = ( 1 2, 1 2), and a linear update model, π(c, x) = λc + (1 −λ)x.","support":"full"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"none"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R3","candidates":{"front_matter":"n/a","evidence_card":"n/a","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"none","findings":[{"finding":"The market‑making profit maximization problem under adverse selection is dual to an optimal transport problem.","kind":"theoretical","quote":"In this paper, we show that finding an optimal market maker is dual to an optimal transport problem, with specific geometric constraints on the transport plan in the dual. We show that the problem of maximizing profit under adverse selection is dual to an optimal transport problem.","support":"full"},{"finding":"Optimal multi‑good mechanisms can bundle trades and accept payments in kind, offering better prices than separate bid‑ask spreads.","kind":"theoretical","quote":"We show that optimal mechanisms for multiple goods and under adverse selection can take advantage of bundling, both improved prices for bundled purchases and sales as well as sometimes accepting payment \"in kind.\" However, they also offer bundled trades at better prices than the separate bids and asks.","support":"full"},{"finding":"Learned mechanisms exhibit mixed bundling and sometimes a continuum of allocations, confirming the richness of the mechanism space.","kind":"theoretical","quote":"We find that some of the learned mechanisms make use of mixed bundling; i.e., trading in goods separately at one price, but also offering bundle discounts or taking payments in kind (accepting a sale of one good to discount the purchase of another). Other mechanisms offer a continuum of allocations in certain regions.","support":"full"},{"finding":"In a two‑good uniform setting with linear adverse‑selection update, the optimal mechanism improves profit by about 10 % (up to 11.4 % for certain λ).","kind":"performance","quote":"For λ = 1 this gives about a 10% increase in profit; for λ = √ 2 3 the increase is highest at 11.4% – a fairly substantial improvement in profit.","support":"full"},{"finding":"The conjectured optimal mechanisms require bundling and in‑kind trades, contrasting with the simple bid‑ask spread optimal in the single‑good case.","kind":"theoretical","quote":"From an economics perspective, our results establish that in some cases, the optimal mechanism for market makers who work across multiple goods must involve making use of their capacity to bundle and to accept goods “in kind”, resulting in significant improvements in profit. This is by contrast with the single-good case, where a much simpler bid-ask spread is optimal.","support":"full"}],"reported_performance":null,"out_of_sample":{"method":"none","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":21,"citations":{"cited_by_in_corpus":1,"by_independent_authors":1,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":1},"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":2,"extracted_at":"2026-11-14T19:58:56Z"},"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/"}