{"paper_id":"2503.00320","title":"Shifting Power: Leveraging LLMs to Simulate Human Aversion in ABMs of Bilateral Financial Exchanges, A bond market study","authors":["Alicia Vidler","Toby Walsh"],"paper_date":"2025-03-01","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2503.00320v2","page_url":"https://thequant.space/flowcharts/shifting-power-leveraging-llms-to-simulate-human-aversion-i/","scores":{"empirical_rigor":6.5,"math_complexity":4,"quadrant":"Street Traders"},"proposes":{"status":"available","statement":"The paper proposes TRIBE, an agent‑based model that integrates a large language model to simulate human‑like aversion and client agency in bilateral government bond markets.","quote":"To address these complexities, we introduce TRIBE an agent-based model augmented with a large language model (LLM) to simulate human-like decision-making in trading environments. Our unique contribution is to contribute a framework and example of how an LLM imple-mentable within an ABM market simulation can add, and represent, these human differences.","summary":"Bilateral markets, such as those for government bonds, involve decentralized and opaque transactions between market makers (MMs) and clients, posing significant challenges for traditional modeling approaches. To address these complexities, we introduce TRIBE an agent-based model augmented with a…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Each market maker is modeled as an individual agent with its own adaptive utility function.","type":"model","stated":"explicit","quote":"Each market maker is modeled as an individual agent with its own adaptive utility function.","support":"full"},{"assumption":"Clients are only available to be contacted with a preset probability p at each time step.","type":"market","stated":"explicit","quote":"In summary: (1) Clients have bonds and cash sourced from a log normal distribution (in line with public data) (2) Many clients are simply not available for a market market to contact at any time step (we model \"client availability\") (3) Within TRIBE(LLM) we try to capture a synthetic client’s personal preference to trade \"right now\". 𝐵, and corresponding values for cash are set based on available data, such as from","support":"full"},{"assumption":"Clients’ desire to be a buyer or seller at any point is drawn from a Bernoulli distribution with equal probability.","type":"statistical","stated":"explicit","quote":"(4) Trade direction: despite bond and cash holdings many clients have needs outside of the model to buy and sell, we ascribe a Bernoulli distribution to the desire to be a buyer or seller at any point. Definition 3.1. as is 𝐷𝑗,𝑡hence an even chance of being a buyer or seller, and an even chance of deciding to trade, determined at each time step.","support":"full"},{"assumption":"The number of market‑making agents is fixed at four.","type":"model","stated":"explicit","quote":"TRIBE ABM: M = ⟨𝑁, L, S, P, D, A, 𝑓,𝑇⟩, where: 𝑁= {𝑎1,𝑎2, . . . ,𝑎𝑛} is the set of market-making agents, with𝑛= 4 in this case.","support":"full"},{"assumption":"Trade intention (“right now”) is generated by querying an LLM with limited prompt information rather than a parametric distribution.","type":"model","stated":"explicit","quote":"Rather than us-ing a distributional assumption with no available data for parametrization, we rely on an LLM for this choice with limited prompt information. To incorporate the dynamic phenomena, we use a call to an LLM at the point at which TRIBE is a Client deciding if they want to trade with a specific MM \"right now\" at a time step.","support":"full"},{"assumption":"Initial client bond and cash holdings follow a log‑normal distribution calibrated to public data.","type":"data","stated":"explicit","quote":"In summary: (1) Clients have bonds and cash sourced from a log normal distribution (in line with public data) (2) Many clients are simply not available for a market market to contact at any time step (we model \"client availability\") (3) Within TRIBE(LLM) we try to capture a synthetic client’s personal preference to trade \"right now\".","support":"full"}]},"datasets":{"status":"available","items":[{"name":"APRA bond holdings balance data","provider":"Australian superannuation regulator","kind":"market-prices","access":"public","period":"December 2023","universe":null,"quote":"𝐵, and corresponding values for cash are set based on available data, such as from the Australian superannuation reg-ulator [3] Client availability at any given time step 𝑡for position (𝑥𝑗,𝑦𝑗) is modeled as: 𝐴(𝑥𝑗,𝑦𝑗) (𝑡) = ( 1 with probability 𝑝 0 with probability 1 −𝑝 where 𝑝is the preset availability probability, and Clients’ roles as buyers or sellers are probabilistically drawn from a Bernoulli distribution, As","support":"partial"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"synthetic"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R4a","candidates":{"front_matter":"n/a","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"Business costs for market makers are set between 0.1 and 0.5 (per simulation parameter).","card_quote":"Table 1: TRIBE ABM: Simulation Parameters Parameter Category Value Grid Size General 50 x 50 (2500 clients) Number of Bankers MM Settings 4 Client Base grid size MM Settings 1 to 50 Business costs MM Settings 0.1 to 0.5 Initial Bonds Range MM Settings 1 to 5 Initial Cash Range MM Settings 1 to 5 Maximum Bonds Client Sett","meaning":"the paper reports no trading or execution result; the cost the extractor found is a model parameter"},"results":{"status":"available","evaluation_type":"none","findings":[{"finding":"Integrating an LLM into the ABM to enhance client agency is feasible.","kind":"theoretical","quote":"Our research outcomes are threefold: First, we demonstrated that GABM with enhanced client agency is feasible.","support":"full"},{"finding":"Even slight trade aversion encoded in the LLM prompt results in a complete cessation of trading activity.","kind":"negative","quote":"Second, LLMs with even slight trade aversion mentioned in a prompt result in no trading activity.","support":"full"},{"finding":"Introducing human‑like variability shifts power to clients and frequently leads to systemic agent collapse and severe instability across simulations.","kind":"negative","quote":"Finally, incorporating human-like variability shows that shifting power to clients can disproportionately affect system dynamics, frequently leading to agent trading collapse. Across 150 simulations, we find that short-term variability induced by LLMs causes significant system instability and a dramatic reduction in client to market maker asset trading.","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":9,"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":1,"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":2,"extracted_at":"2026-11-14T19:52:04Z"},"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/"}