Persona-Trained Monte Carlo: Estimating Market-Outcome Distributions via Swarms of Persona-Conditioned Neural Policy Bots in a Limit Order Book

Jul 16, 2026

Persona-Trained Monte Carlo (PTMC) is a simulation-based method for estimating probability distributions of market-outcome statistics by repeatedly running limit-order-book interactions among swarms of neural policy bots. An outer Monte Carlo loop samples trader-heterogeneity from a learned distribution P; each run instantiates K bots that share a trained policy architecture but differ via persona-conditioned draws. Price paths produced endogenously form Monte Carlo samples, enabling estimation of target functionals such as drawdown, tail index, and crash probability, with quantified Monte Carlo error. PTMC contrasts randomness source and modeling assumptions against classical Monte Carlo, hand-coded ABMs, single-agent RL, and generative/LLM agents, and includes a survey-backed design rationale and a validation roadmap with stress-test comparisons and ethical considerations.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages58

Reading Time2 h 26 min

Introduction

Key methodological proposal

Motivation from financial modeling limits

Departure from existing ABM practice

Positioning versus adjacent paradigms

arXiv :2606 .29556v1 [ cs .LG] 28 Jun 2026 Persona-Trained Monte Carlo: Estimating Market-Outcome Distributions via Swarms of Persona-Conditioned Neural Policy Bots in a Limit Order Book Salavat Ishbulatov Independent researcher [salavat@doplan. ai](salavat@doplan. ai) Abstract We propose Persona-Trained Monte Carlo (PTMC): a method for estimating distributions of market-outcome statistics by repeatedly simulating limit-order-book interaction among swarms of persona-conditioned neural policy bots, with an outer Monte Carlo loop over draws from a learned trader-heterogeneity distribution P. Each simulation run instantiates K bots sharing one trained policy architecture πϕ but conditioned on heterogeneous, individually-sampled persona draws (θ(k), ρ (k) ) ∼ P; bots interact in a continuous double auction, and the resulting price path is one Monte Carlo sample. Repeating this Nruns times over independent personapopulation draws yields an ensemble from which a target functional F (e.g., maximum drawdown, tail index, crash probability) is estimated via ˆµN = Nrns Pi F (pathi) . Randomness therefore enters through three channels—persona draws, within-run action sampling from π ϕ , and optional exogenous shocks—rather than solely through an exogenous price process as in classical Monte Carlo. We distinguish PTMC from four adjacent paradigms: classical Monte Carlo (geometric Brownian motion, randomness only in price), hand-coded agent-based models (fixed behavioral archetypes, no learned P), single-agent reinforcement learning (one optimized policy, not an ensemble over heterogeneity), and large-language-model-based generative agents (language-driven reasoning rather than a compact, shared trained network) . To justify this design, we survey cross-disciplinary foundations—agent-based computational economics, market microstructure, behavioral finance and neuroeconomics, deep RL for trading, generative and LLM-based agents, news-driven trading, systemic risk, econophysics, game theory, and the mathematical machinery of stochastic processes and information theory—connecting each literature explicitly to a specific design choice in π ϕ , the training data, or the validation protocol. We formalize the PTMC estimator and its convergence properties, specify a candidate bot architecture and training objective, and propose a validation methodology including stylized-fact matching, microstructure-and agent-level checks, and historical stress-test comparison, with explicit head-to-head tests against a zero-intelligence baseline. The framework is proposed but not implemented. We contribute a formal estimator, a cross-disciplinary design justification, and a validation roadmap, without reporting new simulation or empirical findings. We address ethical, systemic-risk, and privacy considerations, and conclude with open research questions. Keywords: agent-based computational finance; Monte Carlo simulation; market microstructure; persona-conditioned agents; limit order book; behavioral cloning; reinforcement learning; financial market simulation 1 Introduction This paper proposes Persona-Trained Monte Carlo (PTMC): a method that estimates distributions of market-outcome statistics by repeatedly simulating limit-order-book interaction among swarms of persona-conditioned neural policy bots, with an outer Monte Carlo loop over draws from a learned trader-heterogeneity distribution P. Where classical Monte Carlo in finance averages a payoff functional over random paths of an exogenously specified price process Cox et al. (1979), PTMC averages a target functional over random paths generated endogenously by a population of interacting agents whose heterogeneity is itself a random draw, not a fixed design choice. This is the paper’s central methodological proposal, and the survey that follows exists to justify the design choices that go into it—not the other way around. Financial market prices emerge from the interaction of thous

FAQs

What does PTMC estimate in the proposed framework?

PTMC estimates distributions of market-outcome statistics by simulating limit-order-book interactions and treating resulting price paths as Monte Carlo samples. Target functionals can include maximum drawdown, tail index, spread distribution, or crash probability.

How does PTMC introduce randomness compared with classical Monte Carlo?

Classical Monte Carlo randomizes over exogenous price paths, while PTMC draws trader heterogeneity from a learned distribution P. Randomness also enters through action sampling from the trained policy and optionally via exogenous shocks.

How is the bot swarm constructed in each simulation run?

Each run instantiates K bots that share one trained policy architecture but are individually conditioned on sampled persona draws from P. Bots then interact in a continuous double auction to generate an endogenously produced price path.

Jul 16, 2026

Upload by Ava Thompson