Neural Network-Based Parameter Estimation of a Labour Market Agent-Based Model
M Lopes Alves, Joel Dyer, Doyne Farmer, Michael Wooldridge, Anisoara Calinescu · 2026
Read Neural Network-Based Parameter Estimation of a Labour Market Agent-Based Model by M Lopes Alves, Joel Dyer, Doyne Farmer, Michael Wooldridge, Anisoara Calinescu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-15572.txt.
Introduction
Lopes Alves and colleagues tackle a stubborn cost in agent-based modelling: fitting parameters of large models that simulate complex systems. Labour market models built on job transition networks are a case in point, since sweeping their parameter space by brute force is too costly for decision support. The team tests a simulation-based inference frame that uses neural networks to estimate parameters, applied to a known labour market model.
They run the model on synthetic data and on real United States labour market data, and they compare hand-built summary statistics against summaries learned by an embedded network. The network route recovers the true parameters across data scales in posterior checks and runs more efficiently than standard Bayesian methods. The paper shows neural inference making large agent models practical to fit, not just to run.
Catalog record
| CREATOR | M Lopes Alves, Joel Dyer, Doyne Farmer, Michael Wooldridge, Anisoara Calinescu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-15572
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Neural Network-Based Parameter Estimation of a Labour Market Agent-Based Model M Lopes Alves, Joel Dyer, Doyne Farmer, Michael Wooldridge, Anisoara Calinescu Abstract Agent-based modelling (ABM) is a widespread approach to simulate complex systems. Advancements in computational processing and storage have facilitated the adoption of ABMs across many fields; however, ABMs face challenges that limit their use as decision-support tools. A significant issue is parameter estimation in large-scale ABMs, particularly due to computational constraints on exploring the parameter space. This study evaluates a state-of-the-art simulation-based inference (SBI) framework that uses neural networks (NN) for parameter estimation. This framework is applied to an established labour market ABM based on job transition networks. The ABM is initiated with synthetic datasets and the real U.S. labour market. Next, we compare the effectiveness of summary statistics derived from a list of statistical measures with that learned by an embedded NN. The results demonstrate that the NN-based approach recovers the original parameters when evaluating posterior distributions across various dataset scales and improves efficiency compared to traditional Bayesian methods. Agent-based modelling (ABM) is a computational methodology widely used to simplify the representation of the real world through abstraction and reductionism while focusing on specific tasks or phenomena 14 . ABMs have been particularly valuable in economic research due to their ability to simulate diverse scenarios beyond the constraints of traditional models that often rely on oversimplified assumptions — such as perfect rationality and equilibrium. In labour markets specifically, ABMs have contributed to clarifying how markets are shaped, constantly changing, and respond to shocks such as the economic impacts of social distancing interventions during the COVID-19 pandemic 20 . Despite their potential use, ABMs face a significant issue in parameter estimation. ABM parameters, defined as \boldsymbol{\theta} , are typically distributed across a large parameter space, making their comprehensive exploration challenging. This study addresses this challenge by employing a state-of-the-art black-box simulation-based inference (SBI) framework for ABM parameter estimation, SBI4ABM (Simulated-Based Inference 4 (for) ABM) 9 in a labour market ABM 7 . By applying neural networks (NNs) and Bayesian theory, SBI4ABM can more effectively incorporate uncertainty quantification and prior knowledge. Previous research has explored alternative approaches to parameter estimation in labour market ABMs. Examples include methods minimising the sum of squared errors (SSE) and the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) 10 . Both approaches yield point estimates without accounting for uncertainty - another limitation that motivates the adoption of Bayesian frameworks such as the SBI4ABM. The goal of this paper is to evaluate the capability of the SBI4ABM framework in handling computationally intensive and stochastic ABMs. More specifically, whether the posterior density distributions derived by SBI4ABM accurately place the original ABM parameters in high-density areas. This work also addresses the lack of analyses of NN-based tools applied to ABM parameter estimation 17 ; 1 . 1 ABM parameter estimation Parameter estimation for stochastic ABMs poses significant computational challenges. As the labour market ABM analysed in this work, such ABMs present a large population of agents that might generate emergent behaviour through local interactions, requiring simulations that scale computationally with both agent count and interaction complexity 22 ; 4 . Another challenge lies in high-dimensional parameter spaces. The combinatorial search of possible parameter ranges and combinations raises challenges that traditional optimisation methods struggle to navigate efficiently 3 . In addition, the stochasticity from probabilistic behavioural rules introduces statistical variability problems that necessitate extensive sensitivity analyses and repeated simulations to distinguish systemic patterns from noise 22 ; 18 . 1.1 Simulation-based inference for ABM parameter estimation Simulation-based inference (SBI) is a likelihood-free inference technique extensively used for parameter estimation, particularly when direct computation of the likelihood function is infeasible 12 . By performing extensive simulations across parameter sets, SBI approximates the posterior distribution {p}(\boldsymbol{\theta}|\mathbf{y}) , the probability distribution over parameter values after seeing the data \mathbf{y} , making it especially suitable for ABMs 6 . Summary statistics address the high dimensionality of ABM outputs. These summary statistics can either be handcrafted - a static list of statistical measures that require domain expertise and careful selection 6 - or automatically learned from simulated data by constructing a separate learning module, usually a NN. 1.2 SBI4ABM In this paper, we assess the performance of the SBI4ABM 9 , a framework tailored for parameter estimation of high-dimensional models, such as ABMs applied to economic modelling. SBI4ABM addresses the limitations of traditional methods by integrating advanced machine learning techniques: the neural posterior estimation (NPE) 19 ; 16 ; 11 and the neural ratio estimation (NRE) 13 . NPE directly approximates the posterior density, and NRE approximates the likelihood-to-evidence ratio. The experiments described in later sections focus on the NPE. In NPE, NNs are trained to approximate the global posterior density p(\boldsymbol{\theta}\mid\mathbf{y}) across all possible values of \boldsymbol{\theta} and \mathbf{y} using a limited number of simulations compared to methods such as Approximate Bayesian Computation (ABC). The training for a global density estimator benefits the NN-based methods with amortisation: once trained, the NN can generate many posterior samples without additional ABM simulations. NPE is implemented using normalising flows (NF), a machine learning model to transform a simple base distribution (e.g., Gaussian) into a complex one that closely matches the true posterior, via invertible transformations. The neural estimators incorporated into the SBI4ABM framework have been shown to outperform traditional methods in terms of the accuracy of estimated posterior distributions 9 . 1.3 Labour Market Occupational Mobility and Automation ABM The target ABM is the Labour Market Occupational Mobility and Automation ABM (LM-ABM) 7 . This model simulates the effects of automation on workforce distribution in the U.S. It focuses on the mobility of employees between occupations of an out-of-equilibrium economy as the labour market adjusts to a new steady state after automation shocks. We can describe a labour market as having n\in\mathbb{N} occupations, a workforce distributed in those occupations \mathbf{z}=[z_{1},z_{2},...,z_{n}],\quad z_{i}\in\mathbb{N} , the automation probability per occupation \mathbf{p}=[p_{1},p_{2},...,p_{n}],\quad p_{i}\in[0,1] , and a matrix of transition probabilities between different occupations \mathbf{P_{i,j}}\in\mathbb{R}_{+}^{n\times n} where each element gives the probability of transitioning from occupation i to occupation j . For parameter estimation, an observed value of the labour market, \mathbf{y}\in\mathbb{R}^{n} , should also be provided. During each discrete time step, t , for each occupation i , the ABM computes the economic indicators listed in Table 1 . Table 1: Economic indicators for occupation i at time t Variable Description e_{i,t} Number of employed workers u_{i,t} Number of unemployed workers last employed in occupation i v_{i,t} Number of job vacancies d_{i,t} Labour demand The dynamics of the labour market are governed by shifts in labour demand, which depend on two processes: worker separation (firing) and vacancy creation. Worker separation occurs with rate \delta_{u} , while vacancy creation happens with rate \delta_{v} , contingent upon the number of currently employed workers in the same occupation. Let r\in[0,1] denote the probability that a worker remains in the same occupation each time step. These parameters are the same for all occupations and are presented in Table 2 . Table 2: ABM Parameters Parameters Description \delta_{u} Rate at which employed workers are separated (fired) \delta_{v} Rate at which new vacancies are opened r Probability that a worker stays in the same occupation 2