Towards Credible Agent-Based Policy Simulations: Disentangling Opportunities and Preferences in a Financial Inclusion Case Study of Egypt

Alba Aguilera; Georgina Curto; Nardine Osman; Ahmed Al-Awah

cs.AIOct 3, 2026

Towards Credible Agent-Based Policy Simulations: Disentangling Opportunities and Preferences in a Financial Inclusion Case Study of Egypt

Authors: Alba Aguilera, Georgina Curto, Nardine Osman, Ahmed Al-Awah

Organizations: Artificial Intelligence Research Institute (IIIA-CSIC) · United Nations Economic and Social Commission for Western Asia (UN-ESCWA) · United Nations University Institute in Macau (UNU Macau)

Abstract

Credibility is a central topic for agent-based models intended to support policy-making. Simulations must not only represent the target scenarios and their core dynamics but also demonstrate that their assumptions, parameters, and outputs are empirically grounded and sufficiently accurate for their intended use. This paper addresses this challenge by presenting a general modelling framework, aligned with the Capability Approach, for building credible policy simulations that rely on data and domain-expert knowledge. It then demonstrates how it can be contextualised and implemented to study the social challenge of financial inclusion in Egypt, building on an agent-based model that represents heterogeneous individuals and firms behaving according to their financial states, barriers, opportunities, and preferences. The model is fitted to real-world data in two stages, initialisation and calibration, which respectively build representative synthetic populations and estimate behavioural parameters. By fixing the feasibility parameters, which determine agents' opportunities, and calibrating preference parameters across different population groups, we are able to distinguish and analyse the role of institutional and social barriers in the system, as well as the role of agents' motivations and priorities. This calibration stage provides transparent and group-specific hypotheses about the drivers of observed financial-inclusion gaps, which can further be analysed as gaps between agents' opportunities and realised outcomes, a very relevant insight for policy-making. This paper is thus a step towards improving the credibility and usefulness of policy simulations, strengthening the relationship between the model, the real target system, and the stakeholders who will use it. The code is available at: https://www.comses.net/codebase-release/df8383cb-f49b-4f09-8cce-73603b59adcc/.

Figures & tables

Figure 1: CA-aligned simulation framework from ( Aguilera et al. 2025a ; Aguilera et al. 2026 ) , operationalised for the financial inclusion context in Egypt.

Figure 2: Main agents involved in the financial inclusion context for the Arab region, defined according to domain expert knowledge.

Agent	Central Capabilities $c\in\mathcal{C}$	Actions $a\rightarrow c$	Indicators $i\rightarrow c$
Indiv.	$c_{1}$ : Financial autonomy : freedom to independently own and operate financial instruments.	$a_{1}$ , $\hat{a}_{1}$ open / do not open a formal account	Traditional/digital account ownership rate.
	$c_{2}$ : Financial empowerment : freedom to access credit and borrow from reliable sources for productive growth.	$a_{4}$ , $\hat{a}_{4}$ save / do not save formally $a_{5}$ , $\hat{a}_{5}$ borrow / do not borrow formally $a_{6}$ , $\hat{a}_{6}$ borrow / do not borrow informally	Formal/informal borrowing rate.
	$c_{3}$ : Digital financial literacy : ability to use digital financial services effectively.	$a_{2}$ , $\hat{a}_{2}$ adopt / do not adopt mobile money $a_{3}$ , $\hat{a}_{3}$ use / do not use digital payments	Digital/traditional payments and remittances. NGO reached users.
SMEs	$c_{5}$ : Formalization : freedom to operate as a recognized enterprise.	$a_{7}$ , $\hat{a}_{7}$ obtain / do not obtain a business license $a_{8}$ , $\hat{a}_{8}$ obtain / do not obtain a commercial registration $a_{9}$ , $\hat{a}_{9}$ adopt / do not adopt accounting books	Share of formalized SMEs; share of SMEs with valid license.
	$c_{6}$ : Digital financial literacy : ability to adopt digital channels for business transactions.	$a_{10}$ , $\hat{a}_{10}$ digitize / do not digitize sales/payments	Share of SMEs with digitized sales/payments.
	$c_{7}$ : Employment quality : freedom to create stable and protected jobs.	$a_{11}$ , $\hat{a}_{11}$ hire / do not hire non-household workers $a_{12}$ , $\hat{a}_{12}$ insure / do not insure workers	Share of employer SMEs with insured workers.

Table 1: Central opportunity–action–outcome mapping used in the simulation.

Panel of Figure 3 — Figure 3: Feasibility probability $p^{\alpha}_{i,a,t}$ for each action at initialization, disaggregated by target group, for (a) individuals and (b) small enterprises. The dashed line marks the opportunity threshold $\tau=0.5$ .

Level	Dataset	Size (cases $\times$ columns)	Geographic granularity	Information
Micro	$ds_{1}$ : Global Findex Survey ( World Bank 2025 )	1001 $\times$ 178	Nation	Individual-level socio-demographic characteristics and financial inclusion outcomes : account ownership, saving, borrowing, payments, insurance, digital access and use, and self-reported barriers to formal financial service use.
	$ds_{2}$ : Egypt Labour Market Panel Survey ( Economic Research Forum 2023 )	70,000 $\times$ 4,500 (IND)	Village	Individual- and household-level socio-economic characteristics, labour-market status , and, where available, enterprise and finance-related modules for synthetic population enrichment.
	$ds_{3}$ : Egypt Labor Force Survey ( OAMDI 2025 )	297,000 $\times$ 103 (IND)	Region	Labour-market characteristics used to structure the synthetic population, including employment status, sector, occupation, and income-related variables.
Macro	$ds_{4}$ : Financial Access Survey ( International Monetary Fund 2025 )			Infrastructure and financial service providers: counts of commercial banks, microfinance institutions, credit unions and cooperatives, and insurance corporations.
	$ds_{5}$ : Indicators ( Information Technology Industry Development Agency (ITIDA) 2026 ) and reports ( Research and Markets 2026 )			Non-bank financial inclusion ecosystem: venture capital and accelerator counts, typical ticket size, fintech focus, cohort size, and graduation rate.
	$ds_{6}$ : Indicators ( Financial Regulatory Authority (FRA) 2026 ) and reports ( Entlaq 2024 ; Entrepreneur Middle East 2024 )			Digital and non-bank finance: BNPL providers, active users, market size, digital lenders, average loan size, and default rate.

Table 2: Micro- and macro-level data sources used for initialisation.

Figure 4: Prior and posterior preference parameters $\theta_{j}\in\Theta$ after calibration, for individuals (left) and SMEs (right), ordered by action. Circles show the prior mean while diamonds show the posterior mean across the three random seeds, coloured by the direction of the movement (green: increased; red: decreased; grey: within $10\%$ of the prior); horizontal bars show the across-seed range. The dashed line marks the neutral value $\theta_{j}=1$ .

Movement	Interpretation	Individuals		SMEs
		Parameter	Change	Parameter	Change
Increased	Stronger revealed preference for $a$ (or for remaining in $\hat{a}$ ) than the prior assumed.	$\theta^{\mathrm{indiv}}_{\hat{a}_{3},\mathrm{female}}$	$+55.7\%$	$\theta^{\mathrm{SME}}_{\hat{a}_{7}}$	$+47.8\%$
		$\theta^{\mathrm{indiv}}_{a_{1}}$	$+36.5\%$	$\theta^{\mathrm{SME}}_{\hat{a}_{11}}$	$+39.9\%^{\dagger}$
		$\theta^{\mathrm{indiv}}_{a_{3}}$	$+33.7\%^{\dagger}$	$\theta^{\mathrm{SME}}_{a_{8},\mathrm{low\ capital}}$	$+25.7\%^{\dagger}$
		$\theta^{\mathrm{indiv}}_{a_{5},\mathrm{rural}}$	$+25.6\%^{\dagger}$	$\theta^{\mathrm{SME}}_{a_{9},\mathrm{low\ capital}}$	$+22.6\%^{\dagger}$
		$\theta^{\mathrm{indiv}}_{a_{3},\mathrm{rural}}$	$+25.2\%$	$\theta^{\mathrm{SME}}_{a_{7}}$	$+18.1\%$

Table 3: Largest posterior movements in percentages relative to prior values, by agent type and averaged across the three random seeds. Each movement is a change in an estimated preference (conditional on the fixed feasibility

\Phi_{0}

). Movements marked † have a large across-seed spread (posterior s.d.

>0.30

) and are therefore only weakly identified.

Panel of Figure 5 — Figure 5: Calibrated opportunities (dashed) versus realized outcomes (solid) by target group, for (a) individuals and (b) small enterprises. Colours denote target groups (individuals are red: youth, green: women, blue: rural, amber: low income; and SMEs are green: women-led, blue: rural manager, amber: low capital). For each central capability, its related actions (in Table 1 ) are aggregated by Eq. 5 with equal weights. The distance between the polygons is the capability gap in Eq. 6 . At the action level, the dashed polygon is the share for which the action is a genuine opportunity (feasible above $\tau$ or already realised) and the solid polygon the share that realises it.

Figure 6: Elicitation item to be shared with the domain-expert panel for hypothesis H1 . The expert is shown the parameter movement and our interpretation, and responds accept / revise / reject, with an open prompt to comment on unmodelled barriers that could change it.

Agent	Action	Parameter ( $b_{i,a,t}$ or $b_{a,t}$ )	Value / rule	Justification based on domain-expert knowledge
Baseline feasibility
Individual	$a_{1}$ – $a_{6}$	baseline $b^{ind}_{0,a_{j}}$	$0.85$ ; $0.80\,p^{\alpha}_{i,a_{1},t}$ ; $0.75$ ; $0.65$ ; $0.55$ ; $0.80$	Account opening and informal borrowing are considered the most relatively feasible, while usage actions like mobile-money adoption and digital payments are less feasible. The first depends on account-opening feasibility. Formal borrowing is the less feasible action, as hinted by domain-expert reports.
SME	$a_{7}$ – $a_{12}$	baseline $b^{SME}_{0,a_{j}}$	$0.70$ ; $0.65$ ; $0.60$ ; $0.65$ ; $0.55$ ; $0.50$	Obtaining a business license is the most feasible SME transition, followed by commercial registration and adopting a digital sales channel, then accounting books. Hiring non-household workers and, especially, insuring workers receive the lowest baselines; worker insurance is the lowest because it requires both formal employment relationships and a connection to labour or social-insurance institutions.
Barrier multipliers
Individual	$a_{1}$ , $a_{4}$	not enough money	$\times 0.20$ , $\times 0.30$	Lack of money strongly reduces opening an account and saving formally. Domain-expert reports mark insufficient funds as the most frequent unbanked barrier in Arab LMICs, with Egypt about 90% and Morocco about 75% among the unbanked.
	$a_{1}$	high fees, strict documentation	$\times 0.05$ , $\times 0.10$	High fees and hard documentation requirements make account opening difficult. Global Findex and ESCWA Financial Inclusion Index evidence identify account costs and documentation requirements among the most commonly reported barriers to account ownership in developing economies and Arab countries, although their relative importance is generally lower than insufficient funds ( KlapperEtAl2025 ; ESCWA2025FII ) .

Table 4: Feasibility assumptions, which define action feasibility before preference parameters are applied, classified between baselines, barriers, contextual barriers, and policy multipliers. All complement feasibilities are

1.00

unless changed by policies, final feasibility is clipped to

[0,1]

and context barriers qualitative levels are mapped to

0.3

0.6

and

0.9

quantitative levels. In the 20,000-person synthetic population, about 63% of individuals report at least one active account-access barrier. For individuals without an active reported barrier, feasibility is informed by the baseline.

Agent	Action	Parameter	Value / rule	Justification based on domain-expert knowledge
Barrier multipliers
SME	$a_{7}$ – $a_{9}$	formalization readiness score $r_{f}$	base $0.55$ ; capital low/mid/high: $\times 0.45$ , $\times 1.00$ , $\times 1.35$ ; workers no/yes: $\times 0.75$ , $\times 1.35$ ; age young/mid/older: $\times 0.85$ , $\times 1.00$ , $\times 1.10$ ; manager income low/mid/high: $\times 0.85$ , $\times 1.00$ , $\times 1.10$ ; $r_{f}$ clipped to $[0.15,1.50]$ ; $a_{7},a_{8}:\times r_{f}$ ; $a_{9}:\times(0.65+0.35r_{f})$	Data-informed structural proxy. Since barriers are not explicitely reported in data, we create a formalization readiness score. The score starts at $0.55$ and is adjusted using observed SME characteristics. Small, young, low-capital firms without workers are treated as less ready to formalize. Larger, older, higher-capital firms with non-household workers are treated as more ready. License and registration depend directly on $r_{f}$ ; accounting books depend on it more softly because bookkeeping can occur before full formalization. The direction is supported by ELMPS SME variables and MSME-barrier reports; exact multipliers are modelling assumptions.
SME	$a_{7}$ – $a_{9}$	manager has formal finance	$\times 1.40$ , $\times 1.50$ , $\times 1.20$	Prior use of formal finance increases familiarity with documentation, record keeping, and provider requirements, which makes licensing, registration and bookkeeping more feasible. The penalty for managers without formal-finance experience captures exclusion among small and informal firms (World Bank Enterprise Surveys, Egypt 2020; Morocco 2019).
SME	$a_{10}$ , $a_{9}$	manager digital state	$a_{10}:\times 1.80$ , $a_{9}:\times 1.20$ ; plus $r_{f}:\times 1.05$ , followed by clipping $r_{f}$ to $[0.15,1.50]$	Digital capability is a key determinant of SME adoption of digital payments and sales channels. World Bank and GSMA evidence shows that adoption depends not only on infrastructure but also on managerial skills, confidence, and ability to use digital tools. The bookkeeping effect is smaller because records can still be maintained manually. The readiness bonus reflects complementarities between digitalization and formalization.
	$a_{10}$ , $a_{7}$	rural manager	$a_{10}:\times 0.80$ , $a_{7}:\times 0.90$	Rural firms face weaker connectivity, lower digital adoption among customers, and fewer nearby service providers. The stronger penalty is therefore assigned to digitization. Licensing is less affected because administrative procedures can often be completed through intermediaries or occasional travel, although distance and information costs remain relevant. This is consistent with rural–urban gaps reported in ESCWA financial inclusion and GSMA digital-access evidence.
	$a_{11}$ , $a_{12}$	capital scale	low capital: $a_{11}:\times 0.60$ , $a_{12}:\times 0.50$ ; high capital: $a_{11}:\times 1.50$ , $a_{12}:\times 1.60$	Capital is one of the strongest predictors of firm growth and formal employment. Hiring and worker insurance require sustained wage and compliance costs, while productive investment can sometimes be financed incrementally through retained earnings or supplier credit. Enterprise Survey data consistently show higher investment, employment growth, and formal-finance access among larger and better-capitalized firms, justifying the stronger effects on hiring and insurance.

Table 5: Remaining barrier multipliers for SMEs and context barrier multipliers.

Agent	Action	Parameter(s)	Value ( $\sqrt{R}\rightarrow\theta^{0}$ )	Justification based on observed ratios
Neutral or weakly identified preference priors
Individual	$a_{1}$ – $a_{6}$	$\theta^{\mathrm{indiv}}_{a_{j}}$	$1.00$	Neutral prior for overall individual transition actions. Calibration can move the values up or down using synthetic-person targets, ratios are considered in the complement actions.
SME	$a_{7}$ – $a_{12}$	$\theta^{\mathrm{SME}}_{a_{j}}$	$1.00$	Neutral prior for overall SME transition actions. Calibration can move the values up or down using synthetic-SME targets when available.
Data-informed preference priors below one
Individual	$a_{3}$	$\theta^{\mathrm{indiv}}_{a_{3},\mathrm{female}}$ , $\theta^{\mathrm{indiv}}_{a_{3},\mathrm{rural}}$	$0.965$ , $0.978$	Data-informed priors for digital-use transition actions, with ratios for female and rural subgroups equal to $0.368/0.396=0.931\rightarrow\theta^{0}=0.965$ and $0.378/0.396=0.956\rightarrow\theta^{0}=0.978$ , respectively.
Individual	$\hat{a}_{3}$	$\theta^{\mathrm{indiv}}_{\hat{a}_{3},\mathrm{low\ income}}$	$0.995$	Data-informed prior for non-digital use complement action, with ratio for low-income group equal to $0.598/0.604=0.989\rightarrow\theta^{0}=0.995$ .

Table 6: Preference-prior assumptions classified between neutral or data-informed. Prior mean values

\theta^{0}

used in the simulation are listed by agent type, action set, and associated mathematical parameter. A justification explicitely explain how are they built from observed shrinked ratios

R

in the data, i.e. 20,000-person synthetic population and the associated 4,318-row synthetic SME. Additional shrinkage is considered for extremely large ratios, only when

\sqrt{R}>2

Agent-Based Modeling

Agent-based modeling simulates how interacting, heterogeneous entities and their environments produce patterns in social, economic, biological, and engineered systems.

See all 160 papers

Apr 5, 2026

From Plausible to Causal: Counterfactual Semantics for Policy Evaluation in Simulated Online Communities

LLM-based social simulations can generate believable community interactions, enabling

policy wind tunnels'' where governance interventions are tested before deployment. But believability is not causality. Claims like

intervention

A

reduces escalation'' require causal semantics that current simulation work typically does not specify. We propose adopting the causal counterfactual framework, distinguishing necessary causation (would the outcome have occurred without the intervention?) from sufficient causation (does the intervention reliably produce the outcome?). This distinction maps onto different stakeholder needs: moderators diagnosing incidents require evidence about necessity, while platform designers choosing policies require evidence about sufficiency. We formalize this mapping, show how simulation design can support estimation under explicit assumptions, and argue that the resulting quantities should be interpreted as simulator-conditional causal estimates whose policy relevance depends on simulator fidelity. Establishing this framework now is essential: it helps define what adequate fidelity means and moves the field from simulations that look realistic toward simulations that can support policy changes.

Agam Goyal, Yian Wang, Eshwar Chandrasekharan +1

University of Illinois Urbana-Champaign, Urbana, Illinois, USA

Counterfactual Evaluation Large Language Model-Based Social Simulation Causal Inference Actual Causality

Ask PDF

Sep 23, 2026

PAWS: Policy-driven Agentic World Simulation

Policy interventions propagate through public communication, institutional decisions, and stakeholder responses, yet datasets for financial multi-agent simulation rarely connect these processes to temporally aligned historical evidence. We introduce PAWS, a Policy-driven Agentic World Simulation dataset covering 36 verified U.S. financial and economic policy episodes, 12,727 policy-linked news records, and 65,291 source-grounded stakeholder actions. Each action is linked to its supporting news and represented by a multi-layer event frame capturing its interaction mode, financial-action family and subtype, semantic attributes, and conditional mappings to external taxonomies. Entities are resolved to normalized organizations, and actions are aligned with daily market-return context to support policy-agent simulation replay. On 2,522 stratified action samples, independent AI and human reviewers achieved 89.4% initial agreement on interaction mode, with disagreements subsequently adjudicated. Case studies of the 2008 short-selling ban and 2001 decimalization recover documented policy timelines and associated market patterns across both dense and sparse news settings. A replay study further shows that high accuracy can mask failure to detect rare stakeholder actions, identifying action timing and calibration as central challenges. PAWS provides an auditable substrate for evaluating agent influence, policy-response cascades, and action-outcome alignment in historically grounded financial simulations.

Tiviatis Sim, Jia Hui Woon, Xinming Gao +5

National University of Singapore · Tsinghua University · City University of Hong Kong

Benchmark Design Agent-Based Modeling

Ask PDF

Oct 24, 2026

FraudSociety: A Generative Agent-Based Society for the Financial-Fraud Economy

Financial fraud caused over $15.9 billion in losses in the United States in 2025, according to the Federal Trade Commission~\citep{ftc2026databook}. Understanding such large-scale harm requires jointly modeling informational and financial dynamics. Yet existing simulators typically capture only one side of this interaction, leaving a gap in modeling how fraud unfolds in the real world. To bridge this gap, we introduce FraudSociety, the first LLM-agent-based simulator that jointly models these coupled dynamics. It integrates a dynamic social environment for information exposure and communication, persistent financial ledgers that preserve payment and loss histories, and a Need-Intent-Action architecture for civilian decision making. Fraudsters further adapt their communication and financial actions to individual interaction histories and ledger states, forming a closed feedback loop between the two dynamics. We validate FraudSociety across macro-level outcomes, micro-level behavior, and intervention effects. At the macro level, the simulated fraud outcomes closely match real-world patterns across multiple key statistics. At the micro level, it reproduces strong loss concentration, with the top 10% of victimization records accounting for 81.1% of total losses, consistent with Canadian Anti-Fraud Centre statistics~\citep{cafc2025annual}. Counter-fraud experiments further reveal variation across intervention strategies, with the best-performing personalized intervention reducing the repeat-payment rate from 49.4% to 10.9%. These results establish FraudSociety as both a high-fidelity simulator and a scalable testbed for evaluating anti-fraud strategies under diverse intervention conditions, thereby advancing research on financial fraud and its prevention.

Keyang Chen, Mingxuan Jiang, Zeping Li +10

Fudan University, Shanghai, China · University of Illinois Urbana-Champaign, Urbana, IL, USA

Financial Fraud Detection Agent-Based Modeling Large Language Model-Based Social Simulation Human Behavior Simulation

Ask PDF

Abstract

Agent-Based Modeling

Explore similar work

From Plausible to Causal: Counterfactual Semantics for Policy Evaluation in Simulated Online Communities

PAWS: Policy-driven Agentic World Simulation

FraudSociety: A Generative Agent-Based Society for the Financial-Fraud Economy