In light of the Madoff case, we present an Agent-Based Model of a Ponzi scheme. Agents are initially inclined to invest in the scam because they believe the wealth will increase, even if the fraudster dissipates it without any investment. We stress that the main characteristic of such schemes is the growing discrepancy between the perceived wealth and the actual total amount of money in the impostor's possession.

The tendency gradually reverses and more agents withdraw their wealth (and made-up profits) if trust is lost as a result of hearing negative news about the economy. We look at how long it takes to expose the fraud and file for bankruptcy in relation to the volume of news that enters the market. We also look into the impact of a special agent dubbed Markopolos (inspired by a genuine personage) on the time to bankruptcy because of his capacity to quickly "convince" the agents he encounters to disinvest. Although the Markopolos effect seems to be statistically significant, it is not very strong when it comes to the results of a news flow and the subsequent widespread loss of faith and redemptions.

This study develops and validates an agent-based model (ABM)

of Japan’s new-graduate labor market to examine how educational pol-

icy shapes occupational labor supply under fixed structural constraints.

Student agents carry ability vectors derived from TF-IDF analysis of

university diploma documents and projected onto a seven-dimensional

competency space. The model reproduces the empirically observed rank-

order structure of occupation-level fulfillment rates, demonstrating struc-

tural validity across repeated trials and parameter conditions.

Two policy scenarios are simulated: Specialization, which amplifies faculty-

specific competency profiles, and Generalization, which reduces inter-

faculty variance. The results show that neither scenario improves ag-

gregate labor-supply–demand balance. Instead, Specialization produces

pronounced cross-occupational redistribution, while Generalization leads

to near-complete occupational inertia due to weakened sorting signals.

Sensitivity analysis reveals a clear hierarchy of constraints: institutional

eligibility requirements dominate, followed by social popularity signals,

with educational content exerting only a tertiary influence. Healthcare

occupations remain persistently undersupplied because eligibility con-

straints prevent inflow from outside qualified educational tracks.

These findings indicate that educational policy functions primarily as

a redistributive mechanism under structurally fixed constraints rather

than as a tool for aggregate improvement, highlighting the need for policy

design that explicitly accounts for constraint hierarchies.

The Economic Complexity Index (ECI) is a widely used predictor of national economic growth, but it relies exclusively on goods trade data and cannot capture digital capabilities. We identify an empirical anomaly — the "Kazakhstan Paradox" — where Kazakhstan's ECI declined significantly while GDP per capita grew steadily, coinciding with the rise of Kaspi.kz, a dominant fintech super-app. We propose a Dual Complexity Index (DCI) that incorporates both traditional and digital capability dimensions, and implement it in an agent-based model calibrated on Atlas of Economic Complexity data, Kaspi.kz investor metrics, and World Bank WDI. The DCI model achieves 35% lower MAPE than the ECI-only benchmark. We identify two novel mechanisms: Digital Dutch Disease (oil shocks crowd out digital capability formation) and crisis-driven digital acceleration. Results are validated across three post-Soviet economies: Kazakhstan, Belarus, and Azerbaijan.

Congestion pricing is increasingly adopted by cities worldwide, yet the behavioural mechanisms through which drivers respond to charges remain poorly understood. This study questions how the choice of agent decision rule shapes both the magnitude and spatial distribution of congestion reduction under cordon pricing. We build a proof-of-concept agent-based model on Auckland’s CBD road network and compare three decision rules under a proposed time-of-use (ToU) schedule: 1) a baseline exponential decay, 2) an El Farol bar model capturing bounded rationality, and 3) a Q-learning rule in which agents learn from repeated experience. Congestion is measured through the volume-to-capacity ratio (V/C), where values approaching 1.0 indicate a road operating at capacity. The three rules produce sharply different pathways. Baseline agents deter smoothly and show diminishing returns, reducing inner-cordon peak V/C only marginally (0.38 to 0.36). El Farol agents collectively over- and under-react, and peak V/C actually edges slightly upward under ToU (0.58 to 0.61 inside the cordon), while peak entry rises rather than falls. Q-learning agents, by contrast, con-verge within roughly ten days on individually stable policies: inner-cordon peak V/C falls from 0.65 to 0.39, the peak entry rate drops from 77.7% to 41.7%, and the reduction at the cordon boundary (VC by -0.19) is of the same order as inside the cordon (VC by -0.26), showing no sign of displacement onto peripheral roads. Mechanistically, lower value-of-time drivers relocate their trips to off-peak hours. Rather than prescribing a “correct” driver, the exercise shows that an ABM framework can place competing behavioural theories side by side within a common network and pricing environment, making visible how each assumption reshapes the congestion response and providing a platform on which researchers and policymakers can interrogate their own mental models of how drivers might react to a cordon charge.