Fraudulent actions of a trader or a group of traders can cause substantial disturbance to the market, both directly influencing the price of an asset or indirectly by misinforming other market participants. Such behavior can be a source of systemic risk and increasing distrust for the market participants, consequences that call for viable countermeasures. Building on the foundations provided by the extant literature, this study aims to design an agent-based market model capable of reproducing the behavior of the Bitcoin market during the time of an alleged Bitcoin price manipulation that occurred between 2017 and early 2018. The model includes the mechanisms of a limit order book market and several agents associated with different trading strategies, including a fraudulent agent, initialized from empirical data and who performs market manipulation. The model is validated with respect to the Bitcoin price as well as the amount of Bitcoins obtained by the fraudulent agent and the traded volume. Simulation results provide a satisfactory fit to historical data. Several price dips and volume anomalies are explained by the actions of the fraudulent trader, completing the known body of evidence extracted from blockchain activity. The model suggests that the presence of the fraudulent agent was essential to obtain Bitcoin price development in the given time period; without this agent, it would have been very unlikely that the price had reached the heights as it did in late 2017. The insights gained from the model, especially the connection between liquidity and manipulation efficiency, unfold a discussion on how to prevent illicit behavior.

The paper sketches and elaborates on a framework integrating agent-based modelling with advanced quantitative probabilistic methods based on copula theory. The motivation for such a framework is illustrated on a artificial market functioning with canonical asset pricing models, showing that dependencies specified by copulas can enrich agent-based models to capture both micro-macro effects (e.g. herding behaviour) and macro-level dependencies (e.g. asset price dependencies). In doing that, the paper highlights the theoretical challenges and extensions that would complete and improve the proposal as a tool for risk analysis.

The financial sector continues to experience wide digitalization; the resulting transactional activity creates large amounts of data, in principle enabling public and private actors to better understand the social domain they operate on, possibly facilitating the design of interventions to reduce illegal activity. However, the adversarial nature of frauds and the relatively low amount of observed instances make the problem especially challenging with standard statistical-based methods. To address such fundamental issues to non-compliance detection, this paper presents a proof-of-concept of a methodological framework based on automated discovery of instances of non-compliant behaviour in a simulation environment via grammatical evolution. We illustrate the methodology with an experiment capable of discovering two known types of Ponzi schemes from a modest set of assumptions.

With the uptake of digital services in public and private sectors, the formalization of laws is attracting increasing attention. Yet, non-compliant fraudulent behaviours (money laundering, tax evasion, etc.)—practical realizations of violations of law—remain very difficult to formalize, as one does not know the exact formal rules that define such violations. The present work introduces a methodological framework aiming to discover non-compliance through compressed representations of behaviour, considering a fraudulent agent that explores via simulation the space of possible non-compliant behaviours in a given social domain. The framework is founded on a combination of utility maximization and active learning. We illustrate its application on a simple social domain. The results are promising, and seemingly reduce the gap on fundamental questions in AI and Law, although this comes at the cost of developing complex models of the simulation environment, and sophisticated reasoning models of the fraudulent agent.