Statistical and machine learning for market abuse detection and limit order book modeling

This thesis aims at tackling two classes of financial problems using statistical, computational, and machine learning methods. The first strand of research investigated is related to data-driven approaches to support insider trading detection. The second strand concerns the modeling of the dynamics of limit order books (LOBs).In order to support decision in insider trading detection, we propose two unsupervised machine learning approaches for contextual anomaly detection: clustering and dimensionality reduction based. The approach based on clustering combines two complementary methods. The first identifies discontinuities in the trading activity of an investor in the vicinity of a price sensitive event (PSE) with respect to her own past trading history and to the present trading activity of her peers. The second method aims to identify groups of investors that act coherently around PSEs, pointing to potential insider rings. On the other hand, the approach based on dimensionality reduction follows the reconstruction-based paradigm and its only input is the trading position of each investor active on the asset for which we have a PSE. After determining reconstruction errors related to the trading profiles, several conditions are imposed and potential insiders are identified. As a case study, our methodologies are applied to investor resolved data of Italian stocks around takeover bids, provided by the Commissione Nazionale per le Società e la Borsa.Using the same data, we also empirically investigate the impact of COVID-19 mobility restrictions on the financial investor population. Our results indicate that a sort of regime change in its composition occurred at the time of the lockdown.Concerning LOB modeling, we aim to realistically model the market response to exogenous trades. We propose an explainable and non-Markovian version of the well-known Zero Intelligence model. We include the price dynamics in the sampling of limit order signs and we interpret this choice as the reaction of traders with reservation prices to the price trend. This leads to a concave price path when a metaorder is executed and to a price reversion after the execution ends, as empirically observed. We analyze in depth the mechanism at the root of the arising concavity, the components which constitute the price impact in our model, and the dependence of the results on the two main parameters, shedding light on the financial interpretation of our model.