Asset Publisher

Fintech Research Paper by Consob entitled "A machine learning approach to support decision in insider trading detection" has been published (5 December 2022)

The study concerning two artificial intelligence methods based on unsupervised machine learning, suitable to support the supervisory activity aimed at identifying potential cases of insider trading, has been published in the Consob series of Fintech Research Papers.

The identification of potential market abuse cases is a complex and challenging activity, due to the huge volume of data to be processed and the variety of factors to be taken into account in the assessment of the operational behaviour of each investor. The aim of the analysis is to evaluate the possible continuity/discontinuity of the behaviour and its magnitude, in absolute and relative terms.

The study, developed on anonymous data sets, describes the characteristics of two proofs of concept that, using artificial intelligence methods based on unsupervised machine learning, could be useful in supporting - once testing of the prototypes is complete - the preliminary analysis for identifying parties suspected of insider trading behaviour. This may be followed by an investigation aimed at collecting further useful elements to hypothesise individual cases of abuse.

The first model uses a clustering analysis method. This method enables the identification of those groups of investors whose trading activity within a short time from a price sensitive event is not only carried out seeking reward, but also characterised by operational discontinuity both compared to the prior trading history and to the typical operations of the group to which they belong.

In particular, the clustering analysis first elaborates the trading model of each investor based on selected quantitative parameters (net balance of purchases and sales, trade concentration and exposure). By developing these parameters, the methodology, called k-means clustering, identifies homogeneous groups of investors with reference to a specific time horizon. Finally, by analysing how the position assumed by each investor changes over time, the analysis identifies the parties that, within a short time from a price sensitive event, appear to have changed their trading behaviour (so-called discontinuous investors).

The second model aims to identify small groups of investors that act seeking reward and in a synchronised way within a short time from a price sensitive event (so-called insider ring).

The methodology used, called Statistically Validated Networks - after characterising the trading activity of each investor according to three possible states (buying, selling, buying-selling) - builds a network of investors characterised by synchronous activity in terms of trading status and timing. Starting from a statistically validated network of investors, homogeneous groups of parties with similar activity are identified that have operated seeking reward with respect to a price sensitive event.​​​​​​