Un'Applicazione Deep Learning alla Generazione di Dati Finanziari Sintetici

Financial institutions are generating an ever increasing amount of structured and unstructured data, whose access is heavily restricted mostly due to regulatory requirements, concerning the privacy of the parties involved, or due to business needs. Being financial data the subject of interests for different modelling applications, the generation of data capable of mimicking the underlying properties of the real data without allowing to be mapped back to its real counterparts could help overcome the restricted nature of sensible data. Such data is often called Synthetic, while the processes that produce it are referred as to Generative Processes, an active research field that has been gaining interest for quite some time. In the present work, one of the possible generative processes involving a Deep Learning application is applied to the generation of synthetic financial correlation matrices, which are then compared with their empirical counterparts, by means of well-known stylized facts of correlation matrices, in order to assess whether or not the generative process has properly captured the properties of real financial correlation matrices.