ETD

• Data Analysis
• Digital Twin
• Finance
• Optimisation
• Stochastic Process

The thesis investigates the internal mechanisms governing the behaviour of financial institutions through a multi-scale quantitative approach. The work initially develops a forecasting model for the operational volumes of a bank, starting from a phenomenological analysis of historical time series. Due to the limited size of the available datasets, standard machine-learning methods were unsuitable, leading to a model based on recurring empirical properties such as seasonality and inversion around a characteristic level.
The common dynamics observed across different series suggested the existence of underlying mechanisms responsible for the macroscopic behaviour. The analysis therefore moved to a microscopic framework through the construction of a Digital Twin of a bank department responsible for mortgage disbursement. This model incorporates relevant internal processes, including technological equipment prices and interest-rate dynamics.
Particular attention is devoted to technological pricing, for which a Game-Theoretic model shows that the observed dynamics can emerge from equilibrium in price competition. Finally, the Digital Twin is shown to reproduce series similar to those observed empirically under the stated assumptions. The Model Predictive Path Integral (MPPI) method is then employed to solve the optimisation problem despite the system's nonlinear, non-analytic and discontinuous nature.