Master_Thesis

The core of this work is represented by networks, specifically from a dynamical point of view. Longitudinal networks can be statistically modelled in several ways, most of them developed in the social sciences field. In this essay it is attempted an extension of the use of one of those models, namely the Temporal Exponential Random Graph model,to a different science field: finance. The TERGM is here applied to a stock correlation dataset built on the monthly correlations of the daily returns of thirteen listed tech companies. The model has been tested both from a predictive and a inferential perspective. It is presented that it is possible to model such networks through the use of a TERGM. The results though can vary depending on the selected period of time. Specifically the main issue, exposed through a rolling origin cross validation methodology, is the consistency of the predictive performance over time. Still, the model has shown satisfactory results over small period of time, such as one or two years, both in terms of prediction and inference and tested on an out-of-sample prediction.