Link to the University of Bologna webpage for the course

Thesis

Starting from the research carried out by professors Calvano, Calzolari, Denicolò, Pastorello, we know that algorithmic pricing softwares powered by Reinforcement Learning can autonomously learn to collude. The purpose of thesis is to address the following question:

When and how is it possible to detect the beginning of tacit collusive behaviour between reinforcement algorithmic powered agents?

As you can read in the section dedicated to this project in my website, this is a a binary classification on a time series and the ML tool used to make forecasts is a Random Forest classifier on some engineered moving averages.

Classes

First year

Second year

Third year