Thesis for Bachelor's Degree at Sapienza University of Rome
This thesis explores the design of trading agents and their evaluation through simulated trading on historical market data. Trading agents are systems that automatically perform online trading. Designing such systems is challenging because financial markets are highly dynamic and noisy. In this work we investigate how machine learning techniques can be used to improve trading strategies and decision making.
The thesis is organized into several parts:
- Problem analysis and background
- Introduction to algorithmic trading
- Overview of machine learning techniques used
- System design
- Design of software systems capable of using machine learning
- Architecture of trading agents
- Multi-asset trading
- Design of agents capable of managing multiple financial instruments
- Simulation and evaluation
- Backtesting on historical market data
- Analysis and comparison of the different approaches
For the Python implementation see https://github.com/sdcirri/thesis_project.