Fintech final research project. The study explores how financially inexperienced investors can construct diversified and risk-managed stock portfolios using Large Language Models (LLMs) like ChatGPT, without relying on traditional financial theories such as Modern Portfolio Theory (MPT).
The research includes a comparative simulation of three portfolios:
- Portfolio 1 – Vague Prompt:
- Generated by ChatGPT with a general natural language input (e.g., “I have €1,000 to invest, give me a portfolio”).
- Portfolio 2 – Technical Prompt:
- A second ChatGPT-generated portfolio based on a refined, risk-conscious prompt that specifies a short-term investment horizon and moderately aggressive risk tolerance.
- Portfolio 3 – Human/MPT-Based Portfolio:
- Constructed manually by the researchers using historical performance data, applying MPT principles to create an efficient frontier–based, diversified portfolio.
📊 Tools & Methodology:
- All portfolios were built in Google Colab using Python.
- Financial data was gathered via the Yahoo Finance API (yfinance).
- Evaluation metrics include:
- Sharpe Ratio
- Correlation and Covariance Matrices
- Alpha, Beta
- Risk-Return plots
- Portfolio weights and asset behavior over time
- :
The goal is to compare diversification and risk, across the three portfolios, and understand how LLMs behave when faced with vague vs. technical investment prompts, as well as how their outputs compare to a human-built MPT portfolio.