Projects
Exploring quantitative finance through research and implementation
Fama-French Quantitative Trading System
5+1 Factor Portfolio Implementation
I built the Fama-French Quantitative Trading System as an automated portfolio management platform that applies academic finance research to real-world investing. The system continuously analyzes every company in the S&P 500, ranks them using six well-tested financial factors, and generates data-driven trade recommendations that can be executed through Interactive Brokers with full user confirmation.
This project represents my interest in quantitative investing and data-driven design. By combining financial modeling, algorithmic processing, and safe execution workflows, I created a tool that transforms complex research into a practical, transparent investment strategy.
How It Works
The system automatically collects daily price data and quarterly fundamentals for all S&P 500 companies. It calculates key financial metrics, market sensitivity (beta), size, value-to-book ratio, profitability, investment discipline, and price momentum, then computes a composite score for each stock.
Based on these rankings, it constructs a portfolio of top-scoring stocks and rebalances every quarter. When it's time to update the portfolio, the system generates clear buy and sell recommendations that I can review before confirming trades through the Interactive Brokers API. This setup keeps the process efficient but still under full manual control.
Research & Results
The trading logic is based on the Fama-French Five-Factor Model plus the momentum factor, a foundation of modern asset pricing research. I designed and validated the factor calculations using historical data for all 503 S&P 500 stocks.
- • Python with DuckDB and pandas
- • Yahoo Finance (yfinance) for market data
- • Interactive Brokers API (ib_insync)
- • Processing 500+ stocks in under 5 minutes
- • Managing 16 years of historical data
- • Explicit trade confirmation required
- • Paper-trading mode for testing
- • Fractional shares support
- • Diversification limits
- • Transaction cost tracking
I developed the project in structured stages: building the data infrastructure, implementing the factor engine, constructing the portfolio logic, adding a backtesting module, and finally integrating brokerage access. Each stage focused on modularity, testing, and clear documentation. The finished system processes the full S&P 500 dataset, executes complete rebalances in under ten minutes, and is organized in a clean, production-quality architecture.
Bachelor Thesis
Zip it: Connecting Proximity Bias and Home Bias among American Mutual Funds
This thesis investigates whether mutual funds systematically overweigh securities headquartered in geographic proximity and how this preference is influenced by the presence of national or linguistic borders. Drawing on a panel dataset that links mutual fund holdings to fund and firm headquarters over two decades, this analysis employs high-dimensional fixed effects regressions to isolate the effects of distance, country borders, and language barriers on both the extensive (participation) and intensive (weight) margins of investment.
The results confirm a robust and statistically significant negative relationship between geographic distance and fund holdings. However, this effect loses significance when controlling for persistent fund-security pair characteristics, suggesting that unobserved structural relationships may explain part of the proximity effect. Country borders show a significant reduction in holdings probability and security weights, while language borders are likely not big enough of a deterrent for information exchange among highly educated individuals to offer significant incremental, explanatory power. Additionally, the analysis of a change in distance effect likely suffers from some uncontrolled selection bias, due to the way funds select investments and unconsidered cutoffs to the distance effect. The findings highlight the importance of geography in institutional investment decisions, but also highlight limitations in the used data and modeling frameworks.
Erasmus School of Economics
Erasmus University Rotterdam
Dr. Esad Smajlbegovic
Second assessor: Dr. Theresa Spickers
July 15, 2025