GF Securities (HK)
HongKong
As an Equity Derivatives (EQD) intern, I worked with the trading desk of a market-making team specialising in over-the-counter (OTC) derivatives, including swaps, options, and structured products. This exposed me to both the quantitative and practical aspects of financial markets, as I applied my engineering and programming background in a trading environment. Throughout the placement, I worked on several projects. I developed a machine learning model to predict the annualised basis rate and direction of China's CSI 300 index futures (IF) using statistical and machine learning techniques. I also wrote Python scripts to automate trading desk workflows, including Excel data processing, filtering with Pandas, and PnL calculations across multiple trade files, improving efficiency and reducing manual errors. My final project was building an AI agent that could calculate and analyze the profit and loss (PnL) of derivative trades, allowing my colleagues to query PnL and uncover trading insights. Beyond project work, I spent time shadowing traders, quantitative analysts, and sales colleagues to gain a broader understanding of the lifecycle of OTC derivative products.
My role as an Equity Derivatives intern at GF Securities (HK) combined quantitative analysis, software engineering, and applied AI to support trading desk operations. I contributed across three major technical projects: statistical basis modelling, automated PnL data pipelines, and full-stack AI agent development. I developed multi-factor models using Ridge Regression, Random Forest, and XGBoost to forecast the annualized basis rate and direction for China’s CSI 300 (IF) index futures. I engineered factors across volatility, investor flows, ETF dynamics, and short-selling ratios. To resolve severe multicollinearity, I built factor correlation matrices using Gram-Schmidt orthogonalization and PCA, while leveraging functools.lru_cache to reduce recomputation overhead during rolling window OLS regressions. To automate workflows, I built a PnL engine in Python using OpenPyXL, Pandas, and custom dataclasses. The engine tracked positions across multi-currency portfolios. I integrated the Yahoo Finance API for live market data, added pickle cache layers, and optimized runtime via multiprocessing and multithreading. Building upon this PnL engine, I architected an AI agent from scratch. I created custom tool wrappers (get_database(), sql_query()) around an underlying SQLite database, linking them to DeepSeek via an OpenAI API interface. Then, I deployed the system by connecting a FastAPI backend via Webhooks to Feishu (Lark) and building a bilingual Streamlit dashboard.
This placement has expanded my EIE knowledge by applying computer science and mathematical principles to financial systems in real-life. - Programming & Libraries: Advanced Python (Pandas, NumPy, OpenPyXL), dataclasses, caching (functools.lru_cache, pickle), and performance optimization via multiprocessing and multithreading. - Machine Learning & Statistics: scikit-learn, XGBoost, Random Forest, Ridge Regression, PCA, Gram-Schmidt orthogonalization, distance correlation, and OLS regression analysis. - AI & Prompt Engineering: LLM API integration (DeepSeek/OpenAI interface), system prompt design, tool selection architecture, temperature tuning, and output evaluation. - Backend, APIs & Databases: RESTful API development with FastAPI, Webhooks, relational database schema design using SQLite, SQLCipher, and SQL querying. - Frontend & Tools: Built interactive UI dashboards with Streamlit, integrated enterprise webhooks (Feishu/Lark), used Git, and explored reverse-engineering/memory-scanning tools. - Financial Concepts: Equity derivatives pricing, basis trading strategies, stock splits, continuous compounding, CAPM, and delta/minimum variance hedging logic.
Operating on a fast-paced trading desk required time management and prioritisation. I evaluated projects based on urgency and operational impact, ensuring everything was delivered promptly without compromising reliability. I showed initiative when working on open-ended problems. Seeing that the existing framework was overly complex for my team’s use, I took ownership to build a custom AI agent, independently researching unfamiliar technologies, data pipelines, and backend architectures. Communication was important too, as I tailored to different stakeholders across the firm. I communicated technical implementation details with developers, while shifting focus toward practical business value and user experience when presenting tools to traders and managers. Finally, I adopted an inclusive mindset by considering non-technical users in my tool design. Building accessible interfaces, such as a bilingual Streamlit dashboard, ensured my colleagues could seamlessly query PnL data regardless of their technical background. Shadowing traders and following macroeconomic news have deepened my overall market awareness and understanding of trading operations.