Financial Risk Scoring

Financial markets are flooded with news, and hidden within that text are clues that can signal major stock price swings. This project, in collaboration with Ned Davis Research, tackled that core challenge: how to systematically quantify market sentiment from text to help investors anticipate market drops.

As the project's technical lead, my strategy was two-pronged: build a robust time series foundation from scratch while simultaneously exploring advanced NLP and multivariate time series techniques. I began by architecting a structured data pipeline that transformed raw, messy financial articles into a clean, reliable dataset. This essential foundation was the launchpad for all subsequent time series analysis and modeling.

The core challenge was finding true signals in the text while avoiding the noise of 'clickbait' headlines. My approach was to build a hybrid model. I engineered a unique risk score by using NLP to analyze text at a granular level, and simultaneously developed a time series model to forecast the market's 'fear gauge.' By combining these two powerful signals, our resulting system significantly outperformed simpler sentiment-only models. Ultimately, this work provides a more stable and actionable tool for investors to better anticipate market inflection points.