Research Project Title:
Forecasting Daily Realized Volatility with News Sentiment and Intraday Patterns
abstract:We constructed a rigorous pipeline to train and test NLP algorithms on single stock and index volatility forecasts during prior UROPs, but we did not witness substantial increases in prediction accuracy due to model complexity and data restrictions, which resulted in severe overfitting. As a result, the objective of our SuperUROP study is to incorporate sentiment variables from news articles and social media into our advanced model prediction. Specifically, we want to include sentiment research from Brandwatch into our LSTM and GRU modules, as well as investigate more detailed 5-minute stock price data.
Through the SuperUROP program, I intend to improve my research abilities and deliver unique and significant outcomes. Unlike my previous ML class experience, where we learned about theoretical equations, the more hands-on approach in SuperUROP allows me to study and address problems that arise while dealing with real-world data.