Natural Language Processing Based Business Sentiment Index
My Undergraduate Dissertation : A medium frequency business sentiment analysis tool for forecasting economic indicators in India using Natural Language Processing and econometric modeling.
The problem: Traditional survey-based economic indices are slow, infrequent, and backward-looking. Can NLP applied to business news produce a real-time sentiment signal that actually predicts economic activity?
Economic analysis: Legally scraped business news at scale and built a hybrid index combining RoBERTa-based sentiment classification, time-series econometrics, and a behavioural economics-informed policy-weighted evaluation framework that penalises missed downturns 3× more than false positives — reflecting how policymakers actually weight errors.
TL;DR: The index achieves r=0.77 correlation with Composite PMI, beats the OECD BCI on real-time tracking (F1 0.87 vs 0.82), and Granger-causes 1-month services PMI at p<0.05. First LLM-personalised sentiment index for an emerging market economy.
