How much do polypropylene (PP) prices follow oil – and can OPEC text sentiment help us forecast PP moves?

This project combines market data and LLM-based text analysis to study European polypropylene (PP) prices.
We:

• build clean monthly time series for PP, feedstocks (PGP), and energy benchmarks (Brent, WTI, NatGas)
• extract section-level sentiment from OPEC Monthly Oil Market Reports (MOMR) with FinBERT and GPT-4.1
• create a hybrid sentiment index (demand / supply / price-outlook) from 2019–2025
• test whether that sentiment helps predict PP prices beyond crude and PGP

Spoiler: crude and PGP do most of the heavy lifting; OPEC sentiment adds, at best, modest incremental information – which is still an interesting result.

LLM-polypropylene/
├─ data/           # Raw and intermediate CSVs (prices, OPEC text, bvse plastics reports, etc.)
├─ notebooks/      # 01–08: EDA, baselines, sentiment index, modeling
├─ src/            # Reusable Python modules (EDA, sentiment, modeling utils)
├─ artifacts/      # Cleaned datasets, model-ready tables, metrics
├─ plots/          # PNGs used in the report / slides
└─ scripts/        # One-off helpers (scraping, preprocessing)