Financial Market Sentiment Analysis

Problem

Analysts follow far more news than they can read. A measure of the tone of news per company and day helps them see where coverage turns negative or positive and which headlines to read first.

The goal is a pipeline that scores financial news headlines for sentiment and aggregates the scores per ticker and trading day.

• Data: headlines with publication time and tickers from a news source whose terms allow automated processing.

• Constraints: financial language differs from everyday language ("beat", "short", "guidance"), so general-purpose sentiment tools are not enough; news after the market close belongs to the next trading day; syndicated copies of the same headline must not be counted several times.

• Scope: the result describes the tone of headlines. It is not investment advice and not a price forecast. Any use for trading decisions would need its own out-of-sample evaluation.

Goals

• At least 80 % agreement with a hand-labeled sample of 200 of your own headlines (positive, negative, neutral).

• Clearly better agreement than a general-purpose lexicon baseline (VADER) on the same sample.

• A daily table per ticker with headline count, mean score and positive and negative share, available within an agreed time after the market close.

• Every report built on the output carries the note that it is not investment advice.

Solutions

• News headline sentiment per ticker and day with FinBERT and a VADER baseline (FinBERT (ProsusAI/finbert), Hugging Face Transformers, PyTorch, VADER, pandas) – Scores the tone of financial news headlines and aggregates it into one row per ticker and trading day. The output is a sentiment time series you can chart next to prices or use as one input in your…