Multiclass Sentiment Prediction for Stock Trading

TL;DR

This paper develops a sentiment analysis approach for biotech stock news using crowd-sourced labels and machine learning, demonstrating that sentiment-based trading can outperform the market.

Contribution

It introduces a novel sentiment classification pipeline for biotech stocks and shows its effectiveness in generating superior trading returns.

Findings

Sentiment models achieved high classification accuracy.

Sentiment-based trading strategies outperformed market benchmarks.

Crowd-sourcing effectively labeled complex financial news.

Abstract

Python was used to download and format NewsAPI article data relating to 400 publicly traded, low cap. Biotech companies. Crowd-sourcing was used to label a subset of this data to then train and evaluate a variety of models to classify the public sentiment of each company. The best performing models were then used to show that trading entirely off public sentiment could provide market beating returns.