Press releases represent a valuable resource for financial trading and have long been exploited by researchers for the development of automatic stock price predictors. We hereby propose an NLP-based approach to generate industry-specific lexicons from news documents, with the goal of dynamically capturing, on a daily basis, the correlation between words used in these documents and stock price fluctuations. Furthermore, we design a binary classification algorithm that leverages on our lexicons to predict the magnitude of future price changes, for individual companies. Then, we validate our approach through an experimental study conducted on three different industries of the Standard & Poor's 500 index, by processing press news published by globally renowned sources, and collected within the Dow Jones DNA dataset. Classification results let us quantify the mutual dependence between words and prices, and help us estimate the predictive power of our lexicons.
CARTA Salvatore;
CONSOLI Sergio;
PIRAS Luca;
PODDA Alessandro Sebastian;
REFORGIATO RECUPERO Diego;
2021-01-19
Springer Verlag
JRC120225
1611-3349 (online),
0302-9743 (print),
https://link.springer.com/chapter/10.1007%2F978-3-030-64583-0_16,
https://publications.jrc.ec.europa.eu/repository/handle/JRC120225,
10.1007/978-3-030-64583-0_16 (online),
| Name | Country | City | Type |
|---|
This document is only visible at the Commission level.
You are not authorized to publish or distribute it outside the European Commission.
This is a public document. You can share this publication.
Datasets
| ID | Title | Public URL |
|---|
Dataset collections
| ID | Acronym | Title | Public URL |
|---|
Scripts / source codes
| Description | Public URL |
|---|
Additional supporting files
| File name | Description | File type |
|---|