Toward Overnight Stock Movement Prediction with Contextualized Embedding Incorporating News and Stock

Abstract

This paper targets to predict overnight stock movement by taking contextualized news and stock information into account, using the Pre-trained Language Model (PLM) that was recently popular in Natural Language Processing (NLP) field. We proposed a model in which, given a piece of news and a stock code, the model can predict its overnight stock movement by utilizing combined news-stock embedding. Such embedding consists of (1) the contextualized embedding that contains the semantics of such a piece of news produced by a language model trained on a set of news and its paired stock movement. (2) The contextualized embedding is produced by a PLM trained on the information of stocks. Moreover, we introduce news augmentation on multiple pieces of news for the input and study its effect, respectively.