Abstract
This work presents a convolutional neural network for the prediction of next-day stock fluctuations using company-specific news headlines. Experiments to evaluate model performance using various configurations of word embeddings and convolutional filter widths are reported. The total number of convolutional filters used is far fewer than is common, reducing the dimensionality of the task without loss of accuracy. Furthermore, multiple hidden layers with decreasing dimensionality are employed. A classification accuracy of 61.7% is achieved using pre-learned embeddings, that are fine-tuned during training to represent the specific context of this task. Multiple filter widths are also implemented to detect different length phrases that are key for classification. Trading simulations are conducted using the presented classification results. Initial investments are more than tripled over an 838-day testing period using the optimal classification configuration and a simple trading strategy. Two novel methods are presented to reduce the risk of the trading simulations. Adjustment of the sigmoid class threshold and re-labelling headlines using multiple classes form the basis of these methods. A combination of these approaches is found to be more than double the Average Trade Profit achieved during baseline simulations.
Highlights
Despite suggestions that the stock market is not predictable [1], many investors and researchers seek methods that can provide market fluctuation predictions to aid investment strategy
A much larger training set would be required for general context relationships to be represented in self-learnt embeddings. These observations, suggest that non-static embeddings provide the best configuration because of their ability to be fine-tuned to the task in question and because a more general context of words is retained in the embeddings allowing for better application to both unseen headlines and new tasks
The largest single-day loss from an investment is 11.3%, the model predicts rðzÞmean 1⁄4 0:93 based on the previous day’s headlines. The effect of these incorrect predictions with high rðzÞmean, coupled with a significantly reduced number of trades, leads to lower performance metrics than in the baseline case at values of t [ 0:75. These results demonstrate the shortcomings of making predictions based solely on company headlines, as it is possible for the network to make a positive prediction with high certainty based on a collection of headlines but for a significant loss to be made
Summary
Despite suggestions that the stock market is not predictable [1], many investors and researchers seek methods that can provide market fluctuation predictions to aid investment strategy. Advances in machine learning (ML) and natural language processing (NLP) have led to a shift in focus from technical to fundamental analysis This new approach uses data such as news articles and historical stock prices and is based upon the efficient market hypothesis which states that an asset price reflects all available information [2]. Mittermayer [23] focuses on intra-day predictions, whereas long-term trends are briefly considered in the work of Ding [24] Methods such as support vector machines [22] and complex decision trees [17] remain popular for predictive tasks of this nature.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
