News Trading Strategies with Textual Analysis

Textual analysis can be broken up into two main groups: quantitative and qualitative. Quantitative texual analysis relates to numeric data published by governments and companies, such as economic data or quarterly earnings. Qualitative textual analysis relates to sentiment data in the form of news and social media. In this article, we will consider an application of the latter and discuss potential strategy implications of the results of a study of news-trading strategies using textual analysis.

Feuerriegel and Prendinger (2018) examine the use of textual analysis in a trading strategy by creating and comparing the performance of several models. As a benchmark, they utilize a time-series momentum strategy which places trades solely based on the rate-of-change of the price of an asset and a cross-sectional momentum strategy which considers a portfolio of 20 stocks, buying those with positive monument and selling those with negative momentum. Next, two rules-based strategies are created: (1) a news-sentiment trading strategy which uses a decision support system and places trades based on the sentiment of news releases; and (2) a combination momentum strategy and  textual analysis news trading strategy which uses price momentum and news sentiment signals. Finally, supervised and reinforcement learning models are used to improve upon the weaknesses of the rules-based systems.

Constructing a Textual Analysis Rules-Based News Trading System with a Decision Support System  

A decision support system (DSS) is a set of related computer programs and the data required to assist with analysis and decision-making. Feuerriegel and Prendinger (2018) use a DSS that transforms news content into trading decisions utilizing sentiment analysis. The basic design of a DSS for news trading is illustrated in the diagram below:

Source: Feuerriegel and Prendinger (2018)

The textual analysis news trading system uses a DSS to continuously scan incoming news and make trading decisions based on the sentiment score of the information, going long if extremely positive sentiment is detected and short if extremely negative sentiment is detected.

“When the news sentiment associated with [a] press release is either extremely positive or negative, this implies a strong likelihood of a subsequent stock market reaction in the same direction. We benefit from the stock market reaction if an automated transaction is triggered shortly before the price adjustment.”

Combining Momentum and Textual Analysis News Trading Strategies

Price momentum trading strategies have been shown to produce positive returns as the price of an asset will continue in the direction of an existing trend over time (read more about momentum strategies.) Feuerriegel and Prendinger (2018) combine their textual analysis news trading strategy with a momentum strategy using momentum to filter signals provided by the DSS so that a trade is only made when the sentiment of a news disclosure regarding a particular asset matches the asset’s existing price momentum. Thus, the combined system signals trades only when there is momentum in the same direction as the sentiment identified by the textual analysis news trading system.

Strategy Learning

The news trading strategy and the combined news trading and momentum strategy are both rules-based and therefore inherently lack the ability to adjust to arbitrary patterns. Thus, Feuerriegel and Prendinger (2018) look to supervised learning as a resolution. Random forests are chosen for this study but it is noted that any machine learning strategy would work, as machine learning strategies learn patterns from data and can handle non-linearity.

The authors recognize the potential dilemma caused by the fact that each trading decision influences the next. Therefore, they use a reinforcement learning method called state-action function to define the expected value of each possible action in each state. If the expected value is known, the optimal policy is given by the action that maximizes the optimal policy at the given state. From there, the action-value function is initialized to zero for all states and actions, and the agent successively observes a sequence of ups and downs from a historical dataset.

Results and Strategy Implications

The results of the textual analysis news trading system show positive performance that is statistically significant. Somewhat surprisingly, the results of the combined news trading and momentum strategy show lower performance than just the news trading strategy alone. However the combined strategy does offer lower volatility as a tradeoff. Both machine learning strategies achieved among the highest returns of all strategies studied in this paper. The random forest supervised learning model provided higher returns, but the reinforcement-learning model provided lower risk.

For traders looking to implement a news-trading strategy, including a momentum filter may be a way to reduce volatility. For those implementing a machine learning system to automate their news-trading system, a reinforcement-learning model may offer lower risk than a supervised learning model.


To generate your token:

  1. Register for a free practice account here.
  2. Log into the Trading Station and click on the account ID in the upper right corner.
  3. Click token management and generate your token.


Feuerriegel, Stefan, and Helmut Prendinger. “News-based trading strategies.” Decision Support Systems 90 (2016): 65-74  DOI: 10.1016/j.dss.2016.06.020

Risk Warning: The FXCM Group does not guarantee accuracy and will not accept liability for any loss or damage which arise directly or indirectly from use of or reliance on information contained within the webinars. The FXCM Group may provide general commentary which is not intended as investment advice and must not be construed as such. FX/CFD trading carries a risk of losses in excess of your deposited funds and may not be suitable for all investors. Please ensure that you fully understand the risks involved.