Words matter: News sentiment analysis for financial markets

16 april 2020

By Dr Svetlana Borovkova, Probability & Partners and VU Amsterdam

….the news, and especially its “tone”, or sentiment, not only can change our perception of the reality - it can change reality itself….

Part I

News drives financial markets: the flow of fresh information and opinions, reflected in news items, is consumed by market participants who then act upon this information, buying and selling financial assets. News is a rich source of information, and the sentiment, i.e., the tone of financial news, reflects how people perceive financial markets and how those perceptions change over time.

News forms our perception of reality and, since we act on that perception, reality is -in turn- shaped by news. When we characterize elusive concepts such as “the state of financial markets” or “the health of the economy”, we should realize that the socio-economic reality is the result of human behaviour and interactions. Based on the available information, market participants – we - form our own perception of reality whereupon we act. In this way the loop is closed, and reality is driven by our perception of it. As a result, the news, and especially its “tone”, or sentiment, not only can change our perception of the reality - it can change reality itself.

A great example of this is a run on a bank: imagine you hear on the news that your bank is in trouble; your first reaction will be to start shifting your money elsewhere. Others will react similarly, draining the bank of cash and creating a run on the bank. As cash is the lubricant that allows banks to operate smoothly, your cash-strapped bank could now get into real trouble, regardless of whether the original news story was reliable. Negative media attention can be a self-fulfilling prophecy: by stating that a bank is not “solid”, phantom problems can become very real ones. Time and again, we see financial markets rise and fall on the tone of news and not only, or even not, on factual information.

News sentiment, or the tone of news can reflect upon an individual company or commodity, a sector of the economy, a geographical region or the stock market as a whole. The more assets we hold in our portfolios, the more difficult it becomes for a person to read and interpret all the news relevant to these assets. Some large companies are always in the news: for example, Apple or Microsoft can have several hundreds of news items dedicated to them per week. The situation is even more extreme for commodities: there are hundreds if not thousands of crude oil-related news items every day. And if our portfolio comprises many stocks (such as S&P500 or Eurostoxx 600 portfolios), then no human can possibly read and assimilate such volumes of information.

Over the past decade, there has been great progress in automated interpretation of text. Clever language processing algorithms, powered by fast computers, can now “read” and interpret large volumes of news in real time. This is called Natural Language Processing (NLP): it uses semantic analysis of text, specialized dictionaries and Machine Learning. Analysing and interpreting financial news requires specialist dictionaries related to financial markets: financials of companies, merger and acquisition activities, corporate governance and other such topics. Using such dictionaries for interpreting company-related news is relatively straightforward: a news item which “sounds” positive for a particular company, usually is positive, in that it has potential to drive up the value (and hence, the share price) of this company. The situation is much more complex for commodities: where, news has to correlate with supply or demand, and its sentiment has to be evaluated in relationship to them. For example, what sounds positive (such as “Peace in Middle East”) has potential to drive the oil price down, while negatively sounding news (such as “Oil rig explodes” or “Attack on Iraq’s oil fields”) will almost certainly drive the oil price up. However, commercial news analysis engines – such as those offered by Refinitiv (formerly Thomson Reuters), RavenPack or Bloomberg – are becoming more and more precise in interpreting financial news.

Even when all this clever news interpretation by NLP is done, the work of quantitative analysts is just beginning. In the next instalment of this column, I will tell you more about how we extract actionable sentiment signals from the big and noisy sentiment data and show how useful these signals are in practice.

Probability & Partners is a Risk Advisory Firm offering integrated risk management and quantitative modelling solutions to the financial sector and data-driven companies.