Finding correlations between publicly traded companies is a topic of interest for a variety of actors on the financial markets. Many financial institutions use it to predict asset returns, while the regulators want to the know how the risk of default will spread through the market in times of crises. The classical approach towards building such a network is to look at the various types of business links that exist between firms. This includes customer-supplier relationships, subsidiaries, financial loans and others. The only problem is that discovery of such links is extremely difficult since data on this topic is either outdated, incomplete or not widely available.
On the other hand, there is a constant stream of information available in news articles published by different media outlets. In particular, we can read about new business relationships being formed or old ones being broken apart on a regular basis. This offers an unprecedented opportunity to extract meaningful information from news and form a network of connections between companies on the basis of it. With the help of recent advances in natural language processing techniques (NLP) we set out to fill the gap and build the network of connections in an increasing order of sophistication.
The starting point of our work was to simply count the number of times two companies appear together in the same news story. Building on this, we imposed a time frame in which these counts are considered relevant, so that the network is always up to date. Next, we compare the news stories about companies based on their content as well. So if two companies are reported about in similar stories, but are not mentioned explicitly, we could still detect a link between them. There is no doubt that additional approaches exist for construction of network and this is part of our ongoing research. At the same time, we also wish to build some theoretical framework around this line of work, so we borrow structures from the area of random graph theory that help us formalise the notion of dynamically changing graph.
Once the network of relations is obtained, it is then possible to compare it to the relationship between companies on the financial markets. The stock data is a perfect fit to achieve this goal since there is a vast amount of literature studying the gradual incorporation of new information into the prices. Hence the research questions that we are trying to answer with this line of research are the following:
- Can we reconstruct the firm relation network from news?
- What is the relationship between the network of connections implied by news and the one seen in the financial markets?
- Can we use the news network to measure systemic risk in times of crisis?
While examining the network of connections to determine what is the financial risk of a certain company, we must also look deeper into the stories about each company separately. The diversity of media coverage about a single firm is a relevant factor, since different media outlets produce news stories that are highly differentiated and vary along many dimensions, such as focus, accuracy, depth and style. In order to quantify this difference, we used NLP to build a diversity measure that compares content dissimilarity across the largest news providers for each company using data from EventRegistry. With this measure we investigate what is the relation between the news diversity and firm’s exposure to systemic risk as gauged by beta. Specifically, we aim to answer the following:
- Does diversity of media coverage exert influence on beta;
- If yes, what’s the direction of the influence?
These two lines of research demonstrate how the vast amount of data that is available from the news can be used in different financial applications. The past few years have seen an increase in interest from researchers in this area, but the news data is still an under-explored source of new information and will be a relevant area for future research as well.
Miha Torkar is based at Jožef Stefan Institute 2016-2019, and his research project is Characterising Financial Markets from Event-driven Perspective (WP3)