The news cycle moves too quickly for anyone to keep up with. But for credit risk management in financial institutions, staying on top of the latest news is essential for success. In this post, we’ll explore the technical implications of how to leverage news for better credit risk management. By harnessing news for credit risk management, businesses can more accurately assess risk, identify potential opportunities, and gain a competitive edge.
Context
As a credit officer, it is essential to assess the solvency of client companies in order to evaluate their ability to pay off their debt. This involves performing a thorough analysis of their financing file, also known as a “credit file”, including examining the financial situation of the company, the economic climate of the industry, the purpose of the financing application, and related news and events. The output of this analysis is a credit note or credit score.
However, researching news about a company to gain insight into potential factors that could affect its solvency is a laborious and time-consuming process. To overcome this, Risk Officers can leverage AI techniques to optimize the credit process. This article demonstrates how machine learning can be used to automate the traditional process, creating a streamlined and integrated approach to solving the problem.
Collecting News for Credit risk Management
Assuming you know the sources of news you are interested in, the first challenge is collecting and storing up-to-date news. In most cases, this implies working with multiple news providers or publishers. There are various ways to achieve this, such as scraping and subscribing to news feeds. However, working with a News Signal provider — a SAAS provider that pulls news from various sources, is an option as well. Examples of such providers that can help you stay on top of the latest news include Knowsis and Datayes.
Unpacking the News
Entity Detection
Commonly known as Named Entity Recognition (NER), the goal of this step is to identify any entities that appear in a news article. Specifically, in the financial news context, the entities can be companies, people, places, commodities, … etc. This is the first step in automating the analysis of news.
There are two primary approaches to NER. The first is to match entities from an existing database. This is relatively simple to do; all you have to do is tokenize (break down into words) your news article and then perform lookups against your curated database of entities. However, this method does have a downside — it would not be able to differentiate between cases such as the company Tesla and the scientist Nikola Tesla.
The second common approach is rule-based matching NER. In this approach, you define templates that allow for entity detection with the help of Part-of-Speech tagging. For example, the contexts around the mentions of the word “Tesla” can be used to determine whether it is referring to the company or the scientist.
Article Classification and Event Detection
With an understanding of the entities mentioned in an article, the next step is to understand why. This is the objective of article classification and event detection: to provide a general theme for the news piece and a more granular categorization for each entity mentioned, respectively. Together, these two processes differ yet complement each other, providing a comprehensive understanding of the article.
Article Classification
To properly categorize financial news articles, it is best to use a Machine Learning approach. To do this, you need to define a set of categories and create a training dataset. This dataset should include articles categorized by an expert. Additionally, the text data must be preprocessed, and features must be extracted before training an ML model. With this, you can expect a high degree of accuracy in categorizing financial news articles.
Event Detection
Considering the phrase, “Tesla acquires Twitter for $44b”, if you encounter it in a news article, it’s likely an acquisition event. To detect such events, a rule-based matching approach can be used, taking into account the specific jargon journalists use when referencing certain types of events.
This approach is much more efficient than trying to train a model to classify events, due to the vast difference in volume and granularity between the number of events and the number of article categories. For instance, it is very normal to have a huge news dataset with just 5 news categories but over 6000 event categories.
Article classification and event detection provide a two-level taxonomy for categorization out of the box. If you need a single level of categorization, either one of these processes should be sufficient. However, if you need a multi-level taxonomy, you can simply group similar events from the event level up to the article level. As an example, consider the case of “Tesla acquiring Twitter for $44b”; this could be as follows: business > acquisition and mergers > acquisition.
Scoring News
Attributing scores to articles is one of the most flexible parts of this process, as it depends on our desired outcome. We can score at the article level, event level, or both — although event scores are preferred, as we can use them to derive a score for the entire article. Let us now understand how these event scores can be calculated.
We will consider the impact that news has on an entity’s creditworthiness (credit impact) — be it positive, negative, or neutral. To do this, we get experts to define scores for each event category in our taxonomy.
We can repeat this process for risk impact or sustainability impact, creating a new set of scores for each event. This means that each event has credit, risk, sustainability, and any other scores required. We can then use these scores to create an article score, which gives us an overall idea of what is happening in the story.
In addition to the expert consensus evaluated scores, we can add more scores at the article and event level using sentiment analysis. These techniques do not require any domain expertise, as they are based on the nature and tone of the text.
It is important to note that you may not always have access to the full content of articles, depending on your news source or provider. However, many news providers offer a variety of scores out of the box that can be timesaving.
News Visualization
To bring it all together and effectively generate risk signals, we must provide a means for Credit Risk Officers (CROs) to visualize the insights derived from the news. The best way to do this is by offering a portal where CROs can search for their counterparties and gain access to relevant news and insights. For this to be successful, the following must be taken into consideration:
Intuitive interface: Risk officers are acquainted with news but not with scores. It is important to keep the platform user-friendly and explain these scores displayed with news articles. Ideally, you want to display as few scores as possible in order to reduce confusion.
Efficient filtering: With thousands of articles hitting the news every day, it is imperative that officers can find news about entities they are interested in easily. The way to do this is to provide filters that allow for searching the news feeds by time range, company, location, etc. A step further is to allow for these filters to be saved and notification events sent in a more automated manner.
News recommendations: This is a great nice to have that complements the news filters. As officers continuously get flooded with information, surfacing news that is most relevant to them is a huge time-saver.
Before rounding this up it's noteworthy, that the use of news for credit risk assessment is not a novel concept. However, recent advances in machine learning have opened up new doors for automated possibilities. From gathering news to visualizing the data, we now have the groundwork for utilizing news to improve credit decisions. While this provides a foundation, the next step is to explore more advanced use cases such as Event prediction.
In summary, utilizing news for credit risk management can be of great benefit to financial institutions aiming to expand and reduce risk. By leveraging news and accessing appropriate news sources, businesses can stay up to date on industry and global developments which could influence their credit risk management. By layering the right tactics and tools, businesses can be in an even better position to make smart decisions about their credit risk management and protect their business.
Thanks for reading, feel free to share and connect with me on LinkedIn.
