Reshaping text analytics – is AI a game changer for analysts?

Natural Language Processing (NLP), a branch of Artificial Intelligence (AI) has seen some of the most notable advances in recent years. In this blog, we explore the productivity opportunities presented by NLP in research and portfolio management and look at how organisations can leverage this technology.

1. Natural Language Processing (NLP), a branch of Artificial Intelligence (AI) has seen some of the most notable advances.
   
2. This new technology is expected to enhance our ability to create and solve.
3. Chat GPT has set new standards for the capabilities of NLP.
   

With AI now becoming a central part of our lives, it also brings a number of great opportunities, despite its relative lack of familiarity. Technology continues to progress and as it begins to surpass human intelligence, this new dimension is expected to enhance our ability to create and solve.

With larger amounts of digitised data becoming available to portfolio researchers the search for key pieces of information relevant to decision making has intensified. The typical investment professional is pouring through their investment universe – stocks, bonds, sectors, countries and needs to find the next solid investment idea. The arms race will be won by those who are equipped to make sense of the data more efficiently. Artificially intelligent agents are expected to be integral in our quest to solve complex workflows as they take bigger analysis and decision-making roles in more and more industries. According to research^[1], NLP will grow significantly in the coming years with projections putting that anywhere between reaching 45bn and 91bn in investment with a CAGR of ca 20-27%.

Chat GPT has captured the imagination of not only the technology industry and people worldwide but set new standards for the capabilities of NLP. With the capacity to store information and provide detailed responses to inputs, this latest innovation is one model that has the potential to do the work of what used to be multiple individual NLP models.

It has shown significant promise in understanding complex questions, being able to summarise texts whilst also holding a natural-feeling conversation.

With the advent of tools like AutoGPT, there is even further potential for tools to perform increasingly complex tasks, for example carry out a chain of activities autonomously instead of just responding to individual tasks. In the below, we will discuss some discrete tasks where NLP is able to help but AI agents will go further and be able to intelligently orchestrate a range of tasks in the future.

For a financial workflow, these are powerful capabilities. However, they are initially trumped by the challenge of information overload and pattern matching. Tools like GPT do not come directly loaded with the latest news without a significant engineering backbone to support them. Similarly, they are good at analysing focused pieces of text but struggle to compare those against a vast history or find outliers.

The systematic evaluation of text via NLP and large language models (LLM) brings with it several challenges.

Successfully implementing analyses like the ones described above require a range of NLP/Data Science & tools:

• Resilient and reproducible data pipelines (detect failures)
  
• Entity linking tools
• Backtesting frameworks
• Scalable compute
• ML & DevOps frameworks

Traditionally, equity research analysts spend a considerable amount of time reading company reports, news articles, and other sources of information to gather data. Reuters publishes ca 8,000 news articles every day and our news monitor aggregates over 300,000 news articles, so identifying the most relevant articles to focus can be challenging. In isolation, NLP tools can help determine whether an article is about a specific company and whether that article covers M&A activity and rumours. Collectively applied over 10 years of news history, the same can be used to determine what is the lead time between a rumour and a deal taking place. Further, combined with more deals data, it can help find a correlation between deal pricing and speculations in the lead up to the final deal or the long-term impact to the stock market price of the acquirer. All of these will help for better informed research and a view on the best timing to invest.

Extracting relevant themes that are bespoke for an entity and timely can significantly reduce the time needed to understand all the available content. 

Source: Workspace Signal Search

NLP can super-charge an analyst’s skills by allowing them to quickly filter through vast amounts of data. Topic modelling automatically assigns topics to articles. When done right, the topics can align well with an analyst’s needs. For instance, if an ESG analyst is looking for an investment in renewable energy, nuances between the exact type of ESG topics, is crucial.

Topics like wind farms, regulation, and physical risk, will narrow down 100s of thousands of articles to just a handful and allow the analyst to dig deeper into those more quickly. Additionally, sentiment analysis and fact extraction can further extract specific statements relevant to the analysis done – is the amount of carbon offset highlighted, does that compare well to similar wind farm output in the area or country, etc. Such information can significantly speed up the decision of whether to invest / promote this specific project.

NLP can also be used to extract insights from unstructured data sources, such as social media, news articles, and online forums. This information can be used to identify emerging trends and sentiment towards certain companies, which can be valuable in making better investment decisions. Data extracted via NLP can be used in AI clustering algorithms such as K-means or DBSCAN in combination with the contextual understanding that Large Language Models can lead to discovering relationships between investments from unrelated sectors or other groupings such as GICS.

Another category of text analytics is summarization and translation. Portfolio researchers can extract benefit from reviewing abridged versions of earning calls, emails and news assuming the right legal agreements are in place. Translation is a powerful tool in making research content available across regions.

Another area NLP can be a valuable tool is in the trade ideation process as it can help identify patterns and trends that may not be immediately apparent to human analysts. Examples include sudden spikes in mentions in social media or news, which will require high grade company linking tools and the ability to disambiguate entities – for instance ETH can refer to the cryptocurrency Etherium or the country Ethiopia or ETH Zurich. Additionally, NLP can be used to analyse historical market data and identify patterns and trends that may be indicative of future market movements. As there are too many articles to keep track of, automated approaches to spotting an outlier can help here.

With the ability to automate many tasks to understanding the voice of the customer, it is no surprise that the broad adoption of NLP tools is being encouraged throughout organisations and into products. The benefits of AI to client outcomes are clear and we are actively engaging customers and researchers to find the best way to help our customers.

Both Fundamental and Quantitative investors can benefit from the use of NLP methodologies. Keeping in mind these techniques can be applied proactively as alerts as well as interactively. Alerts can be expressed in multiple contexts from changing sentiment on a company to predicting a sector change. Typically, we see the combination of several AI techniques to achieve these more advanced research outcomes.

By modelling the discussion by its nature, analysts are often interested in specific topics and cases where both analysts and management are positive about the future prospects. In the below, the analyst has leveraged NLP-derived labels about “forward-looking” statements and “positive/negative/disagree” outlook. To drastically reduce the amount of relevant text to review. Proportions reflect 6 months of company earnings call transcripts.

On 12 Jan 2023. Americanas’ CEO Sergio Rial suddenly resigned from the company after announcing a R$20bn “accounting inconsistency”

Source: LSEG news and price data, May 2023

The ability to connect those with the companies associated can directly been linked to stock performance:

• 27 Jan 2016 Oprah Winfrey tweeted about her success with weight loss and strongly endorsed Weight Watchers. Stock price for the company jumped by almost 20%
  
• 20 Apr 2017, Tim Hrenchir, a local Kansas reporter, tweeted about a court case impacting the merger of Westar Energy and Great Plains Energy–Westar Energy dropped 4.5% post-market.
• 13 April 2017 Short-seller Marc Cohodes shared negative opinions of Equitable Group ($EQB) via Twitter, urging Canadian regulators to investigate the company. The company stock dropped by over 40% over the next two weeks.

• 22 Feb – 08 Mar 2022 – German utility provider Uniper SE dropped by 54% due to significant exposure to Russian imports and a number of energy plants in Russia.
• 12 Jan 2023 – Americanas’ CEO Sergio Rial suddenly resigned from the company after announcing a R$20bn “accounting inconsistency”.
• 01 Nov 2022 – Cineworld stock jumped by 170% on news of a successful bankruptcy settlement with landlords and debtors allowing it to borrow more.

We have seen tools which can significantly enhance the work of Research and Portfolio teams through NLP. We are excited to continue innovation in this space and deliver such tools to a wider range of our customers.

[1] https://straitsresearch.com/report/natural-language-processing-market

https://finance.yahoo.com/news/natural-language-processing-nlp-market-070000969.html