How can capital-market-relevant information be extracted from qualitative text data? A new tool simplifies the analysis and identification of annual reports and competitors.
More Time for the Important Things
How can capital-market-relevant information be extracted from qualitative text data? A new tool simplifies the analysis and identification of annual reports and competitors.
Anyone who invests in individual shares wants to be well informed about the companies to which they are entrusting their money. Annual reports are a valuable source of information. They can provide information about the issues management is dealing with and the risks the company is exposed to. Investors can also find company-specific information on sustainability or digitalization.
“Reading hundreds of pages of annual reports every year is time-consuming and inefficient.”
However, much of the information was already available in the previous year and was already priced into the market. “Investors are therefore basically interested in new information,” explains Prof. Sebastian Müller, Professor of Finance at TUM Campus Heilbronn. However, manually reconciling two consecutive annual reports is extremely time-consuming. This also applies to searching for specific topics using word lists.
Qannual: Invest Efficiently Thanks to Quantitative Analysis
Thanks to advances in natural language processing, there are now new methods for addressing these issues. Prof. Müller is conducting research on this together with his doctoral student Christian Breitung. “In current projects, instead of word lists, we use modern methods that are able to take into account the semantics or context of a text with the help of machine learning,” says Prof. Müller. “These are pre-trained language models that can be adapted to different application tasks. With the help of these models, it is possible to identify those sentences in annual reports that contain semantically new information. They can also be used to assign sentences to specific topics, without the need for word lists. Combined, this makes it possible to identify new information on a particular topic.”
“With the help of these functions, annual reports can be analyzed much faster.”
To exploit the potential of the method, Christian Breitung developed the Qannual tool together with TUM alumnus Felix Alexander Müller. This gives users access to the current as well as past annual reports of more than 9,000 companies. In addition to annual reports, Qannual now also offers quarterly reports. Users can selectively display individual sections of an annual report. Sentences with semantically new information are highlighted by default.
Added Value for German Companies
Currently, Qannual mainly lists U.S. companies. In the future, however, companies from other countries will also be added. This could result in new opportunities for German companies. The tool offers several functions that simplify the analysis and identification of competitors. Using the “company finder” function, users can search specifically for listed companies that offer a certain product. Subsequently, the respective annual reports can be clustered according to specific topics using the semantic search.
“Thanks to advances in text analytics, however, completely new fields of application are also conceivable,” says Breitung. In addition to identifying companies with similar business models or risk profiles, it would also be conceivable to forecast sales using a text-based determination of market sentiment.
One of the hurdles to analyzing international capital markets is the nature of language. Multilingual models can be used to identify topic-related information even across national borders. This opens up completely new possibilities for research at TUM Campus Heilbronn. For example, Prof. Müller’s team can analyze whether investors from different countries price certain information differently.