Annual reports are among the most important documents that companies produce in their reporting cycle. They complement many other disclosures that management make to shareholders and other stakeholders throughout the fiscal year. Shareholders, regulators, stakeholders, journalists and financial analysts all use them to assess firms’ health, prospects, and the quality of management decision-making. These reports contain a lot of information in a qualitative format (text). Until relatively recently, the volume of information in the typical annual report and complementary disclosures could be feasibly analysed by reading it manually. Most annual reports comprised 30 or 40 pages maximum in the 1980s and 1990s, however, over the last two decades their average length has grown dramatically and currently stands at approximately 34,000 words. Some reports are more than 300 pages long, containing more than 100,000 words. Analysing a single report is now a significant challenge; reading multiple reports, along with other company disclosures – as many investors and financial analysts are required to do – is practically impossible. This information-processing problem is exacerbated by the unstructured, non-uniform format of annual reports, making it hard to find specific items of information using manual search techniques. It is no longer feasible for an analyst or an academic researcher to read through hundreds (or even dozens) of annual reports and conduct manual analysis. Some sort of automated approach is required: this is where we come in. Working with Lancaster’s School of Computing and Communications (SCC) and Centre for Corpus Approaches to Social Science (CASS), as well as colleagues from the University of Manchester, we have developed an application to dissect and analyse narrative aspects of these reports. The Corporate Financial Information Environment – Financial Report Structure Extractor (CFIE-FRSE) application is designed to access and process large samples of annual reports disclosures electronically. It helps academic researchers perform largesample statistical analyses; but it also has applications for financial market professionals who conduct analysis using large datasets. CFIE-FRSE enables us to cut through hard-to-understand annual report language and aid users in identifying unusual patterns in corporate reports that may help to distinguish long-term financial strength from inflated shortterm profits. The challenge in designing this type of application is that it requires an interdisciplinary approach. On the one hand, computer scientists and computational linguists working alone can only make so much progress because they do not have the technical financial knowledge to build datasets to train and evaluate their algorithms. Designing an algorithm to extract and process complicated financial disclosures is more challenging than tasks such as measuring the general sentiment of social media posts and movie reviews. On the other hand, while accounting and finance specialists have the subject-specific technical knowhow, they typically do not possess the computing skills required to develop sophisticated algorithms and apply AI techniques. Our interdisciplinary approach brings together experts from computing, linguistics and accounting to build an application capable of extracting and analysing complicated financial disclosures. One of the questions we are addressing is: what makes a ‘good’ annual report? What are the language properties, structure and content that ensure an annual report provides useful information for users? Our findings are entirely consistent with accounting regulators’ assessments of high-quality reporting, and what linguists believe makes writing easy to read and understand. We find that discussion of strategy and business models is associated with high-quality annual reports; as measured by a quality award issued by an expert industry body such as the Investor Relations Society (IRS). Strategy is about how the firm creates and maintains value for stakeholders, the key thing a shareholder or stakeholder wants to know. Much of this information is forward-looking and therefore helps inform investment decisions: investors care more about what is going to happen to the business in the future than what has happened in the past. These dimensions speak very naturally to what we think of as being informative financial reporting and reflect current regulatory guidance on how management can make their financial reports more informative to users. On top of that, we identify a set of linguistic features associated with writing style and presentation of information that enables readers to understand a firm’s message more 14 |
RkJQdWJsaXNoZXIy NTI5NzM=