More information is available to the public today than ever before in our history. In an increasingly turbulent world, companies are facing ever increasing scrutiny – not only in relation to their finances, but also to their business practices, brand, and sustainability.
The use of AI and machine learning to interrogate corporate information is changing the way companies are analysed and evaluated, and in turn may be changing the way that companies convey their information, creating a 'feedback effect' which may present challenges of its own.
Using machine learning and AI to analyse corporate information
The analysis of publicly available corporate information is hardly a new concept; nor is the use of computer software and algorithms to assist in this analysis. Algorithmic trading, for example, was introduced to financial markets globally in the 1970s, and is now ubiquitous.
A more recent development has been the use of machine learning and AI – and specifically the use of natural language processing, text analysis, and computational linguistics – to automate and enhance the analysis of publicly available corporate information, which can include information contained in the media, companies' annual reports, public regulatory filings, and even information provided orally by management during investor presentations and conference calls.
The AI programs being deployed for these purposes are able to extract, identify, and quantify subjective information in a financial context using established lexica to classify sentiment, allowing them to pick up positive or negative sentiment, weak and strong modal words and instances where, for example, the phraseology used by a company conveys uncertainty or litigiousness. For information provided orally, AI programs have been similarly developed to analyse the linguistic tendencies, vocal patterns and vocal emotion used by executives and entrepreneurs to identify and quantify similar markers of sentiment.
'How companies speak when machines are listening': the feedback effect
The use of AI and machine learning is changing the way that information is used, and in turn, it is changing the way companies choose to construct and present that information. A recent paper produced by researchers at the National Bureau of Economic Research identifies and analyses the 'feedback effect' of increased machine readership on corporate disclosures (primarily in this case public filings made by companies to the U.S. Securities and Exchange Commission); that is, how companies adjust the way they use language (both in written and oral communications) in response to the knowledge that the information they disseminate to the world is going to be processed and analysed by machines on a far larger scale than by humans.
In their analysis, the authors found that AI readers employed by algorithmic investors have shaped the style and quality of corporate writing to suit the leading algorithmic lexica, specialised dictionaries of words developed by researchers and used by algorithms to identify different sentiments and tones. With knowledge of these lexica, companies – at least those that expect a high machine readership – appear to be adapting the language they use to avoid words that may be perceived by an algorithm as being predictive of, amongst other things, negative opinion, uncertainty, or legal liability and in doing so, trying to present themselves to their stakeholders in the best possible light (whilst still complying with their legal and regulatory obligations).
This 'feedback effect' may present new challenges in the way we analyse company information. Companies can now take advantage of their knowledge of how these algorithms work to adapt – or manipulate – the information presented to ensure that algorithms (and in turn analysts and investors) perceive a company in a certain light, in much the same way that search engine optimisation is used in the marketing and advertising industries. Understanding how the relevant machine works is critical to understanding how changing the input (in this case the corporate information) will affect the output (the perceived investability of the company).
How can we help?
With machine readership of online information steadily increasing, companies should be aware of how public information relating to their business may be construed by non-human eyes. Equally, those in the business of analysing corporate information should be aware of the trend towards adaptive language, and aware of the methods behind the messaging.
Data is the most valuable asset of the twenty-first century. Our corporate and data science teams are on hand to help you with your corporate data, and can work with you to identify where AI and machine learning may help your business. Contact us to find out more.