Alison Thomas investigates how investors use technology to analyse information about companies and what management can do to make their reports easier to use
This article was first published in the May 2019 UK edition of Accounting and Business magazine.
The world of corporate communications evolves slowly. Partly due to restrictive reporting regulations, companies continue to report on their performance primarily using PDFs and PowerPoint presentations. Videos, the occasional interactive Excel spreadsheet and Twitter feeds have crept in. But, in essence, little has changed since the turn of this century.
In contrast, shareholders have long embraced the digital age. Faced with a mountain of data to sift through and analyse, they are early adopters of any technology that expedites a seemingly impossible task. They use data intermediaries to provide a one-stop shop on financials. Economic data series are on tap. Analytical and visualisation tools abound. So there is little surprise that the investment community has been quick to experiment with machine-based learning.
Examples of the use of advanced data analytics by investment professionals grow by the day. The UK’s Financial Reporting Lab’s recently issued report, Artificial intelligence and corporate reporting: How does it measure up?, classifies the various uses of artificial intelligence (AI) by market analysts into three categories: finding data sources, reading and structuring reports, and using algorithms to analyse data. Let’s look at each in turn.
The data sources that investment professionals use are expanding. According to a 2018 survey by Greenwich Associates published in Seismic Shifts: The Future of Investment Research, 70% of CIOs use non-traditional data sources or expect to do so in the next 12 months. ‘Two decades ago, we might have used 10 data feeds. Now it can be up to 100,’ explains David Wright, head of product strategy in BlackRock’s Systematic Active Equity team.
Examples of these non-traditional sources include:
- text and multimedia (including social media, blogs and photos)
- transactions (such as shipment records, point-of-sale records and credit card processing)
- sensors (including satellite information, GPS devices and crop sensors).
The way in which this data is used is limited only by the imagination. Some, for example, use mobile phone activity to monitor store traffic; others buy satellite data to track delivery trucks or e-invoice data to monitor product sales – anything that gives insight into companies’ performance.
Reports and databases
As for reading reports and structuring them into databases, a decade ago text analysis was a clunky affair, but not anymore. ‘Natural language processing has transformed the value of the written word,’ says Wright.
Text readers add meta-tags to unstructured prose to allow investors to tease out insight in a range of ways. ‘The earnings call is a goldmine,’ Wright adds. ‘We look at a wide range of indicators, including the tone of the presentation and how it differs from that used in formal filings. We consider the tense that they use and whether the language differs from that typically used by the company. It has become a really interesting way to spot companies that might outperform.’
The investment community has used algorithms to analyse data for decades. The difference today is one of degree. In the old days, models were relatively crude and based on a limited universe of data. Today they are non-linear, analysing the interdependencies between many disparate streams of data. This allows them to spot relationships that humans wouldn’t intuitively identify. Moreover, AI uses the results of previous analyses to refine models. This reduces the inherent bias and subjectivity of earlier approaches.
Today’s models allow analysts to more quickly identify investments that match their values, ethical standards or specific risk/return objectives. They can look for short-term trading opportunities and signals of unrecognised value creation. The number of applications developed in recent years is extraordinary.
So does this mean the end of professional investors? In 2018, Fidelity International surveyed 905 asset owners about the impact of AI on the investment profession. Of the respondents, 52% think that technology will replace many traditional investment jobs, with 40% predicting that AI will allow them to create portfolios without the need for asset managers. Nearly 70% think that technology will handle asset allocation, and almost half think that AI will be able to judge when to buy and sell shares. But few predict the end of the professional investor.
‘I don’t see myself retiring anytime soon,’ says Ben Peters, fund manager and director at Evenlode Investments. ‘Computers are good at spotting short-terms shifts in sentiment or pricing anomalies. But investing for the long-term takes a more nuanced approach, and computers don’t do nuance.’
Even in BlackRock’s quant-driven team, humans appear to be here to stay. ‘Computers can find spurious relationships between all sorts of unrelated things,’ says Wright. ‘That’s why our process is driven by humans; it’s our researchers that come up with the economic hypotheses that we test, not machines.’
The impact of the investment community’s commitment to digital data on corporate reporting is likely to be a slow burn. There is much discussion in the media about AI’s ability to handle unstructured data. Some speculate that this negates the need to move to digital reporting, such as tagging financial data using XBRL. At some point, the need to structure data may evaporate, but that remains a distant goal.
Need for structure
For now, machine-learning algorithms require clean and well-labelled data. For the very largest investment management companies, the process of adding structure to data is part of their competitive advantage. But for the rest, poorly structured corporate data is hard to find and consequently hard to use.
‘We need structured data,’ explains Peters. ‘In particular, we need globally consistent, comparable financial data. The standard setters have made some progress in that regard, but there is still some way to go.’
As for the narrative report, Peters sees plenty of scope to make that more digitally accessible too. ‘I’m torn. The structure of the 10-K [US annual report] lends itself better to the world of AI than the UK or European style of reporting. But the narrative in a UK report is far richer than that in the 10-K. Having said that, I see no reason why we can’t have the best of both worlds – good narrative that is well-structured.’
The message to the corporate community is clear: the more you can add structure to the information that you report to the investment community the better. The move to XBRL is a start. But it is only a small step forward. The data used in an AI world has been created with digital in mind; corporate reporting has not. It is largely paper-based and not comparable between companies. This is not sustainable and will change.
So companies have two choices. They can be proactive in creating a digital framework for reporting, or they can wait for others to dictate how they tell their story.
Alison Thomas is a consultant.
"The data used in an AI world has been created with digital in mind; corporate reporting has not"