Each month, Ole Søeberg, a board member for Advice Capital’s Vision Fund, sends off his portfolio of companies to an artificial intelligence (AI) lab. The firm runs the price and volume data through an algorithm that compares the stocks against more than 20 years of market performance.
At the end of the process, the algorithm suggests how the portfolio should be weighted to provide increased alpha. The process is so complex that even the developers who designed the AI model do not understand exactly what is going on. But the results speak for themselves. ‘In an average month, it adds 15 basis points,’ Søeberg says.
He provides just one example of how AI and Big Data are changing how investors do their job. The industry is investing billions of dollars a year to acquire the best staff, computer systems and data sources. Investment giants like BlackRock, JPMorgan and Man Group have gone a step further, setting up their own internal AI research centers to steal a march on competitors.
None of this is entirely new for IR teams – the stock market has been dominated by computer trading for years, and keeping track of new industry data is part of the day job for an IRO. The scale of change presents new challenges, however. The machines and data are growing quickly in complexity. The task of answering that staple IR question – what’s happening to the share price? – is getting harder.
While the mention of AI conjures images of computers making their own decisions, its use in the investment process today is mainly as an additional research tool. In a study last December carried out by TABB Group, asset managers cite actionable insight as the most important benefit of AI (selected by 56 percent of respondents), followed by efficiency and automation (28 percent), better client experience (11 percent) and strategy selection (6 percent).
Within AI, the most important subfield for all industries – including the investment world – is machine learning (ML), which refers to the use of algorithms that can execute tasks without the need for human intervention. ML drives most of the AI technology in use today, from facial recognition to fraud detection and cancer screenings.
In that subfield, there is a further subfield that demands attention: deep learning. This is where the AI model contains many different nodes arranged as layers, called a neural network, which mimics how human brains process information. The sophistication of a deep-learning model allows it to extract insight from huge quantities of data.
AI Alpha Lab, the team that works with Søeberg, is using a deep-learning approach to try to predict future stock prices. Mikkel Petersen, chief investment officer at the firm, says he has to explain to potential clients that his firm wants to help them, not take away their jobs. ‘What we say to people is, What you have been doing for 20-30 years, we can now possibly do better,’ he says. ‘It’s a challenging exercise to present it as an empowerment of investors, and not as a replacement. We don’t believe AI will replace humans, but we do believe humans with AI will replace humans without it.’
The company pulls information from a variety of sources, including financial statements, macro indicators, share prices and alternative data, and runs it through a deep-learning model. The aim is to predict where stock prices will be over the next three to six months. Crucially, the firm also attempts to quantify how certain it is about its predictions.
‘The model is not only providing forecasts for securities,’ says Petersen. ‘It’s also telling us how good that forecast is. That’s unique, not only to AI, but also to investing in general. It wasn’t really possible to do this kind of uncertainty estimation just five or 10 years ago. We didn’t have the knowledge or the computer power.’
As with many of the most sophisticated AI techniques, it is not possible to explain exactly why AI Alpha Lab’s model makes the predictions it does. Petersen says that if investors want to be exposed to the new breed of forecasting tools, they need to get used to this lack of explainability.
‘We will never be able to explain these things – it’s simply not the way humans think,’ he says. ‘We cannot think in 1,000 dimensions. But maybe getting close to the truth requires us to think in 1,000 dimensions.’
In their search for an investment edge, investors are not just incorporating new technologies, but also ever-more data. While the Big Data revolution is old news, the hunger for new sources of information remains as fresh as ever. Much of the buzz today centers around alternative data, which refers to any data not traditionally used as part of the investment process – anything from Twitter sentiment to weather patterns and satellite photographs of car parks.
‘If you are a commodity-dependent industry, I can use satellite imagery to project crop yields, and therefore predict commodity prices,’ explains Paulo Salomao, managing director in the asset management practice at Accenture. ‘Or I can monitor oil tanker movements, or look at weather data. All of that information starts to add to the body of knowledge I have about a specific company.’
Sometimes a single piece of information proves to be crucial. In April this year a flight-tracking service identified a private jet from Occidental Petroleum landing in Omaha, the hometown of Warren Buffet. Don Bilson, an analyst at independent research firm Gordon Haskett Research Advisors, quickly released a note titled ‘What was Occidental doing in Omaha?’. The following day, it was revealed that Buffet had taken a $10 bn position in the energy company.
The use of alternative data was pioneered by hedge funds. In 2011 Derwent Capital Markets launched a fund that used tweets to guide investment decisions, though it closed just a year later. Today, institutional investors – whether they are quantitative, fundamental or hybrid in approach – are increasingly adopting alternative datasets as they look for new ways to eke out alpha.
A recent survey conducted by Greenwich Associates on behalf of IHS Markit gives a snapshot of the changes taking place in the industry. In a poll of more than 40 investment specialists, 74 percent say alternative data is starting to have a big impact on institutional investing, while more than half subscribe to at least one alternative data source.
IROs may worry about what unknown data streams are being mined by investors. Given that most information is publicly available, however, crucial insight is unlikely to come from just one source, says Salomao: ‘The trick is around triangulating those different sources of information to extract unique insight.’
Of course, it’s not just alternative data that is running through AI models. Complex algorithms are also being turned on traditional sources of financial information. An example is research automation firm Prattle, which analyzes just about every corporate communication released by companies, from earnings calls and regulatory filings to press releases and speeches by corporate executives. In total, the firm says it processes around 5 mn documents every day.
The idea is to build a unique lexicon for each company. With this information, Prattle attempts to predict how the market will react to a new communication as soon as it is released. What kind of statements might prompt a negative reaction? The crucial factor isn’t whether you have used certain phrases or not, says Evan Schnidman, president and founder of the firm. What’s more important is how changes in language can have a psychological impact.
‘Specifically, if a person tends to be fairly informal in speech and then suddenly shifts to a more formal language pattern, this is typically a sign that he/she is seeking to create distance between himself/herself and the subject matter,’ Schnidman explains. ‘This is typically tied to negative information and thus formal language often causes people to view information as more negative, even if it isn’t.’
As investors rely more on Big Data and AI, the job of the IRO is likely to get tougher. It will be harder to keep track of all the key data sources on your company and industry. And it will be more difficult to understand exactly why an investor is making a buy or sell decision. Indeed, with esoteric ML models, the investor itself may not understand why the decision is being made.
But these trends can also be harnessed for the benefit of IR professionals. In part, this comes from giving IROs access to the same tools the investors are using. Firms like Prattle, after building up a client base among hedge funds and institutions, are now offering their technology to the corporate audience.
‘Corporate clients often want to look at their own communications, as well as those of several competitors,’ explains Schnidman. ‘They particularly like our ‘core comments’ function because it algorithmically identifies the most salient remarks in any given communication, extracting a few sentences to provide context to the scoring.’
Some firms are also developing IR-specific AI technology, focused on areas such as investor targeting. IHS Markit is steering clear of algorithms with low levels of explainability, says Tony Fabrikant, chief technology officer in the corporate solutions business at the firm. ‘We don’t think IR professionals should take it as a given that this investor will buy this many shares of your stock, without understanding why the algorithm is making that prediction,’ he says.
With spending on AI and Big Data growing rapidly, IR teams must keep pace as best they can with the evolving approach of the buy side. A profession that prides itself on building human relationships will need to give increasing thought to how it interacts with machines.
‘As we engage more with machine algorithms, we in the investor relations community will need to understand what kinds of things these algorithms are interested in, and perhaps craft our message without changing the meaning in ways that are more digestible by the machines,’ says Fabrikant. ‘That’s the mid-to-long term evolution of the IR profession.’
SIDEBAR: Handing over control
While most investment processes today are AI-assisted, we can expect more AI-led decision-making over the coming years, says Tony Fabrikant, chief technology officer for the corporate solutions business at IHS Markit. ‘Some mainstream shops have experimental portfolios that are AI-led, though it’s not the majority of money that’s being invested in this way yet,’ he says. ‘But it’s definitely a segment to watch.
‘Before AI, some manually programmed, quantitative algorithms have been fully automated, so it’s reasonable to expect that some of these deep learning-based algorithms will also eventually be fully automated with no human intervention. Especially high-frequency trading – the whole point is to get there before the other guy.’
This is an extract from IR Magazine’s forthcoming guide to the use of AI and Big Data in the investment industry. It was published in the Winter 2019 issue of IR Magazine.