One avenue some are exploring is that of so-called ‘alternative data’, a catch-all term that describes all non-traditional forms of market data.
The list of alternative data types is numerous and includes the following: data scraped from internet news and social media sites, search engine keyword search statistics, satellite imagery, credit card data, footage from strategically placed cameras, measures of retail outlet footfall, and point of sale data.
The use of satellite imagery provides an example of how investors use alternative data. One of its applications is in the field of commodity trading, where images of cornfields can help in the early estimation of crop prices; or images of retail car park spaces are used to estimate retail sales. This alternative data type even found its way onto TV: satellite images of factories were used by fictitious hedge fund manager Bobby Axelrod in the series ‘Billions’ to determine whether a manufacturing stock was worth buying or not.
In the beginning, alternative data was seen as marginal to mainstream finance and most early adopters were quant-based funds but fundamental shops are increasingly using it too, especially for risk management functions.
“It is becoming part of that workflow, and while it is early stages, I would say, there are examples of funds that have successfully adopted the use of alternative data, specifically in risk management,” says Armando Gonzalez, CEO of alternative data vendor RavenPack, which has developed a way of automatedly scanning the news and extracting sentiment statistics.
In terms of adoption, two-thirds of mainstream asset managers have set up some sort of alternative data program, according to a recent industry survey of 59 key players, although their level of development varies greatly.
The majority surveyed - some 39% - were still in the very early stages of adoption, dubbed “Starting” according to the authors’ arbitrarily defined 4-stage “Analytics Power Journey”.
In comparison, only 7% of the asset managers had reached the fourth, most mature, stage of alternative data usage called “Breaking New Ground”.
The process of going from the first to the fourth step in the evolution process takes time - on average 4 years, and alternative data comes with challenges.
But the wait is worth it: those firms with a mature alternative data department tended to be very positive about its value, stating they had “entered a period of sustained value creation,” said the survey’s authors’ Element22, a data analytics company, and Swiss investment bank UBS
Yet a not inconsiderable proportion of those in the initial stages had also seen some benefit, with 11% of those in the “Starting” category saying they had seen gains.
Characteristics of the different stages of the maturation process were as follows: in the first phase, companies would typically do the ‘foundation work’, which involved developing systems for importing and integrating new data and building teams with the necessary skill sets. This was also the stage at which ‘advocates’ for alternative data would attempt to influence hierarchy decision-makers to invest more in the project.
In the second stage of the process, called “Early Stages”, there was usually a marked increase in the process of determining ‘proof of concept’ (POC).
“In the early POCs, asset managers experience more misses than hits, but it is the few hits that build confidence and generate the support that leads to much-needed investment capital to build fully-fledged capabilities,” says the Analytics Power 2019 report.
The third stage, named “Middle of the Journey”, is normally characterized by a surge in investment and growth into the area, once POC has been established.
“Another 14% of participants, who are in the Middle of the Journey, have largely completed their data foundation and are beginning to realize successes with their trials and POCs,” says the report.
The characteristics of the final “Breaking New Ground” stage are firm organizational faith in the value of alternative data, and exponential growth in both size and scope of operations. At this stage "leaders have realized significant success with advanced analytics and alternative data and, as a result, are making orders of magnitude more investment in their programs."
Other findings from the survey showed alternative data use had increased from the previous year, and that 64% of asset managers were using multiple sources, with 20% using more than 10 data sources.
In terms of application, the most common approach - with 42% saying they adopted it - was to incorporate the data into ‘supervised algorithms’ where the final decision was discretionary.
22% of respondents used the data in a “semi-supervised” fashion, whilst “reinforcement and deep learning are each utilized by 19% of surveyed managers”, said the survey.
“Deep learning poses an additional challenge with regard to understanding and documenting the decisions made by the algorithm for governance and security purposes,” it added.
The area of alternative data with the most promise was natural language processing (NLP), in which machines are taught to read and process written text. This was seen as having a “very good potential to create value.”
Some asset managers were planning to aggressively increase the use of NLP through all the stages of the Analytics Power Journey, but the most growth was anticipated by those in the Middle of the Journey.
The House of Quant vs The House of Fund
One hurdle towards more widespread adoption is that alternative data is often large, unstructured, and requires quantitative skills to evaluate, which can be a bar to take up for many fundamentally-orientated funds.
If teams of data scientists are brought in to help prep the data, existing fundamental analysts can often feel threatened, or a clash of cultures can ensue, limiting dynamic synergies.
Gonzalez’s company RavenPack is trying to get around the problem by making its data more digestible from the outset by focusing time and resources on developing an intuitive front-end so non-quantitative clients can just easily interrogate the data as more tech-savvy quant types.
“You often get pockets of quants existing within fundamental funds but there is a lack of communication between the two,” says Gonzalez.
The more evolved asset managers such as BlackRock stress developing a well-functioning, multi-skilled, team as a key requirement for creating a successful alternative data department.
"Having specific (data science) expertise on a team is not enough," says BlackRock, "If a team is to be truly innovative, we believe there must be a culture of collaboration and constructive debate across these disciplines. Members — from senior to the most junior — must be encouraged to work together regularly to improve current data techniques and to test ideas in a beta environment, without fear of failure."
An example of a fund that has bridged the quant-fund divide successfully is Point72, which was set up by the famous billionaire fund manager Steve Cohen.
Point72 operates three distinct approaches to investing: a purely fundamental, a purely systematic, and a blend of the two.
In the blended approach, the fund takes the ‘fundamental intuitions’ of discretionary types and puts them through a process of rigorous hypothesis-testing by data scientists.
If they are proven to add value they pass into a ‘recipe book’ of strategies that traders then apply in a discretionary fashion.
“It is like a ‘systematic best ideas book’ put together by discretionary traders,” says Mathew Granade, the Chief Market Officer at Point72.
The fund focuses a lot on differentiating which functions humans are better at and which computers are better at and divvying up tasks based on the results.
“Our view is that in terms of really being able to interpret fairly nuanced and complicated situations inside a specific company, people are still really, really, good. There are other things that machines do really, well, but if you are going to meet with a management team and interpret a large set of data that has a lot of nuance and specifics to it, the people still beat the machines at that,” says Granade.
Whilst human analysts are good at picking up on subtle cues, “machines are very good at correcting for behavioral bias, portfolio construction, and trade execution,” as well as “quite repetitive, complicated math,” says Granade.
The fund is always trying to figure out how to “marry those two up, in a really smart way,” and Granade thinks this will probably characterize the “next wave of hedge funds.”
Will there come a time when computers completely replace humans in the investment process?
In a way, the process of automation has already begun, “we are already allowing machines to make decisions, in quantitative investing, specifically in high frequency and intraday trading,” says Armando Gonzalez.
Beyond the systematic realm, however, Gonzalez thinks computers will increasingly help analysts to be more efficient and productive, via enabling quicker sifting, prioritizing and organization of information.
“It is important to understand that AI and specifically natural language processing, as well as Big Data analytics, all of these tools are going to help us become more efficient,” says Gonzalez.
The RavenPack platform, which scans, organizes, and derives actionable statistics from online news sources, can save analysts a lot of time they might ordinarily spend reading low-value news sources.
“Think about how much work an analyst spends reading content, and how much time they spend sifting through documents that add no value at the end of the day," says the CEO.
Another more recent application of the RavenPack platform has been to scan and analyze internal content, such as that stored in a fund analysts’ email inboxes or generated through internal online chat conversations.
In a recent project with one discretionary fund, RavenPack found that by analyzing and leveraging internal content in this way they were able to create a completely new source of revenue, which outperformed the fund’s overall performance.
"Think about the number of documents that they (analysts) don’t read - look at their inbox, for example - studies show that more than 99% of a human’s inbox is unread, and you might be thinking, this is interesting, and you should look at that, but you don’t have any time to,” says Gonzalez.
For Point72’s Grenade, the possibility that computers might get good at even the tasks humans do well - such as reading subtle verbal cues and body language - is still only a relatively long-term reality.
When asked whether computers could completely replace humans, or whether human analysts would still be needed “indefinitely”, in a recent interview, Granade answered, “Indefinitely is a very long time. What I will say is that our thesis as a firm right now - over the next, call it, 7-10 years - is that it is people plus machines.”
Yet the reality is that computers are already able to read body language: a computer was taught to differentiate between different hand gestures and body movements using a 500 camera panoptic studio in 2017, in an experiment conducted by Carnegie Mellon University.
Great strides in natural language processing where RavenPack is leading the way now also mean computers can differentiate sentiment in written language, although computer analysis of verbal language remains relatively rudimentary.
Nevertheless, it seems only a matter of time before machines become sophisticated enough to interpret the subtle cues the human brain can read and eventually even take over some of the more nuanced analysis tasks, as described by Point72’s Granade.
When that time comes a new generation of investors could evolve, a generation of robot market wizards, who can effortlessly alchemize more gold from the markets.
Struggling Asset Managers can easily tackle Alternative Data with the RavenPack Analytics Platform, which includes sentiment data on over 250,000 individual entities and 6,800+ market-moving events, available on visual dashboards or via web APIs. Request a trial here.