Good investing is essentially about making good decisions. That means collecting information, organising that information, and finally evaluating it to produce a conclusion.
The problem that investors and asset managers are facing, however, is the sheer amount of information that is out there.
According to Goldman Sachs Asset Management there are now over 10 000 investable companies on stock exchanges in developed countries around the world, and another 3 500 in emerging markets. Taken together, these companies produce over a million pages of annual reports every year. They also host almost 40 000 earnings calls.
Properly evaluating all of that information is impossible, even for a large team of professional analysts. It has, therefore, become incredibly difficult for anyone to be a truly global investor.
Technology may, however, offer the solution. Artificial intelligence, machine learning and natural language processing are providing the means to organise all of this information, and even make decisions based on what is found.
The machines are watching you
Using computers to analyse data is certainly not new. It has been prevalent in other sectors, such as online retail for years.
“If you’ve ever shopped online you’ve been bombarded with other offers before, during and after your purchasing experience,” says executive director at Goldman Sachs Asset Management, Michael Rhodes. “The retail interaction with consumers is drastically changing and using data is helping retailers to recommend more products and services to us.”
By analysing a user’s search history and online behaviour companies such as Amazon and Google are able to create very accurate predictions about what each individual may be interested in. Netflix has also used this incredibly successfully to recommend shows to its users.
Investment firms are also using similar techniques to find ways of producing alpha.
“We use over 100 sources of data to construct specific signals that we think have predictive power to produce alpha in investing,” Rhodes says.
An example is analysing the amount of web traffic of online retailers.
“Web traffic data tells us when consumers are on a website, how often are they going [online], how long they spend on a page, and how often they are clicking,” Rhodes explains. “We find that sometimes we see meaningful spikes in this data, showing large increases in traffic on a website.
“If you’re a retailer, that could be a very strong predictor of sales,” he says. “For most people, if they want to know if a retailer is having a good quarter they have to wait for the sales call. We don’t have to wait. We get this data every day.”
Computers that can read
Natural language processing offers similar potential advantages. This is technology that enables computers to identify significant information in texts and speech.
“All analysts can read, but the problem is that there is too much to read,” Rhodes says. “We can, however, find patterns in words that show the view that people have of a company.”
Through the use of natural language processing, Goldman Sachs Asset Management can cover every earnings call, millions of news articles and hundreds of thousands of pages of research reports.
“We read all of them,” says Rhodes. “They all have predictive power about whether a stock will perform or not, but what’s really interesting is when you combine them. When they all think that a company is going to do well, it probably is.”
Making the trade
Goldman Sachs Asset Management uses this information to inform the decisions it makes on which stocks to buy or sell and how to put them together in portfolios. These final decisions are, however, still made by humans.
Elsewhere, asset managers are using artificial intelligence to go one step further, where the computers also make the decisions about what goes into the fund.
Locally, NMRQL Research launched South Africa’s first machine learning-powered unit trust last year. The NMRQL SCI Balanced Fund uses a series of algorithms to analyse a range of data and make investment choices based on this analysis.
While proponents believe that technology can be more accurate more often than humans alone, they are careful to make it clear that it can never be flawless.
“No prediction in financial markets can ever be 100% certain, so a solution needs to be able to measure the certainty of the model and provide feedback when change occurs or model error strikes,” says Stuart Reid, chief scientist at NMRQL Research.
Making use of this feedback is a key part of the strategy.
“The difference between quantitative investing and machine learning investing goes back to the ability to introduce a fundamental feedback loop between the market and the model, thus creating a mechanism where the model will constantly adapt at the sign of bad performance,” Reid explains.
This constant reevaluation of the models and what they are producing, as well as the fact that computers take the emotion out of decision making are two of the major reasons why supporters believe that these technologies can deliver an edge.
“We have been doing this for 30 years and have a very strong belief that the combination of data and technology is a fantastic way to gain an advantage in the market place,” Rhodes says. “It will be a consistent and persistent source of alpha for a very long time.”
Brought to you by Moneyweb