Every trade has its tools. A plumber uses a wrench, a farmer a plough, a dentist a drill. And quantitative financial analysts at BNP Paribas Asset Management? One valuable tool being used these days is a type of artificial intelligence known as machine learning. It’s more complex than a hammer, so hold on to your hats!
“Machine learning is a sub-category of artificial intelligence that uses algorithms capable of improving their predictive capacity while apprehending a given environment and providing information which we use to improve our decisions,” says Raul Leote de Carvalho, BNP Paribas Asset Management’s Quantitative Research Group Deputy Head.
For the sake of clarity, he punctuates this definition with a simple reminder: “Our decisions are about investing clients’ money in a way that meets their investment objective”.
To achieve that objective, quantitative analysts (quants) have to sift through the huge mountains of data which are now readily available to financial institutions and which might inform, however obliquely, good investment decisions.
Quants do much of this sifting with machine learning (ML) computer models designed to look for differentiating bits of information about specific investment targets such as individual companies or large groups of companies organised into investment funds.
While the term was coined in 1959, machine learning has been actively applied to finance only since the 1980s, when its value as an investment tool became clear.
In the early days, quants were limited to sifting through ‘structured’ or numeric data such as stock prices, company earnings or revenues, or macro-economic data such as gross domestic product (GDP) numbers for different countries.
However, the arrival of more sophisticated models now makes it possible to analyse ‘unstructured’ data as well, including human or natural languagesfound in electronic written materials such as online financial newspapers, company reports and financial statements – a virtually unlimited amount of potentially rich information.
The tricky part is to weed through all this information and come up with only the most interesting bits – a task Raul and his teammates perform by ‘training’ algorithms to search for specific ‘investment signals’ about the financial health of a given company or companies at a specific point in time.
In this way, natural language processing and machine learning can transform vast amounts of news into numerical signals that indicate to what extent articles published on a certain day transmit a positive, neutral or negative sentiment about a company, for example.
“When an algorithm indicates positive sentiment about a company, we can use another algorithm to test to what extent that positive sentiment might translate into higher future prices for the stock of that company,” says Raul Leote de Carvalho. In other words, the more accurately he and his team pinpoint quality investment signals, the better the resulting information on which they can base their investment decisions.
Once they’ve applied the ML algorithms to making the data more useful, BNP Paribas Asset Management’s QRG crew uses still other algorithms to classify that data into categories such that similarities and differences between the different data inputs are highlighted.
This step is critical because only by categorising the data and enabling its differentiation can you arrive at an informational advantage – or visibility on an investment target that is sufficiently different to add value.
“If you’re looking at a variety of metrics to determine whether a company is cheap or expensive – share price, earnings, price-to-earnings, forecasts – and they’re all telling you the same thing about its value, you have no informational advantage,” says Raul Leote de Carvalho. “Machine learning algorithms help us see when we’re looking at information that is too similar to be of use, and when we’re looking at differentiating information that is potentially valuable.”
After the QRG team has sifted through all the data and used machine learning to determine which financial assets portfolio managers should buy, it is faced with another challenge: how the portfolio managers should “size” these investments, or how much of each company’s stock or bond they should buy such that overall portfolio risk is kept to a minimum.
For example, for stock portfolios, at this point, the QRG team will have sifted through an initial 500 stocks or more to finish with a selection of 50 to 100 stocks. They already know they can expect attractive returns from all the chosen stocks. What they don’t know is how each of these stocks will perform in varying market conditions.
By applying machine learning, they can determine which of the stocks usually rise together and fall together. This information enables quants to help managers protect their portfolios by not spending all their cash on stocks that are likely to go down in price at the same time, in the same market conditions.
“That may sound obvious,” says Raul Leote de Carvalho, “but when it comes to investing in financial assets, there’s no such thing as ‘obvious.’ You’re better off using all the help you can get.”