In the late 1940s, Alan Turing first explored the idea that a computer could play chess, to offer up a good example of what these machines could be capable of.
Decades later, a computer for the first time beat a reigning world champion in a classic chess match when Garry Kasparov lost to IBM's Deep Blue in 1997. Deep Blue could evaluate 200 million moves per second. It was another good example of what a machine with artificial intelligence (AI) could be capable of.
Today, this potential is not lost on investors who are increasingly interested in how AI technology can give their trading strategies an edge in an era when high returns relative to the market – or alpha - have been difficult to come by.
While relying on computers for trading strategies is nothing new, with quantitative, or quant, funds mainstream more than a decade ago and automated high-frequency trading triggering a flash crash in 2010, we are entering a new era as AI has once again crossed a threshold in terms of power and accuracy, into what is called "deep learning".
This category of AI is currently very popular among Wall Street banks such as Goldman Sachs and JP Morgan and some hedge funds including Two Sigma and WorldQuant. Deep learning is the ability of artificial neural networks, inspired by the biological brain, to, well, learn. It is what underpins more mainstream applications such as speech recognition and self-driving cars.
The aim of this learning is to better forecast outcomes and help to remove a lot of risk from any activity – in the case of autonomous cars, the goal is taking the risk of an accident while driving down to zero.
Researchers have utilised neural networks to help with the early detection of skin and brain cancer as well as to interpret signs of foetal distress from ultrasound images. Weather patterns, energy grid management, cubesat missions and the effects of earthquakes on structures are all areas where the application of neural network technology has improved forecasting.
Banks and hedge funds expect this technology could soon be a game changer in financial markets where assets managed by systematic funds – those that analyse large amounts of historical data to formulate computer-based trading strategies – already hit a record US$500 billion last year after doubling over the past decade, according to Barclays. A survey from Deutsche Bank, released in March, showed that 79 per cent of investors are allocating funds to systematic strategies, up from 70 per cent a year earlier.
However, deep learning technology is not yet a mainstream investment tool, something which is being addressed by AI-led fintech start-ups such as Alpaca, Inovance and Watstock, the Singapore company founded by Carl Freer, who has been working within the AI sector for the past decade.
“AI is about pattern recognition. Forecasting and predicting more and more accurately. Whether in financial markets or oncology research,” says Mr Freer, giving an example of being able to accurately forecast if oil and gas can be found in one area based on a comparison of the seismic data from another area where you know hydrocarbons have previously been discovered.
Watstock has been developed utilising the technology behind IBM’s extreme learning machine Watson – ELM is a larger category of AI, of which deep learning is a subset. Essentially the Watstock platform runs very large amounts of data through its neural network – the deep learning bit – resulting in forecasts from which trading recommendations are then made. The models are continuously worked on to improve accuracy. The company says its platform can forecast stock market movements for up to 10 days with up to 80 per cent accuracy.
“Nothing, not even quant strategies are 100 per cent guaranteed with AI but the goal there is to remove risk,” says Mr Freer, who has been exploring the appetite for AI models among UAE financial services firms for the past six to nine months with a view to opening a representative office here.
Hedge funds have had a tough period in general in recent years but the first half of this year has been the sector's best since 2009, showing more consistent performance and with 12-month returns in the double digits, according to research last month from the alternative assets industry data provider Preqin. This uptick has not included systematic funds, however, which have not shown the same returns as discretionary strategies, Preqin says. Annualised returns for discretionary funds were at 13.28 per cent in June, more than double that of systematic funds amid lower volatility in European and Asian markets.
Still, Professor Jürgen Schmidhuber, of the Swiss AI institute, whose pioneering work on deep learning has influenced Google, Apple, Microsoft, IBM, China's Baidu and Amazon, says we are on the cusp of AI changing the financial services industry forever.
“Artificial neural networks are getting smarter through experience, and faster and bigger by a factor of 100 per decade per dollar, says Prof Schmidhuber. “That is, in a few decades we should have rather cheap computers with the raw computational power of all human brains combined, and it won’t stop there.”
His company, Nnaisense, has a German subsidiary called Quantenstein, which uses deep learning technology for automated long-term value investing.
“[At Nnaisense] we believe we can pull off the big, practical breakthrough that will change everything, in line with my old motto since the 1980s: ‘build an AI smarter than myself such that I can retire’,” says Prof Schmidhuber.
