A cursory glance at the data would suggest that the headline above is a question to which the answer is no. For now.
The average constituent of Citywire’s Alternative Ucits Managed Futures sector has lost 1.8% over the past three years. The best, the £720 million DB Platinum IV Systematic Alpha Index – which is based on the Winton Trading Strategy – has returned 5.9% through the period. The worst, Man AHL Diversity, is 9.2% down.
But Pierre Gave, managing director of Gavekal Dragonomics, believes we would be wrong to bet against the machines for too much longer.
Trend-following strategies – which will continue to evolve, while survivor bias will begin to flatter the averages – are easier to program than asset allocation models, let alone successful stock-picking algorithms.
Gave accepts that attempts to develop automated systems along the latter two lines ‘have yet to show much success’, but maintains they could be ‘the next frontier’.
He identifies two stumbling blocks that have delayed progress to date. First, the input assumptions were inadequate.
‘When trying to program a computer to take independent decisions, programmers tended to go to experts in their fields and ask them to systematically explain how they came to a certain conclusion,’ Gave said. ‘The developers then coded this knowledge into their programs, which proceeded to fail miserably.’
That disappointment, according to Gave, lies more with the experts than the programmers: perhaps wedded to the idea that they practise an art rather than a science, they spouted truisms and platitudes. In investment, these would typically be about the importance of buying assets at the right price or focusing on quality.
Gave believes the developers have moved on now, though. ‘Basing an artificial intelligence (AI) program on human cognition was doomed to fail. Instead, AI coders now let the program look for patterns itself and set its own rules when trying to solve a problem.’
Evidence of success in doing so with investments is currently scarce, but Gave highlights how such processes have revolutionised medical diagnostics. ‘They look for patterns that would never occur to a human doctor.’
The second problem was insufficient data collection. ‘In order for an AI program to work successfully, it needs to be able to process the same data as a human. And up until recently, continuously feeding the program with new information was just too cumbersome and unproductive. However, today, with the arrival of cloud computing and big data, virtually all the necessary information is now available online and very easy to access.’
The secular trends of digitalisation and faster computers will only reinforce this. ‘The human information advantage over digital competitors is vanishing rapidly,’ Gave said.
He concluded: ‘We are still in the early days of AI implementation in finance. But we would be best served not to scoff at these advances, but to embrace them, however uncomfortable this may be.’
Indeed, Citigroup is known to be working with IBM on its Watson supercomputer – the one that famously won US quiz show Jeopardy! – to develop automated ways of recommending financial products.
Meanwhile, hedge fund giants Bridgewater Associates last year hired David Ferrucci, formerly IBM’s lead researcher on Watson, to help it develop its systems, which indicates just how seriously asset managers are taking the potential of AI.