AI in capital markets: living up to the hype?
According to some of the more conservative claims doing the rounds, artificial intelligence (AI) is going to end much of the insurance industry by powering accident-free driver-less cars, transform agriculture and eradicate poverty with smart farming, and even deliver a cure for cancer. By any of those measures AI should therefore be able to comfortably outwit money managers and deliver superior investment returns. Needless to say it doesn’t always work like that. For one thing, AI and its machine learning (ML) cousin have been around for quite a while now – probably in some form or other for over 50 years. So, what’s different this time around I hear you say?
Many previous initiatives bundled under the AI umbrella were probably more accurately described as computer, high-frequency, low-latency, co-location, or other trading innovations. They certainly leveraged breakthroughs in technology that provided new advantages to the traders and investors that embraced them. But they were still more focused on using the power and speed that IT could then deliver rather than using computers and the algorithms that drive them to learn how to conduct actions that mimicked those of the human brain.
The game changers now are threefold:
- The power and capacity to store data has plummeted in price.
- The sophistication of the analytics required to deliver actionable intelligence from the data has risen considerably.
- The data available for that comparative analysis has multiplied enormously.
The confluence of these developments has enabled data scientists to start applying specific solutions to a host of existing problems. Financial institutions, whether buy- or sell-side, face three broad challenges if they are to remain viable: costs, compliance, and customers. They need to operate at vastly higher levels of efficiency due to shrinking margins, demonstrate compliance with the rafts of new regulatory reform sweeping the industry, and meet the expectations posed by more IT-literate and demanding customers.
In addition, the market is overcrowded with lingering loyalties and outdated working practices. These disguise massive industry inefficiencies that are unlikely to be tolerated for much longer by a growing majority of investors and customers. So can AI come riding to the rescue?
For once at least, the hype looks to have some justification. We are already seeing AI and ML being used in tandem to automate many manual back office processes and drive post-trade processes towards the T+1 settlement and clearing (and faster) that regulators are now expecting. In addition, financial software providers have developed intelligent tools that take this a stage further. For example, Finastra offers a solution that detects and alerts business to errors created in the manual entry of transactions, preventing the need to correct costly exceptions and also avoid potential fraud.
Energy is also now being applied to the application of AI in the middle office where machine readable tools and speech-to-analytics capabilities are being deployed to both implement and oversee compliance with new regulations. Even regulators are catching on and seeing the potential benefits of issuing new rules that are written in ways that can be fully and unambiguously interpreted on-site at banks by machines.
But the final frontier has always been the front office and the pre-trade, decision-making environment. Can a machine really create, identify and deliver a superior trading idea, or portfolio management strategy better than a human being? Many have their doubts. The key however is to have the machines complement the human brain not replace it.
Some of the smart opinion in this space likens what is happening more to evolution than revolution. This is probably the case with much of what we are already seeing with the rise of the robo-advisers. Many of them are fairly low-tech and tend to offer fairly straightforward advice based on binary choices already indicated by individual investors.
For some time now, hedge funds have been deploying “black box” technology which they have portrayed as the “secret sauce” behind their investment approach and have consequently charged chunky fees and commissions for the privilege of doing business with them. The problem is, barring a very few, the magic doesn’t seem to be working. Fund outflows have risen to critical levels as the average return of hedge funds last year slumped to below 6%, less than half what a passive tracker exchange-traded fund (ETF) would have returned if it had mirrored the S&P 500.
On a recent webinar, I talked to Tom Nash, the CEO of MarketsFlow. His company has developed an AI/ML-led solution that optimises portfolio management returns. He told me that the combination of new technology with experienced market professionals can deliver sustainable outperformance. It is no secret that the investment management industry is placing serious bets on technology it believes will make a difference. If Nash is right, these technologies could not only change the industry but accelerate the transformation that is rapidly blurring the lines between the buy-side and its sell-side providers.
Time will certainly tell as the quants and algo traders were certainly in ascendancy until their inability to deal with the price volatility and liquidity drought that accompanied the 2008 financial crisis discredited them. They have staged something of a comeback in the benign, low interest rate, and steadily rising asset prices of the past five years. But is that sustainable and will they be robust enough when the next inevitable market hiatus occurs?
If Nash is correct, the concurrent application of integrated management of risk and volatility alongside a more conservative investment approach that optimises rather than seeks to magnify returns will prove the durability of these new approaches. Working on the basis that this industry, like many others, can actually learn from its mistakes, we could be on the verge of AI finally establishing itself at the sharp end of the capital markets. Certainly the machines now seem able to do so. Would the same be true of the people operating them?