Some asset owners have begun to formulate strategies to engage with artificial intelligence (AI) and machine learning, cognisant of how it promises one day to automate the job of active portfolio management and slash the cost of alpha generation.

For asset managers, the stakes are even higher because finding that killer AI application could yet shape tomorrow's winners and losers.

That's the theory anyway. So although the evolving new technology has yet to disrupt portfolio management in a big way, the industry's AI push is intensifying.

It's why the hiring of data scientists within the operations and investment divisions of asset owners and fund managers is accelerating. If US trends are any guide – and they often are – it is expected to be a major employment growth area in asset management for at least the next five years.

For institutional asset owners, the challenge is to evaluate how AI-managed funds will perform relative to those with mainly human input. Japan’s Government Pension Investment Fund (GPIF) has carried out probably the most detailed study to date as it pressures third-party asset managers to be more transparent, while holding them to strict fee schedules.

For other asset owners seeking to integrate AI into their investment processes, Nicole Musicco, senior managing director with Ontario Teachers’ Pension Plan, breaks it down into three key segments: investment, risk and enablement.

“Of these three areas, it is our view that ‘enablement’ has the most sweeping reach and therefore impact across the investment process. As the word suggests, it enables the investment team to get more out of their investment process from end-to-end,” Musicco said.

In its March 2019 report on AI in investment management, Massachusetts-based Meketa Investment Group echoed Musicco’s suggestion that the bulk of the potential value-add comes from incorporating machine-learning tools into investment processes at the portfolio level.

“The greatest amount of value-add is likely to result from an adoption that is less dependent on manager selection or a specific asset class, but rather a broad implementation that can encompass the entire investment process of the organisation,” it said in the report.

Paul Sandhu, head of multi-asset quant solutions at BNP Paribas Asset Management in Hong Kong told AsianInvestor that asset owners are now asking ‘how does AI help me assess my risk?’.

“In the current market situation, it’s all about hedging the downside. Whenever we start talking about our quant algorithms or safety strategies that measure volatility spikes, investors want to get into the detail of that. What are the models telling us? What are the risk metrics that we are using? And how does that data feed into portfolio management decisions?”

Where investors used to insist on human involvement in that process, Sandhu said they are now happy with a very automated system of controlling that downside risk. “That is moving us in the right direction, because much of the time, you want something that is emotionless and does what it is supposed to do in a given situation.”

ASSET MANAGEMENT RESPONSE

Asset managers have the potential to exploit the AI opportunity because they have massive amounts of data to call on. Chi Kit Chai, head of capital markets and chief investment officer at Ping An Asset Management in Hong Kong told AsianInvestor that he sees asset managers in the near-term trying to harness quality data to catch up with the technology advances to date.

“Asset managers need to master both the engineering aspect and the management of advanced technology in order to succeed. Very few can do that. In the long-term, a few winners will emerge and dominate the market,” he said.

Chai said Ping An has deployed deep-learning AI technology to extract high-level features and non-linear patterns in data to generate alpha. Deep learning is based on artificial neural networks, a class of machine-learning algorithms that uses multiple layers to progressively extract higher level features from the raw input.

“Alphas and information ratios (driven by the AI) have been good and consistent. The models also produce good results in markets that traditional quant strategies have failed. Our non-linear AI techniques can extract features and information that traditional linear quants cannot. After all, the world is not linear, by and large,” he added.

CONVERGENCE THREAT

A less promising outcome for asset managers suggested by the GPIF study is that as AI systems continue to learn from data reflecting market characteristics, there may be a convergence of strategies over time. So there is a high possibility that AI trading systems, based on deep learning or other statistical machine-learning techniques, will eventually approach index-trading behaviour.

Sandhu said that could potentially pose a big problem for asset managers, noting the way markets have behaved over the past decade, with investors “rewarded for being concentrated in a certain asset classes” and diversification seen as a “penalty rather than an advantage.”

“One of the reasons for that is there have been certain correlated asset classes, especially in developed market equities, that have just outperformed everything, with very little volatility,” he said. “So if you tried to do something innovative around US equities, for example, by factor investing, seeking alpha, timing different sectors or types of stock, it’s hard to pull off a better return than just buying the broad market.”

That does a couple of things: It shrinks the potential for alpha because beta is “crushing alpha” and also raises the possibility of regression models trying to optimise returns over history treating the idea that beta performs better than alpha almost as fact.

“What could end up happening if you don’t put enough thought into the AI mechanism is that the AI will behave like an indexer, because it thinks that is where the performance is going to come from,” he said.

The latest edition of AsianInvestor Magazine contains a special report on technology in investment management.