Artificial intelligence: Smart advantage
Global asset managers are committing more resources to the development of funds using artificial intelligence, writes Richard Newell. How will this affect portfolio management?
- A new wave of AI investment funds could disrupt the asset management industry
- Data scientists are in demand
- AI spans data aggregation and interpretation as well as semantic and correlation analysis, and predictive modelling
- This year’s equity market volatility impacted AI funds
- Human decision making is still important, even if AI funds are capable of independent decision-making
The asset management industry begun to recognise, perhaps belatedly in some cases, that it is likely to be massively disrupted by the introduction of artificial intelligence.
It is estimated that algorithms now account for 90% of financial market trading. Yet there are only a few managed funds that are fully using machine-learning technology. There are plenty of funds using artificial intelligence (AI) to churn out ideas, or refine their trading signals, but the exposure they take and the way they execute trades is still controlled by the human fund manager.
The next wave of funds is set to radically disrupt the industry. Progress with machine-learning algorithms, especially through the development of deep learning techniques, is producing a new wave of managed funds that are robotically controlled in the analysis, security selection and trading processes.
Global fund giants such as BlackRock, and hedge fund specialists including Bridgewater, MAN Group and Two Sigma have had exploratory programmes on machine learning capabilities for some time. BlackRock recently set up an AI laboratory in California and has centralised its data science efforts in a new team called Data Science Core.
Paulo Salomao, managing director at Accenture Asset Management, which tracks the financial industry’s AI evolution, says: “Many institutional investors and pension plans are still in the process of rolling out data lakes and robust data governance frameworks. The most advanced peers have mastered these skills and are looking now at deploying AI and machine learning to improve investment decision making, such as producing investment research, supporting security selection in private assets, driving performance attribution and portfolio construction, and talent management.”
The fund giants are hiring a particular type of professional – the data scientist – to help them develop and optimise algorithms. Analysis by Gaurav Chakravorty, CIO of Qplum, a US-based asset management firm that offers AI-based trading strategies, shows that the asset management industry typically takes in about $1trn (€848bn) in revenue, more than half of which is invested in human capital. In an effort to alleviate fee pressure, Chakravorty expects $100bn of annual investment will go into technology-driven roles in the next two to three years. “A typical asset management firm will soon have more machine learning engineers and data scientists than people with experience in financial markets,” he says.
The evolution of AI funds
The potential use of AI in fund management is broad. It reaches from low-complexity data aggregation and interpretation to highly complex semantic analysis, correlation analysis and predictive modelling that can directly feed into investment decisions.
Another recent development is the use of machine-assisted trading in less liquid asset classes or less efficient markets. These have the potential of higher information asymmetries and AI can be used to efficiently identify small but meaningful signals in the underlying noise.
In its current state, AI portfolio management shows potential, but the consensus seems to be that machine learning has only an augmenting role in the investment process at this stage. While AI funds are capable of independent decision-making, the human interface is still an important, if not crucial, element.
The future looks promising though, to judge from the performance of the next generation of hedge funds – a step on from commodity trading advisors (CTAs) and other funds using AI allied to human portfolio managers. On a five-year annualised basis, the average CTA has returned just 2.63%, while the AI hedge-fund universe has returned 10.71%, according to data provider Eurekahedge.
While results so far for fully AI-managed funds are encouraging, the volatility spike in the first quarter of this year was a rude shock for AI funds, according to Mohammad Hassan, head of research at Eurekahedge. “All of last year the VIX index was down below 10 and suddenly this year it shot up to 50,” he says. Whatever was happening in the AI portfolios was suddenly upset by this volatility swing.”
This recent rocky patch supports the argument that AI portfolio management is not sufficiently robust or adaptable to work autonomously. Despite the availability of historical data, these AI portfolios have not been through sufficient training in different market cycles.
It is early days then for fully-fledged AI funds – even in the US where the AIEQ ETF, which invests in US stocks and REITs and uses an AI model developed using IBM’s Watson, was listed last October. The fund’s methodology is based on taking price predictions on single stocks to build the model portfolio. The ETF has a year-to-date return of 5.54% (to mid-May), outperforming the S&P500 by 4.6% (see AI-powered ETFs vs S&P500).
Another US ETF promoter, Buzz, offers a US Sentiment Leaders fund, with the 75 holdings selected by AI technology, having analysed the most-discussed stocks on different media channels. Over 12 months to May, the ETF returned 21.49%, outperforming the S&P 500, which grew 14.63% in that time.
In March, through its iShares division, BlackRock received regulatory approval in the US for seven sector ETFs powered by machine learning. The funds use a new sector classification system which, using data science techniques, allows single companies to span more than one sector, reflecting changes in their business emphasis – Amazon is the example BlackRock uses – that traditional indices may not.
An example of the AI interaction that could help optimise portfolios is an application that can combat behavioural biases in portfolio management. Fintech company Capital.com says behavioural biases can severely impact an investor’s trading results: “Supposedly rational decisions may stem from mental shortcuts that ignore chunks of information, which can then have a significant impact on traders’ results.” Its AI software claims to identify investor biases and offer approaches to overcome them.
Artificial intelligence: Programs that can learn and reason like humans.
Machine learning: Algorithms that can learn without being explicitly programmed.
Artificial neural networks: Interconnected nodes, akin to the vast network of neurons in a brain, which enable a computer to learn from observational data.
Deep learning: Artificial neural networks adapt and learn from vast amounts of data.
Data lakes: A centralised repository for storage of structured and unstructured data at any scale using flat architecture as apposed to files and folders.
The impact on humans
Bridgewater CEO Ray Dalio recognises the need to address his own investment biases and uses his own well-publicised ‘principles’ to compensate for them. His firm has set up a technology division with the express objective of automating every process currently undertaken by people.
Dalio has predicted that 40% of all jobs in financial services are going to be replaced by algorithms over the next 20 years. Another prediction given by US firm Opimas in March 2017 is that the global asset management industry will shrink by 90,000 people (out of total industry figure of 520,000 ) by 2025. The upside, according to the Opimas study, is that 30,000 new jobs will be created for technology and data providers who respond to the financial industry’s new requirements and demands.
James Gautrey, global equities portfolio manager at Schroders says: “With sufficient data, successful early adopters of AI could enjoy a competitive advantage based on lower costs, time to market and insight. New entrants may even emerge if an industry is not moving fast enough. However, sustainable competitive advantage may need to come from proprietary solutions developed in-house.”
Aside from traders, other jobs that will be in jeopardy will include traditional quantitative analysts who use stochastic calculus rather than deterministic systems. But quants who can code will be in demand. Florian Spiegl, co-founder of Hong Kong-based fintech solutions firm FinFabrik is encouraged that the large bank asset managers are starting to put more resources into AI.
He says: “Managers who can code and are aware of machine learning analysis – that is all coming up. And a lot of CTAs are saying they now have something which appears to be faster and provides better signals, so they are incorporating that into their trading systems. I wouldn’t be surprised to see it overtaking traditional systematic strategies.”
Last June, Chinese fund-management firm ChinaAMC began a strategic co-operation with Microsoft to research in-depth the effects of artificial intelligence (AI) in financial services.
Although ChinaAMC says it is too early to reveal any outcomes from this research, it has been working with other technology specialists to develop AI-derived products. In March, it created a Chinese multi-factor managed account operated via artificial intelligence (AI).
The firm teamed up with Hong Kong-based fintech firm Magnum Research, whose Aqumon automated-investment engine was created at Hong Kong University with funding from the Alibaba Entrepreneurs Fund.
By using advanced AI techniques, ChinaAMC’s head of exchange-traded funds (ETFs), Frederick Chu, says the new strategy is able to identify the most effective factors to harness excess returns. Using a machine-learning stock-picking algorithm, it applies a smart beta-like multi-factor approach, making changes according to different market cycles.
The structure of the service points the way forward for AI funds in that the whole process – from data analysis to risk control and trading – is fully automated. That means the strategy can be offered at fees comparable with passive products in China, which are typically between 40bps and 75bps.
It outperformed the CSI 300 index during backtesting for the last two years, and in the first quarter of 2018 performance was up 1.43%, compared with a 5.14% decline for the MSCI China A index. Top sector picks were consumer and healthcare, according to Chu.
Kelvin Lei, chief executive of Magnum Research, accepts that a company with the best AI (say Google or Alibaba) could become one of the global asset-management giants of the future: “The technology around AI algorithms is mature and I think the traditional fund managers need to catch up.” There are no signs so far that they will catch up. Apart from BlackRock’s move to introduce a range AI-driven ETFs in the US (see main article) other big players such as Vanguard are resisting the rise of machines. State Street does not have any AI ETFs at present.
Larry Wang, head of marketing at CSOP Asset Management in Hong Kong, says his firm has been studying AI-themed products for a while and also plans to issue related products, adding it would be investing in AI-related companies.
The future for asset owners
If AI funds can improve their ability to make the big calls more quickly, more consistently and without the behavioural biases that humans are prone too, they will eventually gain the trust of investors. For example, if an AI has a conviction to short China’s property market – a tough call given the many variables – and such a call turns out to be correct, investors are going to start taking notice.
In any case, most buy-side institutions are looking seriously at AI and how it will impact their investment operations. Pension funds will soon have fully AI-managed portfolio options, which might play well in certain styles of investing, such as smart beta. In 2017, Japan’s Government Pension Investment Fund began a research partnership with Sony to study the impact of AI on asset management and how it might take leverage that. Among other things, the study is considering how to use AI-driven dynamic-factor analysis and scenario-based risk management.
Other than in fund management, AI is being used by pension funds in advanced client profiling, similar to the activities of e-commerce companies. We have already seen the emergence of robo-advice as the first widespread usage of AI in the client service side of asset management. As well as disrupting the traditional wealth management platforms, fintech experts see AI feeding in to the millennial generation’s interest in ESG and community engagement in general, which has implications for pension fund members.
In Europe, AI is being used to improve services to scheme participants. PensionDanmark has integrated AI in its critical-illness claims procedure and has developed a digital messaging service for members that harnesses prompts generated by AI.
ChinaAMC works with Microsoft to gain AI advantage
The question remains, what role will deeply human qualities such as intuition continue to play in a highly automated investment process?
Salomao suggests there will still be an element of the human factor required. For those bank asset managers his firm is tracking, he says: “Some of them have left the risk management, position size, trading all down to AI, and what the manager will do is from time to time, maybe once a year, tweak their model to incorporate things like potential market events caused by geopolitics.”
Spiegl says the AI funds of the future will be a significant improvement, as they learn to deal with the different market cycles. “The direction is clearly visible. Machine learning continues to gradually take over the entire investment process of quantitative funds. Machines will ultimately deliver higher speed, quality and precision compared to human investment managers, across different cyclical phases in markets.”