A panel of experts from across the funds industry was asked to comment on the survey.
HELMUT PAULUS, CEO AND MANAGING PARTNER, QUONIAM ASSET MANAGEMENT
Many players are beginning to sense the tremendous impact digital transformation may have on the industry. Accessing information is easier than ever. If machines already ‘understand’ (e.g. ‘Alexa’) what humans are saying and have access to an abundant amount of information, it seems obvious at first glance that machines will be able to take reliable investment decisions in the near future.
Big data and machine learning are definitely not new. The amount, speed and diversity of information available has been increasing over recent decades. In consequence, a small number of asset managers have established themselves which are already systematically processing and analysing available information using state-of-the-art technology and a sophisticated infrastructure. Essential in this context is the identification of relevant, value-adding data, whilst the actual use of all data is less important. Commonly known as ‘quantitative asset managers’, it is not humans who ‘manually’ search for economically plausible and robust correlations but, rather, a suitable machine designed by humans. If such a machine is used in the right way, it is not only capable of continuously evaluating tens of thousands of securities with the newest data, it can even analyse its own forecasting errors, subsequently correcting itself. Thus, machine learning is not new either; what is new is the number of implementable concepts and their performance power – both have grown immensely due to available computing capability.
The industry’s hysteria surrounding these topics is more than understandable, bearing in mind that traditional managers, focusing on fundamentals, have ignored this development for decades, placing the competence of their ‘star’ portfolio managers above everything. Meanwhile, it is evident that it is impossible for humans to fully grasp the scale of global data and the cross-relationships of such data. Thus, on average, they are falling behind digitalised processes, which are becoming increasingly powerful. The industry is attempting to hastily perform a long-overdue paradigm shift, thinking ‘Alexa’ might be able to intelligently invest soon.
Yet this is far from true: most machine-learning technology and big data benefits appear far easier to new arrivals in digital asset management than they actually are. The complexity arising from the combination of volume, variety, and velocity of data should not be underestimated. In fact, the sheer amount of data (hence the term: ‘big data’) is both a curse and a blessing. Only those who have verifiable experience in digital or quantitative asset management will be able to reliably use the benefits of machine-learning technology.
JONATHAN HAMMOND, PARTNER, CATALYST
It is clear that asset managers are optimistic about the use of AI and the more adventurous are exploring possible uses across their firms. Whilst many envisage AI being useful for identifying investments, there is also recognition that AI could be used in areas that are traditionally labour-intensive and difficult to automate.
In general, appreciation of AI is improving as it moves into the mainstream. The survey reveals that the definition is expanding to encompass technologies that have not previously been associated with AI, namely robotic process automation (RPA) and, to a lesser degree, blockchain. Maybe this is part of the key to unlocking AI within investment firms. AI technologies to process text, speech, images and patterns in data are becoming more widely available but their use expects a level of expertise that is in short supply across the industry. As the survey notes, training these AIs is extremely data-hungry and not without its own problems, as IBM recently discovered when they incurred the wrath of the press for using Flickr pictures to train their facial recognition AI, seemingly without the consent of the individuals featured.
The Holy Grail of AI in asset management would be replicating the success of AIs such as DeepMind’s AlphaGo for investment research and decisions. However, deploying AI to optimise investment strategies is challenging. AlphaGo Zero had to play millions of games, sometimes winning, sometimes losing, before it became proficient. Replicating this training strategy with investment markets is almost inconceivable – the feedback loop is too long, potentially years, and the cost of losses could be high. Training an AI using historic data is potentially the only way. Unfortunately, this is not the end of the problem. Once trained and operational, not even the programmers of such AIs are able to comprehend the strategies they are using. Explaining an AI’s unusual investment decision, that may only pay off in many years, to a client or regulator would be challenging.
KEITH PHILLIPS, EXECUTIVE DIRECTOR, THE INVESTMENT ASSOCIATION
Artificial intelligence represents a broad technology category and, whilst there is no single accepted definition, it generally refers to a suite of technologies and modelling techniques that are enabled by adaptive predictive power with a degree of autonomous learning.
To date, AI applications within the industry have mainly centred on realising greater operational efficiencies across front, middle, and back-office operations. However, as traditional sources of differentiation become increasingly commoditised, AI has the ability to provide opportunities that extend far beyond cost-reduction and more efficient operations. Focus is already turning to using big and alternative data sets to generate additional alpha through better structuring of investment strategies, the application of real-time customer segmentation and content tailoring for better funds marketing and distribution. There is also opportunity for continued enhancement of risk management and compliance processes through machine learning-driven automated data analysis. The effect is that traditional cost centres can be transformed into AI-enabled service offerings and, in doing so, release valuable internal resources.
In deploying AI, firms need assess their current technology strategies, infrastructure, governance frameworks, operating models and talent. However, early movers will benefit from the long-term strategic advantage and ultimately capitalise upon the returns that can be achieved.
©2019 funds europe
