By Daniele Grassi, CEO of Axyon.AI
Artificial intelligence (AI) and big data are dramatically reshaping the way in which the financial industry operates. Asset management, in particular, is being disrupted by AI’s ability to mine enormous amounts of data very quickly, which makes it possible for asset managers to generate alpha by reviewing both structured and unstructured data.
Technological innovation has been an ongoing focus for financial services for years. AI, however, represents a major leap forward, as this technology can offer vastly improved efficiency when it comes to the data-heavy tasks present within the industry. At the same time, AI allows firms to move beyond traditional methods of analysis and use more sophisticated indicators that rely on alternative data sources and new, machine learning-based algorithms.
The true advantage of AI is its ability to learn unconventional patterns in large quantities of data without being explicitly programmed to do so. While human processing did most of this work previously, AI makes it possible for asset managers to uncover new and complex insights and make connections that would be impossible for a human to identify.
For example, AI can obtain real-time inflation rates using the online prices of millions of items or estimate agricultural yields by analysing satellite images of specific locations. Asset managers can then use these findings to better inform their business decisions and asset investment.
Deep learning in asset management
AI is able to draw conclusions from disparate sets of data by using ‘deep learning’ algorithms that reproduce the workings of the human brain in processing data and recognising patterns. Deep learning has already achieved widespread success in computer vision, natural language processing, machine translation, as well as speech recognition, and is the most popular area of research in the machine learning field.
It is no surprise, then, that deep learning is being adopted to predict financial market behaviour. Several hedge funds already rely on this technology to receive predictive insights on key variables, which are then used to determine strategies for investment. Systematic macro funds, for example, are starting to deploy deep learning models for forecasting economic variables, such as GDP or inflation rates.
However, more generally, a growing number of investment managers have adopted deep learning practices to reduce human bias within investment choices. This is a hugely appealing aspect of AI in asset management, as it removes any unquantifiable ‘gut feel’ reactions and enables asset managers to base their decisions purely on data driven results.
Deep learning can also be used to identify patterns in other areas, such as economic data, to predict the performance of specific assets. These predictions can range from the very short-term, even just a millisecond into the future, to a more medium-term outlook, such as in several months’ time. The asset manager can also target different performance indicators, like market volatility and the rate of return on investment. AI can give a far better indicator of market outcomes on specific assets, and thus helps to inform investment decisions far better than human analysis ever could.
The true advantage of AI is its ability to learn unconventional patterns in large quantities of data without being explicitly programmed to do so.
Modelling the future
However, in order to remain a leading technology for asset management, AI will need to continually build upon itself. In particular, it will need to anticipate many different variables, ranging from market changes and investment in new industries to general socio-economic changes that impact investors. This is a new frontier for the asset management industry, and one that could yield massive returns if harnessed effectively.
Axyon AI is currently exploring how this process can be achieved using a class of machine learning models known as Generative Adversarial Networks (GANs). GANs can be used to simulate future market scenarios by having two neural networks work against one another. One neural network, known as the ‘generator’, produces fake market scenarios while the other, which is labelled the ‘discriminator’, decides whether that data is real or false. As the discriminator learns to spot the false data, the generator improves its practices to make the next set of data, or market scenario, even more realistic.
Practices like GANs are looking to provide a more sophisticated alternative to the well-known technique of ‘Monte Carlo simulations’, which have traditionally been used to simulate various sources of uncertainty that could affect the value of an asset or portfolio. While these have worked to help with stress testing and sensitivity analysis in the past, GANs could become far more effective at producing a more accurate image of the outcome, especially from the risk perspective.
Building a valid portfolio
It is not just in market analysis where AI supports asset managers, however; its application also helps develop a reliable portfolio for the investor. Currently, many investors are still reliant on the Markovitz mean-variance framework, which was established over 60 years ago and does not completely account for the analysis opportunities in today’s environment. When the Markovitz framework was created, data sets were rare and expensive. By contrast, there is now a wealth of information that investors can use to help inform their decisions in building a reliable portfolio.
As a result, traditional portfolio optimisation is no longer the only way of building an effective range of assets for investment. With AI now able to work with higher degrees of uncertainty and digest large quantities of data, asset managers can better estimate the expected return given a target level of risk.
AI actually enhances the work being done by humans, allowing managers to make better decisions and draw from more reliable data analysis.
With the far-reaching advantages that AI offers to those in asset management, it is easy to see why businesses may be concerned about how these innovations might impact the workforce. While there have been numerous reports on the negative impact of automation, AI has the potential to streamline processes and improve efficiencies within asset management.
Nevertheless, machine learning, AI and deep learning algorithms still require a degree of human oversight to contextualise their findings. While AI offers huge advantages to investors, there is sometimes the chance that its decisions can be based on correlations rather than causations.
For example, AI may draw a link between ice cream sales and sunburn, predicting that when more people buy ice cream, sunburns increase. However, a human understands that good weather will lead to more ice cream sales and people staying outside in the sun for a long period – subsequently leading to a higher rate of sunburn cases. As a result, AI will not remove the human element of data analysis altogether, but will instead support the data-driven, back office functions that have historically been at risk of human error.
Indeed, AI can help active managements in their challenge with passive ones. Machine learning and AI can provide more insight with less human effort, leveraging the increasing available data. Through this automatic analysis of large quantities of data, asset managers will be able to reduce management costs by limiting the manual analysis of data and fundamentally improve the organisation’s processes. In this way, AI actually enhances the work being done by humans, allowing managers to make better decisions and draw from more reliable data analysis.
AI already offers unprecedented benefits for many industries, but there is a wealth of potential for this technology in asset management in particular. As AI develops and improves even further, it will continue to drive innovation within the investment sector.
Although there are still some understandable concerns about the security of jobs with AI adoption, the reality is that top-skilled professionals will be augmented by this technology, rather than replaced by it. As a result, firms will be able to streamline internal processes and enhance the growth of the business, rather than focussing on data-heavy tasks that can be completed far more effectively by a machine.