Artificial Intelligence (AI) is a hot topic in the asset management world with potential applications to portfolio management, trading and risk management. As active managers here at Lion Global Investors, we strive to harness the power of AI to achieve optimal portfolios for our clients. Our research has shown that Machine Learning (ML), which is an application of AI, can add value to our fundamental research processes and enhance alpha generation.
In this article, we would like to explain what AI and ML means and how Lion Global Investors is leveraging on this disruptive technology. But first, let us quickly define what AI and ML means.
DEFINITION OF AI AND ML
Figure 1: Machine Learning is a subset of Artificial Intelligence. Artificial Neural Networks, Decision Trees, Support Vector Machines, etc. are types of Machine Learning algorithms
Source: Lion Global Investors August 2022, Note: LASSO -Least Absolute Shrinkage and Selection Operator
The Oxford Dictionary defines Artificial Intelligence as the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.
Machine Learning is defined as the use and development of computer systems that are able to learn and adapt without following explicit instructions, by using algorithms and statistical models to analyze and draw inferences from patterns in data.
In short, AI refers broadly to the use of computers to mimic human behaviour while ML is a subset of AI referring to algorithms which enable machines to learn and improve to a reasonable degree of accuracy without specific instructions.
SIMPLE EXAMPLE OF AN APPLICATION OF ML
Different people may have different expectations of what AI or ML can do with respect to application to asset management. The more optimistic would expect ML models to run autonomously and totally replace the role of fund managers. However, current ML models are closer to advance automation of repetitive tasks (whether complex or not) without general intelligence or reasoning. ML models, at current stage of development, are not adequate to totally replace the role of a human supervisor.
Below is a simple application of ML to illustrate this.
Let us say we are building a model to forecast the selling price of a condominium unit in Singapore. We could have a training data set of 1,000 transactions over the past 3 years with features like the number of rooms, floor space, freehold or leasehold, distance from the central business district, etc. The model would then predict the price of a new condominium unit using those features. The live model would also be updated with data as new transactions come by.
The model could work well when things are at an equilibrium but not when there is a significant change in the environment. For example, if the government decides to rezone some location in this example, it could render older transaction data obsolete, and the model would become inaccurate till it gets updated with newer data. In this case, a human would probably be more suitable to make the forecast using reasoning and experience.
WHAT ARE THE ADVANTAGES OF APPLYING ML?
The key advantage of applying ML would be the machine’s ability to process large amounts of data and make inferences of complex patterns that an (human) analyst would otherwise be unable to do. The machine would also be totally objective, meaning that its recommendations or forecasts are without human behavioral biases like herd behavior, anchoring or other emotional traits.
There are different ML algorithms with different pros and cons for different objectives. (See Figure 3: Types of AI/ML Techniques applied to asset management). For example, Natural Language Processing, an algorithm suited to extract information from unstructured data source, can be used to meticulously track the frequency of key words, like “recession”, appearing on the news to serve as an indicator of investor sentiments. While other techniques like Artificial Neural Network and Decision trees can be applied to large data sets to forecast asset prices.
As explained above, ML does have its limitations.
ML models draw their training from past data and as such, ML models do not perform well when the investment environment changes drastically. Not to mention during black-swan events when humans would arguably be able to make better decisions. Also, ML models require high quality data to work with as analyzing poor data would only lead to bad decisions. It runs the risk of “rubbish-in, rubbish-out” if data is of poor quality. As such, the cost required to run an AI investment strategy could be significant. And this may be part of the reasons why AI strategies are generally only offered to high-net-worth investors able to afford the higher fees.
APPLICATION OF AI AT LION GLOBAL INVESTORS
Here, at Lion Global Investors, we apply AI and fundamental research to create new and innovative products. It was a natural evolution for Lion Global Investors and it took a number of years to progress from using smart beta in our portfolio construction processes to applying AI techniques. (LGI has been managing a Global Disruptive Innovation Fund using smart beta in the portfolio construction process for the past 5 years).
Our basic motivations for doing so are:
i) To leverage on the advances of technology to improve our investment performance, and
ii) To make AI augmented products easily accessible to the public.
We aim to democratize innovative financial products and contribute to the retirement planning efforts of the population here in Singapore.
Figure 2: Balancing the advantages of Humans and Machines at Lion Global Investors
Source: Lion Global Investors August 2022, Note: FM – Fund Manager, AI – Artificial Intelligence
To that end, we are investing resources to build our capabilities in Artificial Intelligence of Investing (AIOI). We have set up a dedicated AIOI team led by Ms Ong Ai Ling, who has over 16 years of investment experience and is the portfolio manager for the LionGlobal Disruptive Innovation Fund that invests in themes such as AI, Fintech and Gene Therapy. The AIOI team currently works on different ML models to complement fundamental research to help deliver smarter investment products that are affordable to the general population.
ARE THERE RISKS TO APPLYING AI TO THE INVESTMENT PROCESS?
The models developed by our AIOI team are not “black boxes” and these models are maintained by Ai Ling and her team of data scientists. The results from the AIOI models are also constantly reviewed not only by the AIOI team but also with the fundamentally driven investment team. Sector specialists would review the output against the fundamentals of the specific sectors to interpret the rationale for the models’ decisions.
The AIOI team would also be constantly cleaning up the data used as inputs to ensure the consistency of the data from our vendors.
The development of AI has indeed come a long way with advancements in ML algorithms, ever increasing computational power and accessibility of data. And we, at Lion Global Investors, are also benefitting from the advancement of this technology by applying AI to our investment processes. Our aim being to continue harnessing the strengths of unbiased computational power of AI and the reasoning skills of our experienced team of fundamental managers to offer new AI enhanced products to our clients.
Figure 3: Types of AI/ML Techniques applied to asset management
Source: CFA Institute Research Foundation, 28 August 2020