Are investment managers ready for Artificial Intelligence?

ONLY 10 per cent of fund managers or analysts have used AI or Machine Learning in the past year, according to a 2019 CFA Institute survey of the adoption of Artificial Intelligence (AI) in the investment management industry.

People may wonder why investment managers are late adopters of AI. After all, we have seen many other industries recruit data scientists and engineers. It is doubly intriguing when we consider that fund managers are supposed to keep abreast of the latest trends, and most would agree that AI has significant potential to affect the value of the companies they invest in. Why then the hesitation in adopting AI?

The first wave of AI adoption

We would argue that investment managers were not incorrect to drag their feet. The reason has to do with the underlying technology. Many of the current AI applications can be traced to a technological breakthrough in 2012, when a Deep Neural Network model achieved a significant improvement in Image Recognition tasks. This kickstarted a wave of innovation in the field of computer vision. However, investment professionals do not deal with many images, and the easier decision was to outsource to third-party vendors rather than build internally.

Furthermore, applying AI to investment strategies turned out to be more difficult than expected, because problems in financial markets are fundamentally different from computer vision. A common complaint was that AI could explain past events very well, but the results did not generalise well to future unseen events. This made it hard to tell when AI worked and when it did not. As there was potential to be misled, practitioners had to maintain elevated levels of scepticism towards AI. As a result, the first wave of AI innovation passed by without disrupting business models.

A second AI wave is now underway

However, the situation is rapidly changing. A second wave of AI innovation is now underway which is different from the first. In 2018, researchers made a breakthrough where an AI model obtained a significant improvement in language tasks, kickstarting a wave of innovation in the field of Natural Language Processing. Importantly, this generation of models can be easily adopted by industry. Google released a powerful Natural Language model called BERT under an open-source licence that made it possible for anyone to build on BERT and adapt it to suit their specific use cases.

These developments made several investment managers sit upright. Fund managers spend a lot of their time processing textual data, such as emails, research reports, earnings transcripts, and sustainability reports. They could no longer claim such skills were not core to their business. Furthermore, applying AI to textual data was transparent. For instance, one had simply to read the underlying text to know whether the model was classifying sentiment correctly.

Being late to the game, investment managers find themselves playing catch up. At this point it is again tempting to turn to outsourcing for lower costs and faster speed to market. Vendors who invested early in AI found themselves ahead of the industry and could offer helpful solutions. However, we argue that core skills should not be outsourced as it eventually leads to a loss of skill, a costly dependency on the vendor, and an erosion of competitive advantage.

Fortunately, there is a window of opportunity to catch up. The impact of Covid-19 has been harsher on businesses than financial markets. As a result, industry revenues have not been pressured to an extent necessitating widespread cost-cutting. At the same time, labour market dislocations have increased talent availability. Far-sighted leaders would do well to ramp up AI initiatives during this breather. In as little as a year's time, we can expect to see successful firms distinguish themselves in AI adoption.

There's no best way to adopt AI

Having started the data science practice at my workplace, I am often asked how data scientists can help with AI adoption. The debate usually centres on the question of whether to jump straight into high-impact AI projects, or to prioritise low-hanging fruits that can be harvested without AI.

Unfortunately, there is no best way to adopt AI. The argument to jump straight into high-impact AI projects is driven by the compressed timeline of AI innovations. One oft-heard remark is that a month is a long time in the field of Neural Networks - many innovations can happen within that window. For example, no sooner had we completed a project to summarise our emails using BERT than we learn that BERT may soon be superseded by another breakthrough. Open AI, a company backed by Tesla founder Elon Musk, recently released GPT-3, a model with billions of parameters and easily surpassing BERT in terms of computational intensity. Such rapid developments make the best AI masterplan go awry. If we do not get our hands dirty on AI projects, then we may find ourselves left behind quickly.

On the other hand, the argument to reach for the low-hanging fruits is driven by practical considerations. In the early stages of a firm's journey towards AI, simple innovations without AI is often sufficient for productivity gains.

Good starting projects can be as simple as getting data scientists to support fund managers with data-driven tools. My colleagues and I are fundamental investors who embarked on a journey towards data-driven decisions. We utilise quantitative multi-factor models to quantify our senior fund manager's intuition and transform it into investment signals to select securities for our portfolios.

After converting data into signals, data scientists often encounter demands to present their findings in an app or dashboard format. Therefore, investment dashboards can be a natural second project. Our investment dashboards open the "black box" of quantitative investing to our investment teams for inspection. This transparency really helped to improve confidence in our quantitative models and turned out to be a key factor behind our successful effort to launch our first quantitative fund with Environmental, Social and Governance features.

Other managers use dashboards to integrate new insights into existing processes. For example, according to the Financial Times, UBS Asset Management found that companies that lost a vote on executive remuneration were much more likely to suffer poor share price performance. The finding is then integrated into its investment dashboard that is accessible to portfolio managers and analysts.

We expect investment dashboards to become increasingly popular. Data scientists can construct web-based dashboards without having to learn another programming language, thanks to the maturity of open-source libraries such as R Shiny or Python Dash. Data providers have taken note and are forming developer communities to provide more support to the initiative.

In our experience, simple initiatives like (i) leveraging more data to support decisions and (ii) utilising dashboards to disseminate investment signals are sufficient to boost productivity of most investment teams, without having to dive into AI.

The CFA report notes that a fifth of analysts and fund managers are undergoing training in AI or data analytics. This suggests significant growth in the field in the coming years. Significant productivity gains are likely for those who can integrate AI into their investment processes, with repercussions for those who cannot. It would not be surprising if investors started asking their asset managers, "How many data scientists do you have in your team?"

This article is adapted from a talk the writer gave to CFA Society Singapore in June 2020. The opinions expressed are his own.