Automation, artificial intelligence (AI), machine learning (ML) and cloud technologies are fast becoming essential tools for the financial services industry – not least within asset management and securities services. These tools are already being used to enhance a variety of processes, from front-office activities to compliance and operations in the back office. They promise to usher in a new age of competitive services and efficient processing based on digital intelligence.

However, realising the transformational potential of these technologies requires advance planning and preparation. The importance of ensuring that AI, ML and cloud systems are properly governed – based on good knowledge of the underlying technology, rules, ethical standards, and human domain expertise – cannot be overstated. It is only with this proper foundation that market infrastructure participants, from broker-dealers to global custodians, can harness the full value of automation.

To take a closer look at where automation is heading in securities services, four industry experts joined Mathew Kathayanat, Head of Product, APAC at Deutsche Bank, for ‘Unlocking the value of Automation in Asset Management’ the first webinar in the Securities Services APAC leadership series hosted by The Asset.

Recent ‘black swan’ events, such as the Covid-19 pandemic, have accelerated asset managers’ plans to address key areas where technology could help increase the efficiency of operational processes, boost portfolio performance and improve risk management. The impact can already be seen throughout the industry.

The speed with which AI and ML can produce meaningful analytics from historically unstructured data is an exciting gain to many in the asset management industry. For firms seeking growth by outperforming the market, big data – and particularly so-called “alternative data” – can offer new strategies for generating alpha. Alternative data is information used by investors outside of the conventional sources offered by corporations and established industry data vendors. It can consist of anything from social media posts, satellite photos and news stories, through to granular cash-flow data –provided it also offers an insight into potential risk or investment performance.²

As Chauwei Yak, Founding Partner at GAO Capital outlined, AI and ML can automate and filter data and information from the daily news cycle. This can provide fast insight into market sentiment as well as saving time and resources for the asset managers. According to a London Stock Exchange Group survey in 2021 of 463 Data scientists, quants, model governance professionals and C-level executives working in financial services, more companies are now working to integrate unstructured data into their models, and fewer work exclusively with structured data (see Figure 2).³

AI and ML also play a growing role in meeting demand for better insights into environmental, social and governance (ESG) performance. According to Bloomberg Intelligence, by 2025, ESG assets under management could account for more than a third of the anticipated worldwide total of US$140.5trn.⁴ China and India have made commitments to net zero, while Singapore and Hong Kong are now putting mandatory ESG into reporting regulations. The result for asset managers is increasing pressure to measure ESG criteria in their portfolios – AI and ML are proving important tools in the space, helping to automate the collection and interpretation of alternative data.

For example, AI and ML can create alternative datasets in ESG asset management. A programme taught to read the transcripts of a business's quarterly earnings calls could determine the tone and message of the CEO's words using natural-language processing to quickly identify those parts of the conversation where the CEO talks about ESG-related topics. Asset managers can then use those extracts to quickly drill down into the relevant information and determine how dedicated an organisation is about mitigating environmental risks and to what degree it is addressing the task in a structured and credible way.

Yet, while enthusiasm for these technologies might be high among asset managers and their service providers, obstacles remain. Questions of ethics, security, data sovereignty and regulatory guardrails in the industry will require careful thought and negotiation as institutions look to innovate. Though the potential competitive advantage of using unstructured big data may seem seductive, the complexities of a data scientist’s day-to-day job is, “not as sexy as people think,” according to Rebecca Lim, Founding Partner at AutoML Capital. This, in part, is because data scientists must be constantly vigilant of AI ‘black box’ problems, AI ‘conceptual drift’ and ‘inferred bias.’

What did she mean by these concepts? An AI “black box” is where humans are no longer able to adequately explain or interpret the outputs of an AI model. This type of situation typically represents grounds for ethical and governance concerns. Lim noted the importance of “explainable AI” also known as “white box” construction. This means that the ML or AI process is broken down into, small, easily understandable steps.

Similarly, conceptual drift, or model drift is where the relationship between two variables in the input data starts to change. For example, the relationship between two input variables might shift from multiplication to addition without this being labelled to the model; a scenario the Deutsche Bank white paper, Unleashing the power of AI in securities services explains. As a result, the accuracy of the model outputs will start to drift. This effect can be amplified by the model’s attempts to learn from the new inputs, potentially leading to strange outcomes if not managed correctly. To mitigate the risk of conceptual drift, data scientists must be proactive in supervising their AI and ML tools, identifying and addressing any apparent drifts, and ensuring the data being fed in is relevant and explainable.

The third challenge – inferred bias – refers to the potential for discriminatory conclusions to be drawn from statistical calculations and other attributes based on limited data. This is among the most prominent challenges, according to Kathayanat, who noted: “The area where we haven't yet got a complete handle as an industry is how do we prevent inherent biases in AI?”

“You need to keep humans in the loop”
Robert Coughlan, Finance Sector Lead, Google Cloud

France’s regulator Authorite des Marchés Financiers (AMF) claimed that algorithms might perpetuate prejudices and result in the financial exclusion of certain groups. For instance, models based on consumer behaviour historically work less effectively for younger customers with a limited history.⁵ While correcting bias in AI and ML remains a considerable challenge, it was suggested that one way to quickly avoid discriminatory bias – while adhering to most markets’ data privacy and sovereignty laws – is to “mask the data.” That is, to modify sensitive data so that it is rendered unusable to unauthorised intruders but remains useable by the asset manager’s software or authorised workers.

While the panelists reflected that the need to supervise, monitor and question the underlying processes of AI and ML technology will continue in the years to come, there is little debate as to whether or not the enterprise is worthwhile.

In the years ahead, ethical, regulatory, and social responsibility will remain at the core of the AI and ML asset management decision-making process – across portfolio, strategic and regulatory levels. “You need to keep humans in the loop,” concluded Coughlan. “AI is not a silver bullet, it’s a modern-day workflow.”

• Digital transformation and automation are increasing asset management productivity and assets under management (AUM) levels
• More financial services firms rely on structured and unstructured data for market analysis
• AI and ML are also playing an increasing role in meeting demand for better insights into environmental, social and governance (ESG) performance
• However, model drift remains a risk and data scientists should be proactive in supervising their AI and ML tools, identifying and addressing any apparent drifts

Sources

¹ See deloitte.com
² See dowjones.com
³ See lseg.group
⁴ See bloomberg.com
⁵ See iosco.org