AsianInvesterAsianInvester
Advertisement

GPIF ramping up AI efforts to assess fund partners

Japan’s public pension fund is testing new technology on its investment portfolio, which is advancing its plans to add artificial intelligence into its investment management.
GPIF ramping up AI efforts to assess fund partners

Japan’s Government Public Investment Fund (GPIF) is ramping up efforts to implement artificial intelligence (AI) into its investment management processes, to better assess the management styles and performance of its external fund managers.

The world’s largest pension fund, which had assets ¥‎159.2 trillion ($1.49 trillion) in assets under management at the end of June, has outsourced the AI project to Tokyo-based Sony Computer Science Laboratories (Sony CSL), which has been working on it since being hired in November 2017

The company was founded in 1988 for the sole purpose of conducting research relating to computer science.

A spokeswoman for GPIF told AsianInvestor the aim of the commissioned AI research is to help the pension fund optimise its fund manager structure.

“Based on the result of the current commissioned research, we will examine the future use of AI,” she said, referring further information requests about the project to Sony CSL.

Takao Tajiri, project leader of GPIF's AI project at Sony CSL, said his company has completed the first stage of developing a prototype AI system, which consisted of theoretically testing some small data sets. He told AsianInvestor he and his team are constructing a blueprint of a system that would apply the theory into practice within GPIF's business processes.

Takao Tajiri

“Through the introduction of such a system, GPIF should be able to conduct more prudent and data-driven selection and monitoring over fund managers,” Tajiri said. “In addition, this may foster more constructive and in-depth dialogue between GPIF and fund managers, which will improve the robustness and performance of investment practices at GPIF in the long run.”

The research and testing are so far taking place by applying it on parts of GPIF's domestic equity holdings. 

ARTIFICIAL ASSESSMENT

GPIF is interested in seeing how AI can improve its overall business practices, and in particular more accurately assess the fund managers it works with.

Artificial intelligence is the simulation of human intelligence processes by machines, especially computer systems. These processes include learning (the acquisition of information and rules for using the information), reasoning (using rules to reach approximate or definite conclusions) and self-correction. 

Sony CSL’s project essentially seeks to test the possibilities and implications of applying AI to the long-term management of pension assets. GPIF believes that artificial intelligence programmes can reduce the potential for human subjectivity when it comes to selecting and monitoring fund management partners, and more accurately divine whether they are investing in accordance with the policies they were hired to pursue.

This could come in handy as the pension fund seeks to develop its performance-based fee structure further.  

Selecting and monitoring external fund managers are some of the most important tasks for GPIF. The pension fund is legally obliged to outsource all of its assets for external management, and most of them are passively managed.

Currently, it has to do so by using track records and qualitative explanations. It believes that AI can greatly improve its ability to accurately assess these fund houses.

GOING FURTHER

As part of its current fund manager oversight, GPIF has developed and maintains what it has dubbed a “manager structure”. This refers to a structure it employs to manage organisations or asset managers, along overseeing the associated allocation or reallocation of assets to be managed on its behalf.

One example of associated allocations are GPIF’s overseas real estate investments. To date the pension fund has only invested into such assets via a fund of funds structure through CBRE Global Investment Partners. Alternatives made up 0.35% of total portfolio as of the end of June, while overseas real estate investments stood at ¥54.5 billion as of March 31.

Sony CSL has had to make various definitions as part of developing and maintain the manager structure for the AI programme, Tajiri explained. That includes defining the characteristics of each manager, characterising their management behaviour with respect to various aspects of the economic background, and determining whether they are investing in a manner that is consistent with the policy that GPIF hired them to execute.

Doing so is important for GPIF, as it can become exposed to unanticipated risks if individual fund managers start to veer away from the mandates they were hired to pursue. It could become a particularly acute problem were multiple fund managers to shift away from their investment briefs in a similar manner.

The Sony CSL team began by developing the prototype system to focus on this. It uses deep learning to detect the investment style of managers from a small set of Japanese equity trading behaviour data that is collected daily by GPIF. It’s now moved into the second phase. The behaviour data assessed includes trading items, timing, volume, unrealised gain and losses.

“We are in the process to widen and deepen the prototype, whose basic foundation has been already confirmed,” Tajiri said. “Secondly, we are developing another layer of component that distinguishes performance caused by managers' investment skill from mere luck, such as good market circumstances.

“Thirdly, we combine these components into an integrated system in order to provide an additional new way of evaluating fund managers, which will augment GPIF officers' business processes.”

Institutional investors across the world will be likely to follow the impact of the AI prototype with great interest. If successful, it could help revolutionise how GPIF and other asset owners pick their external fund management partners.

¬ Haymarket Media Limited. All rights reserved.
Advertisement