AsianInvesterAsianInvester

GPIF says AI a 'black box', posing transparency challenges for $1.8 trillion portfolio

The need for transparency makes it no easy task for Japan's national pension fund to fully leverage artificial intelligence tools in asset management processes, its CIO tells AsianInvestor.
GPIF says AI a 'black box', posing transparency challenges for $1.8 trillion portfolio

High standards for transparency make it challenging for Japan’s Government Pension Investment Fund (GPIF) to leverage artificial intelligence (AI) in its asset management, according to Eiji Ueda, chief investment officer and executive managing director.

Eiji Ueda
GPIF

“After we make an investment and look at the outcome in hindsight, we need to disclose how we did the investment, and how the outcome came about. AI makes investment disclosure very hard for us because AI is like a black box where we cannot monitor every step of the process,” Ueda told AsianInvestor.

Given the size and diversity of the Tokyo-based behemoth’s portfolio, AI tools could offer a helping hand to manage and oversee global investments. As such, AI technology holds some relevance for GPIF.

“When managing big data, AI can be a helpful tool for checking if data points are missing or are recorded wrongly, for instance. But we are not applying it to the decision-making process itself for investments,” Ueda said.

As of June 30, GPIF holds ¥254.7 trillion ($1.8 trillion) worth of assets.

BIG DATA

Being a government entity, GPIF must ensure that all stakeholders can review and understand its investment processes. This scrutiny reflects the pension fund’s commitment to delivering returns and creating value through informed investment decisions.

“We are a public entity and whenever we make decisions, especially active investments, we must make sure we have reasonable evidence that we will perform better than our benchmarks,” Ueda said.

He noted that the same fundamental concerns about transparency also apply to passive investments in indices.

ALSO READ: How GPIF uses active management to achieve ESG goals

While AI presents transparency challenges, GPIF focuses on analyzing and learning from big data - large and diverse datasets that are huge in volume and grow at ever-increasing rates over time.

“We do require a lot of data analysis for investment decisions, but we are more using big data than AI for that,” Ueda said.

Big data is utilised in machine learning, predictive modelling, and other advanced analytics to address business challenges and make informed decisions.

“We want to know exactly how we analyse that data, so when we see the investment outcome, we can always look back to see whether we made the right decisions, find any missing pieces and learn from the process. Using AI more will make that process increasingly difficult,” Ueda said.

AI STUDY

GPIF’s approach to AI is built on thorough research on the relevance of disruptive technology and its potential scope for enhancing the pension fund’s asset management efforts.

As early as fiscal 2017, Japan's giant pension fund initiated a study to explore the potential of AI within its investment management practices. This involved analyzing the trading data of GPIF’s active managers for both domestic and foreign equities using machine learning techniques.

The results demonstrated the potential to identify changes in managers’ investment behaviour and differences between asset managers that conventional indicators, such as the investment style, have not been able to capture.

ALSO READ: How Japan’s GPIF avoids inflated values in alternatives

Despite all the hype, the use of AI remains limited in ensuring transparency.

“As of today, it is hard for us to use AI on the complete decision-making processes, even with the most advanced technology available in the AI space,” Ueda said.

A study, commissioned by Sony Computer Science Laboratories, showed the potential to classify investment styles into patterns and identify style drifting. From fiscal 2019 to fiscal 2020, the study was further developed to quantify aspects that are difficult to measure numerically, such as uniqueness and behavioural habits.

The objective was to assess the consistency of investment behaviours of active asset managers over time. Additionally, the study aimed to evaluate the degrees of similarities among asset managers was conducted.

ALSO READ: What GPIF thinks about crypto, other niche assets

¬ Haymarket Media Limited. All rights reserved.