AI can add value to almost every part of the investment process, and portfolio management purposes can provide support for information gathering and analytics from non-traditional and inaccessible data sources, according to a new CFA Institute Research and Policy Center report.
Report, title Pensions in the age of artificial intelligenceexamined selected academic literature and outlined the possibilities for AI strengthening up and down the pension chain. This highlights the owners how AI can be used to modernize operations, as well as the issues that may arise.
Using case studies from the Government Pension Investment Fund (GPIF), this report highlights AI and machine learning (ML) as useful tools for increasing the accuracy of asset owner manager selection and reviews.
The world's largest asset owner, GPIF has used a variety of external managers with uneven performance over the past decade. Although we relied on a small number of internal human experts to select and evaluate active managers, in 2017, GPIF tried multi-stage programs to better identify and evaluate manager styles using techniques such as deep learning.
“…An AI system allows GPIF to assess fund managers' investment styles more thoroughly, accurately and efficiently, providing quantitative metrics that were previously only available as an explanation of qualitative fund management,” the CFA Institute report states.
“These technologies demonstrate the possibility that GPIFs can access a wider range of asset managers and capital benefits by relying on internal data-driven analyses to determine the performance of fund managers rather than relying on qualitative explanations of performance or policy.
“This helps to eliminate bias against fund managers who have a history of working with GPIFs that are suitable for selling products and large companies.”
Information gathering and analysis is one of the most common ways for asset owners to use AI, even when using programs such as Natural Language Processing (NLP) or “increase the scope of available information.”
For example, NLP can call revenues to screen transcripts to identify changes in company positioning, but sentiment analysis can be used to predict market and investor responses to corporate events.
Similarly, AI can be used to scan and formulate ESG insights by collecting and analyzing a large amount of structured, unstructured data from portfolio companies. The availability and consistency of company ESG data is far the biggest challenge for pension funds that want to exercise stewardship, and a human-driven approach with AI support can significantly improve the efficiency of corporate analysis. [See also Can artificial intelligence (AI) help stewardship resourcing?]
Connect the dots
Ultimately, the most valuable benefit for fiduciary investors who are aware of risk and responsibility is knowing what is on the horizon, and AI has proven use cases in both market risk management and credit risk management.
The report explained that AI is suitable for identifying correlations and providing forecasts of market crashes, as “market behavior is nonlinear and emerges from dynamic interactions across the system.” ML technologies such as random forests and artificial neural networks can effectively identify signs of a recession.
AI and ML could also help improve the accuracy of credit risk assessments, the report says.
However, AI use is not without risk. Just like humans, the CFA Institute emphasized that there is bias and variance in AI models. “Model bias refers to the inconsistency between model predictions and actual values, while variance refers to the generalization and sensitivity of the model to variation in training data.”
“Ideally, models will have low bias and low variance, but this scenario is not always possible because there is a trade-off between bias and variance.”
A strong bias suggests “wearing”. In other words, although the model is not sufficiently complex to the training data, a strong variance suggests “overfit”. In other words, the model is too complicated.
“Because market data is often irregular over time and the statistical characteristics of such data also change over time, it can be difficult to generate models that fit appropriately with variations of these data and generate accurate predictions,” the CFA Institute said in its report.
“Like other AI and ML applications, it is important to target the output generated by AI to the right test and supervision.”
