Professional Documents
Culture Documents
AI-Portfolio-Insights-and-the-Future-of-Risk-Management
AI-Portfolio-Insights-and-the-Future-of-Risk-Management
© 2024 MSCI Inc. All rights reserved. Please refer to the disclaimer at the end of this document.
AI PORTFOLIO INSIGHTS AND THE FUTURE OF RISK MANAGEMENT |
MONTH YEAR
Andrew DeMond
Managing Director, MSCI Research
Manuel Rueda
Executive Director, MSCI Research
Greg Recine
Vice President, MSCI Research
Samuel López
Senior Associate, MSCI Research
Vicente Albíter
Senior Associate, MSCI Research
Will Baker
Vice President, MSCI Research
Zach Tokura
Vice President, MSCI Research
The authors would like to thank Mayank Gupta, Zsolt Nika, Gabriel Cardoso, Attila Agod, Tamas
Szarvas and Peter Shepard for their contributions to this work.
© 2024 MSCI Inc. All rights reserved. Please refer to the disclaimer at the end of this document.
MSCI.COM | PAGE 2 OF 13
AI PORTFOLIO INSIGHTS AND THE FUTURE OF RISK MANAGEMENT |
MONTH YEAR
Executive summary
The role of risk management for institutional investors is evolving. Risk leaders are looking to
broaden their teams’ focus beyond traditional risk measurement and increase their influence with
investment teams.
To support this transformation, MSCI is launching its first solution powered by generative AI, MSCI AI
Portfolio Insights. It enables risk leaders to scale up their impact with automated analysis that speeds
up their work and enhances the value of their risk data for investment decision-making.
In this paper we introduce AI Portfolio Insights, review the underlying methodology and describe
MSCI’s approach for using AI-powered tools reliably and safely.
© 2024 MSCI Inc. All rights reserved. Please refer to the disclaimer at the end of this document.
MSCI.COM | PAGE 3 OF 13
AI PORTFOLIO INSIGHTS AND THE FUTURE OF RISK MANAGEMENT |
MONTH YEAR
Introduction
Risk officers face the daily problem of distilling analytical reports into a set of prioritized, explainable
and actionable insights. This problem is growing more acute as risk leaders seek to move beyond
reporting and compliance and become more influential in their firms’ investment-decision process.
Yet most teams rely on labor-intensive practices for extracting insights from their risk systems,
delaying their contributions and reducing their impact with investment teams.
MSCI AI Portfolio Insights gives investment professionals a head start on their day by highlighting
what’s important and why across a multi-portfolio investment program. It provides a dynamic front
page to the analytics generated by MSCI’s models, connecting portfolio risk numbers to explanatory
market and issuer news. The MSCI AI Portfolio Insights agent allows users to then dive deeper into
portfolio reports in seconds by answering their plain-language questions, replacing hours to days of
tedious data fetching and emails.
Several new technologies enable these benefits: large-language models (LLMs), proprietary data-
curation algorithms and, foundationally, the cloud-based MSCI AI Portfolio Insights results
warehouse. To meet the high standard for accuracy set by risk professionals, MSCI AI Portfolio
Insights connects these components in a system designed to provide reliable results and safeguard
private data.
1We conducted 14 client meetings, speaking with more than 40 risk professionals across the Americas, EMEA and APAC, including
CROs and investment-risk leaders and their respective teams.
© 2024 MSCI Inc. All rights reserved. Please refer to the disclaimer at the end of this document.
MSCI.COM | PAGE 4 OF 13
AI PORTFOLIO INSIGHTS AND THE FUTURE OF RISK MANAGEMENT |
MONTH YEAR
This summarization feature relies on a rules-based data-curation algorithm to identify the most
important portfolios, factors and issuers each day based on changes in tracking error, with an
expanded set of monitorable metrics to be enabled in future releases.
© 2024 MSCI Inc. All rights reserved. Please refer to the disclaimer at the end of this document.
MSCI.COM | PAGE 5 OF 13
AI PORTFOLIO INSIGHTS AND THE FUTURE OF RISK MANAGEMENT |
MONTH YEAR
Exhibit 2: Portfolios are prioritized based on changes in risk relative to both their past behavior and
user-defined peer portfolios
To help quickly identify the reasons for these changes in risk, users can dig deeper through more-
granular summaries across multiple dimensions (as shown in exhibit 3 below). For each portfolio,
users can look at more detailed information in the “Portfolio Summary,” “Issuer Summary” and “Risk
Decomposition” sections of the front page, which help identify the main contributors to risk in a
portfolio and how sentiment2 for top portfolio constituents has changed over time.
© 2024 MSCI Inc. All rights reserved. Please refer to the disclaimer at the end of this document.
MSCI.COM | PAGE 6 OF 13
AI PORTFOLIO INSIGHTS AND THE FUTURE OF RISK MANAGEMENT |
MONTH YEAR
Exhibit 3: Detailed summaries focused on top portfolios, issuers and risk drivers break down the
high-level summary
The second component of the solution is designed to support users in executing ad hoc analysis
more efficiently. While the AI-generated summaries and figures are designed to answer many
standard questions, for more bespoke queries we’ve integrated an AI agent with a natural-language
interface that helps risk teams get the answers they need quickly.
An important illustration of this benefit is the AI agent’s ability to answer questions that require
analysis across multiple portfolios — for example, looking for low-risk portfolios, the client might ask
“Show me all the portfolios with negative active beta factor exposure.” The agent handles the tasks
of data fetching, reaggregation and analysis instead of the client, giving the client the answer in a
few seconds instead of requiring an hour of work.
© 2024 MSCI Inc. All rights reserved. Please refer to the disclaimer at the end of this document.
MSCI.COM | PAGE 7 OF 13
AI PORTFOLIO INSIGHTS AND THE FUTURE OF RISK MANAGEMENT |
MONTH YEAR
Exhibit 4: Get answers about portfolio analytics from the AI Portfolio Insights agent
Can you show me all portfolios hat has been momentum s ris
with a negative active beta factor contribution to this portfolio over
e posure the last month
1The AI agent understands queries using the language of risk professionals and facilitates follow-on
analysis of portfolios, factors and issuers. It is specialized for this task, so questions outside the
scope of risk management are not supported.
The goal with enabling this type of automation is to democratize and speed up access to risk
analytics. Through this simplified user experience, risk teams will be well positioned to support their
investment teams in a more expansive advisory capacity.
Methodology
Insight, reliability and safety
Risk officers and portfolio managers require accurate, timely information delivered with industry-
standard data safety and privacy. LLMs, the key advancement that enables modern AI functionality,
have the well-known potential to generate plausible but factually incorrect outputs, or
“hallucinations.” Moreover, they generally lack specialized domain knowledge. Thus, the key design
goals for the MSCI AI Portfolio Insights system were threefold: enable rapid understanding of the
most important changes in portfolio risk every day, minimize inaccuracies and protect client data.
Analytical foundations
MSCI AI Portfolio Insights is built on MSCI’s portfolio-risk analytics, covering the U.S. and global
equity-factor models.3 The automated system draws from the cloud-based data warehouse of client
reports that powers MSCI Insights dashboards, ensuring consistency of results across multiple
views of the data. Issuer news stories are sourced from approximately 20 global business
3AI Portfolio Insights currently covers portfolios that are at least 85% equity (i.e. unit delta positions) by weight and covered with
equity factor models.
© 2024 MSCI Inc. All rights reserved. Please refer to the disclaimer at the end of this document.
MSCI.COM | PAGE 8 OF 13
AI PORTFOLIO INSIGHTS AND THE FUTURE OF RISK MANAGEMENT |
MONTH YEAR
publications via an aggregated news feed.4 We process the news to identify mentioned issuers,
remove irrelevant stories and assess the sentiment (positive, neutral or negative) for each company
and story using an LLM-based process.5
Exhibit 5: Rules-based algorithms monitor portfolios for unexpected changes in risk analytics
MSCI 2AI Portfolio Insights analyzes time series of portfolio analytics for unusual and important
changes in risk metrics. The resulting datapoints are prioritized every day and combined with
explanatory information from risk models and news to create portfolio summaries.
In the context of risk management, we define high-priority portfolios and risk statistics as being both
novel and important and use quantitative metrics and a rules-based algorithm to identify the most
salient information on a given day. This approach allows us to provide useful information in a variety
of market conditions: On big market days MSCI AI Portfolio Insights focuses on the portfolios that
stand out from the pack even when everything looks like an outlier, while on average market days it
finds anything unusual. If nothing stands out, MSCI AI Portfolio Insights says so and doesn’t mince
words.
Central to this approach is the concept of a portfolio group, or a set of comparable portfolios that
are monitored together. Portfolio groups are defined by the user. They would normally share the
same risk model and target comparable levels of active risk. The tool identifies portfolios as notable
by scoring them based on risk levels compared to their recent past6 and by comparison of day-over-
day risk changes to other portfolios in their group cross-sectionally. We rank portfolios based on a
combination of these time-series and cross-sectional scores7 after thresholding the scores to filter
out unusual but inconsequential moves.
The key facts explaining the changes in risk in notable portfolios are identified via risk
decomposition into changes in holdings, exposures, volatilities and correlations. We can
algorithmically extract common themes affecting multiple portfolios in the group, for example a
4The sources have been selected to maximize global issuer coverage and include reputable publishers such as Reuters and the
Economic Times, as well as several press-release distributors (PR Newswire, Business Wire, etc.).
5
The specific model in use at the time of writing is GPT-4 from OpenAI accessed via an enterprise endpoint in Microsoft Azure.
Issuer-sentiment scores are defined by scoring positive, neutral and negative stories as +1, 0 and -1 and summing over all scores for
an issuer on a given day.
|(𝑥𝑖 −𝑥̂𝑖 )|
6 Mathematically we define outliers in any statistic 𝑥𝑖 as large deviations from expectations measured by a 𝑧-score: 𝑧 = ,
𝜎̂𝑖
where 𝜎̂𝑖 is the standard deviation and 𝑥̂𝑖 is the expected value of 𝑥𝑖 . 𝑥̂𝑖 is computed with a moving average or trend projection.
7 We define the rank 𝑅𝑛 of portfolio 𝑛 based on a weighted average of its 1-day tracking error change cross-sectional 𝑧 score 𝑧𝑋𝑆 and
the time-series 𝑧 score of the tracking error 𝑧𝑇𝐸 , i.e. 𝑅𝑛 = |𝑤𝑇𝐸 𝑧𝑇𝐸 + 𝑤𝑋𝑆 𝑧𝑋𝑆 |. The rank 𝑅𝑛 is further normalized for display purposes.
© 2024 MSCI Inc. All rights reserved. Please refer to the disclaimer at the end of this document.
MSCI.COM | PAGE 9 OF 13
AI PORTFOLIO INSIGHTS AND THE FUTURE OF RISK MANAGEMENT |
MONTH YEAR
specific issuer or factor, and highlight them as separate facts for summarization. The algorithm
prioritizes relevant facts based on their impact and overall prevalence among portfolios in the group.
When available and relevant, the system adds issuer news summaries to the information set. We
then pass the final collection of information to a series of LLMs prompted to create both multi- and
single-portfolio narratives containing the highlighted facts. We have organized the LLMs to
safeguard against incoherent statements and provide fluent, stylistically appropriate accounts of the
key drivers of portfolio risk.
Agent
The MSCI AI Portfolio Insights agent is a task doer8 — assign it a task and get a result faster than
previously possible. With the same design goals in mind for the agent as the summarizer — i.e., rapid
insight, minimal errors and data safety — we’ve invested significant engineering into producing an AI
system that meets the needs of risk professionals.
Exhibit 6: The agent uses context to answer questions via a multistep process
User Question
Data Manager
Portfolio, Asset, Model, etc.
AI agent Response
3The AI agent answers user questions via a multistep process. After adding relevant context, the
agent assembles the necessary portfolio data and generates the code required to produce a useful
response.
8 Note that the agent currently does not support interactivity – from one question to the next it does not retain any memory so re-
typing or augmenting questions is required.
© 2024 MSCI Inc. All rights reserved. Please refer to the disclaimer at the end of this document.
MSCI.COM | PAGE 10 OF 13
AI PORTFOLIO INSIGHTS AND THE FUTURE OF RISK MANAGEMENT |
MONTH YEAR
Rapid analysis of a complicated portfolio is challenging and, as a monolithic task, beyond the
capabilities of current LLMs. By segmenting tasks into several steps — such as question
categorization, dataset selection, data fetching and analysis — we break down complex problems
into simpler tasks (as shown in exhibit 6). This approach reduces the likelihood of hallucinations by
the LLM, enhancing the accuracy and reliability of the responses. Many steps taken by the agent
focus on adding context, including specific names, dates and relevant datasets required to answer
the question.
Natural questions li e “what is ris iest name in my portfolio today ” are imprecise, but using
contextual clues from the page the agent can replace “my portfolio” with a specific portfolio name
and replace “today” with the highlighted analysis date. Also, unlike nonspecialist models, MSCI AI
Portfolio Insights knows what “riskiest” means. Thus the analysis is aligned with standard reporting
conventions and industry jargon. To support this workflow we have created detailed response
guidelines to minimize the risk of hallucinations and increase the accuracy and usefulness of the
LLM’s responses. These guidelines are regularly updated and refined to support new analysis and
report types. If the agent can’t or shouldn’t answer a question, such as if as ed for off-topic results
or investment advice, it will decline to respond.
For data safety, we don’t allow LLMs to access client data directly. Instead, a specialized LLM writes
code that is executed in a privileged environment to fetch and analyze client data, ensuring that the
code and results are auditable. This transparency helps maintain the integrity and reliability of the
analysis. We don’t train any LLM or deep-learning model on client data, and all client data is
segregated into separate cloud domains to preclude any cross-leakage of information.9
Validation
MSCI’s approach to model testing applies equally to the three pillars of MSCI AI Portfolio Insights —
data curation, summarization and the interactive agent. Each undergoes continuous model
validation and ongoing quality improvement. We evaluate the data-curation algorithm using a set of
quality metrics that correlate with interesting and engaging content, such as the proportion of events
with associated news, comparison against human-labeled datasets, and measures of
repetitiousness such as autocorrelation of 𝑧-scores. Human testers read summaries and score them
for groundedness (lack of falsehoods), coherence (no conflicting statements), relevance (accuracy
regarding what happened and why) and fluency (natural text). We measure the performance of the
agent against a battery of validation questions that encompass a range of tasks, from basic data
fetching to complex analyses generalized from interactions with testers. The validation framework
includes automatic evaluation with humans in the loop to align the results with client expectations.
Conclusion
The future of risk is being shaped by the demand for faster and more accessible analytics. MSCI AI
Portfolio Insights is a key innovation designed to help risk leaders at institutional investors meet
these demands. By leveraging generative-AI technology, it introduces more automation into risk
management, speeding up the workday and improving the value of risk data for investment decision-
making.
9 We do maintain and analyze segregated, client-specific logs of agent interactions for quality assurance.
© 2024 MSCI Inc. All rights reserved. Please refer to the disclaimer at the end of this document.
MSCI.COM | PAGE 11 OF 13
AI PORTFOLIO INSIGHTS AND THE FUTURE OF RISK MANAGEMENT |
MONTH YEAR
Contact us
msci.com/contact-us
ASIA PACIFIC
China + 86 21 61326611
Hong Kong + 852 2844 9333
India + 91 22 6784 9160
Malaysia 1800818185 *
South Korea + 82 70 4769 4231
Singapore + 65 67011177
Australia + 612 9033 9333
Taiwan 008 0112 7513 *
Thailand 0018 0015 6207 7181 *
Japan + 81 3 4579 0333
* toll-free
not be modified, reverse-engineered, reproduced or redisseminated in whole or in part without prior written permission from MSCI. All rights in the
Information are reserved by MSCI and/or its Information Providers.
The Information may not be used to create derivative works or to verify or correct other data or information. For example (but without limitation), the
Information may not be used to create indexes, databases, risk models, analytics, software, or in connection with the issuing, offering, sponsoring,
managing or marketing of any securities, portfolios, financial products or other investment vehicles utilizing or based on, linked to, tracking or otherwise
derived from the Information or any other MSCI data, information, products or services.
The user of the Information assumes the entire risk of any use it may make or permit to be made of the Information. NONE OF THE INFORMATION
PROVIDERS MAKES ANY EXPRESS OR IMPLIED WARRANTIES OR REPRESENTATIONS WITH RESPECT TO THE INFORMATION (OR THE RESULT S TO
BE OBTAINED BY THE USE THEREOF), AND TO THE MAXIMUM EXTENT PERMITTED BY APPLICABLE LAW, EACH INFORMATION PROVIDER EXPRESSLY
DISCLAIMS ALL IMPLIED WARRANTIES (INCLUDING, WITHOUT LIMITATION, ANY IMPLIED WARRANTIES OF ORIGINALITY, ACCURACY, TIMELINESS,
NON-INFRINGEMENT, COMPLETENESS, MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE) WITH RESPECT TO ANY OF THE
INFORMATION.
Without limiting any of the foregoing and to the maximum extent permitted by applicable law, in no event shall any Information Provider have any liability
regarding any of the Information for any direct, indirect, special, punitive, consequential (including lost profits) or any other damages even if notified of
the possibility of such damages. The foregoing shall not exclude or limit any liability that may not by applicable law be excluded or limited, including
without limitation (as applicable), any liability for death or personal injury to the extent that such injury results from the negligence or willful default of
itself, its servants, agents or sub-contractors.
Information containing any historical information, data or analysis should not be taken as an indication or guarantee of any future performance, analysis,
forecast or prediction. Past performance does not guarantee future results.
The Information may include “Signals,” defined as quantitative attributes or the product of methods or formulas that describe or are derived from
calculations using historical data. Neither these Signals nor any description of historical data are intended to provide investment advice or a
recommendation to make (or refrain from making) any investment decision or asset allocation and should not be relied upon as such. Signals are
inherently backward-looking because of their use of historical data, and they are not intended to predict the future. The relevance, correlations and
accuracy of Signals frequently will change materially.
The Information should not be relied on and is not a substitute for the skill, judgment and experience of the user, its management, employees, advisors
and/or clients when making investment and other business decisions. All Information is impersonal and not tailored to the needs of any person, entity
or group of persons.
None of the Information constitutes an offer to sell (or a solicitation of an offer to buy), any security, financial product or other investment vehicle or any
trading strategy.
It is not possible to invest directly in an index. Exposure to an asset class or trading strategy or other category represented by an index is only available
through third party investable instruments (if any) based on that index. MSCI does not issue, sponsor, endorse, market, offer, review or otherwise express
any opinion regarding any fund, ETF, derivative or other security, investment, financial product or trading strategy that is based on, linked to or seeks to
provide an investment return related to the performance of any MSCI inde (collectively, “Inde Lin ed Investments”). MSCI ma es no assurance that
any Index Linked Investments will accurately track index performance or provide positive investment returns. MSCI Inc. is not an investment adviser or
fiduciary and MSCI makes no representation regarding the advisability of investing in any Index Linked Investments.
Index returns do not represent the results of actual trading of investible assets/securities. MSCI maintains and calculates indexes, but does not manage
actual assets. The calculation of indexes and index returns may deviate from the stated methodology. Index returns do not reflect payment of any sales
charges or fees an investor may pay to purchase the securities underlying the index or Index Linked Investments. The imposition of these fees and
charges would cause the performance of an Index Linked Investment to be different than the MSCI index performance.
The Information may contain back tested data. Back-tested performance is not actual performance, but is hypothetical. There are frequently material
differences between back tested performance results and actual results subsequently achieved by any investment strategy.
Constituents of MSCI equity indexes are listed companies, which are included in or excluded from the indexes according to the application of the relevant
index methodologies. Accordingly, constituents in MSCI equity indexes may include MSCI Inc., clients of MSCI or suppliers to MSCI. Inclusion of a
security within an MSCI index is not a recommendation by MSCI to buy, sell, or hold such security, nor is it considered to be investment advice.
Data and information produced by various affiliates of MSCI Inc., including MSCI ESG Research LLC and Barra LLC, may be used in calculating certain
MSCI indexes. More information can be found in the relevant index methodologies on www.msci.com.
MSCI receives compensation in connection with licensing its inde es to third parties. MSCI Inc.’s revenue includes fees based on assets in Index Linked
Investments. Information can be found in MSCI Inc.’s company filings on the Investor Relations section of msci.com.
MSCI ESG Research LLC is a Registered Investment Adviser under the Investment Advisers Act of 1940 and a subsidiary of MSCI Inc. Neither MSCI nor
any of its products or services recommends, endorses, approves or otherwise expresses any opinion regarding any issuer, securities, financial products
or instruments or trading strategies and MSCI’s products or services are not a recommendation to ma e (or refrain from ma ing) any kind of investment
decision and may not be relied on as such, provided that applicable products or services from MSCI ESG Research may constitute investment advice.
MSCI ESG Research materials, including materials utilized in any MSCI ESG Indexes or other products, have not been submitted to, nor received approval
from, the United States Securities and Exchange Commission or any other regulatory body. MSCI ESG and climate ratings, research and data are produced
by MSCI ESG Research LLC, a subsidiary of MSCI Inc. MSCI ESG Indexes, Analytics and Real Estate are products of MSCI Inc. tha t utilize information
from MSCI ESG Research LLC. MSCI Indexes are administered by MSCI Limited (UK) and MSCI Deutschland GmbH.
Please note that the issuers mentioned in MSCI ESG Research materials sometimes have commercial relationships with MSCI ESG Research and/or
MSCI Inc. (collectively, “MSCI”) and that these relationships create potential conflicts of interest. In some cases, the issuers or their affiliates purchase
research or other products or services from one or more MSCI affiliates. In other cases, MSCI ESG Research rates financial products such as mutual
funds or ETFs that are managed by MSCI’s clients or their affiliates, or are based on MSCI Inc. Indexes. In addition, constituents in MSCI Inc. equity
indexes include companies that subscribe to MSCI products or services. In some cases, MSCI clients pay fees based in whole or part on the assets they
manage. MSCI ESG Research has taken a number of steps to mitigate potential conflicts of interest and safeguard the integrity and independence of its
research and ratings. More information about these conflict mitigation measures is available in our Form ADV, available at
https://adviserinfo.sec.gov/firm/summary/169222.
Any use of or access to products, services or information of MSCI requires a license from MSCI. MSCI, Barra, RiskMetrics, IPD and other MSCI brands
and product names are the trademarks, service marks, or registered trademarks of MSCI or its subsidiaries in the United States and other jurisdictions.
The Global Industry Classification Standard (GICS) was developed by and is the exclusive property of MSCI and S&P Global Market Intelligence. “Global
Industry Classification Standard (GICS)” is a service mar of MSCI and S&P Global Market Intelligence.
MIFID2/MIFIR notice: MSCI ESG Research LLC does not distribute or act as an intermediary for financial instruments or structured deposits, nor does it
deal on its own account, provide execution services for others or manage client accounts. No MSCI ESG Research product or service supports, promotes
or is intended to support or promote any such activity. MSCI ESG Research is an independent provider of ESG data.
Privacy notice: For information about how MSCI collects and uses personal data, please refer to our Privacy Notice at https://www.msci.com/privacy-
pledge.