Summary

The project examined whether credit markets react when insurers stop renewing policies in California counties hit by wildfire and flood. It combined state non-renewal data with municipal bond issuance and MSCI corporate bond data from 2015 to 2023.

Key takeaways

Sanoh Yusuf, a doctoral student in economics at The University of Osaka, found that a one percentage point rise in a county’s non-renewal rate was associated with municipal bond spreads that were 11 basis points wider, and that above a non-renewal rate of about 10%, counties became less likely to issue. Corporate bond spreads showed no significant reaction, which he attributes to companies holding assets outside the affected areas.

Quote

Summary

The project asked whether disasters that force people from their homes show up in what governments pay to borrow. It isolated internal displacement, rather than physical damage or lost GDP, and compared each country with itself over time.

Key takeaways

Eve Maynard, a master’s student in statistics at University of Bath, found that a disaster displacing 1% of a country’s population widens the deficit by around 0.16% of GDP the following year, with the budget recovering within two to three years. Bond markets barely register it: only the largest events showed up in credit ratings, and only three to five years later. She recommends a large-event displacement flag in sovereign climate monitoring.

Quote

Summary

The project tested whether companies that adapt to physical climate risk are rewarded with narrower credit spreads. It grouped issuers facing high physical risk by their adaptation response and compared spreads after stripping out conventional drivers such as rating, sector and duration.

Key takeaways

Palak Nagdev, a master’s student at the MIT Sloan School of Management, found that the relationship holds in parts of the market, such as consumer discretionary issuers exposed to flooding, but runs the other way elsewhere. Of 35 statistically significant results across 156 combinations of sector, hazard and year, two thirds pointed in the opposite direction, which she attributes to differences in hazard, sector economics, the visibility of adaptation and timing.

Quote

Summary

The project asked whether the structure of a company’s supply chain is linked to its risk. It built a score that separates chains where every link must work from true diversification, then tested it on 1,264 MSCI World companies over a decade’s worth of data.

Key takeaways

Zachary Too, a master’s student in financial analysis at London Business School, found that when the score was used to choose 50 stocks alongside a single concentrated bet, it roughly halved the portfolio’s volatility while keeping about 60% of the return, and beat 488 of 500 random baskets. He is clear that the effect is modest and the results are flattered by testing on companies that are in the index today. He recommends tagging supply-chain links by type so the score can be estimated from real disruptions.

Quote

Summary

The project asked what a knowledge graph makes possible with supply-chain data that flat tables do not. It rebuilt the data for around 1,300 MSCI World companies as a knowledge graph and benchmarked it against a conventional SQL database on the same task.

Key takeaways

Shishir Iyer, who recently completed a master’s degree in computer science at UC San Diego, found that the graph database scaled far better as the dataset grew when finding the basket of companies with the least shared exposure to critical supply nodes. The SQL version proved surprisingly short to write, so the advantage of graphs lies in performance and intuitiveness rather than simpler code. He recommends exploring how a graph database could fit into MSCI’s data pipeline.

Quote

Summary

The project set out to map the commodities, including critical minerals, that sit behind the broad sector exposures in MSCI’s supply-chain data. It built a crosswalk from sector relationships to likely commodities using public sources such as the U.S. Geological Survey and the International Energy Agency.

Key takeaways

Moritz-Felix Lehofer, a bachelor’s student in economics at LMU Munich, resolved about 20,000 of some 59,000 mappings to a specific commodity. Applied to three companies, the prototype produced distinctly different modeled commodity profiles from the same sector data. He is clear that these are modeled exposures rather than observed purchases, and proposes adding production and trade data next to map geographic concentration and potential choke points.

Quote

Summary

The project asked how much of the financial analyst’s job AI could do by 2030, and how much is held back by rules rather than capability. It scored 30 analyst tasks on two questions in order: can a machine do it, and is the company allowed to hand it over?

Key takeaways

Enzo Kekligian, a master’s student in climate change, management and finance at Imperial Business School, found that five tasks can be automated in today’s environment, 22 are shared with AI and three remain human. In eight of the shared tasks a required signature was the only barrier, and changing approval rules moved more tasks to automation than improving the technology. He concludes that AI exposure scores need a rules dimension as well as a capability one.

Quote

Summary

The project asked whether AI could turn corporate disclosure from an obligation into an exchange that benefits both companies and data users. It built a working prototype on top of MSCI’s issuer platform, extracting data from uploaded documents and mapping it to reporting frameworks on request.

Key takeaways

Aditya Sinha, a master’s student in computer science at the University of Texas, found that AI works best as the front door rather than the decision-maker. In the prototype, extracted data stays an editable draft, nothing is saved without human approval, AI use is opt-in, and companies get coverage insights back in return for their data. He recommends building an analyst view next, so reviewers can correct the mapping and the system learns from it.

Quote