Main Working Papers

Generative artificial intelligence and machine learning are poised to reshape how investors make decisions in private markets. Yet the same technologies that promise deeper insight also risk amplifying the sector’s long-standing distortions. AI systems produce polished narratives that conceal misleading inputs. In markets where outcomes unfold over a decade, this dynamic could reward the most persuasive storytellers rather than the most capable managers. The challenge is therefore not simply to adopt technology, but to govern it by embedding rigorous testing and human expertise to ensure that automation enhances, rather than imitates, judgment.

We develop a theory of private equity continuation vehicles (CVs), a rapidly growing exit route in which general partners (GPs) transfer portfolio companies from an existing fund to a newly created fund they manage. CVs can enhance efficiency by extending the holding period of high-potential firms, but they also allow GPs to extract rents by exploiting information advantages and their intermediary role between legacy and new limited partners (LPs). Our model shows how these frictions give rise to inefficient CVs, which either continue failed firms, or retain already successful firms. We characterize the optimal fund contracts that limit such distortions and highlight how GP and LP coinvestment shape outcomes. The analysis generates predictions for when CVs arise, the quality of assets they contain, and their distributional consequences across investor groups, offering guidance for contract design and for evaluating the welfare effects of GP-led secondaries.

Private equity fund managers issue interim valuations for portfolio companies, which tend to be overly smooth. These valuations are accompanied by a report averaging 10 sentences. We compile a new international dataset of over 15,000 interim performance reports from 1,500 exited investments. Our analysis shows that the average sentiment in each report is strongly correlated with future returns. Sentiment is more significant during fundraising periods, when investors are most likely to focus on these reports. Machine learning algorithms show that the informational value of these reports is also substantial when tested out of sample. Our results indicate that fund managers strategically employ qualitative information to convey expectations to their investors. This behavior is consistent with a double agency dynamic, where both fund managers and investors juggle competing incentives to balance transparency with reputational concerns.

Unfunded capital commitments expose institutional investors to liquidity risk. Using hand-collected data from 323 U.S. university endowments, we show that these commitments are large, procyclical, and weakly offset by cash and credit lines. During the 2008 Financial Crisis, endowments with high commitments and low liquidity coverage underperformed peers, cut spending on faculty and scholarships, and declined in rankings. In normal times, however, liquidity coverage imposes a return cost. These findings reveal a trade-off between return and liquidity in the presence of contingent liabilities, with real implications for institutional stability and financial policy.

Using one million employee reviews, we show that ownership is related to employee satisfaction, with VC-backing being the most significant effect. Employee satisfaction is abnormally high in the presence of a VC and decreases when the VC exits. Textual analysis of reviews shows that VC-backed employees enjoy the supportive culture and Human Resources policies, but complain about their compensation and the challenging work environment. When the VC exits, the topics mentioned in the reviews and the satisfaction score become similar to those of similar companies, but some of the initial differences persist. VC presence therefore has some lasting impact.