OMI Finance Talks

19 November 2026 @1pm BST

Pricing AI as a Service

Jiahua Wu
Imperial College London

Abstract

We study AI as a Service, in which a monopolistic provider sells AI capabilities that perform tasks for downstream buyers. Buyers privately differ in both their value from the service and the probability that the AI resolves their task. The seller designs a menu of capabilities and outcome-contingent prices and controls how true task outcomes are mapped into reported outcomes; buyers may dispute and verify the reported outcome at a cost. We first compare two pricing formats commonly used in practice: usage-based pricing (UBP), under which the buyer pays per use regardless of the outcome, and pay-for-success pricing (PFS), under which the buyer pays only when the deployed AI reports the task as successful. When the AI-reported outcome fully aligns with the buyer’s task outcome, PFS becomes more attractive as per-success values become less heterogeneous and higher per-success values coincide with higher success probabilities. In a customer-service application, where an AI front end resolves requests before they join the human service queue, PFS is favored when AI capability and human service are complements, and UBP is favored when they are substitutes. Allowing the seller to choose the billable success metric can further increase PFS profit. When buyer verification is sufficiently costly, a single success payment combined with an appropriately chosen metric can reproduce the fully aligned limited-liability two-price benchmark, so optimal PFS weakly dominates UBP. Our main result shows that this misalignment motive disappears in the unrestricted contract space: separate success and failure prices make full outcome alignment without loss of optimality and eliminate dispute waste. We characterize the optimal mechanism, which reduces to a convex program under mild conditions: the lowest- and highest-success-probability types receive first-best allocations, while interior types may be distorted upward or downward according to curvature virtual types. These results extend to settings with more than two outcomes.

Bio

Jiahua Wu is an Associate Professor of Operations at Imperial Business School, Imperial College London. His research examines how pricing, information, and other incentives shape performance in markets and operations. Recent projects focus on technology-enabled markets, including online platforms and AI-enabled services.

24 November 2026 @1pm BST

Artificial Intelligence in the Boardroom

Daniel Ferreira
LSE

Abstract

We analyze how the use of artificial intelligence affects the monitoring and advisory relationship between CEOs and corporate boards. AI can serve as a private advisor to CEOs, partially substituting for board advice, and thus reducing CEOs’ incentives to share information with directors. In equilibrium, firms respond by reducing board independence to restore information flows. Lower monitoring intensity decreases CEO turnover, making CEOs more entrenched. Lower dismissal risk allows firms to reduce CEO compensation. We show that AI adoption by the board mitigates some of the negative aspects of the CEO’s AI use, but cannot restore efficient levels of board monitoring. Our analysis suggests that the adoption of AI in the boardroom may have unintended consequences for corporate governance

Bio

Daniel Ferreira is Professor of Finance at LSE and Fellow of the European Corporate Governance Institute. He is also a department editor of Management Science, a council member of the Society for Financial Studies, a research fellow at CEPR, and an associate editor of The Economic Journal. He served as Head of the Department of Finance from 2020 to 2023. Daniel’s research lies at the intersection of corporate finance and organisational economics, with a particular focus on corporate governance. He has been at LSE since 2006 and holds a PhD in Economics from The University of Chicago.

5 May 2026 @2pm BST - In Person

Harjoat Bhamra,
Imperial College London

Leverage Dynamics and Learning about Economic Crises

(joint with Artur Anschukov and Lars-Alexander Kuehn)

Abstract

Models of learning about economic crises generate risk premia that rise at the onset of a crisis but then fall as belief uncertainty fades. Yet, empirical risk premia remain elevated during crises. We resolve this tension with leverage dynamics generated by the impact of learning on optimal capital structure decisions within a representative agent consumption-based model. Optimal leverage creates a feedback effect: learning increases risk premia, thereby depressing prices and further raising leverage. We structurally estimate the model and show it closely matches the joint dynamics of consumption, equity risk premia, credit risk, and leverage, especially during crises.

Bio

Harjoat Bhamra is a Professor of Finance at Imperial College Business School. Harjoat’s research focuses on dynamic models on corporate investment, capital structure, macro-finance, and monetary policy. He has published in leading journals such as American Economic Review, Review of Financial Studies, Journal of Economic Theory, and Journal of Monetary Economics. He is an Associate Editor for Management Science. Harjoat earned his PhD at London Business School.

Personal Website

30 April 2026 @2pm BST

High Frequency Liquidity Provision in Different Volatility Regimes (joint with Mehmet Saglam)

Yacine Aït-Sahalia,
Princeton University

Abstract

High frequency market makers (HFMMs) often serve as the dominant liquidity provider in equity, futures and, increasingly, other financial markets. They do so largely in the absence of constraints on their decisions to provide or remove liquidity, which they are able to modify on very short time scales. The question of how their liquidity provision can be expected to vary across different volatility regimes has significant implications for market stability. Do HFMMs provide less liquidity in extreme volatility environments? We address this question using a theoretical model with fully rational, optimizing HFMMs and provide empirical evidence.

Bio

Yacine Ait-Sahalia is the Otto Hack 1903 Professor of Finance and Economics and the inaugural director of the Bendheim Center for Finance at Princeton University. His primary areas of research are financial econometrics and mathematical statistics. He has made important contributions to the estimation and testing of continuous-time models in financial economics, and has served on the editorial boards of several leading journals, including Econometrica, Annals of Statistics, Journal of Finance, Review of Financial Studies, Journal of Econometrics, Finance and Stochastics, and Journal of Financial Econometrics. Wikipedia: https://en.wikipedia.org/wiki/Yacine_A%C3%AFt-Sahalia

3 March 2026 @11am GMT

Nancy Xu,
Boston College

Forecasting International Stock Market Variances

Abstract

We first examine 320 different forecasting models for international monthly stock return volatilities, using high frequency realized variances and the implied option variance as the predictor variables. We evaluate linear and non-linear models, and logarithmic transformed and weighted least squares estimation approaches. A logarithmically transformed Corsi (2009) model combined with the option implied variance (“lm4 log”) is robustly, across countries and time, among the best forecasting models. This conclusion survives when estimating a multitude of alternative models, including panel and MIDAS models, models embedding quarticity, leverage, downside, and jump risk. While not always the very best, the “lm4 log” model always generates forecasts extremely highly correlated with the very best model for a particular sample.

Bio

Nancy R. Xu is an Assistant Professor of Finance at the Boston College (Carroll School of Management), a Research Affiliate of Centre for Economic Policy Research (CEPR), and a Visiting Scholar at the Federal Reserve Bank of Boston. Her research lies at the intersection of asset pricing, macro-finance, and political economy, with a recent focus on how fiscal institutions, policy uncertainty, and political frictions shape risk premia and capital allocation. She has also served as a research consultant to a European Central Bank and previously interned at the Federal Reserve Bank of New York. She will join University of Cambridge Judge Business School as an Associate Professor in Summer 2026.

24 February 2026 @4pm GMT

Pedro Bordalo,
Saïd Business School, University of Oxford

Memory and Expectations

Abstract

We address the joint hypothesis problem in cross-sectional asset pricing by using measured analyst expectations of earnings growth. We construct a firm-level measure of Expectations Based Returns (EBRs) that uses analyst forecast errors and revisions and shuts down any cross-sectional differences in required returns. We obtain three results. First, variation in EBRs accounts for a large chunk of cross-sectional return spreads in value, investment, size, and momentum factors. Second, time variation in these spreads is predictable from that in EBRs, holding constant scaled price variables (as proxies for time varying required returns). Third, firm characteristics often seen as capturing risk premia predict disappointment of expectations and low EBRs. Overall, return spreads typically attributed to exotic risk factors are explained by predictable movements in non-rational expectations of firms’ earnings growth.

18 February 2026 @4pm GMT

Martin Herdegen,
University of Warwick

Portfolio Optimisation under Transaction Costs with Recursive Preferences

Abstract

We study the optimal investment-consumption problem for an agent whose preferences are governed by Epstein-Zin stochastic differential utility and who invests in a constant-parameter Black-Scholes-Merton market. We assume that purchases and sales of the risky asset are subject to proportional transaction costs. We fully characterise all parameter combinations for which the problem is well posed (which may depend on the level of transaction costs). We also provide a full verification argument that relies on no additional technical assumptions and uses primary methods only. Even in the special case of power utility, our arguments are significantly simpler and more elegant than the results in the extant literature. A novel key idea is to parametrise consumption and the value function in terms of the shadow fraction of wealth. The talk is based on joint work with David Hobson and Alex Tse.