Main Conference Day OneTuesday, 17 November 2026 - GMT (Greenwich Mean Time, GMTZ)

The industrialization of quantitative investing and the rise of QIS platforms have dramatically expanded the universe of investable strategies. As implementation becomes increasingly standardized, the key challenge is no longer to generate alpha, but to select the most appropriate strategy for prevailing market conditions. This talk explores how the quant problem is shifting from alpha discovery to alpha selection, and discusses possible frameworks for defining performance and quantifying risk in a world of almost limitless choice.

We analyse liquidity risk in decentralised finance (DeFi) lending protocols, focusing on bad debt arising from toxic liquidations and liquidity pool depletion caused by large-scale liquidations or bank-run-like dynamics. We examine how inadequate liquidity management can generate systemic vulnerabilities and pro-cyclical feedback effects, even in systems that appear robust under normal conditions. Although over-collateralisation is intended to protect lending platforms, it remains unclear whether this mechanism alone is sufficient to prevent bad debt accumulation or liquidity shortages capable of triggering contagion across the wider DeFi ecosystem. To address these issues, the paper develops an agent-based model of an Aave-like decentralised lending protocol to analyse the incidence, scale, and dynamics of liquidity risk events. Particular attention is given to the interaction between liquidation mechanisms, protocol parameters, and user incentives in generating instability and pro-cyclical behaviour.

We look at XVA through the lens that underlies deep hedging, viewing it as the result of hedging under a utility function. This reveals a generalisation of standard XVAs, generating the classic CVA and FVA, along with HVA (hedging costs) and a number of non-classical contributions that arise from the realities of a real-world hedging strategy. The full framework leads to formulae that can be solved by approximation or via a deep learning approach.