In briefSocial welfare and price discovery in double auction markets explained
SummaryDouble auctions are known from lab experiments to drive prices toward competitive equilibrium, but no theory explained this. The paper studies dynamic double auctions in a pure exchange economy where agents bid indifference prices implied by their holdings and preferences. It shows that Walras equilibria coincide with the fixed points of the double auction, and that repeated auctions produce bounded sequences of allocations and prices whose cluster points are Walras equilibria with transfers.
In briefVerifying Agent Claims by Evidence, Not Aggregate Scores
SummaryThis paper proposes a claim-specific verification audit for modular agents that plan, act, check and refine, scoring the evidence behind each conclusion rather than an overall task score. Each conclusion receives one of four verdicts (supported, unsupported, unresolved, or not evaluated) along with its supporting evidence and scope. The authors apply oracle policies and component replacement to a constrained portfolio-allocation agent in a synthetic market with hidden regimes. The main contribution is the protocol itself, and the empirical findings apply only to the agent and environment studied.
In briefPPO-HRAP: A Risk-Controlled Trading Policy Blending PPO with Regime-Aware Exposure
SummaryThis paper proposes PPO-HRAP, a reinforcement learning trading method that combines Proximal Policy Optimization with an interpretable, volatility-aware regime prior. Its reward blends portfolio log return, a VIX-conditioned drawdown penalty, target-exposure deviation, and turnover cost. On a held-out 2020-2022 SPY test window, it reports 27.62% total return and reduces maximum drawdown from 34.10% (Buy and Hold) to 18.47%. The authors note high turnover remains, and cross-asset results on QQQ and DIA are single-run only.
In briefPricing American options under stochastic local volatility and stochastic correlation using RBSDEs
SummaryThis paper studies American option pricing in stochastic local volatility (SLV) models extended with stochastic correlation driven by an additional stochastic process. The authors derive the corresponding reflected forward-backward stochastic differential equations (RBSDEs) and solve them with data-driven numerical methods for pricing and hedging. They also provide a convergence analysis of the method and numerical experiments assessing the extended models and the accuracy and efficiency of the approach.
In briefOptimal design of catastrophe bonds and their use for climate change risk
SummaryThis paper treats the payout function of catastrophe bonds as a decision variable and studies their optimal design from the sponsor's perspective. It shows that a piecewise linear indemnity function is optimal under expected-disutility and spectral-risk-measure frameworks, reducing the design problem to a one-dimensional choice of attachment point. Calibrated with U.S. severe storm loss data, the optimal designs are reported to achieve substantial risk reductions across most risk criteria.
In briefBeyond Supra-Competitive Outcomes: Collusive Behaviour in Deep RL Trading Games
SummaryThis paper studies a two-player, finite-horizon liquidation game in which independent reinforcement learning agents reach costs below the Nash benchmark. The authors identify a learned punishment mechanism: when one agent deviates, the other accelerates liquidation, costing the deviator more than it gains while leaving the punisher's average payoff largely unchanged. Two formal checks show the punishment outweighs the deviation gain and that the change in trading explains the loss, supporting a collusive interpretation.
