In briefHow Well Do LLMs Perform in Peer Review? Evidence from Finance and Accounting
SummaryThe paper has GPT-4o referee papers from the 2024 AFA and NBER finance meetings and compares its verdicts with publication in top finance journals over the next two years. Papers scored one standard deviation higher were about 44% more likely to be published, net of author reputation. The predictive information came from judgements of correctness and exposition, not originality or significance. The authors conclude LLMs can check whether research is right but not whether it matters.
In briefRisk Measures for the Climate Transition Age: Why Symmetric Risk Budgets Fall Short
SummaryThis paper argues that common risk measures such as variance, standard deviation and tracking error penalise outperformance as heavily as underperformance, which conflicts with climate transition investing. The authors propose downside-only alternatives, including semi-variance, standard risk and trailing error. They also suggest specific mandate design changes that would let institutional investors pursue transition opportunities within risk-aware frameworks.
Lars Dijkstra, Andreas G. F. Hoepner, Fabiola I. Schneider et al.
In briefPortfolio decarbonization when carbon and transition information is incomplete
SummaryThis paper studies how an investor can dynamically decarbonize a portfolio when firms' carbon intensities and the transition state are only partly observable. The investor maximizes expected utility of terminal wealth minus a penalty on perceived cumulative financed emissions. In a numerical example, incomplete information has limited effects when prior beliefs are well specified, but biased beliefs cause welfare losses and opposite portfolio tilts that affect the chance of exceeding the emissions target differently.
In briefOn integrability of local utility maximizing strategies
SummaryThis paper studies local utility maximizing strategies in stochastic control, where a local optimizer is derived from a global criterion such as expected utility. The existence of such a local optimizer can be established under very mild conditions, so the key question is when it has good integrability properties. The authors show that the reasonable asymptotic elasticity condition of Kramkov and Schachermayer (1999), applied to the local utility function, plays a central role.
In briefStructured SDEs for Realized GARCH: A Continuous-Time Volatility Model
SummaryThis paper proposes a continuous-time stochastic volatility model for forecasting next-day integrated variance. It feeds a jointly evolving realized-measure state into latent variance dynamics and adds correlated Brownian innovations linked to returns. Parameters are estimated via simulation-based predictive learning, with Monte Carlo simulation used for forecasts. Across 17 equity indices, the model shows strong out-of-sample performance against several benchmarks, with realized-measure feedback contributing most of the gain and return-linked Brownian dependence adding smaller complementary improvements.
In briefBacktesting Expected Shortfall without VaR forecasts in time-series models
SummaryThis paper proposes two e-backtesting methods for Expected Shortfall that use only observed losses and ES forecasts, with no Value-at-Risk forecast required. One targets iid losses, and the other targets time series with an AR-GARCH conditional VaR structure. Under the stated assumptions, both methods control false rejections at any stopping time. Simulations and an application to daily NASDAQ losses compare the methods with VaR-ES e-processes and fixed-sample ES backtests.
WhyProposes VaR-free sequential e-backtests for ES under AR-GARCH, a novel formulation with proofs, simulations and NASDAQ application; useful for risk management.
In briefDesigning Independent Variation: Guided Iteration for Price Experiments in Correlated Panels
SummaryThis paper argues that estimation error in short, correlated pricing panels is driven mainly by trajectory dispersion rather than sampling noise, so only independently assigned price variation can resolve the identification shortfall. It proposes a guided iteration method that uses a cyclic price-perturbation scheme under a revenue-at-risk budget to decide how much variation to generate, where, and for how long. The method is applied to Dominick's retail data, with theoretical results and empirical comparisons reported.
In briefDirichlet sub-mixtures: a new view of multiplicative risk models with Pareto margins
SummaryThe paper builds a class of multiplicative background risk models with Pareto margins by combining Marshall-Olkin exponential and additive gamma risk factor structures, each including idiosyncratic and systemic background factors. Under an admissibility condition, the joint survival function is shown to be a finite sub-mixture of Pareto survival functions, with Dirichlet-probability weights over orderings that sum to at most one. The authors also study tail behavior, stating that asymptotic tail dependence and exceedance decay rates are governed by the gamma risk factors alone.
In briefThe Buffett Effect: How Famous Investors' Stock Disclosures Move Markets
SummaryThis paper examines whether disclosed new stock positions by recognizable investors trigger larger abnormal returns than similar disclosures from non-recognizable managers. The authors build a recognition index from Wikipedia pageviews and news mentions and compare 24 well-known funds with 24 matched control funds from 2000 to 2025. They find a positive disclosure premium that grows with recognition, and portfolios copying the most recognized investors earn significant alpha lasting months.
