arXiv:2401. 14483v4 Announce Type: replace Abstract: In the current practices of machine learning, the evaluation of forecasts has become a cornerstone of scientific progress.
arXiv:2605. 26703v2 Announce Type: replace-cross Abstract: The classic concept of "calibrated forecasts" and its more recent refinement, "calibeating," are defined with respect to the standard quadratic scoring rule.
The paper investigates how different proper scoring rules influence the performance and behavior of large language model (LLM) forecasters. Five scoring rules were compared as training objectives for binary forecasts of real-world events, revealing that while they all theoretically incentivize truthful probability reporting, they produce models with varying calibration, probability usage, and bias, information, and noise profiles. The Brier-trained model achieved the lowest Brier score and highest AUC-ROC, whereas the log-trained model achieved the best log score and lowest calibration error, indicating that scoring rule choice can shape both forecast accuracy and error structure.
arXiv:2607. 00164v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards can in principle train calibrated probabilistic forecasters, since a proper scoring rule such as the Brier score is computed from outcomes alone and is minimized in expectation by the true probability.
arXiv:2607. 06166v1 Announce Type: new Abstract: Prediction markets aggregate dispersed beliefs into prices that act as probabilistic forecasts of uncertain events.
The paper argues that traditional weather forecast evaluations, which focus on statistical comparisons between forecasts and observations, do not adequately capture how forecasts influence real-world decisions. It introduces decision calibration, a framework that assesses probabilistic forecast performance from the decision-maker’s perspective. Using this framework, the authors compare a machine learning model to a classical numerical weather prediction model across various weather-dependent decision tasks, finding that forecast-level performance does not reliably predict decision-level outcomes and that model rankings can shift depending on the decision context.
The paper introduces rankECE, a new metric for assessing calibration error in predictive models. Unlike the widely used Expected Calibration Error (ECE), rankECE compares predictions with neighboring probability values, offering theoretical guarantees and empirical evidence that it better approximates ECE than traditional binned methods.
The paper demonstrates that the outcome of a forecasting leaderboard is largely determined by the evaluator’s design choices rather than the models themselves. By fixing the data, horizon, and period, the authors varied three key evaluation decisions—unit of analysis, error pooling, and scoring metric—and showed that each can reverse or eliminate the apparent superiority of any forecasting method. The study also evaluates the practical impact of these choices on a deployed system, revealing that the selection rule captures a significant portion of the potential performance gain, and confirms the findings on an external public dataset.
arXiv:2607. 17765v1 Announce Type: cross Abstract: We introduce WC2026-Agents, a benchmark and dataset for evaluating large language models (LLMs) as autonomous forecasting agents on real, future events.
arXiv:2607. 03015v1 Announce Type: new Abstract: Forecasting future events has attracted growing attention as a testbed for general-purpose AI.
arXiv:2607. 10972v1 Announce Type: new Abstract: Many evaluations of model outputs rely either on contracts checkable at evaluation time or on feedback that arrives within the operating loop.
Many evaluations of model outputs rely either on contracts checkable at evaluation time or on feedback that arrives within the operating loop. We study the complementary setting in which ground truth is delayed, censored, or private, so deterministic code cannot check correctness at scoring time and must instead issue a code-owned provisional forecast.
arXiv:2606. 15917v1 Announce Type: new Abstract: We use Group Relative Policy Optimization (GRPO), a recently devised sample and memory efficient reinforcement learning method, to finetune pretrained LLMs in the range of 1.