Researchers have introduced FinVerse, a benchmark for evaluating the forecasting ability of time-series foundation models in financial domains. Unlike existing benchmarks, FinVerse takes into account the economic relevance of each time series and assigns specific evaluation metrics based on its underlying meaning. This approach aims to provide more realistic evaluations of model performance. The benchmark includes 116,897 financial time series with over 171 million observations, and an analysis of 43 public models shows that strong performance under generic criteria does not necessarily translate to useful financial forecasts. --- Why it matters: This matters because current benchmarks often focus on uniform error-based metrics, which may not accurately reflect real-world decision-making. FinVerse provides a more nuanced evaluation framework that can help researchers and developers improve the forecasting ability of their models in financial domains. Source: https://arxiv.org/abs/2608.03259