Researchers propose a new method called SPACE for multivariate time-series forecasting that improves upon existing methods by directly estimating the covariance geometry from the current forecast sample cloud. This approach allows for more accurate and efficient prediction regions, especially in situations where the underlying distribution is shifting. The authors claim that their method outperforms competing wrappers on diverse datasets. --- Why it matters: This matters to researchers and engineers working on AI because it provides a new tool for improving the accuracy of time-series forecasting, which is crucial in many applications such as finance, energy management, and weather forecasting. Source: https://arxiv.org/abs/2608.17333