Researchers propose a new approach to financial reasoning using time series models. They create a taxonomy of capabilities for these models and develop FinSTaR, a system that can assess the current state or predict future behavior in financial markets. FinSTaR is trained on a benchmark dataset of S&P stocks and achieves higher accuracy than other models. The authors also show that different types of reasoning are complementary and mutually reinforcing when combined. --- Why it matters: This matters to researchers in AI because it addresses a long-standing challenge in applying time series models to financial domains, where unique characteristics often lead to poor performance. FinSTaR's success could enable more accurate financial forecasting and decision-making. Source: https://arxiv.org/abs/2605.03460