FinSTaR: A Financial Time Series Reasoning Model

Researchers propose FinSTaR (Financial Time Series Thinking and Reasoning), a model trained to address the failure of general time series reasoning models (TSRMs) in the financial domain. They introduce a 2x2 capability taxonomy for TSRMs, crossing single-entity vs. multi-entity analysis with assessment of current state vs. prediction of future behavior. This taxonomy is instantiated as ten financial reasoning tasks forming the FinTSR-Bench benchmark based on S&P stocks. FinSTaR is trained on FinTSR-Bench with distinct chain-of-thought (CoT) strategies: for deterministic assessment tasks, it uses Compute-in-CoT, a programmatic CoT that enables models to derive answers from observable data. The work is published on arXiv as paper 2605.03460.