{"paper":{"title":"FinSTaR: Towards Financial Reasoning with Time Series Reasoning Models","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"A 2x2 taxonomy of time series capabilities with tailored chain-of-thought strategies enables 78.9 percent accuracy on financial reasoning tasks from S&P stocks.","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Dongwan Kang, Hwanil Choi, Jaehoon Lee, Jun Seo, Minjae Kim, Seunghan Lee, Soonyoung Lee, Sungdong Yoo, Tae Yoon Lim, Wonbin Ahn","submitted_at":"2026-05-05T07:46:39Z","abstract_excerpt":"Time series (TS) reasoning models (TSRMs) have shown promising capabilities in general domains, yet they consistently fail on financial domain, which exhibit unique characteristics. We propose a general 2x2 capability taxonomy for TSRMs by crossing 1) single-entity vs. multi-entity analysis with 2) assessment of the current state vs. prediction of future behavior. We instantiate this taxonomy in the financial domain -- where the distinction between deterministic assessment and stochastic prediction is particularly critical -- as ten financial reasoning tasks, forming the FinTSR-Bench benchmark"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The proposed method achieves 78.9% average accuracy on FinTSR-Bench, substantially outperforming LLM and TSRM baselines. Furthermore, we show that the four capability categories are complementary and mutually reinforcing through joint training, and that Scenario-Aware CoT consistently improves prediction accuracy over standard CoT.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the ten tasks constructed from S&P stocks adequately capture the distinctive challenges of financial reasoning and that the deterministic-versus-stochastic distinction is the primary reason current models underperform.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"FinSTaR reaches 78.9% accuracy on a new financial time series reasoning benchmark by applying Compute-in-CoT for deterministic assessments and Scenario-Aware CoT for stochastic predictions.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A 2x2 taxonomy of time series capabilities with tailored chain-of-thought strategies enables 78.9 percent accuracy on financial reasoning tasks from S&P stocks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9e643edef0ce0de4df87f401f8add687028b68cde690bdcca2893d52b84e7a1e"},"source":{"id":"2605.03460","kind":"arxiv","version":2},"verdict":{"id":"43461c49-c64c-4b6e-a79d-0e695271a4f2","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-07T16:47:45.280055Z","strongest_claim":"The proposed method achieves 78.9% average accuracy on FinTSR-Bench, substantially outperforming LLM and TSRM baselines. Furthermore, we show that the four capability categories are complementary and mutually reinforcing through joint training, and that Scenario-Aware CoT consistently improves prediction accuracy over standard CoT.","one_line_summary":"FinSTaR reaches 78.9% accuracy on a new financial time series reasoning benchmark by applying Compute-in-CoT for deterministic assessments and Scenario-Aware CoT for stochastic predictions.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the ten tasks constructed from S&P stocks adequately capture the distinctive challenges of financial reasoning and that the deterministic-versus-stochastic distinction is the primary reason current models underperform.","pith_extraction_headline":"A 2x2 taxonomy of time series capabilities with tailored chain-of-thought strategies enables 78.9 percent accuracy on financial reasoning tasks from S&P stocks."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.03460/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T13:40:49.333291Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-20T01:01:21.890033Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T15:23:21.754179Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"931eeaeb87e60099f7fd16af8722e29263ba4a58d0a71c144438dfc378345c36"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}