Financial Time Series Reasoning on FinTSR-Bench (Test A)
99.3Drawdown ScoreFinSTaR
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FinSTaRChain of Thought (CoT)=true2026.05 | 99.3 | 93.5 | 99.3 | 88.1 | 57.2 | 80.5 | 74.4 | 79.5 | 51.5 | 65.8 | 78.9 | |
| Qwen2.5-7B (w/ CoT)Training=SFT on FinTSR-Bench, Chain of Thought (CoT)=true2026.05 | 96.4 | 46.6 | 92.3 | 35.4 | 47.6 | 51.1 | 57.2 | 48.7 | 47.3 | 55 | 57.8 | |
| Gemma-2-9BType=Language Model2026.05 | 79.6 | 0 | 59.6 | 29.4 | 45.4 | 29.4 | 22.3 | 17.2 | 0 | 0.1 | 28.3 | |
| Qwen2.5-7B (w/o CoT)Training=SFT on FinTSR-Bench, Chain of Thought (CoT)=false2026.05 | 77.5 | 40.4 | 75.2 | 34.7 | 53 | 53.7 | 67.1 | 60.6 | 50.7 | 51.3 | 56.4 | |
| TimeOmni-1-7BType=TS Reasoning Model2026.05 | 75.7 | 35.6 | 50.5 | 31.5 | 49.1 | 54.1 | 59.3 | 58.2 | 50.7 | 48.4 | 51.3 | |
| Qwen2.5-7BType=Language Model2026.05 | 72.9 | 37.1 | 76.9 | 26.9 | 52.2 | 51.2 | 54.1 | 44.9 | 49.2 | 50.4 | 51.6 | |
| Llama-3.1-8BType=Language Model2026.05 | 57.2 | 30.3 | 75.7 | 1.4 | 47.3 | 50.2 | 47.3 | 48 | 29.9 | 33.6 | 42.1 | |
| Random2026.05 | 25 | 33.3 | 20 | 33.3 | 50 | 50 | 50 | 50 | 50 | 50 | 41.2 | |
| Mistral-7BType=Language Model2026.05 | 8.8 | 0.7 | 1 | 0 | 45.3 | 52.4 | 4.9 | 40.4 | 23.7 | 34.8 | 21.2 | |
| TimeMQA-LlamaType=TS Language Model2026.05 | 6.9 | 7.5 | 6.6 | 3.3 | 16 | 20.4 | 17.6 | 7.2 | 8.7 | 9.7 | 10.4 | |
| TimeMQA-QwenType=TS Language Model2026.05 | 5.9 | 1.6 | 3.3 | 0.4 | 1.2 | 2.1 | 1.6 | 1.7 | 1.4 | 2 | 2.1 | |
| TimeMQA-MistralType=TS Language Model2026.05 | 5.5 | 1.6 | 4.7 | 1.3 | 2.5 | 11.5 | 10.8 | 11 | 9.9 | 5.4 | 6.4 | |
| Phi-3.5-miniType=Language Model2026.05 | 1.8 | 0.4 | 9.6 | 8.2 | 0.2 | 0.8 | 1.2 | 2 | 1.5 | 1.7 | 2.7 |