Build models that make better decisions.
Understand your dataset, select appropriate models, and evaluate predictions by real-world utility, not prediction error alone.
Try the tools without an account: 5 successful runs per rolling week.
| Model | FIS | MAPE | Sharpe | Max DD |
|---|---|---|---|---|
| Candidate A Utility leader | 0.65 | 15.3% | 1.12 | -18.2% |
| Candidate B Error leader | 0.51 | 0.09% | 0.87 | -22.4% |
| Candidate C | 0.48 | 12.8% | 0.94 | -19.7% |
| Candidate D | 0.43 | 8.2% | 0.76 | -25.1% |
The lowest-error candidate is not always strongest for a real decision. QuantSynth keeps prediction quality and decision utility visible side by side.
Why? Error metrics measure numerical closeness; decision metrics add directional, event, and utility context.
Illustrative values only. Validate rankings on your own held-out data.
Evidence, with the caveats visible
One workflow, before and after training
After training: Evaluation
Compare predictions with ground truth. Inspect error, FIS/CER decision utility, and, only when applicable, explicit trading assumptions.
- FIS/CER decision metrics
- MAE, RMSE, MASE, FIS, CER, and model comparisons
- Finance metrics only for suitable finance inputs
Before training: Decision Intelligence
Inspect dataset health, target and feature signals, leakage warnings, preprocessing guidance, and suitable model families.
- Dataset health score
- Automated model recommendations
- EDA + preprocessing automation
AutoML for Trading
A future execution layer for training and benchmarking candidates. It is not currently available for live model training or deployment.
- Candidate benchmarking is being designed
- No launch date is being promised
- Decision Intelligence remains available now
From dataset to defensible decision
Diagnose the dataset
Upload a CSV, confirm the target, and review quality, leakage, feature, and validation warnings before training.
Train outside Quantsynth
Use the suggested baselines and model families in your own reproducible training pipeline.
Evaluate held-out predictions
Upload predictions with actual outcomes, compare statistical quality and decision utility, then validate assumptions before deployment.
Diagnostics and model guidance
Baselines, challengers, validation
Error, utility, risk, limitations