Forecasting
Compare features, horizons, model families, and training schemes for time-dependent prediction.
Autoresearch for financial models under real constraints.
Give Reverie your datasets, candidate modeling space, evaluation windows, and objective. It trains and compares models while respecting the risk and operating constraints you define.
Where Reverie fits
Model research where evaluation design matters as much as fit.
Financial models must survive time, costs, changing conditions, and operational limits. Reverie can search a broad action space while applying the same disciplined evaluation to every candidate.
Risk modeling
Optimize predictive quality alongside calibration, stability, tail behavior, and policy constraints.
Ranking and allocation
Search models that prioritize opportunities or allocate resources under practical limits.
Run a disciplined model search across methods, features, and market conditions.
Optimization target
Define the forecasting, ranking, pricing, allocation, or risk objective and its guardrails.
Financial data
Connect structured time series, event data, fundamentals, alternative data, or internal datasets.
Action space
Specify candidate features, architectures, estimators, losses, and training procedures.
Robust evaluation
Train and compare models across the windows, scenarios, costs, and constraints your team requires.
Reviewable outputs
Receive trained models, comparative metrics, artifacts, and a report covering the research lineage.
Example project
Search forecasting models with walk-forward evaluation.
The project definition keeps every candidate grounded in the same information set, timing assumptions, costs, and operating constraints.
- Objective
- Improve out-of-sample forecast quality while satisfying stability, latency, and risk thresholds.
- Dataset
- Time-stamped panel data, event features, fundamentals, internal signals, and explicit availability dates.
- Action space
- Feature transforms, lookback windows, estimators, neural architectures, objectives, regularization, and ensembles.
- Evaluation
- Walk-forward windows with leakage controls, regime analysis, transaction or decision costs, and capacity constraints.
- Outputs
- Trained candidates, reproducible backtests, robustness analysis, and a report comparing performance and tradeoffs.
Optimize for more
than one headline metric.
Financial model quality depends on stability, costs, risk, and behavior across regimes. Reverie lets your evaluation encode the full set of requirements.
Time-aware. Define evaluation windows and validation procedures that reflect how the model will actually be used.
Constraint-aware. Include turnover, latency, capacity, stability, risk, and other practical requirements.
Auditable. Preserve the configuration, data path, training run, and result behind every candidate.
Delivered to your team
A complete package for model review and validation.
Candidate models
Trained artifacts, features, configurations, data windows, and the environments needed to reproduce each result.
Robustness evidence
Performance by period and regime, calibration, sensitivity, failure modes, and constraint compliance.
Research report
A traceable summary of the search, rejected directions, leading candidates, and the reasons behind final selection.