From a strategy idea to a structured spec
FiClaw first breaks the natural-language description into universe, factors, signals, rebalancing, risk control and parameter constraints, rather than generating hard-to-review code snippets directly.
Methodology
How FiClaw organizes a reviewable quant research process
This page explains FiClaw's working boundaries: AI can raise research efficiency, but every backtest, parameter optimization and strategy conclusion should be reviewable by the team.
Generated code must enter backtest validation
Strategy code only has research value once it passes a historical backtest and metric review. FiClaw keeps code generation, backtest submission and result diagnosis in one flow.
Parameter optimization targets robustness, not a single-point best
Parameter search watches return, drawdown, Sharpe, win rate, turnover and neighborhood stability together, avoiding chasing the highest return of a single backtest.
Key conclusions need human review
FiClaw output is for research evaluation and team review. It is not investment advice and does not replace risk approval or pre-live validation.
Risk Boundary
Risk and boundaries
- Historical backtests do not promise returns and cannot guarantee future performance.
- Model-generated code, diagnosis and parameter suggestions may contain errors or omissions.
- Before going live, a strategy should pass out-of-sample validation, transaction-cost sensitivity analysis and human risk review.
- Actions involving trade execution, funds or client assets should not be automated directly by AI.
See examples, then run your own strategy
The example reports show how FiClaw organizes prompt, strategy summary, metrics and diagnosis into a reviewable result.