vectorbt.
Vectorized backtesting for research-speed iteration at parameter-sweep scale.
Overview
vectorbt reframes backtesting as array computation: strategies are expressed as NumPy/pandas operations over entire parameter grids simultaneously, making thousand-variant sweeps feasible on a laptop. It trades event-driven realism for iteration speed — the correct trade in the hypothesis-screening phase of research.
Where it fits
- Screening signal ideas across large parameter grids before committing to slow, realistic simulation.
- Researchers fluent in pandas who want backtesting to feel like data analysis.
- Robustness analysis: heatmaps of performance across parameter space rather than a single cherry-picked path.
Constraints to weigh
- Vectorized fills idealize execution; results overstate capacity and understate costs versus event-driven simulation.
- Path-dependent logic (queueing, partial fills, intraday risk rules) fits the paradigm poorly.
- The open-source version and the commercial PRO fork differ in features — check which docs you are reading.
In an agentic workflow
Parameter-sweep tooling is a double-edged instrument for agents: it accelerates honest robustness analysis and industrial-scale overfitting equally. Pair sweep results with pre-committed out-of-sample rules — the validation discipline documented in the governance framework.
This profile is an independent editorial description; verify current pricing, licensing, and capabilities directly with the provider. Not affiliated with or endorsed by vectorbt.