Compare

sigc vs vectorbt

Vectorised backtesting on top of NumPy/Numba in Python.

vectorbt is a Python library that gives you very fast vectorised backtests by leaning hard on NumPy and Numba JIT compilation. It is excellent for portfolio simulations and parameter sweeps inside a notebook. sigc is a typed compiled DSL with a Rust runtime and a content-addressed cache. The two solve different problems.

Feature

sigc

vectorbt

Language

Domain-specific .sig DSL (compiled)

Python library, idiomatic NumPy/pandas

Type checking

Compile-time shape and dtype checks before execution

Runtime: shape errors surface during the simulation

Execution engine

Rust + Polars/Arrow + Rayon, SIMD kernels (AVX2/AVX-512 where available)

NumPy + Numba JIT compilation; broadcasts across parameter grids

Determinism / caching

Content-addressed cache (sled + blake3); identical IR + inputs = identical bytes

No built-in content-addressed cache; reproducibility is up to the user

Backtester features

Long-short rank weights, vol-targeting, transaction costs (bps + square-root slippage), benchmark attribution

Portfolio simulation with stop-loss, take-profit, signal-based entries/exits, and broad metrics

Parameter optimisation

Params block; GridSearch via the CLI / Rust embedding

First-class: parameter sweeps are a core idiom (broadcasting across the parameter axis)

Production mode

Same binary as `sigc daemon`; nng REQ/REP on tcp://127.0.0.1:7240

Library; production deployment is BYO

Risk / circuit breakers

sigc.yaml: max drawdown, max position, kill switch, order rate limits

Not in scope as a library; you wire it up

Embed in another system

Rust crate; Strategy::from_file / Backtest API

Python import; works wherever Python works

Language ecosystem

Rust; pysigc bindings for notebooks

Python-first, deep integration with the data-science stack

License

MIT

Apache 2.0 (vectorbt) / commercial (vectorbt.pro)

Pick vectorbt when

You live in a Python data-science workflow, you run heavy parameter sweeps inside notebooks, and your strategy logic stays inside the Python ecosystem end-to-end.

Pick sigc when

You want compile-time guarantees on signal shape, deterministic content-addressed caching for reproducibility, and the same binary running research and production over nng.

Honest note

vectorbt is faster than naive pandas at parameter sweeps because it broadcasts across the parameter axis. sigc does not try to compete on parameter-sweep throughput; it competes on type safety, reproducibility, and a unified research-to-production binary. Both libraries can be embedded in larger systems; the embedding language is the dividing line.

Back to overview Read the docs