VectorBT

What it is

vectorbt backtests by broadcasting rather than by looping. A strategy is a pair of boolean signal arrays; a thousand parameter combinations are a thousand columns of the same array, and one Portfolio.from_signals call simulates them together through compiled kernels. Grid searches that take an event-driven backtester hours finish in seconds. The price is a mental model most people find alien: you stop thinking in bars and start thinking in matrices, and a strategy with genuine path dependence is awkward to express.

Liveness, in numbers: the repository is not archived, v1.0.0 landed on 22 April 2026 and added an optional Rust backend alongside Numba, v1.1.0 followed on 5 July 2026 with Python 3.14, pandas 3 and NumPy 2.4 support, and the last commit to master was 2 August 2026. 121 issues are open, excluding pull requests.

Pricing

Nothing here is for sale — pip install vectorbt, no account and no key. The licence is what to read instead of a price list. It is Apache 2.0 with the Commons Clause v1.0, which withholds only the right to Sell: to provide third parties, for a fee, a product or service whose value derives substantially from the software. Trading a firm's own capital on it is not selling it. Charging clients for a service built on it is, and needs a written agreement with the author.

Data & coverage

vectorbt ships no market data and no feed of its own. The optional [full] extra installs yfinance, python-binance, ccxt and alpaca-py for the built-in connectors; anything else arrives as a pandas DataFrame with a datetime index, which is all the library requires. Asset classes, markets and latency are properties of your feed, not of this tool.

Integrations

TA-Lib, ta and pandas-ta-classic for indicators, QuantStats for tearsheets, Plotly and ipywidgets for heatmaps, Ray for distributed sweeps, and schedule plus python-telegram-bot for unattended updates and notifications. The [rust] extra installs the vectorbt-rust engine, which dispatches per call and falls back to Numba where Rust has no kernel.

Limitations

• No live trading. No broker connection, no order routing, no paper account — execution is somebody else's problem.
• A vectorised simulator fills at bar prices. Intrabar sequencing, queue position and partial fills are approximations, and stops are checked against OHLC rather than against the tape.
• Not OSI open source, which matters wherever a policy says open source and means it.
• Python 3.11 to 3.14 only, and v1.1.0 expects pandas 3 and NumPy 2.4 or newer. An older stack stays on the 0.28 line.

Alternatives

Backtesting.py is far smaller and event-driven, Backtrader is the classic looping engine, and NautilusTrader is the one to reach for when the same code has to place real orders. VectorBT PRO is the author's paid product — a private repository from $25 a month, and a much narrower licence: it grants Private Use only, defined as use from which no "monetary compensation, commercial advantage, or for-profit purpose" is obtained, and limits an organisation to internal evaluation, prototyping and R&D "provided it does not result in a commercial product, service, or outcome". The Commons Clause here stops at forbidding sale to third parties, so for a firm trading its own capital the free package is the more permissive of the two.

Compare it against the rest of the backtesting frameworks.