Backtrader

What it is

Backtrader is an event-driven backtesting framework for Python. You subclass bt.Strategy, hand data feeds and indicators to a Cerebro engine, and it walks the bars one at a time calling next(), filling orders through a simulated broker. Event-driven rather than vectorised is the design decision that defines it: it costs speed and buys order semantics a vectorised engine cannot express — limit, stop, stop-limit, trailing stop, OCO and bracket orders, slippage models, volume-based filling, and continuous cash adjustment for futures-like instruments. 122 indicators ship with it, TA-Lib plugs in, and the documentation is unusually complete for a project of this size.

It is also frozen. Last commit to master and last PyPI release are both version 1.9.78.123, dated 19 April 2023. The repository is not archived, GitHub issues are turned off, and 63 pull requests are open — the newest from July 2026, one of them a May 2026 fix for a Python 3.10+ deprecation that is still unmerged. setup.py declares neither python_requires nor a single dependency, so pip drops it into any interpreter without a word of complaint; the package's own classifiers stop at Python 3.7. The community forum the README names as the support channel in place of issues returned a 522 on every attempt on 14 September 2026.

What the simulated broker models is unusually complete for a free library: percentage or fixed slippage with separate switches for whether opening prices and limit orders slip, volume fillers that cap a fill at a share of the bar's volume, commission schemes that charge a percentage or a fixed amount per contract and can add a yearly interest rate for holding a short or a leveraged position, and a cheat-on-open mode for strategies that size off the opening print. Fifteen analyzers — Sharpe, SQN, VWR, Calmar, drawdown, time return, transactions, a trade breakdown — turn a run into statistics without a second library, and optstrategy spreads a parameter grid across processes.

Data & coverage

None of its own. Built-in feeds read CSV files, pandas DataFrames, Yahoo, and the live stores; everything else is a feed you write or install. Resampling, replay and multi-timeframe are built in, as are trading calendars for non-US sessions, so the framework is market-agnostic and the coverage is whatever you pay someone else for.

Integrations

Live data and trading through Interactive Brokers (IbPy), VisualChart and Oanda v1. Plotting is matplotlib. pip install backtrader pulls no dependencies at all unless you ask for plotting, and a btrun command-line runner is installed alongside the library.

Limitations

• Unmaintained since April 2023. Every Python, numpy or matplotlib incompatibility from here is yours to patch or pin around.
• The Oanda store targets an API Oanda retired, and the IB store needs IbPy, archived since January 2017; pyfolio integration is marked deprecated in the project's own README.
• Event-driven means slow. Sweeping a parameter grid over a large universe is where people leave for a vectorised engine.
• No options modelling — no chains, no greeks, no expiry handling. An Options support pull request has been open since September 2025.
• GPL-3.0-or-later. Embedding it in a closed-source product is a licensing decision, not a free one.

Alternatives

Backtesting.py and VectorBT for speed on parameter sweeps, NautilusTrader and QuantConnect LEAN for a maintained event-driven engine with live brokers, Zipline-reloaded for the pipeline API. Within the backtrader family itself, backtrader_next is the fork with 2026 releases.

The rest are in the backtesting frameworks category.