Zipline-reloaded

What it is

Zipline is an event-driven backtester for Python. You write initialize and handle_data, it walks the trading calendar bar by bar, fills orders through a slippage and commission model, and hands back a pandas DataFrame. The piece with no close equivalent elsewhere is the Pipeline API: a declarative way to compute a factor across the whole universe on every bar without writing the cross-sectional loop yourself.

Quantopian wrote it, ran a crowd-sourced fund on it, and closed in late 2020 — the original repository's last commit is dated 14 October 2020. Zipline-reloaded is Stefan Jansen's fork, kept alive largely because his book Machine Learning for Algorithmic Trading runs on it.

Pricing

Free, Apache-2.0, nothing to buy. The real cost is data. The package ships Cython extensions, so a pip install needs a compiler where no wheel matches; conda-forge is the route the docs push.

Data & coverage

Exactly two bundles are registered out of the box: quandl, which downloads the Nasdaq Data Link WIKI dataset with a free API key, and csvdir, which reads your own CSVs. The docs state that WIKI was discontinued in early 2018 and no longer updates — so the default bundle is US daily bars that stop eight years ago. Fine for the tutorial, not for research.

Current data is a bundle you build or install. zipline-norgatedata (3.1.2, January 2025) and zipline-polygon-bundle (0.2.4, May 2025) are the third-party ones on PyPI, and each assumes you already pay that vendor. The --trading-calendar flag accepts any exchange_calendars code — XNYS is the default and the help text names XLON — but the data behind it is yours to supply.

Integrations

The same maintainer publishes the rest of the old Quantopian stack: pyfolio-reloaded for tearsheets, alphalens-reloaded for factor analysis, empyrical-reloaded for the risk statistics. zipline run -o writes a pandas pickle; every other format is your own to_csv.

Limitations

• No live trading and no paper trading. There is no broker adapter in the package at all.
• Equities and futures are the only asset types the engine knows. No options, ever.
• Daily or minute bars only — no tick data and no intrabar fill modelling.
• The hosted documentation lags the code: it still brands itself "Zipline 3.0" and says Python 3.8–3.11, while 3.1.1 requires 3.10 and supports 3.13.
• Maintenance is one person plus dependabot. Last release July 2025, last commit to main 13 November 2025.

Alternatives

Backtrader ships broker adapters, VectorBT is built for vectorised parameter sweeps, and QuantConnect's LEAN solves the data problem for you at the cost of running on their terms.