The whole stack is free. Seriously.

The lingua franca of quantitative research. pandas is the library that makes time-series data — prices, returns, indicators — feel like a spreadsheet you can program.

Foundation. Every tool below assumes it.

Non-negotiable. If you learn exactly one thing from this page, learn pandas properly: indexing by time, resampling, rolling windows, and joins. Half of all backtest bugs are secretly pandas mistakes.

NumPy is the fast array math underneath pandas; Jupyter notebooks are the standard scratchpad for exploratory research — code, charts, and notes in one document.

Foundation. Where hypotheses get their first, roughest test.

Use notebooks for exploration, then move anything you trust into plain Python files with tests. Strategies that live their whole lives in a notebook tend to hide their bugs there too.

A vectorized backtesting library: it tests entire parameter grids as array operations, so thousands of strategy variants run in seconds instead of hours.

Fast, wide screening of ideas and parameter spaces.

Best-in-class for the survey stage — 'is there anything here at all?' Its speed is also its trap: the easier it is to test 10,000 variants, the more aggressively you must deflate whatever wins. Speed of testing and risk of overfitting are the same dial.

A classic event-driven backtesting framework: it steps through history bar by bar, the way live trading actually unfolds, with orders, positions, and broker simulation.

Realistic single-strategy backtests with order-level detail.

The gentlest introduction to event-driven thinking, with a decade of community answers for every error message. Development has slowed, but for learning the architecture of a trading system it remains the friendliest on-ramp.

A professional-grade, event-driven platform with a Rust core: the same code that backtests a strategy can run it live against a real broker.

The backtest-to-live bridge, at production quality.

The steepest learning curve on this page and the highest ceiling. Worth it when a strategy has already earned its way past the screening stage and you are engineering for real execution — not before.

An open-source algorithmic trading engine with a hosted platform on top: institutional-quality historical data for equities, options, futures, and FX, plus cloud backtesting.

Serious multi-asset backtesting when clean data is the bottleneck.

The hosted free tier is the cheapest legitimate access to survivorship-bias-free options data a beginner can get. The trade-off is platform lock-in and their cloud; the engine itself is open source and self-hostable.

A Python wrapper for Yahoo Finance data — free daily bars for stocks, ETFs, indices, and FX going back decades.

First dataset. Free, instant, good enough to learn on.

Fine for learning and daily-bar research; not fine for anything precise. Splits, dividends, and revisions are handled inconsistently, and it is an unofficial API that breaks periodically. Know what it is: a classroom dataset, not a production feed.

One unified API over a hundred-plus crypto exchanges — same code for fetching candles and order books everywhere, and the easiest free source of high-resolution intraday data.

Free intraday data and exchange connectivity practice.

Even if you never intend to trade crypto, ccxt is the cheapest way to practice against real, live, minute-resolution markets with a real exchange API. The engineering lessons transfer directly to any broker integration.