QSTrader

What it is

QSTrader is a Python backtesting engine shaped like an institutional book rather than like a script. An alpha model emits signals, a portfolio construction model turns them into target weights, an order sizer converts weights to share counts, a risk model can veto, and a simulated broker fills the result against the next bar. Replacing one module and leaving the other four alone is the whole appeal — and the reason a twenty-line moving-average idea costs more setup here than in a vectorised backtester.

What it simulates is narrow: schedule-driven, daily-bar, long/short cash equities and ETFs. The bundled examples are buy-and-hold, 60/40, a long/short book and a momentum tactical asset allocation; output is a matplotlib tearsheet plus a JSON statistics dump.

The headline fact is maintenance. The repository is not archived and the README still calls the software actively developed, but the last commit to master is 24 June 2024, release v0.3.0 went to PyPI the same day, and nothing has been pushed to any branch since 30 June 2024 — development stopped in May 2020 and advanced-algorithmic-trading in August 2020. Ten pull requests are open. The oldest has been waiting since January 2017; the newest was opened in July 2026 and has had no maintainer reply. Read it as a finished artefact with good bones, not a project you can file a bug against.

Pricing

Free, MIT-licensed, nothing to buy. QuantStart's actual business is the content sold beside it: the Advanced Algorithmic Trading ebook at $49, or $99 with source code, and the Quantcademy membership forum at $35 a month or $350 a year. The ebook is the trap — it teaches the earlier QSTrader parked on the advanced-algorithmic-trading branch, a different codebase from the package pip installs.

Data & coverage

None of its own. CSVDailyBarDataSource reads a directory of daily OHLCV CSVs in Yahoo's column layout, adjusting open and close for corporate actions — and since adjust_prices defaults to True, a file with no Adj Close column raises rather than falling back. There is no vendor adapter and no download helper.

Non-US data will load, but the simulation is US-shaped underneath. Each daily bar is split into an open and a close timestamped 14:30 and 21:00 UTC, and SimulatedExchange hardcodes those same hours with a standing TODO: Eliminate hardcoding of NYSE above them. There is no exchange calendar of any kind.

Limitations

• Daily bars only. DailyBusinessDaySimulationEngine walks Monday to Friday and its own docstring admits it ignores regional holidays, US or UK.
• Equities and cash are the only asset types; the data source's asset_type argument carries the comment "TODO: Unused at this stage and currently hardcoded to Equity".
• Market orders and nothing else. Order carries a quantity and an optional commission — no price, no order type, no time in force — and MarketOrderExecutionAlgorithm is the only execution algo shipped. Stops, limits and brackets are not expressible.
• No slippage and no market impact. SimulatedBroker accepts slippage_model and market_impact_model arguments and then assigns None to both, marked # TODO: Implement. Bid and ask are set to the same number by the data source, under its own comment that it is unable to distinguish them, so every fill crosses a zero spread. The default fee_model is ZeroFeeModel, which means a default backtest trades free.
• Cash is not enforced. If the order exceeds the portfolio's cash, the broker prints a warning and transacts anyway, leaving a negative balance. Nothing is scaled down.
• No live trading, no paper trading, no broker integration of any kind. The community offered three — Interactive Brokers, Oanda and IG — as pull requests in 2017, and all three are still open.
• Python 3.9 to 3.12 declared. Nothing newer is claimed or tested.
• Four pages of documentation. The rest is reading the source.

Alternatives

For daily-bar research with a live maintainer, zipline-reloaded or Backtesting.py; for allocation and rebalancing in the same shape, bt; for event-driven work that reaches real execution, NautilusTrader. The fork trading-strategy-qstrader exists on PyPI precisely because the original was unmaintained — but it stopped in February 2022 and is staler than what it forked. The rest of the field is in the backtesting category.