awesome-systematic-trading

A curated catalog of the systematic-trading field, maintained by the paperswithbacktest organization: 97 libraries and packages, 40+ strategies from academic papers, 55 books, 23 videos, plus blogs and courses. ~~~8.8k★ / 1.2k forks; created February 2022, last pushed January 2025. That makes it a static map of the field as of early 2025, not a live one.~~ Re-checked 2026-09-27 against the GitHub API and the live README: the repo came back to life. 14,467★ / 1,737 forks, last pushed 2026-09-03, with merges through late August and September 2026 (24 libraries and 4 engineering blogs imported from another awesome list on 09-03, a monthly link check added in August). The library count is now 136, dead and dormant projects are flagged inline (zipline "dormant since 2024-02", backtrader "dormant since 2024-08"), and a Japanese README joined the Chinese one. Still no license file.

The replication record (added to the README by 2026-09)

The README now opens with numbers the July version did not carry. The maintainers say they coded and ran 4,843 papers over each paper's own full history, and report: median replication Sharpe 0.37; 48% clear a t-statistic of 1.96; median test window 34 years; median beta +0.17 to the S&P 500, which takes the median information ratio to 0.21; and across 2,838 papers with data on both sides of publication, no measurable post-publication decay once the market period is controlled for. These are the maker's own figures about its own replication pipeline, with the method on the company's wiki, so they sit at T3 like the rest of the page. Nobody outside paperswithbacktest has re-run them. The decay finding is the one to hold loosely: it cuts against the academic literature on anomalies weakening after publication, and here it rests on one vendor's catalogue, not on a paper anyone reviewed. It is still the first source in this spoke that states its N, and that is edge #15's question (see synthesis).

This is the first source here that describes the whole field rather than one system. Every prior ingest was a single tool (banbot, tensortrade, trademaster) or a single practitioner (multithreaded-ml-trading-robot, smolinski-h1-2026-aar); this one is the map they sit on.

How it carves up the software

The library taxonomy is itself a claim about the field's shape, and it lines up with the spoke's predict → decide → execute stack while adding pieces the corpus never had:

Backtesting and live trading, split three ways — event-driven frameworks (vnpy, zipline, backtrader, QuantConnect's Lean, nautilus_trader, backtesting.py, hftbacktest, plus Go and Rust entries), vector-based frameworks (vectorbt, pysystemtrade, bt), and crypto (Freqtrade, Jesse, OctoBot, Hummingbot). The event-driven / vector-based split is the significant one — see vectorized-backtesting.
Trading bots — alpha models, many unmaintained (the list says so).
Analytics, subdivided into indicators, metrics computation, optimization, pricing, and risk. The last three are the corpus's blind spot: portfolio optimization (PyPortfolioOpt, Riskfolio-Lib, Deepdow), derivative pricing (Google's tf-quant-finance, FinancePy, a QuantLib wrapper), and risk analytics (pyfolio) are established tool categories this wiki has no pages for.
Broker APIs · data sources · databases · graph computation · machine learning · time-series analysis · visualization — the plumbing under all of it.

Ranking inside each table is by GitHub stars, which sorts by adoption rather than quality.

The strategy tables — a paper, a Sharpe, and runnable code

The strategy half is the more unusual part. As read in July 2026, each entry carried a title, Sharpe ratio, volatility, rebalancing frequency, a QuantConnect implementation, and a link to the source paper, sorted by descending Sharpe. Time Series Momentum sat at 0.576 with 20.5% volatility, rebalanced monthly; Paired Switching at 0.691 and 9.5%, quarterly. ~~The list also keeps entries that didn't work: Short Term Reversal with Futures is in there at Sharpe -0.05. It reports what the paper found rather than curating winners,~~ The July list kept entries that did not work (Short Term Reversal with Futures at Sharpe −0.05), reporting what the paper found rather than curating winners, the opposite habit from the self-reported numbers this corpus keeps running into (see synthesis, "Where's the evidence?").

The strategies have since moved off GitHub to paperswithbacktest.com; the README keeps a "previous list" and points at the site. Update 2026-09-27: the strategy table is back on GitHub, regenerated by a script from the replication catalogue. It shows the 61 strongest of 1,687 replications that clear t = 1.96 over at least 10 years, capped at 12 per asset class, Sharpe gross of costs. The July table, losers included, is gone. The new one is filtered to winners by construction, though it prints the t-statistic beside each Sharpe and says outright that half the catalogue does not clear it. The business model still shows through: each row links to paperswithbacktest.com, where the code lives.

Why it matters here

It names the incumbents. The spoke's engine pages are all young projects (banbot, tensortrade, trademaster); zipline, backtrader, Lean, vnpy, nautilus_trader and vectorbt are the field's established frameworks and none had a page. They're now page candidates with a reason attached.
It narrows the crypto skew the synthesis flagged: crypto is one subsection of one category here, next to equities, bonds, commodities, currencies and REITs.
It supplies the missing layers — risk, pricing, portfolio construction — that the "beyond crypto" open question named as unsourced.

Tier

T3 — a community link catalog, not analysis. Its value is coverage and page candidates, the same role awesome-db-tools plays in ../operational-databases-wiki. Carry two caveats with anything taken from it. ~~The repo has been untouched since January 2025, so the tool rankings and the "not maintained" notes are 18 months stale.~~ (Retracted 2026-09-27: maintained again since August 2026, with dormancy now flagged per entry.) And the Sharpe figures are the papers' own backtested results, inheriting whatever the papers assumed.

algorithmic-trading · vectorized-backtesting · backtesting · event-driven-trading · paperswithbacktest · strategy-optimization · synthesis