ffn: a financial toolbox whose own example zeroes two of its five tickers

yfinance is a runtime dependency, not an optional extra

The runtime dependency list runs to eight entries: decorator, matplotlib, numpy, pandas, scikit-learn, scipy, tabulate and yfinance. The last of those is a market data downloader, so installing the analysis library reaches for a network client whether or not the caller ever asks for a price. The README's account of the stack, that ffn stands on the shoulders of giants such as Pandas, Numpy and Scipy, does not mention scikit-learn or matplotlib, although both are declared. The lower bounds are uneven as well. `decorator>=5.0.7`, `pandas>=0.19`, `scikit-learn>=0.15`, `scipy>=0.15` and `tabulate>=0.7.5` all carry a floor, while numpy is written as bare `numpy` and yfinance as `yfinance>=0.2`, the loosest bound in the set.

requires-python opens at 3.9 while the classifiers stop at 3.13

The metadata sets the interpreter floor at `>=3.9` and then enumerates individual versions as classifiers, 3.9, 3.10, 3.11, 3.12 and 3.13, with nothing listed for 3.14. The same block labels the project `Development Status :: 4 - Beta`, which is a packaging classification rather than a statement about the release history: the version in the file is 1.3.0 and three tags are published. Keywords sit at that level too, running from python and finance through quant, quant finance, algotrading and algorithmic trading, so the intended audience is spelled out in the distribution rather than in prose. Authorship is a single entry, one named maintainer with an email address in the author table.

v1.2.0 and v1.2.1 were published fifteen seconds apart

The three recent releases are v1.2.0 published on 2026-09-07 at 18:17:35, v1.2.1 published the same day at 18:17:50, and v1.3.0 published on 2026-10-03. Two tags fifteen seconds apart read like a packaging correction applied on the spot rather than two planned releases. The version number itself is held in two places in pyproject.toml: the project table's `version` field and the bumpversion table's `current_version`, both reading 1.3.0, with the bumpversion section ending partway through its next key. The tool that keeps them aligned is `bump-my-version` in the develop extra, paired with a `show-version` target in the Makefile. The default branch is `master` rather than `main`.

ty is installed as a checker, and no make target invokes it

The develop extra installs `ty` among two dozen build tools, and the Makefile gives it a target of its own:

ty check ffn

That target's own comment calls the check advisory, and nothing calls the target. `lint` is defined as `lint-py lint-docs`, `checks` is `check-dist` alone, and `test` is `python -m pytest tests`, so type checking sits outside every aggregate target while ruff and codespell run on each `make lint`. The docs linter covers `README.md docs/development.md docs/source/*.md` and nothing else, which leaves any other markdown in the tree unchecked and unspell checked. The benchmark suite has the same shape, a separate target running pytest with `--benchmark-only` against a `benchmarks/` directory.

The headline example returns zero weight for two of five tickers

The entire demonstration is three lines of work:

import ffn
returns = ffn.get('aapl,msft,c,gs,ge', start='2010-01-01').to_returns().dropna()
print(returns.calc_mean_var_weights().as_format('.2%'))

The block underneath it is static text headed as example output, and two of the five positions come back at `-0.00%`: `c` and `gs`. The remaining three carry aapl at 62.54%, ge at 36.19% and msft at 1.26%. The symbols are Yahoo style tickers, which is the reason the downloader appears among the runtime dependencies rather than behind an extra. The printed frame ends with `dtype: object`, so what the reader sees is a formatted frame of strings rather than numbers, and the weights were produced by a single call, `calc_mean_var_weights`, that the README does not describe further.

uv drives the Makefile while the install notes point at Anaconda

End users are told to run one command and nothing else:

pip install ffn

They are then told that because ffn has many dependencies, the Anaconda Scientific Python Distribution is strongly recommended, on the grounds that it ships many of the required packages and pip itself. Contributors go another way entirely. The Makefile drives `uv pip install -e '.[develop]'` to set up development, `uv pip install -r pyproject.toml --extra develop` for prerequisites, `uv pip install .` to install, and `python -m build -n` to build. Nothing in the repository root records resolved versions: no lock file, no requirements file, and no uv configuration file sit beside pyproject.toml.

bt takes the backtesting half, and the tree is template generated

The README draws its own boundary in the first paragraph: readers who want a full backtesting framework are pointed at bt, which is built atop ffn. That is the division of labour in a sentence, measurement, weighting and data transformation here, strategy testing in the other package. The repository root is short, thirteen entries including `benchmarks/`, `docs/`, `ffn/`, `tests/` and a `.copier-answers.yaml`. That last file, together with `copier`, `yardang` and `klink` in the develop extra and a contribution note about Copier template updates in the development guide, means part of the tree is generated rather than written by hand, so an edit to a generated file is an edit that can be overwritten.

Editorial conclusion

ffn is a bag of analysis functions with a thin layer of its own around them, and its packaging metadata is more informative than its promotional copy: an unpinned numpy, a market data downloader in the runtime dependencies, a Beta classifier on a 1.3.0 version, and a type checker that no make target calls. Reach for it when the work is performance measurement, portfolio weighting or data reshaping and you would rather keep working in Pandas than adopt an ORM. Look elsewhere when you need a strategy backtester, since the project hands that job to bt in its first paragraph. Before installing, read the dependency list and decide whether yfinance belongs in your environment, and check whether the files you plan to edit were generated from the Copier template.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. pmorissette/ffn on GitHub
4. README
5. Releases