intangiblecapital

Load the Ewens, Peters and Wang (2024) knowledge and organization capital stocks for Compustat firms.

pip install intangiblecapital

import intangiblecapital as ic
stocks = ic.load()                    # checks the manifest; downloads each release file once
stocks = ic.load(release="20260924")  # pin a release for reproducibility
params = ic.parameters()              # depreciation rates and gamma by SIC
merged = ic.merge_compustat(funda)    # attach stocks to your Compustat frame on gvkey, fyear
ic.releases()                         # what releases exist

load() returns these columns in order:

Missing stocks are NaN, not zero. The historical release="090123" reads the Stata file and returns the same schema, with empty notes because that file has none. stocks.attrs records release, release_date, source, and citation.

Use ic.merge_compustat(funda) when your gvkey is an integer or a zero-padded string such as "001004". Fiscal years can be integers or whole-number floats such as 2019.0. The function preserves your input key columns and their dtypes, leaves the input unchanged, and returns a new frame with a reset index. Missing keys do not match; nonnumeric or fractional keys raise ValueError. For other input column names, use gvkey="firm_id", fyear="year". The default how="left" keeps all input rows; how="inner" keeps only matches. Other join types are not supported. Duplicate input keys are allowed; duplicate stock keys raise pandas.errors.MergeError. Rename or remove existing knowCapital, orgCapital, note, _ic_gvkey, or _ic_fyear columns before calling. For a direct pandas merge, normalize both frames' gvkey and fyear to Int64.

Every unpinned load() fetches the small release manifest, so a new release becomes available automatically. Each release file is cached at ~/.cache/intangiblecapital; set INTANGIBLECAPITAL_CACHE to choose another directory, or inspect ic.cache_dir(). load(release="20260924") uses the cached manifest and file, checking online if the tag is absent. A cached pin works offline; an unpinned load requires a reachable manifest. load(refresh=True) re-downloads the manifest and selected file. Pins assume the maintainer keeps each published release file unchanged.

releases() returns release dictionaries newest date first, including tag, file, date, and other manifest metadata. Use releases(refresh=False) for the cached list. parameters() returns the fixed 2023 parameter table, independent of the stock release: sic (Int64, without leading zeroes), knowDepr, organDepr, gamma (float64 rates/shares), and industry5 (string). Use parameters(refresh=True) to re-download it. Parameter files do not update automatically.

Unknown release tags and invalid schemas raise ValueError; download failures raise OSError with the URL and cache location. Retry with refresh=True if a cached file is damaged. Failed downloads preserve any previous cached file.

Requires Python 3.9+ and pandas 1.5+. To run the offline smoke/regression tests from a checkout, use IC_LOCAL=1 python3 python/tests_smoke.py at the repository root. Set IC_REPO_ROOT=/path/to/checkout to read data elsewhere. Add --build to also check wheel and source-distribution packaging using already-installed setuptools>=77 and wheel; no dependencies are downloaded. Build/publish tooling must target the python/ subdirectory, which contains pyproject.toml and its README.

Documentation, figures and the LLM prompt: https://intangiblesdata.org. Method and code: https://github.com/michaelewens/Intangible-capital-stocks.

Citation: Ewens, Michael, Ryan Peters and Sean Wang. "Measuring Intangible Capital with Market Prices." Management Science 71.1 (2024): 407-427. https://doi.org/10.1287/mnsc.2021.02058

License: the package code is MIT. The data it downloads are released under Creative Commons Attribution-NonCommercial 4.0 (see license.md in the repository); cite the paper when you use them.