BLS Release Dates — Documentation
Technical references for BLS data programs and their release methodologies, designed to support Bayesian NFP nowcasting.
Usage
1. Download releases and build release_dates
Writes data/releases/{ces,sae,qcew}/*.htm and data/release_dates.parquet (columns: publication, ref_date, vintage_date).
2. Build vintage_dates (revisions)
From release_dates.parquet, build a dataset with revision codes:
Writes data/vintage_dates.parquet with columns: publication, ref_date, vintage_date, revision, benchmark_revision.
- CES: revisions 0, 1, 2, and 9 (benchmark for March ref_dates only).
- SAE: revisions 0, 1, and 9 (benchmark twice for April–September: first at March Y+1, second at March Y+2).
- QCEW: by quarter of ref_date — Q1: 0,1,2,3,4; Q2: 0,1,2,3; Q3: 0,1,2; Q4: 0,1.
benchmark_revision: 0 = not a benchmark row; 1 = first benchmark; 2 = second benchmark (SAE re-replacement only).
(For recent ref_dates, not all revisions may exist yet.)
3. Read the parquet data
If the parquet files have been created, read them as polars DataFrames:
from bls_release_dates import read_release_dates, read_vintage_dates
# Default paths: data/release_dates.parquet, data/vintage_dates.parquet
releases = read_release_dates() # DataFrame or None if not created yet
vintages = read_vintage_dates() # DataFrame or None if not created yet
if releases is not None:
print(releases.head())
if vintages is not None:
print(vintages.head())
# Custom path
releases = read_release_dates("path/to/release_dates.parquet")
Using in other projects
Install without publishing to PyPI:
- Path (sibling repo or same machine): From the other project root, run
uv add ../bls-release-dates. Or in that project'spyproject.toml:"bls-release-dates @ file:///path/to/bls-release-dates". - Git:
"bls-release-dates @ git+https://github.com/lowmason/bls-release-dates.git@v0.1.0"(pin to a tag for reproducible installs). - Wheel: In this repo run
uv build, then in the other project:uv add /path/to/bls-release-dates/dist/bls_release_dates-0.1.0-py3-none-any.whl.