nfer-macrocast 0.1.0

Decomposing ML Forecast Gains in Macroeconomic Forecasting

macrocast

Decomposing ML Forecast Gains in Macroeconomic Forecasting.

An open-source Python (+ R) framework for systematic evaluation of machine learning methods in macroeconomic forecasting, with built-in support for the FRED-MD, FRED-QD, and FRED-SD database ecosystem.

---

Release Status

Surface	State
Latest PyPI release	Pending first nfer-macrocast publish
Current repository version	0.1.0
PyPI-safe scope today	Data layer and package metadata cleanup
Repo-head, not yet published	Forecasting pipeline, evaluation layer, macrocastR companion work

The repository is ahead of the current PyPI release. Install from source if you need unreleased pipeline or evaluation features. The PyPI distribution name is nfer-macrocast. The Python import package remains macrocast.

---

Installation

pip install nfer-macrocast
# or with all extras
pip install nfer-macrocast[all]

---

Quick Start

import macrocast as mc

# Load and transform FRED-MD (latest vintage, cached locally)
md = mc.load_fred_md()
md_t = md.transform()

print(md_t)
# MacroFrame(dataset='FRED-MD', vintage='current', T=790, N=128,
#            period=1959-01-01 to 2024-10-01, status=transformed)

# Subset by variable group
output = md_t.group("output_income")   # INDPRO, RPI, ...
prices = md_t.group("prices")          # CPI, PPI, ...

# Check missing values
report = md.missing_report()
print(report[["n_leading", "n_trailing", "n_intermittent"]].head())

# Method chaining
md_ready = (
    mc.load_fred_md()
    .trim(start="1970-01", end="2023-12")
    .handle_missing("trim_start")
    .transform()
)

# Load a specific vintage
md_2020 = mc.load_fred_md(vintage="2020-01")

# FRED-QD (quarterly)
qd = mc.load_fred_qd()

# FRED-SD (state-level)
sd = mc.load_fred_sd(states=["CA", "TX"], variables=["UR"])

---

Documentation

Full documentation is available at nanyeonk.github.io/macrocast.

R Package Status

macrocastR remains source-only and is not CRAN-ready yet.

• man/ pages and a full roxygen regeneration/review step are still pending.
• The R CI workflow currently runs R CMD check in a non-failing mode because documentation artifacts are incomplete.
• Registry publishing and credentialed release work should happen after those packaging gaps are closed.

---

License

MIT