Replications

This track shows end-to-end examples of running a research-grade macroforecast study from raw FRED data through bit-exact replicable manifest. Three layers:

1. Bundled example walkthroughs – detailed walk through one of examples/recipes/*.yaml, explaining every layer choice and the resulting artifact tree.

2. Bundled paper baselines – a published paper’s ridge / model horse-race ported to macroforecast schema. Recipe ships in the repo, runs end-to-end, replicates bit-exact. The user swaps the bundled smoke panel for real FRED data when reproducing the paper’s figures.

3. Research replications – four studies the project maintainer conducted with macroforecast. Each replication includes the exact recipe YAML, command line, expected artifacts, and pointers back to the published paper / preprint.

Use these as a template when you want to write your own publication-grade replication study.

Walkthroughs

• Example walkthrough – minimal ridge

Replication tier codes

Each replication page carries a tier code:

Tier	Meaning
SMOKE	Recipe runs end-to-end on stock install; bit-exact replicate passes on bundled smoke panel
SCHEMA-FAITHFUL	Recipe axes match the paper’s experimental design at the YAML level
NUMERICAL	Numerical results (MSEs, rankings) verified against paper-reported figures on full FRED data

A page may hold multiple tiers (e.g., SMOKE + SCHEMA-FAITHFUL).

Bundled paper baselines

• Goulet-Coulombe (2021) – The Macroeconomy as a Random Forest — SMOKE + SCHEMA-FAITHFUL

Research replications

Coming in v0.9.1 (additional 4+ replication walkthroughs covering published macro-forecasting papers).