ML Economic Forecasting Replication Package v1.0.0

ML Economic Forecasting Replication Package v1.0.0

This release archives the replication materials for:

Beyond Average Accuracy: Statistical Distinguishability and Temporal Concentration in Data-Rich Macroeconomic Forecasting

Included

• forecasting and statistical-validation code;
• frozen human-readable outputs for four paper-core FULL runs;
• paper table and figure-data exports;
• sanitized validation records;
• the R 4.6.0 dependency environment in renv.lock;
• public file-policy, frozen-output, and SHA-256 inventory validators; and
• the working-paper PDF.

Validation

• internal research release: 139 checks, zero failures, PASS;
• clean-clone renv restoration and project function loading: PASS;
• independent Python validation of frozen outputs and paper exports: PASS;
• complete four-target FULL rerun from a clean clone: completed;
• frozen FRED-MD checksum verification: PASS;
• SSRN research-protocol validation after the rerun: PASS; and
• GitHub Actions release checks: PASS.

Data

The source FRED-MD CSV is not redistributed. The dated June 2026 vintage is identified by MD5:

8591dd9f169f7aeb45b7c91782fbd947

See DATA_AVAILABILITY.md for acquisition and licensing guidance.

Dashboard

https://slackquant.shinyapps.io/ml_econ_forecasting/

Reproducibility scope

The public package excludes private development RDS objects, logs, raw source data, and Shiny deployment source. Exact bitwise identity across operating systems is not guaranteed for stochastic learners and bootstrap procedures; the package instead combines frozen-output validation, a locked software environment, and a successful clean-clone end-to-end rerun.

Archival note

The published v1.0.0 release and its associated version DOI contain the then-current manuscript file, paper/Beyond_Average_Accuracy_Preprint_v0.10.5.pdf.

The SSRN-submitted final manuscript, paper/Beyond_Average_Accuracy_Preprint_v1.0.pdf, is available on the repository’s default branch and through the SSRN record.

No forecasting code, frozen research outputs, or paper-export tables changed between the tagged release and the final-manuscript update.