Replication package for Low-Dimensional Asset-Pricing Networks: Identification, Stability, and Economic Use
收藏资源简介：
Version 1.1.0 replication package for Low-Dimensional Asset-Pricing Networks: Identification, Stability, and Economic Use, by Jinpeng Wang and Yan Jiang. Code, frozen configurations, synthetic results, aggregate evidence, figure/table inputs and manuscript sources follow the article section order. This revision adds two bounded post-study exploratory controls: architecture-matched random-network geometry and a same-panel six-feature ablation with 60 trained networks. It distinguishes absolute rank from rank relative to layer width and isolates feature masking from historical pipeline differences. The separate retrospective temporal exercise is reported in Supplement D. The earlier failed pricing-function criterion, archived results and chronology qualifications are retained. Synthetic geometry calibration is integrated into this package. Public exhibit replay regenerates the paper without WRDS access. A separate aggregate-only adapter recalculates the new feature-control statistics using released managed-factor returns plus public benchmark returns obtained by the replicator. Its numerical results were verified against the original evaluator. Full retraining still requires lawfully acquired source data. Licensed stock records, security-level panels/predictions, checkpoints, credentials and institutional connection details are excluded. Original and released SHA-256 values are recorded. Legacy Core-92 development overlaps January 2020; Core-86 bridges benchmark and self-built characteristics; the temporal exercise is selected-sample and retrospective; the theoretical result concerns finite known loading spans. Later controls use familiar historical data. No untouched research-wide holdout, journal acceptance or cross-platform bitwise retraining claim is made. Version 1.0.1 remains preserved under DOI 10.5281/zenodo.23064062 and its GitHub tag. Original aggregate data/documentation: CC BY 4.0. Source code: MIT. Third-party inputs and dependencies retain their terms. GitHub: https://github.com/extradimen/asset-pricing-dimension-replication. Correspondence: jiangyan@gcu.edu.cn. Download and reconstruction: this Zenodo deposit stores the release ZIP as 18 numbered parts (.part001 through .part018). Download all parts plus archive_parts.json and join_archive.py into one folder, then run python3 join_archive.py. The script verifies every part and the reconstructed ZIP against SHA-256 checksums. README_ZENODO.txt provides instructions. The reconstructed archive is asset-pricing-dimension-replication-1.1.0.zip (18,314,848 bytes; SHA-256 fe52ba58556bc9c4976b3ad2efb9fbec00c1dfc46b00675c717cc46f31b6e260).
