Replication package for "Where Private Equity Risk Lives: Portfolio-Mediated Incidents and the Unit-of-Observation Problem at Non-Listed Buyout Managers"
Abstract
Replication package for the paper "Where Private Equity Risk Lives: Portfolio-Mediated Incidents and the Unit-of-Observation Problem at Non-Listed Buyout Managers". The package contains everything needed to audit and re-derive the paper's results: the pre-registration with its complete amendment log, all analysis code, every aggregated table behind every number in the manuscript, the figures, and a script that mechanically checks each number in the text against the result files (109 checks, all passing). Scope: 68 months of Korean news-based ESG incident data (January 2021 to July 2026) for one buyout manager and four domestic peers measured under identical rules, comprising 229,043 classified articles. Incident attribution is rule-based against a published asset master. The relevance classification was validated two ways: a full independent reclassification by a second model (inter-classifier kappa 0.703 on the binary retain/drop decision) and a 400-item human double-coding (inter-coder kappa 0.795; against the human consensus the production classifier reaches kappa 0.611 and the second classifier 0.827). The human-coding sample is released as identifiers, codes and reliability tables, without article text. Article full texts and crawled article bodies are not released. The news corpus is licensed from the Korea Press Foundation database and redistributing article text would breach its terms of use and publishers' copyright. The exact collection settings are released instead, so that a researcher with database access can retrieve equivalent source material. All aggregated outputs the paper depends on are included, so every figure and table can be re-derived and every number verified without the corpus. This paper has a companion research report, "When a Buyout Fails in Public", deposited separately. The two share a public-authority action database, an asset master and a data-loading module; those shared components are duplicated into each package under shared/ so that each stands alone, and the copies are byte-identical across the two deposits (SHA-256 digests in MANIFEST.md). Code within this deposit is licensed MIT; see LICENSE.md.
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 12 September 2026
- Language
- EN
- Document type
- Dataset