遇见数据集

JKP-Style Factor Returns for China A-Shares

收藏
Zenodo2026-11-04 更新2026-11-05 收录
官方服务:

资源简介:

Monthly long-short returns of 145 of the 153 JKP factors (Jensen, Kelly and Pedersen, 2023, "Is There a Replication Crisis in Finance?", Journal of Finance 78(5)) and their 13 themes, constructed for China A-shares (Shanghai and Shenzhen main boards, ChiNext and STAR; Beijing excluded). Sample: holding months 2000-02 to 2026-09. The returns are the author's own calculations following the JKP methodology. They are NOT official JKP data and are not endorsed by the JKP authors. Only aggregated monthly portfolio returns are released; no stock-level characteristics or returns, and no official JKP factor returns. Factor names, references, directions and theme assignments are taken from the JKP code repository (github.com/bkelly-lab/jkp-data, CC BY-NC 4.0) with attribution; see LICENSE_DATA.txt. Construction (JKP-style, "P1"): characteristics are built point-in-time from daily market data and multi-version financial statements (statements visible from the first trading day after announcement). Each month-end stocks are sorted into terciles with breakpoints from non-micro stocks (main-board 20th/80th size percentiles replace NYSE percentiles). Long-short = top minus bottom tercile times the JKP direction. Three weightings: equal, value, and capped value (cap at the main-board 80th size percentile; main version). Excess returns over a monthly risk-free rate, winsorized at 0.1%/99.9% each month; missing holding-month returns count as zero; ST/PT delistings receive -30%. Gross returns, no trading costs. Theme returns are equal-weighted averages of member factors. Validation on 5,556 stocks / 877,194 stock-months: correlation with the official JKP China factors (vw_cap) has median 0.842 across 145 factors; point-in-time invariance holds bit for bit at three test dates; 195/195 hand-calculated values match; portfolio identities hold within 1e-12. Caveats: point-in-time financial statements start in 2004, so accounting factors have thin early samples (the column n_stocks_min is provided for filtering). Eight factors cannot be constructed (missing share-repurchase cash, gross PP&E, or point-in-time employee history). Raw market and financial data come from a commercial data vendor (compiled from exchange data and public company filings) and are not redistributed. Files: jkp_ashare_factors_monthly_v1.csv (145 factors), jkp_ashare_themes_monthly_v1.csv (13 themes), factor_details_v1.csv (all 153 factors with status and coverage), README_DATA_EN.md and README_DATA_CN.md (documentation in English and Chinese), LICENSE_DATA.txt, MANIFEST_V1.json (hashes and checks). License: CC BY-NC 4.0 (non-commercial use, including academic research and publications, with attribution). For commercial use, please contact the author.Please also cite Jensen, Kelly and Pedersen (2023).

提供机构:
Zenodo
创建时间:
2026-11-04
二维码
社区交流群
二维码
科研交流群
商业服务