EDINET Point-in-Time Japan Equities Dataset

Survivorship-bias-free, Point-in-Time Japanese annual securities filings (Parquet & DuckDB) with exact filing timestamps (submit_datetime) and year-over-year risk disclosure drift scores (risk_change_score).

Request Full Dataset / Early Access GitHub (Sample & Pipeline) Hugging Face Sample

Key Design Properties

Strict Point-in-Time (No Look-Ahead Bias)

Amendment filings (doc_type_code = 130) never overwrite earlier filing history. Filtering by submit_datetime <= as_of returns only the figures actually known to the market on that date.

Survivorship-Bias-Free & Text Drift

Retains delisted issuers from official EDINET submissions and precomputes year-over-year cosine distance on the “Business Risks” section using intfloat/multilingual-e5-large.

148-Company Cohort Validation (2025 vs 2026 Filings)

Measured across 148 Japanese listed companies comparing 2025 and 2026 annual securities reports (有価証券報告書):

  • Median (50th percentile): 0.001543 (bottom 25% at 0.000253 or lower, indicating largely unchanged risk text)
  • Maximum observed drift: 0.023372 (roughly 15× the median)
  • Top separation examples: Issuers with major operational or risk-disclosure revisions stood out clearly, such as ANA Holdings (0.023372) and Daito Trust Construction (0.018248).

Schema (pit_metrics Table)

Column Type Description
doc_idStringEDINET document ID (e.g., S100YDF3)
edinet_codeStringEDINET issuer code (e.g., E01441)
sec_codeString5-digit securities code (e.g., 59730)
filer_nameStringIssuer legal name
doc_type_codeStringFiling type (120: Annual Report, 130: Amendment)
is_amendmentBooleanAmendment filing flag
period_endDateFiscal period end (YYYY-MM-DD)
submit_datetimeDatetimeSubmission timestamp (Point-in-Time cutoff key)
metric_nameStringNormalized metric (net_sales, operating_income, ordinary_income, net_income, net_assets, total_assets, operating_cf)
metric_valueFloat64Financial value in JPY
risk_change_scoreFloat64Year-over-year cosine distance of Business Risks text

Quick Start (Free Sample: 10 Companies, 2 Years)

Download sample/edinet_pit_sample.parquet or sample/edinet_pit_sample.duckdb from GitHub or Hugging Face and query Point-in-Time snapshots locally:

SELECT sec_code, filer_name, period_end, metric_name, metric_value, risk_change_score
FROM read_parquet('sample/edinet_pit_sample.parquet')
WHERE submit_datetime <= TIMESTAMP '2026-06-26 00:00:00'
QUALIFY ROW_NUMBER() OVER (
    PARTITION BY edinet_code, period_end, metric_name
    ORDER BY submit_datetime DESC
) = 1;

日本語概要 / Full Archive Early Access

全上場企業(上場廃止含む)・過去 5〜10 年分のフルデータセット(初期買い切り ¥20,000 想定 / 月次更新サブスク ¥3,500 想定)の先行リリース案内および追加指標リクエストを受け付けています。必要な期間や銘柄範囲・希望フォーマットをお知らせください。