Since fiscal years ending March 2023, Japan has required listed companies to include a standardized diversity table in their annual reports (Japan's equivalent of a US 10-K on SEC EDGAR).
Every report publishes three things in one place: female manager ratio, male paternity leave rate, and the gender pay gap (women's average annual pay as a percentage of men's — broken down by all workers, regular full-time staff, and non-regular staff).
Pull ten well-known Japanese companies from those filings and chart them, and the differences — and a few surprising data quirks — jump right off the screen:
This post shows how to pull the same numbers and build the chart in about 30 lines of Python. You don't need to know anything about Japanese stocks — this is about public company disclosures, not prices or trading.
Disclosure: I'm Sohei, and I build ZENITH Code, the API used here. It turns Japanese company filings into JSON. This series is me showing what you can build with that data. The free plan is enough for everything in this post.
What jumps out
Here are the latest parent-company figures from each company's annual report (pay gap = women's average annual pay as a % of men's; 100% means parity):
| Company | Code | Fiscal period | Female managers | Male paternity leave | Pay gap (All) | Pay gap (Regular) | Subsidiaries in JSON |
|---|---|---|---|---|---|---|---|
| Shiseido | 4911 | 2025-12 | 38.9% | 91.1% | 81.1% | 92.4% | 6 |
| Rakuten Group | 4755 | 2025-12 | 33.5% | 73.3% | 78.0% | 81.6% | 13 |
| Mercari | 4385 | 2026-06 | 25.2% | 86.4% | 69.1% | 71.4% | 0 |
| Fast Retailing (UNIQLO) | 9983 | 2025-08 | 21.6% | 75.0% | 60.6% | 65.1% | 5 |
| Sony Group | 6758 | 2026-03 | 20.5% | 84.0% | 82.8% | 83.4% | 9 |
| Mitsubishi Corp | 8058 | 2026-03 | 13.1% | 171.7% | 64.7% | 65.2% | 42 |
| Capcom | 9697 | 2026-03 | 11.8% | 79.7% | 62.6% | 81.3% | 0 |
| Nintendo | 7974 | 2026-03 | 5.8% | 82.0% | 72.7% | 73.6% | 2 |
| Toyota | 7203 | 2026-03 | 3.9% | 79.0% | 67.0% | 66.8% | 9 |
| Honda | 7267 | 2026-03 | 3.6% | 116.8% | 72.8% | 74.3% | 1 |
- A 10× spread in female managers. Cosmetics giant Shiseido sits at 38.9%, Rakuten at 33.5%, Mercari at 25.2%, Fast Retailing (UNIQLO) at 21.6%, and Sony Group at 20.5%. At the other end, Japan's manufacturing icons are still in single digits: Nintendo 5.8%, Toyota 3.9%, and Honda 3.6%.
- Data quirk #1: All workers vs. regular full-time staff. Looking only at
wageGapAllcan mislead you. At Capcom, the all-worker figure is 62.6%, but among regular full-time employees (wageGapRegular) it jumps +18.7 points to 81.3%. At Shiseido, it jumps from 81.1% to 92.4% (+11.3 points). The difference comes from workforce mix (a higher share of women in non-regular or part-time roles). Because the API separateswageGapAll,wageGapRegular, andwageGapNonRegular, you can spot that immediately.
- Data quirk #2: Why Honda's paternity leave is 116.8% (and Mitsubishi Corp's is 171.7%). At first glance this looks like a parser bug. It isn't — it's the statutory formula. Under Japanese rules, the rate is (men who took paternity leave in fiscal year N) ÷ (men whose spouse gave birth in fiscal year N). When employees whose children were born late in year N-1 take leave in year N, the numerator can easily exceed the denominator.
- Bonus: Subsidiaries are in the same response. Nintendo's JSON doesn't stop at the parent company — its
subsidiariesarray includes Mario Club Co., Ltd. and MONOLITH SOFTWARE INC. (the studio behind Xenoblade Chronicles), both at 100% male paternity leave. Mitsubishi Corp returns 42 subsidiaries in one call.
Getting the data, step by step
1. Get an API key
Sign up for ZENITH Code (free plan, no credit card) and create an API key in the dashboard:
export ZENITH_API_KEY=your-key-here 2. Find a company's 4-character code
curl -H "x-api-key: $ZENITH_API_KEY" \ "https://api.zenithcode.net/stable/search-name?query=Nintendo" 3. Ask for the diversity metrics
curl -H "x-api-key: $ZENITH_API_KEY" \ "https://api.zenithcode.net/stable/diversity?symbol=7974&limit=1" Here is the real latest row for Nintendo (7974), with null fields and a few metadata fields trimmed for readability:
{ "symbol": "7974", "fiscalPeriod": "2026-03", "period": "FY", "docId": "S100Y9NX", "submittedAt": "2026-06-25T06:30:00Z", "sourceUrl": "https://disclosure2.edinet-fsa.go.jp/WZEK0040.aspx?S100Y9NX", "reportingCompany": { "femaleManagerRatio": 0.058, "maleChildcareLeaveRatio": 0.82, "maleChildcareLeaveBasis": "childcare-leave", "wageGapAll": 0.727, "wageGapRegular": 0.736, "wageGapNonRegular": 0.842 }, "subsidiaries": [ { "name": "マリオクラブ㈱", "nameEn": "Mario Club Co., Ltd.", "maleChildcareLeaveRatio": 1, "wageGapAll": 0.759, "wageGapRegular": 0.759 }, { "name": "㈱モノリスソフト", "nameEn": "MONOLITH SOFTWARE INC.", "maleChildcareLeaveRatio": 1, "wageGapAll": 0.748, "wageGapRegular": 0.723, "wageGapNonRegular": 0.866 } ] } Ratios are fractions (0.727 = 72.7%), and sourceUrl links directly to the EDINET filing.
4. What it costs
/stable/diversity costs 1 token per request. Pulling all 10 companies plus a name search is 11 tokens out of the 5,000 tokens/month free plan.
The code
Requires Python 3.9+ and matplotlib (pip install matplotlib). Everything else is standard library.
import json, os, urllib.error, urllib.request import matplotlib.pyplot as plt API = "https://api.zenithcode.net/stable/diversity" COMPANIES = [ ("4911", "Shiseido"), ("4755", "Rakuten Group"), ("4385", "Mercari"), ("9983", "Fast Retailing"), ("6758", "Sony Group"), ("8058", "Mitsubishi Corp"), ("9697", "Capcom"), ("7974", "Nintendo"), ("7203", "Toyota"), ("7267", "Honda"), ("6861", "Keyence"), # femaleManagerRatio is null — skipped safely ] def latest_diversity(symbol): url = f"{API}?symbol={symbol}&limit=1" req = urllib.request.Request(url, headers={"x-api-key": os.environ["ZENITH_API_KEY"]}) try: with urllib.request.urlopen(req) as res: rows = json.load(res) return rows[0] if rows else None except urllib.error.HTTPError as e: if e.code == 404: return None raise points = [] for sym, label in COMPANIES: row = latest_diversity(sym) rc = (row or {}).get("reportingCompany") or {} mgr, gap = rc.get("femaleManagerRatio"), rc.get("wageGapAll") if mgr is None or gap is None: print(f"Skip {sym} ({label}): missing femaleManagerRatio or wageGapAll") continue points.append((label, mgr * 100, gap * 100)) fig, ax = plt.subplots(figsize=(8, 5)) for label, mgr, gap in points: ax.scatter(mgr, gap, s=80, color="#2563eb") ax.annotate(label, (mgr, gap), xytext=(6, 4), textcoords="offset points", fontsize=9) ax.set_xlabel("Female manager ratio (%)") ax.set_ylabel("Women's pay as % of men's (all workers)") ax.set_title("Japanese listed companies: female managers vs. gender pay gap") ax.grid(True, linestyle="--", alpha=0.4) fig.tight_layout() fig.savefig("diversity.png", dpi=150) print(f"Saved diversity.png ({len(points)} companies)") Three things to watch out for
1. Individual fields can be null (and unknown codes return 404).
For example, industrial sensor giant Keyence (6861 — from part 2) reports wageGapAll: 0.432 (43.2%) and maleChildcareLeaveRatio: 0.847, but femaleManagerRatio is null because its annual report doesn't include that item at all (the parent-company table covers only paternity leave and the pay gap). Always check for None instead of assuming every field is populated.
2. Don't cap paternity leave at 100% blindly.
As shown above, values like 1.168 (116.8%) and 1.717 (171.7%) are valid statutory filings, not data errors. Also check maleChildcareLeaveBasis when comparing across companies, since firms can report either statutory childcare leave or a broader leave definition.
3. reportingCompany is the parent entity only.
Just like the salary figures in part 2, reportingCompany covers the filing parent company. For holding companies (Sony Group, Fast Retailing, Mitsubishi Corp), check the subsidiaries array to see what the operating units actually look like.
Why I built this
Japanese company filings are public, but they are hard to use, even inside Japan. The reports are in Japanese, the XBRL follows a Japan-specific taxonomy, line items are tagged differently across companies and accounting standards, and quarterly figures are cumulative. I spent about two years parsing all of that until it could be served reliably as an API.
The other reason is price. Japanese financial and stock market data has mostly been sold through terminals and data feeds priced for banks and funds. I wanted developers, students, and small teams to be able to just try it.
So ZENITH Code has:
- About 4,000 listed Japanese companies: annual reports filed on EDINET (Japan's EDGAR) and quarterly earnings releases, parsed into JSON.
- End-of-day stock prices for Japanese listed companies on the same API key, provided under formal agreements with the exchanges.
- One set of field names across Japanese GAAP, IFRS, and US GAAP.
- Direct filing links (
sourceUrl/provenance) on every row so you can verify any number against the original report. - English company names and docs, plus an MCP server so Claude or ChatGPT can query these filings directly (its own post later in this series).
The free plan (5,000 tokens a month, no card) covers almost every endpoint. Try it, and if something in the data looks wrong or confusing, tell me in the comments — I read all of them.
Try it with no account
If you want to explore before signing up, the examples repository has a demo mode with fictional companies:
git clone https://github.com/Crab-grass/zenith-code-examples.git cd zenith-code-examples python3 python/zenith_examples.py statements --symbol DEMOA --demo Next in this series: Japanese companies own each other — drawing cross-shareholdings as a network graph.
Data provided by ZENITH Code (sources: TDnet, EDINET and company disclosures; processed by ZENITH Code). ZENITH Code does not warrant the accuracy, completeness or timeliness of the data, and nothing in this post is investment advice.


