Asia Pacific Markets

Asia-Pacific markets (ex-mainland China)

Mainland A-shares are a separate skill (fin-china). Everything here is the rest of the region, where the libraries are thinner and the market-structure differences are larger.

Two verdicts up front: outside Korea, no free Asian source gives a survivorship-bias-free universe, and only J-Quants and TEJ offer genuine point-in-time fundamentals. ib_async is the practical cross-market execution answer.

1. Libraries — verified 2026-09-04

🔴 Dead but still widely recommended: nsepy (2020 — its last commit is literally titled "Deprecation notice"), investpy (2022), ib-insync (repo archived: true), asx (2022), mojito2 (2023), korea-investment-stock (self-deprecated on PyPI), and Angel One smartapi-python (stale ~19 months, and its PyPI homepage repo 404s).

🚨 Do not exist on PyPI (all verified 404): kabusapi, kabusapi-python, kabuslib, jquantsapi, TejToolAPI, SGXData. There is no maintained Python SDK for Kabu Station.

🚨 Name collisions: jpx on PyPI is a JMESPath engine, not Japan Exchange. sgx is an evolutionary-algorithm package, not Singapore Exchange. XBSE in exchange_calendars is Bucharest, not Bombay.

2. Two licence/packaging findings worth correcting

✅ futu-api is sdist-only — but the usual conclusion is wrong. All 84 files across every release since 2018 are sdists, yet the package is pure Python (no ext_modules, zero .c/.pyx/.so/.pyd in the archive), so it installs fine on Windows. The real traps are: it requires the FutuOpenD daemon running separately, python_requires is unset, and protobuf>=3.20.0 is unpinned.

🚨 shioaji has no licence at all. PyPI license and license_expression are both None, and the GitHub repo root has no LICENSE file and no source — it ships closed-source binary wheels. Windows win_amd64 is supported; there are no musllinux wheels, so it fails on Alpine.

3. 🚨 exchange_calendars — four verified defects

exchange_calendars 4.13.2 is genuinely well maintained (regular per-venue holiday commits). These are real defects anyway, all reproduced by executing it locally:

1. 🔴 Korean CSAT late opens stop in 2021. On the national university entrance exam day KRX opens one hour late. The released list contains 30 dates ending 2020-12-03, and even master ends at 2021-11-18. ✅ Executed: XKRX.session_open("2024-11-14") returns 09:00; the true open was 10:00. CSAT days for 2022–2026 are all wrong. Any intraday model keyed to "first bar of the day" is an hour off on those dates.
2. 🔴 XSES models no lunch break ever (has_break=False), though SGX had one until 2011 — the source admits it.
3. 🔴 The default date bounds are a moving target. ✅ GLOBAL_DEFAULT_START / _END are computed from pd.Timestamp.now() at import as today − 20 years / today + 1 year. The same code returns a different calendar tomorrow. → always pass explicit start= and end=.
4. 🔴 XBOM has 245 sessions in 2026 and ZERO in 2027 — Indian holidays are announced annually and hand-maintained. Same for XSES. By contrast XHKG, XTAI, XTKS, XKRX, XASX extend to today+1y because their holidays are rule-generated.

🚨 There is no XNSE. ✅ "XNSE" in get_calendar_names() → False. NSE handles the overwhelming majority of Indian volume and has no calendar — you must use XBOM (BSE) as a proxy. Holidays are near-identical in practice but not guaranteed, and special sessions like Muhurat trading (the ~1-hour Diwali session) are exchange-specific. Using XBOM for NSE is a reasonable approximation, silently made.

✅ What it gets right (also executed): Japan's close moving 15:00 → 15:30 on 2024-11-05; Hong Kong's open 10:00 → 09:30 on 2011-03-07; Korea's 2025-01-27 temporary holiday; Taiwan's LNY closure.

🔴 Do not trust it for: Korean CSAT opens 2022+; Singapore pre-2011 intraday; Indian dates after 2026; anything NSE-specific; Stock Connect trading days (no such calendar exists); HK typhoon suspensions (same-day events no static calendar can carry).

4. 🚨 Date-dependent regimes that silently break backtests

Key every rule to a date. These changed within most backtest windows:

5. Per-market notes

Hong Kong. 🚨 Lot sizes vary per stock — there is no universal round lot, and a backtest assuming 100 shares is wrong for most names. T+2 settlement, no daily price limits, but a VCM (Volatility Control Mechanism): ±5%–50% thresholds by tier with a 5-minute cooling-off, covering HSCI Large/Mid/SmallCap constituents plus SPACs and ETFs. Stock Connect adds quota and eligibility constraints, and no Stock Connect trading calendar exists in any library.

Taiwan. ±10% daily limit; 09:00–13:30 with no lunch break; a 14:00–14:30 after-hours fixed-price session; six tick-size tiers. FinMind (2,780★) is the most-starred library; shioaji is the broker path but is closed-source and unlicensed.

Japan. J-Quants is confirmed official JPX. ✅ Its free tier gives 2 years of data with a 12-week embargo — usable for research, not for anything near-live. jquants-api-client is the client. No maintained Kabu Station SDK exists. Lunch break 11:30–12:30.

Korea — the bright spot. 🔑 FinanceDataReader provides not just the delisted list but price history for delisted names (DataReader('KRX-DELISTING:036360')), plus administrative-issue lists. This is the best free survivorship-bias story in this entire domain. ⚠️ pykrx is no longer credential-free — its README (May 2026) now requires KRX_ID/KRX_PW, actual Korea Exchange member credentials, for authenticated datasets. This will break existing CI.

India. 🚨 STT (securities transaction tax) materially changes intraday economics — a cost model ported from US equities will overstate net returns. Circuit breakers are index-level and per-stock. kiteconnect (Zerodha) is the most usable broker API; nsepython is GPL-3.0; jugaad-data ships LICENSE.YOLO.md, which is effectively public domain but not OSI-approved — flag it if you have a licence allowlist. The official KIS repo (1,590★) declares no licence at all.

Singapore / Australia. Thin library coverage. yfinance for ASX, trading-ig for IG Markets, and Interactive Brokers as the practical answer for both data and execution.

6. Cross-cutting

Everything in ../../fin-core/skills/research-integrity-guards/SKILL.md applies, and the settlement, price-limit and lot-size rules above are the local equivalents of the A-share rules in ../../fin-china/skills/china-ashare-data/SKILL.md.