Lib Yfinance

yfinance

The default free source, the best-maintained one, and the one whose defaults have changed most. Roughly 40% of its changelog is "fix Yahoo changed X" — treat schema drift as normal operation.

The trap that costs you money

🚨 auto_adjust=True is now the hard default in yf.download(), and there is no Adj Close column. OHLC come back already adjusted; df['Adj Close'] raises KeyError on ≥1.0. Code written before 2025 does not error — it returns different numbers.

Ticker.history(auto_adjust=True) has defaulted True since 0.1.26 — only download() differed. Always pass auto_adjust explicitly so the number does not depend on the installed version.

Signature traps in download()

• end is EXCLUSIVE. Verbatim: "for end='2023-01-01', the last data point will be on '2022-12-31'".
• multi_level_index=True since 0.2.48 → even one ticker returns MultiIndex columns. Pass False for a flat frame.
• group_by='column' (default) → field-major ('Close','AAPL'); 'ticker' → ticker-major.
• progress=True writes to stdout — set False in pipelines.
• start defaults to "99 years ago"; period's default is the literal sentinel string '1mo if start & end None'.
• repair=True needs pip install yfinance[repair] (scipy + scikit-learn); a plain install lacks them. It fixes GBp/ZAc/ILA 100× sub-unit mixups, phantom dividends, split repair, and capital-gains double counting (1.1.0).

🚨 Intraday timezones moved in 1.4.0

ignore_tz docstring, verbatim: "Default depends on interval. Intraday = False. Day+ = True … if False (the intraday default), the index is converted to the most common exchange timezone among the requested tickers (before 1.4.0, this case always returned UTC instead)." Intraday download() returned UTC before 1.4.0 (2026-05-23) and exchange-local since. A stored intraday dataset spanning that boundary has shifted bars. Set ignore_tz explicitly.

🚨 Survivorship, and the data licence

Yahoo drops delisted symbols. yfinance's own YFTickerMissingError carries a "possibly delisted; " prefix — the library guesses at delisting because Yahoo will not say. Do not build a backtest universe from it. README, verbatim: "yfinance is not affiliated, endorsed, or vetted by Yahoo, Inc." … "the Yahoo! finance API is intended for personal use only." Apache-2.0 covers the code only — redistributing or commercializing the data is a separate legal question.

Rate limits and the curl_cffi pin

yfinance.exceptions.YFRateLimitError — "Too Many Requests. Rate limited. Try after a while." There is no published Yahoo rate limit; it is undocumented, IP-based and moves. Any "N requests/hour is safe" number is folklore. yf.config.network.retries defaults to 0.

🚨 curl_cffi remains a hard dependency at 1.7.0 despite 1.4.0's "make it optional" — optional at runtime, still installed. yfinance <1.5.2 breaks against curl_cffi ≥0.16 (fixed in 1.5.2), and curl_cffi is at 0.16.3. stockdex pins curl_cffi==0.12.0 and therefore cannot co-install.

First answer to "yfinance keeps 429-ing me": yfinance-cache (0.9.3, 2026-08-26, MIT) — calendar- aware caching that only re-fetches genuinely new bars. Structural alternative: defeatbeta-api, a HuggingFace parquet snapshot queried with DuckDB — no scraping, no rate limit, ~weekly freshness.

Other API notes

Ticker(ticker, session=None) — the old proxy= constructor kwarg is gone; proxies go through yf.config.network.proxy. New in 1.x: yf.Search, yf.Lookup, yf.Market, yf.Calendars, yf.Auth (1.4.0), yf.WebSocket/AsyncWebSocket, yf.screen/EquityQuery, yf.Sector/Industry, Ticker.valuation, ttm_income_stmt, and Ticker.get_shares_full(start, end) (a series, unlike the Ticker.shares property). 30m bars are fetched as 15m and resampled (a documented Yahoo workaround); intraday history is capped — "Intraday data cannot extend last 60 days", 1m ~7 days.

Minimal correct call

import yfinance as yf
df = yf.download(
    ["AAPL", "MSFT"],
    start="2020-01-01", end="2024-01-01",   # end is EXCLUSIVE
    interval="1d",
    auto_adjust=True,        # explicit: flipped in 0.2.51, hardened at 1.0
    multi_level_index=False, # flat columns even for one ticker
    ignore_tz=True,          # explicit: intraday default changed in 1.4.0
    progress=False,
)
# No 'Adj Close' column exists when auto_adjust=True.
# This universe has NO delisted names — do not backtest a screen on it.

See also

• ../../../fin-core/skills/market-data-sourcing/SKILL.md — vendor choice, survivorship-free universes
• ../../../fin-core/skills/market-data-sourcing/references/yfinance.md — the source card
• ../../../fin-core/skills/research-integrity-guards/references/adjustment-conventions.md — split/dividend conventions

Where this sits

This file is the deep dive on one library and assumes the choice is already made. For which library to pick, how it compares with the alternatives, and the traps that span several of them, the entry point is the domain skill market-data-sourcing (../../../fin-core/skills/market-data-sourcing/SKILL.md).