How to download historical stock prices in Python

The short way

With yfinance, installed with pip install yfinance, this is the whole of it:

import yfinance as yf

prices = yf.download("AAPL MSFT", start="2020-01-01", end="2026-09-01")
prices.to_csv("prices.csv")

You get a pandas DataFrame, no key, no account and no bill. That is why every tutorial uses it, and for reading a chart or sketching a strategy it is the correct answer.

One parameter in there is doing more than it looks. auto_adjust defaults to True, so the prices you just downloaded are already adjusted for splits and dividends — which is usually what you want and is emphatically not what the exchange printed on the day. Pass auto_adjust=False to get the raw series, and actions=True to get the splits and dividends alongside it so you can do the adjusting yourself.

What the options are

The choice is not really between libraries. It is between having a contract and not having one.

No contract. yfinance reads Yahoo's undocumented endpoints. There is no agreement behind it, no support, no uptime commitment and no published quota, and when Yahoo changes something every user breaks simultaneously until a volunteer maintainer ships a release.

A cheap contract. Tiingo sells end-of-day equity history back to 1962; its free Starter plan is 50 requests an hour, 1,000 a day and 500 unique symbols a month, and the $30 Power plan raises that to 10,000 an hour — personal and internal use only, with the commercial licence at $50. EODHD covers 60-plus exchanges from $19.99 a month, and its unit of account is worth reading twice: the 100,000 daily allowance is calls, not requests, and one fundamentals or options request costs ten of them.

A broker's feed. Alpaca Market Data is free forever at 200 requests a minute, but the free tier is one exchange — IEX — and it withholds SIP data from the last 15 minutes. The $99 Algo Trader Plus plan is the full SIP and OPRA at 10,000 requests a minute. If you are going to trade through Alpaca anyway, this is one account rather than two.

The exchange feed itself. Databento is the shape when bars are not the product and you want the order book, billed per gigabyte of what it sends you rather than per month. Its Python client needs Python 3.10 or later; there is no Go, Java, C# or JavaScript client at all.

Free with no strings, in a narrow lane. Alpha Vantage gives a real key with real terms for nothing, at 25 requests a day — small, but a contract, and verified open-source and educational projects are exempt from the cap entirely.

Where this breaks

Splits and dividends. This is the one that makes a correct-looking script wrong.

A stock that splits four-for-one drops 75% overnight without anybody losing money. A dividend takes the price down by roughly its own size on the ex-date. A price series that records those two events as returns will show you crashes that never happened, and a backtest run on it will either buy the dip or stop out — repeatedly, and always on the same days.

So every provider offers an adjusted series, and here is the part that costs the afternoon: they do not agree, and they do not default the same way. yfinance adjusts by default. Others return the raw series unless you ask. Some adjust for splits only and leave dividends alone, which is a third answer again. Downloading the same ticker from two providers and diffing the closes is a perfectly reasonable first thing to do, and the difference you find is not a bug in either of them. Why adjusted close differs is the full version of this.

Two smaller walls behind it. Intraday history is short: yfinance cannot go past the last 60 days at any intraday interval, and most cheap plans are end-of-day only. And rate limits bind on the backfill, not on the daily job — pulling five years for 500 tickers is 500 requests in a burst, which is ten hours on Tiingo's free tier and twenty days on Alpha Vantage's.

If you outgrow this

The signals that you have, in the order they usually arrive.

Your script broke and nobody owes you anything. That is the moment to stop using an unofficial endpoint, whatever it costs. Any of the contracted providers above is under ten dollars a month at the entry tier.

You need delisted tickers. A universe assembled from the names that exist today has survivorship bias baked into every backtest run on it. Norgate Data carries delisted securities and index constituent membership, which is the fix.

You need the same data your broker fills against. Then the question is no longer historical and you are shopping for a live feed, where the exchange's fee schedule rather than the vendor's price list sets the number — see why real-time stock data is so expensive.

You are redistributing. The moment a price reaches somebody who is not you — a public page, a newsletter, an app — you need a licence that says so, and personal-use plans like Tiingo's Power tier explicitly do not. Read that line before building, not after.

The rest of the field is in market data APIs, narrowable by latency, coverage and price.