Getting Historical Equity Data Without Paying for a Bloomberg Terminal
yfinance is a Python wrapper around Yahoo Finance's unofficial API. It pulls OHLCV data, fundamental metrics, options chains, and some macro indicators. It is free, pip-installable, and fast enough for most retail-scale workflows. The trade-off is that Yahoo does not guarantee uptime or data accuracy, so you cannot run production systems on it without safeguards. Install it once with pip install yfinance. That is the entire barrier to entry. From there, pulling a few years of daily closing prices for a stock takes three lines of code.
Unlock Market Data With Yfinance In Python A Gamechanger For Traders And Investors
The simplest call looks like this: import yfinance as yf This returns a pandas DataFrame with a DatetimeIndex and columns for Open, High, Low, Close, Adj Close, and Volume. Adj Close is the one you want for backtests because it accounts for dividends and splits. Raw Close will mislead you if the stock pays dividends or undergoes a split during your period.
data = yf.download("AAPL", start="2020-01-01", end="2024-01-01")
For a single ticker with explicit columns, use the Ticker object instead: tick = yf.Ticker("MSFT") The
hist = tick.history(period="max")period parameter accepts strings like "1d", "5d", "1mo", "3mo", "6mo", "1y", "2y", "5y", "10y", and "ytd". If you need a custom range outside these buckets, fall back to start and end with ISO date strings. The Ticker approach also gives you access to financials, balance sheets, cash flow statements, and analyst recommendations through properties like income_stmt and recommendations.
Get the Full Details
Batch downloading multiple tickers is where yfinance starts to feel useful rather than just convenient: tickers = ["SPY", "QQQ", "IWM", "VTI", "VOO"] This returns a MultiIndex DataFrame with ticker as the outer level and date as the inner level. It saves you from writing a loop that makes eight requests instead of one. Yahoo's backend handles it, and you avoid burning through rate limits unnecessarily.
batch = yf.download(tickers, start="2019-01-01", end="2024-01-01", group_by="ticker")
There is one thing that catches almost everyone off the first time. start and end are interpreted in UTC, but the resulting timestamps get localized to your system timezone after the fact. If you are in a negative UTC offset zone, you will lose the last trading day of your range. I wasted half a morning debugging a backtest that appeared to be missing a weekend when it was actually missing the final Friday because the UTC conversion shifted it past my end boundary. The workaround is to set end to midnight on the day after your desired cutoff, or better yet, append tz="US/Eastern" to the download call to lock the index into a specific market timezone before the shift happens. Another thing that matters more than the basic usage. When you pull options data with tick.options, yfinance returns a list of expiration dates. You then pass one of those strings into tick.option_chain(date). The chain object gives you call and put dataframes with strike, last price, bid, ask, volume, open interest, and implied volatility. This alone replaces a $200 per month data subscription for anyone doing basic options screening. I ran into a case where tick.history() returned zero rows for a particular small-cap stock despite the ticker being valid on Yahoo's website. The issue was that the stock had a recent name change or symbol correction, and yfinance resolved to the old symbol which now pointed to a defunct listing. Running yf.Ticker("XYZ").info first and checking longName and symbol against what Yahoo actually returned caught the mismatch before I wasted time debugging the download. I now wrap my initial fetch in a try-except that logs the resolved symbol and compares it to the requested one. If they differ, I alert and skip.
Where yfinance Actually Breaks
The tool is unreliable in several specific scenarios, and pretending otherwise gets people hurt in live trading. Yahoo changed their API version mid-2024. The older v8 endpoint supported intraday data with minute-level granularity. The current v7 endpoint only returns daily bars unless you request 1-day intervals, and even then the intraday bars are limited to the last 7 days with no guarantee of consistency. If your strategy requires minute data, yfinance is not your solution. Use Polygon, IEX Cloud, or Interactive Brokers instead. Fractional shares and certain ETFs with complex structures sometimes return incomplete OHLCV sequences. I discovered this the hard way with a commodity-backed ETF that had missing volume rows on several rebalance dates because the underlying constituent changes caused Yahoo to drop volume data while keeping price. The price column looked normal, but volume was NaN. Anyone filtering by volume thresholds without checking for nulls would silently exclude those days from analysis and draw wrong conclusions about liquidity patterns. Real-time data is another blind spot. The latest Close price you get from yfinance is typically 15 to 30 minutes delayed during market hours, depending on the exchange and the ticker. Pre-market and after-hours pricing is inconsistent. The field exists in the raw JSON response but yfinance does not expose it in a stable way across ticker types. If you need live quotes, call a broker API directly.
The fundamental data pulled via financials, balance_sheet, and cashflow is quarterly and annual but comes with a significant lag. Earnings releases hit Yahoo's database hours after they appear on SEC EDGAR, and restatements sometimes take days to propagate. For fundamental screening based on the most recent quarter, verify the date stamp in tick.info.get("financialsCurrency") and cross-reference with the filing date if precision matters.
Practical Patterns That Actually Work
Vectorized returns calculation beats looping over rows every time. Given a DataFrame df with an Adj Close column: df["returns"] = df["Adj Close"].pct_change() Use
df["log_returns"] = np.log(df["Adj Close"] / df["Adj Close"].shift(1))np.log for geometric returns in backtests. Simple percentage returns compound incorrectly when you aggregate across multiple periods. Log returns add cleanly. This is basic portfolio math but people still mess it up in scripts.
When pulling data for a watchlist, always set auto_adjust=True in the Ticker constructor if you want adjusted prices by default. The default is already True for history(), but when you pull info or financials, the adjust flag does not apply the same way. Being explicit prevents subtle inconsistencies between your price series and your fundamentals series. Caching downloaded data locally is not optional if you are doing iterative development. Yahoo rate-limits aggressive requests, and the error messages are opaque. I use a simple SQLite cache keyed by ticker plus date range. Before each fetch, I query the cache. If the requested range is fully covered, I return the cached slice. If it is partially covered, I fetch only the missing tail and append. This cuts redundant network calls by roughly 80 percent in a typical research session and avoids the temporary ban that shows up as a 429 status with no data returned. For event-driven strategies, rolling forward fills matter. If you buy a stock on day N based on a signal calculated from day N close data, you cannot execute at day N close. The signal is only knowable after the market closes. Use df["signal"].shift(1) to align signals with the next day's open, or df["signal"].shift(-1) if you are simulating intraday execution at close. Forgetting this shift is the single most common source of lookahead bias in retail backtests. I have seen it in GitHub repos from people who should know better.
There is a quiet feature worth knowing about. The holders property returns institutional ownership and fund holders. The major_holders property gives percentages like insiders owned, institutions owned, float held by institutions, and shares outstanding. This data updates irregularly and is sourced from periodic SEC filings, not real-time. Treat it as a snapshot, not a live feed. Calendar data through tick.calendar provides earnings dates and dividend dates. The earnings table includes estimated EPS and reported EPS when available. I use this to filter out blackout windows before earnings reports when my strategy normally stays flat. The dates are approximate and can shift. Always cross-check with the company's investor relations page before committing capital based on an automated schedule. If you need continuous futures data for commodities or indices, yfinance will not give you rolled contracts. You get a single ticker like CL=F for crude oil futures, but it does not auto-roll. The price series will show artificial gaps at contract expiry. I build roll-adjusted series by downloading individual contract months and stitching them together using a front-month detection rule based on open interest. It is more work but necessary for any futures backtest that spans multiple expirations.
The Honest Assessment
yfinance covers enough ground to be the default starting point for equity and options research in Python. It is fast to set up, well-documented enough, and integrates cleanly with pandas and numpy. It is not a production-grade data source. The API is unofficial, the accuracy guarantees are nonexistent, and the downtime happens without warning. If you are building a tool that other people depend on, or if you are executing trades based on the data, you need a fallback source and a validation layer. For most people doing personal research, screeners, or educational backtests, it is more than sufficient. The friction is in understanding its limits and coding around them. The value is in getting accessible data without signing up for paid tiers or writing your own web scrapers. That balance is why it persists as the go-to option despite its flaws.
