Where to Get Historical Dividend Data for Stocks

August 19, 2026

Four sources carry historical dividends, and they are not interchangeable. A company’s own investor relations page is authoritative but covers one company. SEC filings report the total cash a company paid over a period, not the amount per share attached to any particular date. Market data APIs carry a per-share cash amount stamped on the ex-dividend date, which is the form every yield calculation and total-return series actually needs. Commercial terminals carry all of it and price accordingly. For anything programmatic, the API route is the only one that scales, and the field to look for is a per-share cash amount on the ex-dividend date.

What does a single dividend record contain?

A dividend has four dates and one amount, and confusing them produces wrong numbers rather than approximate ones.

The declaration date is when the board announces the payment. The ex-dividend date is the first session on which a buyer does not receive it; the price drops by roughly the dividend amount that morning, which is why this date and no other belongs in a return calculation. The record date determines who sits on the register. The payment date is when cash actually arrives, often several weeks later.

Most datasets key on the ex-dividend date because that is the date the market prices. Some carry all four. A source that gives you only a payment date and an amount cannot be joined cleanly to a price series without introducing a timing error of a few weeks, which matters for anything measured at daily or monthly frequency.

The amount has its own trap. A regular quarterly payment and a one-off special distribution both arrive as “a dividend” in most feeds, with nothing separating them. One large special can double a company’s apparent trailing yield and then vanish the next year, which reads as a dividend cut when nothing was cut.

Why do two sources report different dividend yields for the same stock?

Usually because they are computing different things under the same label.

Trailing yield sums the cash paid over the last twelve months. Indicated yield takes the most recent regular payment and multiplies by the payment frequency. Forward yield uses an estimate of the next twelve months. For a company that has just raised its dividend, those numbers can sit two or three percentage points apart, and none of them is wrong.

The second cause is adjustment. If the price in the denominator is a dividend-adjusted price rather than the traded price, the result is not the yield an investor would receive. Split adjustment cuts the other way: a per-share dividend paid before a four-for-one split needs restating onto the current share basis before it can be compared with a recent one, and not every source does that consistently.

What are the options in Python?

Alpha Vantage wins a point here worth stating plainly: it returns all four dates on every record, which xfinlink does not. If your work turns on the gap between record date and payment date, that endpoint is the better tool, and it is cheap to test against with the documented demo key.

The tradeoff runs the other way once the unit of work stops being one symbol. Dividends are rarely the end goal; they feed a yield screen, a payout study, or a total-return series, each of which also needs prices for the same companies over the same window. A dividend feed that arrives separately from the price series has to be joined to it, and that join is where ticker changes and share-class mix-ups do their damage.

How do you pull a dividend history and compute a yield?

The dividend column arrives inside the price frame, on the ex-date row, in the same units as the traded close beside it.

import pandas as pd
import xfinlink as xfl

xfl.set_api_key("YOUR_API_KEY")  # free at https://xfinlink.com/signup

df = xfl.prices("KO", start="2015-01-01", end="2026-08-01",
                fields=["close", "adj_close", "return_daily", "dividend"])
df["date"] = pd.to_datetime(df["date"])

cash = df[df["dividend"].fillna(0) > 0][["date", "dividend", "close"]]
last12 = cash[cash["date"] >= cash["date"].max() - pd.Timedelta(days=365)]["dividend"].sum()
last_close = df.iloc[-1]["close"]
print(f"trailing 12-month cash {last12:.2f}, yield {100 * last12 / last_close:.2f}%")

price_ret = df.iloc[-1]["adj_close"] / df.iloc[0]["adj_close"] - 1
total_ret = (1 + df["return_daily"].fillna(0)).prod() - 1
print(f"price return {100 * price_ret:.1f}%, total return {100 * total_ret:.1f}%")

ex-dividend dates on file: 46 from 2015-03-12 to 2026-06-15
      date  dividend  close
2025-06-13      0.51  71.02
2025-09-15      0.51  66.21
2025-12-01      0.51  71.95
2026-03-13      0.53  77.34
2026-06-15      0.53  80.91

trailing 12-month cash dividend: 2.08
last close 87.59 on 2026-07-31  ->  yield 2.37%

2015-01-02 to 2026-07-31
  price return  (adj_close):    107.9%
  total return  (return_daily): 185.1%
  dividends contributed:         77.2 points

Two things in that output repay a close reading. The quarterly rate steps from 0.51 to 0.53 at the March 2026 ex-date, which is why the trailing sum of 2.08 sits between the 2.04 paid across 2025 and the 2.12 the new rate implies. Over eleven and a half years, dividends account for 77 of Coca-Cola’s 185 points of total return. A study that measures performance on price alone discards about forty percent of what the shareholder earned.

What quietly breaks a dividend backtest?

Two things, and neither announces itself.

The first is survivorship. Build a universe from today’s list of dividend payers, run it back ten years, and you have selected on the outcome: companies that cut and then fell out of the index are absent from the sample, so the strategy looks safer than it was. A point-in-time roster fixes it. xfl.index("sp500", as_of="2015-12-31") returns the members as they stood that day, including the ones that later left.

The second is identity. Tickers get reassigned, and a company that changes its symbol splits into two partial histories whenever the join key is a string of letters. Keying on entity_id keeps dividends and prices attached to the company rather than to the letters it happened to trade under. The measured version of that problem is in our write-up on ticker recycling, and survivorship bias in backtesting covers the first.

For a worked example using both, does a high dividend yield predict a dividend cut rebuilds eleven years of dividend rates from ex-date cash across point-in-time index members. Field definitions sit in the docs, and history depth by plan is on the pricing page.

FAQ

Does adjusted close already include dividends?
Not in every dataset, and the difference is large. In xfinlink, adj_close is adjusted for splits only, so it stays a clean price series, while return_daily is the total return including dividends. Check which convention a source uses before comparing performance numbers across providers.

How do I separate regular dividends from specials?
Neither of the feeds in the table above carries a flag for it, so infer it. Take the year’s payments, compute the median, and discard anything far above it; a payment more than roughly one and a half times the year’s median is almost always a special.

Can I get dividend data for delisted companies?
Only if the source keeps the delisted company’s price history at all, which many do not. This is the survivorship problem in a different guise, and it is worth testing on a known case before trusting a backtest built on the answer.

Built with xfinlink — free financial data API for Python. pip install -U xfinlink