Twelve years of options chains, probed rather than promised

How far back it actually goes

Catalogues tend to quote a start year. We went looking for the first day that answers. Probing SPY chains backwards: 2013-01, 2013-06, 2014-01 and 2014-03 return 404. 2014-09-02 returns 206 contracts in a single expiry series, and every year after it answers. So the honest boundary is September 2014 — twelve consecutive years to today.

The blue rows are the twelve-year archive: US index and equity options — the big ETFs (SPY, QQQ, IWM, DIA), the cash indices (SPX, NDX, RUT, XSP, VIX) and the large single names. The two green rows are ours alone, and much shorter: five-minute intraday snapshots we capture ourselves, and CME futures options, which are not in any third-party options archive at any tier.

How deep one expiry goes

Depth matters more than start date. An archive that reaches 2014 but carries forty strikes is a different product from one that carries four hundred. One June expiry of SPY, one strike ladder, queried once per year:

Between 206 and 504 contracts in a single expiry series, rising as the strike grids widened over the decade. The variation between years is real — it follows how many strikes were listed around spot at that moment, not a gap in the data.

What each row carries

Per contract, per day, the archive returns the full quote rather than a single price:

• bid · ask — and bid_size · ask_size, so you can weight by resting size instead of assuming the mid
• bid_exchange · ask_exchange — which venue posted it
• bid_condition · ask_condition — the quote condition codes, which is how you exclude a quote that was not firm
• open · high · low · close · volume · count
• oi — open interest, and iv, delta, gamma, theta, vega
• last_trade and created — two separate timestamps

That last line is the one worth pausing on, because it is the difference between a clean backtest and a subtly broken one. The moment a row was written and the moment it last traded are not the same moment, and on an illiquid contract they can be hours apart. A dataset that gives you one timestamp is not simpler — it is hiding which of the two you are holding. If you are pairing greeks against a price, you need to know whether that price is from this minute or from Tuesday.

How to tell a usable options archive from a free one

Five checks, in the order that catches the most problems soonest:

1. Probe the claimed start date. Not the catalogue — the API. Ours answers from 2014-09-02 and 404s before it, and now you know because we ran it.
2. Count contracts in one expiry. A thin ladder is the most common quiet defect: the archive technically covers the date, and the strikes you need are missing.
3. Look for size next to price. A bid with no bid_size cannot be distinguished from a bid nobody would fill.
4. Ask how many timestamps there are. One is a warning. It means quote age is unknowable.
5. Check whether open interest is aligned to the right session. OI is published the next morning for the previous close. An archive that files it under the wrong day shifts every opening-versus-closing inference by one session.

What it cannot do

These limits are not ours specifically — they are properties of the consolidated tape, and any vendor claiming otherwise is selling an inference as a field.

There is no participant identity. OPRA carries no field naming a firm, a desk or a customer type. Every "institutions were buying" headline is a classification built on condition codes and size, ours included — we publish the measured error of our own version next to it rather than presenting it as observation.

There is no order book. Options quotes are top of book. Resting depth and queue position do not exist in this data at any price. If you need that on futures, market-by-order is a different product.

Open interest lags by a session. It is published the following morning. Opening versus closing is therefore an inference from an OI change, and it is a good one — but it is arithmetic, not a reported field.

And one finding of our own, since it decides whether this data is worth buying at all: across roughly two hundred strategy tests we have published, options data has repeatedly carried a real edge where order flow carried none — the put/call contrarian signal survives costs and out-of-sample testing, while twelve order-flow indicators did not. We would rather tell you which half of our own catalogue tested well.