Equity Options: subscriptions and chain data — QuantConnect / LEAN

Two decisions shape every equity-options algorithm, and each has two answers that look interchangeable but carry different data: how you subscribe (add_option vs add_option_contract) and how you read the chain (the slice's option_chains vs the option_chain() method). Picking the wrong member of either pair produces algorithms that trade on day-old numbers without any error.

Greeks/IV freshness — the core rule

There are two different option datasets behind the APIs (doc: "There is a daily, pre-calculated dataset based on the end of the previous trading day. There is also a stream of values that are calculated on-the-fly as your algorithm runs."):

The operating rule that follows: any intraday decision — entry sizing, delta hedging, IV-threshold checks, strike selection at a decision time — should read the slice's chain. The option_chain() method is for daily contract discovery (find which contracts exist, filter by prior-day delta/OI, then subscribe); using it for intraday sizing or greeks means every number is one day stale, and nothing errors to tell you. Greeks on slice contracts are computed lazily — accessing greeks costs nothing; accessing greeks.delta triggers the calculation — so only read the greeks you need.

Pick ONE route — and subscribe only what the strategy can trade

Decide the route once, in initialize. Mixing them is usually a smell:

• Universe route (add_option + filter): a universe's purpose is its slice chain. If all selection, pricing, and sizing happen through option_chain() anyway, the minute-resolution subscriptions are pure cost — either read the slice for the decisions that need live data, or drop the universe and subscribe the picks directly.
• Discovery route (option_chain() + add_option_contract): the daily chain picks the contracts; subscribe each pick. Match the data to the decision: the option_chain() rows are previous-close values — fine when daily-granularity data suits the strategy (screening, ranking, a daily-cadence rule that tolerates day-old marks), but when the strategy calls for decision-time values (intraday sizing, hedging, entry marks), read the subscribed picks' live quotes from the slice. Whichever you use, know which one you are using.
• A subscription earns its cost by being read or being held: a contract you own needs its subscription (position pricing, fills, expiry processing) even if you never read a chain. What doesn't earn its cost is breadth — a wide add_option universe where the algorithm neither reads the slice chain nor holds more than its few picks.
• option_chain() needs no option subscription — the underlying's equity subscription is sufficient. Never add add_option, a filter, or add_universe_options just to make a chain available; an empty chain has some other cause.

Size the subscription to the trade, not to the chain:

• Prefer a strategy-shaped filter when one matches the structure — u.iron_condor(30, 5, 10), u.straddle(30), u.call_spread(30, 5), and friends subscribe only the legs the strategy needs.
• Otherwise derive the strike span and DTE window from the structure's own numbers (target deltas, wing widths, expiry rules), adding a delta-band filter (u.delta(0.05, 0.30)) where the structure is delta-targeted. The span should be the tightest one that always contains the tradable candidates — not a safety blanket.
• strikes(-100, 100), or an unfiltered expiration(0, 500) window, subscribes thousands of minute-resolution contracts — a cost bug even when the algorithm is otherwise correct.

Subscription route 1: add_option — a filtered basket (universe)

For strategies that pick contracts from a chain at decision time (spreads, condors, straddles, rolling structures), subscribe once in initialize:

self.universe_settings.asynchronous = True
option = self.add_option("SPY")                 # underlying auto-subscribed RAW
option.set_filter(lambda u: u.include_weeklys().strikes(-3, 3).expiration(15, 60))
self._symbol = option.symbol                    # canonical Symbol — save it

• The canonical option.symbol is not tradable — it is the key into data.option_chains and the first argument of every OptionStrategies factory.
• Filter timing: selection "usually runs at the first bar of every day"; a newly selected contract's data arrives "in the next Slice". So the chain can be absent on any given event — always guard: chain = data.option_chains.get(self._symbol); if not chain: return.
• Default filter if you never call set_filter: standards + weeklys, within 1 strike of the underlying price, expiring within 35 days — almost never what a spec wants; set the filter explicitly.
• Filter building blocks: strikes(min, max) (relative strike counts around the money, not dollars), expiration(min_days, max_days), calls_only()/puts_only(), weeklys_only()/standards_only()/include_weeklys(), front_month()/back_month(), greek/OI filters delta(min, max), iv(min, max), open_interest(min, max) (these compare prior-day EOD values — use them for coarse pre-selection, then pick precisely from live slice greeks), and strategy-shaped helpers (u.straddle(30), u.iron_condor(30, 5, 10)) that narrow the universe to exactly the legs a named strategy needs.
• Resolution rule: the option resolution must be ≥ the underlying Equity's resolution (minute equity → minute/hour/daily options). Contract subscriptions support minute, hour, and daily; minute is the default.

Reading the chain at a decision time:

def on_data(self, data: Slice) -> None:
    chain = data.option_chains.get(self._symbol)
    if not chain:
        return
    spot = chain.underlying.price
    atm_call = min((c for c in chain if c.right == OptionRight.CALL),
                   key=lambda c: abs(c.strike - spot))
    delta = atm_call.greeks.delta        # live price-model value, computed on request
    mid = (atm_call.bid_price + atm_call.ask_price) / 2

In a Scheduled Event there is no slice parameter — read the same object through self.current_slice.option_chains.

Subscription route 2: add_option_contract — individual known contracts

For strategies that hold a few specific contracts found by daily screening, discover with the option_chain() method, then subscribe each pick:

# In a scheduled event — option_chain() carries PRIOR-DAY EOD greeks: fine for
# discovering contracts, wrong for intraday sizing or hedging.
chain = self.option_chain(self._underlying, flatten=True).data_frame
expiry = chain.expiry.min()
symbol = chain[(chain.expiry == expiry) & (chain.right == OptionRight.CALL) &
               (chain.delta > 0.3)].sort_values("openinterest").index[-1]
self.add_option_contract(symbol)

The three idioms that must accompany this route:

1. Seed prices — a contract subscribed this time step has no data until the next slice ("you'll need to wait until the next Slice to receive data and trade the contract"). Set self.settings.seed_initial_prices = True in initialize to trade it immediately.
2. RAW underlying — subscribe the underlying with self.add_equity("SPY", data_normalization_mode=DataNormalizationMode.RAW) so strike-vs-price comparisons share a scale. If you skip this, LEAN auto-subscribes (or force-switches an existing subscription) to RAW anyway — see the history caveat below.
3. Volatility warm-up — set_warm_up in initialize cannot anticipate contracts added later; warm the underlying's volatility model in a security initializer (feed ~30 daily bars through security.volatility_model.update(...)) so slice-chain IV/greeks are sane from the first read. On the add_option route the simple self.set_warm_up(31, Resolution.DAILY) suffices (default volatility model needs 30+1 daily bars).

To read a subscribed contract from the slice, index the chain with the canonical: data.option_chains.get(self._contract_symbol.canonical). remove_option_contract(symbol) cancels its open orders and liquidates the position.

The RAW-underlying history caveat

Because any option subscription forces the underlying Equity to RAW normalization, a history() call on that underlying now returns raw closes — dividend gaps included. Any realized-volatility, daily-return, or overnight-gap calculation on an option underlying must request adjusted data explicitly: self.history(self._underlying, 22, Resolution.DAILY, data_normalization_mode=DataNormalizationMode.ADJUSTED). Do not "fix" this by subscribing the underlying ADJUSTED — LEAN silently switches it back to RAW.

Common mistakes

• Sizing or hedging from option_chain() intraday. Its marks and greeks are the previous close; position sizes computed from them are provably off versus decision-time quotes. Decision-time numbers come from the slice's chain.
• Trading the canonical symbol, or passing a contract symbol where the canonical is required (chain lookup, OptionStrategies factories).
• No if not chain: guard — the chain is legitimately absent before the first daily selection and whenever the filter turns over.
• Forgetting include_weeklys() when a spec's DTE window (e.g. 21–45 days, closest to 30) needs weekly expiries to hit its target.
• Carrying a wide universe you never read — subscribing a broad chain while doing all selection, pricing, and sizing through option_chain(): every subscription beyond the contracts actually held is pure cost. Read the universe's slice chain for the decisions that need live data, or drop the universe and use add_option_contract on the picks.
• Subscribing a huge chain as a tradability crutch (±100 strikes, unbounded expiries) so any discovered contract can be ordered directly — subscribe the picks with add_option_contract instead, and let a strategy-shaped or structure-derived filter keep the universe minimal.
• Computing RV/returns on the force-RAW underlying without an explicit adjusted-normalization history request.
• Only American-style US equity options are supported; strikes in the filter API are relative counts, not dollar offsets — strikes(-1, 1) means one strike either side of the money.