Trading strategies·event

Scheduled vs unscheduled events

∞structural
Reviewed 4 June 2026. As of 2026: a permanent feature of the market, not an edge that decays.

A scheduled release (you know when, not what) is a different problem from breaking news (you know neither). The microstructure around each, spread widening and liquidity withdrawal, differs sharply.

The idea

Scheduled vs unscheduled events annotated diagramfigure
A scheduled release (you know when, not what) is a different problem from breaking news (you know neither). The microstructure around each, spread widening and liquidity withdrawal, differs sharply.

Reference figure. This concept is explained in prose and diagram; the interactive widgets live on the flagship pages it links to under Where this fits.

What is a scheduled event?

A scheduled event is one whose timing is public but whose content is the surprise: economic releases (payrolls, CPI, GDP), central-bank decisions (FOMC, ECB), and company earnings. Because you know the exact timestamp, you can pre-position, pre-stage orders, and pre-build the parser that turns the released number into a trade in microseconds. You trade the surprise, not the level. This is the prepared half of directional event trading.

Intuition first. Everyone knows non-farm payrolls prints at 08:30 ET on a known Friday; what no one knows is the number. The market has already priced the consensus, so the tradable move is the gap between the actual figure and that consensus, the surprise. A figure that lands exactly on consensus barely moves the price; a large miss moves it sharply. A strong number is bullish only if it beats what was already priced in.

The repricing is driven by the surprise (actual minus expected), not by the actual figure alone. Event traders track the consensus forecast precisely so they can pre-compute the mapping from surprise size to expected move.
ΔP  ≈  β ( actual−expected )⏟surprise\Delta P \;\approx\; \beta \,\underbrace{(\,\text{actual} - \text{expected}\,)}_{\text{surprise}}

What preparation buys you is the whole edge. A known timestamp means you can pre-stage everything: a parser that reads the release format, a decision rule mapping surprise to order, and orders pre-validated and ready to fire. The race at T=0T=0 is then purely who submits the correct order first among firms that all prepared. This is a microsecond game decided by systems and colocation, not by thinking on the spot. By asset class: rates and FX react to macro releases and central banks; single stocks to earnings; index futures to macro; crypto to scheduled protocol and macro events; and prediction markets to the resolution of the very events they are written on.

What is an unscheduled event?

An unscheduled event is one you cannot anticipate: a breaking headline, a surprise announcement, a geopolitical shock, a large unexpected order. You know neither when nor what, so there is no pre-positioning, only the fastest correct reaction as the news arrives. This is the home of machine-readable news feeds and NLP: parse the headline and act before the field finishes reading it.

Intuition first. A CEO resignation, an M&A leak, a regulatory action, a war headline: these arrive without a calendar slot. You cannot pre-position because there is nothing to position for until it happens. The entire edge is in the reaction: detect the news, interpret it correctly, and submit the order faster than competitors.

Interpretation dominates here far more than for scheduled events. A scheduled release is a number in a known format, trivial to parse, the surprise is arithmetic. An unscheduled headline is unstructured language: "Company X explores sale" might mean a takeover (bullish) or a distress sale (bearish). Reading it correctly and fast is the hard part, which is why machine-readable news and NLP/LLMs matter most here. The false-signal problem is acute: react to a headline that turns out to be misread, stale, a duplicate, or fake, and you take a directional loss. Speed without correctness is negative edge: the discipline is a confidence threshold, firing only when the parse is unambiguous.

Expected P&L on an unscheduled reaction is the move you capture when your read is right, minus the directional loss when it is wrong, weighted by your interpretation accuracy. A faster but less accurate parser can have lower expected edge than a slower, more correct one.
E[π]  =  p ΔP⏟correct read  −  (1−p) ℓ⏟false signal,p=P(read correct)\mathbb{E}[\pi] \;=\; \underbrace{p\,\Delta P}_{\text{correct read}} \;-\; \underbrace{(1-p)\,\ell}_{\text{false signal}}, \qquad p = P(\text{read correct})

Crypto and prediction markets are unscheduled-event-heavy: 24/7, headline-driven, and (on prediction markets) the contracts are the events. A surprise news item can resolve or sharply reprice a contract instantly; see also crypto, where the same headline sensitivity runs around the clock.

How latency and pre-positioning differ between the two

For scheduled events, latency is a prepared race: parsers and orders are pre-staged, so the winner is decided in microseconds among firms that all did their homework, and pre-positioning on a forecast is possible (but it is a prediction bet, not an event-reaction bet). For unscheduled events there is no pre-staging, only raw detect-interpret-react latency, where correct interpretation is the bottleneck.

Scheduled: the prepared race. Everything that can be done before T=0T=0 is: feed handlers warmed, parser compiled, decision rule tabulated, orders pre-validated. At T=0T=0 the only work is read-the-number, look-up-the-rule, fire. The latency budget is dominated by the path to the matching engine (colocation/FPGA), not by computation. Pre-positioning (taking a position before the release on a forecast) is a separate, riskier game: you are betting on the surprise's direction and exposed if you are wrong, and to the pre-release liquidity thinning.

Unscheduled: the reaction race. Nothing is pre-staged because nothing is known. The latency chain is: news arrives on the feed, parse/interpret (the slow, hard step), decide, submit. Here interpretation latency (NLP/LLM inference) can dwarf network latency, and the trade-off between speed and correctness is sharp. Firms tune a confidence threshold: act instantly on unambiguous machine-readable items, defer or skip ambiguous ones.

Scheduled latency is dominated by the network path to the matching engine; unscheduled latency is dominated by interpretation. The bottleneck moves from the wire to the model, which is why AI shifts the unscheduled game and barely touches the scheduled one.
tsched≈tnet,tunsched≈tinterpret+tnet,tinterpret≫tnett_{\text{sched}} \approx t_{\text{net}}, \qquad t_{\text{unsched}} \approx t_{\text{interpret}} + t_{\text{net}}, \quad t_{\text{interpret}} \gg t_{\text{net}}

The shared truth: in both cases the slow drift after the event is largely arbitraged in liquid markets, so the edge is the jump window. If you cannot be early and correct, you are providing liquidity to those who are, and, as a market maker, you defend by widening or pulling around known event times, the pre-release spread widening you can observe directly.

How each is actually traded

Scheduled events are traded by pre-staging a parser and a surprise-to-order rule, then winning the microsecond race at the known timestamp (and optionally pre-positioning on a forecast). Unscheduled events are traded by ingesting machine-readable news, interpreting it correctly and fast, and firing only on high-confidence parses. Both defend against false signals and the pre/post-event liquidity collapse.

The scheduled playbook: subscribe to the lowest-latency release channel; pre-compile the parser for that release's exact format; tabulate surprise to trade; pre-validate orders; at T=0T=0, read-lookup-fire. Manage the pre-release liquidity thinning (wider spreads, less depth) and the risk that the surprise is within noise (do not trade tiny surprises that will not clear costs). The unscheduled playbook: ingest elementized/machine-readable news (news trading); run NLP/sentiment/LLM classification; gate on confidence; fire only on unambiguous, novel, market-moving items; deduplicate against feeds you have already seen. The whole risk is a wrong or stale read.

For both, model fills against a thin, fast-moving book: at event time you are trading into vanishing liquidity, so realistic fill modelling is essential (see backtesting & simulation). And both interact with circuit breakers: a large enough event-driven move can trip a halt, changing the game mid-trade. The trade arrivals themselves cluster in bursts around the event rather than on the clock: the statistical signature is a point process, modelled with autoregressive conditional duration (Engle & Russell, 1998); see irregular time.

Worked example

Two synthetic events for contrast, as of 2026, with illustrative numbers. Scheduled: payrolls. Release at 08:30:00.000 ET. Consensus +180k+180\text{k}; actual +250k+250\text{k}, a +70k+70\text{k} upside surprise. The pre-computed rule maps a +70k+70\text{k} surprise to about +0.30%+0.30\% in the index future. Liquidity thinned from 08:29:55 (spread 1→31 \to 3 ticks). At 08:30:00.008 the number prints; a prepared firm reads it, looks up the rule, and lifts asks in the first ∼20\sim 20 ms, capturing most of the +0.30%+0.30\%. A reactor at 08:30:00.250 finds the move done and pays the wide spread, a net loss.

Unscheduled: a takeover headline. At 11:43:17 a wire prints "Company X to be acquired at a premium". No warning. A machine-readable-news system classifies it as a confirmed acquisition (high confidence) in ∼5–40\sim 5\text{–}40 ms of inference, then buys X before slower readers finish. A competitor whose NLP flags it "ambiguous" waits and misses the first 1.5%1.5\% pop: correct caution, but no profit. A third system misreads a similarly-worded "Company X denies takeover talk" as bullish and buys into a drop, the false-signal loss that defines this game.

In both, the captured edge is the move inside the jump window minus the widened spread you crossed, positive only if you were early and correct. Interpretation latency is the bottleneck on the unscheduled side; network latency on the scheduled side.
π  ≈  ΔPjump⋅1[ early  ∧  correct ]  −  swide\pi \;\approx\; \Delta P_{\text{jump}}\cdot \mathbf{1}[\,\text{early}\;\wedge\;\text{correct}\,] \;-\; s_{\text{wide}}

This page sits within the event-trading topic; the live event/duration toy (IX-DURATION) lives on irregular time, and this page is diagram-only. Numbers are synthetic and illustrative. Real surprise-to-move mappings, latency windows and false-signal rates must be measured per release/feed and dated. Educational only, not investment advice.

Where this fits

↑ Up · building block of Directional event trading ↑ Up · building block of Trading strategies ↔ Across · composes with News trading ↔ Across · composes with Irregular time & point processes ↔ Across · composes with Circuit breakers → Apply · makes money in Prediction markets (Polymarket) → Apply · makes money in Equities & futures ⤓ Build / Buy · tool needed Datasets & tools