Directional event trading
◆still alphaTake a directional position around an event (release, earnings, breaking news) before the price fully reprices. The slow-drift edge is gone; the surviving edge is latency-to-react and correct interpretation.
The idea
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 event trading, precisely?
Event trading is taking a directional position to profit from the price move an event causes. An event is any discrete arrival of information (a data release, earnings, an FOMC decision, a headline) that forces the market to reprice. You either pre-position for an expected move or react fastest to the actual one. The bet is on direction, not on a spread reverting.
Intuition first. Most of the time price wanders on incremental order flow. Occasionally a chunk of information arrives all at once (non-farm payrolls, a rate decision, a merger headline) and the "right" price moves discontinuously. Event trading is positioning to be on the correct side of that jump, or to react to it before everyone else has finished.
The contrast with the other primary families is the cleanest way to place it. Statistical arbitrage bets a relationship reverts and stays market-neutral; market making earns the spread while staying roughly flat. Event trading deliberately takes directional risk for a short, information-rich window. It is the one primary family where you want a view on which way the price goes. Why HFT cares: the repricing is fast and the window to act is short, so whoever parses the information and submits the correct order first captures the move before the price has fully adjusted. That makes event trading a latency-and-interpretation problem, the natural home of machine-readable news and, increasingly, NLP/LLMs.
What counts as an "event"?
An event is any discrete information arrival that forces a repricing. Two broad kinds. Scheduled: you know when, not what (economic releases, earnings, central-bank decisions). And unscheduled: you know neither (breaking news, headlines, shocks). They are traded differently: scheduled events let you pre-position and pre-build parsers; unscheduled events are pure latency-to-react.
Scheduled events have a public calendar; the content is the surprise. Non-farm payrolls, CPI, FOMC rate decisions, GDP, company earnings, index rebalances. You know the exact timestamp, so you can pre-position, pre-stage orders, and pre-build the parser that turns the number into a trade in microseconds. The game is reacting to the surprise (actual vs consensus) faster and more correctly than the field; the full treatment is on scheduled vs unscheduled events. Unscheduled events (a headline, a sudden block trade, a geopolitical shock) give you neither timing nor content in advance, so there is no pre-positioning, only the fastest correct reaction as the news breaks. This is where machine-readable feeds and NLP earn their keep: news trading.
The third member, event arbitrage, is structural: taking the other side of a predictable repricing (an index/ETF rebalance, a merger spread, a known cash-flow event) where the "event" is a scheduled mechanical flow rather than new information. Treatment on event arbitrage. Across all three, trades arrive in bursts around events, not on the clock; the statistical signature is event clustering, modelled as a point process (Engle & Russell ACD, 1998). See irregular time.
How does the repricing actually happen?
At the event timestamp, informed and automated participants update their fair value and act simultaneously. Resting quotes that are now stale get picked off; market makers widen or pull; the order book thins; and the price jumps to the new level in milliseconds. The slow "drift" afterwards (once a reliable edge) is now largely arbitraged in liquid markets.
T−: the run-up. Liquidity often thins before a scheduled release: market makers widen or pull quotes to avoid being run over by the surprise, which you can see as a pre-release spread widening. Some participants pre-position on a forecast; that is a prediction bet, not an event-reaction bet. T=0: the jump. The number prints, fair value moves discontinuously, and the fastest correct responders lift or hit the now-stale resting liquidity before market makers can reprice. This is the brief window where the move is captured, and where latency arbitrage overlaps, picking off stale quotes across venues that update at different speeds.
T+: the drift (mostly gone). Classically, prices kept drifting in the surprise's direction for minutes (post-earnings-announcement drift, per Bernard & Thomas, 1989, and post-release momentum), giving slower traders an edge. In liquid 2026 markets that drift is largely arbitraged away; the repricing completes in milliseconds. Residual drift survives only in less efficient corners: small caps, illiquid names, some crypto and prediction markets. The honest framing: the edge has migrated from the drift (slow, now dead) to the jump (fast, latency-and-interpretation gated). If you cannot react in milliseconds with a correct read, you are the liquidity, not the predator.
What's in the family? The map of event strategies
The event family splits by what you know in advance and what you're trading. Scheduled-vs-unscheduled covers the timing axis; news trading covers reacting to machine-readable information; event arbitrage covers taking the other side of predictable, structural repricings. All share the core: position around a discrete information/flow event to capture the move it causes.
Scheduled vs unscheduled events trades the surprise in a calendar release versus a breaking shock: pre-positioning plus parser speed for the scheduled, pure reaction speed for the unscheduled; live, latency-gated, drift mostly gone. News trading trades machine-readable news (elementized feeds, NLP/sentiment, increasingly LLMs) where the edge is in correct speed, not raw speed; live but crowded. Event arbitrage trades predictable structural repricings (index/ETF rebalances and merger spreads) by modelling the forced flow or the deal odds; merger arb is structural, rebalance arb crowded.
The unifying picture is that each member positions around a discrete event, but the certainty differs, from the fully-known mechanical flow of an index rebalance (event arb) to the fully-unknown breaking headline (unscheduled news). The more predictable the event, the more it resembles arbitrage; the less predictable, the more it is a pure speed race. Prediction markets are a special case: their entire substrate is events. A prediction-market contract pays out on whether an event happens, so every contract is an event trade and the whole family ports directly onto them. See event arbitrage for cross-asset event positioning.
Is event trading still profitable in 2026?
Yes, but the edge has narrowed to speed and correctness. Slow post-event drift in liquid markets is arbitraged away. What survives: being among the first to react correctly to scheduled releases and breaking news (a latency + interpretation problem), structural event arbitrage (mergers, rebalances), and event trading in less-efficient venues such as small caps, crypto and prediction markets.
Dead / commoditised: the slow drift trade (read the release a minute late and ride the residual move) in liquid large caps and major FX/rates. The jump completes in milliseconds. Live, for the equipped: millisecond reaction to scheduled releases with a pre-built parser; machine-readable news trading with a correct read; event arbitrage on rebalances and mergers; and event trading on prediction markets and thinner crypto venues where repricing is slower.
What AI changes: NLP/LLMs genuinely help interpret unstructured news faster and more correctly (the interpretation half of the edge), but they add latency (inference time) and false-signal risk, and they commoditise the easy reads. See news trading and what AI changes for HFT. For the brand-level verdict see is HFT still profitable in 2026.
Worked example
A schematic scheduled-event trade, as of 2026 (synthetic and illustrative). A rate decision is due at 14:00:00.000. Consensus is "hold"; the surprise is a 25bp cut. Pre-event: from 13:59:55 liquidity thins, the best-bid/ask depth halves and the spread widens from 1 tick to 3 ticks as market makers de-risk.
T=0 (14:00:00.012): the decision prints. The "correct" price is about higher. The fastest correct responders lift the now-stale asks in the first ms, capturing the move before market makers reprice. A reactor at 14:00:00.300 finds the price already at the new level, too late for the jump.
Captured edge: approximately the move you caught in the jump window minus the widened spread you paid to take it. React in 50 ms and you bank most of the ; react in 200 ms and the jump is done, so you captured nothing and paid the wide spread, a loss. The whole P&L lives in being fast and correct. The live news/event toys for this family sit on the concept pages (IX-DURATION on irregular time); this guide is diagram-only. Real jump sizes, latency windows and residual drift must be measured per instrument and event type and dated; these figures are synthetic, educational only, and not investment advice.