How Aftershock Measures Market Reactions

An event study compares how a group of related stocks moved around a specific date to how the overall market moved on the same days. The difference, the sector's move beyond the market, is what we call the abnormal return. Adding those abnormal returns across a short window gives the cumulative abnormal return (CAR): the piece of the sector's move that isn't explained by the broad market.

This matters because on any given week, everything moves together to some degree. Isolating the event's effect stops us from mistaking a rising tide for a real reaction to news.

Markets react to information, not to press releases. By the time an announcement is issued the price has usually already moved, sometimes days before. If we measured from the announcement date we would systematically understate the reaction.

Example: for the Strait of Hormuz closure, the formal announcement was on 4 March 2026. But vessel-tracking anomalies were public on 26 February and wire reports of restrictions circulated on 28 February. Oil, tanker and airline stocks had already repriced by 1 March. Anchoring measurement to 28 February, the information date, recovers the full reaction; anchoring to 4 March would miss most of it.

Every sector basket has its own normal weekly swing. A move only counts as statistically significant when it sits clearly outside that basket's usual range, big enough that we can't reasonably explain it as ordinary week-to-week variation.

When a move isn't significant, Aftershock greys it out and tags it "not significant". The direction may still be interesting, but you shouldn't rely on it. This is why our reports often highlight one clear signal in a sea of muted moves rather than pretending every reaction is meaningful.

Direction is only half the story. A sector can end the week roughly flat and still have traded wildly along the way, and a rising VIX tells you the whole market got scared even when the tape looks calm on close. Aftershock reports two volatility signals alongside the directional move:

• VIX (market-wide fear). The change in the benchmark option-implied volatility index around the event window. A sharp rise means investors were paying up for protection across the whole market.
• Realized sector volatility. The ratio of a sector's actual daily-price swings after the event to its swings before. A ratio well above 1× means the sector's price became noticeably more erratic, even if the direction ended up muted.

Volatility answers a different question than "which way did it move", and both matter when sizing a decision under uncertainty.

Every event moves through three states as market data accumulates:

• Breaking. The event just happened. There is no measurable reaction yet, so we show what similar past events actually did, clearly labelled as history, not a forecast.
• Developing. A partial window of price data exists. Reactions are shown with a "provisional" flag on the significance test.
• Settled. The full 30-day window is complete. The measured event now joins the precedent library and is used to inform the next similar shock.

Multiple things happen every week. When a Fed decision, an earnings surprise and a geopolitical shock all land in the same window, a market move can't cleanly be pinned to one of them. Where relevant, reports flag confounding events so you can weigh the attribution yourself instead of assuming a single cause.

Every settled event joins a library of precedents. When a new breaking event arrives, the engine looks up structurally similar past events and reports what actually happened in those windows, with the same significance tests applied to both. This is why a breaking report can show numbers on day one, none of them predictions, all of them measured history.

A precedent is only kept if it passes its own measurement. When a historical parallel fails the same statistical test we apply to live events, it is discarded rather than published. What you see in a precedent list is the surviving evidence, not everything the engine considered.

Aftershock measures the past. It never predicts the future. When we don't have enough data, too few comparable events, too little price history, sources that disagree, we say so, in the confidence footer of each report.

The tool exists to inform decisions, not to make them. It does not give investment advice.

Building Aftershock surfaced three failure modes that were easy to miss on paper:

• Announcement-vs-information dates. Early iterations measured from official announcement dates and consistently under-reported reactions. Switching to the information date recovered signals that had already played out in prices.
• Benchmark contamination. When an event is big enough to move the whole market, the "market" benchmark itself contains part of the reaction. We adjust the benchmark composition for events with wide macro impact to avoid subtracting the effect from itself.
• Matching the basket to the event. A generic "energy" basket misses the point of a shipping-route disruption. Every event type is paired with baskets whose exposures actually correspond to the shock, tanker operators for chokepoints, foundry-exposed semis for export controls, and so on.