Event · Expectations, surprises, and execution timing
Study post-event returns using advance calendars and control groups.
Observation
Volatility around unlocks, listings, upgrades, or macro releases does not mean the event caused the entire move. It may have been expected, or the whole market may already have been moving together. Distinguish scheduled from unexpected events.
Mechanism
The trade concerns actual information relative to prior expectations and forced rebalancing demand. If fully priced, merely “the event happened” need not create an edge. Unrealized events, delays, and cancellations belong in the sample too.
Hypothesis
Use a calendar published beforehand, defining pre/post windows and abnormal-return baselines. Researching surprise requires traceable advance expectations. Without them, study event windows only; hindsight “better than expected” interpretations cannot become signals.
Data
Save times of the initial announcement, every amendment, actual occurrence, and data arrival, using then-current versions. Include failures and cancellations. Flag overlapping events and same-day systemic shocks rather than enlarging effects by selecting clean winners.
Signal
Scheduled events may signal within a known window; unexpected events signal only after the news arrives. Check quote freshness and executability for every signal. With delayed news, do not backfill ideal entries; stop prices cannot erase gap losses.
Backtest
Compare control assets with similar regimes and risk, or calculate abnormal returns from factors estimated beforehand. Report event counts, not merely minute rows: minutes within one event are not independent events. Sample uncertainty intervals by events or correlated event clusters.
Falsification
Shift event dates to produce placebo windows and check whether the result captures ordinary trends. Add actual publication delays and worse fills. An effect found only in hindsight-selected positive events, or also in controls, lacks sufficient event-specific explanation.
Exercise and project acceptance
Create a teaching calendar containing one on-time, one delayed, and one canceled event. Freeze a common rule and work through all three. Submit visibility times, signals, unfilled orders, abnormal returns, and counterexamples per event. Choosing not to trade must be allowed; do not invent counterparties to fill the sample.
Return to the research protocol and register approvals under institutional governance.