Alerting on Price Moves Without Generating Noise
An alert that fires constantly is ignored, which makes it worse than no alert. How to define a threshold that reflects significance rather than movement.
The purpose of an alert is to make you look at something. If it fires when nothing matters, you stop looking, and the alert system has made you worse off than having none.
Why naive thresholds fail
"Alert on a five-cent move" sounds reasonable and fails in both directions.
At 50 cents, five cents is an ordinary fluctuation in an active market, often just the spread being crossed in one direction. You will get constant alerts.
At 3 cents, five cents is impossible. You will get none, on exactly the markets where a move is most informative in relative terms.
Better definitions of significance
Relative to the spread. A move of less than the spread is not a move; it is which side traded. Require at least a spread's worth of movement before anything counts. Mid price versus last trade.
Relative to the market's own volatility. Compute how much this market typically moves over the window in question, and alert on unusual moves for this market rather than on a universal number. This automatically handles the difference between a quiet long-dated contract and an active short-dated one.
In log-odds rather than cents. A move from 2 to 4 is a doubling of implied odds; a move from 48 to 50 is nearly nothing. Log-odds treats both correctly with a single threshold, which is exactly what a bounded price scale requires. Percentage points versus per cent.
Sustained rather than instantaneous. Require the new level to persist for a few observations. This removes single-print noise and the bid-ask bounce entirely.
Weighted by liquidity. A five-cent move on a market with no depth is not news. Gate every alert on a minimum book size. Thin market warning signs.
Alerts worth having
Sustained move on a liquid market. The core signal, and the one behind a top movers view. Top movers.
Cross-venue divergence that persists. Not the instantaneous gap, which is mostly timing. A gap that survives several consecutive observations on two fresh, liquid quotes. Persistent versus temporary gaps.
Internal inconsistency. A strike ladder out of order, or a multi-outcome set summing far from one. These are almost always staleness and they are free to check. How multi-outcome markets sum.
Freshness failures. The most important alert in the whole system, and the one people build last. A source that has stopped updating while everything keeps running is the failure that silently corrupts every downstream number. Detecting stale data.
Keeping the volume down
Deduplicate. One alert per market per move, not one per observation.
Cool down. After firing, suppress that market for a period.
Suppress around known events. Everything moves when a figure is published. That is not news; it is the schedule.
Batch the low-priority ones. A digest is fine for anything that does not need a response within minutes.
Track your own precision. Record how often an alert led you to do anything. If the answer is never, the threshold is wrong, and that is a fixable measurement rather than a matter of taste.
Next
Educational content only. Nothing here is financial advice, a solicitation, or an offer to trade. Prediction markets can move against you and a position can lose its entire value. Where this guide describes a difference between venues it describes a theoretical cross-market edge in observed prices, which is not a profit: fees, spreads, capital lockup, transfer time and differing settlement rules all come out of it first.
Keep reading
Data and Automation · 3 min readNormalising Market Data Across VenuesTwo venues describe the same idea with different shapes. A documented intermediate representation is the difference between a comparison and a guess.
Data and Automation · 3 min readBuilding a Market Dashboard That Tells the TruthDesign notes for a page that shows prediction market data without overstating it: which fields to show, what to filter, and what to label loudly.
Data and Automation · 3 min readDetecting Stale Data Before It Reaches AnyoneA pipeline that keeps running while its source stops updating is worse than one that fails loudly. How to measure freshness and what to do when it slips.
Prediction Market Basics · 4 min readYes and No Shares: Two Sides of One PriceYes and No are not two markets. They are two halves of one contract whose prices must sum to a dollar, and that identity explains a surprising amount.