Volume Persistence
Volume persistence measures how consistently a prediction market sustains trading activity across time. Instead of focusing on one-off spikes, it asks whether a market continues to see volume day after day (or hour after hour), indicating sustained attention, participation, and usable liquidity for analysis.
In prediction markets, persistence is often a quality filter: two markets can have the same total volume, but one may be “bursty” (a single news spike) while the other is continuously active.
Why it matters in prediction markets
Volume persistence is useful because it helps you interpret probabilities and price moves with context:
- Signal reliability: Persistent volume suggests ongoing participation, making probability updates more interpretable and less likely to be thin-market noise.
- Execution realism: If you’re trading (or simulating trading), persistent volume usually implies better odds of getting fills with lower slippage over time.
- Market health monitoring: Persistence can identify which markets remain “alive” throughout an event lifecycle versus those that go dormant.
- Comparability across events: Persistence normalizes “attention shape” so you can compare markets that peak at different times.
Volume persistence vs. total volume
- Total volume answers: How much traded?
- Volume persistence answers: How steadily did it trade across the time window?
A market with $1M total volume concentrated in 30 minutes may be less useful for ongoing monitoring than a market with $300k spread steadily across days.
Common ways to measure volume persistence
There is no single universal definition; teams typically implement a metric aligned to their use case. Common approaches include:
- Non-zero activity rate (coverage):
- Share of time buckets with volume > 0.
- Example: % of hours in the last 7 days with at least one trade.
- Rolling-window stability:
- Compare volume across rolling windows (e.g., today vs. yesterday) using correlation or ratio constraints.
- Decay / half-life models:
- Apply exponential decay to past volume so recent consistency is weighted more.
- A “half-life” parameter controls how quickly old activity stops mattering.
- Burstiness / concentration penalty:
- Penalize markets where most volume occurs in a small fraction of buckets.
- Implementations often use entropy-style measures or a concentration index.
- Participation-adjusted persistence (when available):
- Pair volume with participant counts or active trader metrics to reduce the impact of a few large trades.
How to use it (practical applications)
- Filtering for analytics: Only compute higher-level indicators (efficiency, reaction metrics, etc.) on markets above a persistence threshold.
- Detecting “dead” markets: Drop or down-rank markets whose persistence collapses after an initial spike.
- Separating information from noise: Combine persistence with probability movement; persistent volume + sustained probability change is more likely to reflect a true information signal.
- Benchmarking liquidity conditions: A market can show tight spreads briefly, but persistence helps identify whether that liquidity is durable.
Pitfalls and interpretation notes
- News-driven events are naturally bursty: Low persistence does not automatically mean “bad,” especially for binary events with a single decisive catalyst.
- Wash trading and incentive programs: Persistence can be artificially inflated by incentives or manipulative patterns; validate with additional checks.
- Time-of-day effects: Many markets have predictable activity cycles; choose bucket sizes that match your interpretation.
Volume Persistence and FinFeedAPI
FinFeedAPI’s Prediction Markets API provides time-stamped prediction market price/probability and activity data you can use to compute volume persistence over any window (hourly, daily, or event-to-date). Typical workflows:
- Pull market-level trades/volume series
- Bucket into consistent intervals
- Compute persistence metrics (coverage, decay-weighted consistency, concentration)
- Use persistence as a filter or feature in market monitoring and modeling