Pump & Dump Detection Methodology
By Rishabh Narang · Last updated April 16, 2026
How the Pump & Dump Detector signal works, and why it works.
The Problem
Price pumps combined with fragile liquidity, concentrated supply, aggressive perp positioning, and sell-side flow reveal manipulation patterns that are hard to see from spot price alone. Three types of actors commonly drive this:
- Hedge funds buy locked tokens at 80%+ discount, then short perpetual futures to hedge. On a typical 6-month lockup, this generates approximately 40% annualized return. They have no incentive for the price to rise.
- Market makers operating on loan models pump the price to attract retail buyers, then aggressively sell on perpetuals to lock in profit while maintaining inventory-neutral exposure.
- Team members pump their own token through coordinated activity while quietly shorting through perps, profiting from both the pump and the eventual dump.
Industry research confirms the scale: studies of 16,000+ unlock events show 90% result in price declines. Billions in locked token transactions happen annually, all requiring derivatives hedging.
The Signal
The model looks for manipulation phases: setup, markup, distribution, and dump risk. Negative funding during a rally remains a strong signal, but it is now combined with open interest, basis, true taker-flow/CVD, top-holder concentration, liquidity fragility, and DEX flow behavior.
No single condition alone triggers the strongest readings. The signal is designed to require independent confirmation across market structure, positioning, and flow so it does not become a simple momentum or funding-rate leaderboard.
The model also looks for camouflage patterns: holder fragmentation, sudden holder-count inflation, old-wallet-dominated flow, high volume-to-liquidity churn, and visible concentration falling while the token still trades like a controlled market.
How We Score
Each detected coin receives a composite score from 0 to 10 based on multiple signal families:
Funding, open interest, basis, true taker-flow/CVD, liquidation context, and cross-venue breadth.
Holder concentration, dev/insider/bundler/sniper ownership, liquidity depth, volume-to-liquidity churn, and wallet-fragmentation camouflage.
Setup, markup, distribution, and dump-risk phases inferred from price path, sell-side flow, and persistence.
We publish the signal families, validation process, and backtest results, but keep exact weights and thresholds private so teams cannot easily tune behavior to avoid detection.
Backtest Results
The pump & dump score is intentionally selective. It fires only when independent market-structure, derivatives, holder, liquidity, and flow signals align strongly enough to clear coverage and quality gates. Over a 12-month walk-forward replay across 394 coins, every time the signal fired, we tracked what happened next: 86.4% of flagged coins declined within 30 days with a median drop of -24.6%, a 10.6 percentage-point edge over a random coin from the same universe on the same entry dates. At 60 days the hit rate rises to 90.9%. A top-5 equal-weight short basket rebalanced weekly would have returned +1,279% with a 21% max drawdown, largely because the signal spent most days in cash — only activating for genuine extremes.
Key result
of flagged coins declined within 30 days
median price move 60 days after the signal · 90.9% hit rate
equal-weight portfolio return over 12 months, max drawdown −20.5%
Walk-forward replay across 394 coins from April 16, 2025 to February 15, 2026. Every coin is re-scored end-of-day using only data available at that moment — no lookahead. Refreshed monthly; last computed April 16, 2026.
Headline: Every flagged coin
Portfolio simulation — equal-weight basket, how a user would trade it
Every 7 days, capital is split evenly across the top 5 highest-scored flagged coins (score ≥ 4) and held short until the next rebalance. Positions overlap and compound. This is the most realistic “could I trade this?” view — it caps positions at what real capital allows and handles concurrent signals naturally.
Top-5 equal-weight short basket, rebalanced every 7 days
Final: $1378.71 (+1278.7%)Max DD: −20.5%
Starts at $100. At every rebalance (every 7 days), the portfolio is evenly divided across the 5 highest-scored flagged coins (score ≥ 4). If fewer than 5 coins are flagged, the basket is smaller. Portfolio compounds period-over-period. No fees or slippage. Illustrative — not a recommendation.
Signal precision — every crossing as one independent trade
Each step in the curve below is the P&L of a single $100 notional short held for 30 days, entered the day after the score crossed the threshold. Non-compounding — this isolates the edge of the signal itself, independent of any capital-allocation strategy.
Equity curve — short $100 notional per signal, 30-day horizon
Final: $559.96 (+460.0%)
Simulates a naive strategy: on every signal event, enter a $100 notional short at the next-day close and exit 30 days later. No fees or slippage. Illustrative — not a recommendation.
Case studies — the calls that played out
The five largest drops among flagged coins and the five times the signal went against us. Both are shown in full — nothing cherry-picked.
Highest-conviction wins (30-day horizon)
| Coin | Entry date | Score | Avg funding (72h) | 30d return |
|---|---|---|---|---|
| DashDASH | Nov 5, 2025 | 7.2 | -0.077% | -59.8% |
| AptosAPT | Oct 7, 2025 | 4.9 | -0.052% | -49.1% |
| DashDASH | Jan 18, 2026 | 7.0 | -0.272% | -48.9% |
| AlgorandALGO | Jan 7, 2026 | 4.4 | -0.321% | -37.7% |
| My Neighbor AliceALICE | Oct 10, 2025 | 4.2 | -0.099% | -33.6% |
Calls that went against us
| Coin | Entry date | Score | Avg funding (72h) | 30d return |
|---|---|---|---|---|
| ZcashZEC | Oct 31, 2025 | 5.6 | -0.056% | +32.9% |
| FusionistACE | Nov 30, 2025 | 4.1 | -0.182% | +11.8% |
| AergoAERGO | Dec 15, 2025 | 4.3 | -0.380% | +1.1% |
| Aerodrome FinanceAERO | Jun 20, 2025 | 4.2 | -0.036% | -2.8% |
| Basic AttentionBAT | Nov 27, 2025 | 4.5 | -0.053% | -5.2% |
Consistency across horizons and score buckets
Hit rate and price move measured at 7, 14, 30, and 60 days, broken down by bucket. The highlighted row is the strongest combination of hit rate and sample size.
| Horizon | Bucket | N | Hit rate | vs baseline | Median price move | Return edge |
|---|---|---|---|---|---|---|
| 7d | Strong (≥7) | 2 | 100.0% | +47.7 pp | -30.3% | -29.6% |
| 7d | Moderate (≥4) | 20 | 70.0% | +2.3 pp | -6.6% | -2.8% |
| 7d | All flagged | 22 | 72.7% | +6.4 pp | -9.2% | -5.5% |
| 14d | Strong (≥7) | 2 | 100.0% | +25.1 pp | -38.8% | -25.9% |
| 14d | Moderate (≥4) | 20 | 85.0% | +16.4 pp | -12.5% | -6.0% |
| 14d | All flagged | 22 | 86.4% | +17.2 pp | -14.9% | -8.5% |
| 30d | Strong (≥7) | 2 | 100.0% | +16.1 pp | -54.3% | -34.0% |
| 30d | Moderate (≥4) | 20 | 85.0% | +10.1 pp | -22.8% | -5.5% |
| 30d | All flagged | 22 | 86.4% | +10.6 pp | -24.6% | -7.4% |
| 60d | Strong (≥7) | 2 | 100.0% | +16.0 pp | -60.5% | -36.1% |
| 60d | Moderate (≥4) | 20 | 90.0% | +7.5 pp | -29.0% | -4.1% |
| 60dbest | All flagged | 22 | 90.9% | +8.3 pp | -31.7% | -6.7% |
How we avoided biasing the results
- Walk-forward replay: at each day D, only data up to D is used to generate signals
- 14-day warm-up: a coin must be below threshold for 14 consecutive days before its first crossing counts (eliminates 'discovery' bias at window start)
- 30-day lockout: after any signal, that coin is suppressed for 30d (prevents overlapping trades)
- Universe baseline uses the same coins at the same entry dates (controls for market beta)
- Includes all coins that traded during the window — delisted coins are NOT dropped (avoids survivorship bias)
Price source: correlation_prices (daily market close). Entry rule: close[signal_day + 1] (realistic next-day execution). Portfolio sim excludes transaction costs — real returns will be lower after fees, borrow, and slippage.
Disclaimer
This is an experimental signal derived from public market, derivatives, and on-chain context. It does not constitute financial advice or an accusation of market manipulation. Always do your own research.
Methodology Questions
What signals does the model use?
The model combines derivatives positioning, open interest, basis, true taker-flow/CVD, price path, holder concentration, liquidity quality, and DEX flow context.
Does the methodology disclose exact weights?
No. We explain the signal families and validation process, but do not publish the exact formula or weights because public disclosure makes the leaderboard easier to game.
How does concentration affect the signal?
High top-holder, insider, dev, bundler, or sniper ownership increases risk because smaller amounts of coordinated trading can move price and accelerate dumps.
What triggers the pump and dump signal?
A coin can be flagged during setup, markup, distribution, or dump-risk phases when multiple independent market-structure and derivatives signals align across sufficient data coverage.
Who are the typical actors behind pump and dump patterns?
Three types of actors drive this pattern: hedge funds buying locked tokens at 80%+ discount and shorting perps to hedge, market makers pumping price to attract retail while selling on perpetuals, and team members pumping their own token while quietly shorting through perps.
How was the signal backtested?
We replayed the signal walk-forward over 12 months across 394 coins using daily market prices. Flagged coins hit 86.4% at 30 days with median -24.6% — a 10.6 percentage-point edge over a random coin from the same universe on the same entry dates. At 60 days the hit rate rises to 90.9%. A top-5 equal-weight short basket rebalanced weekly would have returned +1,279% with a 21% max drawdown, largely because the signal spent most days in cash. A 14-day warm-up eliminates 'discovery' bias at the window start.