How Our Scanner Detects Both Pumps and Dumps Across 2,300 Coins

Most crypto volume scanners only look for one thing: coins going up fast. They detect a pump, buy in, and hope to ride the wave. The problem? More than half of all volume spikes happen during crashes, not rallies. By ignoring dumps, traditional pump bots miss the biggest opportunities in crypto.

We built DeepAlpha's scanner to trade both directions. When it detects a pump, it goes LONG. When it detects a dump, it goes SHORT. And the results from our 3-month backtest across 287 coins speak for themselves.

The Problem with Traditional Pump Bots

Traditional pump scanners work like this: detect volume spike, buy, set take profit, wait. This approach has a fundamental flaw that we discovered after backtesting across 3 months of data on 287 coins:

Pump-only strategies have a profit factor below 1.0 on longer timeframes. They work for a few weeks, then the losses accumulate. Why? Because by the time you detect the pump and enter, the price has already moved. You are buying near the top.

Our Approach: Bi-Directional Volume Scanner

Instead of only looking for coins going up with high volume, our scanner looks for any unusual volume activity and determines the direction:

LONG (Pump Detection)

• Volume spike 5x+ above 20-period average
• Small candle body (volume arrives BEFORE price moves)
• RSI below 75 (not exhausted yet)
• Price has not already pumped more than 5% in the last hour

The key insight: we enter when volume spikes but price has barely moved. This means someone is accumulating before the move happens. We get in early, not late.

SHORT (Dump Detection)

• Volume spike 3x+ above average
• Price already dropping (24h change below -3%)
• RSI below 45 (bearish momentum confirmed)
• Red candle with negative 1-hour change

Dumps are more predictable than pumps. When panic selling starts with high volume, the momentum tends to continue. Our SHORT model captures this cascading effect.

The Numbers: 3-Month Backtest

We backtested the combined LONG + SHORT scanner on 287 coins with 3 months of hourly data, testing thousands of parameter combinations on GPU infrastructure:

The SHORT (dump) scanner generated 15x more profit than the LONG (pump) scanner. Crashes are more predictable and more profitable to trade than rallies.

GPU-Optimized Parameters

Every parameter in our scanner is data-driven, not guessed. We tested thousands of combinations of stop-loss, take-profit, hold time, and trailing stop configurations using GPU-accelerated backtesting:

• LONG: SL 5%, TP 4%, max hold 2 hours, trailing after +1%
• SHORT: SL 5%, TP 4%, max hold 4 hours, trailing after +1%

The optimal hold time is surprisingly short. This is not swing trading: it is precision scalping on volume anomalies, entering and exiting within hours.

Why SHORT Outperforms LONG

In our testing, the SHORT scanner contributed $19,161 versus $1,268 for the LONG scanner. There are structural reasons for this asymmetry:

1. Panic is more predictable than greed. When a coin starts dumping with high volume, the selling cascade tends to continue as stop-losses trigger more selling.
2. Dumps happen faster. A coin might take days to pump 20%, but it can drop 10% in an hour. Faster moves mean faster profits for the scanner.
3. Less competition. Most retail traders and bots only go LONG. The SHORT side has less crowding and better fills.

Risk Management

Volume scanning is inherently risky. Not every volume spike leads to a tradeable move. Our risk management includes:

• Maximum 2 scanner positions at any time
• Separate from AI positions (does not affect the core strategy)
• Strict SL on every trade (5%)
• Maximum hold time enforced (2-4 hours)
• Cooldown after losses (no revenge trading)
• Daily loss limit shared with the main bot