How Grid Trading Bots Work — The Strategy Behind V3.5, V3.7, and V3.8

Published April 7, 2026 · 8 min read

Three of the four live CoinClaw bots use grid trading. V3.5 Grid, V3.7 Narrow Scalper, and the newly approved V3.8 ETH Grid all place orders on a grid — but they do it differently. Here's how the strategy works and what makes each bot unique.

Grid Trading in 60 Seconds

Imagine the current price of BTC is $68,000. A grid bot places:

• Buy orders below the current price: $67,500, $67,000, $66,500...
• Sell orders above the current price: $68,500, $69,000, $69,500...

When price drops to $67,500, the bot buys. When price bounces back to $68,500, the bot sells. That's one completed grid trade — profit equals the grid step ($1,000) minus fees.

The bot doesn't predict direction. It profits from oscillation. As long as price moves up and down within the grid range, the bot captures small profits on each swing.

When Grid Trading Works

• Ranging markets — price bouncing between support and resistance. This is the sweet spot.
• High volatility within a range — more oscillation means more completed trades.

When Grid Trading Fails

• Strong trends — if price drops straight through all buy levels, the bot accumulates a losing position with no sells to offset it.
• Low volatility — if price barely moves, no grid levels get hit and the bot sits idle.
• Flash crashes — a sudden 20% drop fills all buy orders at once, creating a large underwater position.

V3.5 Grid — The Original

V3.5 is the simplest grid bot. It uses ATR (Average True Range) to set the grid step size — when volatility is high, the grid levels spread out; when volatility is low, they tighten. The grid covers ±12% around the current price with 10 levels on each side.

The problem: V3.5 failed Gate 1 with p = 0.938. That means there's a 93.8% chance its returns are indistinguishable from random. It's been running with real money since early March but has generated zero realized PnL — the grid levels haven't been hit because price has been trending rather than ranging.

Auto-recentre: When price drifts more than 8% from the grid midpoint, V3.5 cancels all orders and rebuilds the grid around the new price. This has triggered multiple times, which means the bot keeps resetting rather than completing trades.

V3.7 Narrow Scalper — Tighter Grid, More Trades

V3.7 takes the opposite approach from V3.5: instead of a wide grid with large steps, it uses a narrow grid with small steps. The idea is to capture more frequent, smaller profits from minor price oscillations.

The trade-off: Narrow grids complete more trades in ranging markets, but they're more vulnerable to trends. A small directional move can push price outside the entire grid range, leaving the bot with an underwater position and no sell orders to recover.

Like V3.5, V3.7 has zero realized PnL so far. Both bots were recently reset after a ccxt compatibility fix (see our behind-the-scenes article), so they're starting fresh.

V3.8 ETH Grid — The Validated One

V3.8 is the most sophisticated grid bot in the competition. Two key differences from V3.5 and V3.7:

1. Regime Filtering

V3.8 detects the current market regime — bull, bear, or sideways — and adjusts its behavior accordingly. In a bull regime (where it performs best, Sharpe = +0.218), it trades actively. In bear or sideways regimes, it reduces position size or pauses entirely.

This is the critical innovation. V3.5 and V3.7 trade the same way regardless of market conditions. V3.8 adapts. Grid trading works best in ranging markets, so a regime filter that avoids strong trends should reduce the biggest risk factor.

2. ETH Instead of BTC

V3.8 trades ETH/USDT, not BTC. ETH has different volatility characteristics — it tends to have larger percentage swings than BTC, which means more grid levels get hit per cycle. The trade-off is higher risk during drawdowns.

Why V3.8 Passed Validation

The validation numbers tell the story:

• p = 0.003 — only 0.3% chance the returns are random. V3.5 scored 0.938 (basically random).
• WFE = 2.559 — out-of-sample performance was 2.5x better than in-sample. This suggests the strategy captures a real market dynamic, not just overfitting to historical data.
• Bull Sharpe = +0.218 — positive risk-adjusted returns in bull markets. The regime filter correctly identifies when to trade.

The Competition So Far

Here's how the grid bots stack up against the non-grid strategies:

Grid trading dominates the competition — 5 of 6 active bots use some form of it. The only non-grid strategy (BTC Trend, an EMA crossover) is the worst performer at -$97.18. That doesn't prove grid trading is better — it might just mean this particular trend-following implementation doesn't work in the current market.

Bottom Line

Grid trading is simple in concept but the details matter. V3.5's wide grid hasn't completed a single trade. V3.7's narrow grid hasn't either. V3.8's regime filter is the key differentiator — it tries to solve the fundamental problem with grid trading (it fails in trends) by detecting trends and stepping aside.

Whether that works with real money is the question. V3.8 is deploying now. We'll have data soon.