Grid strategy handbook

1. The core assumption behind grids

Grid trading in one sentence: split a price band into N rungs, place a paired buy and sell at each rung; when price falls a rung triggers a buy, when price rises a rung triggers a sell, every completed buy-sell cycle prints one rung's worth of spread minus fees.

The whole strategy assumes price comes back. If BTC chops between 60k and 80k, the grid harvests every leg. If BTC breaks 80k and rips to 100k, the grid sells out rung by rung, and at 80k you are flat and watching the move without you — that is the "missed the move" failure mode. If BTC breaks 60k to the downside and runs to 40k, the grid buys all the way down to 60k, then the bot is fully invested with a heavy unrealised loss — that is the "trapped long" failure mode.

The deep structure of a grid is "short volatility". You make money when realised vol stays inside the band and price mean-reverts; you lose money when vol explodes and price never comes back. Once you have that mental model, you know when to run a grid and when to switch it off.

For the textbook angle, Investopedia's grid trading article walks through the FX-market origins, and the shorter Investopedia grid trading definition covers the generic framing you can port to crypto.

2. The OKX grid bot parameter sheet

To create a grid on OKX (Strategy → Strategy Trading → Spot Grid / Futures Grid), you fill in:

Price band. Upper and lower price, defining where the grid runs. Pick 60,000 and 80,000 and the bot only operates inside that band.

Number of grids. How many rungs split the band. 50 grids on a 60k-80k band gives you $400 between rungs (arithmetic). OKX typically recommends 10-100 grids.

Arithmetic vs geometric. Arithmetic keeps a fixed USDT step ($400). Geometric keeps a fixed percentage step (0.5%). More on this in Section 3.

Investment amount. Total USDT you commit. The system splits it across the rungs given the current price.

Leverage toggle (futures only). Perp grids can run 2-5x leverage for amplified yield — and amplified blowup risk.

Before you press Confirm, OKX shows a "Strategy Estimate" — expected per-grid yield, expected annual return, capital required near current price. For a new trader, that summary line tells you immediately whether your parameter combo even makes sense.

3. Arithmetic vs geometric — which fits your pair

Arithmetic grids hold the dollar step constant. On BTC 60k-80k / 50 grids, every rung is $400 apart. Simple, but the percentage step is uneven — 60,000 to 60,400 is +0.67%, while 79,600 to 80,000 is +0.50%. The same $400 means different percentage moves at different price levels.

Geometric grids hold the percentage step constant. Set 0.5% and the rungs become 60,000, 60,300, 60,601.5, ... up to 80,000. Denser at the bottom, sparser at the top, which keeps the "probability of a trigger" roughly even across rungs.

Our desk's rule: BTC and ETH (lower percentage realised vol) — go arithmetic. SOL, DOGE, smaller-cap tokens (higher realised vol) — go geometric. The reason is that on a high-vol asset, a doubling in price is not unusual, and an arithmetic grid up there gets so sparse that the trigger frequency collapses.

4. Setting density — the profit-versus-fee tradeoff

Too dense and per-grid yield gets crushed by fees. OKX spot fees at the default tier are 0.08% taker / 0.06% maker. A round-trip is 0.14%. If your rung step is only 0.2%, gross is 0.2%, fees are 0.14%, net is 0.06%. Any realised-vol drop and the grid turns negative.

Too sparse and the grid does not trigger enough. BTC 60k-80k cut into 10 grids puts $2,000 between rungs. Price needs to walk $2,000 to fire once. If BTC only does a couple of $2,000 round trips in 30 days, your total per-grid output is, well, two trips' worth — a few dozen USDT.

Heuristic from running these in production: per-grid gross should run 3-5x your round-trip fee. With OKX spot at 0.16% taker round-trip, that means rungs at least 0.5-0.8% apart. BTC 60k-80k spans 33%. 33% / 0.5% = 66 grids upper bound; 33% / 0.8% = 41 grids lower bound. So 50 grids is exactly the sensible middle.

SOL 100-200 at 50 grids puts $2 between rungs, which is 1-2% — also reasonable. Push to 100 grids and you are at $1, or 0.5-1% — still tradeable. Push to 200 grids and you are at $0.50, or 0.25-0.5%, and fees take more than half.

5. Worked example: BTC 60k-80k / 50 grids / 30 days

Take a common parameter set and pull the PnL apart layer by layer. Everything below is computed from the fee structure and the formulas given earlier in this article — it is not a record of what any particular bot printed.

Parameters. BTC-USDT spot grid, band 60,000-80,000, arithmetic, 50 grids ($400 per rung), invested 1,000 USDT, no leverage. Assume the bot starts with spot in the bottom third of the band, somewhere near 63k.

Layer one — grid PnL. Capital per rung is 1,000 / 50 = 20 USDT. A $400 rung around the 60-68k area is 0.5-0.67%, so one complete buy-and-sell round trip grosses roughly 0.10 USDT and nets about 0.06 after the 0.04 in fees. Assume 89 triggers over the 30 days (trigger ratio 89 / (50 × 2) ≈ 89%, which is already an optimistic read on a chop month) and cumulative grid PnL lands near 89 × 0.06 ≈ 5.3 USDT — 0.53% on capital.

Layer two — unrealised holding PnL. Whatever the grid bought low is still on the books. If the closing print sits above your average fill, that shows up as an unrealised gain. It moves with the market and evaporates the moment price drops back under your average. The grid did not earn it; direction did.

Stack the two layers and you get the number the strategy page shows you — which is exactly why they have to be read apart. The grid's own contribution is the half-a-percent of vol yield. The rest is the same thing you would have collected by simply holding. Whether a grid is worth running comes down to that difference, not the headline figure, and most of the double-digit monthly screenshots circulating online are quietly crediting the grid with directional PnL.

Run the same parameter set through August 2024 and the conclusion flips. BTC came into the month at $62,500 and printed a session low of $48,830 on 2024-08-05, straight through the 60,000 floor of the band. By the time price hit 60k the grid was already fully invested, with every fill somewhere inside 60,000-80,000. Mark-to-market at the low is at least −18.6% (Section 6 shows where that bound comes from). Even after BTC clawed back into the mid-50s over the following weeks, the grid stayed dormant — it cannot operate below its lower bound.

Two takeaways. First, the PnL difference between a chop month and a trend month is huge. Second, picking the wrong band — putting the lower bound on what you assumed was support and then having the market break support — turns the grid into a "buy more on the way down" machine.

6. How a trend slices a grid — 2024-08-05 replayed

The 2024-08-05 crash is the cleanest "grid sliced by a trend" case of recent years. The week before, BTC chopped tight between 60k and 67k — perfect grid territory. If you had launched a 50-grid BTC spot grid on 60k-70k on 2024-08-01, the trigger rate that week would have been excellent.

On 2024-08-05, carry-trade unwind took BTC from $57,200 at 02:00 UTC to a session low of $48,830 at 06:00 UTC — and the month had opened around $62,500. Your grid filled every rung on the way down and was fully invested before price even reached the 60,000 floor. You do not need to know the average fill to put a floor under the damage: assume the most generous case, that every single rung filled at the very bottom of the band, 60,000. Against the session low that is still (48,830 − 60,000) / 60,000 = −18.6%. The true average fill sits above 60,000, so the real number is worse than that.

The strategy report would log this as "unrealised drawdown", but the real story is "the central assumption broke". The grid says price will come back. On 2024-08-05 week, it did not. BTC needed 21 days to retake $60,000. During those 21 days the grid was 100% allocated at the band floor, with zero new triggers.

Liquidation data for 2024-08-05 is on CoinGlass: roughly $1.05B of derivatives liquidations across the day, about 82% of it longs. Events of this size happen a few times a year, which means every quarter you run a grid you should expect to face at least one slice event.

Two practical countermeasures. One: set the lower band 20-25% below the current price to give yourself headroom. Two: pair the grid with a trend filter — pause it when BTC drops below the 200-day moving average, restart when the trend confirms back. Both cost you a little chop-month yield, but both keep you alive through the tail events.

7. Spot grid vs perp grid — the leverage tradeoff

OKX gives you both. Spot grid is long-only (buy under the band, sell over) with no liquidation risk — worst case is the spot price falls toward zero. Perp grid runs bidirectional (neutral) or one-sided (long-only / short-only) and can layer 2-5x leverage.

Leveraged perp grids look attractive — same capital, theoretically 5x the grid yield. But the slice risk magnifies the same way. The BTC 60k-80k spot grid took at least a −18.6% mark-to-market hit at the 2024-08-05 low. At 5x that is roughly 93% of margin gone — and the maintenance-margin trigger fires well before margin reaches zero, so a 5x perp grid on the same band would already have been closed out at market somewhere on the way down.

Our desk's hard rule: year one is spot grid only. Get a feel for the realised distribution of chop returns vs trend drawdowns. Year two, you can experiment with 2x perp grids (you want at least a 2x buffer above the slice). Anything above 3x is not a beginner instrument.

One more thing on perp grids — beyond the grid PnL you still pay or receive funding. A long-side perp grid pays positive funding every 8 hours. A short-side perp grid pays negative funding. Over 30 days, accumulated funding can be 10-30% of the grid's gross profit.

8. Five checks before you press Start

By this section you should be able to design grid parameters on your own. Before you click Create, run this five-line checklist.

One. Is the 30-day realised vol low to mid? If daily realised vol is over 4%, your slice risk is high — wait it out.

Two. Is the lower band 15-25% below the current price? Headroom for a tail event.

Three. Is per-grid gross at least 3x the round-trip fee? OKX's grid-create screen prints this directly under "single grid yield".

Four. Are you OK losing 30% on this allocation? A drawdown of the 2024-08-05 size is not rare — confirm your risk appetite before committing.

Five. Do you know when you would stop the bot manually? BTC breaking the 200-day MA, the band's lower bound printing, a market-wide event — any one of those should trigger an immediate pause.

All five pass, then create. To pre-size the trade, open the grid PnL budget tool. To pair this with the directional logic, the stop-loss / position-size paradox piece contrasts grid logic (no stop unless you set one) with directional logic ("hard stop") side by side.