Grid Trading Strategy

The choices and what each one trades off

A grid bot looks like a few boxes to fill in, but each box changes what the bot earns and how it fails. Here they are one at a time.

The range

The range is the lowest and highest price the grid covers. A narrow range packs the lines close, so each price move completes more grids and the profit per move is larger. But price needs to fall only a little to leave through the bottom, where the bot is holding the asset at a loss. A wide range survives bigger moves, but the same money is spread over a larger area, so less of it is working near the current price.

The number of grids

More grids inside the same range means smaller spacing. Each completed grid earns less but there are more of them. Past a point the spacing falls below the fee on a round trip and every cycle loses, as grid bot fees and costs shows. Fewer grids means bigger spacing, fewer completions and more money per order.

Spacing type

Arithmetic spacing keeps the price gap the same. Geometric spacing keeps the percent gap the same. In a wide range, an arithmetic grid has a larger percent gap at the bottom than at the top. Whether that suits an asset depends on how it behaves, and which type exists depends on the bot.

Investment size

The amount invested, divided by the number of grids, is the money behind each order. More money per grid means more profit per completed grid and more loss per unit of price fall. It also decides how much is exposed when price leaves the range.

Stops and trailing options

Some platforms let a bot close all positions if price falls to a chosen stop level, or shift the range up as price rises (trailing). A stop turns an open-ended loss into a defined one, at the cost of locking it in and leaving the bot out if price then recovers. Trailing keeps the bot in the action when price rises, at the cost of selling lower-priced units earlier and buying back at higher ones. What exists varies by platform and no option removes the risk.

Trend filters

A filter pauses new orders when a rule says the market is trending, for example when price is far from a moving average. It can reduce the pile-up in a strong move. It also pauses the bot at times when price would have returned, and its rule is another setting that can be over-tuned.

Three made-up settings, side by side

The same 3,000 of capital, an asset at 100, and three settings. All numbers are a made-up example. Capital per grid is 3,000 divided by the number of grids. Spacing is the range divided by the number of grids. The last two columns are how far price can fall or rise from 100 before it leaves the range.

Setting A leaves the range after a 5 percent fall; B and C after 20 percent. A and B have the same spacing, but A puts four times the money behind each order, so it earns four times as much per completion and has a quarter of the room. C has the same room as B and half the spacing, so it completes more often for less each time. Calculated by us on 27 Oct 2026, exact.

What each setting earns per completed grid

Now take a made-up round trip fee of 0.2 percent. Net percent is spacing percent minus 0.2. Net money is capital per grid times that percent.

Setting C is the cautionary one. Its spacing is 1.00 percent, a fifth of which is eaten by the fee, and each grid has only 75.00 behind it. Per completed grid it earns 0.60, against 10.80 for A. It would need many more completions to catch up, and that depends entirely on how price moves.

How to compare settings honestly

The temptation is to try many settings on one stretch of price history and pick the one with the best curve. That finds the setting that suited that stretch, not one that suits the future. This is overfitting, covered in the overfitting guide.

• Test every setting on several assets and several periods, including periods when price trended through the range.
• Include fees, spread and funding on every fill, not only the first.
• Record the worst unrealised loss and how long the bot stayed outside the range, not only the final profit.
• Set aside a period the settings were never tuned on, run each setting there once, and accept what comes out.
• Then run the chosen setting forward on live prices with no real money first. The page on forward testing explains how.

Use the grid calculator to check the arithmetic of any setting before it goes into a test. Tested results of our own will appear on this site when they are ready, with the data, dates and costs. None are published yet, so no figure here is a result from a bot.

What this page does not say

It does not say which setting is best, or that any of them will earn. A narrow range is not wrong and a wide range is not safe. They fail in different ways, and a reader who knows how each one fails can test for it.