A grid trading bot buys low and sells high inside a set range, over and over. We ran the same grid on BTC during a 60-day range, a 60-day fall and a 60-day rally to see exactly where it earns and where it loses.
November 16, 2026
14
min to read
The short answer
A grid trading bot places buy and sell orders at fixed price steps across a range, buying small dips and selling small rises. It earns often while the price stays inside its range, and falls short when the range breaks: selling out to cash and missing a rally, or buying all the way down into a loss it still holds.
A grid trading bot splits a price range into evenly spaced lines and places a standing order at each one: buy at the lines below the current price, sell at the lines above it. Whichever line the price touches next, it fills, and the bot immediately arms the opposite order one line away, ready to fill again the next time the price crosses it. It never asks whether the asset is a good buy. It only asks where the price sits inside its grid. That makes it one of the common kinds in our guide to trading bots, and the opposite of dollar-cost averaging, which buys on a calendar whatever the price.
That makes it a bet on movement, not direction: a grid earns from a price that keeps crossing its lines back and forth, whichever way it is headed on any single day. The trade-off is the one this guide backtests: a grid has no view on whether the price will still be inside its range next week, and when it is not, the same mechanism that earned small, steady amounts stops helping.
| Term | What it means |
|---|---|
| Grid line | One price level in the ladder, set when the grid is built. |
| Slot | The money and, once bought, the coins sitting between two adjacent lines. |
| Range | The highest and lowest line; set from recent prices, by hand or by a rule. |
| Fill | One buy or one sell at a line. |
| Round trip | A slot’s buy and its matching sell, together. |
Picture ten lines spread evenly across a range, with nine slots of equal money between them. At the start, the slots whose line sits below the current price are empty, waiting to buy; the slots above it hold inventory already, waiting to sell. From there the loop is the same every time: price falls to a line with an empty slot above it, the bot buys; price rises to a line with a filled slot below it, the bot sells and re-arms that slot to buy again if the price comes back down.
Each round trip is small on purpose. The gap between two adjacent lines is a few percent of the price (4.7% in the range test below), so one fill rarely earns much. The bot’s edge, if it has one that day, comes from doing this often, not from any single trade being large.
Hyperliquid’s own order types include a scale order: “multiple limit orders in a set price range,” placed in one action. That covers the first half of a grid, the opening ladder. It does not re-arm itself: once a scale order’s legs fill, they are done, so keeping a grid running, buying low and selling high over and over as price moves back and forth, takes a bot or an agent that watches each fill and places the next order, which is why “grid bot” rather than “grid order” is the usual term. Our guide to Hyperliquid trading bots covers hosted options that do this for you.
A grid is one of several fixed rules traders automate and leave running. Our wider backtest of six automated setups tests the same grid design alongside trend-following, breakout and dip-buying rules over one shared window, if you want to see how it compares with rules that do take a view on direction.
We built one grid and ran it, unchanged, on three 60-day stretches of BTC chosen to put it in a different kind of market each time:
This is a long-only, fully cash-backed simulation that holds the coins themselves. There is no leverage, so no liquidation, and no funding, the hourly payment between a perpetual’s longs and shorts, which applies only to a perp position. Ethy’s own Accumulation Grid template trades spot tokens, which never carry funding either.
Same grid, same costs, three different 60-day stretches of BTC:
| Window | Dates | Grid return | Grid’s worst drop | Buy and hold | Round trips | Still open |
|---|---|---|---|---|---|---|
| Ranging | March 6 to May 4, 2024 | +7.7% | −2.9% | −0.2% | 14 (14 won) | 2 of 9 slots |
| Trended down | October 7 to December 5, 2025 | −21.4% | −25.4% | −28.5% | 13 (13 won) | 9 of 9 slots |
| Trended up | September 12 to November 10, 2024 | +7.4% | −1.7% | +40.3% | 15 (15 won) | 0 of 9 slots |
The grid beat both cash and buy and hold only in the range: up +7.7% while BTC itself moved almost nowhere net. In both directions away from a range, it fell short, though not the same way. Falling, it ended −21.4%, worse than sitting in cash but better than buy and hold’s −28.5%, because it started in cash and bought one line at a time on the way down, not all at the top. Rallying, it made a small +7.4% and then stopped, while buy and hold made +40.3%: the grid ran out of grid.
From March 6, 2024 to May 4, 2024, BTC opened at $63,918 and closed the window at almost the same price, −0.2% for buy and hold, while swinging through a much wider band along the way. That gap, between a flat net move and a wide range, is exactly what a grid is built to collect: it closed fourteen round trips, every one of them a winner, averaging +4.88% each, for a total of +7.7% with only a −2.9% drop along the way, far gentler than BTC’s own swings.
Its first round trip bought at $63,358 and sold at $66,376 the same day, March 6, 2024, for +4.55%; its last bought at $60,341 on April 30 and sold at $63,358 on May 4, 2024. Two of its nine slots were still holding bought coins when the window ended, worth −1.6% each on average: a reminder that even in its best case, a grid can end a window holding a small loss on whatever it bought last.
From October 7, 2025 to December 5, 2025, BTC opened the window at $124,711, near its peak for the window, and closed at $89,294, down 28.5%. None of the grid’s nine slots started bought, since the price opened at the very top of its own range. As BTC fell, slot after slot bought in behind it, and thirteen of those buys even closed for a small gain before the fall resumed, averaging only +1.57% each, a fraction of the ranging window’s average round trip, because the band each line covers is much narrower next to BTC’s price at these levels.
What matters more is what did not sell. By the close, all 9 slots were still holding BTC bought on the way down, worth −23.2% each on average, as deep as −28.2% on the one bought highest. Every closed round trip in this window was a winner, which would look, from the trade log alone, like a strategy with no losing trades at all. The account was still down 21.4%, because a 100% win rate on the trades that did sell says nothing about the inventory still sitting unsold.
From September 12, 2024 to November 10, 2024, BTC opened at $57,287 and closed the window at $80,458, up 40.3%. This time the grid’s trouble was not holding losers; it was running out of sellers. Price pushed through the grid’s top line on October 14, just past the window’s halfway point, and from then on every slot was empty and every line was below the market, with no upper line left to sell into and no reason for price to fall back to a buy line. The grid finished the window exactly as it started that day: fully in cash, earning nothing more while BTC kept climbing.
By the close it had made +7.4% from fifteen completed round trips, all winners, averaging +4.30% each. Buy and hold made +40.3%. The grid was not wrong about any single trade; it simply had nothing left to do for the back half of a rally it had no way to see coming.
Two settings decide most of what a grid does: how many lines it has, and how wide a range those lines cover. We reran all three windows with more and fewer lines, and with the range set from a shorter and a longer stretch of prior prices:
| Setting | Ranging | Trended down | Trended up |
|---|---|---|---|
| 5 lines, 30-day range | +3.2% | −19.9% | +6.8% |
| 10 lines, 30-day range (baseline) | +7.7% | −21.4% | +7.4% |
| 20 lines, 30-day range | +5.4% | −22.4% | +6.6% |
| 10 lines, 14-day range | +9.3% | −21.4% | +3.2% |
| 10 lines, 60-day range | +5.8% | −20.4% | +10.0% |
Going from 5 to 10 lines helped in the range, because the finer grid caught more of the back-and-forth. Going on to 20 lines gave some of that back in every window we tested, in the range and in both trends, because each of its slots held about half as much money and sold only half as far above its buy, so a swing the 10-line grid caught in full paid the 20-line grid about half as much. A finer grid comes out ahead only on swings too small to reach a coarser grid’s next line.
A shorter, 14-day lookback helped in the range: a tighter band that matched the market’s recent, narrower swings caught more round trips. The same short lookback hurt in the rally, because a narrower range gave the grid less room above the price, so it ran out of upper lines and sold out to cash even sooner. A longer, 60-day lookback did the opposite in each case: it gave up some of the range’s extra round trips, but in the rally it gave the grid more room above the start price, so it rode further before running out of lines. Neither setting won in every window. Whatever sets your range, know what it assumes about how far the price might move before it gets there.
Ethy is an AI trading agent you ask before you trade. Its Strategy Library includes an Accumulation Grid template, which trades spot tokens rather than a perpetual contract, so it never carries funding or a liquidation price. It is not set up the same way as the custom grid tested in this guide, so read its exact rules in the preview before you judge it against anything above. Ethy’s grid bot shows how its ladder fills and when it sells.
Ethy is software for research and analysis. It does not give investment advice, and nothing here is a recommendation to run a grid on any particular market. Automated trading bots are also a common subject of overpromising: the CFTC’s customer advisory warns that promises of outsized or sure-thing returns are red flags of fraud, and FINRA warns that some auto-trading services “promote vague and unverified claims of impressive past market returns.” A backtest like this one, dated and showing its losses as plainly as its gains, is the opposite of that kind of claim.
It can, specifically while a market trades inside a range: our backtest made +7.7% on a 60-day BTC range. The same grid lost money when BTC fell hard and trailed far behind buy and hold when it rallied hard, so “does it work” depends entirely on whether the market stays inside the grid’s range.
There is no single best setting across every market. In our tests, 10 lines beat both 5 and 20 in the range; a narrower range caught more trades in a range but ran out of room sooner in a rally. Match the range to how far you think the price could realistically move, and expect to be wrong sometimes.
Yes, when the price falls through the whole grid: every slot ends up holding coins bought on the way down, worth less than they cost. When the price rises through the whole grid, it keeps its gains but falls behind: every slot sells out to cash and earns nothing further while the price keeps climbing without it.
Our tests used 10 lines (9 trading slots) as a baseline and compared 5 and 20. Going from 5 to 10 helped in a range; going on to 20 gave some of that back everywhere we tested, because each of its smaller slots earns only the narrow gap to the next line. Past a point, more lines mean more trades, each earning less.
No. DCA (dollar-cost averaging) buys a fixed amount on a fixed schedule and never sells. A grid buys and sells repeatedly inside a price range on no fixed schedule at all, trading only when the price actually crosses one of its lines.
Ethy’s Strategy Library includes an Accumulation Grid template for spot tokens, which you can preview, backtest and run on paper before it trades for real. It is not configured the same way as the custom grid tested in this guide.
Trading involves risk, including the loss of your money. Ethy AI is a research and trading tool and does not give investment advice or recommendations. Backtests are simulations on historical data and do not predict future results.
Deploy your AI trading agent in one click.
