Grid trading: buy at each line down, sell one line up

Grid trading draws price lines a fixed ratio apart, buys one slot each time the price falls through a line, and sells that slot one line above where it was bought. In Korea it is also called tteolsaopal, short for "buy when it falls, sell when it rises". This article states the exact rule this site's backtest uses, how it differs from exchange grid bots and from the infinite buying method, and why it falls behind in rising markets, using QQQ and TQQQ results.

Rules

• Price lines: start at the first day's open (P0) and repeat up and down at the same ratio without end. With 5% spacing, lines sit at P0 × 1.05, P0 × 1.05², P0 ÷ 1.05 and so on, each line 1.05 times the one below (geometric spacing).
• First buy: one slot at the first day's open.
• Buy: each time the price falls through a line, buy one slot if that line's slot is not already held. Nothing is bought on the way up.
• Sell: each slot is sold when the price reaches the line one above where it was bought. Every slot is bought and sold on its own.
• Slot size: a slot is the account value at the time of purchase divided by the slot count, and no more slots than the slot count are held at once. Spacings of 3%, 5% and 10% use 32, 20 and 11 slots, so the grid can keep buying at every line down to about −60% from the first slot. If every slot is full or the cash left is less than half a slot, nothing is bought.
• Fills: a line that is touched fills at the line price. Within a daily bar, an up day is assumed to move open, low, high, close, and a down day open, high, low, close. If the open has already passed a line, the fill is at the open.

Each completed slot earns about 4.7%: the 5% spacing minus the 0.3% round-trip cost. A slot is 1/20 of the account when bought, so each sale adds about 0.2% to the account. A grid's return is set by how many times the price travels between lines, and has little to do with how far the price rises.

How it differs from exchange grid bots

Grid bots on crypto exchanges start with an upper and lower price limit and a number of grids. The grids are split either by equal price differences (arithmetic) or by equal ratios (geometric). Binance's spot grid buys the coins needed for the sell orders above the current price when the bot is created, and places no orders while the price is outside the range. Some bots can also move the range up as the price rises (trailing up).

This article's rule differs in three ways. It has no price range, so it buys again whenever the price falls through a line, even above a new high. It starts with a single slot (5% of the account). Its spacing is geometric. With arithmetic spacing, each grid becomes a smaller percentage as the price rises: lines ₩1,000 apart are 10% apart at ₩10,000 but 5% apart at ₩20,000.

How it differs from the infinite buying method

Laoer's infinite buying method (무한매수법), often mentioned alongside grids in Korean communities, is not a grid. As commonly described, it splits the capital into 40 parts, buys one part near the close every day, and sells the whole position once the price is 10% above the average cost (for TQQQ), then starts over. Details differ by version. A grid lets price lines decide when to buy and sells each slot on its own. The infinite buying method lets the calendar decide when to buy and sells everything at once against one average cost. It is closer to scaling in with a profit target.

The stock share fills only as far as the price has fallen from its high

A grid buys only on the way down and sells on the way up, so the number of slots it holds ends up set by how far the price sits below its high for the period. Each drop of about 5% from the high adds a slot, and if the price keeps rising past the high every slot is sold and only cash is left. That is why it holds almost no stock in a rising market.

Over the last 10 years of QQQ (October 2016 to September 2026), the 5% grid's average stock share was 10%. QQQ was within 5% of its high on 62% of the days, and the stock share averaged 5.2% on those days. The grid sold a slot 112 times in 10 years. Compounding 112 sales that each add about 0.2% gives about 30%, and the actual 10-year return was 31.6%. The same money simply held ended at $66,564.

Backtest results (as of September 2026)

The starting money was $10,000, with nothing added later. Dividends were reinvested on the ex-dividend date, a 0.15% cost was charged on every buy and sell, and cash earned no interest. The last 10 years run from October 2016 to September 2026, and the full QQQ period from January 2000 to September 2026.

Over TQQQ's full period (March 2010 to September 2026), the grid ended at $113,870 (15.8% a year) and buy and hold at $3,482,470 (42.3% a year). The monthly-updated numbers and the 5-year period are in Grid trading vs. holding backtest.

Where it led and where it lagged

Shifting a three-year window one month at a time, the grid ended higher than buy and hold in 62 of 295 QQQ windows (21%) and 17 of 164 TQQQ windows (10%). Every QQQ window where the grid led was one where holding returned 3.2% a year or less, and the grid led in all 51 windows where holding lost money. All of those windows started between 1999 and 2002 or between 2005 and early 2008, the stretches that ran through the dot-com crash and the 2008 financial crisis.

The grid was furthest ahead for the window starting April 2000: over the next three years, holding returned −38.5% a year and the grid −17.4%. It was furthest behind for the window starting January 2019: holding 38.9% a year, the grid 3.7%. On TQQQ, every window where the grid led started between January 2020 and May 2022 and ran through the 2022 decline. The window starting November 2021 returned −2.8% a year for holding and 9.6% for the grid.

If the same average share had simply been kept

To separate the trading method from the low share, the grid was compared with keeping its average stock share fixed over the last 10 years, reset on the first trading day of each month. On QQQ, a fixed 10% returned 2.1% a year and the grid 2.8%. On TQQQ, a fixed 27% returned 14.0% and the grid 16.7%. Most of the gap to buy and hold came from the low share, and the grid did a little better than the same average share. That difference can be read as what the grid gained from the price moving up and down, and it was larger on the more volatile TQQQ.

In exchange, TQQQ's max drawdown was −55.9% for the grid and −31.7% for the fixed 27%. A grid fills up as the price falls, so its share is largest near the bottom. On the 300 days (12% of the 10 years) when TQQQ was 60% or more below its high, the grid's stock share averaged 70%, and it peaked at 79% on March 12, 2020.

Spacing and costs

Wider spacing meant fewer fills and slightly better results. Over the full QQQ period, the annual return was 1.9% at 3% spacing, 2.3% at 5% and 2.5% at 10%, with 1,182, 530 and 168 fills. The 3% grid paid $676 in costs and the 10% grid $307. Costs were not the main reason, though. Over the last 10 years of QQQ, even with no costs the 3% grid returned 2.8% a year, below the 10% grid's 3.5%. A wide grid sells a slot only after a 10% rise, so it held stock longer during rebounds (average stock share over the last 10 years: 9% at 3% spacing, 13% at 10%). TQQQ's full period was also 14.3% a year at 3% and 16.8% at 10%. The full table by spacing is in Grid spacing 3%, 5% and 10% compared.

Interest on cash and taxes

This calculation gives cash no interest. The grid held about 90% cash on average over the last 10 years, and the 3-month US Treasury bill averaged 2.4% a year over the same period. Adding interest would raise the annual return by roughly 2.2 points, to about 5% a year. The gap to buy and hold (20.9% a year) stays large. Taxes work the other way. For Korean residents, the tax rate on foreign stock gains above ₩2.5 million a year is 22% (including local income tax). Buy and hold defers tax until the sale, but a grid realizes a gain every time it sells a slot, so tax can arise every year.

Related concepts and search terms

Grid trading puts a few known ideas into a price-line rule.

• Grid trading, grid bot: the name widely used in forex and crypto automated trading. Exchanges offer settings such as arithmetic and geometric spacing, a price range and trailing up.
• Tteolsaopal: a Korean shorthand for "buy when it falls, sell when it rises".
• Mean reversion: the assumption that a price that moves up and down returns to where it was. A grid is a bet on this assumption and runs opposite to trend following, which buys breakouts to new highs (Turtle breakout, golden cross).
• Rebalancing bonus, volatility harvesting: the idea that returning volatile assets to set weights can earn a little more than the weighted average of their returns. William Bernstein laid it out in a 1996 article (original). The margin by which the grid beat the same average share above is this effect.
• Shannon's Demon: an example attributed to Claude Shannon. A stock that keeps doubling or halving and goes nowhere can still grow an account that is split half and half with cash and rebalanced every time. David Luenberger called the same effect volatility pumping.
• Value averaging: buy more when the holding is below a target value and sell when it is above. It also buys low and sells high, but its yardstick is a target amount set over time, not price lines (more in The basket method).
• Scaling in: buying in steps as the price falls. The buy side of a grid is close to a ladder that adds a slot for every 5% drop from the high (see the drawdown ladder in Leverage cross-rebalancing).

Search terms: grid trading, grid bot, geometric grid, mean reversion, rebalancing bonus, volatility harvesting, Shannon's demon, tteolsaopal, infinite buying method, 무한매수법.

Drawbacks and risks

• It holds almost no stock in a rising market. Over the last 10 years of QQQ, the average stock share was 10%.
• A decline deeper than the slots cover leaves nothing more to buy with. In this calculation that depth is about −60%. QQQ started in January 2000 and ran into the dot-com crash, so all 20 slots filled, and the grid could not buy at all from 2002 through 2006 and in 2009. Over the full period, the grid could not buy (every slot full or not enough cash) on 31.2% of trading days for QQQ and 7.2% for TQQQ.
• Its share is largest at the bottom. TQQQ's max drawdown over the last 10 years was −55.9% for the grid, deeper than the −31.7% of keeping the same average share fixed.
• It trades a lot. The 3% grid on TQQQ filled 4,352 times over the full period and paid $7,132 in costs (from $10,000 at the start).
• The calculation assumes a price order within each daily bar. If the real order differed, same-day buys and sells would differ.
• Each sale realizes a gain, so tax can arise every year.