Home › Learn › Grid EAs, Honestly

Grid EAs, Honestly: How They Make Money and How They Blow Accounts

By Zubare Khan · Updated 9 September 2026 · 6 min read · Algo

Grid and martingale expert advisors produce the smoothest equity curves in retail trading, and they are responsible for a large share of blown accounts. Both of those things are true for the same reason. This is the mechanism, with the arithmetic written out.

What a grid actually does

A grid strategy places orders at fixed intervals and profits from price oscillating between them.

In its simplest form: buy at the current price. If price falls by a set step, buy again. If it falls another step, buy again. Each new entry lowers the average price of the combined position. When price eventually rallies far enough to put the basket into profit, close everything at once and start over.

Note what the strategy never does: take a loss on a single position. There is no stop-loss on any individual order. A losing position is not closed, it is averaged.

A martingale grid adds one thing: each new entry is larger than the last, usually double. This makes the average price fall faster, so a smaller bounce is needed to get out.

Why it works, and why the curve looks so good

Markets range far more of the time than they trend. Any strategy that bets on mean reversion is betting on the common case, so it wins frequently — a well-tuned grid can close profitable baskets for weeks on end.

The equity curve looks extraordinary, and the reason is a reporting artefact worth understanding precisely. Closed trades are all profits. The losses have not been realised; they are sitting open, as floating drawdown, invisible on a closed-trade report.

So a grid account produces:

• A near-100% win rate on closed trades, because a basket is only ever closed in profit.
• A smooth, rising balance line.
• An equity line that periodically plunges and recovers — and equity is the one that matters.

Any published grid result showing balance rather than equity, or win rate rather than maximum floating drawdown, is not describing the strategy. When you evaluate one of these, the only numbers worth reading are maximum floating drawdown and the largest number of grid levels ever opened. Everything else is decoration.

The arithmetic of a grid that goes wrong

Here is the part that is almost never shown. Take a martingale grid on gold: first entry 0.01 lots, doubling at each level, grid step $5 of gold price. Now suppose gold falls $50 against you — a single ordinary session on this instrument.

By then, ten positions are open. Their sizes double each time, so they are 0.01, 0.02, 0.04, 0.08, 0.16, 0.32, 0.64, 1.28, 2.56 and 5.12 lots. That is 10.23 lots total — a thousand times the position you started with.

A starting position of one hundredth of a lot, and a fifty-dollar move in gold, produces a floating loss of over ten thousand dollars. Nothing unusual happened. Gold did not gap, no bank failed, there was no flash crash. It simply trended for one session in the direction the grid was not positioned for.

And the margin required to hold 10.23 lots is a separate problem on top: at a gold price of $3,000 that is roughly three million dollars of notional, which even at 1:500 leverage demands about $6,000 of margin you must still have free. Run out of margin and the broker closes the positions for you, at the worst possible moment, by definition.

This is why grid accounts do not decline gradually. They work, work, work, and then end.

A fixed-lot grid is safer, and still not safe

Run the same scenario without doubling — 0.01 lots at every level. Ten positions, 0.10 lots total, floating loss of $275 on the same $50 move. Survivable on a modest account.

Fixed-lot grids are therefore vastly more defensible than martingale ones. But the exposure still grows linearly with distance, the recovery still requires a bounce, and the fundamental bet has not changed: the market will come back before my account runs out. Over enough time, on enough instruments, sometimes it does not.

What actually kills them

Not randomness. Specific, recognisable conditions:

• A sustained trend. The strategy's only enemy. Gold can trend for days on a rate repricing or a geopolitical event, and a grid on the wrong side adds size the entire way.
• A gap. Weekend gaps skip grid levels entirely — the positions open at once, far from where the logic expected them, and stops that do not exist cannot help.
• A volatility regime change. A grid step calibrated to a quiet month is meaningless in a violent one. The same parameters that were conservative become aggressive without any parameter changing.
• Running out of margin before running out of thesis. The bounce may well come. It is irrelevant if the stop-out came first.
• Swap. A grid that holds a large one-directional basket for days pays negative swap on all of it, nightly. Slow bleed on top of floating loss.

If you are going to run one anyway

Grids are not fraud and they are not useless. They are a high-probability, negatively-skewed bet: many small wins, rare very large losses. That is a legitimate shape of return, and plenty of professional strategies have it — but only when it is sized and bounded deliberately. The controls that matter:

• A hard basket stop. A maximum floating loss, in money, at which everything closes regardless of how convinced the logic is. This converts an unbounded tail into a bounded one. It is the single most important control and the one most commercial EAs omit, because including it makes the backtest look worse.
• A level cap. A maximum number of open grid positions, after which no more are added.
• Fixed or slowly-increasing lots, not doubling. Doubling is what turns a bad day into a terminal one.
• Capital that reflects the worst case, not the average one. Work out the floating loss at your level cap, as in the table above, and hold enough that it does not breach margin. This is what a stated minimum capital requirement is actually for — it is a risk floor, not a suggested deposit.
• A news and session filter. Not adding levels into a scheduled high-impact release removes a category of worst-case entries for free.
• Monitoring. Never leave one running unattended on money you cannot lose. “Set and forget” is a marketing phrase, not a risk policy.

How to evaluate someone else's grid EA

When a grid strategy is offered to you — free, paid, or bundled with anything — four questions settle it:

1. Show me the equity curve, not the balance curve. If only balance is shown, assume the equity curve is being hidden for a reason.
2. What was the maximum floating drawdown, in percent of the account? If the answer is not a number, there is no answer.
3. How many levels deep has it ever gone, and what is the cap? “It has never needed more than four” is not a cap.
4. Does it double? If yes, read the table above again and decide whether you can fund it.

A win rate, a screenshot, a month of results, and a number of users are not evidence about any of this. They are evidence that the strategy has not yet met the condition that ends it.

And test it yourself before it touches money — on a demo account, over a period containing a real trend, with honest spread assumptions. How to do that without deceiving yourself is a whole subject of its own, covered in backtesting in MT5 without fooling yourself.

Zubare Khan

I trade gold, forex and index CFDs, sell short-dated option volatility, and build the MT5 and TradingView tools published on this site. Everything here is written from my own screen time and my own losses — not from a content brief. More about how I trade and write.