Grid, Martingale and Averaging Down: Frequent Wins, One Trend Can Wipe You Out

Grid, Martingale and averaging down all open more orders when price moves against you, instead of accepting a loss. These systems win very often. The equity curve climbs smoothly for months, until the day price runs in one long trend. Then a single loss wipes out all the profit, sometimes the whole account. This article walks through each method with numbers, then looks at what we found when we actually tested them on gold.

Three names, one mechanism

• Grid places orders in evenly spaced levels, for example every 5 dollars. Whenever price moves against you to the next level, it opens another order there. It then closes the whole basket (a group of orders closed together) once the combined profit hits the target.
• Martingale increases the lot size every time price goes against you, most often by doubling it. The new trade is big enough that a small bounce covers the loss on all the earlier ones.
• Averaging down / recovery trades: your first trade is in the red, so you open another in the same direction at a better price. This pulls the average price closer to the market. It is usually done by hand.

All three share one thing: they refuse to take the loss while it is still small, and they add risk in the direction the market is telling you that you are wrong.

Example: doubling the lot over 8 trades

Say you Buy gold at 0.01 lot. Every time price drops 5 dollars you open a new Buy at double the lot, and you close the whole basket when the combined profit reaches 5 dollars. One lot of gold is 100 ounces, so a 1 dollar price move changes your balance by 100 dollars per lot.

Trade	Lot this trade	Total lot	Price below first trade ($)	Average price below first trade ($)	Floating loss when this trade opens ($)
1	0.01	0.01	0	0	0
2	0.02	0.03	5	3.33	5
3	0.04	0.07	10	7.14	20
4	0.08	0.15	15	11.33	55
5	0.16	0.31	20	15.81	130
6	0.32	0.63	25	20.48	285
7	0.64	1.27	30	25.28	600
8	1.28	2.55	35	30.16	1,235

Total lot after N trades = 0.01 × (2^N − 1). At trade 8 that is 0.01 × 255 = 2.55 lots, 255 times bigger than the first trade, even though price has only moved 35 dollars against you.

Why it looks convincing: the average price follows you down

Look at the average price column. After trade 8, the average sits 30.16 dollars below the first trade, only about 5 dollars from the latest entry. So a small bounce is enough to close the basket in profit. This is why Martingale wins almost every round. Markets mostly swing back and forth, and 5 dollar bounces happen all the time.

The problem is the day price does not bounce. Let price keep falling to 60 dollars below the first trade, and stop adding trades after trade 8.

• Floating loss = total lot × 100 × (average price − current price)
• = 2.55 × 100 × (60 − 30.16) = 255 × 29.84 ≈ 7,610 dollars

Check it trade by trade (lot × 100 × distance against you): 0.01 × 100 × 60 = 60 · 0.02 × 100 × 55 = 110 · 0.04 × 100 × 50 = 200 · 0.08 × 100 × 45 = 360 · 0.16 × 100 × 40 = 640 · 0.32 × 100 × 35 = 1,120 · 0.64 × 100 × 30 = 1,920 · 1.28 × 100 × 25 = 3,200. Total 7,610 dollars, which matches.

A basket that wins 5 dollars at a time needs 7,610 ÷ 5 = 1,522 wins to pay for one losing round. And 60 dollars is not a long move for gold. In the next sections you will see a month where price ran about 1,000 dollars in one direction.

Grid and Martingale do not reduce risk. They just move it all into a single day.

Equal-lot Grids do not escape it either

Some people think that if you do not double the lot and just open the same size at each level, you are safe. An equal-lot grid does grow more slowly, but the loss still grows faster than a straight line. Every earlier trade loses more each time price moves further against you, while new trades keep being added.

Our testing boils it down to one line. A winning basket only earns the distance from breakeven to the next level, which is short. A losing basket loses the full width of the grid times every lot you hold. In one example we worked out, the basket made 3 dollars when it won and lost 68 dollars when it lost. That is an RR (Risk:Reward, how much you can win compared with how much you risk) of 1:22.7, so you need to win 95.8% of the time just to break even. See how breakeven is worked out in Win rate and breakeven.

Our test: a grid buying bounces at Fibonacci levels

We tested a system that trades bounces off the Fibonacci levels of several ZigZag swings. Each trade was 0.01 lot. If price went the wrong way, it added at the next level, at least 2 dollars apart, up to 10 trades. It closed the basket when the combined profit equalled one trade moving 5 dollars. The test ran on gold (XAUUSD) M1 for 2026 with real tick data and a starting balance of 10,000 dollars.

With no basket Stop loss: in early January 2026 the system had opened all 10 Sell trades between 4,330 and 4,379. Then gold ran up to 5,369 on 28 January. The account hit Stop out (the broker force-closes positions because margin runs out) with a loss of 10,007 dollars. The account was gone in the first month.

Rough check: 10 trades × 0.01 = 0.10 lot. Price moved about 1,000 dollars against it. Loss ≈ 0.10 × 100 × 1,000 = 10,000 dollars, the whole account.

With a 50 dollar basket Stop loss: the account did not blow up in one go, but it still lost money.

Figure	Value
Number of trades	26,669 trades
Win rate	65%
Profit factor (gross profit ÷ gross loss)	0.89
Net result	−8,851 dollars, negative every month
Estimated spread cost	0.27 × 26,669 ≈ 7,200 dollars

It won almost two out of three trades and still lost, because the profit per trade was tiny. The spread (the gap between the buy and sell price) cost about 0.27 dollars per trade at 0.01 lot. Over more than twenty thousand trades, that becomes a big number. There is more on this in Small timeframes and the cost of spread. And even with zero spread, the result would still be about 1,650 dollars in the red (8,851 − 7,200).

These are all backtest results. Past results do not predict future results, and these numbers are not returns you should expect.

Other grid systems: same ending

We tried several other families of grid systems on gold with a 3,000 dollar account: levels spaced by ATR, Limit orders at the 50% level of a swing, and levels placed on the Fibonacci levels of larger swings, on M1, M15 and H1. Nearly every setting lost the whole account, losing 3,000–3,430 dollars. Some lost more than the account because Drawdown went past 100%. None came close to surviving.

The clearest example on win rate was a grid system on gold M5, 0.01 lot per trade, 2026, real ticks. It won 68.2% of 12,392 trades and lost 2,996 dollars out of a 3,000 account. Another system on M15 with a 3,000 dollar account won 32 trades in a row and never closed a single loss on its own over 7 months of 2026. But when we ran it on 2022 and 2025 (older data that is not real tick data), it blew the account in both years. A long winning streak like that is not skill. It is a loss that simply has not been closed yet. More on this in Losing streaks are normal.

Another trap is the data you test on. Here is the Limit-at-50% system on gold M1 for 2026, with identical settings in each run.

Data used	Net result	Profit factor	Drawdown
Simulated ticks	+27,038	1.62	3.4%
Real ticks	−3,001	0.80	100%

More than 30,000 dollars apart, from the data alone. A system that holds many trades with no per-trade Stop loss is very sensitive to the tick-by-tick order of prices, so simulated ticks fool it easily. Read more in Tick data quality in MT5. Backtest results do not predict future results and are not returns you should expect.

Recovery trades make even a good entry worse

A fair question: if the original entry already makes money, would adding a recovery trade help? We tested a system that buys a pullback to 61.8% of a larger swing with the Stop loss at 78.6%. We compared it with the same system that, on reaching 78.6%, opens a second trade at double the size instead of cutting the first one. Gold M1, 2026, real ticks, 0.10 lot per trade.

Version (full 24 hours)	Trades	Net profit	Profit factor	Max drawdown
Single trade, no recovery	225	+8,754	1.35	4,416
Recovery ×2 · basket Stop loss 88.6	355	+5,549	1.13	8,275

Even in the full-day version that still made money, the recovery trade cut the profit and nearly doubled the Drawdown (how far the account falls from its peak). The reason is that the 78.6% level has no edge. Adding lots at a point with no edge is pure added risk. In another test (gold M1, January–July 2026, real ticks, 0.02 lot per trade, 3,000 dollar account), a pullback system that was already losing −1,644 dollars became −2,125 after adding a trend filter, and −3,014 after also adding recovery trades. Everything we added made it worse.

None of these backtest results predict future results, and they are not returns you should expect.

A bigger lot is an amplifier, not a profit maker. If the entry has no edge, amplifying it still ends negative.

What survives longer

The systems that survived in our testing look like the opposite: a real Stop loss on every trade, a known maximum loss per trade, fewer wins, but winners bigger than losers. Use Position size from your Stop loss so every trade risks the same amount, and pick the Stop loss distance using Stop loss: wide or tight?

Common mistakes

• Trusting the win rate. Winning 65–68% and still blowing the account really happened in our tests. Ask what the worst loss is.
• Only watching Balance. Balance (closed results) climbs smoothly because the losing trades have not been closed yet. Watch Equity (the account value including open trades) too.
• Testing on simulated data or a short period. A grid that looks good on simulated ticks or a single year may just not have met the trend that kills it yet. See When the market changes regime.
• Thinking a basket Stop loss fixes it. It limits the damage, but it does not create an edge. The result is still negative.
• Averaging down by hand with no plan. The first recovery trade is usually followed by a second and a third, each adding lots in the wrong direction.