Take Profit Near or Far? Gold Backtests Point to a Sweet Spot

Take Profit (TP) is the price you set to close a trade in profit. The short answer to “near or far?” is that both can be wrong. A very near target wins often but earns little compared with the risk. A target right at the old high, or one that is too far away, rarely gets hit. In our backtests, the results formed a hill with a peak, and the peak sat in a different place for each system. This article explains why, with the numbers we measured.

Why near targets win often

Suppose price moves completely at random, with no one having an edge. The math says the chance that price reaches the TP before the Stop loss (SL, the point where you cut the loss) is

Chance of hitting TP = SL distance ÷ (SL distance + TP distance)

If the SL is 10 dollars away

TP distance	Chance of hitting TP if price is random
3 dollars	10 ÷ 13 = 76.9%
10 dollars	10 ÷ 20 = 50%
20 dollars	10 ÷ 30 = 33.3%
30 dollars	10 ÷ 40 = 25%

The further the target, the lower the chance of reaching it. That is why near targets win often. But winning often does not mean making money. Work out the expectancy of a 3 dollar TP at 0.01 lot of gold (1 ounce, so a 1 dollar move makes or loses 1 dollar)

• Win: 0.769 × 3 = 2.31
• Loss: 0.231 × 10 = 2.31
• Expectancy = 2.31 − 2.31 = 0

Every distance in the table gives the same zero. If price is random, no TP gives you an edge. The difference comes from costs and from whatever edge you actually have, which is the next section. If comparing win rate with RR (Risk:Reward, how much you can win compared with how much you risk) is new to you, read Win rate and breakeven alongside this.

Expectancy, step by step

Expectancy is the average amount you make or lose per trade over many trades.

Expectancy = chance of winning × profit − chance of losing × loss − costs

Assume a spread cost (the gap between the buy and sell price) of 0.30 dollars per trade, close to the roughly 0.27 dollars per trade at 0.01 lot we measured on gold. The SL is still 10 dollars away.

Step 1: find the win rate you need to break even = (SL distance + spread) ÷ (SL distance + TP distance)

TP distance	If random	Needed to break even	Must beat random by
3 dollars	76.9%	10.3 ÷ 13 = 79.2%	2.3 points
10 dollars	50.0%	10.3 ÷ 20 = 51.5%	1.5 points
20 dollars	33.3%	10.3 ÷ 30 = 34.3%	1.0 points

Near targets need a bigger edge over random, because the spread takes a large share of the profit per trade. 0.30 is 10% of a 3 dollar profit, but only 1.5% of a 20 dollar profit.

Step 2: give the system the same 3 point edge over random

TP 3 dollars, chance of winning 76.92% + 3 = 79.92%

• Win: 0.7992 × 3 = 2.40
• Loss: 0.2008 × 10 = 2.01
• Expectancy = 2.40 − 2.01 − 0.30 = +0.09 dollars per trade

TP 20 dollars, chance of winning 33.33% + 3 = 36.33%

• Win: 0.3633 × 20 = 7.27
• Loss: 0.6367 × 10 = 6.37
• Expectancy = 7.27 − 6.37 − 0.30 = +0.60 dollars per trade

Check with the shortcut: expectancy = edge over random × (SL distance + TP distance) − spread

• TP 3: 0.03 × 13 − 0.30 = 0.39 − 0.30 = 0.09, matches
• TP 20: 0.03 × 30 − 0.30 = 0.90 − 0.30 = 0.60, matches

Same edge, but the far target makes almost 7 times as much. There is an important catch, though. That 3 point edge is not constant at every distance. Past a certain point, real systems usually beat random by less, or even lose to it, which is exactly what we see in the data below.

Evidence 1: a near profit cap wipes out profit

We took a set of 3 pullback systems on large ZigZag swings (swings at least 1% of price long) that normally set the TP at a Fibonacci level. We forced the TP to be no more than 3 or 5 dollars of price away. Backtest on gold (XAUUSD) M1 for 2026 with real tick data, 0.10 lot per trade, starting balance 10,000 dollars.

TP	Trades	Win rate	Net profit	Max drawdown	Profit ÷ Drawdown
Max 3 dollars	1,229	83%	+2,182	2,835	0.77
Max 5 dollars	890	76%	+3,685	2,974	1.24
At the Fibonacci level	362	42%	+12,461	3,941	3.16

Drawdown is the largest drop in the account from a peak along the way. The profit cap did make the systems win more often, but profit fell to just 2,182 ÷ 12,461 = 17.5% and 3,685 ÷ 12,461 = 29.6% of the original, while drawdown only dropped about 25–28%. The number of trades went up because trades closed early and re-entered. The more you trade, the more often you pay the spread.

Another system buys pullbacks within the range of the last 150 bars, gold M1, tested continuously from 1 January 2022 to 20 August 2026, 0.01 lot per trade, 3,000 dollar account. The only change was the TP. With the target at just 61.8% of the way back to the top of the range, the account lost 3,001 dollars, a full wipeout (100% drawdown), with a 44.1% win rate. With the target past the top of the range, it made +2,400 dollars with a 38.2% drawdown and a 20.9% win rate.

All results in this article are backtests. Past results do not predict future results, and these numbers are not returns you should expect.

Evidence 2: a target right at the old high does worse

If a near cap is bad, maybe going all the way is better. Not for Fibonacci pullback trades (refresher on Fibonacci levels).

We counted every time price pulled back to touch the 61.8 level of a large swing on gold M1 in 2022–2026, which happened 357 times. Assume the SL at 88.6 (a risk of 26.8 swing units), and check how often price reached each target before hitting the SL.

Target	Profit (swing units)	Actually reached	If random	Edge over random × (SL + TP)
50	11.8	74%	69%	0.05 × 38.6 = 1.93
38.2	23.6	57%	53%	0.04 × 50.4 = 2.02
23.6, 11.8, 0	38.2 or more	Below random		Negative

The last column is the shortcut from the previous section (before spread, and approximate because the percentages are rounded). Target 38.2, which is about 38% of the way back (23.6 ÷ 61.8), gives the highest value. Past halfway, the chance of reaching the target drops below random, so expectancy turns negative. This is a count of events with none of the system’s other filters. It is useful for choosing a target distance, but it is not a real profit figure.

Backtests of real systems point the same way. Pullback systems in the 61.8–78.6 zone of a large swing (some also require an FVG, a price gap, explained in the ICT article), gold M1, 0.10 lot per trade, 10,000 dollar account. 2022–2025 used 1-minute data and January to September 2026 used real ticks, almost 5 years combined.

Version	TP 38.2	TP 0 (full way)
SL 100, no FVG	+1,605	−3,657
SL 100, with FVG	+2,566	−1,938
SL 88.6, no FVG	+20	−4,742
SL 88.6, with FVG	+1,592	−2,710

In all 4 pairs, the full target lost money. The win rate fell from 35–54% to 16–30%. Backtest results do not predict future results and are not returns you should expect.

Evidence 3: too far also drops off

A Donchian channel breakout system (entering when price breaks out of a range) on M15 sets its target as a Fibonacci extension (a level beyond the top of the swing, for example 161.8 is 61.8% of the swing past the top). We swept 4 target sizes on gold from 5 January to 25 July 2026, real tick data, 0.02 lot per trade, 3,000 dollar account.

The win rate fell from 28% at target 161.8 to 16% at target 423.6. Far targets only get hit in months with a long trend, so months where price swings in a range turn into losses.

Another system buys pullbacks at 61.8 and measures the target as a share of the swing (100 = back to the old high, above 100 = beyond the old high). Gold M1, tested continuously for about 4.6 years, 3,000 dollar account. We compare using profit ÷ drawdown, because the 250 row used 0.02 lot while the others used 0.01 lot, and this ratio does not depend much on lot size.

Target	45	61.8	250	400
Profit ÷ Drawdown	0.39	1.22	4.43	0.39

For this system the peak of the hill sits well beyond the old high, unlike the earlier table where the peak was less than halfway. But the shape is the same: too near is poor, and too far falls off a cliff. Backtest results do not predict future results and are not returns you should expect.

There is no TP distance that is right for every system. But every system we tested had a hill shape, with the nearest and furthest targets both doing poorly.

Nudging the TP barely matters

When price nearly reaches the TP and then turns back, many traders want to pull the TP in a little. We tried this on a combined set of 3 pullback systems, 6,000 dollar account, results for 2024–2026 combined (2026 used real ticks).

Pull TP in by	3-year total profit	vs no change
No change	16,890	—
One spread	16,943	+0.3%
3% of TP distance	16,897	+0.04%
6% of TP distance	16,622	−1.6%
12% of TP distance	16,392	−2.9%

A small pull changes things by only about 0.3% or less, which is within the noise. Pulling it in further keeps making it worse. Backtest results do not predict future results. The work that pays off is choosing the right target range, not polishing decimals.

How to choose a TP for your own system

1. Write a clear target rule, for example a Fibonacci level or a multiple of the SL distance.
2. Test a wide range of distances, from very near to well beyond the old high.
3. Look at net profit, profit ÷ drawdown and the number of positive months, not just the win rate.
4. Pick a point on the hill where the neighbouring values are also good, not a single sharp peak (more in Sharp peak vs plateau).
5. If the target is far, be ready for long losing streaks (Losing streaks are normal).

Use the RR calculator at the end of this article to see what win rate each TP distance needs to break even.

Common mistakes

• Choosing a near TP because you want to win often. The 83% win rate in the first table made almost 6 times less profit than the 42% win rate.
• Setting the TP right at the old high because it seems logical. In our data, the full target lost money in every pullback version we tested.
• Stretching the target because you saw trades on the chart that ran far. Those trades do happen, but not often enough to pay for all the ones that fall short.
• Forgetting costs. The nearer the target, the bigger the share of profit the spread eats, and it shows even more on small timeframes (Small timeframes and the cost of spread).
• Using one system’s TP distance on another system. The peak is in a different place every time, so you have to test again.