What to do when your gold position is underwater? The arithmetic of averaging, stopping out and getting back to even

Start with one number: the snapshot at 18:07 Beijing on 2026-10-21 put gold at 4,116.96 USD/oz, or 883.9 per gram, 883,900 per kilogram. If you bought at 4,300 you are down 4.3%; at 4,200, down 2.0%; if you chased above 5,000 you are down close to 18%. All three are "underwater", and all three need completely different treatment, because one of them needs a single hourly candle to recover while another needs an entire trend.

This article is not about direction, only arithmetic, and it answers three questions: how far the price must rise to get you back to even, where averaging down can pull your cost to and what it costs you, and how far your margin can actually hold out. All the statistics come from 514 trading days and 38,768 fifteen-minute candles of XAUUSD between 22 December 2024 and 14 August 2026, and every amount can be recomputed with your own numbers.

1. First work out which kind of underwater you are

"Underwater" is one word but three completely different conditions in an account. Use the wrong remedy and you lose a little more; use the worst one and you get liquidated.

Type one: directionally underwater (fully paid spot, physical gold)

You bought 100 grams of bars for 100,000, gold falls 10%, and the book value is 90,000. There is no mechanical risk here — no liquidation line, no overnight interest, and the bars do not evaporate. What you pay is opportunity cost (that money could have done something else) and storage cost (safe deposit, transport, assay). The only cure is time, and time is the one thing physical gold never runs short of.

Type two: leveraged and underwater (CFDs, margined spot)

You open one lot long on a 20,000 USD account, notional 411,696 USD. Gold falls 4.26% and your margin is wiped out (section five works through it). The nature of the problem has changed entirely: whether you were right no longer matters; whether you survive until being right does. Leverage translates "floating loss" into "countdown".

Type three: underwater in time (positions with a holding cost)

Some positions are directionally fine and never liquidated, but bleed every day. One lot long financed at 4% a year costs about 45.7 USD a night, 1,372 USD a month. Three months of range-trading with gold going nowhere and your account is 4,116 USD lighter — exactly 1% of the gold price. This is the most insidious kind, because the screen says "flat" while you have already paid three bills.

2. The arithmetic of recovery: why a 50% loss needs a 100% gain

The most basic and most ignored lesson. Losses and recoveries are not symmetric: after a 50% fall you need a 100% rise to get back to even, because the denominator shrank — 100,000 down to 50,000 means a rise from 50,000 to 100,000, which is 5 ÷ 5 = 100%.

The right-hand column is the nominal price implied by today's 4,116.96 USD/oz mid, shown only to illustrate magnitude, not as a forecast.

The most useful way to read this table is backwards: it tells you where your stop belongs. If the hardest recovery you can accept is "a 25% rally", your stop is at −20%; if your ceiling is "an 11% rally", the stop belongs at −10%. Most people set stops by asking "will this level get hit", not "if it gets hit, what kind of move do I need to come back" — the second question is the real cost.

Worse still is the mental accounting. Losing 1,000 USD from a 1,000 USD profit and losing 1,000 USD from flat feel different: the second hurts roughly twice as much. Which explains why underwater traders gravitate to two wrong moves — averaging down (turning a loss into "not finished yet") and cancelling the stop (turning a realised loss into an unrealised one). Both postpone facing the arithmetic.

3. 514 trading days: how deep gold's drawdowns go, and how long they take to recover

Everything below is measured from the raw 15-minute candles, not from memory. The window runs 22 December 2024 to 14 August 2026 — 514 trading days, a low of 2,596.07 and a high of 5,595.32 (29 January 2026), with the close rising from 2,624.29 to 4,371.41, a gain of 66.6%. That is a clear bull market — in other words, these drawdown figures were measured in a bull market, and a bear market will only be worse.

Method: close versus the highest close of the previous 90 sessions; one count per drawdown episode; recovery means the intraday price touches that high again; 180 days is the observation cap. The sample covers only 20 months and should not be extrapolated to longer cycles.

Three conclusions worth remembering.

First, shallow drawdowns are normal and deep ones are tail events. Drawdowns beyond 5% occurred 8 times in 20 months — roughly every ten weeks — and 88% recovered within six months, median 53 days. That means the great majority of underwater positions resolve themselves inside a quarter, which is exactly why averaging down "appears to work" in a bull market. But drawdowns beyond 10% occurred only 3 times with just 1 recovery, and nothing beyond 15% ever recovered. Mistaking "most cases" for "all cases" is where averaging down starts losing serious money.

Second, time underwater is far longer than people imagine. On 434 of the 514 trading days (84%) the close was below the prior high; genuine new highs occurred on only 80 days. There were 36 underwater episodes in all, with a median of just 2.5 days — most pullbacks end in two or three sessions. But the longest ran 169 trading days, from the 28 January 2026 high of 5,513.02 down to 3,985.78 on 16 July, a maximum drawdown of −27.7%.

Third, averages lie. On the average alone, the 20 sessions after a 10% drawdown returned +3.0%, which looks like a buy signal. But two of those three cases kept falling; the average was dragged up by a single fast rebound. The smaller the sample, the less reliable the average — with three observations the distribution matters, the mean does not.

4. The real maths of averaging down: average cost, rally needed, and what it costs

The formula is trivial: new average = (original cost + added cost) ÷ total quantity. It genuinely lowers your average, but by far less than intuition suggests, and the price you pay is hidden.

Equal-size averaging: a 10% fall only lowers the average by 5%

Worked example: buy 0.50 lots (50 oz) at 4,200, then add another 0.50 lots after a fall of X%. Total position afterwards is 1.00 lots (100 oz).

Look at the last column: after an equal-size top-up your total exposure has doubled. You had 50 oz, 210,000 notional; now you have 100 oz, 336,000 notional (at 3,360). The average cost fell from 4,200 to 3,780 — apparently 10% better — while your risk exposure rose 60%. That is the essence of averaging down: it is not "reducing cost", it is "buying a lower cost line with a bigger position".

Doubling down: the fourth step eats the whole account

A more common approach is "add on every fall and double each time". Starting again at 0.20 lots and 4,200, doubling the position after every 5% decline:

Based on the XAUUSD standard contract of 1 lot = 100 oz; notional exposure = cumulative ounces × current price.

It looks wonderful: the average slides from 4,200 all the way to 3,536, so the whole book only needs a 5.2% rally to be "back to even". Add two facts and it collapses:

• Margin. The 620 oz at step four requires 20,832 USD of margin at 1:100 — a 20,000 USD account cannot even open it. And if it could, used margin would be 100% and a further 1% move in gold would liquidate it.
• Floating loss. At step four your actual floating loss is 109,120 USD, 5.5 times the size of a 20,000 USD account. The account would have been liquidated at step three; step four is unreachable.

Three numbers to compute before you add

1. Total notional after the add ÷ account equity — above 3×, do not add; that is adding leverage, not averaging.
2. Distance from the new average to the liquidation line — run it through the margin calculator. If it is under twice the average daily range (1.85% today, about 76 USD), you are handing the account to noise.
3. The floating loss if it falls another 10% — compute it with the drawdown calculator. If that number exceeds a year of your disposable income, any averaging down is gambling.

5. Leverage decides whether you are allowed to wait

Many people assume the leverage multiple determines risk. One worked example shows how wrong that is.

Example: a 20,000 USD account, gold at 4,200, and a platform that liquidates below a 50% margin level (an industry-common setting).

Read columns three to six: for the same 0.50 lots, going from 1:100 to 1:500 cuts margin used from 2,100 to 420 — an 80% reduction — while the liquidation distance moves only from 9.02% to 9.42%. The reason is simple: liquidation distance is set by position size ÷ account equity; the leverage multiple only sets how much margin gets frozen. 1:500 does not make you more resilient, it just lets you open a bigger position with the same money.

Put those two figures together and the risk becomes concrete:

• The average daily range in this sample is 1.85% (median 1.55%, 90th percentile 3.34%), which in dollars is 75.2 / 58.3 / 141.6.
• One lot at 1:100 has a liquidation distance of 4.26%, about 179 USD.
• Which means a single 90th-percentile day (3.34%, 141.6 USD) is enough to remove eight-tenths of a one-lot position's buffer, and two in a row finish it off. The longest losing streak in the sample was six sessions.

6. Overnight interest and weekend gaps: the hidden bill for waiting

Overnight interest (swap): one charge a day, invisible but certain

CFD holding cost is accrued as "notional × cost of funds ÷ 360", usually paid on longs and received on shorts (it varies by platform; the published swap points govern). At the 2026-10-21 snapshot of 4,116.96:

• One lot notional = 100 oz × 4,116.96 = 411,696 USD
• At 4% a year, daily cost ≈ 411,696 × 4% ÷ 360 = 45.7 USD
• 30 days = 1,372 USD; 90 days = 4,116 USD (exactly 1% of the gold price)

Turn that into "how much gold has to rise just to pay the interest": 45.7 ÷ 100 oz = 0.457 USD/oz/day. One lot long must gain 0.011% a day to break even. It sounds trivial, but underwater positions are usually held for months — 90 days is 1%, a year 4%. This is why "hold a leveraged position long term and wait to recover" is the worst possible exit strategy in arithmetic terms: you wait for the price to come back while the cost pushes the break-even line away.

Weekend gaps: your stop only takes effect at the Monday open

Gold trades continuously Monday to Friday and stops over the weekend. Geopolitical events, central-bank comments and dollar moves during the closure are priced in at once at the Monday open, and a gap does not trigger your stop price, it triggers the traded price — your stop sits at 4,000, Monday opens at 3,900, and you are filled at 3,900.

Measured over 85 weekends in this sample: average gap +0.01%, median +0.04%, largest positive +1.28%, largest negative −2.00% (1 February 2026); 7% of weekends (6 of them) gapped more than 1%.

−2.00% on a one-lot position is an 8,234 USD hole, 41% of a 20,000 USD account. With two lots that single gap punches through the liquidation line and the account is at zero one second after the open. So before holding over a weekend the job is not "read the direction" but "do the lot maths": size down until a worst-case 2% gap costs no more than 3% of the account. On 20,000 USD that is 600 ÷ (4,116.96 × 2%) = no more than 0.073 lots — a number that surprises most people.

7. Three ways out and four conditions for adding

Reduce all the arithmetic above to a decision. Once underwater there are only three paths, and no fourth.

Path A: take the loss and leave (for leveraged positions where the thesis has broken)

Any one of these triggers is enough: the entry reason no longer holds (you were betting on a data point or a meeting and it went the other way); the position's risk exceeds 2% of the account; or the floating loss has you widening stops, adding margin and losing sleep. The cost of leaving is certain (current loss plus spread); the cost of staying is uncertain and possibly larger. A certain 10% loss beats an uncertain 50% one.

Path B: cut, do not move (for positions that still have a case but are too large)

Halve the position, bring effective leverage under 2×, and keep the remainder. The benefit is doing two things at once: doubling the distance to liquidation and halving the overnight cost. The drawback is that the average cost and the rally needed to recover are unchanged. This is the most underrated path — most people choose only between "cut everything" and "keep everything" and forget the middle.

Path C: conditional averaging down (for unleveraged or very low leverage, with long-horizon money)

Adding is not a feeling; it is four conditions that must all hold:

1. Position condition: after the add, total notional ≤ 3× account equity and used margin < 20% of the account.
2. Spacing condition: the add is at least 1.5× the average daily range away. Today's average daily range is 1.85%, so at least 2.8%, about 115 USD. Adding every 20 USD of decline treats spread and noise as opportunity.
3. Count condition: at most twice, each no larger than the original position. From the third add onwards you are not investing, you are running a martingale.
4. Invalidation condition: before adding, write down "if it falls to X I exit everything" and place it as a real order. Averaging without an invalidation level is unlimited staking by another name.

8. Three variables specific to overseas Chinese

Time zone: your sleep covers the worst hours

Gold's most volatile window runs from the London afternoon into the New York morning, which is 20:00 to 02:00 Beijing. In North America that is morning to midday — fine. In Australia or New Zealand it is deep night, which produces the classic underwater scenario: down 3% at bedtime, liquidated by the time you wake. The fix is not sleeping less but placing stops in advance and sizing positions so they do not need watching, plus checking the session table for the active window in your own time zone.

Tax: losses do not always offset — the most commonly missed point

• United States: retail clients cannot legally trade gold CFDs (post Dodd-Frank); the available instruments are futures, ETFs and physical. Futures are taxed under the 60/40 rule, physical gold and silver as collectibles with long-term gains up to 28%.
• Australia: CFD profits are taxed as ordinary income and losses generally offset only capital gains, not salary; assets held over 12 months may get the 50% discount, but CFDs usually do not qualify.
• United Kingdom: CFDs are treated under capital gains tax, and losses can be offset against capital gains in the same or later years, not against income.
• Canada: 50% of capital gains is included in taxable income, and losses likewise offset only capital gains.
• Hong Kong / Singapore: no capital gains tax, though frequent trading may be treated as trading income — a different character altogether.

The common thread: the offset value of a loss is far below the tax cost of a gain. Expecting losses to make up for trading mistakes turns a 100 loss into a 20 tax saving.

Funding: topping up margin is where things most often go wrong

Deposits must come from a bank account in your own name. Paying from a spouse's, parent's or friend's account, or through third-party payment providers and OTC crypto transfers, triggers AML review: at best the funds are returned and frozen for 30 to 90 days, at worst the account is closed and reported. When you are underwater what you need most is money that arrives immediately, and this is precisely the route that jams when used wrongly. Cross-border transfers add another layer: the mainland's 50,000 USD annual facilitation quota is for current-account purposes only, and splitting transfers or borrowing another person's quota is a breach; once confirmed you land on a watch list that affects all your future FX business.

9. An actionable checklist

On the day you go underwater, work through these eight items in order rather than trading on feel:

1. Compute effective leverage: notional ÷ equity. Above 3×, cut first and discuss the rest later.
2. Compute the liquidation distance: put the current price and lot size into the margin calculator to get "how many dollars of decline liquidates me".
3. Compare with the average daily range: 1.85% today (about 76 USD). A liquidation distance under twice that average hands your fate to a single candle.
4. Compute the rally needed to recover: use the table in section two, then check how long such a move has historically taken (section three's recovery days).
5. Compute the holding cost: nightly swap plus amortised spread, expressed as "how much it must rise each day just to break even".
6. Write the invalidation level and place the order: not "a number in your head" but a stop that exists on the server.
7. Choose the path: A leave / B halve / C conditional add — pick one and write down the reason and the date.
8. Record it: log the entry reason, current loss, decision and invalidation level in a trading journal. Next time you are underwater, that record will be worth more than any indicator.

10. FAQ

Related tools