Analysis

Sell gold below its 10-month moving average, buy it back above: thirty-six years put to the test

Published November 15, 2026

It is the best-known trend-following rule: stay invested as long as the price is above its average of the last ten months, get out as soon as it drops below. We applied it to gold since 1990, one decision a month, trading costs included, against someone who buys once and never sells.

Holding gold above its 10-month moving average, against staying invested Gold (Ounce), in US dollars, from October 1, 1990 to November 2, 2026 — ×10.4 against ×5.6. 50 100 200 400 800 1600 1995 2000 2005 2010 2015 2020 2025 ×10.4 invested throughout ×5.6 10-month moving average Rebased to 100 on 1 October 1990. Costs of 0.5% per order. Log scale.
Two value curves since October 1990, rebased to 100: staying invested throughout reaches 1,037, the moving-average rule 561.

From October 1990 to November 2026, with costs of 0.5% per order, the rule turns $1,000 into $5,613. Someone who bought on the same day and never touched it again has $10,370 — nearly double, with one order instead of sixty.

From 1 October 1990 to 2 November 2026, in US dollars, costs of 0.5% per order. Same period and same prices for both.
Strategy Total return Annualised Volatility Largest drawdown
10-month moving average ×5.61 +4.90% a year 11.54% −39.14%
Invested throughout ×10.37 +6.70% a year 14.04% −40.68%

The rule, word for word

Once a month, at the start of the month, note the gold price and compute the average of the last ten readings, the current one included. Price above the average: hold gold, buying it if you have none. Price below: hold none, selling if you do. Then look at nothing until the following month.

The history starts in January 1990; a first average needs ten readings, so the first decision falls on 1 October 1990. On that day gold is below its average, and it stays there a long time: the rule first buys only in July 1992, at $351.98. The last reading is that of 2 November 2026. In between, 434 readings, 60 orders, thirty completed round trips.

On the other side, someone buys gold on 1 October 1990 and leaves it alone. Both pay 0.5% on every order: once for the holder, sixty times for the rule.

The signal does see something

Unlike many rules, this one is not blind. In the months after a reading above the average, gold gains +0.857% on average and rises 143 times out of 264. In the months after a reading below, it gains only +0.257% and rises 80 times out of 169 — less than one time in two.

Annualised, that is roughly 10.8% on one side and 3.1% on the other. The gap is clear on paper; it is less so statistically, because an ordinary month for gold strays 4% from its mean. Welch’s test gives a t of 1.54: better than pure chance, not enough to rule it out with confidence.

Suppose the effect is real. One problem remains, and it is the whole problem: +0.257% a month is still positive. In the months when the rule was out, gold kept rising — more slowly, but rising. Leaving an asset while it gains 3% a year in order to earn nothing at all is not protection, it is a shortfall.

Twenty-three buy-backs out of twenty-nine above the sale price

It is the simplest measure of what an exit is worth: at what price did you get back in? The rule left and came back twenty-nine times. Twenty-three times, it bought back higher than it had sold. On average, 1.68% higher — before the two commissions.

The mechanism is always the same. Gold slips a little, drops below its average, the rule sells. The next month or the one after, it moves back above, and the rule buys back higher: sold at $328.02 in April 2003, bought back at $355.22 in May; sold at $586.42 in October 2006, bought back at $628.11 in November; sold at $1,776 in November 2020, bought back at $1,900 in December.

The six useful exits do exist, and two of them are real successes: sold at $385.34 in June 1996 and bought back at $308.20 in April 1998, 20.02% lower; sold at $1,578 in February 2013 and bought back at $1,296 a year later, 17.87% lower. Those are gold’s two big declines over the period, and the rule avoided a good part of them. They are not enough to pay for the other twenty-three.

Eighteen losing round trips out of thirty

Of the thirty completed round trips, only twelve are winners, and ten after costs. The median round trip loses 0.95%. That is the signature of all trend following: many small losses, a few very large gains that carry the whole.

Those large gains are real. +123.17% from December 2022 to July 2026. +82.54% from January 2009 to December 2011. +38.64% from December 2018 to November 2020. Three periods that account for most of the result on their own, and during which the rule did exactly what is expected of it: nothing.

The rest of the time, it gets whipsawed. Between December 2020 and June 2022, five purchases, five sales, five losses in a row — −6.11%, −7.18%, −3.47%, −0.44%, −5.29% — in a market that, overall, went nowhere.

What costs change

Final value of 1 invested on 1 October 1990, by the cost charged on each order. The rule places 60 orders, the permanent holder a single one.
Cost per order 10-month moving average Invested throughout
0% ×7.58 ×10.42
0.1% ×7.14 ×10.41
0.5% ×5.61 ×10.37
1% ×4.15 ×10.32
2% ×2.26 ×10.21

Sixty orders: here costs really matter

We used 0.5% per order: more than a listed gold-backed product costs at an online broker, bid-ask spread included, and well below physical gold, where the gap between the selling price and the buy-back price of a coin commonly reaches several per cent.

Sixty orders at 0.5% take a quarter off the final value: from ×7.58 with no costs to ×5.61. At 1% per order, ×4.15 is left. At 2%, ×2.26 — at physical-gold rates, this rule destroys more than three quarters of what simply keeping it earns.

But even free of charge, it loses: ×7.58 against ×10.42. Costs widen the gap; they do not create it.

The expected protection did not come

You do not follow a trend to earn more, its advocates reply, but to avoid the big declines. This is where the result is most surprising: the rule’s largest drawdown is −39.14%, that of gold held without interruption −40.68%. A point and a half apart.

Here is why. Gold peaks at $1,787 in August 2011. The rule only sells in December, at $1,567: a ten-month average reacts slowly, and 12% is already gone. Then come five years of false starts — six round trips between 2012 and 2016, all losers after costs. Each rebound moves back above the average, the rule buys, the rebound fails, it sells lower. It avoided the 2013 fall, and gave the equivalent back in small change.

The permanent portfolio hits its low in December 2015; the rule in June 2017, eighteen months later. Its volatility is lower — 11.54% against 14.04% — but set against that volatility, its annualised return gives 0.42, against 0.48 for buy-and-forget.

One decade out of four

The rule had its decade: the 1990s, when gold fell almost without interruption. From October 1990 to January 2000 it loses 15.88% while gold simply held loses 25.73%. That is exactly the market it is made for — a long, steady decline.

The next three are bull markets, and it gives up ground every time: +188.33% against +293.22% in the 2000s, +24.53% against +41.77% in the 2010s, +85.83% against +150.45% since 2020.

Its latest decision dates from 1 July 2026: it sold at $4,068.30, after the pullback that followed the March peak. On 2 November 2026 gold trades at $3,969.60, 2.43% lower. It is out of the market, and for now it is right.

What if the cash earned interest during the absences?

The rule spends 169 months out of 433 away from gold — fourteen years out of thirty-six. The calculation above leaves that money idle at 0%, the harshest assumption for it.

At 2% a year, cash lifts it to ×7.42. At 4%, to ×9.75: still behind. The break-even is at 4.46% a year, net and risk-free, in each of the months spent out. The dollar offered that in the early 1990s and briefly since 2023; not between 2009 and 2021, a period during which the rule was nonetheless out a good part of the time.

The thirty round trips, one by one

Gold price in US dollars at the month’s reference point. The last column takes off 0.5% on the purchase and 0.5% on the sale. Since July 2026 the rule has been out of the market.
Bought Price Sold Price Return After costs
Jul 1992 $351.98 Aug 1992 $343.34 −2.45% −3.43%
Apr 1993 $341.69 Aug 1994 $379.94 +11.19% +10.09%
Sep 1994 $389.92 Dec 1994 $379.43 −2.69% −3.66%
Apr 1995 $391.13 Aug 1995 $383.28 −2.01% −2.98%
Dec 1995 $387.32 Jun 1996 $385.34 −0.51% −1.50%
Apr 1998 $308.20 May 1998 $298.90 −3.02% −3.98%
Oct 1998 $294.70 Nov 1998 $293.40 −0.44% −1.43%
Oct 1999 $311.50 Apr 2000 $279.79 −10.18% −11.08%
May 2001 $268.04 Apr 2003 $328.02 +22.38% +21.16%
May 2003 $355.22 May 2004 $383.84 +8.06% +6.98%
Jul 2004 $398.92 Aug 2004 $400.19 +0.32% −0.68%
Sep 2004 $405.47 May 2005 $422.37 +4.17% +3.13%
Jun 2005 $430.41 Jul 2005 $424.42 −1.39% −2.38%
Aug 2005 $437.95 Oct 2006 $586.42 +33.90% +32.57%
Nov 2006 $628.11 Aug 2008 $839.12 +33.59% +32.26%
Jan 2009 $858.42 Dec 2011 $1,567.00 +82.54% +80.72%
Jan 2012 $1,657.00 Mar 2012 $1,662.00 +0.30% −0.70%
Sep 2012 $1,742.00 Feb 2013 $1,578.00 −9.41% −10.32%
Feb 2014 $1,296.00 May 2014 $1,257.00 −3.01% −3.98%
Jun 2014 $1,316.00 Sep 2014 $1,239.00 −5.85% −6.79%
Jan 2015 $1,261.00 Feb 2015 $1,213.00 −3.81% −4.77%
Feb 2016 $1,237.00 Nov 2016 $1,178.00 −4.77% −5.72%
Apr 2017 $1,268.00 May 2018 $1,298.00 +2.37% +1.34%
Dec 2018 $1,281.00 Nov 2020 $1,776.00 +38.64% +37.26%
Dec 2020 $1,900.00 Feb 2021 $1,784.00 −6.11% −7.04%
May 2021 $1,908.00 Jun 2021 $1,771.00 −7.18% −8.11%
Aug 2021 $1,814.00 Sep 2021 $1,751.00 −3.47% −4.44%
Dec 2021 $1,805.00 Jan 2022 $1,797.00 −0.44% −1.44%
Feb 2022 $1,908.00 Jun 2022 $1,807.00 −5.29% −6.24%
Dec 2022 $1,823.00 Jul 2026 $4,068.30 +123.17% +120.94%

What this test does not say

It does not say that trend following is worthless. It says that, on gold and over thirty-six years, this particular rule earned less than doing nothing, without noticeably reducing the largest drawdown. On an asset with longer and deeper falls the same rule may give something else; this test covers gold only.

It says nothing about other settings. We tested ten months because that is the rule as stated, and did not look for the length that would have given the best result: finding it on the same history would be the product of the sorting, not of the market. As a check only, six-, eight- and twelve-month averages also lose to simply buying.

Nor does it say anything about a rule that looked at the price every day. One decision a month reacts late; one decision a day would react sooner, and place far more orders.

Finally, everything is in US dollars and no tax is deducted. In most countries every sale of gold is a taxable event: thirty sales for the rule, none for the holder. The real gap is therefore wider than the one on this page.

Frequently asked questions

Should you sell gold when it falls below its 10-month moving average?

Over 1990–2026 it cost money. The rule gives ×5.61, with costs of 0.5% per order, against ×10.37 for gold held without interruption. Twenty-three times out of twenty-nine, it bought back higher than it had sold.

Does the 10-month moving average protect against gold’s big declines?

Very little over this period: a largest drawdown of −39.14% for the rule, against −40.68% for gold simply held. It avoided the 2013 fall but lost the equivalent in six losing round trips between 2012 and 2016.

Does the signal have any predictive value?

A weak one, but not zero. Gold gains 0.857% on average in the months after a reading above the average, against 0.257% below. Welch’s t is 1.54. The trouble is that 0.257% a month is still positive: being out has a cost.

How many orders does the rule require?

Sixty in thirty-six years: thirty purchases and thirty sales, a little under two a year on average but very unevenly spread — five round trips between December 2020 and June 2022.

Do trading costs explain the gap?

Only in part. With no costs at all, the rule ends at ×7.58 against ×10.42: it already loses. At 0.5% per order it drops to ×5.61, at 1% to ×4.15, at 2% to ×2.26.

What if cash earned interest while the rule was out?

At 2% a year the rule reaches ×7.42; at 4%, ×9.75. It would have taken 4.46% a year, net and risk-free, over the 169 months spent out of the market to match gold simply held.

Method and sources

Gold prices (GC=F contract), in US dollars, read from Finamax’s public API and stopped on 2 November 2026. Each month is represented by its reference point, the convention of /api/prices. The moving average is the arithmetic mean of the last ten points, the current month included; a price exactly equal to its average triggers no order.

That reference point deserves a note. Before 2026 the database keeps a single price per month, dated the 1st, and that price is not a quote from the 1st: set against the contract’s daily closes, it coincides with the month’s average until 2010, and with a value close to its close thereafter. Before 2026 the rule therefore decides on a price it could not have known exactly on that day. We reran the calculation executing every decision one month later, which leaves only the past to read: ×7.18 for the rule against ×10.35, and a largest drawdown of −34.85%. The rule does slightly better, and the conclusion does not change.

Both strategies start with the same capital on 1 October 1990 and are valued on the same points. Costs are 0.5% of the amount of each order, purchase and sale alike. Cash earns nothing unless stated otherwise. No tax is deducted.

Annualised return, volatility and largest drawdown are computed by shared/financeMath.js, the module from which the site’s comparator draws its own indicators. The backtest lives in shared/signalBacktest.js, and the report is regenerated with node scripts/backtest-signal.js --rule trend --fee 0.005 --end 2026-11-02.

Past performance is no guide to future performance. This page is an analysis of historical data and does not constitute investment advice.