Buy gold on 1 August, sell it on 1 March: thirty-six years put to the test

The rule, word for word

Every 1 August, buy spot gold. Every following 1 March, sell it and leave the cash idle until the next 1 August. Then start again. No conditions, no thresholds, no exceptions: that is the whole point of a calendar rule, and it is what makes it testable.

The first purchase falls on 1 August 1990, the last sale on 2 March 2026. Between the two, thirty-six completed cycles. A thirty-seventh is open — gold was bought on 3 August 2026 at $4,033.70 — but it is not closed, and a position you have not sold has no return yet: it is left out of the count.

The comparison that matters is not “does gold rise between August and March?” but “is doing this better than doing nothing?”. Both strategies therefore start from the same price on 1 August 1990, stop at the same one on 2 March 2026, and read exactly the same quotes. The seasonal one simply ignores what happens during the five months it sits out.

The season looks real… on average

At first glance it works. Of the thirty-six cycles, twenty-four are winners — two in three. The average cycle returns +6.07%, the median one +3.59%. The five months left out return only +2.56% on average and +0.28% at the median: barely one positive stretch in two, eighteen out of thirty-five.

Annualised, the seven months held are worth +10.62% and the five others +6.26%. The gap does point the way the rule claims. The trouble lies elsewhere: +6.26% a year is not nothing, and it is precisely what the strategy declines to collect.

Over the 427 months of the period it is invested for only 252 — seven months in twelve. For the other 175, the capital does no work. An asset that rises on average in both seasons always punishes whoever steps out of one of them, even the weaker one: that is the arithmetic of compounding, and it is the whole distance between ×6.55 and ×13.42.

Cash would have needed to pay 5% a year

The objection writes itself: between March and August the money need not sit idle. Parked somewhere that pays interest, it earns something, and the sums change.

They change less than one hopes. At 2% a year on cash, the strategy reaches ×8.75 and +6.28% a year — still behind. Break-even sits at 5.04% a year, net, earned without risk, five months a year for thirty-six years. Such a rate existed in the early 1990s and vanished for the decade that followed the 2008 crisis; averaged over the period, risk-free dollar rates stayed well below it.

And even that threshold flatters the rule, because it ignores what applying it costs: thirty-six round trips, seventy-two orders, each with its spread between buying and selling price, its brokerage or custody fees, and a sale that is almost everywhere a taxable event. None of that is deducted from the figures on this page. The real gap is therefore wider than the one shown here, not narrower.

And the season may not be one at all

So far the rule was losing the race, but the gap between the two seasons looked real. It does not survive inspection.

Month by month, gold gains an average of 0.822% during the seven “season” months and 0.493% during the other five. The gap is three tenths of a point a month — against standard deviations of 3.89% and 4.12% respectively. In other words, the spread of an ordinary month is twelve times larger than the effect being hunted. A Welch test on the difference between the two means gives a t of 0.83: nowhere near the level at which chance can be ruled out. Across thirty-six years of prices, this season is indistinguishable from noise.

Spread over time, the picture is blunter still. The average cycle is worth −1.56% in the 1990s, +13.90% in the 2000s, +1.22% in the 2010s and +13.81% since 2020. The rule did not work every other decade: it worked during gold bull markets, and sat flat the rest of the time. In both decades where it shone, the out-of-season months were positive too — +4.94% and +6.54% on average per stretch. What the strategy captured was not a season: it was a trend, and it captured seven twelfths of it.

One last detail, and it is a nice one: August is the least reliable of the seven. It is the only one with a negative median (−0.25%), the one with the fewest rises — sixteen out of thirty-six — and its +0.25% average rests on a handful of exceptions rather than on the typical month. Gold’s three strong months over the period are December (+1.76%), January (+1.29%) and September (+1.15%). Buying on 1 August means stepping in just before the most uncertain stretch of the very season you are trying to exploit, to catch its heart a month later.

What the rule actually buys: calm

It is not without effect, and its effect lies somewhere other than where it is advertised. Out of the market five months a year, the seasonal portfolio takes fewer knocks: annualised volatility of 10.45% against 13.81%, and a largest drawdown of −27.04% against −40.68%.

Both suffered from the same episode — gold’s long slide between 2011 and the end of 2015 — but not to the same depth. Whoever stayed invested watched more than 40% of their capital melt away and waited years to see their starting point again; the seasonal investor lost a quarter. For anyone who measures risk by what has to be endured without selling, that is not a detail.

But even on that ground the rule does not come out ahead: set against its volatility, its annualised return gives 0.52, versus 0.55 for buy-and-forget. It cuts risk and return in roughly the same proportion. Anyone wanting exactly that gets it more simply by holding less gold.

The two cycles worth remembering

The best is 2007: bought at $664.12 on 1 August 2007, gold was worth $968.42 on 1 March 2008 — +45.82% in seven months, the flight to gold in the first tremors of the financial crisis. The next one, August 2008 to March 2009, still returns +10.13% while equity markets collapse.

The worst is 2020: bought at $1,967.00 on 1 August 2020, within days of the all-time high of the era, sold at $1,731.00 on 1 March 2021, −12.00%. A calendar rule does not look at the price: it buys the peaks too.

The twelve losing cycles are 1990, 1991, 1992, 1996, 1997, 2000, 2011, 2012, 2013, 2014, 2016 and 2020. Five in a row, near enough, between 2011 and 2016: when gold falls, the season protects nothing.

All thirty-six cycles, one by one

What this test does not say

It says nothing about what gold will do. Thirty-six cycles and four hundred and twenty-seven months are enough to rule out a clear effect, not enough to establish a faint one: if gold has a seasonality, it is smaller than thirty-six years of history can show.

Nor does it say anything about other calendar boundaries. We tested the rule as it is stated — 1 August, 1 March — and not the other hundred and thirty-one pairs of months available. Hunting for the best of them over the same thirty-six years would certainly turn up one that beats buy-and-hold: that would be the result of the search, not of the market. Which is exactly why we did not go looking.

Finally, everything is computed in US dollars, the currency gold is quoted in. For an investor who counts in euros or pounds, both strategies would also carry currency movements — at different moments, since they do not enter and leave on the same dates.