Two months out from the Congress, the volume of gold price forecasting has become difficult to ignore. Most of it shares a structural flaw: a number is produced, the reasoning behind it is not, and no condition is offered that would cause the author to change their mind. This edition sets out the alternative we use — four drivers, the public data behind each, and the conditions under which we would abandon the whole thing.
One: real yields, not inflation
Gold pays nothing. The entire cost of holding it is the yield you gave up elsewhere, and the relevant comparison is the real yield rather than the nominal one. Falling real yields lower that opportunity cost and gold typically re-rates; durably rising real yields do the opposite.
This is where most retail commentary goes wrong. Gold does not respond to inflation as such. Inflation rising alongside faster-rising nominal rates produces higher real yields, which is historically unhelpful for the metal. If you only ever check one series, check the 10-year Treasury inflation-indexed yield, published daily and free on FRED as DFII10.
The relationship is a tendency, not an identity. It breaks whenever a second driver takes control — and every break since 2022 has been informative rather than anomalous.
Two: the official-sector bid
After roughly two decades as net sellers, central banks turned net buyers around 2010 and have stayed there, with annual net purchases above 1,000 tonnes in 2022, 2023 and 2024 on World Gold Council reporting. What makes this matter more than the tonnage suggests is behavioural: a reserve committee executing a multi-year mandate does not suspend it because spot rallied. It may slow execution; it rarely reverses.
Before quoting any figure, ask two questions. Is it gross or net — because one large seller can mask several determined buyers. And what reporting lag applies, because disclosure is voluntary in practice and revisions are routine rather than suspicious.
Three: flows and positioning
This is the fast driver and the source of most of the volatility that makes gold look erratic. Aggregate ETF holdings publish daily; managed-money net positioning appears each Friday in the CFTC Commitments of Traders report, as of the prior Tuesday. Read together they tell you how crowded a move is — a rally into record net length rests on an unstable base, while a rally with falling ETF tonnage is being driven by something other than Western portfolio demand.
Four: physical demand and the supply response
Gold is a stock-to-flow market: an estimated 216,000 tonnes above ground against annual mine supply of roughly 3,000 to 3,700 tonnes. New production adds around 1.5 to 2 per cent to the pool each year, which is why mine disruption headlines rarely move price and why the behaviour of existing holders matters far more.
Recycling is the elastic part and the reason spikes tend to be self-limiting — scrap has typically supplied a quarter to a third of annual supply and responds within weeks of a price move. For a real-time read on whether physical demand is absorbing or resisting a rally, watch the Shanghai and Mumbai premiums or discounts to the London benchmark.
How the four drivers interact
The reason a four-driver framework beats a single-variable one is that the drivers operate on different clocks and through different channels. Real yields work through opportunity cost and act on the ceiling: while a safe asset pays a compounding real return, there is a level above which price-sensitive Western capital stops adding. The official sector works through absorption and acts on the floor: a reserve committee buying to a mandate removes float from the market irrespective of the level. Flows and positioning determine the path between the two, and physical demand determines how much of a move sticks once the flow stops.
Written that way, most of the apparent contradictions in gold commentary dissolve. A market with a rising floor and an unchanged ceiling compresses into a range and then breaks — which describes the 2013–2019 period reasonably well. A market with a falling ceiling and a rising floor trends, which describes 2019–2020. A market with a rising ceiling and a rising floor grinds higher with violent corrections, which is the more useful description of the present configuration than any narrative about a single cause.
Weighting the drivers by horizon
If you are looking out days to weeks, positioning and flow explain most of the variance and macro explains almost none; this is the horizon on which gold looks irrational to people using a macro model. Over one to four quarters, real yields dominate and positioning becomes noise. Over multiple years, the official-sector bid and the marginal cost of production dominate and both of the faster drivers wash out entirely.
The practical consequence is that you must state your horizon before you interpret any series, because the same data point supports opposite conclusions on different clocks. Record net managed-money length is bearish for the next month and irrelevant for the next five years. Sustained official-sector accumulation is decisive over five years and worthless as a signal for next Tuesday. Most public disagreement about gold is not disagreement about facts; it is two people using the same facts on different horizons.
The drivers disagree, and that is the signal
They do not take turns and they are not additive. Positioning dominates over days, real yields over quarters, the official bid and the cost curve over years. The genuinely useful moments are the disagreements: gold rising while real yields also rise means the standard mechanism is not in charge, and the usual suspects are sovereign accumulation or a currency event. A broken correlation is a prompt to investigate, never proof the framework is broken.
What would prove us wrong
Stated now so it can be checked later. The framework needs rebuilding, not defending, if two of the following hold for two consecutive quarters: sustained 10-year real yields above roughly 2.5 per cent with stable inflation expectations and no gold drawdown; reported official-sector net purchases below roughly 400 tonnes annualised with no change in the character of corrections; deep physical discounts in both Shanghai and Mumbai during a continuing rally; or record managed-money length coinciding with a durable trend rather than a reversal.
Where each driver can mislead you
Real yields: the published series is a US ten-year real yield, and gold is held globally. Long stretches in which euro-area or Japanese real yields sat far below US levels produced gold demand that the US series alone could not explain. Treat DFII10 as the best single proxy, not as the whole world.
Official-sector demand: reporting is voluntary in practice, lagged, and revised. Purchases executed over the counter through the Bank for International Settlements or a small number of bullion banks can be invisible for quarters. A weak print is therefore weak evidence, and a strong print is history rather than news.
Flows and positioning: both are stale on release and both measure consequence rather than cause. ETF tonnage records demand already expressed through the share price; positioning categories are legal classifications rather than motives, so a large short can be a producer hedge or the offsetting leg of an over-the-counter long.
Physical demand: regional premiums are contaminated by local policy. An Indian discount can reflect an import-duty change or the wedding calendar rather than any view on gold, and a Shanghai premium can reflect quota administration rather than retail appetite.
Your monthly page
Once a month, write down four things: the 10-year real yield and its three-month change, the latest quarterly official-sector net purchase figure, aggregate ETF tonnage and its four-week direction, and whether Shanghai and Mumbai sit at a premium or a discount. Then one sentence on which drivers agree and which disagree. Twelve entries later you will own an auditable record of your own reasoning, which is worth more than any target price — including ours.
On the programme
This framework is the spine of the macro and market-structure sessions on 10 October. Participants have been asked to bring the series they actually watch and to state their own falsification conditions on the record. Speakers are announced individually as contracts complete.
Housekeeping
The long-form version of this framework is now published in the Magazine, with the full driver-by-driver treatment and an FAQ. General Admission to the Congress remains free; Pro and VIP places are arranged by manual invoice after you submit the registration form. Nothing in this edition is investment, tax or legal advice.