Gold Demand and Supply
Balance Analysis

In Q3 2022, this line item came in at over 300 tonnes. The entire global ETF outflow that quarter was around 227 tonnes. The residual was bigger than the category it was supposed to be supplementing. Metals Focus, which compiles much of the underlying data for the WGC, doesn't pretend to have full visibility into what drives this number. Neither does the WGC. It represents the net effect of everything that happens in the London OTC market that doesn't get captured elsewhere: sovereign wealth funds building positions through LBMA member banks, family offices buying or selling multi-tonne lots, central bank transactions that haven't been disclosed yet, inventory shuffles between commercial vaults. The plug figure in the balance sheet is, in certain quarters, the largest demand-side or supply-side entry on the page.

Gold's stock-to-flow ratio runs around 60 to 70. Annual mine production of 3,500 to 3,600 tonnes against roughly 210,000 tonnes above ground. Less than 2% per year. Copper's stock-to-flow is around 0.5 to 1. Everyone who writes about gold mentions this. The implication takes longer to absorb: quarterly mine supply fluctuations, the stuff that fills up most sell-side research notes, are analytically close to meaningless for pricing purposes. A 5% swing in annual mine output is a 0.08% change in total available stock. The time and attention the industry devotes to tracking mine production quarter by quarter, mine by mine, is out of all proportion to its explanatory power.

And here the analysis immediately runs into a wall, because nobody knows what fraction of that 210,000 tonnes is in a state where the holder would consider selling at any price. Temple gold in India. Bars in central bank vaults that haven't moved since the 1960s. Rural hoards across Southeast Asia. Sunken ships. All counted in above-ground stock, none of it responsive to price signals.

The rest of this piece spends most of its time on physical flow tracking, because that's where informational asymmetry in gold analysis is widest. The demand-side categories, jewelry, industrial, investment, central banks, are competently covered in every WGC report and every bank research note. Repeating that coverage adds nothing. What's missing from most published analysis is granular attention to where physical metal is actually going, in what form, through what channels, and what the discrepancies between different data sources reveal.

Koos Jansen, when he was publishing regularly on the Bullion Star blog, did more to make Shanghai Gold Exchange withdrawal data accessible to Western analysts than any institutional research outfit. His work tracking SGE weekly withdrawals and cross-referencing them with Chinese import data via Hong Kong customs created a factual foundation for understanding Chinese physical demand that didn't exist in English-language analysis before roughly 2013. The data infrastructure he built is still the template that most independent gold analysts use. The WGC's own China demand estimates diverge from SGE withdrawal data in ways that have never been fully reconciled, and the gap is large enough to matter.

Swiss customs data. Published monthly by the Swiss Federal Customs Administration, broken down by country of origin and destination, in kilograms. Four refineries handle the bulk of global gold processing: Valcambi, PAMP, Argor-Heraeus, Metalor. When 400-ounce London Good Delivery Bars need to become 1-kilogram bars for Asian markets, they transit through these facilities. A surge in UK-to-Switzerland gold flows means London vaults are draining toward Asia. Switzerland-to-India and Switzerland-to-China flows map demand intensity in those markets with a lag of about a month. Nick Laird at Goldchartsrus compiles and charts this data in a way that makes the patterns visible over time. The Swiss customs dataset, combined with SGE delivery data and London vault reporting, provides a triangulated view of East-West gold flow that is far superior to the WGC balance sheet for tracking physical demand shifts in near-real-time.

London vault data comes from two sources that can be cross-referenced productively. LBMA publishes aggregate vault holdings monthly. The Bank of England publishes its own vault data with about a two-month lag. BoE vaults hold overwhelmingly central bank gold and some large institutional custody. Subtracting BoE holdings from the LBMA aggregate gives a rough approximation of what's in commercial vaults, the float available for market transactions. When the LBMA total is declining and the BoE component is stable, commercial vaults are losing metal. That metal is being shipped out, most likely to Swiss refineries for recasting and onward delivery to Asia. This happened persistently through 2022 and 2023, and the commercial vault decline ran well ahead of any price signal.

COMEX vault reporting. This gets less attention than it deserves outside of a fairly small community of analysts (Jesse from Jesse's Café Américain used to track it obsessively, and BullionStar has continued to publish regular breakdowns). COMEX vaults hold gold in two categories: Registered, meaning warranted and available for delivery against futures contracts, and Eligible, meaning the metal meets exchange specifications and sits in an approved vault, not listed for delivery. The ratio between these two categories is a behavioral indicator. A persistent decline in the Registered share means vault holders are actively pulling metal off the deliverable list. They're taking bars that could be delivered against a futures contract and removing that optionality. Between 2020 and 2024, this ratio went through dramatic swings that tracked with periods of physical market tightness far more closely than the gold price itself did. COMEX total vault gold peaked around 40 million ounces in early 2022 and then declined substantially as metal was withdrawn and, in many cases, shipped out of the U.S. entirely.

London's gold market runs on unallocated accounts. The holder has a claim on gold at an LBMA member bank. No specific bars assigned. Unsecured creditor status. The bank holds physical in its vault and runs a book of claims against a much larger notional amount. LBMA daily clearing has historically run in the range of 18 to 20 million ounces per day. London vault holdings are around 25 to 30 million ounces depending on the period. The clearing volume relative to the physical backing is a leverage ratio, and it's high. Ronan Manly at BullionStar has done more detailed work on this than anyone else in the public domain, pulling together LBMA clearing statistics, vault data, and BoE custody figures to estimate the effective coverage ratio. His conclusions have been disputed by LBMA members. The data he uses is the LBMA's own.

March 2020 broke the linkage between COMEX and London prices in a way that most market participants had assumed was impossible. COMEX specifies 100-ounce bars. London's standard delivery unit is the 400-ounce Good Delivery Bar. Normally arbitrage keeps the two markets aligned to within a couple of dollars. In March 2020, with air freight disrupted and Swiss refineries at reduced capacity, the supply of 100-ounce bars couldn't match COMEX delivery needs. The spread blew to $70 at its peak. CME rushed to introduce a new 400-ounce contract (the Enhanced Delivery contract) to bridge the gap. COMEX subsequently built up physical vault holdings to levels far above historical norms. That inventory buildup was a retroactive admission that the previous system had been operating with inadequate physical backing.

Grade decline gets covered thoroughly in every annual gold mining review. Under 1 g/t average now at major operations, down from ~1.5 g/t twenty years ago, irreversible because high-grade ore bodies are preferentially depleted. Discovery-to-production cycles of ten to fifteen years. All of this is standard.

The capital allocation picture is covered less. Gold miners have been generating record free cash flow. Newmont, Barrick, Agnico Eagle, the major names, have used that cash flow overwhelmingly for shareholder returns and acquisitions of existing producing assets. Greenfield development spending, the investment that would eventually become the next generation of mines, has not kept pace with the depletion of existing reserves. S&P Global's mine pipeline data shows a persistent thinning of advanced-stage projects. The reason isn't geological. Plenty of discoveries remain undeveloped. The reason is that the capital markets punish miners who announce large, long-dated capital commitments. Share prices drop on development project announcements. They rise on dividend increases and buyback programs. So management teams, rationally responding to market signals, starve the project pipeline.

The industry's reserve replacement ratio has been running below 1.0 for several years. GFMS and Metals Focus both flag this in their annual surveys. It doesn't drive short-term trading. It drives the ten-year supply outlook.

Russia, 300+ tonnes per year. Post-2022, a substantial share stays domestic. If more producing countries follow resource-nationalist policies, globally tradeable mine supply diverges further from total production. This is a slow-moving variable but directionally clear.

AISC has risen for a decade. The cost floor under gold prices keeps stepping higher. Relevant on multi-year horizons.

1,100 to 1,300 tonnes a year normally, more in sharp price rallies. The behavioral dynamics here don't get enough credit for their complexity. Scrap selling in India and China involves 22K and 24K gold held as family wealth with emotional and cultural attachment. Price has to exceed psychological thresholds and hold there before selling volumes respond. The anchoring effect is strong: a family that almost sold at $2,000 and missed the window doesn't sell at $1,900. They wait for $2,000 again.

Indian demand in particular correlates with monsoon quality, because agricultural income drives rural purchasing power, and rural India is the backbone of physical gold demand in that country. Import duty arbitrage through smuggling (primarily via the Middle East) adds a layer that no official data captures. The WGC India demand estimate minus customs-recorded imports is the standard industry proxy for smuggling volume. Nobody publishes the calculation formally. Everyone in the business knows the gap is there and what it means.

ETF holdings data is the most-watched and most-misinterpreted gold demand indicator. AP hedging mechanics determine whether an ETF inflow hits the physical market. An AP can source physical in London OTC to back new share creation, which is genuine physical demand. Or the AP can hedge via COMEX futures, rolling positions, never touching physical. Same ETF inflow headline, entirely different physical market implication. GLD's prospectus allows for both. Chinese gold ETFs operate through SGE physical delivery, so their flows are fully physically backed.

When Eastern and Western ETFs add simultaneously, the aggregate physical demand impact is not the sum of the flows as reported. The Eastern component is 100% physical. The Western component is some unknown mix.

Net buyers since 2010, buying intensifying post-2020. Channels are opaque. BIS gold desk, direct purchases from domestic miners, undisclosed intermediaries. China's reported reserves consistently lag estimates based on import data and SGE flows. The IMF's IFS database and national self-reports don't agree. The visible portion of central bank demand is a lower bound on actual purchases at all times.

SGE price versus London adjusted for freight and insurance. When the premium is positive and widening, Chinese demand is outrunning supply through normal channels. Arbitrageurs respond by accelerating London-to-Switzerland-to-Shanghai shipments. When the premium flips negative, flow decelerates. This signal updates daily, is anchored by physical arbitrage, carries no methodological lag.

Backwardation in gold, near-month above far-month, is one of the rarest events in commodity markets. April 2013, post-crash, saw a brief instance. It signals that someone needs physical metal so urgently they'll pay more for delivery now than delivery later. The information content of a single day of gold backwardation exceeds that of an entire quarterly supply-demand report.

The 10-year TIPS yield has had a persistent negative correlation with gold prices for twenty years. When real rates fall, gold's opportunity cost falls, price rises. This relationship weakened in 2022-2023 as central bank buying and geopolitical demand introduced a structural bid that was insensitive to rates. The weakening of the TIPS correlation is itself a supply-demand story: a class of demand (official sector) emerged that doesn't respond to the variable (real rates) that previously dominated pricing.

Mine supply: plateaued production, declining grades, capital starvation of greenfield projects, resource nationalism reducing internationally tradeable output. Cost floor rising with AISC.

Demand: central bank buying structurally persistent. Post-ZIRP institutional allocation to gold ratcheted higher and hasn't reversed. Younger cohorts accessing gold through ETFs, tokenized products, DeFi, at near-zero friction. The demand base is widening at the margin.

Deep-sea mining is a variable on a twenty-year horizon. Precious metal deposits exist on the ocean floor outside all reserve estimates. Technology and legal framework are nowhere near commercial viability. Not relevant to any current analysis. Relevant to anyone who claims to think in decades.

Two markets in parallel. Paper: COMEX futures, London unallocated accounts, ETF shares, option books, algorithmically traded, near-infinite liquidity, priced off macro variables and fund flows. Physical: mine output, refinery throughput, shipping logistics, vault inventories, the accumulated decisions of hundreds of millions of individual holders across dozens of countries. Arbitrage holds them together almost all of the time. March 2020, the post-crash scramble of April-May 2013, the persistent Shanghai premium episodes of 2022-2023, these are the moments when the two markets separate. Each time, the divergence has resolved with paper prices adjusting toward the physical reality. Not once has physical capitulated to paper over any sustained period.

Most of what gets published as gold supply-demand analysis serves the function of describing last quarter's weather. The balance sheet is assembled after the fact from incomplete data, published with a lag, and organized into categories that conceal as much as they reveal. Its structural information, the slow shifts in mine supply trajectory, the directional change in official sector behavior, the rising cost floor, is valid on a multi-year basis. Its ability to explain price moves in any given quarter is close to zero, and the residual item is the receipt for everything it can't explain.