Gold prices are driven primarily by central banks', ETFs', and speculators' willingness to hold the asset, rather than by the traditional balance of production and consumption

Overview

The report systematically reviews the structure of global gold trading, supply characteristics, and holder behavior, and establishes a pricing framework based on three types of conviction-based holders: ETFs, central banks, and speculators. Its central conclusion is that gold is hardly consumed, while new mine supply is stable and price-inelastic. Prices therefore depend primarily on who is willing to hold the existing gold stock and what price is required to induce incumbent holders to sell.

Core views

London and New York are the principal centers of global gold price formation. London is the center of the physical gold market, where the over-the-counter market primarily trades large 400-ounce bars weighing approximately 12.5 kilograms. Participants include central banks, sovereign wealth funds, and bullion banks; most of the gold backing physical gold ETFs worldwide is also stored in London vaults. New York's COMEX, by contrast, is a speculative paper-gold market, with futures contracts denominated in 100-ounce units weighing approximately 3 kilograms. They are usually settled financially, with relatively little physical delivery. The two markets are connected through the New York–London price spread via the Exchange for Physical, or EFP, mechanism: traders can buy spot gold in London and sell futures in New York, or execute the reverse trade, with arbitrage bringing prices back into alignment. This connection depends on Swiss refining capacity and cross-border transportation. When COMEX requires delivery, London's 400-ounce bars are generally shipped to Switzerland, recast into 100-ounce bars, and then transported to the United States. In March 2020, the shutdown of Swiss refineries and suspension of commercial flights caused the price spread between the two markets to widen sharply. In late 2024, concerns that the United States might impose broad tariffs on imports including gold again created delivery risks, causing New York prices to exceed London prices. Traders therefore shipped gold to New York in advance, with some bars even bypassing Switzerland and traveling directly from the United Kingdom to the United States. Switzerland also performs the format conversion between institutional and retail markets: when ETFs or central banks buy, small bars are melted and cast into 400-ounce bars before being shipped to London; when institutional demand weakens, bars flow from London to Switzerland, where they are recast and shipped to New York or retail markets such as China and India. Gold differs from commodities such as oil and natural gas, which are consumed. Almost all gold ever mined still exists, with an above-ground stock of approximately 220,000 tonnes. Annual new production amounts to only slightly more than 1% of this stock and is generally stable and insensitive to prices. Gold mining has high fixed costs, making it difficult to cut production rapidly in bear markets or simply increase output in bull markets. Mine production must be planned years in advance and is constrained by processing capacity and safety requirements; a new mine may take decades to progress from discovery to production. Ore grades have also declined from approximately 12 grams of gold per tonne of rock in the 1950s to around 3 grams currently, meaning that producing the same amount of gold requires more rock, energy, labor, and capital. Of the existing gold stock, approximately 22% is locked in investment vaults and approximately 17% is held as central bank reserves. Much of the remainder exists as jewelry, particularly in emerging markets. In these markets, gold jewelry is viewed more as household savings than as mere ornamentation. Because gold is not consumed, the traditional “production minus consumption” supply-demand model cannot adequately explain its price. The market clears primarily through transfers of the existing stock among holders, making the identity of the marginal buyer more important than the small amount of new production. The report accordingly divides buyers into two categories. “Conviction-based buyers” include central banks, gold ETFs, and speculators. They allocate to gold based on macroeconomic views or hedging needs, are relatively price-insensitive, and determine the price trend. “Opportunistic buyers” are primarily emerging-market households, which buy when prices are attractive and withdraw when prices rise but rarely become net sellers. They therefore mainly provide support during declines and resistance during rallies, affecting the magnitude of volatility rather than the direction of the trend. Export restrictions in emerging markets also make it difficult for household gold holdings to return to the international market. Goldman Sachs' empirical model shows that conviction-based flows—including central bank and ETF flows and COMEX managed-money net-long positions—can explain approximately 70% of monthly gold price movements. The rule of thumb is that net purchases of 100 tonnes of gold by conviction-based holders correspond to an approximately 1.7% increase in the gold price, with a 95% confidence interval of 1.5% to 2%. The monthly ordinary least squares regression sample covers January 2014 to August 2024, while another related sample covers January 2007 to May 2025. The report summarizes pricing pressure as “conviction-based net purchases divided by mine supply.” Because mine supply is stable, changes in conviction-based net purchases explain nearly all monthly price direction. The model residuals correlate with emerging-market economic growth, suggesting that wealth growth may raise the price floor or ceiling but cannot explain the direction of the trend. Gold also does not follow the ordinary commodity principle that “high prices cure high prices,” because high prices neither stimulate mine supply rapidly nor necessarily induce existing holders to sell. A price rally ends only when conviction-based buyers' reasons for holding gold weaken. The report uses the 1970s as an example: after the US dollar ceased to be convertible into gold at $42 per ounce in 1971, the monetary system shifted from collateral backing to reliance on policy credibility. Inflation subsequently rose, real interest rates turned deeply negative, and gold increased fivefold over the decade. The 1979 Iranian Revolution and the United States' freezing of the Iranian central bank's reserves further weakened confidence in the neutrality of dollar reserves, driving gold to $850 per ounce in January 1980, approximately 20 times its former price in less than a decade. Gold declined by more than 20% only after Volcker raised interest rates sharply and restored credibility in price stability. This history illustrates that gold more directly hedges the credibility of monetary and geopolitical institutions rather than mechanically tracking inflation. The three types of conviction-based flows have different drivers. Physically backed gold ETFs represent demand from long-term Western investors. Their holdings change slowly and are sensitive to interest rates because gold itself pays no interest. Goldman Sachs' vector autoregression analysis shows that a 25-basis-point cut in the US policy rate can gradually generate approximately 60 tonnes of ETF demand over 6 months. Capital generally shifts only after rates have actually declined rather than trading ahead of the move. During recessions or periods of stress, ETF holdings may also remain significantly and persistently above levels implied by interest rates. The so-called “interest rate–gold” relationship primarily arises from the close relationship between ETF holdings and interest rates; this relationship became more apparent after gold ETFs expanded Western investor participation in the mid-2000s. The apparent breakdown of the “interest rate–gold” relationship after 2022 resulted from divergent flows among different buyers. High interest rates did reduce gold ETF holdings, but central bank purchases increased approximately fivefold, enough to outweigh ETF outflows and reset the level of gold prices. Central bank gold purchases usually occur in long-term cycles: central banks increase gold allocations when the monetary neutrality of other reserve assets is questioned due to fiscal sustainability concerns or when their geopolitical neutrality is impaired by sanctions risk. If gold is stored domestically, it does not depend on an issuer's creditworthiness and is difficult for foreign governments to freeze. The financial crisis and the subsequent quantitative and fiscal easing deepened concerns about US fiscal sustainability, turning central banks from net sellers of gold into net buyers. After approximately $300 billion of Russian reserve assets were frozen in 2022, the pace of central bank gold purchases accelerated sharply, particularly among emerging-market central banks. The report views this as a direct response to the risk that reserve assets can be frozen. The urgency of central bank gold purchases may decline only if the credibility of the reserve system is restored or a new dominant currency establishes credibility, causing the opportunity cost of holding less liquid, low-yielding gold to once again exceed its strategic value. Because central banks may not disclose their gold purchases promptly or completely, Goldman Sachs uses UK customs data to produce real-time estimates of reported and unreported central bank demand. The rationale is that the United Kingdom itself has no significant gold mines or accredited refineries producing qualified 400-ounce bars, so gold traded in London's over-the-counter market must be imported. Changes in London vault holdings align almost exactly with the United Kingdom's net gold imports, indicating that UK statistics may include some sovereign transactions. The real-time estimate consists of two components: first, UK gold-bar exports to countries other than Switzerland, used as a proxy for gold shipped directly to central banks' domestic vaults; and second, the gap between UK-reported exports to Switzerland and Swiss-reported imports from the United Kingdom, used to identify central bank gold routed through or stored in Switzerland. Since mid-2022, the first type of flow, particularly the portion destined for emerging markets, has increased at an unprecedented rate. For China, this estimate follows a timing pattern similar to the People's Bank of China's reported purchases but generally begins earlier, reaches higher levels, and persists longer. Speculative positioning represents “fast money” that is established and unwound rapidly. COMEX managed-money net positions fluctuate around their long-term average and tend to mean-revert after large increases or decreases. The report therefore views extreme speculative positioning as a source of short-term volatility around fair value anchored by ETFs and central banks rather than as a persistent change in value. Ahead of major events with a wide distribution of potential outcomes, speculators may temporarily allocate to gold, as before the 2016 Brexit referendum and the 2024 US election. Once the outcome becomes clear, positions often decline and gold may be sold. Speculative capital is also highly sensitive to news. During sharp equity-market declines, gold's high liquidity can make it a source of funds for meeting margin calls, causing prices and positions to fall initially. If uncertainty persists, gold may subsequently rebound; the report cites the 2008 global financial crisis, the March 2020 pandemic shock, and the April 2025 “Liberation Day” as examples. China and India are major retail gold markets, but import controls can cause local prices to diverge from the London benchmark. China manages retail gold imports through quotas; government gold purchases are not subject to these quotas, and gold cannot be exported after entering China. India uses relatively high import tariffs to raise import costs and suppress demand while also restricting some gold exports. Depending on local supply and demand conditions, these barriers generate premiums or discounts relative to London prices. The UAE is more of a growing trading hub than a major center of end demand, and gold there often trades at a discount to London, partly because of different sourcing standards that may permit Russian mined gold, which has been prohibited from the London market since July 2022.

Analysis framework

The report first explains how gold is traded through physical and futures flows among London, New York, and Switzerland. It then uses the enormous above-ground stock and stable, inelastic mine supply to explain why the traditional commodity supply-demand model does not apply. It subsequently divides holders into conviction-based and opportunistic groups, quantifies the effect of conviction-based flows on prices through a market-clearing identity and monthly regressions, and separately examines ETFs' interest-rate sensitivity, the fiscal and geopolitical drivers of central bank purchases, and the mean-reverting characteristics of speculative positioning. Finally, it uses UK and Swiss customs data to construct a real-time estimate of central bank gold purchases and explains how import barriers in markets such as China and India cause local prices to diverge from the global benchmark.

Methodology notes

• Industry/Sector Analysis FrameworkSupply-demand framework

  Market-clearing framework based on transfers of the existing gold stock

  Rather than using the ordinary commodity model of production minus consumption, the report matches conviction-based and opportunistic purchases against sales from the existing stock and mine supply, explaining gold prices through transfers of ownership among holders.

• Quantitative/Factor/Portfolio Theory

  Monthly ordinary least squares regression

  The report uses monthly percentage changes in gold prices from January 2014 to August 2024 as the dependent variable and central bank flows, gold ETF flows, and COMEX managed-money net-long flows as explanatory variables to estimate their marginal price impact and explanatory power.

• Quantitative/Factor/Portfolio Theory

  Vector autoregression model

  The report uses a vector autoregression model to estimate the dynamic cumulative effect of a decline in the US policy rate on Western gold ETF holdings, controlling for interest-rate expectations using the US two-year Treasury yield as a proxy.

• Quantitative/Factor/Portfolio Theory

  Real-time estimate of central bank gold purchases using UK customs data

  The report estimates central bank gold purchases that may not have been disclosed promptly using UK gold-bar exports to destinations outside Switzerland and the statistical gap between UK exports to Switzerland and Swiss imports from the United Kingdom.

• Quantitative/Factor/Portfolio Theory

  Mean-reversion analysis of speculative positioning

  The report observes that COMEX managed-money net positions fluctuate around their long-term average and accordingly treats extreme positioning as a source of short-term volatility rather than a change in the long-term fair value determined by ETFs and central banks.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Gold

  The report's core research asset, whose price is dominated by conviction-based net purchases from central banks, gold ETFs, and speculators, and moderated by opportunistic demand from emerging markets.

  Strengths

  Mine supply grows slowly and is price-inelastic; as neutral collateral that does not depend on issuer credit, gold can play a role when the credibility of the reserve system declines or sanctions risk rises.

  Weaknesses

  Gold itself pays no interest, and ETF demand can be suppressed by higher policy rates and opportunity costs; compared with higher-yielding reserve assets, gold has lower liquidity and yield.

  Comparison

  Unlike commodities such as oil and natural gas that are consumed, gold is primarily stored and transferred, so the traditional production-consumption supply-demand model has limited explanatory power.

  Risks

  Extreme speculative positioning, the resolution of major events, equity-market margin pressure, and disruptions to London-New York physical transportation or tariffs may all cause short-term price volatility.

Key data

• Above-ground gold stockApproximately 220,000 tonnesAlmost all gold ever mined still exists.
• Annual new mine supplySlightly more than 1% of the existing stockProduction is relatively stable and price-inelastic.
• Share held in investment vaultsApproximately 22%Estimated share of the total above-ground gold stock.
• Share held as central bank reservesApproximately 17%Estimated share of the total above-ground gold stock.
• Change in ore gradeDeclined from approximately 12 grams/tonne in the 1950s to approximately 3 grams/tonne currentlyThis indicates that producing the same amount of gold requires more rock, energy, labor, and capital.
• Explanatory power of conviction-based flows70%The proportion of monthly gold price movements explained by central bank, ETF, and speculative flows.
• Price sensitivity to flowsNet purchases of 100 tonnes correspond to an approximately 1.7% increase in the gold priceThe 95% confidence interval is 1.5% to 2%.
• Monthly regression sample periodJanuary 2014 to August 2024Used to estimate the impact of central bank, ETF, and speculative flows on monthly gold price movements.
• ETF demand associated with an interest-rate declineA 25-basis-point rate cut corresponds to approximately 60 tonnesDemand is expected to form gradually within 6 months.
• Change in central bank gold purchases after 2022Increased approximately fivefoldThe scale outweighed the contemporaneous decline in ETF holdings caused by high interest rates.
• Frozen Russian reservesApproximately $300 billionFrozen in 2022; the report views this as a major trigger for accelerated gold purchases by emerging-market central banks.
• London physical gold-bar specification400 ounces, approximately 12.5 kilogramsThe primary specification for London's over-the-counter market and large institutional transactions.
• COMEX futures unit100 ounces, approximately 3 kilogramsThe trading and delivery specification in the New York futures market.
• 1971 US dollar-to-gold conversion price$42/troy ounceThe price benchmark before the US dollar ceased to be convertible into gold.
• Gold price in January 1980$850/troy ounceIt rose to approximately 20 times its former price in less than a decade; gold declined by more than 20% after Volcker restored credibility in price stability.

Impact & implications

The report argues that the key to determining gold's direction is not forecasting short-term increases or decreases in mine production, but estimating changes in the net holdings of central banks, ETFs, and speculators. Actual US rate cuts can gradually support ETF demand; concerns about fiscal sustainability, sanctions risk, and declining credibility of the reserve system can drive long-term central bank purchases; and speculative positioning mainly amplifies short-term volatility. Emerging-market household wealth and local import policies can alter price support, resistance, and regional premiums or discounts, but usually do not determine the global trend. Because supply does not respond rapidly to high prices, a decline in gold prices is more likely to require a change in conviction-based buyers' macroeconomic or hedging rationale.

Risks

• Speculative positioning may mean-revert after reaching extreme levels, causing rapid short-term reversals in prices and holdings.
• Once the outcome of a major uncertain event becomes clear, gold positions established beforehand may be unwound and trigger selling.
• During sharp equity-market declines, investors may sell highly liquid gold to meet margin requirements, causing gold prices to fall initially.
• When Swiss refining capacity, commercial transportation, or import tariffs are disrupted, the price linkage between New York and London may temporarily break down and widen the EFP spread.
• If the credibility of the reserve system is restored, the opportunity cost of central banks holding low-yielding, less liquid gold may outweigh its strategic value, reducing the urgency of gold purchases.

What to watch

• Track actual changes in the US policy rate and the subsequent response of Western physically backed gold ETF holdings over the following 6 months.
• Track disclosed central bank gold purchases and potentially undisclosed demand reflected in UK gold-bar exports to destinations outside Switzerland.
• Monitor the statistical gap between UK exports to Switzerland and Swiss imports from the United Kingdom to assess central bank gold routed through or stored in Switzerland.
• Track COMEX managed-money net-long positioning relative to its long-term average and changes in positioning before and after major events.
• Monitor whether fiscal sustainability concerns, sanctions risk, and the neutrality of reserve assets continue to support central bank gold purchases.
• Track China's import quotas, India's import tariffs, and local gold premiums or discounts relative to the London benchmark.