How to read this. The panels share a timeline but not units. The inflection everywhere is 2022: the year Western governments froze Russia's reserves. Official institutions hold roughly the same ~$3.9 T of Treasuries they held in 2015, so all net growth in foreign holdings has come from private investors, a more yield-driven and price-sensitive bid. Gold's rising reserve value owes more to price than to tonnage, and the price panel is the multiplier behind the crossover.

What is estimated. The reserve-composition panel puts everything over one denominator, total global reserves with gold at market value: currency lines are COFER shares of allocated FX reserves scaled by (1 − gold share); the gold-share series is interpolated between published anchors (~10% through the 2010s, 16.5% in 2023, 20% in 2024, 27% at end-2025, per ECB/WGC). The "nontraditional tail" is everything outside the dollar, euro, yen and sterling (AUD, CAD, CNY, KRW, Nordics, SGD…, per Arslanalp–Eichengreen). Note the 2025 wrinkle: the tail's share of *total* reserves dips even as its share of FX reserves grows, because gold's revaluation expands the denominator: the solvency layer crowding the working-capital layer. The gold-value series are computed as tonnes × year-end spot price. World official tonnage is interpolated between IMF/WGC totals (32,700 t in 2015, ~35,000 t in 2020, ~36,340 t at end-2025) minus the constant US holding of 8,133 t, so the world series matches the scope of the TIC "foreign official" Treasury series. China's dashed marker uses Jan Nieuwenhuijs's estimate that the PBoC actually holds ~5,411 t. The PBoC buys abroad without customs traces, and China treats true reserves as a state secret. The in-country stock chart reflects that China prohibits exporting domestically mined and imported bullion: what enters China stays. The USA panel counts metal physically on US soil regardless of owner: ~5,900 t in the New York Fed's vault belongs to 36 foreign central banks and international institutions, and the repatriation trend runs against the US: France alone moved 129 t out during 2025–26, and some 6,900 t has gone home from foreign custody since 1972. The US private figure is a rough household-survey-based estimate; ETF gold (mostly vaulted in London) is excluded. Where China's in-country stock only accumulates, the US stock can leave. London completes the picture: the LBMA's vault statistics show 9,392 t in the city (May 2026): around 5,000 t custodied at the Bank of England mostly for foreign central banks, an estimated ~2,500 t backing gold ETFs (≈60–65% of the world's ~3,930 t of ETF gold, including SPDR GLD at HSBC's vault), and the rest as commercial trading stock. The UK itself owns just 310 t. London is a warehouse for everyone else's metal. Switzerland is the inverse of China: the SNB's 1,040 t hasn't changed since 2008 (and only ~70% of it sits in Switzerland: 20% is at the Bank of England, 10% in Ottawa), yet the price surge alone quadrupled its value. The Swiss total combines the SNB with the University of St. Gallen/philoro estimate that households privately hold ~920 t in bars, coins and jewellery (their narrower 2024 survey, excluding jewellery, found ~200 t) and 341 t in Swiss gold ETFs, about 255 g of gold per resident, among the highest anywhere. Not counted: metal Swiss banks and freeports vault for foreign clients, which is real but unquantified in a country that refines ~two-thirds of the world's gold. Swiss-held Treasuries (TIC) include private custody, not just the SNB.

How to read this. Every series is a price divided by the gold price in the same currency at the same date, so it answers one question: how much metal would you have had to hand over? Falling lines mean gold buys more. The striking feature of the American panel is that all five lines have the same shape, peaking in 2001 and bottoming in 2026, because the shape is gold’s own price history, not five separate stories about houses, cars and food. 2001 was the end of gold’s twenty-year bear market, the year the UK finished selling half its reserves near $275/oz.

The finding that cuts against a gold-bug reading. Over thirty years gold rose 10.4× and the S&P 500 rose 10.1×. The median new US house rose 2.9× and median household income 2.4×. So gold did not outrun everything. It kept pace with US equities while both outran wages and consumer goods by roughly four times. Gold’s buying-power gain is mostly a story about financial assets repricing against the real economy. On that reading gold has done its job of preserving claim on real goods, without being an especially good investment relative to simply owning the index.

Why Australia is different. In AUD terms gold rose 11.5× since 1996, slightly more than in USD. But the Sydney median house rose 7.4×, against 2.9× for the US median new house. Australian housing absorbed most of what gold gained, so a Sydney buyer holding gold since 1996 improved their position by about 1.6× where an American improved by 3.5×. Gold has been a far weaker hedge against Australian property than against American property, and anyone buying gold in Australia as a housing hedge should price that in. The conclusion survives the source ambiguity below: substituting Domain’s higher Sydney figure (~$1.79m) still leaves only a 27% improvement.

What is spliced, estimated or mismatched. The 2026 column mixes vintages, because series publish on different clocks: gold is spot on 20 Jul 2026, houses are Q1 2026, cars June 2026, the S&P 17 Jul 2026, tuition the 2025–26 academic year, while groceries and income are 2024, and the latest the BLS and Census have published. Using the current $4,019 spot is the conservative choice: at January 2026’s $5,595 peak the median house was 72 oz, not 100. Houses: the US series is Census/HUD MSPUS, which covers new houses only and is mix-distorted, because builders have shifted to smaller homes, which is why it shows just 2.9× over thirty years; the NAR existing-home median rose about 3.5× over the same period, so the true US improvement is somewhat smaller than plotted. Cars: there is no continuous US series. 1996–2006 is CNW Research via BTS; 2011 onward is Edmunds/KBB transaction prices. The line is drawn with a deliberate break at that join rather than chained through it. No Australian average-car series exists at all. The "$47,462 Australian average" circulating online is Cox Automotive’s US figure in USD. Sydney: the weakest series here. 1996 and 2001 are Abelson & Chung’s reconstruction (~4% below the ABS basis); 2006–2021 are ABS 6416.0 unstratified medians, itself rebased onto CoreLogic data in 2014; and 2026 is a Cotality hedonic median value, which models the whole dwelling stock rather than actual sales. Three current "Sydney median house price" figures span about $230,000, so treat the level as indicative and the trend as the finding. Bitcoin is year-end close where the other series are annual averages, and its 2011 value comes from a market trading under 100 BTC a day.