Gold  against stocks

Gold was indeed a high-Sharpe asset during certain windows from 2025 to early 2026, and long-term bonds did under perform. However, as of September 7, 2026, gold does not “beat tech stocks” over 12 months. NVDA and the S&P have higher TTM returns than those shown in the table.

The illustrative recalibration (TTM window, $r_f = 4.5%$) yields Sharpe ratios: Gold ≈ 1.16, S&P 500 ≈ 0.91, NVDA ≈ 0.74, Brent ≈ 1.12, US10Y ≈ -0.93. An honest reading shows gold remains excellent against duration, competitive with the S&P, but no longer dominates NVDA/S&P as starkly as previously claimed. Brent, after accounting for real returns, also has a high TTM Sharpe, albeit with higher volatility and geopolitical drawdown. Over five years (more robust window, ETF data), a recent public calculation shows GLD Sharpe ≈ 0.81 versus SPY ≈ 0.52 and TLT ≈ -0.76. Here, gold’s relative superiority over equities and especially long-term bonds is better established.

Gold’s appeal in 2025-2026 is driven by three key factors. Firstly, an inverse duration effect from rising US 10Y yields (~4.3% to ~4.8%) and long JGB yields, making bonds unattractive while gold lacks nominal duration. Secondly, systemic hedging demand, with gold hitting intra-period records near $5,000-$5,500 (spiking January-February 2026), linked to “shadow leverage” from AI/off-balance-sheet debt and private credit stress. Thirdly, a weak USD (periodic DXY declines) and official sector purchases create a scarcity premium.

Using Markov-switching regimes, two states are identified. Regime 1 (“growth/equity”) shows positive gold-SPX correlation. Regime 2 (“stress/duration shock”) sees gold-SPX correlation near zero or negative, with negative gold-TLT correlation. The TTM 2025-2026 mixes these, driving gold’s rally to February 2026, a correction through March-July, and a rebound in August. This explains a misleading Sharpe ratio of 2.71 when anchored to the August peak instead of September 7. Gold’s realized volatility (14-19%) is structurally lower than NVDA (40-60%) or Brent (spiking March-May 2026), offering a genuine risk-adjusted advantage, not a static 38% return.

On the recent window, the Student copula for gold-stocks shows moderate lower tail dependence, not a correlation of 1. Gold is not a perfect intraday hedge; it is more effective for duration and credit shocks. Conversely, the conditional correlation gold-S-ACBI (AI credit stress) remains relevant for NVEO. When the AI credit factor tightens, gold captures part of the hedging flow; BTC does so more unstably (short-squeeze then mean-reversion).

The Markowitz, Black-Litterman, and CVaR models are used for portfolio optimization. The problem is to maximize risk-adjusted returns. Using raw trailing 12-month (TTM) returns, the optimizer overweights gold excessively due to window bias. After recalibrating, an efficient allocation includes 15-25% gold, 40-50% equities, 15-25% short-duration/cash, underweight long-duration bonds, and 5-10% energy. CVaR shows gold reduces drawdowns but doesn’t replace equity risk premiums. Demographic shifts (baby boomer spending peak) support gold as a store of value, but not monotonically.

Gold has been, and remains, the best Sharpe stabilizer against duration and systemic stress. It has not “beaten tech” over the 12 months ending September 7, 2026. A peer in New York or Zurich would dismiss the chart for window bias, omission of $r_f$, and TTM errors on NVDA/S&P/Brent. The portfolio thesis (sleeve or physical + underweight long duration + S-AGRI / S-XSR indices) survives this correction; the figure 2.71 does not.