Skip to content

The Golden Dilemma

Authors: Claude B. Erb, Campbell R. Harvey | Year: 2013 | Journal: Financial Analysts Journal, 69(4), 10-42

Thesis

Gold has no cash flows, no earnings, and no coupon. Its expected real return should therefore be approximately zero over the long run. The "golden dilemma" is that gold's popularity as an inflation hedge and safe haven is largely unsupported by rigorous empirical evidence. Erb and Harvey show that gold's real price has mean-reverted over very long horizons, that gold is a poor short-run inflation hedge (correlation with unexpected inflation is near zero at monthly/quarterly frequencies), and that the gold-to-CPI ratio is the single best predictor of subsequent 10-year real gold returns. The implication: at elevated gold/CPI ratios, forward returns are poor.

Key Math

The core valuation framework rests on the gold-to-CPI ratio as a mean-reversion anchor:

\[R_{t \to t+10}^{\text{real}} = \alpha + \beta \cdot \ln\!\left(\frac{P_t^{\text{gold}}}{\text{CPI}_t}\right) + \epsilon_t\]

Where \(\beta < 0\) (estimated around \(-0.85\)), indicating that high real gold prices predict low subsequent real returns. They also decompose gold futures returns into:

\[r_{\text{futures}} = \underbrace{\Delta S}_{\text{spot return}} + \underbrace{(F - S)/S}_{\text{roll yield}}\]

Roll yield for gold is persistently negative (contango), dragging total futures returns below spot appreciation.

Data & Method

  • Monthly gold prices 1975-2012, extended with historical data back to 1791 for the real price series.
  • CPI-U for deflation. Gold/CPI ratio tested as predictor via OLS with Newey-West standard errors (overlapping 10-year windows).
  • Cross-sectional comparison of gold vs. TIPS, commodities, equities, and bonds as inflation hedges using rolling correlations at 1m, 1y, and 5y horizons.
  • Convenience yield model from futures term structure (COMEX front and deferred contracts).

Our Replication Verdict

CONFIRMED -- The gold/CPI ratio remains a robust long-horizon predictor. Extending the sample to 2025 does not break the relationship; the ratio peaked in 2011 at ~8.5x and again approached elevated levels in 2024. Short-run inflation-hedging failure is well-documented and replicated in our data. Caveat: The predictor requires 5-10 year horizons to be useful -- it is useless for tactical trading at weekly/monthly frequencies. Also, structural demand shifts (central bank buying post-2022) may create regime changes the historical model cannot capture.

Signal Mapping

  • Valuation overlay for the Strategic Allocation layer. The gold/CPI ratio feeds into a regime classifier that adjusts position sizing: when the ratio exceeds 2 standard deviations above its rolling 20-year mean, strategic gold allocation is reduced.
  • NOT used as a tactical signal -- horizon is too long. Consumed by the portfolio construction module as a risk budget input.
  • Related to carry signal: the contango/backwardation state from futures term structure feeds the Carry strategy (SS5.2).

References

  • Erb, C.B. & Harvey, C.R. (2013). "The Golden Dilemma." Financial Analysts Journal, 69(4), 10-42. DOI: 10.2469/faj.v69.n4.1
  • Erb, C.B. & Harvey, C.R. (2016). "The Golden Dilemma Revisited." Working paper.
  • Barro, R.J. & Misra, S. (2016). "Gold Returns." Economic Journal, 126(594), 1293-1317.
  • O'Connor, F.A. et al. (2015). "The Financial Economics of Gold -- A Survey." International Review of Financial Analysis, 41, 186-205.