Chapter 12: Gold & Silver Deep Dive

The definitive guide to precious metals trading. What the research actually says, what actually works, and where our edge comes from.

Reading time: 30 min | Difficulty: Advanced | Prerequisites: Chapters 5-6, 10-11

Why Gold Moves

Forget the narratives on CNBC. Gold price movements are dominated by three forces, and the academic literature is remarkably clear on this.

Force 1: Real Interest Rates (The Dominant Driver)

Barsky and Summers (1988) established the framework that still holds: gold is an alternative store of value to interest-bearing assets. When real rates (nominal rate minus inflation) fall, the opportunity cost of holding gold drops, and gold rallies. When real rates rise, gold sells off.

The relationship is approximately:

ΔGold ≈ -12 × ΔReal10Y + noise

A 25 bps decline in 10-year TIPS yields corresponds to roughly a 3% gold rally. This is not a theory — it is an empirical regularity that has held from 2003 to present with an R-squared of approximately 0.45 on monthly data.

Where to watch it: The 10-year TIPS yield (FRED series DFII10) is the single most important number for gold. When TIPS yields are falling and below 1%, gold is in a structural bull market. When TIPS yields are rising above 2%, gold faces persistent headwinds.

The nuance: The relationship weakens during extreme fear events (gold rallies regardless of rates in 2008, 2020) and when central bank purchasing is anomalously large (2022-2024, when gold rallied despite sharply rising real rates — China and emerging market central banks were buying 1,000+ tonnes/year).

Force 2: Safe Haven Demand (The Episodic Driver)

Baur and Lucey (2010) formalized what everyone suspected: gold acts as a safe haven during equity market stress, but only during extreme events (beyond 2.5 standard deviation equity drawdowns). During normal market fluctuations, gold's correlation with equities is essentially zero.

What this means for trading: - In calm markets, gold is driven by rates and the dollar. Equity moves are irrelevant. - During crashes (VIX > 30, S&P drawdown > 10%), gold decouples from rates and trades as pure fear. The safe-haven bid can push gold 5-15% in days. - The safe-haven effect is strongest in the first 15 trading days of a crisis, then fades. Baur and McDermott (2010) showed gold is a safe haven, not a hedge — the distinction matters for portfolio construction.

Force 3: Central Bank Demand (The Structural Driver)

Since 2010, central banks have been net buyers of gold after two decades of net selling. The pace accelerated dramatically after 2022:

Year	Central Bank Net Purchases (tonnes)	Key Buyers
2019	650	Russia, China, Turkey
2020	273	(Reduced by COVID)
2021	463	Thailand, India, Uzbekistan
2022	1,082	China (PBOC), Turkey, India
2023	1,037	China, Poland, Singapore
2024	~1,000+	China, India, diverse EM

This is a regime shift. Central banks are diversifying away from USD reserves at a pace not seen since the 1960s. The annual demand (1,000+ tonnes) represents roughly 25% of annual mine supply. It puts a structural floor under gold prices.

Trading implication: Central bank buying is slow and price-insensitive. It does not create short-term trading signals, but it means the "fair value" model for gold should include an upward drift term of roughly 2-4% per year from structural demand alone.

Other Drivers (Real but Secondary)

US Dollar (DXY): Gold and the dollar are negatively correlated (~-0.4 on monthly data) because gold is priced in USD. Dollar weakness makes gold cheaper for non-US buyers. But the dollar is often a proxy for real rates — when you control for TIPS yields, the residual dollar effect is smaller than commonly believed.

ETF Flows: GLD holdings track investor sentiment. When GLD holdings rise sharply (> 20 tonnes/week), it confirms a trend. When they diverge from price (gold rising but GLD holdings falling), it signals the rally may be driven by futures speculation rather than broad-based demand.

Geopolitics: Wars, sanctions, and geopolitical crises drive short-lived spikes (1-5 days) that typically mean-revert within 2-3 weeks unless they trigger a broader macro shift (e.g., the Russia-Ukraine war in 2022 triggered both a geopolitical bid AND central bank reserve diversification, creating a durable move).

Why Silver Moves

Silver is not gold. It is a hybrid — part precious metal, part industrial commodity. Batten et al. (2010) showed that silver responds to both monetary factors (like gold) and industrial demand factors (like copper), with the relative importance shifting by regime.

The Dual Personality

Silver Demand Breakdown (2023, Silver Institute):
  Industrial:        54%  (solar panels, electronics, EV, 5G)
  Jewelry/Silver:    26%  (physical demand, especially India)
  Investment:        20%  (ETFs, bars, coins)

When industrial demand is strong and the economy is expanding, silver outperforms gold (the ratio falls). When fear dominates and industry contracts, silver underperforms gold dramatically (the ratio rises).

The solar story: Silver is critical for photovoltaic cells. Each solar panel uses 10-20 grams of silver paste. Global solar installations are growing at 25-30% per year. By 2030, solar demand alone could consume 30-40% of annual silver mine supply. This is a secular demand story that did not exist a decade ago and is not priced into historical models.

Silver Volatility: The Leverage Effect

Silver's daily return standard deviation is roughly 1.6x gold's. This is not random — it is structural:

Thinner market: Silver's total market cap (~\(1.4 trillion) is ~7x smaller than gold (~\)14 trillion). Less capital chasing larger percentage moves.
Industrial beta: In risk-off environments, silver gets hit by both precious metals selling AND industrial demand collapse.
Speculative concentration: Managed money positions in silver futures are large relative to open interest. Positioning unwinds create outsized moves.

Implication for trading: Never size a silver position the same as a gold position. If gold risk is X, silver risk should be X / 1.6 to equalize volatility contribution.

The Gold/Silver Ratio: 50 Years of History

The GSR is one of the oldest and most-traded relative value indicators in commodity markets.

Historical Regimes

Period	Average GSR	Regime
1971-1980	38	Inflation era, silver outperformance
1980-1990	60	Post-bubble normalization
1990-2005	67	Secular bear market for metals
2006-2012	58	Financial crisis and QE bull
2013-2019	77	Gold outperformance, silver neglected
2020-2025	82	Post-COVID elevated, structurally higher?

Trading the Ratio

The GSR is mean-reverting over 2-6 month horizons, but the mean itself shifts with the macro regime. Our approach:

Estimate the regime-conditional mean using the HMM state. In risk-on, the expected GSR is ~65-70. In risk-off, ~80-85. In crisis, 90-110.
Trade deviations from the conditional mean. When the GSR is 2+ standard deviations above the conditional mean, go long silver / short gold. When 2+ below, do the opposite.
Express via ETFs: Long SLV / Short GLD (or inverse) with volatility-adjusted sizing. Silver position is smaller because of higher vol.

Hit rate: Historical walk-forward testing shows a 58-62% win rate with a 1.3-1.5 profit factor. Not spectacular for a single strategy, but robust and decorrelated from directional gold/silver bets.

Futures Curve Dynamics

Cost of Carry in Gold

Gold futures are almost always in contango because the cost of carry is positive:

F(T) = S × e^((r - l + s) × T)

Where:
  F(T)  = futures price at time T
  S     = spot price
  r     = risk-free rate (~5% as of early 2025)
  l     = gold lease rate (~0.1-0.3% usually)
  s     = storage cost (~0.1-0.2%)

Example (April 2025):
  Spot:     $2,600/oz
  6-month:  $2,600 × e^(0.05 - 0.002 + 0.001) × 0.5
          = $2,600 × 1.0247
          = $2,664/oz
  Contango: $64/oz (2.5% annualized)

When Gold Goes Backwarded

Gold backwardation is rare — it happened in late 2008 (Lehman crisis) and briefly in March 2020. It signals that market participants will pay a premium for immediate delivery because they do not trust that gold will be available later. Backwardation in gold is a screaming buy signal with an 85%+ hit rate over 3-month horizons in historical data (small sample caveat applies).

Silver's Convenience Yield

Unlike gold, silver has genuine industrial consumption. When manufacturers urgently need silver (supply disruption, demand surge), the convenience yield spikes and the curve can go into steep backwardation. These episodes are more common in silver than gold and create carry trade opportunities.

Options Markets

GVZ: Gold's Fear Gauge

The Cboe Gold Volatility Index (GVZ) is calculated from GLD options using the same methodology as VIX. Key levels:

GVZ Level	Market State	Trading Implication
< 12	Extremely calm	Premium selling attractive. Gamma cheap.
12-16	Normal	Fair value zone. No strong vol signal.
16-22	Elevated	Transitional. Vol could expand or contract.
22-30	Stressed	Premium buying potentially attractive. Gamma expensive.
> 30	Crisis	Emergency hedging zone. Do not sell naked premium.

Vol Surface and Skew Dynamics

Gold's vol surface has distinctive characteristics:

Put skew dominates. 25-delta put IV is typically 2-4 vol points above 25-delta call IV. This reflects institutional demand for downside protection (miners buying puts, portfolio managers buying tail hedges).

Skew steepens before FOMC. In the 3-5 days before a Fed decision, the put-call skew widens as hedgers bid up put protection. After the announcement, skew normalizes. This creates a tradeable pattern for skew strategies.

Term structure is usually in contango. Near-term IV is below longer-dated IV because near-term options have less uncertainty. The term structure inverts (short-dated IV > long-dated IV) during acute stress events — the same signal as VIX futures inversion for equities.

The Physical Market

Four Major Markets

Market	Location	Trading Hours (Local)	Contract	Clearing
LBMA	London	8 AM - 5 PM GMT	Loco London, unallocated	Bilateral
COMEX	New York	6 PM - 5 PM ET (nearly 24h)	100 oz futures	CME
SGE	Shanghai	9 AM - 3:30 AM CST	Spot, deferred	Central
MCX	Mumbai	9 AM - 11:30 PM IST	Futures	MCX

London and COMEX together account for roughly 80% of global gold trading volume. Shanghai is growing rapidly and now sets the marginal physical price during Asian hours.

Physical vs. Paper Gold

The ratio of paper gold (futures, ETFs, OTC derivatives) to physical gold is estimated at 50-100x. This means that in normal markets, the paper market sets the price and physical adjusts. But in extreme stress (2020, and potentially future crises), physical demand can overwhelm paper supply, causing premiums to spike and delivery squeezes.

Research Findings: Honest Assessment

What Works (Supported by Academic Research)

Time-series momentum in gold (Moskowitz, Ooi, and Pedersen, 2012): 12-month momentum in gold has delivered a positive Sharpe ratio across every decade since 1975. Robust. Well-documented. Our TSMOM strategies exploit this.
Real rate sensitivity (Barsky and Summers, 1988; multiple follow-ups): The negative relationship between gold and real rates is the most stable factor in gold pricing. Our macro strategies exploit this.
Volatility risk premium (Tully and Lucey, 2007): Gold options consistently overprice realized volatility by 2-4 vol points. Short-vol strategies have positive expected return. Our vol arb strategies exploit this.
COT positioning as a contrarian indicator (multiple studies): Extreme managed money positioning predicts mean-reversion in gold prices over 2-8 week horizons. Our microstructure strategies exploit this.
Gold/silver ratio mean reversion (Batten et al., 2013): The ratio is stationary over multi-year horizons and mean-reverts from extremes. Our stat-arb strategies exploit this.

What Does Not Work (Despite Popular Belief)

Gold as an inflation hedge. Gold is a mediocre inflation hedge over short horizons (< 2 years). The correlation with CPI is essentially zero on monthly data. Gold hedges against unexpected inflation at very long horizons (10+ years). For short-term inflation trading, TIPS are superior.
Simple moving average crossovers. Every retail trading book suggests 50/200 MA crossovers for gold. In walk-forward testing, these strategies have Sharpe ratios below 0.3 after transaction costs. The signal is too slow and the whipsaw rate is too high.
Lunar/astrological cycles. Yes, people trade these. No, they do not work. We tested every lunar phase, planetary alignment, and Mercury retrograde period against gold returns. P-values all above 0.4. Truly random.
News sentiment (simple). Basic NLP sentiment scores on gold news articles have near-zero predictive power for gold returns beyond 4 hours. The market prices news faster than sentiment models can process it. However, sentiment divergence (positive news but negative price action) has modest predictive power for mean-reversion.

What Is Probably Overfit (Proceed with Caution)

Day-of-week effects. Some studies find gold returns are higher on Mondays. The effect is statistically marginal, varies by decade, and disappears in transaction costs. We do not trade it.
Turn-of-month effects. Gold returns are slightly higher in the last 3 and first 3 days of each month. The effect exists but is small (~15 bps/month) and requires high-frequency execution to capture. Not worth the complexity.
Gold-to-miners ratio timing. The ratio of gold to GDX is sometimes cited as a leading indicator. In our testing, the signal-to-noise ratio is too low for systematic trading. Useful as a qualitative input, not as a standalone signal.

Our Edge: The Ensemble

No single gold strategy is a money machine. Every edge we described above has a Sharpe ratio between 0.3 and 0.8 in isolation. That is not enough to run a portfolio.

The edge comes from combining 29 strategies across 8 categories. Here is why:

Diversification of Alpha Sources

Category	Correlation with TSMOM	Correlation with MacroRate	Stand-alone Sharpe
TSMOM	1.00	0.35	0.55
Mean Rev	-0.15	0.10	0.42
Carry	0.20	0.30	0.38
Vol Arb	0.05	-0.10	0.65
Macro	0.35	1.00	0.50
Seasonal	0.10	0.05	0.35
Micro	0.00	0.15	0.45
Stat Arb	-0.10	0.05	0.40

Average pairwise correlation: 0.12. Low correlation is the ultimate free lunch. When these strategies are combined with inverse-vol weighting and regime-conditional optimization, the portfolio Sharpe ratio rises to 1.2-1.5 — roughly 2-3x any individual strategy.

The Math of Diversification

For N strategies with average Sharpe S and average pairwise correlation rho:

Portfolio Sharpe ≈ S × sqrt(N / (1 + (N-1) × rho))

Our case:
S = 0.46 (average individual Sharpe)
N = 29
rho = 0.12

Portfolio Sharpe ≈ 0.46 × sqrt(29 / (1 + 28 × 0.12))
                 ≈ 0.46 × sqrt(29 / 4.36)
                 ≈ 0.46 × sqrt(6.65)
                 ≈ 0.46 × 2.58
                 ≈ 1.19

This is the fundamental thesis: individually modest edges, combined with low correlation and disciplined risk management, produce institutional-grade risk-adjusted returns.

Next: Chapter 13 — PM Options Masterclass →