Chapter 11: Gold/Silver Trading Glossary
Every term that matters for trading precious metals systematically. Organized by domain, not alphabetically. Use Ctrl+F to find anything.
Reading time: 15 min (reference) | Difficulty: All levels
Physical Markets & Market Structure
LBMA Fix (London Bullion Market Association Fix) The benchmark gold price set twice daily (10:30 AM and 3:00 PM London time) by ICE Benchmark Administration. Used as the reference price for physical gold contracts, ETF NAV calculations, and central bank reserves valuation. The AM fix is more liquid than the PM fix. Silver has one daily fix at 12:00 noon London.
COMEX (Commodity Exchange) The primary futures exchange for gold and silver, a division of CME Group. Gold futures (GC) trade in 100 troy ounce contracts; silver futures (SI) in 5,000 troy ounce contracts. COMEX warehouse stocks are the most-watched physical supply indicator in North America.
Registered vs. Eligible COMEX warehouse inventory has two categories. Registered metal has warrants attached — it is available for delivery against futures contracts. Eligible metal meets purity/weight standards but has no warrant — the owner has not made it available for delivery. When registered stocks drop sharply, it signals potential physical tightness and can drive backwardation in the futures curve.
Shanghai Gold Exchange (SGE) China's primary physical gold market. Trades in RMB-denominated contracts. The Shanghai Premium (SGE price minus LBMA price, converted) indicates physical demand intensity in China. Premiums above $20/oz suggest strong physical demand; discounts suggest weak demand or capital outflow restrictions.
Indian Premium The premium or discount of gold in India versus international prices. India is the world's second-largest gold consumer. Premiums spike during Diwali (October-November) and the wedding season (November-February). A $5-15/oz premium is normal; $30+ indicates exceptional demand.
Mint Sales Monthly coin and bar sales from sovereign mints (US Mint, Royal Canadian Mint, Perth Mint). Retail demand indicator. Eagle and Maple Leaf sales above 200K oz/month signal elevated safe-haven demand. Useful as a contrarian indicator at extremes — when everyone is buying coins, the trade may be crowded.
Gold/Silver Ratio (GSR) The price of gold divided by the price of silver. Historically averages 55-65 over long periods. Has ranged from 15 (1980 Hunt Brothers peak) to 125 (March 2020 COVID crash). Above 80 is historically extreme and mean-reverts — a signal to go long silver vs. short gold. Below 50 signals the reverse. The ratio is regime-dependent: it rises in risk-off environments (gold outperforms silver) and falls in reflationary environments (silver outperforms).
Futures & Derivatives
Backwardation When near-month futures trade at a premium to deferred contracts (front > back). In gold, backwardation is rare and significant — it indicates physical supply stress, where holders will not lend gold at any lease rate. Backwardation episodes (e.g., late 2008, early 2020) are historically associated with aggressive price rallies.
Contango The normal state for gold futures: deferred contracts trade at a premium to near-month. The spread approximately equals the cost of carry (risk-free rate minus lease rate). Gold is almost always in contango because storage costs are low and there is no consumption demand to create convenience yield.
Roll Yield The return earned (or lost) when rolling a futures position from an expiring contract to the next. In contango, long positions suffer negative roll yield (buying the more expensive deferred contract). In backwardation, long positions earn positive roll yield. For gold ETFs like GLD (which hold physical), roll yield is zero — they do not hold futures.
Basis The difference between the futures price and the spot price: Basis = Futures - Spot. In gold, the basis reflects cost of carry. A widening basis relative to the risk-free rate can indicate increased demand for futures (speculators) or decreased supply (physical tightness).
Cost of Carry For gold: approximately Risk-Free Rate - Gold Lease Rate + Storage Cost. Since storage is minimal and lease rates are usually small, gold futures are priced close to spot × (1 + risk-free rate × time to expiry).
Convenience Yield The non-monetary benefit of holding physical commodity rather than a futures contract. For gold, convenience yield is usually near zero (gold is not consumed). For silver, convenience yield can spike when industrial users urgently need metal, creating sharper backwardation episodes.
Lease Rate The interest rate charged to borrow physical gold. Central banks and bullion banks lend gold. GOFO (Gold Forward Offered Rate) was historically the benchmark; after its discontinuation in 2015, lease rates are estimated from futures-spot relationships. Rising lease rates indicate physical tightness.
Positioning & Sentiment Data
COT Report (Commitments of Traders) Published weekly by the CFTC every Friday at 3:30 PM ET (data as of Tuesday close). Shows futures and options positions for three categories.
Managed Money (COT) Hedge funds, CTAs, and systematic traders. The "speculative" category. When managed money net long positioning reaches extreme levels (> 90th percentile of historical range), the trade is crowded and vulnerable to a positioning squeeze.
Commercials (COT) Producers (miners) and consumers (jewelers, industrials). They hedge against price moves. Commercials are structurally short gold (miners hedging production). Extreme commercial shorting relative to history is a bearish signal — smart hedgers are locking in high prices.
Hedging Pressure The asymmetry between commercial hedging and speculative positioning. When miners aggressively hedge forward (increase shorts), it provides selling pressure in the futures market. The Hedging Pressure Premium is the return earned by taking the other side of hedgers' positions — the theoretical basis for carry strategies.
Non-Reportable (COT) Small traders below the CFTC reporting threshold. Historically the worst forecasters. Extreme non-reportable net positioning is a reliable contrarian indicator.
Volatility & Options
IV (Implied Volatility) The market's forecast of future realized volatility, extracted from option prices via Black-Scholes or similar models. For gold, IV is quoted via the GVZ index. Average gold IV: 14-18% in calm markets, 25-35% during stress. Silver IV runs 1.5-2x gold IV due to higher beta.
RV (Realized Volatility) Actual historical volatility calculated from returns. Compare against IV to determine the volatility risk premium. Typically calculated as the annualized standard deviation of daily log returns over a trailing window (21-day or 63-day).
Volatility Risk Premium (VRP) The persistent gap where IV > RV. For gold, IV overprices realized vol by 2-4 percentage points on average. This is the foundation of every short-vol strategy. The VRP exists because option buyers pay for insurance (crash protection) and hedgers (miners) are natural buyers of puts.
GVZ (Cboe Gold Volatility Index) The VIX equivalent for gold. Calculated from GLD option prices using the same VIX methodology. GVZ below 14 = low vol, good environment for selling premium. GVZ above 25 = elevated, potentially good for buying protection or vol mean-reversion trades.
Skew The difference in IV between out-of-the-money puts and calls at the same delta. Gold typically has negative skew (puts are more expensive than calls) because institutional demand for downside protection exceeds speculative demand for upside calls. The skew steepens sharply before FOMC meetings.
Gamma Scalping A delta-neutral strategy: buy options (long gamma) and dynamically hedge the delta by trading the underlying. Profitable when realized vol exceeds implied vol. In gold, gamma scalping around FOMC/CPI announcements captures the realized move that often exceeds the implied move priced by the market.
Backtesting & Model Validation
PBO (Probability of Backtest Overfitting) A statistical framework (Bailey et al., 2014) that estimates the probability that a strategy's in-sample performance will not persist out-of-sample. Calculated by evaluating all possible train/test splits combinatorially. PBO > 0.5 means the strategy is more likely overfit than not. Our threshold: PBO < 0.35.
DSR (Deflated Sharpe Ratio) Adjusts the Sharpe ratio for the number of strategies tested (multiple testing bias), non-normality of returns, and the length of the track record. A Sharpe of 1.5 tested once is significant; a Sharpe of 1.5 found after testing 500 variations is likely noise. Our threshold: DSR p-value < 0.05.
CPCV (Combinatorial Purged Cross-Validation) An advanced cross-validation method that tests all possible train/test combinations while purging overlapping data to prevent leakage. Standard k-fold CV fails for financial time series because of autocorrelation and look-ahead bias. CPCV with an embargo period (gap between train and test) is the gold standard.
Walk-Forward Validation Train on period 1, test on period 2. Then train on periods 1+2, test on period 3. And so on. Mimics live trading where you always train on past data and trade on unseen future data. We use anchored walk-forward (always include all historical data in training) with 63-day test windows.
Regime-Stratified Sharpe The Sharpe ratio calculated separately for each market regime (risk-on, risk-off, crisis, transition). A strategy with an overall Sharpe of 1.0 might have 1.5 in risk-off and 0.2 in risk-on — useful to know before relying on it. We require positive regime-stratified Sharpe in at least 3 of 4 regimes.
Meta-Labelling A technique from Marcos Lopez de Prado: a primary model generates directional signals (+1/-1), and a secondary model predicts the probability that the primary signal is correct. The secondary model can use features the primary model does not: regime state, strategy agreement, positioning, calendar effects. This converts binary signals into sized, confidence-weighted bets.
Triple-Barrier Method A labelling scheme where each trade has three exit conditions: take-profit barrier (above), stop-loss barrier (below), and time barrier (maximum holding period). The label is determined by which barrier is hit first. This creates more realistic training labels than simple forward returns.
Regime Classification
HMM (Hidden Markov Model) A probabilistic model that assumes the market moves through unobservable (hidden) states, and that observable returns are generated by different distributions in each state. We use a 4-state HMM: risk-on (low vol, positive drift), risk-off (elevated vol, negative drift), crisis (extreme vol, negative drift, correlation spike), and transition (mixed signals).
BOCPD (Bayesian Online Change-Point Detection) A real-time algorithm that detects abrupt changes in the statistical properties of a time series. Unlike the HMM (which classifies the current state), BOCPD answers a different question: "has something fundamentally changed right now?" When BOCPD fires with high probability, the system increases uncertainty estimates and widens risk bands.
Regime Classifier The combined output of HMM and BOCPD. The HMM provides state probabilities (smooth, backward-looking). BOCPD provides change-point probabilities (sharp, forward-looking). Together, they give the system both situational awareness and early warning.
Execution & Transaction Cost Analysis
Almgren-Chriss Model The foundational optimal execution framework. Balances two opposing forces: executing quickly (reduces timing risk but increases market impact) versus executing slowly (reduces impact but increases risk of price moving away). The optimal trajectory depends on urgency, volatility, and market depth.
TCA (Transaction Cost Analysis) Post-trade analysis measuring total execution cost: spread cost + market impact + timing cost + opportunity cost. Our TCA target: total cost < 15 bps for GLD, < 25 bps for silver ETFs, < 5 bps for gold futures.
Implementation Shortfall (IS) The difference between the paper return (signal price) and the actual return (fill price). IS = (Fill Price - Signal Price) / Signal Price. Positive IS means you paid more than the signal price. Target: IS < 5 bps average.
Arrival Price The mid-price at the moment the order reaches the exchange. Used as the benchmark for measuring execution quality. All our slippage calculations are measured against arrival price.
Risk Management
CVaR (Conditional Value-at-Risk) Also called Expected Shortfall. While VaR says "the most you will lose at the 95th percentile is X," CVaR says "if you are in that worst 5%, the expected loss is Y." CVaR is always larger than VaR and better captures tail risk. We use 95% CVaR as our primary risk measure.
EVT (Extreme Value Theory) Statistical framework for modeling the tails of return distributions. Uses the Generalized Pareto Distribution (GPD) fitted to returns beyond a high threshold. EVT-based VaR and CVaR better capture the fat-tailed nature of gold returns than Gaussian assumptions.
Filtered Historical Simulation (FHS) A VaR estimation method that combines GARCH-filtered returns with historical simulation. First, fit a GARCH model to capture current volatility clustering. Then, standardize historical returns by their GARCH-estimated volatility. Finally, simulate forward using the current volatility level. Produces more accurate tail estimates than pure historical simulation or pure parametric methods.
Factor Model Decomposes portfolio returns into systematic factors (gold beta, momentum, carry, vol) and idiosyncratic residual. Used for risk decomposition — understanding whether portfolio risk comes from intended factor exposures or unintended concentrations.
Operational & Safety
Kill-Switch Tiers Three levels of automated risk intervention. Tier 1 (soft cap): position reduction at -8% drawdown. Tier 2 (hard cap): full flatten at -12% drawdown. Tier 3 (dead-man's switch): flatten on loss of heartbeat > 15 minutes. See Chapter 9 for full details.
Reconciliation The process of comparing the system's internal position state with the broker's records. Performed at every daemon startup and continuously during market hours. Mismatches are logged and flagged — never auto-corrected — because the broker is the source of truth for actual positions.
Dead-Man's Switch A safety mechanism that triggers protective action when the system fails to check in. If no heartbeat is received for 15 minutes, the system assumes a crash and flattens all positions via the broker's native close-all endpoint. Named after the train operator safety device.
Two-Person Rule Critical actions (kill-switch reset, risk parameter changes, strategy addition/removal) require confirmation from both OpA and OpB. Prevents emotional or impulsive decisions, especially after drawdowns. Borrowed from nuclear command and institutional fund governance.