Metal Markets: Gold and Copper Quant Trading
Gold and copper are both liquid metals, but their return drivers are not interchangeable. Gold is primarily a monetary and reserve asset with jewelry and industrial demand at the margin; copper is an industrial input whose price reflects global growth, electrification, inventories, and supply disruptions. A metals strategy improves when it models each market in its own economic coordinates before searching for common factors.
Gold: real rates, currency, and optional demand
Gold has no contractual cash flow, so opportunity cost matters. Real yields, the U.S. dollar, central-bank purchases, risk aversion, and positioning are common explanatory variables. Their signs can change across regimes: a dollar move driven by growth is different from a dollar move driven by risk aversion.
r_gold = alpha + beta_r × Δreal_yield + beta_fx × Δdollar
+ beta_risk × risk_shock + beta_flow × ETF_flow + error
Fit this as a conditional model, allowing coefficients to vary with volatility or inflation regime. A static regression frequently produces unstable residuals because it mixes inflation hedging, safe-haven demand, and liquidity shocks. Gold options may be a cleaner instrument when the thesis is a convex tail event rather than a forecast of the next futures return. Surface construction and risk-neutral diagnostics are detailed in implied volatility surface.
Copper: inventories and the global cycle
Copper’s physical market has warehouse inventories, treatment charges, scrap supply, mine disruptions, and consumption across construction, grids, vehicles, and manufacturing. Exchange stocks are informative but incomplete: reported inventory can migrate between locations or be economically unavailable. Model inventories jointly with the curve and spot basis rather than treating a warehouse series as total supply.
| Feature | Gold interpretation | Copper interpretation |
|---|---|---|
| Real yields | Opportunity cost | Macro-growth proxy |
| Dollar | Pricing currency and safe-haven channel | Global demand and affordability |
| Inventories | Limited direct role | Scarcity and financing signal |
| Curve | Lease rate/financing | Storage, scarcity, and delivery stress |
| Supply shock | Mine/refinery effect | Often first-order effect |
A copper curve inversion can signal nearby tightness, but it can also arise from a localized delivery constraint. Test whether the signal predicts the relevant futures contract after costs, not merely a continuous spot series.
Relative-value signals
Gold-copper ratios are popular because they summarize a monetary-versus-growth narrative. Their usefulness depends on the horizon. A z-score is not a structural equilibrium: monetary policy, Chinese demand, and mine supply can permanently shift the ratio. Combine the ratio with explanatory states and use slow re-estimation.
import numpy as np
def robust_zscore(x, median, mad):
return (x - median) / max(1.4826 * mad, 1e-9)
def volatility_scaled_weight(signal, forecast_vol, target_vol=0.10):
return signal * target_vol / max(forecast_vol, 1e-6)
Relative value should be beta-aware. A long gold/short copper book can be a deliberate risk-off expression, not market neutral. Hedge broad dollar or equity exposures only after measuring how they behave in the intended stress regime.
Data construction and roll effects
Metal datasets often splice contracts, London and exchange fixings, spot indications, and warehouse statistics. Each series has timezone and settlement conventions. Align observations to tradable decision times and preserve the original publication lag. For futures, roll return can dominate a slow signal, especially when contango or backwardation persists.
The broad curve toolkit in commodity spread trading applies, but metals add delivery-location and warrant mechanics. A backtest that simply subtracts front-month prices will not capture financing, delivery optionality, or roll execution.
Risk design
Gold can gap on monetary-policy surprises; copper can gap on mine disruptions, tariffs, or Chinese demand news. Their correlations rise or fall precisely when a diversified portfolio needs them most. Build scenario tests around real-rate shocks, dollar spikes, recession, supply curtailment, and inventory draws. Do not infer capacity from average daily volume: open interest and book depth can shrink around delivery and macro events.
Margin-aware sizing is important for leveraged futures. Volatility targeting responds after a shock, while variation margin is due during it. Reserve liquidity for stressed moves as discussed in leverage and margin.
Key takeaways
- Gold and copper require distinct structural models despite sharing commodity-market infrastructure.
- Gold’s macro betas are conditional; copper signals should combine inventories, curve, and industrial demand.
- Ratios are regime indicators, not stationary laws.
- Preserve point-in-time timestamps and contract roll mechanics in all research.
- Size for policy, supply, and liquidity gap risk rather than average volatility alone.