Copper Physical-Futures Hedge Optimization & Basis Risk Model

The minimum-variance LME–COMEX hedge removes 33.1% to 33.8% of daily variance out of sample, while a naive one-for-one hedge removes only 0.6% on the same days. The paired in-sample result is 34.1%, so the performance survives an honest forward test. Measured Friday to Friday, effectiveness rises to 70.0% as the London–New York closing-time mismatch washes out.

The remaining risk is economically meaningful: optimal daily hedging still leaves residual volatility of $0.0433/lb, while the daily change in the unhedged LME–COMEX basis has a standard deviation of $0.0531/lb.

A company that owns physical copper loses money when the copper price falls. To protect itself it can sell copper futures: if the price drops, the loss on the metal it holds is offset by a gain on the futures it sold short. The catch is that the metal it holds (LME cash copper in London) and the contract it sells (COMEX futures in New York) are not the same thing, and their prices do not move in perfect lockstep. The hedge cancels most of the price risk but never all of it, and that leftover is what this project measures.

Three quantities do the work.

Three daily series, all reduced to dollars per pound and aligned on shared trading days: 1,875 days from 2 January 2019 to 12 August 2026. LME is quoted in dollars per tonne and is divided by 2,204.62. The calendar mismatch drops 49 LME rows, 42 COMEX rows and 36 CPER rows, and no gaps are forward-filled. Outliers are kept: the 2024 squeeze and the 2025 tariff dislocation are signal, not noise to be cleaned away.

The primary LME–COMEX hedge removed 34.1% of daily variance in-sample, and the honest one-step-ahead results stayed close at 33.1% to 33.8% out of sample. Shorting one pound of COMEX copper for each pound of LME exposure removed only 0.6% on the same days. The small deterioration is consistent with a relationship that is stable rather than overfit: the independently estimated hedge ratio stays between 0.45 and 0.56 in all five market regimes below.

On 375 Friday-to-Friday weeks, the in-sample hedge ratio rises to 0.7276 and effectiveness doubles to +70.0% (R² 0.6996); even the naive 1:1 hedge removes +60.2%. The reason is asynchronous closes. LME cash fixes around 1 p.m. London time, COMEX around 1 p.m. New York time, and CPER at 4 p.m. New York. News arriving between those closes can land in one leg’s “day” before the other’s, weakening same-day covariance. Across a week that timing mismatch mostly washes out, so the measured relationship strengthens and h* moves closer to the one-for-one exposure of two copper prices.

Counterintuitively, the violent regimes hedge better at the daily frequency. The closing-time mismatch is roughly constant in size, so it can dominate small moves in calm markets but is swamped when both copper markets make large common moves. COVID 2020 is the clearest warning against reading its +23.1% daily result as a broken hedge: its weekly correlation is approximately 0.92, the highest of the five regimes.

In mid-May 2024 a short squeeze hit COMEX copper. The setup was a delivery trap. High financing costs had drained US exchange inventories, leaving the market with little physical cushion, while investment funds bought copper futures aggressively on expected demand from AI data centers, power grids and electric vehicles. Physical desks stood on the other side of that flow: short COMEX futures against cheaper overseas metal, planning to ship it to the US and deliver it against their shorts.

When the market moved, the escape routes closed. Much of the world’s spare metal was of brands not registered for COMEX delivery, fresh sanctions on Russian metal had narrowed what the LME system could supply, and ships from South America take weeks while margin calls arrive daily. Unable to deliver, shorts had to buy their positions back, and the US futures price tore away from the rest of the world’s copper. In the ten trading days from 30 April to 14 May, the COMEX front month rose $0.3890/lb while LME cash rose only $0.0469/lb. For a desk long 1,000 tonnes of physical copper and short COMEX futures dollar for dollar, the position built to be safe became the problem: a mark-to-market loss of −$0.3421/lb, about −$754,097 in two weeks.

Where May 2024 was a ten-day accident, 2025 was a seven-month policy trade. Early in the year, traders came to expect that a US Section 232 investigation would end in a 50% tariff on imported refined copper. Anyone who could land cathode in a US warehouse before the tariff hit would own tariff-free metal in a tariffed market, so desks raced to ship copper in, and that buying pushed COMEX to a premium over LME with no precedent in the sample. The basis ground steadily lower until it reached −$1.3237/lb on 25 July 2025, the widest gap in seven and a half years of data. A desk hedged dollar for dollar was dragged along the whole way: from 2 January to 25 July the position bled −$1.2749/lb, or −$2,810,701 on 1,000 tonnes, at the worst mark. Nothing about its copper had changed; it was simply short the expensive leg of a spread that policy expectations kept stretching, posting margin against a loss that grew for seven months.