Before You Compare Gold Returns, Audit the Contract Roll

The question comes before the chart

Suppose two screens disagree about how much gold moved. The useful first question is not which screen is wrong. It is whether both screens are measuring the same instrument over the same interval. One may show a spot quote, another a particular delivery month, and a third a continuous futures series whose underlying contract has changed.

For research, write a one-sentence measurement instruction before downloading anything: compare one named spot feed and one named futures contract between two exact UTC timestamps. Keep bid, ask, midpoint and last trade distinct. This small discipline prevents a visually compelling chart from becoming an undocumented mixture of prices.

A roll changes the contract, not just its label

CME describes a rollover as closing exposure in one contract month and establishing exposure in a later month. Futures expire; the delivery and expiry arrangements belong to the individual contract. That is a different operation from a chart vendor joining two historical series together.

My audit keeps two separate records. The instrument record lists the exchange, symbol, delivery month and relevant expiry information. The series record lists the vendor, roll trigger, adjustment method and retrieval date. A trader’s actual roll transaction belongs in a third execution record, with both legs and costs. These three records should never be silently substituted for one another.

Source notes: [1] CME Group — Understanding Futures Expiration & Contract Roll

A deliberately hypothetical discontinuity

Imagine contract A is quoted at 4,000 dollars per ounce and contract B at 4,015 at the same instant. A continuous chart switches from A to B. Its unadjusted line can jump by 15 dollars even if neither contract moved during the switch. That is a constructed example, not a historical gold quotation or an observed market return.

Now suppose B reaches 4,025 at the next observation. The within-contract move from 4,015 is 10 dollars, about 0.249%. Comparing 4,025 with the earlier A quote of 4,000 instead produces 25 dollars, or 0.625%. The latter bundles the difference between contracts with the subsequent movement. Neither arithmetic result tells you an account’s profit without specifying what was held and traded.

Choose the series for the job

For an event study, the cleanest starting point is a single contract with adequate observations throughout the event window. If an event falls near a roll, flag it and inspect both contracts. For a longer study, document a consistent roll rule before seeing which version gives the most attractive result.

Adjustment can make a continuous chart easier to analyse, but its historical levels may no longer equal contemporaneous transaction prices. Preserve raw contract files alongside any transformed series. If you later change the roll rule, create a new dataset version instead of overwriting the evidence behind the original analysis.

Make the spot comparison genuinely synchronous

CME identifies its standard GC contract as 100 troy ounces. That specification describes a contract’s size; it does not make every screen labelled gold the same product. Price units, contract exposure and feed definitions still need checking.

Compare simultaneous observations, then report the futures-minus-spot difference with both source labels. If one feed is delayed, carry that delay into the interpretation. A settlement value and a later spot snapshot are useful references, but their difference is not a clean same-time basis measurement. Missing synchronous data should produce an unavailable result rather than an interpolated certainty.

Source notes: [2] CME Group — Gold Futures

The minimum audit trail

Keep the raw file, original filename, retrieval time, timezone convention, row count and a checksum. Add a small table identifying each contract switch and the prices on both sides. Check duplicate timestamps, out-of-order records, unexplained missing sessions and abrupt changes in decimal precision.

The final research note should state the instrument, comparison window and roll treatment in ordinary language. If the conclusion changes when a reasonable alternative roll rule is used, report that sensitivity. A result that survives a transparent measurement audit is more useful than an impressive return assembled from incompatible observations.