What the gold positioning report measures, and what it misses

What the report actually counts

The weekly positioning report starts from a number the exchange already publishes: open interest, the count of futures contracts still open. The report takes that total and splits it by who is holding it. Every open contract has a long side and a short side, so the categories must balance exactly against each other. There is no aggregate net position for the market as a whole, only a distribution of the same contracts across different kinds of holder.

Each group is reported three ways: gross long, gross short, and spreading, which covers positions that are long one delivery month and short another. Holders above a size threshold set by the regulator have to report; everything below that line is grouped together as non-reportable. Gold is published in a futures-only version and a combined version that folds in options on a delta-adjusted basis. The same week can read differently depending on which table you opened, which is why mixing the two mid-analysis is a quiet way to fool yourself.

Who sits in each reporting category

Four reporting groups matter. Producer, merchant, processor and user covers firms whose main business is handling physical metal: miners, refiners, fabricators, jewellery groups. They use futures to fix a price for metal they already own or expect to own. Swap dealers are intermediaries. They run a book of over-the-counter trades with clients and lay the resulting exposure off on the exchange, so their futures position is usually the shadow of somebody else's trade.

Managed money is the discretionary and systematic fund community: trading advisers, hedge funds, trend followers. Other reportables is a catch-all for large traders who do not fit the first three, including family offices and corporate treasuries.

The classification follows how a firm is registered, not how it is behaving this week. A refiner can take a directional view. A fund can run a genuine hedge. A swap dealer can be flat in risk terms while showing a huge gross position. Treating one bucket as informed and another as naive is a story laid over the data, not something the data states.

Why the two headline lines are mirror images

Plot managed money net position against commercial net position and it looks like a standoff: funds pile long while hedgers pile short, and the two curves move like reflections in water. They are reflections. Because longs and shorts sum to zero across all categories, a large fund net long must be matched somewhere, and firms with metal to sell forward are the natural counterparty. The mirror is arithmetic before it is anything else.

This matters because the most common headline built on this data, a record commercial short, is usually just a restatement of record fund buying. It is not a declaration that producers expect a fall. A miner selling forward is fixing a price for ounces still coming out of the ground, which is a decision about certainty rather than a forecast. Reading the hedger side as a bearish opinion is where positioning analysis most often goes wrong. The honest version is narrower: the market has become more one-sided than usual, and one group is carrying the other side of it.

The reporting lag you cannot trade around

Positions are snapshotted at the close on Tuesday. The report is released on the Friday afternoon of the same week. By the time you read it, three sessions have passed, and in many weeks one of them carried the month's main data release. If gold has moved hard on the Wednesday and Thursday, the positioning in front of you belongs to a market that no longer exists.

That delay removes a whole class of use. You cannot time an entry with it, and you cannot use it to explain a move that happened after the snapshot, which is the single most common abuse of the series. What survives the lag is slower: the direction of travel across several weeks, and whether the market is sitting near the stretched end of its own recent range of positioning. Those are weekly review questions, not Monday morning triggers. Anything faster than that is you reading tea leaves with a three day head start given to everybody else.

Crowding is a condition, not a trigger

A stretched position tells you about fuel, not about ignition. When a large share of open interest sits on one side, the holders on that side share a risk and will reach for the exit through the same door. That makes sharp, fast unwinds more plausible than they are in a balanced book. It says nothing about when, and nothing about which way the first push comes.

Extremes also persist. Positioning can sit at the high end of its own range for weeks while price grinds higher, because a trend is exactly the environment that attracts trend followers. Selling because the funds look crowded is a bet against momentum with no timing component attached, which is a slow and expensive way to be eventually right. The usable shape is conditional. If positioning is stretched and price then breaks structure against the crowd, the follow through tends to travel faster than the same break in a balanced market. The break is the trigger. Positioning only changes what you expect once the break has happened. Keeping participation separate from direction is the habit that protects you here.

Normalising the series before you compare

Raw contract counts are not comparable across long stretches of time. Open interest grows as the market grows, participation changes, and a figure that looked extreme years ago can be unremarkable now. Two simple adjustments help.

• Express the net position as a share of total open interest, so the series is scaled to the size of the market rather than to its history.
• Compare against a rolling window of the last year or two instead of against an all-time record that was set in a different market.

Both are improvements. Neither creates a threshold. There is no level at which positioning becomes a sell, and any fixed line you draw will be fitted to the sample you drew it on, which is the same trap as any other over-optimised parameter. Treat the normalised series as a description of how unusual the current distribution is, and stop there. The reasoning in why a score is not a probability applies to positioning readings with full force.

What the report cannot see

The report covers exchange-traded futures and options. It does not cover the much larger over-the-counter market, where a great deal of gold risk actually sits. Forwards, unallocated account balances and bespoke bank trades never appear. A swap dealer short can be the hedge against an over-the-counter long sold to a pension fund, and the client side of that trade is simply not in the file.

It also tells you nothing about entry prices, stop levels or holding intent. Two funds with identical net longs can be a month into a position and a day into a position, with completely different pain thresholds and completely different behaviour when price turns. And because exchange futures are one venue for the same metal rather than the whole market, the link between that book and the price on your screen needs care. If that distinction is new, spot versus futures in gold is the groundwork.

A sane place for it in a weekly routine

Used honestly, this is a context note you write once a week and then mostly leave alone. A workable routine is short. Open the futures-only table. Note whether managed money net length is rising, falling or flat. Note whether it sits in the middle or at the edge of its recent range. Write one line in the journal. Then go back to the chart and let structure decide what you actually do.

The report never produces an entry. What it can change is how much respect you give a level and how quickly you expect a failure to travel once it starts. When positioning is balanced, expect ordinary follow through. When it is stretched and price breaks the wrong way for the crowd, expect speed and worse fills. Beyond that the file is history. Watch the market you are actually in on the live chart, and keep the positioning note where it belongs, in the weekly review rather than the execution window.