10.12 Positioning: the CFTC Commitments of Traders report and MCX open interest

Motivation

Commodity commentary is full of sentences like "funds are record long gold" or "speculators have capitulated in crude". These come from one public dataset, the US CFTC's Commitments of Traders (COT) report. This section says what the report measures, how to turn it into a number you can compare across years, what the published evidence says about its predictive value, and what the Indian equivalent on MCX does and does not tell you. The short version: positioning is a good description of who holds risk, a weak predictor of returns on average, and most useful as a warning about crowding at extremes.

10.12.1What the report is#

Every futures position is a contract between a long and a short. Clearing members and brokers report to the CFTC the positions of each trader whose holding in any one contract month reaches a reporting level set by CFTC regulation; once a trader crosses that level in any month, the firm reports the trader's whole position in that commodity (CFTC, Explanatory Notes, read 6 Nov 2026). The CFTC aggregates these into categories and publishes the totals.

Timing The report gives positions as of the close on Tuesday and is released on Friday at 3:30 p.m. US Eastern time (CFTC, About the COT Reports, read 6 Nov 2026). In India that is Friday night or early Saturday. By the time you read it the data are three days old, and anything that happened on Wednesday to Friday is not in it.

Futures-only and combined Each report comes in a futures-only version and a futures-and-options-combined version. In the combined version option positions are converted to futures equivalents with delta factors supplied by the exchanges, so a long call counts as a fraction of a long future. Futures-only history starts in 1986 and combined history in 1995 (CFTC).

Report families

Report	Categories	Coverage	Since
Legacy	Commercial; non-commercial (long, short, spreading); nonreportable	all markets	futures-only data from 1986
Supplemental	adds Commodity Index Traders	selected agricultural markets, combined only	2007 (history from 2006)
Disaggregated	Producer/Merchant/Processor/User; Swap Dealers; Managed Money; Other Reportables; nonreportable	physical commodities with at least 20 traders above the reporting level	published from 4 Sep 2009; history back to 13 Jun 2006
Traders in Financial Futures (TFF)	Dealer/Intermediary; Asset Manager/Institutional; Leveraged Funds; Other Reportables; nonreportable	financial futures (currencies, rates, equity indices)	2009

Sources: CFTC, About the COT Reports and Disaggregated Explanatory Notes, read 6 Nov 2026. The TFF start year is from general knowledge of the CFTC programme and was not re-read this session.

The disaggregated categories (CFTC wording, paraphrased only where noted):

Producer/Merchant/Processor/User: an entity that "predominantly engages in the production, processing, packing or handling of a physical commodity" and uses futures to manage the associated risk. Gold miners, refiners, jewellers and bullion banks' physical arms; oil producers and refiners.
Swap Dealer: an entity that "deals primarily in swaps for a commodity and uses the futures markets to manage or hedge the risk" of those swaps. Its counterparties include index investors and commercial hedgers who trade over the counter, so swap-dealer positions are partly a mirror of OTC client flow.
Managed Money: a registered commodity trading advisor (CTA), a registered commodity pool operator (CPO), or an unregistered fund identified by the CFTC, "engaged in managing and conducting organized futures trading on behalf of clients". This is the "speculator" or "fund" number in the press.
Other Reportables: every reportable trader not in the first three categories.
Nonreportable: open interest not held by reportable traders, computed as total open interest minus reportable positions.

The CFTC classifies a trader by its predominant business, not by the purpose of each position. A producer can speculate and a fund can hedge. Classifications are not kept historically, so the back-filled 2006–2009 data use later classifications (CFTC).

10.12.2The accounting identity: someone is always on the other side#

Proposition 10.12aggregate net position is zero

Let categories $c = 1, \dots, C$ (including nonreportables) hold long $L_c$ , short $S_c$ and spread $P_c$ contracts in a market with open interest $\mathrm{OI}$ . With spreading counted once on each side,

\mathrm{OI} = \sum_c (L_c + P_c) = \sum_c (S_c + P_c), \qquad\text{hence}\qquad \sum_c N_c = 0,\quad N_c := L_c - S_c. \tag{10.22}

Proof

Open interest is the number of contracts entered into and not offset. Each such contract has exactly one long holder and one short holder, so the total long position across all holders equals the total short position equals $\mathrm{OI}$ . A spread position is an equal long and short held by one trader, so it appears once in the long total and once in the short total. Partition holders into the categories and sum: the first two equalities follow. Subtracting them gives $\sum_c (L_c - S_c) = 0$ .∎

The identity says that "funds are record long" always means "the other categories together are record short". In gold and crude the usual counterparty of managed money is the producer/merchant category (hedging future output) and the swap dealers. Problem 10.20 works through the arithmetic. It also means positioning by itself cannot say which side is "right": a record managed-money long is a record hedger short, and hedgers are not trying to predict prices.

10.12.3Normalising positioning#

Raw net contracts are not comparable over time, because open interest itself trends (it roughly follows the growth of the market and of index investing). Three normalisations are standard.

Definitionpositioning measures

For a category with net position $N_t$ in week $t$ :

Share of open interest: $x_t = N_t/\mathrm{OI}_t \in [-1, 1]$ .
Rolling $z$ -score: with a trailing window of $W$ weeks (156 weeks, three years, is common), $z_t = \frac{x_t - \bar x_{t,W}}{s_{t,W}}, \qquad \bar x_{t,W} = \frac1W\sum_{j=1}^{W} x_{t-j},\quad s_{t,W}^2 = \frac1{W-1}\sum_{j=1}^{W}(x_{t-j} - \bar x_{t,W})^2. \tag{10.23}$
Rolling percentile: $\pi_t = \frac1W\#\{j \in \{1,\dots,W\} : x_{t-j} < x_t\}$ . A min–max version, $(x_t - \min)/(\max - \min)$ over the window, is popular with traders as the "COT index".

The window excludes the current week, so $z_t$ is computable on the release date without look-ahead.

Why use percentiles The $z$ -score assumes the window's distribution is roughly symmetric with a stable spread. Positioning series are persistent and regime-dependent: managed-money gold length was structurally higher in some multi-year bull phases than in others. A percentile is invariant to any increasing transformation of $x$ , so it does not care whether you use contracts, share of OI or tonnes, and one extreme week cannot inflate the denominator as it does for $z$ . Its drawback is that it saturates at 0 and 1: "100th percentile" says nothing about how far outside the window the reading is.

Converting to physical units One COMEX gold contract is 100 troy ounces (CME contract specification; not re-read this session). One tonne is $10^6/31.1034768 = 32{,}150.75$ troy ounces. A net long of 135,000 contracts is $135{,}000 \times 100/32{,}150.75 \approx 420$ tonnes, which you can set against the 4,256 t held by gold ETFs at the end of Q3 2026 (§Facts) or against a year of Indian jewellery demand. Physical units turn "record long" into a quantity whose liquidation the market would have to absorb.

Worked exampleillustrative numbers, not CFTC data

A gold report shows managed money long 180,000, short 45,000, open interest 520,000. Then $N = 135{,}000$ and $x = 0.2596$ . If the trailing 156-week mean of $x$ is 0.12 with standard deviation 0.07, then $z = (0.2596 - 0.12)/0.07 = 2.00$ ; if 149 of the 156 prior readings were lower, $\pi = 0.955$ . Problem 10.19 asks you to finish the example.

10.12.4What the evidence says#

The studies below use US futures and the CFTC categories. The results are stated as narrowly as their abstracts allow.

Positions follow prices. Sanders, Boris and Manfredo (2004), "Hedgers, funds, and small speculators in the energy futures markets: an analysis of the CFTC's Commitments of Traders reports", Energy Economics 26(3): 425–445, run Granger-causality tests in crude oil, unleaded gasoline, heating oil and natural gas. They find positive correlation between returns and non-commercial positions and negative correlation for commercials. Non-commercial net positions rise in the week after price increases, and "traders' net positions do not lead market returns in general". Büyükşahin and Harris (2011), "Do speculators drive crude oil futures prices?", The Energy Journal 32(2): 167–202, use the CFTC's non-public trader-level data and find "little evidence" that hedge fund and other non-commercial position changes Granger-cause price changes; the results "suggest that price changes precede their position changes".
Sentiment indices: continuation for large speculators, reversal for hedgers. Wang (2001), Journal of Futures Markets 21(10): 929–952, studies six agricultural futures markets: large-speculator sentiment forecasts price continuation, large-hedger sentiment forecasts reversals, and small-trader sentiment has little forecasting power. Wang (2003), Applied Financial Economics 13(12): 891–898, finds the same pattern in S&P 500 index futures and adds that "extreme large trader sentiments ... tend to provide more reliable forecasts". Note the sign: in these studies an extreme speculator long is bullish, not the contrarian signal the folk rule assumes.
Two premiums with opposite signs. Kang, Rouwenhorst and Tang (2020), "A tale of two premiums: the role of hedgers and speculators in commodity futures markets", Journal of Finance 75(1): 377–417, show that "short-term position changes are driven mainly by the liquidity demands of noncommercial traders, while long-term variation is driven primarily by the hedging demands of commercial traders", and that the two components "influence expected futures returns with opposite signs". When speculators rush in over a week or two they pay for liquidity, which predicts weaker subsequent returns. When hedgers are persistently short they pay speculators an insurance premium, which predicts stronger returns for the long side.

How to read this as a trader Three conclusions survive.

On average, positioning is a lagged momentum indicator. Managed money is dominated by trend-followers (M04 · 4.3), so its net position is close to a smoothed function of past returns. A regression of future returns on positioning partly re-discovers time-series momentum, and you should control for past returns before crediting positioning with anything.
Fast changes are costly; levels are not a timing signal. The Kang–Rouwenhorst–Tang result says the information is in the speed of position changes, not in the level, and it points against chasing a sudden speculative inflow.
Extremes measure crowding, not direction. A position at the 98th percentile is not a forecast that prices will fall. It is a statement about how much forced selling there could be if the trend breaks or margins rise (10.10 on margins, M03 on ruin). Use it to size and to set stop discipline, not to call tops. This is a risk-management use, and it is consistent with the evidence: the studies above find weak average predictability, but none of them says a crowded position unwinds gently.

10.12.5Testing a positioning signal properly#

Suppose you want to test whether $z_t$ predicts the next $h$ -week return $R_{t+1,t+h}$ . Four traps:

Release timing. Positions are as of Tuesday, released Friday. Measure returns from the Friday close (or Monday open for MCX traders) after the release, never from Tuesday.
Overlap. With $h > 1$ weekly observations overlap; Proposition 4.4 (M04 · 4.1) shows the naive standard error is too small by a factor of about $\sqrt h$ . Use non-overlapping observations or Newey–West errors with at least $h - 1$ lags.
Persistent regressor. $z_t$ is highly autocorrelated and its innovations are correlated with returns (positions respond to prices). In that setting the OLS slope in a predictive regression is biased in small samples (Stambaugh 1999, Journal of Financial Economics), the same issue that afflicts dividend-yield regressions. Bootstrap the slope under the null with the regressor's own AR(1) dynamics.
Choice of parameters. Window $W$ , threshold (90th or 95th percentile), category (managed money or non-commercial), futures-only or combined, horizon $h$ : each choice is another test. Count them and apply M04 · 4.6.

Then hold out the last third of the sample, untouched, and run the frozen rule once.

10.12.6MCX open interest and category-wise disclosure#

MCX publishes open interest for every contract daily, but open interest alone has no category split, and every change in OI has a buyer and a seller. The popular four-quadrant reading ("price up and OI up is long build-up, price down and OI up is short build-up, ...") is a convention, not an inference: when OI rises, new longs and new shorts were opened in equal number, and the price change says only which side was more eager.

Category-wise data SEBI circular SEBI/HO/CDMRD/DNPMP/CIR/P/2019/08 of 4 January 2019 requires commodity exchanges to publish open interest and turnover by participant category, commodity by commodity: Farmers/FPOs, Value Chain Participants (VCPs), Proprietary traders, Domestic Financial Institutional Investors, Foreign participants, and Others. Disclosure was weekly from 1 Oct 2019 and has been daily by 6:00 p.m. on T+1 from 1 April 2020; a category with fewer than 10 participants may be shown as "less than 10". A second annexure requires daily disclosure of the share of open interest held by the top 3, 5 and 10 participants, long and short, for futures and options (taxguru reproduction of the circular, read 6 Nov 2026; the provisions are now in SEBI's master circular for commodity derivatives of 4 Aug 2023). MCX's own pages blocked automated fetching (10.10), so check the current report format on the MCX website.

How it differs from the COT report

Classification by client type, not business purpose. "VCP" is closest to the CFTC's producer/merchant category; "Proprietary traders" are broker-dealers' own books; there is no managed-money category, because Indian commodity funds and AIFs are small, and most retail speculation sits in "Others".
Price discovery is offshore. MCX gold and crude are priced off COMEX and NYMEX through import parity and the RBI/FBIL reference rate (10.5, 10.10). Indian positioning therefore tells you about Indian hedging and arbitrage demand, not about the global marginal buyer. The CFTC report for COMEX gold and NYMEX WTI remains the relevant positioning dataset for the dollar price.
Concentration data are more useful than in the US. The top-3/5/10 shares are a direct crowding measure: if the top 5 hold 40% of the short side in a thin contract, a squeeze (10.11.3) is a real possibility.
Daily, T+1. Indian data are more timely than the weekly, three-day-lagged COT.