Futures & Market Research Glossary

Open interest is the number of futures contracts that remain outstanding in a delivery month or market because they have not yet been offset or extinguished through final settlement or delivery, as applicable. Every futures contract has a long side and a short side, but that long-short pair counts as one unit of market-wide open interest rather than two. Open interest is therefore a stock of outstanding contracts at a point in time, unlike trading volume, which is a flow of contracts transacted during a period.

Why heavy trading can leave open interest unchanged

Open interest changes according to whether the two sides of a transaction are opening or closing positions. If a new long is matched with a new short, one new outstanding contract is created and open interest rises by one. If an existing long and an existing short both close, one outstanding contract is extinguished and open interest falls by one. If one side opens while the other closes, the outstanding contract count remains unchanged: exposure has effectively passed from one participant to another rather than creating or eliminating a contract. The transaction still contributes to trading volume.

Those mechanics also explain the main interpretation limit. Rising open interest says that more contracts are outstanding; it does not, by itself, say whether the market is becoming bullish or bearish, because the aggregate long and short sides belong to the same stock of contracts. Open interest becomes more informative when read alongside price, spreads, liquidity and position data rather than converted into a directional signal on its own.

Published open interest may lag the session

The date attached to an open-interest figure therefore matters as much as the number itself. CME Group, for example, currently releases a preliminary Daily Volume and Open Interest Report at the end of each trading day and official data the following morning. Its futures settlement page always displays the previous trading day's open interest even though settlement and volume information relate to the stated trade date.

A researcher who aligns today's price move with an open-interest figure without checking its date stamp can therefore introduce a one-session timing error. That detail matters when comparing contract months, studying participation around rolls or expiry, or building systematic indicators from daily futures data.

Options open interest is the number of option contracts that remain outstanding rather than having been closed, exercised and assigned, expired or otherwise terminated. It is normally measured for a specific option series—meaning a particular underlying, call or put type, strike and expiration—and can also be aggregated across strikes, expirations or an underlying. Each outstanding contract represents both a long position and a corresponding short position, but open interest counts that matched contract once, not twice.

Why trading volume and open interest can move differently

Open interest originates from trading, but it is not a measure of how many contracts traded during the session. What matters is whether the buyer and seller are opening new positions or closing existing ones. Consider a single matched trade of one option contract. If the buyer opens a new long position and the seller opens a new short position, a new outstanding contract is created, so open interest rises by one. If instead the buyer is closing an existing short position while the seller is closing an existing long position, an outstanding contract is eliminated, so open interest falls by one. If one side opens a new position while the other side closes an existing position, one outstanding position is replaced by another and total open interest is unchanged.

For example, suppose open interest in a particular call option is 1,000 contracts before a trade. An opening buyer matched with an opening seller increases it to 1,001. A closing buyer matched with a closing seller reduces it to 999. An opening buyer matched with a closing seller—or a closing buyer matched with an opening seller—leaves it at 1,000. In every case, one contract may have traded and therefore added one contract to trading volume, yet the effect on open interest is different.

This distinction also limits directional interpretation. High or rising open interest shows that more contracts are outstanding, but it does not reveal by itself whether market participants are collectively bullish or bearish. Every open option contract necessarily has a long side and a short side. Nor does large open interest at a strike prove that the positions were established for the same reason: outright speculation, hedging, volatility trading and multi-leg strategies can all contribute to the same published total.

Open interest is position data, not a live liquidity quote

For U.S. listed options cleared by OCC, displayed series-level open-interest figures are derived from the previous day's settlement. OCC notes, for example, that figures displayed on a Friday morning represent open interest following Thursday's matching executions. Opening and closing transactions are consolidated in calculating open interest, while exercised options that result in assignment cease to remain open positions. Consequently, the published figure should not be treated as a real-time count of positions during the current trading session.

Open interest can help identify where outstanding exposure is concentrated and can provide useful context for participation, positioning research and expiration analysis, but it should not be used as a stand-alone measure of execution liquidity. A contract with substantial open interest can still have an unattractive bid-ask spread or limited displayed size, while an option with modest open interest can sometimes trade efficiently when active market makers are quoting it. For execution, current bids, offers, quoted size and trading conditions describe the immediate market more directly than the stock of contracts remaining open after prior clearing activity.

Other Reportables is a CFTC reportable-trader classification used in both the Disaggregated Commitments of Traders (COT) report and the Traders in Financial Futures (TFF) report. In each family, it is a residual category: a reportable trader is placed there when CFTC staff does not place it in one of that family’s other three reportable categories. The shared label therefore does not mean that the underlying participant mix is the same across the two report families.

Residual category, different frameworks

In the Disaggregated COT, the other three reportable categories are Producer/Merchant/Processor/User, Swap Dealer and Managed Money. In TFF, they are Dealer/Intermediary, Asset Manager/Institutional and Leveraged Funds. TFF places Other Reportables on the buy side and states that traders in this category mostly use the markets to hedge business risks; examples include corporate treasuries, central banks, smaller banks, mortgage originators and credit unions. Those examples should not be transferred to the Disaggregated COT category. Other Reportables is defined within each report family, not as one universal economic type.

Classification applies to the trader, not to the motive of every individual position. CFTC staff use Form 40 information and, where appropriate, other information about market activities, then assign reportable traders according to predominant activity. Multi-functional organizations and separately reportable entities can therefore be classified differently, and classifications can change as activities, traded markets, client bases or information available to the Commission change. A long or short balance in Other Reportables should not automatically be labeled speculative, hedging, bullish or bearish.

What the published positions actually show

For Other Reportables, both Disaggregated COT and TFF report long, short and spreading positions. Spreading is a computed amount based on offsetting long and short positions: offsetting futures in different calendar months, or offsetting futures and options in the same or different calendar months in the combined report. Any residual exposure is reported in the long or short column, and inter-market spreads are not included in this spreading measure. Because spreading can be a partial activity, the same trader can appear in an outright long or short trader count and in the spreading trader count, so column-level trader counts should not be added as though they represented mutually exclusive traders.

The classification is also separate from report format: Disaggregated COT and TFF are published in futures-only and futures-and-options-combined forms. For research, the deeper issue is composition through time. Changes in Other Reportables can reflect changes in positions as well as changes in classification. There is an additional historical limitation: when the CFTC constructed historical Disaggregated and TFF data back to June 2006, it stated that it did not maintain a history of large-trader classifications and therefore used then-recent classifications to backcast earlier positions. The CFTC cautioned that this reduced classification accuracy further back in the historical series. Other Reportables is therefore most useful when the analyst knows which COT family is being examined, what the category can represent there, and whether historical composition is being assumed to be more stable than the underlying methodology supports.

Overfitting occurs when a model, rule or research process adapts so closely to the particular sample used for development that it captures noise, chance relationships or sample-specific quirks as though they were persistent structure. The result can look impressively accurate in-sample while generalizing poorly to genuinely unseen data. In market research, the danger extends beyond complex machine-learning models: a trading rule can be overfit simply because enough indicators, parameter combinations, seasonal windows, filters or markets were tried until one happened to produce an attractive historical result.

Why a good backtest can be part of the problem

The central issue is not merely that a model has many parameters. Overfitting depends on the relationship between model flexibility, the amount and quality of data, and how extensively the researcher searches among alternatives. Even a simple strategy can become overfit when many variants are tested and only the best historical performer is retained. Each additional trial creates another opportunity to select a configuration that benefited from random variation. Research on financial backtests therefore treats the number of alternatives considered as economically important information: apparent performance can be inflated by selection itself, even when no individual calculation is incorrect.

This makes overfitting different from look-ahead bias. Look-ahead bias contaminates a historical test with information that would not have been available at the time. An overfit test can use perfectly point-in-time data and still fail because the research process has extracted more apparent structure from the historical sample than the data can reliably support.

The test set can be overfit too

Holding out data is useful only while that data remains genuinely independent of model development. If researchers repeatedly inspect validation or test results and then alter features, parameters, filters or model choices in response, information from that supposedly unseen sample gradually influences the model-selection process. The evaluation set has effectively become another development set. Validation designs that separate model selection from performance estimation—such as nested validation or a genuinely untouched final holdout—help address this problem, provided the evaluation procedure is appropriate for the data and is not itself repeatedly mined for favorable results.

Overfitting is therefore best understood as a failure of generalization, not simply as excessive mathematical complexity. Warning signs include a large gap between development and out-of-sample performance, unstable results under small specification changes, and performance that depends heavily on one historical period or parameter choice. None proves overfitting by itself: structural breaks, changing market regimes or ordinary sampling variation can also weaken out-of-sample results. The practical question is whether the claimed relationship survives evidence that did not participate—directly or indirectly—in discovering it.