COT’s twice-weekly plan: check what “previous” compares

Twice-weekly records change what “previous” means

On October 27, 2026, the U.S. Commodity Futures Trading Commission (CFTC) announced twice-weekly Commitments of Traders (COT) publication by year-end: Monday data on Wednesday, Wednesday data on Friday, at 15:30 ET. Latency would fall from three days to two.[S1] The start date is unspecified.

The change concerns more than how often a reader receives news. Moving down one row will no longer necessarily represent the same comparison interval. The familiar correspondence between a change from the previous observation and a change over one week breaks. Even if a display retains a previous-change label, the dates must establish how much time lies between its endpoints.

Readers want to know whether positioning has become more one-sided, persisted or reversed in between. Additional observations may reveal a round trip that was previously invisible. They will not reveal every transaction between two snapshots. The useful test is which additional question the denser information can answer.

This article uses one fictional ledger with observation dates and publication times. Readers can calculate adjacent changes, same-weekday changes, the latest four rows and a fixed date window, then apply the selection rules to their own records. The result is a check of the interval being compared and the information available at the time, rather than a trading answer.

A position change needs two dates

Open positions describe contracts remaining at a point in time. The CFTC’s current explanation distinguishes reporting-firm position data from participant classifications based on business purpose. A classification does not identify the purpose of each trade or reveal individual motives.[S2] Earlier delivery would not turn this statistic into a complete transaction history.

If a category’s net position moves from 100 to 160, the difference is 60. Without the dates represented by those endpoints, neither its pace nor its comparability with another difference is clear. A date forms part of the comparison itself and should remain attached to the calculated result.

The existing guide to long, short and net positions explains the basic formula. The task here is to specify the counterpart selected for subtraction. Using the previous row and using the same weekday a week earlier answer different questions even when the arithmetic is identical.

Matching the contract, category and futures-only or futures-and-options coverage remains a separate requirement. Dates alone cannot make different series comparable. The exercise holds these dimensions constant and changes the date selection. With real data, verify that those fixed conditions actually hold before interpreting the difference.

Build three comparisons from one ledger

Every position, long and short value in Figure 1 is fictional. None is a forecast or an unpublished CFTC reading. The dates follow the 2027 calendar for an exercise; they are not a commencement date or an actual release schedule. The example assumes ordinary weeks without holiday changes, a matched series and contracts as the unit. Net equals long minus short.

Short positions remain at 300 while long positions move through 390, 400, 460, 430, 420, 450 and 480. The resulting nets are 90, 100, 160, 130, 120, 150 and 180. Holding one side constant makes the differences easier to examine. Both sides can change in practice, so the simplification says nothing about how actual participants trade.

Choose the comparison before examining its outcome: adjacent observations, the same weekday or a fixed date interval. The ledger is not a device for selecting favorable endpoints afterward. Precise arithmetic can reinforce a mistaken interpretation when comparisons serving different purposes share one label. Test both the calculation and its fit with the question.

To use your own CSV or table, first match the observation date, publication timestamp, series identity and net columns. Do not infer an absent publication timestamp from the observation date; preserve the gap. Replacing missing observations with zero mixes the effect of denser sampling with a data-handling assumption. Working on a copy while retaining the source makes later choices traceable.

Adjacent differences cover unequal intervals

In the exercise, Monday, February 22 to Wednesday, February 24 spans two calendar days, and net rises from 100 to 160: an increase of 60. The next interval, ending Monday, March 1, spans five calendar days and falls by 30 to 130. The following two-day interval falls by 10 to 120. Reading the signs differs from comparing their sizes over equal intervals.

Two and five calendar days must not become two and five trading days. One interval crosses a weekend, while actual markets have holidays and differing trading hours. Dates establish elapsed calendar time; a trading-day count needs a separate calendar. Specify the type of day before comparing pace.

Dividing a difference by elapsed days yields an average endpoint change per day. It does not yield daily net buying. A large increase followed by a reversal can have the same endpoints as a gradual move. An adjustment for interval length must not manufacture an unobserved daily transaction history.

A report saying “up 60 since the previous observation” becomes more reusable when it states the endpoints: February 22 to 24. If the question concerns a week, construct a separate same-weekday comparison. Keeping both allows the latest short move and the change remaining over a longer interval to retain their different meanings.

Matching weekdays shows the change remaining over a week

Select Wednesdays only: net falls from 160 on February 24 to 120 on March 3, then rises to 180 on March 10. The weekly changes are minus 40 and plus 60. Mondays show a different sequence: 100 to 130 to 150, or increases of 30 and 20. Endpoints on different weekdays can therefore produce different weekly changes within the same ledger.

This is not an invitation to find the weekday with the strongest trend. Decide beforehand whether a weekday follows an established reporting process or a relevant event cycle. Switching weekdays after seeing the outcome lets the selection itself shape the conclusion, despite apparently consistent calculations.

Nor does disagreement between weekday comparisons prove that one number is wrong. In the first part of the exercise, an increase early in the week and a subsequent decline both occur. Different endpoint windows contain different parts of that path. When signs disagree, displaying the dates can explain more than collapsing the calculations into one measure.

Sum adjacent changes back to the same endpoints

Selecting a fixed weekday from newly available observations differs from replacing the historical Tuesday series with Wednesday values. An unobserved past weekday cannot be recreated by changing its label. Preserve historical observation dates and record when each new comparison rule becomes usable rather than pretending to convert the past.

The same-Wednesday change can be checked by adding the intervening adjacent differences. February 24–March 3 gives minus 30 over five calendar days plus minus 10 over two: minus 40. March 3–10 gives plus 30 plus 30: plus 60. Intermediate values cancel because the series and endpoints match. Failure to reconcile points toward mixed series, gaps or endpoint selection. Reconciliation still does not reconstruct transaction volume.

Identical weekly endpoints can conceal different paths

Figure 4 compares two fictional paths with the same Wednesdays. Path A runs through 160 on February 24, 130 on March 1, 120 on March 3, 150 on March 8 and 180 on March 10. Path B changes the two Mondays to 80 and 220. The Wednesday values remain 160, 120 and 180, so Wednesday-to-Wednesday changes are identical.

The added Mondays reveal different states across the weekend-spanning intervals. A’s 130 and B’s 80, and A’s 150 and B’s 220, cannot be distinguished using Wednesdays alone. The value of denser observation lies in separating paths that previously looked identical, rather than guaranteeing a better forecast of the final point.

That difference may matter when examining business hedging conditions: the same final net can coexist with a different concentration of positioning or burdens in between. Aggregate categories cannot establish that a particular company paid additional margin or lacked cash. Connecting a position path to business consequences still requires evidence about the firm’s financing conditions.

The rows follow observation-date order, but their spacing is not proportional to elapsed days. No daily values are interpolated between observations. Adding two Mondays still leaves intermediate round trips unobserved. This reading recognizes the information gained without filling an unobserved path as though it had been measured.

Endpoint changes cannot reconstruct turnover or profit

A fall in net from 160 to 120 can follow substantial trading by several participants or a smaller adjustment in holdings. Both can leave the same difference of minus 40. Comparing endpoint stocks does not uniquely recover transaction volume within the interval.

Keeping long and short columns reveals more than net alone, but still does not identify every counterparty or transaction behind the remaining position. Category movements and entry or exit from the reporting population also matter. The CFTC’s FAQ identifies classification and participant changes among explanations for changes between reports.[S2]

Profit requires further information: execution prices, the holding path, settlement and costs. Replacing the sign of a quantity change with the sign of profit silently assumes a price path as well. A denser aggregate position report does not become an income statement.

Keep the explanation of remaining positions separate from the explanation of transactions or profit. If evidence answers the first question but trade-level records are absent for the second, record the particular missing information. This keeps one number from being asked to perform more work than it can support.

The latest four rows do not define a constant period

Figure 5 fixes the endpoint at March 10. The latest four rows are March 1, 3, 8 and 10, spanning nine calendar days. Select Wednesdays from the same ledger and take the latest four observations instead: February 17 and 24, March 3 and 10. Their endpoints span 21 calendar days. Equal counts contain different amounts of time.

The latter is a fixed-weekday reference for the exercise, not a reconstruction of the actual historical Tuesday COT series. It thins the fictional Wednesdays to compare row counts at the same endpoint. It cannot convert historical data into observations for a weekday that was never measured.

This difference affects both averages and comparisons of how unusual the current position is. Adding denser observations to a fixed-count window ejects old observations sooner. A changing assessment may reflect a different economic state, a shorter historical comparison period, or both. Preserve those possibilities rather than assigning the whole change to the market.

Fifty-two regularly spaced weekly observations contain 51 intervals, spanning 51 weeks. Fifty-two alternating Monday/Wednesday observations contain 51 intervals and span either 25 weeks plus two days or 25 weeks plus five days, depending on phase. These are calendar calculations without holidays or gaps, not exactly 26 weeks. Preserve both the row count and the actual endpoints.

Fix the dates, and the count changes instead

The inclusive February 24–March 10 window contains five exercise observations: 160, 130, 120, 150 and 180. Restrict the same date interval to Wednesdays and three remain: 160, 120 and 180. If the question concerns information within a fixed period, a changing count is itself part of the result.

Giving each of the five observations equal weight counts each observation once. It differs from treating a value as persisting until the next observation. The latter adds a holding-between-observations assumption. Without actual daily positions, do not silently introduce that assumption and call the result an average daily state.

The appropriate aggregation depends on the question. A distribution of published observations can use observations as units. An assessment of holdings over elapsed time raises a separate problem concerning the unobserved intervals. More rows do not automatically define a superior population for every purpose.

The same formula receives different selected sets

The existing percentile and z-score guide explains historical-window choices. The additional task here is checking whether an unchanged period label contains a different pattern of observation dates. Save the first date, last date, count and weekday composition as well as the window’s name so the change can be reconstructed.

Simple means provide another exercise check. The latest four rows total 580, averaging 145. The five-row fixed-date window totals 740, averaging 148. Its three Wednesdays total 460, averaging approximately 153.33. Different selected sets enter the same formula. These are fictional ledger checks, not market readings or recommendations of an averaging method. List the selected dates before interpreting the difference.

A transition changes density within the same history

Appending denser observations to an older weekly history creates a window with sparse and dense sections. In a fixed-count calculation, a short recent period can occupy as many rows as a much longer older period. Treating all rows as weekly silently changes the time weighting.

Increasing the frequency of the past needs underlying observations. Duplicating a historical Tuesday into Monday and Wednesday rows adds no information. Linear interpolation is an estimate, not an observation. If interpolation serves a legitimate purpose, keep it separate from the original series and state the assumption creating the additional rows.

Discarding extra observations can restore a similar count, but may change the weekday and sacrifices information about the shorter path. Do not thin solely to make histories look alike. Design the continuity comparison separately from the use made of the extra observations.

Which earlier conditions remain in the window?

Compare whether a conclusion survives both a fixed-period and fixed-count window during transition. If it changes materially, inspect which earlier episodes leave the sample and which recent episodes enter more densely. Record the changed selection rule and affected period rather than concealing the change in conclusion.

Density can also alter which past conditions a window effectively remembers. A short recent period may be counted more often while an earlier turbulent episode drops out sooner. That can reflect current market conditions, window construction or both. It is too early to conclude that turbulence has disappeared without separating these comparisons.

Two observations need not be two independent pieces of evidence

Nearby observations from one week may strongly reflect the same market environment and participants. Doubling the row count does not establish a proportional increase in independent information. Adding correlated records differs from observing a genuinely different set of conditions.

The Mondays separating paths A and B in Figure 4 add information about their differences. An extra observation nearly identical to Wednesday’s may instead corroborate persistence. Both can be useful, but serve different roles. Ask which question the additional point advances before doubling confidence along with the count.

The explanation of sample counts and dependence provides background. Here the issue takes a concrete form in weekday placement and continuity of the same contract and category. Testing whether a judgment based on two points survives in other weeks or regimes requires more than adding rows.

Which value is usable at 3:29 p.m. on March 3?

Figure 6 assumes fictional ordinary-week publication two days after each exercise observation, at 3:30 p.m. ET. At 3:29 p.m. on March 3, the March 1 value of 130 has not been published. The latest available row is February 24’s 160. The value of 120 observed on March 3 awaits a still later release.

At 3:30 p.m., 130 becomes available. A new row appearing now still describes March 1. Reading it as a state reached at the instant of publication merges the observation and availability clocks. Preserve the information date to explain what changes across the release boundary.

The eligibility rule is publication timestamp less than or equal to judgment timestamp. Observation date before judgment date is insufficient. Match time zones when timestamps share a calendar date. If delivery is late, evidence of actual availability must be distinguished from the scheduled timestamp.

Public availability and the time needed to use it

This matters when assessing whether a past judgment was reasonable at the time. A table acquired later includes rows unavailable then. Coexistence in one file is not evidence of simultaneous historical availability. The point-in-time comparison procedure helps preserve those relationships as records.

Passing the availability test does not establish instantaneous use in a trade or meeting. Acquisition, checking, distribution and reading each take time. The exercise’s 3:30 p.m. boundary identifies public-information eligibility, not completion of every user’s processing. Assessing operational benefit also requires the delay from publication to actual use.

Keep scheduled release, retrieval and judgment times separate

A scheduled release helps organize preparation and processing. Retrieval time records when a particular workflow obtained the material. Judgment time records when someone used it in an explanation or decision. Even an on-time release leaves these timestamps different if acquisition occurs later.

Suppose the exercise’s 3:30 p.m. release is retrieved at 4 p.m. and a memo is prepared at 4:10. The memo’s information set can be matched to that retrieval record. Retrieval at 4 p.m. alone does not prove that identical content was public at 3:30. An analysis requiring scheduled and actual availability needs evidence connecting them.

International users should also avoid treating a local time label as a permanently fixed UTC offset. Retain both Eastern Time attribution and the time zone of the consuming process to preserve date-crossing relationships. The guide to sessions and daylight-saving errors explains why conversion assumptions matter.

The purpose is not to burden a report with timestamps. It is to distinguish a value discrepancy, accidental inclusion of an unpublished row and simple retrieval delay when results disagree. Combining everything under one “data date” removes evidence needed to reproduce the problem.

A missing row is not a day without position changes

Remove the fictional March 1 value of 130. The previous available observation for March 3’s 120 becomes February 24’s 160, producing a fall of 40. The March 1–3 change has not expanded from minus 10 to minus 40; the endpoints have changed. Inspecting the date sequence explains the difference.

Carrying the previous value forward introduces another assumption. Filling March 1 with 160 creates an artificial zero-change interval, followed by a decline of 40. It shifts where the change appears to occur. A convenience introduced to run a calculation must not survive as an observed position.

A contract disappearing from a report differs from a failure to retrieve or match its file. The first calls for examining coverage; the second, acquisition and joins. Leave an unexplained gap explicitly unresolved rather than inserting zero or a carried-forward value. This preserves the next question to investigate.

When joining sources, match series identity as well as observation date. A shared date does not make different categories or aggregation bases the same row. The guide to data coverage and date boundaries helps frame the comparison conditions, including treatment of gaps.

A release response and a post-observation price change are different questions

If the fictional March 1 value is published on March 3, studying prices after its publication uses the release as the starting point. Studying prices from March 1 instead asks what happened after the positions were observed. Switching origins changes the relationship under examination.

A price move before publication cannot be explained as a response to that public release. Nor does a price move after observation establish that the position predicted it. Whether the value was available at the time and what other information was available remain separate conditions.

When several statistics or policy announcements occur near each other, attributing a price change to one source becomes harder. Proximity can identify a comparison to investigate but does not complete a causal argument. Record the chosen release boundary and examine other contemporaneous information.

Position comparisons alone do not establish price responses or predictive performance. Adding fictional price reactions can make a conditional illustration resemble observed results. Examine market implications through the availability and granularity channels first; profit and direction require separate evidence and a suitable testing design.

Reconcile your own file with eight fields

Figure 7 is a worksheet for reading and manual transcription. It does not accept inputs or upload, process or store CSVs. Readers supply their records and note series identity, observation date, publication timestamp, comparison rule, endpoints, count, gaps and result label. Confidential account details or individual orders need not be sent to this site.

First preserve the source’s column names and isolate one series. Then match the observation-date ordering to the selection made by publication time. Do not assume the two orders coincide; retain identifiers linking the same observation. Before removing duplicate rows across files, establish whether they are repeat retrievals or different aggregation scopes.

Write the selected counterpart alongside a short label such as “previous” or “weekly.” For March 8’s 150, identify whether it is compared with March 3’s 120 or March 1’s 130. Once the counterpart is explicit, plus 30 and plus 20 coexist as different comparisons without contradiction.

Finally match the output label to the calculation. Check whether an adjacent difference is labeled weekly, an observation-date state is described as a publication-date state, or a filled value appears as observed. A formula check alone can miss a correct value with a wrong explanation. Complete the worksheet by checking that explanation too.

Passing the same number between two departments

Consider procurement explaining raw-material price conditions and treasury examining hedging, both using one COT series. Procurement may use same-weekday weekly differences while treasury uses the latest adjacent change. Different amounts need not mean one calculation is wrong. The problem arises when both travel under the same “change from previous” label.

Agreement on definitions does not require agreement on conclusions. Series identity, endpoints, availability and selection rule let each team preserve the comparison appropriate to its task. Procurement’s negotiation horizon may differ from treasury’s cash-management horizon; recording that difference is more useful than suppressing it.

Transmission to international firms and information providers begins with such handoffs. Users abroad incorporate a U.S. release into local meetings and processing deadlines. A new observation arriving before a meeting can expand its information set; arriving afterward will not change that meeting’s explanation. A faster schedule and usable operational benefit require different measures.

The guide to matching macro sources by question and timestamp provides background. More detailed COT alone cannot establish procurement costs if prices, inventories and contract horizons do not match. Identify what each source describes and record which decision the added observation actually changed.

The benefit of earlier information and the cost of changing processing

An extra observation may shorten reliance on an older state. For a path that changes in between and returns near the endpoint, Figure 4 shows how more states become distinguishable. That is an information benefit, not proof of improved price-forecast accuracy.

Updating, checking, explaining and redistributing data also require work. Some processes may need only another date; others need comparisons rebuilt around their implicit one-row-per-week assumption. Locate those assumptions rather than estimating the impact from an organization’s size alone.

Information-service value cannot be measured solely by update count. Acquisition, series matching, period management and explanations impose separate costs. Two services updating equally often may leave users with very different checking burdens if one makes the relationship between dates and calculations traceable.

Whose completed task defines the benefit?

The cost need not permanently reduce margins. Reusable checks and clear rules can lower later rework. Assess whether users can make the required comparison accurately rather than rewarding update counts. No common cost estimate or price-pass-through rate is established here; actual processes and contracts determine those questions.

Counting updates alone can miss work transferred to users. A publisher may finish an update while consumers still investigate column names and reference periods, leaving time to a usable judgment unchanged. Conversely, clear comparison rules can save more rework than the extra update count suggests. Decide whether the benefit ends at delivery or at completion of the user’s task.

What must still be checked within a shorter process?

The CFTC’s May 5 request for comment described checks for missing or erroneous data, reporting-firm inquiries, classification corrections and aggregate validation. It also asked about accuracy and process changes when publishing more recent information.[S3] Those questions do not establish that the planned delivery will be inaccurate; they frame the problem of preserving checks while improving speed.

Automation may shorten a sequence, but accelerating ordinary records does not explain unresolved exceptions. Separate detection, receipt of a reporting firm’s response and authorization for publication. Identify which step changes instead of evaluating the entire process by elapsed time alone.

Consumers should look beyond receipt time too: required contracts, category continuity, unexpected gaps and later explanatory information matter. Faster updates can increase the importance of keeping versions and partial retrievals distinct. Preserve acquisition records alongside the source’s reference time.

The separate News analysis of provisional industrial production and revision ranges offers a useful comparison. It does not determine COT processing or correction policy: one statistic’s revision practice cannot be transferred to another. The shared question is how speed relates to the strength of evidence needed for a decision.

More transparency still needs disciplined attribution

The May consultation also asked whether greater frequency or recency could make participant identity or proprietary information easier to infer.[S3] Asking about the risk is not evidence of a realized harm under this plan. It provides context for evaluating protection of the reported population alongside information benefits.

A larger aggregate category position does not identify a particular company or country as the buyer. Multiple participants and specific coverage definitions stand behind the total. Denser observations do not authorize stronger claims about identifiable parties.

In a smaller market, combining a few rows with other public information may appear persuasive. Plausibility is not verification. Consider whether other parties or paths could explain the same aggregate result. Reporting the change supported within the aggregate’s scope is more defensible than escalating an unverified attribution.

Questions for counterparties need not presume their behavior from an aggregate. Ask which price risk they manage, over what period and under which contract conditions. Those questions lead to the counterparty’s own records. Public totals can initiate an inquiry while the operational conclusion remains tied to the relevant party’s evidence.

There are reasons to retain a weekly comparison

Using every additional observation with equal weight is not the only reasonable response. An organization examining longer changes can retain a fixed-weekday or fixed-period report and use the added observations to investigate shorter paths. Different purposes can justify maintaining two views.

A fixed weekday carries its own sampling pattern, but continuity helps match new judgments with earlier ones. Additional observations can reveal states that pattern misses. Giving the continuity view and the denser view different roles avoids forcing all information into a single indicator.

Maintaining two views could complicate communication or invite cherry-picking. Pre-specify their uses and selection rules: for example, same-weekday changes in the regular report and adjacent changes in a clearly labeled update. Do not reverse their priority after seeing the outcome. That operating rule can matter more than another formula.

If extra observations rarely alter relevant assessments and add processing burden, a narrower use may be appropriate. Supporting the frequency increase does not require assuming identical benefits for all users. Test which question each added observation advances.

SG Group View: follow the comparison rule, not just the new row

SG Group would assess the plan through more than delivery count. Additional observations can distinguish previously hidden paths while creating work to preserve the meaning of weekly changes and historical windows. Readers need material that explains the interval being compared as well as delivering a newer value.

Three tests help: does the extra observation distinguish previously indistinguishable states; can users keep same-weekday and adjacent changes separate; and can they trace endpoints and density across the transition? These need not improve together. Each requires its own evidence.

The emphasis on transition burden would weaken if comparison errors remain small, additional observations consistently reveal useful distinctions and processing remains stable. Attention could shift toward the information benefit. If unchanged weekly-change and history labels conceal incompatible calculations, adaptation deserves greater weight instead.

Improved market liquidity or price formation lies further downstream. User responses, contracts and other information can alter the effect of a denser public dataset. Figure 8 separates evidence about comparison rules, operational use and market outcomes. Additional observations alone cannot establish the last result.

The next documents should establish commencement, coverage and exceptions

Before operational use, establish the first observation and release actually available under the new format. A planned commencement and a published file support different conclusions. When a first file is identified, match its reference date, publication time and coverage to the explanatory document. Passing a planned date is not proof of operation.

Match report types and contract coverage to implementation material. Do not expand a short announcement into a promise about every file specification, category or historical series. Only after confirming that a particular contract and format enter the scope can a user define the required adaptation.

Holiday changes, delays, gaps and the transition boundary do not follow automatically from an ordinary-week exercise. Generating dates from the weekday pattern does not establish that observations or releases occurred. Compare official scheduling information with acquired files and retain the affected exceptions.

Existing SG Group guides or displays do not by themselves establish support for the new arrangement. This article provides manual checks, not automatic transfer of data or settings into another function. Distinguish an educational entry point from the currently offered ability to process a changed data structure.

Can you return “up 30” to the right explanation?

Use the fictional March 8 value of 150 for a final check. The preceding observation is March 3’s 120, giving an adjacent increase of 30. The previous Monday is March 1’s 130, giving a weekly increase of 20. These cannot both become an explanation saying “up 30 over the week.”

First turn a counterpart selected by row number back into a date. A calculation storing only “previous value” may select another endpoint after a gap or additional observation. Saving March 3 beside the result makes the comparison testable. For the same-weekday change, record why March 1 was selected.

Second establish when the explanation became possible. The exercise publishes the March 8 observation on March 10, so 150 cannot enter a March 9 judgment. Replace an observation-date-only condition with a publication-time condition and the latest eligible row changes. The timing distinction becomes an actual selection.

From historical selection back to your source

Third inspect the historical range. Choosing the latest four rows rather than the inclusive February 24–March 10 interval changes whether 160 enters. Values appearing on the same screen need not all enter the calculation. Confirm the application of the selection rule as well as the presence of the numbers.

Fourth apply the checks to your own source. Hold contract, category and aggregation coverage constant, and use rows whose observation dates and publication times can be traced back. Investigate missing timestamps or coverage before substituting estimates. The worksheet helps record those gaps locally.

These tasks remain personal to the file and purpose even after a summary explains the plan. Delivery frequency alone does not decide what the extra row establishes or which comparison should continue. Return each number to its observation and availability dates, then match the result’s label to the calculation.