Data releases in practice — how to read surprises and market reactions
Before a release, you have forecasts and earlier data. Afterwards, you have the new reading, sometimes revisions, and the price reaction. A timestamped record made before the event helps keep these layers separate. We build that record, calculate the surprise and examine what the market reaction can tell us. We assess execution risk separately: a chart candle does not show the spread or the price at which an order could actually have been filled.

Before publication, fix the comparison
A release changes the information available to you. To assess that change, preserve the benchmark before seeing the result: series, unit, seasonal adjustment, survey provider, forecast timestamp and the version of the previous observation.
Consider a wholly hypothetical monthly activity index. At 09:55 local source time, your archived survey median predicts growth of 0.4% from the previously published index level of 124.0. That implies a comparison level of 124.0×1.004=124.496. Your note says to check the full release at 10:00, then compare the same two-year rate instrument after 5 minutes. Its starting yield is 3.40%; a rise of at least 2 basis points is your chosen condition for support for a rate-channel hypothesis. This threshold is an exercise rule, without an estimated success rate.
Save the time the material actually reached your data source. The BLS review of release procedures illustrates why a scheduled release and observed public availability need separate fields; it supplies none of this invented index's values.
A faster growth rate can accompany a lower level
At 10:00 the release revises the prior index to 123.6 and reports current growth of 0.6%. The growth surprise is +0.2 percentage points: 0.6−0.4. But the new level is 123.6×1.006=124.3416, which lies 0.1544 index points below your pre-release benchmark of 124.496. These comparisons answer different questions.
| Record | Previous level | Growth | Implied current level |
|---|---|---|---|
| Before release | 124.0 | 0.4% forecast | 124.496 |
| Published package | 123.6 revised | 0.6% actual | 124.3416 |
Do not add the revision in index points to a surprise in percentage points. Reconstruct levels first, with matching definitions and adequate precision. Neither result alone says whether EUR/USD should rise: the currency concerned, the other currency, policy expectations and other simultaneous news still matter. Keep the arithmetic result separate from its economic interpretation.
At 10:05 the same instrument yields 3.41%. That is +1 basis point, below the +2 basis point condition fixed beforehand. The selected rate-channel hypothesis lacks support under this rule. It does not follow that the data contained no information or that all rate expectations stayed unchanged. Other maturities or premia could have moved; one instrument cannot identify the entire transmission mechanism.
A quote is an observation; a fill needs assumptions
Here is a separate hypothetical EUR/USD quote path, in USD per EUR. One pip is 0.0001. The times are measured from the release becoming available at our source.
| Time | Bid | Ask |
|---|---|---|
| Immediately before | 1.0848 | 1.0850 |
| First received update | 1.0854 | 1.0860 |
| +30 seconds | 1.0857 | 1.0859 |
| +5 minutes | 1.0862 | 1.0864 |
The first post-release spread is 6 pips, compared with 2 pips before publication. A chart using only the midpoint hides this difference. For this calculation, assume a purchase of 20,000 EUR fills in full at the +30-second Ask of 1.0859, followed by a sale in full at the +5-minute Bid of 1.0862. Assume no commission, financing or additional slippage. Profit is 20,000×(1.0862−1.0859)=6 USD. The 3-pip executable change already includes the two-sided quote cost; subtracting another spread would count it twice.
The midpoint rose 5 pips over those same two observations. That is a price statistic, not this position's return. Quotes alone do not prove that the assumed size could have filled at either price. Record size, latency, order type, accepted fills and fees when moving from a teaching calculation to actual execution. A favourable outcome in one invented path provides no evidence for a repeatable release strategy.
Three exercises with worked answers
The exercise data are hypothetical. Choose an answer, then check the calculation.
From one release to a complete reading process
The following sections examine survey dispersion, release clocks, simultaneous reports, revisions and historical documents. Use the worked record above to keep each observation's unit, timestamp and evidential limit visible.
1. What is known before the first number arrives
Start the week with the release calendar, then preserve your benchmarks before each event. Record what you know while the outcome is still unknown. Otherwise, a later price move can make an explanation feel obvious even when it was absent from the original note.
Your calendar review and notes should answer six questions. The publishing institution’s schedule supplies the date; a commercial calendar helps collect events, but its importance labels and forecasts need separate verification.
- Which releases are expected to have the greatest impact this week? Employment reports, inflation and decisions by major central banks are examples. Check where this assessment comes from: importance labels in economic calendars (“high impact”, “tier 1”) are a provider’s convention, rather than a measurement. Two services may label the same report differently, and its actual impact changes with the economic cycle — inflation is read differently when the central bank is fighting it and when it is already near target.
- Which currency does each release concern, and what is the exact time in the source’s time zone? Record the release time in the publishing institution’s local time zone and convert from there. If the Bank of Canada decides on Wednesday and UK inflation is released on Thursday, these are two separate events involving two different currencies. Both USD/CAD and GBP/USD also contain the dollar, so each pair’s reaction needs to be decomposed.
- How has the priced rate path changed since the previous release of this series? A news agency’s consensus is a starting point, but answers a different question from instrument pricing. Record both: the survey median with its closing date, and what interest rate contracts price immediately before the release. Last week’s exchange rate movement cannot establish positioning — that is the backward inference from price discussed in section 3.
- What other releases occur on the same day at a similar time? If two reports concerning different currencies appear minutes apart, attributing a move to either requires the procedure in section 5. Record this beforehand: afterwards, selecting the “right” release becomes a fit to what has already happened.
- Which releases warrant a full record? Choose those that could change the answer to your question: the assessment of interest rates, growth, external financing or risk premia. Refer to a specific series and central bank communication instead of treating the rate channel as the sole criterion. Also record other events within the measurement window; they may complicate attribution.
- What does a ready-made surprise index describe? Before adding it to your notes, check the series, signs, weights, scale and decay window for older observations. A positive unemployment surprise may be signed oppositely to a positive employment surprise. A high index therefore does not mean every raw number exceeded forecasts. It describes an aggregate of defined surprises; it measures neither positions nor the next report’s likely result.
2. A template for preparing for a release
The weekly plan provides a framework. Before a specific release, close and timestamp your note: add subsequent information separately. Your knowledge is still incomplete — you do not know everyone else’s forecasts or exposures — but you preserve what was actually available from your source.
Three possible readings
Before publication, record a reading below, near and above your chosen benchmark. Set and explicitly justify the boundaries according to your question. Minimum and maximum forecasts provide a supporting description of the panel: a reading within that range can still differ substantially from the median, especially when the extremes represent individual forecasts.
| Scenario | What to record beforehand | What to check afterwards |
|---|---|---|
| Above the benchmark | Difference in the reading’s units; forecast distribution | Headline, other fields, revisions and instrument reactions |
| Near the benchmark | Chosen range and rationale; do not equate it with min–max | Whether details or revisions add new information |
| Below the benchmark | Difference and an explicit condition for changing the interpretation | The same fields as in the other scenarios, without reversing an exchange rate rule |
For each scenario, record what would undermine its interpretation. Add this condition to the note alongside the table, with the instrument and observation window. It gives the later review a test that was fixed before the result was known.
The employment data package — which layers to read first
The NFP headline, earnings, revisions and CPS data are published together. The diagram below shows the reader’s workflow, rather than the order in which report components become available. The reaction of rate instruments is a market observation, rather than a BLS field. The record may remain ambiguous even after the whole report has been read.
You can scroll the diagram horizontally or open it at full size.
Checklist for 30–60 minutes before a release
- Forecasts preserved. Record the provider, aggregate value, forecast count, distribution and retrieval time. Preserve both the previous reading known then and any later revision; do not replace the value in the pre-release note.
- Informal expectations. Commentary published after the survey closes may describe a different forecast, but does not measure the entire market’s expectations. In a hypothetical example, a survey at 180,000, later commentary at 210,000 and a reading of 195,000 produce differences of +15,000 and −15,000 respectively. Both benchmarks are meaningful only with a preserved source and timestamp; later exchange rate movements cannot reconstruct them.
- Exchange rate movements in recent hours — an observation rather than a positioning conclusion. A pre-release move is sometimes interpreted as positioning for a particular outcome, but price alone does not reveal who holds which side. Record direction, magnitude and times; the third layer in section 3, rather than the chart, supports inferences about pricing.
- COT positioning — context rather than today’s position snapshot. The U.S. CFTC’s Commitments of Traders report describes positions as of Tuesday and is normally released on Friday at 15:30 New York time. Several days have therefore elapsed by a subsequent data release. Check the report version: Leveraged Funds and Asset Manager/Institutional are categories in Traders in Financial Futures, while Commercial and Non-commercial belong to the older Legacy report. Mixing the two creates a picture found in neither. An extreme position in one category is sometimes interpreted as indicating the room left for a further move in the same direction, but that interpretation concerns positions several days earlier, rather than at the release time. See article 6.6.
- Rate instrument pricing. Record the instrument, meeting, price, conversion model and observation time. A probability moving from 40% to 70% changes by 30 percentage points. Only the expected rate is measured in basis points: with two outcomes — no change or a 25 bp cut — the expected change moves from −10 bp to −17.5 bp, a change of −7.5 bp. This is a hypothetical calculation, rather than an archived FedWatch result.[5]
- The broker’s spread. Open your platform and check whether the spread is normal. It can widen before publication, before any number is announced — a response to the approaching event rather than to the data’s content.
- Other data that day. Check whether anything else is released alongside the U.S. employment report. Canadian employment data are sometimes released at the same source time (8:30 U.S. Eastern Time, also 8:30 Canadian Eastern Time). If so, USD/CAD reflects two releases and cannot by itself establish which caused its reaction.
- Scenario record closed. Timestamp the three scenarios, abandonment condition and observation window before publication. Afterwards, use the record for comparison, rather than filling it in.
- Scope of the conclusion recorded. An interest rate contract price is not a direct survey about NFP or CPI. Preserve its level and timestamp, but do not work backwards to a single “number the market expected”.
Publisher-local time, UTC and your display zone in 2026
Preserve the publisher’s date and named source zone before converting. America/New_York uses EST (UTC−5) or EDT (UTC−4); “ET” is not one fixed offset. Convert the instant to UTC, then to your chosen IANA display zone. The date-specific ET converter in lesson 4.5 accepts the source date and hour, including 15:30 for a COT example.
| New York dates in 2026 | Source offset | 08:30 New York → UTC | 15:30 New York → UTC |
|---|---|---|---|
| 1 January–7 March | EST, UTC−5 | 13:30 UTC | 20:30 UTC |
| 8 March–31 October | EDT, UTC−4 | 12:30 UTC | 19:30 UTC |
| 1 November–31 December | EST, UTC−5 | 13:30 UTC | 20:30 UTC |
UTC does not change its clocks. Your selected display zone can have its own DST dates, a different offset or no seasonal change. Confirm the publisher’s actual schedule and any holiday exception separately; an offset conversion cannot establish a release date.
Regional case study: Europe/Warsaw and the US–European DST mismatch
| Dates in 2026 | 8:30 New York in Warsaw | 15:30 New York (COT) in Warsaw |
|---|---|---|
| 1 January–7 March | 14:30 | 21:30 |
| 8–28 March | 13:30 | 20:30 |
| 29 March–24 October | 14:30 | 21:30 |
| 25–31 October | 13:30 | 20:30 |
| 1 November–31 December | 14:30 | 21:30 |
On Fridays 13, 20 and 27 March and 30 October, the standard COT release time of 15:30 ET is 20:30 in Warsaw. On 6 November it returns to 21:30. Before each event, also check holidays and schedule changes. The CFTC’s 2026 schedule includes releases on 16 and 30 November instead of the usual Friday. The schedule gives the publication date; read the position snapshot date from the specific file.
The scheduled time and actual availability can differ
A documented example illustrates the distinction. The BLS scheduled its preliminary CES benchmark announcement for 21 August 2024 at 10:00 ET. The official Team of Experts report of 25 November states that technical problems delayed its appearance on the website until 10:32; some staff had already disclosed it in telephone responses. Source: “Delayed Release of CES Preliminary Benchmark Announcement”.
In your journal, preserve the scheduled time, the documented public availability time and the receipt time in your own source separately. Do not describe movements between 10:00 and 10:32 as reactions to a webpage that had not yet provided the information. This episode does not establish that every earlier move reflects a leak or that all regular BLS releases suffer similar delays.
3. How to measure a surprise
A “large surprise” without a measure is a judgment, rather than an established finding. Before calling a reading surprising, calculate how far it differs from expectations — and state your benchmark. There are three layers, each answering a different question.
First layer — the difference in the reading’s units. In a hypothetical example with an initial reading of 263,000 jobs and a median of 250,000, the difference is +13,000. Subtraction is straightforward, but the result depends on the series, SA/NSA status, reading vintage and preserved survey. Today’s post-benchmark value does not substantiate a historical first release. The next question concerns scale: relative to which sample would +13,000 be a large difference?
Second layer — the chosen denominator. Historical standardisation divides the surprise by the standard deviation of past surprises for the same series, using one forecast source and first-release vintages. Disclose the window, observation count, SA/NSA status, unit and choice of population or sample standard deviation. Dispersion in today’s forecast survey is a different measure: it describes agreement among economists, rather than the typical past error. A reading’s position relative to the minimum and maximum alone does not tell you how many forecasters erred in either direction; individual forecasts are needed.
Third layer — the instrument’s reaction. The change in the implied rate after publication is a separate observation in basis points. It requires the same contract, model and specified before/after window. It may support a hypothesis of changing expectations, but prices also contain premia and market effects. An unchanged price does not prove that the data contain no new information.[5]
A ready-made surprise index may use different weights, signs and rates of decay. Its name alone does not reconstruct its methodology. Record the provider’s definition, version date and the scope of documentation reviewed. If the method is unavailable, use the index as identified external context; do not present its points as your own standardised surprise measure.
The same median, different agreement among economists
Take two complete, hypothetical NFP surveys in thousands of jobs. A: 160, 170, 180, 190, 200. B: 100, 140, 180, 220, 260. Both have a median of 180. A reading of 195 differs from the median by +15. However, the population standard deviation of forecasts is 14.1421 in A and 56.5685 in B: sqrt(Σ(x−mean)^2/n). Dividing the difference by these denominators gives 1.0607 and 0.2652.
This standardises against current forecasts, rather than historical errors. In both surveys, 195 lies within min–max, but exceeds four of five forecasts in A and three of five in B. Min–max alone does not show that distribution. For a historical measure, each earlier event requires its own first reading, a survey closed beforehand and a consistent unit.
Your own historical surprise measure: show the full sample
Take five earlier hypothetical events from the same series in thousands of jobs. Closed survey medians are 100, 120, 110, 130 and 140; first readings are 80, 110, 110, 140 and 160. Errors are −20, −10, 0, +10 and +20. The mean error is zero, and squared deviations sum to 1,000. The population standard deviation is sqrt(1000/5)=14.1421; the sample standard deviation is sqrt(1000/4)=15.8114. A new surprise of +15 therefore has a scale of 1.0607 or 0.9487 respectively. “One sigma” is incomplete without the convention.
The exercise provides the complete history. In a real sample, each row requires its own archived median with a closing date and the statistical office’s first announcement. The standardisation window for a new reading ends before that reading: including its error in the denominator changes the question. Do not join post-benchmark data to earlier consensus forecasts or mix SA with NSA. A change in provider or survey composition may require a separate sample. With five observations, the result is unstable; here it serves only to demonstrate the calculation.
The survey, your forecast and a measurable outcome
The survey median and your forecast may differ. In the model, a median of 180,000, your forecast of 205,000 and a reading of 195,000 yield a survey surprise of +15,000, your forecast error of −10,000 and an absolute forecast error of 10,000. These are three numbers from the same event. Do not change your benchmark afterwards to “I knew it would exceed 180”. An interval preserved before publication also needs a recorded definition: a forecast interval, survey range and hypothesis tolerance are not interchangeable.
Evaluating the process also requires a list of omitted events. If you record only spectacular reactions or successful forecasts, the journal is not a representative sample. Measure the same errors over a defined event set and preserve missing data. An accurate macro forecast does not imply an accurate price forecast: the latter concerns a different outcome, window and information set.
A reaction measurement protocol defined before seeing the outcome
Choose the pair, provider, price field and timestamp rule. For example, specify windows ending 30 seconds, 5 minutes and 60 minutes after the event — these are measurement choices, rather than guaranteed market phases. Define the pre-release baseline: the last available quote or the average over a specified interval. Do not later select the local low that produces the largest move. Disclose resolution, missing ticks, time zone and clock delay. The same price labelled with receipt time or source time may fall in different windows.
Calculate the point move relative to the baseline. In a hypothetical EUR/USD mid-price move from 1.10000 to 1.10110, the change is +0.00110 USD per EUR, or +11 pips with pip=0.0001, and +0.10%. Do not automatically transfer that pip definition to USD/JPY. If you have only one-minute candles, do not claim what happened after 100 milliseconds. For a narrower window, record “insufficient resolution”.
Finally, distinguish the result from its cause. The observed move contains everything that happened within the window. Even several instruments moving consistently with a rate hypothesis demonstrate consistency rather than full control for other news. Comparisons with non-release days or other events require a sample specified in advance; one chart does not establish an empirical rule.
4. What changes after a release, and what does not
Two things change around a release, but in different ways — confusing them is a common source of poor conclusions.
Execution — how execution costs affect position size, how to choose an order type and assess a broker’s execution model — is covered in separate articles: trading around releases in the strategies section, position sizing in the risk section and execution quality in the broker section. Here, keep quoting conditions separate from the report’s content: a wider spread changes the visible trading cost, but does not explain the economic reading.
You can scroll the diagram horizontally or open it at full size.
The spread diagram is a hypothetical example of spread widening followed by narrowing, without statistically established timings. It is not a result from Chaboud et al. Their study examined prices, volume and volatility on EBS in 1999–2004; it did not estimate the broker spread curve shown here.[2] Check the actual path against your own source’s bid/ask data, including for readings matching the survey.
5. How to read the reaction — and its absence
Use the same observation windows for each release: first record the initial change, then compare it with later prices and rate instruments. The windows organise measurement; they do not reveal when particular traders read a component or why they traded. Article 6.3 explains the indicators and report construction.
Window 1 (0–30 seconds): the initial measured change
The first window captures the move from publication onwards. Its speed may exceed an ordinary chart’s resolution, and quoting conditions may change. All report fields are already available; their processing times differ. Research confirms rapid price adjustments in specific samples, but does not prove that every first impulse overshoots or that every subsequent reversal is caused by report details.[3] The 0–30 second, 30 second–5 minute and 5–60 minute intervals used here are chosen observation windows, rather than universal market phases.
Record a move before the official release time as an observation rather than proof of a leak. The Department of Labor permanently ended its press lock-up from 3 June 2020; the Federal Register of 27 May gives the rationale and scope.[11] This establishes neither identical access delays for every recipient nor a single cause of an earlier move. Research on other procedures and instruments does not automatically apply to today’s BLS releases. Without a time source, price series and examination of alternatives, the explanation remains a hypothesis.
You can scroll the diagram horizontally or open it at full size.
Window 2 (30 seconds–5 minutes): compare prices with the package
In the second chosen window, compare the full report with instrument reactions. Check whether including earnings and revisions changes the data interpretation. Even a consistent package does not guarantee the exchange rate’s direction. A candle does not reveal participants’ identities or intentions; additional trade and order data would be needed. You can, however, record how much of the first move remains, with a timestamp and price definition.
Observations consistent with the first move persisting — a description, without a forecast:
- The move persists after the first minute without a clear retracement. No percentage threshold inherently separates a “shallow correction” from a “reversal”; a threshold given without research on a disclosed sample is an unsupported number. The observation is qualitative: a shallow, brief retracement means something different from giving back most of the impulse.
- An activity indicator rises. Check whether it is tick volume, exchange turnover or another provider measure. The number of quote updates is neither global OTC volume nor evidence that a large participant is buying the currency.
- The relevant country’s two-year bond yield moves in the same direction as its currency. For a currency pair, however, the relative change in rate pricing in both countries matters, rather than a yield movement on one side alone. Record both changes in basis points and their observation times. Other participants’ ranking of signals cannot be inferred from a chart and should not be asserted.
- Other pairs containing the currency move in the same direction. Be careful with the dollar index as “confirmation” of EUR/USD: the euro has a dominant weight, so that pair itself moves the index and the comparison counts the same information twice. Pairs containing a third currency and interest rate instrument reactions help narrow the source of the move, but do not establish it on their own.
Observations that weaken the earlier interpretation — also without establishing the cause:
- The exchange rate gives back a substantial part of the first impulse within the first minutes and does not regain it after the whole package has been read.
- The exchange rate moves while the chosen rate instrument barely changes. This weakens an explanation relying solely on that channel, but does not prove that the market read only the headline. Other channels, different liquidity in the instruments and an overly broad window are possible explanations.
- Report components point in the opposite direction to the headline, and rate instruments react consistently with the components rather than the headline.
Window 3 (5–60 minutes): what remains of the initial reaction
In the third window, compare the exchange rate change with the initial move and rate pricing. Quoting conditions may return to earlier levels, but this must be measured rather than assumed. A sustained move accompanied by a changed implied rate is consistent with one transmission mechanism. A price reversal with an unchanged rate is also an observation; it does not by itself prove that report details “invalidated” the headline. Check other events and keep the outcome separate from its interpretation.
Spread: visible cost and quoting conditions
You can observe the spread without reading the report. A narrower bid/ask spread means a lower visible cost of immediate trading in that source, but does not by itself reveal available volume or execution certainty. Define “return” beforehand: baseline, tolerance, number of consecutive quotes and window. Comparisons with earlier events require the same pair, provider and measurement method. We do not assign one normal duration to all FX markets.
Rapid spread narrowing does not prove the exchange rate interpretation is correct. Persistent widening likewise does not identify a cause: volatility, time of day, quoted volume changes and other news may matter. Distinguishing these mechanisms requires additional measurements. This lesson records quoting conditions; it does not derive an entry indicator from the timing of their changes.
Revisions — why they can matter more than the current reading
Updated values for earlier months are normally released alongside the new reading. They are not a technical footnote: a revision changes the picture that the current reading merely adds to. The comparison must be numerical to show which part of the release contributes more information.
• Current-month surprise: +20,000.
• Previous-month revisions: −60,000.
• Final employment-level difference: 20 − 60 = −40 thousand versus the pre-release expected path based on the 175,000 median and then-known data.
In this model, the published path is 195 − 60 = 135 thousand jobs above the previously known level, but 195 − 60 − 175 = −40 thousand below the pre-release expected path. Both comparisons use a common starting point and revisions to the same summed path.
The headline is “above expectations”, while the employment path incorporating the reading and revisions is 40,000 below that constructed expectation. This does not mean the known employment level itself fell by 40,000 or that “the whole release was weak”: earnings growth, unemployment and other components are outside this calculation. Revisions and the current reading must concern the same quantity and unit; adding a surprise in thousands of jobs to a percentage revision is meaningless. Values are illustrative; actual releases require the statistical office’s announcement.
Real release: what was known about July 2024 payrolls?
A later download can give the right month with the wrong information set. Follow one seasonally adjusted CES estimate through four archived BLS releases. The figures below are monthly changes in thousands of nonfarm payroll jobs, not numbers of people in the household survey.
- Before 2 August: the 5 July release put June’s increase at +206 thousand and May’s at +218 thousand. July’s payroll reading was not yet published.
- 2 August, 08:30 ET: the July report first estimated July at +114 thousand. It also revised May to +216 and June to +179 thousand: a combined −29 thousand for those two earlier monthly changes. The same package reported unemployment at 4.3% (CPS), hourly earnings growth of 0.2% month on month and 3.6% year on year (private payrolls, CES). These are separate fields, not a single score.
- 6 September, 08:30 ET: July was revised to +89 thousand. That −25 thousand adjustment was new in September; it was unavailable for an August decision.
- 4 October, 08:30 ET: the next release raised July to +144 thousand, a +55 thousand revision. This completed the two routine monthly revisions, not all possible future benchmark or seasonal revisions.
The times are the archived releases’ embargo times: 08:30 America/New_York (EDT, UTC−4), or 12:30 UTC on all four dates. As a regional display example, the same instant is 14:30 Europe/Warsaw (CEST). They are not measurements of a website’s first response or a trader’s receipt time. The CSV of source values and release dates retains each vintage, series and unit; the source-local and regional timestamps include their UTC offsets, alongside an explicit UTC column.
Stop the clock before reading the answer
- At 08:29 ET on 2 August, which June estimate could the note contain: +206 or +179 thousand?
- At 08:29 ET on 6 September, which published July estimate was available? Calculate the revision announced one minute later.
- Can this evidence establish an August survey surprise, a dollar move or an attainable trade price?
Show the answer and the limits of the record
1. +206 thousand. +179 belongs to the 2 August release. 2. +114 thousand, followed by 89 − 114 = −25 thousand. Using +144 in either earlier note would import October information. 3. No. The bundle preserves neither a contemporaneous forecast survey nor a timestamped FX quote feed. It supports a reconstruction of the published report, with no price reaction assigned.
For the 2 August note, keep three entries: the new July estimate, the revisions to earlier months and the CPS/earnings fields. Do not relabel −29 thousand of revisions as a July surprise, and do not mix these real observations with the hypothetical +195/−60 thousand calculation above.
No reaction as information
A reading that clearly differs from expectations but leaves the exchange rate unchanged is an observation requiring an explanation, rather than an event to omit. Several possible reasons lead to different conclusions:
- The reading was priced earlier. Partial data released in preceding days may have shifted expectations above the survey median — in which case a “survey surprise” differs from a surprise relative to pricing.
- Inconsistent components. The headline may differ from forecasts in one direction while earnings and revisions differ in the other. There is no single sum of the “information balance” until we specify a model, weights and expectations for these fields. This is a hypothesis of an ambiguous package, rather than proof of zero information.
- A channel beyond the nearest rate decision. The data may leave the nearest meeting’s pricing unchanged while still affecting later policy, growth or risk. Check the entire selected horizon. A small contract change does not prove the release is irrelevant to the bank or that other instruments contain no new expectations.
- Market attention is elsewhere. A simultaneous event with greater impact masks the release reaction.
- Earlier positions may matter. Participants’ exposures may affect the reaction. However, they are not fully observable, and confirmation of a forecast can lead to either closing or increasing positions. COT data from several days earlier cannot resolve this for today’s entire FX market. Record it as a hypothesis subject to a data limitation.
- The reading’s credibility is questioned. A one-off spike attributed to temporary factors may be treated as noise expected to disappear in the next release or a revision.
You can test hypotheses against available sources, but cannot resolve all of them. The report does not reveal the full expectations and exposures in OTC markets. The rate curve, components, calendar and statistical office’s commentary help narrow explanations; they do not establish causal identification. Where data are missing, record “cause undetermined” instead of saying the market ignored the release.
“Buy the rumour, sell the fact” — when the phrase has meaning
“Buy the rumour, sell the fact” describes a possible path: a move before an event and a reversal afterwards. Confirmation can reduce uncertainty even when it matches the median. It also does not imply that all existing holders want to close their positions. Testing this narrative requires a record of expectations, prices and alternative events; chart shape alone does not confirm it.
The phrase fails in several recognisable situations. An event that was not fully priced can still convey information when confirmed. Confirmation may arrive with new information, such as a revised projection or revision shifting expectations for subsequent months. A reversal may also be attributed to this mechanism merely because it occurred, although any number of other factors could explain it. The test is the same throughout this article: could you show before the event that it was priced, and did the priced rate path change afterwards?
Two releases at the same time
When data from two countries arrive at similar times, attribution requires separating the currency pair’s sides. An exchange rate is a relative price, so its reaction does not identify which side moved.
Compare pairs containing a third currency and both countries’ rate instruments in the same window. Remember the triangular relationship: EUR/PLN is linked to EUR/USD and USD/PLN. Three rates are not three independent pieces of evidence. A common global factor can move several pairs simultaneously. This review narrows hypotheses, but does not prove which side produced a single impulse.
A contradictory package — when a release remains inconclusive
A textbook example assumes every part of a report points the same way. In practice, they often differ: the headline exceeds the forecast range, a component considered leading falls below it, and earlier months’ revisions point elsewhere again. This is neither an exceptional nor necessarily temporary situation; it is one possible release outcome.
An inconsistent package may remain inconclusive. Record which components differ from expectations and the changes observed in a specified window. An unchanged rate curve does not mean that “the market also found no information”; it describes a particular price and chosen channel.
Two qualifications apply. First, saying “the market cannot decide” attributes an unobservable mental state to it; what is observable is only a narrow exchange rate range and unchanged rate pricing. Second, quantifying those moves requires a quote series with a named provider and timestamps. Without it, the description remains qualitative — more honest than figures recalled from memory.
A consistent report may coincide with currency weakness despite a higher implied rate. This pattern does not fit a simple explanation based on interest rates alone. Growth, risk premia and other news are possible hypotheses; without additional data, do not declare one the winner. An exchange rate is a relative price, so also examine the other side of the pair.
What a candle cannot show
A hypothetical candle has O 100, H 110, L 90 and C 105. Both 100→110→90→105 and 100→90→110→105 fit it, with the extremes occurring in the opposite order. OHLC does not show when a quote update occurred, the bid/ask spread, available volume or whether a particular order would have been filled.
Attach the candle screenshot to your journal as a description of price in a selected source. Execution costs require bid/ask quotes, timestamps and — for actual execution analysis — an order confirmation. Mid, bid, ask and transaction prices may generate different candles. Name the field first, then compare windows.
6. The scenario journal
After a release, it is easy to fit your memory of earlier expectations to the result. A timestamped record lets you compare what you actually wrote with the subsequent reading. Preserved archives, explicit criteria and consistent event selection further limit the bias. We do not claim that one entry removes it entirely or that a particular illusion appears after a fixed number of releases.
A scenario journal differs from a trading journal: it contains no entries, trade results or profit calculations. It should answer one question: did what I considered likely before publication relate to what happened?
| Field | Before: closed record | After: separate addition |
|---|---|---|
| Event | Hypothetical year-on-year CPI, NSA; 12 March 2026, 08:30 America/New_York (EDT) = 12:30 UTC | This is an exercise date, rather than a claim about the actual CPI schedule |
| Note timestamp | 2026-03-12T12:20:00Z | 2026-03-12T12:35:00Z |
| Forecasts | Five model forecasts: 2.9; 3.0; 3.1; 3.2; 3.3%. Median 3.1%; source=exercise | Model reading 3.4%: surprise +0.3 percentage points |
| Report vintage | Preserved previous value and document link | First reading; revisions added separately |
| Rate instrument | Code, expiry, price, model and measurement time | Same instrument; change in bp and, separately, probability in percentage points |
| FX price | Provider, pair, bid/ask or mid, time zone | Comparison in the declared window; missing data=no measurement |
| Hypotheses | What changes the interpretation; competing events | Observation, then possible interpretation, then what remains undetermined |
Regional display example: in Europe/Warsaw on this hypothetical date, the pre-release note is 13:20 CET, release 13:30 CET and added observation 13:35 CET. These are the same instants as 12:20, 12:30 and 12:35 UTC, not additional observations.
Where to obtain material for the pre-release section: institutions maintain their own schedules — in Poland, the Statistics Poland publication calendar and Monetary Policy Council meeting schedule;[6][7] in the United States, statistical agencies’ calendars; in the euro area, Eurostat and European Central Bank calendars. News agencies and bank research departments publish medians and forecast dispersion. The priced rate path is inferred from the instruments described in article 6.2. The tools section reviews calendars and services bringing these data together.
7. Common mistakes when reading a release
These mistakes concern reading a release rather than trading it. Each leads to a conclusion that looks like analysis but results from omitting information available at publication.
- No recorded benchmark. Without the survey median, forecast dispersion and current rate path pricing, there is nothing against which to compare the reading. “The data were strong” then describes a headline impression rather than a difference from expectations — the difference that analysis concerns.
- Using a stale forecast survey. A survey closed several days earlier may omit information that arrived before the release. Preserve its closing time and any newer forecasts separately. Rate instrument prices can help assess a channel, but do not reveal the exact payroll or inflation reading that every participant expected.
- Treating an unchanged rate as no information. Data can change the economic picture while the nearest expected rate remains similar. Record the report’s content and the selected instrument’s lack of reaction separately. Do not replace them with a “calendar event” label.
- Drawing a conclusion from the headline alone. In a hypothetical package, +210,000 jobs coexist with a −90,000 revision to the previous month. Both numbers belong to the same release. Without all fields and forecasts, do not label the whole report strong or weak; neither do we assume a fixed time by which recipients will have read the details.
- Fitting the explanation after the move. A detailed report lets you choose a field matching the subsequent exchange rate and omit others. A pre-release record limits that freedom. Fixed windows, the full set of fields and competing-event analysis also help. Where data are inconclusive, not every price change needs an established explanation.
- Attributing a move to whichever release happened nearest. When two countries publish data on the same day, a pair’s reaction does not identify which side supplied the information. Comparisons with pairs containing a third currency and changes in rate path pricing narrow the source, but portfolio flows or a common risk factor may also be acting.
- No pre-release record. After a dozen or more entries, compare your recorded hypotheses with outcomes. We do not assume what proportion will prove correct. The journal helps check whether memory omits unsuccessful interpretations and whether conditions for changing them were applied.
8. Regional case study: Poland’s MPC and Statistics Poland
For Polish releases, begin with the NBP or Statistics Poland calendar, the available forecast panel and the particular instrument used to observe rate expectations. A zloty pair contains two currencies: record domestic news and events affecting the other currency within the same window.
Start with the calendar. The table sets out information available from institutional sources: who publishes, when and for which period. It includes no “typical move” or “typical spread” columns: these are statistical claims requiring a quote series from a named provider, a defined measurement window and an event count. Without these, the figures would be unsupported.
| Release | Schedule source | What to check for each event |
|---|---|---|
| MPC decision and post-meeting statement | NBP: MPC decisions | Date and availability time separately; the annual schedule guarantees no fixed time |
| NBP President’s press conference | NBP conference announcement | Conference date and time, separately from the statement |
| CPI, GDP, output, retail sales | Statistics Poland: publication calendar | Full name, period, flash/full vintage and time assigned to the entry |
| Balance of payments | NBP: balance of payments statistics | Month, scope, announcement time and revision status |
The decision and press conference are separate information events
Record the rate decision, post-meeting statement and NBP conference as separate events. The conference often takes place the next day, but NBP announces its date and time separately; do not assume a fixed 24-hour interval. Communication can add information even when the rate is unchanged. Studying a specific move requires actual timestamps for all these events, rather than only the meeting date.
Domestic data against the global backdrop
The zloty responds to global risk appetite independently of domestic data. The practical consequence is the same as with simultaneous releases: EUR/PLN alone does not tell you whether the zloty or euro moved. Comparing another zloty pair and EUR/USD within the same window helps. A favourable domestic reading coinciding with weaker global sentiment may not strengthen the zloty. Examine both channels separately; a lack of appreciation alone does not prove the domestic reading was irrelevant.
For a Polish rate derivative, record its reference rate and contract specification. The WIBOR-to-POLSTR transition makes this essential: comparing prices before and after a benchmark change can join different instruments. Article 6.2 gives the transition timetable and source. Do not infer identical exposure from a similar product name.
Supporting questions
What should I do when two releases arrive at the same time?
Preserve both reports’ timestamps and compare several related exchange rates and rate instruments. This helps test the explanation’s consistency, but pairs connected by a price triangle are not independent confirmations. Without high-frequency data and controls for other information, do not attribute the move to one release.
Why can the same publisher-local hour appear at different times in my zone?
Eastern Time follows America/New_York. The same 08:30 local hour is 13:30 UTC under EST (UTC−5) or 12:30 UTC under EDT (UTC−4). In 2026, New York changes clocks on 8 March and 1 November. Your selected IANA zone can follow a different calendar or no DST at all, so convert the exact date through UTC. Retain the publisher’s date, source zone and offset, the UTC instant and the selected display zone. The Europe/Warsaw regional case shows the separate US–European mismatch; it is not a default location for the reader.
FAQ — frequently asked questions
How long before a release should I prepare?
Preparation time depends on source availability. Evaluating a forecast requires preserving its content and pre-release timestamp. A note added afterwards is still a valuable observation, but does not prove earlier knowledge. Label both stages: ex ante — forecast and conditions; ex post — reading, prices and interpretation with its limits.
How does consensus differ from instrument pricing?
Consensus summarises a named provider’s survey, often as a median. Instrument pricing is a price at a particular moment, converted into an implied rate or probability by a model. Do not equate them. Without additional assumptions, they cannot reconstruct a single macro forecast held by the whole market. Preserve the survey source, timestamp and selected contract price separately.
How should I read revisions to previous readings?
In a hypothetical example, jobs exceed the forecast by 20,000, while revisions to the previous two increases total −60,000. If the forecast current increase was to be added to the old employment level, the new published level is 40,000 below that benchmark: +20 −60=−40. This calculates the final level difference from a common starting point, rather than an automatic “NFP quality indicator”. Do not similarly add percentage growth rates, revisions to different series or a benchmark revision without checking the full levels.
Why do good data sometimes weaken a currency?
A strong report may coincide with an opposite exchange rate movement. Changes in expected policy, growth assessments and risk premia are separate channels; the other currency in the pair also matters. A higher rate alongside a weaker currency does not alone prove that the growth channel dominated. Check hypotheses within the recorded window and leave the cause undetermined if the evidence is inconclusive.
What does no reaction to a significant reading mean?
No movement is an observation within a specified window under a particular price definition. It may reflect earlier expectations, report details or other events. Test these explanations separately; full positioning and expectations data are unavailable. Do not equate an unchanged rate instrument with no economic information.
How can I check afterwards whether my scenario was accurate?
Compare the closed pre-release record with the reading, revisions and prices within the specified window. Assess the data forecast and the transmission-channel hypothesis separately. Also check the condition for changing your interpretation. A dozen or more entries may reveal omitted fields or an inconsistent process, but form a small sample dependent on event selection. They establish neither a stable accuracy rate nor which releases are “worth” analysing.
Sources
Main sources
- Martin Evans and Richard Lyons, “How Is Macro News Transmitted to Exchange Rates?”, Journal of Financial Economics, Vol. 88, No. 1, 2008 — analysis of macroeconomic news transmission to exchange rates, including order flow and information asymmetry.
- Alain Chaboud et al., IFDP 823, 2004 — EBS prices, volume and volatility, EUR/USD and USD/JPY, January 1999–January 2004. The abstract and data and results sections were reviewed. The paper is not the source of the broker spread curve drawn here. Source document.
- Torben Andersen, Tim Bollerslev, Francis Diebold and Clara Vega, “Real-Time Price Discovery in Global Stock, Bond and Foreign Exchange Markets”, Journal of International Economics, Vol. 73, 2007 — market reactions to macroeconomic data at second and minute frequencies. Source document.
- BIS, OTC foreign exchange turnover in April 2025, preliminary results of 30 September 2025 — OTC statistical coverage and the distinction between turnover and positions. It is not a source of global volume observable in a retail candle. Source document.
Supporting data and reports
- CME, FedWatch methodology — probabilities implied by 30-Day Fed Funds futures under specified assumptions. The methodology was reviewed; archived probabilities and macro consensus forecasts were not retrieved. Source document.
- NBP, “MPC Meeting Schedule”, annual publication — meeting dates of Poland’s Monetary Policy Council. The schedule gives dates; statement and press conference times are not guaranteed and require current NBP announcements.
- Statistics Poland, “Publication Calendar”, current data — schedule of Polish macroeconomic releases, including flash CPI, GDP and industrial production.
- Carlo Rosa, “The High-Frequency Response of Exchange Rates to Monetary Policy Actions and Statements”, Journal of Banking & Finance, Vol. 35, No. 2, 2011 — empirical analysis of exchange rate reactions to monetary policy decisions and statements at minute frequencies, separating the decision’s impact from accompanying communication.
- Gennaro Bernile, Jianfeng Hu and Yuehua Tang, “Can information be locked up? Informed trading ahead of macro-news announcements”, Journal of Financial Economics 121(3), September 2016, pp. 496–520, DOI 10.1016/j.jfineco.2015.09.012 — study of buyer- versus seller-initiated E-mini S&P 500 volume during embargoes before scheduled FOMC announcements. The authors found results consistent with trading informed by the subsequent policy surprise, but no such evidence before the FOMC embargo began or during NFP, PPI and GDP lock-ups in the sample. This is not a measurement of FX deviations from “fair value”. The official abstract and metadata at the University of Miami and the publisher preview were reviewed; access to the full paper is not claimed.
- Global Foreign Exchange Committee, “Execution Principles Working Group — Report on Last Look”, August 2021 — voluntary standards on last-look windows, client disclosures and limiting abuse; a basis for FX Global Code disclosures. Source document.
- U.S. Department of Labor, “Announcing Discontinuation of the DOL Lock-Up Facility for Participating News Media Organizations With Pre-Release Access to Statistical Information”, Federal Register, 27 May 2020, and the Bureau of Labor Statistics announcement ending the procedure effective 3 June 2020 — sources for the removal of early press access to Department of Labor data. Source document.