A macro backtest is only meaningful when its inputs match the information that was available before each decision. Today's revised history can be useful for studying the economy while being unsuitable as a substitute for the original release. Begin with the information set, then define the strategy, costs and benchmark before evaluating performance.
This lesson is for FX traders learning a repeatable research process. The companion exercise follows five checks; each check should leave a short evidence note.
Freeze the research question
Specify the indicator, exact instrument, currency-pair quotation, signal, decision horizon and exit rule before inspecting performance. Write down how ambiguous or missing observations will be handled. A spot reference series does not automatically represent executable broker prices or a futures contract. The instrument and data must match the question, and a later rule change creates a new research version rather than improving an untouched original test.
Use the correct vintage
Distinguish the first release from later revisions. Retain the value, period, source and evidence of historical availability for every input. For a forecast-based signal, also retain the comparable forecast captured before publication. If that evidence is missing, leave the event ineligible for the relevant test. Replacing it with the current latest value or a forecast generated afterward introduces information the historical decision could not use.
Work through the stored example
In the stored example, Monthly payroll change is 133 for 2026-08-31 and 29 for 2026-09-30, measured in thousands of jobs. These are reference-period observations. The chart reports the monthly change, rather than the total employment level. Before using a point in a historical signal, open its release and revision record. The plotted latest value alone cannot establish what was known at the earlier decision date.
Model when a decision was possible
The reference period is not the publication date. A planned release time is not evidence of actual publication, and a database capture clock does not automatically establish the publisher's first availability. Use the available evidence to define when the signal could have been acted on, then add a realistic execution delay. If suitable timing evidence or price resolution is absent, choose a research question the data can support.
Include costs and financing
Define bid-ask spread, slippage, fees and financing consistently with the chosen instrument. Do not add policy-rate differences to spot returns and call the result realised carry. Vary plausible costs and delays to see whether the conclusion depends on optimistic execution. A result that disappears under small changes is still useful evidence, but it should not be presented as a robust trading edge.
Compare with passive exposure
Include a matched buy-and-hold benchmark for the same instrument, dates, capital basis and cost conventions. For FX, state exactly what long or short passive exposure means and distinguish spot price return from financed total return. Also examine exposure and drawdown so that a strategy's apparent advantage is not merely a different amount of time at risk. A benchmark clarifies the claim instead of automatically validating it.
Preserve an honest evaluation
Keep a separate evaluation sample, record every explored variant and retain the trade ledger and reproducibility inputs. Report returns, drawdown, exposure and cost sensitivity together; any performance claim needs the full appropriate research report. The lesson's stored chart illustrates the difference between observation history and historical information availability. It does not report a completed strategy test, estimated profitability or evidence that the illustrated rules outperform passive exposure.
Write the information contract for the strategy
Before evaluating a rule, define the information it needs at every decision. Name the indicator, transformation, vintage, expectation benchmark if any and the point at which the input is eligible. Then define the exact instrument and quotation direction. A macroeconomic thesis can be sensible while the available data is insufficient to evaluate the proposed execution horizon.
Separate the economic hypothesis from the trading rule. The hypothesis might concern how a policy-relevant surprise changes the relative outlook. The rule must specify how that information becomes a position, when the position can be entered and how it exits. Freeze those choices before looking at the evaluation sample. Otherwise, a sequence of economically plausible adjustments can become a search for whichever implementation happens to fit the historical outcome.
Reconstruct the historical information set
Today's latest series can contain revisions and later corrections. For a point-in-time test, retain the estimate that was available before the decision, together with evidence of its availability. A row's reference month is not enough. For forecast-based signals, also retain the expectation targeted to the same definition and period, recorded before the release.
Document what happens when the required evidence is missing. An event may remain eligible for a descriptive latest-vintage chart while being ineligible for a strategy that needs the first release or a pre-release forecast. Keeping those uses separate preserves useful observations without pretending they support every research question. Report exclusions and their reasons. The final sample should be explainable from the information contract rather than from a preference for events with attractive-looking outcomes.
Match the price evidence to the proposed implementation
Identify whether the price series represents a daily official reference, an intraday observation, a tradable quote or an executed transaction. Those sources support different questions. A daily reference series can be useful for a longer-horizon study, but it cannot establish the spread, slippage or opening-second fill at a particular broker.
Specify the execution delay and the price-selection method before evaluating the rule. Preserve gaps and simultaneous events, and avoid filling an unavailable execution price with a convenient later mark unless the method explicitly requires that choice. If the available resolution does not match the strategy horizon, revise the research question or obtain appropriate evidence. More precise-looking timestamps do not by themselves create a more realistic execution model.
Build costs and passive exposure into the design
Choose costs and financing conventions that match the instrument. A spot price return is different from financed total return. A policy-rate differential is not a universal substitute for broker financing or a forward-market implementation. Write the spread, fees, slippage and holding-cost assumptions so they can be varied in a sensitivity check.
Define a matched buy-and-hold benchmark over the same dates, instrument and capital basis. State the direction of the passive FX exposure. Compare exposure and drawdown alongside return so a difference in time at risk is not mistaken for a complete improvement. The benchmark is part of the question the test is designed to answer. It should be specified before results are selected, rather than introduced only if it makes the strategy look favourable.
Preserve the experiment and evaluate it honestly
Keep a record of every rule version explored and separate development from evaluation. A rule altered after inspecting the evaluation period is a new experiment. Retain the inputs, transformations, trade ledger and configuration needed to reproduce each reported result. This makes the distinction between exploratory findings and a held-out test visible.
Report performance and risk together, including costs, leverage or exposure assumptions and sensitivity to plausible implementation choices. A few attractive trades or a strong-looking chart do not establish a dependable edge. This lesson explains the research process and does not claim a completed profitable strategy. Its practical output is a clear experiment specification: what the rule could know, how it would act, which benchmark it must beat and what evidence would justify retaining or rejecting the result.
Guided research workshop
Research question: Could the historical rule have used these inputs and prices under the stated implementation?
Use the stored example above and its original source as the starting evidence. Work through the following fields in order, writing a short answer for each. Where a field needs information beyond the chart, identify the report or expectation record to inspect before reaching a conclusion.
Frozen experiment
Write the economic hypothesis and trading rule separately. Specify the instrument, pair quotation, decision horizon, entry, exit and missing-data treatment. Preserve this version before evaluating results. A later change creates a new experiment. This makes it possible to distinguish a defined test from a sequence of plausible modifications selected because they fit the completed historical chart.
Eligible information
For every decision, identify the required indicator vintage and expectation evidence where relevant. Retain the period, source and availability context. A latest-vintage series may remain useful for a descriptive chart while being ineligible for a first-release strategy. The exercise should state exclusions transparently rather than substituting later knowledge when a required historical input is missing.
Executable price question
Match the price source to the implementation horizon. Distinguish daily references, observations, quotes and transactions. Define an execution delay and price-selection rule before inspecting outcomes. An apparent precision in the data timestamp does not establish the spread, fill or slippage a particular account could have achieved. If the resolution is unsuitable, revise the research question instead of inventing an execution price.
Costs and benchmark
Specify spread, fees, slippage and instrument-appropriate financing. Add matched passive exposure over the same dates and capital basis, with the direction of the FX holding made explicit. Include exposure and drawdown in the comparison. This field should make the performance claim meaningful before any result is selected, rather than introducing favourable assumptions only after a strategy looks promising.
Evaluation record
Retain the data, configuration, explored variants and trade ledger. Separate development from evaluation, and report sensitivity to plausible costs and delays. The completed specification should let another researcher reproduce the decisions and understand the limits of the evidence. A negative or unstable result remains useful when the experiment is documented honestly and is not promoted into a dependable edge.
A practical decision reference
| Field or situation | What it means | Research response |
|---|---|---|
| Signal input | Was the relevant vintage available? | Retain period, source and information evidence. |
| Forecast input | Did a matching benchmark predate release? | Exclude unsupported surprise calculations. |
| Entry price | Does the source support the horizon? | Model a stated executable-price rule. |
| Costs | What would the instrument incur? | Test plausible spreads, delays and financing. |
| Benchmark | What passive exposure is matched? | Align dates, capital, costs and direction. |
An example of the finished research note
The following wording illustrates the structure of a completed research note. It is a research-plan example, rather than a statement of the current market outlook. Replace its general descriptions with the observations, sources and information dates from the exercise.
The experiment defines the information available before each decision, the exact instrument and the rules for acting on the signal. Original vintages and forecasts are retained where required, while missing evidence produces stated exclusions. Prices, execution delay, financing and costs match the intended implementation. A passive benchmark is evaluated on the same basis. The research record preserves explored variants and reproducibility inputs so any reported advantage can be assessed alongside risk and sensitivity rather than as an isolated return figure.
Keep the source observations and the original version of the note with the conclusion. At the next update, identify the specific new evidence that changes the interpretation. This gives the review a stable starting point and makes the reasoning understandable to another reader.
Work through the exercise
- Freeze instrument and trading rules. Record the source and the reason for the choice.
- Verify original vintages and forecasts. Record the source and the reason for the choice.
- Align decisions to available information. Record the source and the reason for the choice.
- Include execution costs and financing. Record the source and the reason for the choice.
- Compare a matched passive benchmark. Record the source and the reason for the choice.
Use the relevant histories and dashboards at FXMacroData to repeat the comparison. Continue with the related research guide when the evidence raises a more detailed question.
Sources
Educational material. The examples do not establish a profitable strategy or a recommendation for a particular position.
AI was used for research, writing, creating graphs and editing.