Methodology
NORD publishes independent research on Former Soviet Union and Middle East–North Africa energy markets. This page states how that research is built, how its claims are checked afterwards, and how the indicators are constructed. It is the reference for every other page on this site.
NORD is not a Price Reporting Agency and does not publish prices. What appears here are analytical constructs and research judgements, documented so that a reader who disagrees can see where the disagreement lies.
Sources
Every factual claim carries a source. A source here means five things: a named organisation, a specific publication, a locator, an access date, and a direct quotation of the passage relied upon. The locator is a URL for anything on the web, and for a document supplied on request it is that document's reference and a page number. Anything short of five elements does not enter a published analysis. Where an item matters analytically but cannot be sourced to that level, it is held back until it can be, and the gap is noted in the text.
A figure that is calculated rather than quoted is published as a calculation: the inputs it derives from and the operation applied to them are recorded alongside it. A source is not credited with a number it did not state.
Where sources conflict, primary data from system operators and official statistical offices takes precedence over aggregated third-party reporting. Discrepancies material to a published figure are documented in the publication, with both values shown.
No confidential trading data, non-public contract terms or proprietary broker feeds are used. That limits what NORD can say. It also means any competent reader can check the work, which matters more.
Accuracy tracking
Research notes make dated, falsifiable claims and are then rarely revisited. The incentive runs one way, and a reader has no way to know whether an analyst has been right before. Every NORD note is therefore scored against observed outcomes on a published schedule, using a fixed vocabulary.
A prediction is recorded as confirmed when the claim held, confirmed and exceeded when it held and understated the magnitude of what followed, partial when the direction was right and the magnitude was not, understated when the mechanism identified was correct but its weight was materially underestimated, incorrect when the claim failed, and too early when the period has not yet produced evidence either way. The vocabulary is fixed; new categories are not invented to accommodate an awkward result.
Direction and magnitude are tracked separately. A prediction can be right about where a variable moves and wrong about how far, and collapsing the two into correct-or-incorrect destroys the information that matters for the next note.
Both sides of a scored line carry a source. The observation cites its publication; the prediction is quoted from the note being assessed, with a page number. An assessment that cannot quote the claim it is scoring is scoring its own paraphrase of it.
An observation must postdate the claim it tests. A reading taken before a note published cannot be evidence for what that note predicted, however apt it looks.
A prediction and a threshold are different objects. A prediction says where a variable will go; a threshold says what a reading means once it arrives. A threshold can confirm or falsify a prediction and is never itself scored as one.
What a note failed to anticipate is published alongside what it got right. Unanticipated events are listed with their own sources and an explicit statement of how they change the framework the note proposed. Defects found in the note being assessed are published in the same way, whether or not they touch a scored prediction.
These rules are enforced by a program, not by care. It reads the structured file an assessment is built from and refuses any figure it cannot trace to a quotation, any prediction that does not cite its note by page, any value a record asserts that its own quotation does not contain, and any passage the published text puts in quotation marks that no stored source holds. The source appendix will not build against a file that fails it, because an appendix asserts a standard and cannot be generated from data that does not meet it.
Assessments carry an aggregate score, which is the least informative number in them. The weakest calls, and what those calls changed in the method, are the substance.
The NORD Med-TTF Basis Indicator
The indicator is a monthly estimate of the price differential between Southern European gas hubs and the Dutch TTF benchmark, weighted by Algerian supply exposure. It is an analytical construct for research and structural analysis. No transaction should be priced or settled on it.
Reference prices are day-ahead, not front-month. A basis measured on front-month contracts embeds the relative shape of four forward curves, with their differing storage expectations, contract structures and risk premia, on top of the physical premium. That is a different object from the structural basis this indicator measures. Day-ahead prices the gas that is actually moving between these hubs. It also happens to be public, so the reproducibility claimed on this page is available to any reader, not only to one holding a market data licence.
Weights are fixed and announced in advance. Destination weights are computed once per publication year from Algerian supply volumes over the twelve months ending the previous November, and published before the year they apply to. They do not move within the year. A weight derived from contemporaneous flow would fall precisely when a supply disruption made that destination's price signal most informative, damping the event the indicator exists to detect.
Two adjustments are applied. The transport adjustment is the weighted cost of moving gas from TTF to the southern hubs, taken from published tariffs and carried with its sign, so that negative capacity auction clearing prices pass through unchanged instead of being floored at zero. The seasonal adjustment removes the cyclical demand component. It is estimated against weighted heating degree-days, with an explicit term for the 2021–2023 supply disruption, and not against the price history: a calendar-month average fitted across those years would encode the crisis as seasonality and strip it out of the structural basis.
Values are published with an uncertainty band. The adjustments are estimated, the weighting is a methodological choice, and one input arrives late. The published value carries a standard error propagated from all three, and is rounded to one decimal place.
Flash and final. A provisional value is published monthly, built on venue day-ahead data, and revised to final when the official monthly hub series for that month is released. Each published value states which it is. Where a month's degree-days are a climatological normal instead of a realised observation, the value says so and the band widens.
Attribution is estimated, not assumed. The residual left after the adjustments is not the Algerian component. Everything the adjustments fail to capture lands in it, including LNG arbitrage, alternative Mediterranean supply, storage dynamics and measurement error. The Algerian contribution is therefore estimated explicitly, with a confidence interval and a stated share of the indicator's variance, and published as a separate series from the basis.
The full specification of that estimate, covering regressors, standard-error treatment and the variance decomposition, publishes alongside the first indicator value, once it has been calibrated on observed data instead of a test panel. A specification that has not yet met real data is still a draft, and publishing it as settled would misrepresent it. Nothing in the basis construction above waits on this: the indicator is fully specified on this page and reproducible today from public sources.
Revisions and versioning
Published values are revised when input data is corrected, when a system operator restates, or when a computational error is found. Revisions are published with a timestamp and a note stating the previous value, the new value and the cause.
Every published value carries a cryptographic digest of the inputs it was computed from. Two parties can therefore establish whether a disagreement concerns the data or the method before they argue about the number.
Material changes to a calculation method, meaning weighting, adjustments or source hierarchy, are published as a new methodology version with a changelog. Clarifications and corrections are issued as point releases.
Independence
NORD's published research is never commissioned, sponsored or funded by any party with an interest in its conclusions. Commissioned work undertaken for clients is delivered privately and is never published as NORD research, and where a client relationship bears on a subject NORD publishes on, the relationship is disclosed at the point of publication. NORD holds no equity, trading position or financial interest in any entity it writes about.
Feedback
Corrections, methodological objections and source challenges are welcome and are answered. Substantive methodology feedback is documented and, where accepted, incorporated into the next version with credit to the person who raised it.
Versions
This page is version 1.2, effective 14 September 2026. It is the first version published on this site. Versions 1.0 and 1.1 were internal drafts, logged here because the method changed materially between them and a reader is entitled to see what changed before the first value is computed.
1.2, 14 September 2026. Three rules added to accuracy tracking after the first assessment was run: a prediction is quoted from its note with a page number, an observation must postdate the claim it tests, and a threshold is never scored as a forecast. The source standard now names a locator rather than a URL, so a document supplied on request can be cited by reference and page. Derived figures are published as calculations. The enforcement program is described. Each of these came from an error the first assessment made and caught before publishing.
The independence clause was written for this version alongside the Advisory page. It separates published research, which is never commissioned, from client work, which is never published as NORD research, and commits to disclosing a client relationship that bears on a published subject.
1.1, 11 September 2026. Superseded version 1.0, which was drafted in September 2026 and never published. Five things changed in the indicator specification. Reference prices moved from front-month to day-ahead settlements, because a front-month basis measures forward curve shape alongside the physical premium, and because day-ahead data is public and the reproducibility claim depends on that. Destination weights became an annual vector announced in advance, replacing a rolling average of realised flow that fell just as a supply disruption made the affected hub most informative. The seasonal adjustment is now estimated against weighted heating degree-days with an explicit term for the 2021-2023 disruption, replacing a rolling calendar-month average that absorbed crisis level into seasonality. Published values carry an uncertainty band, and one decimal place instead of two. The Algerian contribution is estimated explicitly instead of being read off the post-adjustment residual.
Four of those five changes alter the computed number. Version 1.0 never left draft and no value was ever published under it, so no series requires restatement. Superseded drafts are retained and available on request.