What is backcasting in the NACE Rev 2.1 transition?

Backcasting is the restatement of earlier statistics, originally compiled on NACE Rev. 2, on the NACE Rev. 2.1 basis, so that a historical series can be read without a spurious break at the changeover. Eurostat describes backcasting on its transition page as part of the transition, because for many statistics there is a need for historical time series using the same NACE version, and parallel datasets presenting current statistics broken down by NACE Rev. 2.1 will be published for many domains.

The backcasting manual, published by Eurostat in its 2025 edition, gives the methods for producing those consistent series. Backcasting is a statistical method, not a reclassification of individual businesses.

Why is backcasting needed at all?

A classification change breaks a time series. An activity that sat in one class under NACE Rev. 2 may sit in a different class, or in two classes, under NACE Rev. 2.1. A series that simply switches version at a given reference period will show a jump that reflects the classification change, not a change in the economy. Eurostat states on its transition page that, as there are considerable differences between the two versions, backcasting methods will need to be applied to previous statistics so that they can be presented by NACE Rev. 2.1. The need is therefore not cosmetic. Without backcasting, a reader cannot tell whether a rise or a fall in a category reflects activity, reclassification, or a mix of the two. With backcasting, the earlier periods are reformulated on the new basis, and the comparison across the changeover becomes a comparison of like with like, within the limits of the method used.

What does the backcasting manual cover?

The Eurostat backcasting manual, 2025 edition, provides guidance to producers of official statistics that need to produce consistent historical time series for an updated classification. It gives an overview of methods that can be used to reformulate statistics presented according to the old version in terms of the new version, and all its examples are based on NACE. The manual describes the conditions that affect the choice of method, including the availability of microdata or aggregated data double coded under both versions, the number of reference periods for which double coded data are available, the frequency of the series, the level of detail required, and cost, staffing and timeliness considerations. It also describes the need to reconcile potential breaks in backcasted series. The manual is written for producers of statistics. A reader of statistics does not need to apply its methods, but does need to know that a backcast series is a reformulation, and that its method should be stated in the metadata.

What is double coding, and why does it matter for backcasting?

Double coding means coding the same unit, using the same activity description, under both the old and the new version of the classification, for at least one point in time. The backcasting manual assumes that all units recorded in the Statistical Business Register are double coded in that way for at least one point in time. Double coding provides the bridge between the versions: it shows, for the same units, where the activity sat under each version. Without double coding, a backcast must rely on aggregate shares or other estimation methods, which are less direct. A reader who sees a backcast series should therefore check whether the metadata states that double coding was used, and for which reference periods. A series backcast from aggregate shares is not the same kind of evidence as a series built from double coded microdata, and the metadata should let the reader tell the difference.

Concept	Meaning in this guide
Backcasting	Restating earlier periods on the new classification basis
Double coding	Coding the same unit under both versions, from the same activity description
Break in series	A jump caused by the classification change, not by a change in activity
Restated series	A series reformulated on the new basis, with the method stated in the metadata

The table keeps the four terms distinct. A break in series is a feature to be explained, not a finding. A restated series is a product of a stated method, not a new observation of the past.

How is a backcast series read?

A backcast series is read with its method note. The reader records the NACE version, the reference periods covered, whether the series was backcast or double reported, and any break that remains after backcasting. Where a class split, the reader checks the correspondence table mapping type, because a one to many mapping means the backcast allocation between the new classes depends on the method and the source data, not on the table alone. Where a class merged, the reverse care applies. The explanatory note for each new class states what it includes and excludes, and it remains the record of meaning. A backcast figure should never be quoted without its version and method note, because the same reference period can have two published values, one on each basis, and the values will differ where the classification changed.

What does backcasting not do?

Backcasting does not change the entry held for a specific business, and it does not decide the code for any unit. It restates aggregates. The entry for a specific unit is held in the business register, and it is confirmed with the national statistical institute or the register that holds the record. Backcasting also does not remove every break. The manual discusses reconciliation of breaks, and a published series may still carry a break flag where the method cannot fully bridge the change. A break flag is information, not a fault, and it should be quoted with the figure it qualifies.

Check the dataset metadata for the version, the method, and any break flag, check the correspondence table for the mapping type, and confirm a single entry with the institute or register that holds it.