Services Production index (base 2015)

Paru le :Paru le21/07/2024

Traitement statistique

Source data

The main source data are the monthly tax returns of VAT-registered businesses. The volume indices also use price indices.

Frequency of data collection


Data collection

The monthly VAT returns are provided by Dgfip to INSEE according to the terms defined in the agreement signed between the two parties.

Data collection period


Survey unit

Legal unit

Sampling method

not applicable

Sample size

not applicable

Data validation

The data are validated before being sent to Eurostat. This validation is done in two stages: the checking (and if necessary correction) of individual data and then the checking of the aggregated indices. This second check may lead to further corrections of the individual data.

Data compilation

The elementary (sub-class S) value index of the month is calculated by applying the change in turnover (TO) of the companies between the month m and the same month one year bebore to the index value of the same month one year before: I(m) = I(m-12) * TO(m, S)/TO(m-12, S), S is the sector of legal units (cf. documentation on methodology).

Production indices (volume) are calculated as the ratio of the elementary turnover index in value to an associated price deflator. Aggregated indices are calculated as Laspeyres indices with a two-weighting system (cf. methodology).

Seasonal adjustment

The raw indices are seasonnaly and working-day adjusted (SA-WDA) using the X13 ARIMA program available in JDemetra +. The WD adjustment (trading days, leap year) and the seasonal adjustment decomposition are calculated at the 4-digit level of the NACE Rev. 2. The upper levels are obtained by aggregating the series (indirect method), in the same way as the agregation of raw data.

The Reg ARIMA calendar adjustment is used by constructing working day regressors based on the French national calendar (which takes into account working days specific to France).

Outliers (additive outliers, temporary changes, level shifts, seasonal outliers) are fixed in the past and are detected automatically ont the past 12 months onwards. The critical value for outlier detection, the filter length and the model/filter selection depend on the series and may have to be changed manually to improve the quality of the seasonal correction. This was the case to neutralize some particular points associated with the 2020-2021 health crisis (lockdowns for example), which would have induced an unjustified distortion of the seasonal coefficients over the past.

Either additive or multiplicative decomposition can be used. The seasonal adjustment models are reexamined every year (favouring stability) and the parameters are re-estimated every month.

Each month the SA-WDA data are revised from 2012. For the seasonal adjustment of indices in the recent past, the models are now estimated over a reduced sub-period (from 2005 onwards), in accordance with Eurostat guidelines, and in order to reinforce the robustness of the seasonal adjustment. The data before 2012 are fixed in evolution, in accordance with Eurostat's guidelines (avoid revisions over a too long period).