Chapter 8 - Conduct of a Production Campaign
Each month, INSEE collects all the data needed to compile the index, verifies the data, builds the elementary
series based on the collected data, analyses them and applies adjustments and corrections where necessary,
calculates the indices based on several classifications, seasonally adjusts them and, finally, publishes and
disseminates them to a wide range of users. These complex operations are based on well-established processes
and applications.
1- Operations Management and Data Collection
Historically, the industrial production index has been the responsibility of INSEE. The decree governing the
organisation of INSEE dated 4 August 2016 specifies that the “Short-term statistics” department of the
Directorate of Business Statistics is responsible for compiling activity indices (IPI, turnover indices, etc.) in
industry, trade and services and the various price indices (purchase and sale) collected from enterprises in
industry and services.
The IPI – and, more generally, all related processes, methods and applications – is managed by the “Short-term
business indicators” (in French, Indicateurs conjoncturels d’activité, or ICA) division within this department.
The task of collecting individual production data (dispatch of campaign opening letters, receipt of results,
reminders to late responders, imputation of non-response s…) and of producing the elementary series is handled
at the Normandy Regional Directorate within the National Business Statistics Service (in French, Service des
statistiques nationales d’entreprises, or SSNE) located in Caen for “industry excluding food and agriculture” 44.
Surveys of industrial food and agriculture businesses are managed by the Statistics and Forecasting Service
(SSP) of the Ministry of Agriculture located in Toulouse45, while the collection of energy data is the
responsibility of the Statistical Data and Studies Service (SDES) of the Ministry for the Ecological and Inclusive
Transition, as is the collection of construction data, with collection delegated in this case to the French Building
Federation (FFB) and the French Federation of Public Works (FNTP).
Because of the sheer range of collection bodies and production indicators and the fact that a large number of
product groups are tracked, the process of managing the indicator is a highly complex process.
A customised questionnaire is developed for each enterprise included in the sample. The questionnaire focuses
on the products for which the enterprise was included in the sample. A letter is sent each month to the enterprises
included in the sample to inform them that the questionnaires are available. The vast majority of enterprises
respond online (98% of responses), using the COLTRANE portal dedicated to official business statistics surveys.
If they wish to respond by post, they must make an explicit request. Mail responses mainly come from small
businesses with no Internet access.
2- Legal Basis and Litigation Management
The EMB46 is compulsory. It has been issued with a general interest and statistical quality label. It is included in
the programme of official statistics surveys to which companies must respond under the 1951 Act, which
provides for a dispute resolution procedure in the event of non-response. Companies’ responses are essential for
the quality of the index, and therefore, for short-term diagnostic purposes.
A litigation procedure is set up by the various bodies responsible for overseeing the conduct of surveys to
guarantee a high response rate and thereby ensure that the IPI is of good quality. Thus, for example, for the
EMBs managed by INSEE, the response rate is over 87% (although the rate varies slightly from one month to
the next). In the event of repeated non-responses, the litigation procedure involves serving non-respondents with
a formal notice followed by a notification of non-response with acknowledgement of receipt. The
44Except for certain branches, such as the manufacture of basic iron and steel and of ferro-alloys, where collection is
conducted by partner professional bodies (see Chapter 2).
45As with the “non-food and agriculture” part, data on some branches are collected by partner professional bodies.
46Or, more precisely, the monthly branch surveys since they cover all collection bodies.
56
acknowledgements of receipt are sent to the Litigation Committee at the National Council for Statistical
Information (CNIS), which then sends a letter of referral to the enterprises concerned. A final decision is taken at
a meeting of the Litigation Committee depending on the response provided by the enterprises and their position
with regard to the dispute (repeat offender or not). The decision may result in a fine being imposed on non-
responding enterprises.
3- Data Processing
The role of the unit responsible for calculating the IPI is to collect all the data required to calculate the index, to
produce the aggregated indices (based on different classification levels) using the data collected, to assess them,
to seasonally adjust the indices and, finally, to publish and disseminate them in several formats. It also responds
to internal and external requests for information on the indices.
The integration of the data and the calculation of the indices are mainly based on two integrated applications for
calculation and analysis. These integrated applications enhance the security of the index production process. The
gross indices are recalculated each month from January of the previous year based on current production and
revisions to historical data may be incorporated over a longer period if necessary.
As noted above, the National Business Statistics Service is responsible for the first part of the processing as part
of a monthly campaign to calculate the IPI indices. Sector clerks are responsible for a set of branches within the
scope of the index. Their first duty is to check that enterprises have submitted their responses to the monthly
surveys, which are received either directly (if the survey is managed by INSEE) or indirectly through OPAs.
They are required to ensure that all the results are sent to INSEE within the deadlines set for the production of
the IPI and to follow up with correspondents if necessary.
All declarations of production received are processed and assessed by clerks and heads of units. Raw responses
from enterprises to the monthly branch surveys may be altered without appeal from the enterprise if the error is
obvious (unit error) or after appeal by the enterprise, if necessary. Once the individual data have been verified
and validated, the industrial production indices are calculated (taking into account imputations in the event of
non-response), accompanied by an analysis of the indices at all classification levels to identify possible atypical
changes and to account for (or correct) them as part of a top-down approach from the most aggregated indices
down to individual data.
4- Sequence of a Monthly Campaign
For publication on D-Day (no later than the 10th day) of month M and relating to production in month M-2:
At (D-45), the campaign is launched. Thus, for example, in the case of the publication of the January index on 10
March, the campaign is launched around 25 January. The questionnaires are published online on the collection
site and a letter (letter of notification) is also sent out to businesses informing them that the questionnaire for the
month under review is available and requesting them to respond.
From (D-45) to (D-15), the SSNE collects the data: r eceipt and verification of the responses submitted by
participating enterprises. In the event of very significant changes, the enterprise in question is requested to check
the accuracy of the data and to identify any reporting errors. Non-responding enterprises are sent reminders,
particularly the largest entities. The collection process continues throughout the month.
From (D-14) to (D-6), the SSNE c alculates the elementary series based on the responses provided by enterprises
and non-response imputations and analyses changes and revisions. It may also be required to correct automatic
imputations. It sends reminders to enterprises that have reported atypical changes and, if necessary, corrects their
responses. It takes note of comments provided to explain changes and revisions. It also continues to send
reminders to non-responding enterprises.
From around (D-20) onwards, the unit responsible for calculating the IPI begins the initial index calculations,
analyses the series and identifies any problems. Following the initial calculations, the indices are then updated as
new data become available for advance estimates (see below). The unit also uploads external series: SSP data
(food and agriculture), SDES data (manufacture of coke, energy), SDES data collected by the FFB and FNTP
(construction) and DARES data on temporary work in construction. It then incorporates the price series for the
57
series monitored in invoicing terms and the productivity coefficients for the series tracked in hourly terms. The
IPI section also calculates the gross aggregate and seasonally and working-day adjusted indices. Finally,
contributions to changes and revisions are calculated.
Between (D-5) and (D-2), the IPI section conducts the final analysis of the gross aggregate and seasonally and
working-day adjusted indices. To do so, it draws on the changes that contributed the most to the change in the
overall index, i.e. the most significant changes or those affecting branches with a high weight. The aim is to
understand and explain them. In other words, are the observed changes the result of atypical production by a
business? Are they due to the treatment of seasonality and working days? This work can result in the SSNE
sending a reminder to the enterprise and, if necessary, in correcting the declaration. Similarly, the section focuses
on understanding the most significant revisions: can they be explained by late declarations by enterprises,
corrections to past declarations or updates to seasonal adjustment models? (see below)
From (D-2) to (D-1), the dissemination is validated and prepared. After a final calculation of the IPI, the IPI
section drafts the “ Informations rapides” by way of commentary on developments in the IPI and any relevant
revisions before submitting the publication to the formal approval process. It then uploads the data (long series)
to the relevant database at insee.fr (“Indices and time series”), with INSEE then sending the series to Eurostat on
D-day. The section ensures that operations run smoothly until publication.
The whole thing remains under embargo until D-Day at 8:45!
The media or external or internal users may then request further details on developments in the IPI, which
happens relatively frequently.
5- Implementation of Early Estimates for the IPI
Since 2013, INSEE has been producing early estimates of the IPI for internal purposes. These indices are
calculated at month end (30 days) to update the cyclical diagnosis established in the context of business cycle
scores and points. These early estimates also help to secure the process by providing initial estimates well in
advance of publication.
In addition, since 2016, in the first month of each quarter (January, April, July, October), the IPI section also
calculates an early estimate at 23 or 25 days (indices for December, March, June, September). The indices for the
third month of the quarter enable INSEE’s quarterly accounts division to calculate an initial estimate of GDP
using as much information as possible. The estimate is published 30 days after the end of the quarter (a system in
place since the end of 2016) as opposed to 45 days previously.
6- Analysis
The analysis conducted by the IPI section aims first of all to understand changes in total industrial output
compared to the previous month, adjusted for seasonal and working-day variations.
The seasonally adjusted change in the index of total industrial production results from the changes to its
components, in particular within the classes forming it, these being the most detailed seasonally adjusted series 47.
The IPI section focuses its analysis primarily on the classes making the biggest contributions and conducts
detailed examinations of the changes applicable to them.
The analysis of class changes begins with the determination of the elementary series that contributes (or
contribute) the most. For example, if the seasonally and wording-day adjusted index for “Manufacture of
machinery for food, beverage and tobacco processing” decreases between two consecutive months, it is
necessary to determine which subseries within it account for the decrease.
The elementary series contributing to the seasonally adjusted change cannot be determined immediately, notably
because a link must be established between a series at the “class” level whose seasonally adjusted change is to be
47As a reminder, below the class level, there is the NAF subclass level and the elementary series level (see Chapters 3 and 5).
These last two levels are only available in “gross” form, i. e. before seasonal and working-day adjustment.
58
explained and elementary series that are not adjusted (the minimum level of seasonal adjustment being NACE-
4).
To facilitate the work involved, several analytical tools are available. The key tool is the contribution to year-on-
year change and the year-on-year change rate, which are calculated for each elementary series, the idea being
that an atypical monthly change will be reflected in the year-on-year change or the year-on-year change rate,
even if the relationship is not perfect. Various other graphical and display tools are provided to facilitate the
analysis, in particular the annual superposition of the indices of the elementary series, thereby highlighting
additive outliers in relation to the usual seasonal profile.
Once the elementary series explaining changes in the series at the “class” level has (or have) been determined, it
is then necessary to understand which enterprise (or enterprises) accounts (account) for this change and why.
Here too, the key tool is the year-on-year analysis of business output. If necessary, the enterprise may be
contacted by the SSNE to explain or correct the atypical individual change.
With regard to revisions, the IPI section focuses on the series that contributed the most. The aim is to establish
the reason for the revision based on the information provided by the SSNE. The process of analysing revisions
can lead to the detection of anomalies, which are then corrected.
A significant part of the analytical work carried out by the IPI section consists in examining the treatment of
seasonal and working-day adjustments, whether in terms of changes in the last month or revisions to previous
months. The aim is to ensure that the models carry out appropriate and relevant operations. Adjustments may be
carried out if necessary, including by adding or removing an outlier. For example, a significant change in the
index over the past can be excluded from the calculation of seasonality if it relates to a clearly identified
exceptional event (temporary closure of a plant leading to a fall in the index, which therefore does not relate to a
seasonal phenomenon, etc.).
7- Revisions
The Purpose of Revisions
Indices are generally revised, albeit within a limited range (see below), several times after their initial release, for
several reasons:
the declarations of certain enterprises are received late by the body conducting the survey;
errors in the interpretation of questionnaires by respondents or data entry errors (whether by the
responding enterprise or the collecting body) may not be detected during the initial calculations;
seasonal and working-day adjustments naturally lead to revisions to the seasonally and working-day
adjusted indices as new monthly data become available to refine the estimates;
INSEE conducts the annual production survey, which provides more detailed and complementary
results for the year preceding the last period covered by the IPI; monthly and annual data are then
compared, which may result in revisions to the IPIs.
The amount and scale of revisions decreases as one moves further away from the month in question. Successive
revisions lead to a gradual convergence towards a more accurate estimate of production (with the receipt, in
particular, of late responses). In line with the European Statistics Code of Practice, and for the sake of
transparency, revisions made to recent months are reported in the monthly publication (see Chapter 9).
Late Declarations
For some products tracked among a large population of enterprises, non-responding enterprises may, at the time
of the first release, account for between 10% and 20% of the month’s total production.
59
The tools available provide a means of estimating the output of non-responding enterprises in addition to that of
responding enterprises and of calculating the production statistics for month (m). The initial statistic for month
(m) is termed “provisional”.
The following month, late respondents submit their production declaration for month m. The imputations are
then replaced by the responses received. This then becomes a “corrected” statistic for month (m).
Some enterprises may not respond to surveys for a relatively long period of time: in such cases, the estimate
applied over several months will be weaker, and when the enterprises in question do respond, the corrections
made may have a significant retroactive effect, particularly in cases where the enterprise carries significant
weight in terms of production on national soil.
Correction of Response Errors
Inevitably, some enterprises make mistakes in their declarations, whether material errors, errors in the
interpretation of the explanatory notes to the questionnaires or errors related to staff changes or changes in
internal information systems.
When these errors are detected (either by the responding enterprise or by INSEE following checks carried out on
the data), amended declarations are drawn up, resulting in revisions to the indices.
In this context, the availability (the following year) of detailed annual data from the EAP can also lead to
reporting errors being detected and in values from past series being reviewed.
Corrections Related to the Updating of Seasonal Adjustment Models
Seasonal adjustment models and seasonal coefficients are re-estimated each month since knowledge of the index
for month m+1 enables models and estimates to be further refined. Even without any change in the raw data,
knowledge of the index for an additional month has an (admittedly limited) impact on the estimation of the trend,
on the estimation of seasonality coefficients and, more generally, on all stages of the seasonal adjustment process
(see Chapter 7). In particular, the detection of additive outliers (outliers, level changes, etc.) is updated over the
last few months, with potentially significant effects on the estimation of model parameters.
Taken together, all of these factors result in moderate revisions: on average over the period 2015-201848, changes
in the industrial index of the previous month was revised by 0.16 percentage points (see Table 1), either upwards
or downwards, pointing to a limited range compared to the standard deviation of the series (1.4 points over the
period). The bias (average of revisions) is close to 0 and not significant.
Of course, revisions made to industrial production indices may offset each other: in other words, an upward
revision to changes in one series may be offset by a downward revision to changes in another series. Similarly,
revisions to the raw data of one series may be mitigated by a revision arising from updates to the seasonal
adjustment models in the same or another series. Thus, we see that the revision of the development in the
seasonally and working-day adjusted manufacturing index is lower on average than the revision of the change
due to the raw data (0.16 compared to 0.22 points), with the updating of the seasonal adjustment models
generally mitigating the revision of the raw data.
48More specifically, January 2015-June 2018, and excluding months associated with a major change in methodology (January
2018: base change; March 2016: update of weights). Average of the revisions in absolute terms applied to the change in the
manufacturing index published the previous month. As an example, when the March 2018 IPI was published, the change in the
manufacturing index between January and February 2018 was revised by +0.09 points. Having initially been estimated at -0,6%,
it was increased to -0,5 % in the April publication. The +0.09 point revision breaks down as follows: + 0.05 points related to the
revision of raw data and + 0.04 points related to the updating of the seasonal and working-day adjustment coefficients.
60
Table 1: Revisions Affecting the Manufacturing Index for the Last Published Month (Absolute
Revisions, in Points)
Average of revisions
in absolute value
1st quartile Median 3rd quartile
Revisions – manufacturing –
Total
0.16 0.08 0.14 0.20
Including revision due to
updating of raw data
0.17 0.06 0.12 0.27
Including revision due to
updating of seasonal
adjustment models
0.10 0.04 0.07 0.13
Reading note: the average of the revisions in absolute value (in other words, without taking into account the
direction, i.e. upwards or downwards, but only the range) of the change in the manufacturing index in the previous
month (for example, the change in the manufacturing index between August and September, during the October
index calculation campaign) is 0.16 points. 25% of the revisions are lower than +0.08 points and 75% are below
+0.20 points. These statistics were calculated over the period January 2015-June 2018, excluding months
corresponding to a major change in methodology.
61
Chapter 9 - Publication and Dissemination
1- Publication of the Industrial Production Index: Informations Rapides
The IPI is published monthly by INSEE no later than 40 days after the end of month m in the Informations
Rapides series. However, if m+40 falls on a Saturday, a Sunday or a public holiday, publications are announced
and scheduled for the previous Friday at 8:45 am. This document presents the main results by major branches
and refers to the revisions made to the previous month.
The INSEE press office is responsible for communicating it to the media. As for the other main economic
indicators, monthly and four-monthly schedules are also drawn up:
https://www.insee.fr/en/information/2107811?debut=0
The IPI results are made available on the INSEE website ( www.insee.fr) on the Home page under the heading
“Key indicators” as soon as the embargo is lifted (at 8:45 am on the day the index is published). Users can then
access the main information of the monthly campaign in both French and English, download the relevant
‘Informations Rapides’ in the form of a printable file and access the long series (see following pages).
62 63