A key issue for statisticians is knowing whether the statistic they produce actually represents the population they want to cover. To that end, they need reference data, which describe the population covered and which they can compare with their own data. Registers play this role by identifying all statistical units of interest.

For the production of business statistics, the inter-administrative register SIRENE has long been the go-to reference in France (https://www.insee.fr/fr/information/1972043). From 2008, a complementary register dedicated specifically to the needs of public statisticians entered the landscape: SIRUS (Système d'immatriculation au répertoire des unités statistiques – Statistical Business Register) therefore has different coverage and different observation units from those of SIRENE. Backbone of business statistics, it centralises many information from various statistical processes. It offers a wide range of services, both to survey managers and user information systems.

Collecting information via an appropriate questionnaire, correcting non-responses or disaggregating statistics according to various criteria will not be relevant if the population based on which these statistics are calculated is too specific to draw any general information. To avoid this pitfall, the survey manager randomly draws a sample from a sampling frame. This must include all units of interest for the survey: this will be, for example, all the units with an industrial activity for the French Annual Production Survey (Enquête annuelle de production – EAP) or all the companies created in a given half-year for the French New Enterprises Information System (système d'information sur les nouvelles entreprises – SINE).

While each survey uses a sampling frame according to its field of interest, it is necessary, in order to constitute that sampling frame, to rely on a repository that exhaustively lists the statistical units: a register.

In order to meet the need to draw sample bases, however, the register must go further than the basic listing of statistical units; it is important to gather other information to refine the coverage of the sampling frames to be created. For the annual production survey mentioned above, it is necessary, for example, to know which companies carry out an industrial activity, even if it is not their main activity; for the New Enterprises Information System, it is essential to know the date of creation of the company.

The amount of information thus centralised in the register can be large: variables to identify market oriented units (i.e. the coverage of business statistics), but also the legal status of the company, the sector of activity, the category of company within the meaning of the Law on the Modernisation of the Economy, the institutional sector, market oriented or employer indicator, its turnover, number of employees, etc.

This census of statistical units is essential for statistics produced based on surveys, but not only for them: it is just as necessary for producing business statistics based on administrative data. The field of administrative data covers only the units subject to the administrative obligation that generated the source and it is necessary to compare it with the population that the user wishes to reflect: it is then necessary to correct, where required, the coverage biases of the administrative source (Rivière, 2018). For example, not all companies file tax returns; this is the case for micro-enterprises. Therefore, any business statistics based on these tax returns must be adjusted to take into account the turnover of these micro-enterprises.

In the case of France, statisticians can rely on the SIRENE (Système informatique pour le répertoire des entreprises et des établissements – Information system on business and establishment register) inter-administrative register, which has been managed by INSEE since 1973: each company must, in accordance with the French commercial law, be registered in SIRENE and report any change in its circumstances (a change of address, for example, or the creation of an establishment in a new location). In the management of SIRENE, particular attention is paid to avoid duplicate registrations. Since 1997, the use of the SIREN identifier has been mandatory for all administrative declarations and formalities for companies, which not only guarantees the coverage of the register but also that the information is up to date and accurate. The scope of SIRENE is even broader than the business statistics one’s, i.e. all companies involved in the market production of goods or services: since 1983, both public companies and administrations have been required to register in the SIRENE register. Associations, when they are employers, whether they apply for aid or pay taxes, are also registered.

This SIRENE register is therefore an essential resource for creating a register for statistical purposes. Most countries can rely on an administrative register, but it is rarely managed by the national statistical institute, as is the case in France. This peculiarity makes it possible to take into account statistical needs: the possibility of carrying out quality improvement operations on addresses, of ensuring the correct activity code is used, etc. In the absence of such a register as a starting point, a substantial amount, of work is required to set up a statistical business register: matching and merging various administrative sources makes it possible, through comparison, to attempt to go beyond the coverage limitations of each source. Companies can also be identified by a population census, or by harvesting data from, for example, websites or private directories (Ouvrir dans un nouvel ongletUNECE, 2015).

Thanks to its freshness and expansive coverage, SIRENE has long been used by statisticians who have directly linked their statistical processes with this register.

However, using the same register for both administrative and statistical purposes poses several difficulties. Indeed, the management of an administrative register is not the same as that of a statistical register. For example, it is not possible to change the value of the variable for a statistical purpose; any change to the SIRENE register may have legal implications for the company and it is important that it aligns with the company's declaration. However, statisticians may exclude a company from their sampling frame (especially in the case of cessation of activity) to avoid any unnecessary re-interviewing in the future. However, this would of course be unthinkable in the context of an administrative register, given the consequences for the company in the case of erroneous cessation.

Similarly, requiring SIRENE to integrate new variables for statistical purposes increases its management burden and adds uncertainty regarding the status of its statistical variables and their use. Let us take the example of the APE (Activité principale exercée – Principal Activity) code of a company: this variable is included in the SIRENE register, but its purpose is primarily statistical. However, its presence in the SIRENE register makes it public information that can be easily used by anyone, to obtain aid or subsidies or to open up entitlements or obligations, for example. As a result, this "administrative" use and its practical consequences lead many companies to contest the coding of their main activity by INSEE. This information remains in the SIRENE inter-administrative register, together with the company category, another statistical variable; but both show the difficult balance that must be established when disseminating statistical variables at individual level.

Due to these difficulties, INSEE wanted to separate the statistical and administrative uses of the registers. A statistical register, SIRUS, has been set up since 2008, in addition to the SIRENE register. The separation will be completed with the switch to SIRENE 4, scheduled for 2023, whose use will be purely administrative (Alviset, 2020). All variables with a statistical purpose will be stored in SIRUS, which will become the entry point for all statistical needs; in contrast, SIRUS is used for statistical needs only. It is thus an internal tool of the Official Statistical Service that is not intended to be disseminated externally (box 1).

Two other reasons justified the separation between administrative and statistical registers. The first is the desire to better define and control the coverage of business statistics. Due to its administrative use, SIRENE is very attuned to the needs to identify units concerned by the collection of tax or social security contributions.

A certain number of legal units (see below) are registered to fulfil administrative obligations without having any real economic consistency (for example, non-trading resource companies [sociétés civiles de moyens] for the management of shared resources in the liberal professions).

As a result, the useful coverage of SIRUS is narrower than that of SIRENE. It excludes, for example, managing partners, natural persons who are landlords, the non-trading resource companies mentioned above, the majority of non-trading property investment companies, associations that are not employers, etc. While the SIRENE register had more than 14 million active legal units in July 2020, SIRUS had around 8.4 million (figure 1).

Figure 1 – The SIRUS register in figures

The coverage of SIRUS is, however, a little broader than that of business statistics: SIRUS thus lists employer units, including those that are non-commercial, including public or association units, etc. Thus, of the 8.4 million legal units active in SIRUS, just over one million correspond to non-commercial and non-productive legal units outside the scope of business statistics.

The second reason behind the creation of SIRUS is the definition of the unit of interest of the register: what SIRENE lists is not always what interests statisticians. Indeed, SIRENE registers legal units and their establishments. For example, a company that sells shoes will form a legal unit with several establishments (its multiple shops) that will all be registered in SIRENE.

However, this legal unit, defined from a legal point of view, is not necessarily the most relevant unit for analysing the activity of companies, their decisions and their strategies: with the development of groups of companies, the legal units held by others may lose all or part of their decision-making autonomy. In our previous example, let's imagine that the shops are owned by a group that also has a central purchasing body and an entity that advertises the brand: in order to understand the economic decision, it will be more beneficial to analyse these legal units (shops, central purchasing body and advertising company) all together. The financial links between the legal units are used as the basis for defining the outline of this group of companies. The group is a new statistical unit, but it has no real legal existence and is therefore not included in SIRENE.

In addition to this notion of a group (defined by financial links), economists and statisticians have established the notion of an enterprise, in the economic sense, defined as "the smallest combination of legal units that is an organisational unit producing goods or services, which benefits from a certain degree of autonomy in decision making, especially for the allocation of its current resources". Again using our example, let us imagine that the shoe sales group also owns sports centres, for financial reasons, but they are managed completely independently of its shoe shops. It will be more relevant, from an economic point of view, to consider within the group, firstly, an enterprise in charge of the shoe sales (including the shop, the central purchasing body and the advertising company) and, secondly, an enterprise that includes the network of sports centres (box 2). This enterprise, in the economic sense, is a convention: it has no legal reality, but from the point of view of economists this convention is the most suitable way to describe the decisions of the company in a relevant manner. From a practical point of view, the delineation of these enterprises, in the economic sense, is constructed by statisticians, using so-called profiling methods, whether manual or automatic (Haag, 2019).

In France, structural business statistics have been constructed since financial year 2017 on the basis of the enterprise in the economic sense, rather than legal units. This change has led to a re-evaluation of certain economic aggregates, such as production, because monetary flows with no real economic consistency between legal units of the same group are rendered null (Chanteloup and Haag, 2019): thus, taking into account enterprises leads to a "drop" of about 7% in total turnover (intra-company sales are no longer counted with the transition to the notion of an enterprise). In addition, the turnover breakdown by activity is amended in favour of industry, because the delineation of industrial enterprises include not only industrial legal units, but also trade and services units, activities that "support" the enterprise's primary industrial activity.

Compared to the SIRENE inter-administrative register which lists the legal units, SIRUS lists these different statistical units and identifies the links between them: the establishments of a legal unit, the legal units, the enterprises in the economic sense and their belonging to a group.

In order to facilitate its handling and maintenance, only the information frequently used by statisticians is stored in SIRUS, i.e. the main characteristics of the units (date of creation, primary activity, number of employees, etc.) and the links between those units. More detailed information should be sought elsewhere: a register of groups, together with data on restructuring, complete the SIRUS register.

The comprehensiveness of SIRUS in terms of statistical units makes SIRUS the reference register for designing sampling frames. However, the centralisation of all this information on companies allows other uses of SIRUS: we calculate, for example, the notion of statistical cessation of activities by companies.

The fact that a company is no longer present in the administrative sources used to update SIRUS or that the data relating to that company have not been updated for some time are an indication that the company has ceased its activity. This information is crucial for the accuracy of surveys that use SIRUS: such a company, which has statistically ceased activity, will not respond to a survey. It should not even give result in the adjustment of its non-response (e.g. an extrapolation of its turnover based on the responding companies). Identifying statistical cessations of activity improves the quality of survey results and, obviously, saves unnecessary collection costs.

The continuous updating of information in SIRUS also makes it the suitable source for studying business demographics: not only the number of company creations, but also stocks and cessations of activity. SIRUS has been used for the production of these statistics since 2022 (INSEE, 2022). Thus, business demographics are now calculated using the concept of enterprises and benefit from the processing of statistical cessations of activity. In addition, this makes it possible to have a more consistent coverage with other business statistics productions.

In France, survey data and administrative data are almost always based on the SIREN number, so it is easy to match them to combine the information. However, there are still rare cases in which this information is unavailable (one example among others is the road freight survey, which samples vehicles) or even erroneous. In this case, it is necessary to successfully find the identity of the company (the French also talk of "sireniser" the information, i.e. linking the information with a SIREN identifier), based on the known characteristics of the company in the source used, such as the name of the company or its establishment address. SIRUS, which gathers a lot of information about each company, is the ideal tool for identifying them and retrieving a large set of data describing them.

Finally, a new use has been assigned to SIRUS: measuring the statistical burden on companies. For each company, SIRUS compiles the various official statistics surveys for which it was interviewed, as well as the time the company spent responding; the question is indeed mandatory and always asked at the end of each survey, since the Warsmann Commission on reducing administrative procedures for companies (Ouvrir dans un nouvel ongletFrench National Assembly, 2011). The statistical burden of responding to surveys is a valuable information that allows to negatively coordinate samples, i.e., when drawing up these samples, having a higher probability of choosing companies for which the burden is lower. This makes it possible to spread the burden of surveys between companies in a fairer way.

The construction of SIRUS is based on the SIRENE register enhanced by many other statistical processes, which themselves rely on SIRUS (to draw up samples, identify units, define their coverage, etc.). SIRUS is thus at the heart of a complex system.

Indeed, SIRUS uses information from various administrative sources to enrich the knowledge of its units: VAT data, tax returns, employment data and data from the UCN (URSSAF Caisse Nationale – French Social Security and Family Allowance Contribution Collection Office) making it possible to identify micro-enterprises, as well as customs data to identify importing or exporting companies (figure 2).

Figure 2 – SIRUS, at the heart of statistical information systems

In return, the centralisation of all this information makes it possible to describe the units in detail and to ensure their completeness; it also makes it possible to provide a common register for all business statistics in order to make them comparable although they are produced from different processes. With them all based on SIRUS, the definition of the activity, of the company category and what is included or not included in their coverage should be identical from one survey to another.

However, behind this ambition there are of course difficulties: Sirus is continuously updated using various sources of information. The changes made to it reflect both an economic reality (for example, in December 2021, the active units are not all the same as a year earlier) and the availability of information, which is sometimes delayed (the cessation of activity of a company in December 2021 is known only in March 2022). SIRUS thus logs the information chronologically to take into account the changes over time due to real economic events that impact units and also due to the updates of the sources: the changes thus have a processing date and an effective date (or a validity date, in economic reality). However, given the number of sources involved in updating SIRUS (more than ten), it is not possible to log all variables and changes chronologically, so as to give SIRUS a certain degree of manageability.

There is also another difficulty with the internal consistency of the register: some variables, such as the company category or the coverage of business statistics, are calculated by SIRUS on the basis of the different pieces of information provided by each source. The interdependence between variables in the SIRUS register can be such that it is not always easy to ensure that a correction is spread across the entire process. Therefore, a correction can create a dilemma between the potential inconsistency it can create within the register and the concern for accuracy and freshness (having the latest information from a data source). Sometimes it is necessary, as a last resort, to restore the database using an earlier version from before an erroneous piece of information was taken into account.

Finally, all these sources of information can change units for the same characteristic (an annual structural survey can change the primary activity code whereas the company has requested to the SIRENE register a change of this code for administrative reasons). Therefore, it is necessary to weigh up these different updates to know to which of them SIRUS should give precedence: should the most recent information be chosen? Or should some sources be considered to have priority for certain coverages?

To manage these conflicts, a set of rules has been established aiming to determine, for each variable, the order of priority for each source based on the proposed effective date of the update (SIRENE cannot, for example, update the activity of a company for year N if this company was surveyed in the same year by an annual structural survey).

The stakes are high because in order to be usable, SIRUS must provide its users with the freshest and richest information possible, but it must also be consistent, easy to use and traceable (the user wants to be able to recover the sample frame as he defined it at a given time (T) using SIRUS).

Although business registers are not a "statistic" provided to Eurostat by national statistical institutes, they are the subject of much attention in European and international work, given their central role in the production of all other business statistics: manuals and working groups aim to harmonise and improve the quality of these registers (Ouvrir dans un nouvel ongletUnece, 2015; Ouvrir dans un nouvel ongletEurostat, 2021).

Eurostat or the United Nations Statistical Commission thus encourage statistical institutes to carry out self-assessments of the maturity of their business registers: the coverage of those registers, not only in terms of institutional sectors, but also of statistical units is an element of this assessment, as well as the number of sources taken into account, the frequency with which they are updated, their internal and external consistency, their geographical detail, the legal instruments on which they are based and their interoperability (Ouvrir dans un nouvel ongletCollet and Iskrenova, 2021; Ouvrir dans un nouvel ongletAl Kafri and Hermans, 2021).

While these registers are a challenge at European and international level, it is also because the relevance of their national dimension is being called into question by globalisation (Ouvrir dans un nouvel ongletSturm, 2021). By definition, national business registers cover the national territory: the French register therefore contains the French legal units and what is called the "French footprint" of groups and enterprises in the economic sense. Indeed, some companies are transnational and some of the legal units on French territory may belong to a company that has its decision-making unit based outside France. In this case, French statistics and registers track only the footprint of the company located on French territory, meaning that they reflect only a part of the economic choices made by the company.

Due to the fact that it may be more relevant, when attempting to understand and analyse the decisions of these companies, to view them from a supranational level, a European Group Register (EGR) is being built by the various European national statistical institutes under the aegis of Eurostat; similar attempts are being made at the global level by the OECD (Analytical Database on Individual Multinationals and Affiliates, ADIMA) and the United Nations Statistical Commission (Global Enterprise Group Register), taking into account the considerable weight of multinational companies in the global economy (Ouvrir dans un nouvel ongletSnyder, Di Matteo, Pilgrim and Ostolaza, 2021).

The European Group Register is being constructed based on national registers and makes it possible to define the delineation of European groups. Ultimately, interoperability between these national registers, including SIRUS, and the European Group Register will be required, i.e. SIRUS will need to be able to mark European groups with the identifiers from the European group register and ensure the consistency of the data of the two registers in the coverage of multinational companies. These developments are foreseen by the new European regulation on Structural Business Statistics (Colin, 2019) which also raised the level of requirements for national business registers in terms of the variables to be covered (whether mandatory or recommended, such as the Special Purpose Entities indicator) and reaffirmed the central role of these registers, as the real backbone of business statistics, to ensure the consistency of data over time, across Member States and across different domains (short-term statistics, structural business statistics, business demographics, trade statistics by business characteristics, etc.).

Driven by these regulations, SIRUS should therefore be enriched with new variables in the upcoming years: the intra-Community VAT number, the LEI (Legal Entity Identifier) and the identifier of companies and groups from the European Group Register (EGR), in order to facilitate matching with other sources.

Another significant challenge for SIRUS is the redesign of the SIRENE inter-administrative register to which it is linked: as of January 2023, the ecosystem of business formalities will be thoroughly changed, with the introduction of a one-stop shop for companies that will replace all business formality centres and bring about the complete digitalisation of these formalities. The SIRENE register will be thoroughly redesigned to take this new context into account (Alviset, 2020). The challenge for SIRUS is to continue producing the same services while its primary source will be greatly changed.