Handbook on Extended Supply and Use Tables and Extended Input‑Output Tables

3. Compiling extended supply and use tables using enterprise data

Abstract

This chapter describes how to compile extended supply and use tables using enterprise data. After a brief overview of the process, it provides an explanation of the six steps of the process: 1) categorising enterprises and choosing industries to be disaggregated; 2) disaggregating total industry-level output and intermediate use; 3) disaggregating the product distribution of output and intermediate use; 4) deriving extended valuation tables and an extended use table at basic prices; 5) disaggregating use into use of imports and of domestic production; and 6) disaggregating the use of domestic input. Sample tables illustrate each step.

Introduction

This chapter provides guidance to extended supply and use table (ESUT) compilers that have access to detailed microdata, in particular at the enterprise level. In general, this includes the microdata underlying the construction of regular (SUTs), such as the economic census, administrative data (annual accounts, tax data, customs data, social security records, etc.) or specific surveys (on business statistics, of industrial production, on households, etc.). Compilers may also have access to additional data on aspects that enterprise data do not cover, e.g. information on the informal or the illegal economy. The key idea in the approach described here is to construct ESUTs by disaggregating columns and rows of the regular SUTs, using as much as possible data to capture the heterogeneity between different types of enterprises. The chapter proceeds by explaining how to do this step by step for an individual country, building on earlier experiences of national statistical organisations (NSOs) which have compiled ESUTs. Since the resulting ESUTs may be unbalanced, it may be necessary to use a balancing procedure, as explained in Chapter 5. The results of this approach are ESUTs that can be used as input for deriving an extended input-output table (EIOT) – the input-output table (IOT) where industries are disaggregated by type of enterprise. Chapter 6 will explain how to do this.

Overview of the process

In theory, ESUTs for an individual country could be compiled entirely bottom-up based on the data sources and methods used to compile the regular SUTs, following the concepts of national accounts, while distinguishing categories of enterprises within industries. Such an approach would also use the same statistical units (establishments are recommended) as the regular SUTs. However, although this is the ideal (and therefore recommended) way to compile ESUTs, it is likely to be very resource-intensive and time‑consuming. There are no known examples of ESUTs compiled bottom-up but they could align well with the regular SUTs if planned from the beginning of the process. This chapter describes a top-down method as an alternative approach to compiling ESUTs using microdata. The description builds on experiences of countries that have compiled ESUTs, among them Belgium (Hambÿe, Hertveldt and Michel, 2018[1]), Denmark (Nilsson, Rørmose Jensen and Holst Jensen, 2018[2]), Italy (Sallusti and Cuicchio, 2023[3]), Mexico (INEGI, 2023[4]) and the Netherlands (Chong et al., 2018[5]). This is the recommended approach when microdata are available. Chapter 4 focuses on an approach to compiling ESUTs when microdata are not available.

The method described in this chapter consists of breaking down regular SUTs into ESUTs for an individual country. It takes the country’s regular SUTs as a starting point. Compilers leverage whatever microdata they have at their disposal to bring within-industry enterprise heterogeneity into the regular SUTs. Hence, in this process, the disaggregation of industries and product categories is based as much as possible on enterprise-level data. However, when no data are available, disaggregations may be based on assumptions, including straightforward proportionality assumptions. In addition, depending on the aims for constructing ESUTs and the country characteristics, certain disaggregations may not be necessary. It is up to the compiler to determine whether the assumptions made in the ESUTs’ construction process are compatible with the pursued statistical and analytical aims. Furthermore, the compiler should also keep in mind the plausibility of the results of each step. Chong et al. (2016[6]) gives the example that an enterprise with less than 50 employees is unlikely to build a complete sea tanker. In such cases, one can substitute the “irregular” product-type of enterprise combinations with other products produced by the industry at hand. If that is not possible, one can shift supply and use to a more likely type of enterprise using expert judgement.

A schematic overview

Table 3.1 provides a schematic overview of the ESUT compilation process covering the steps to be taken, the data to be used in each step and the results of the steps. It is based on the presentation of ESUT compilation work for Belgium in Michel et al. (2019[7]). It follows rather closely the steps of the construction process of the regular SUTs at basic prices according to the recommendations in the United Nations Handbook on Supply and Use Tables and Input-Output Tables with Extensions and Applications (United Nations, 2018[8]). Steps 1, 2 and 3 lead to a “column-extended” supply table in basic prices and a “column-extended” use table at purchasers’ prices. The derivation of extended valuation tables in Step 4 allows obtaining a “column-extended” use table at basic prices. Steps 5 and 6 cover the disaggregation of the rows or product categories by origin (imports and domestic production of the different types of enterprises) and result in fully extended SUTs at basic prices. Each step is explained in greater detail below.

Table 3.1. Steps in the extended supply and use table compilation process

Steps	1	2	3	4	5	6
Workflow	Categorising enterprises and choosing the industries to be disaggregated	Disaggregating total industry-level output and intermediate use	Disaggregating the product distribution of output and intermediate use	Deriving extended valuation tables and an extended use table at basic prices	Disaggregating use into use of imports and of domestic production	Disaggregating the use of domestic output by producing enterprise category
Data	Enterprise-level data on the extension criteria	Enterprise-level data on total output and purchases	Enterprise-level data on output and purchases by product category	Enterprise-level data on margins, and taxes and subsidies on products	Enterprise-level data on imports by product category	Enterprise-level data on exports by product category and transaction data (value-added tax)
Result	List of enterprises by type	Column totals by type of enterprise in supply and use tables (SUTs)	Columns by type of enterprise → Column-extended SUTs at purchasers’ prices	Column-extended valuation tables and use table at basic prices	Extended import flow table	Rows by type of enterprise in use table for domestic output → fully extended SUTs (rows and columns)

Source: Adapted from Michel et al. (2019[7]).

Note that Steps 4, 5, and 6 are, strictly speaking, not necessary when deriving an ESUT. However, they provide the basis for deriving the EIOT. To compile an EIOT, one needs to have an extended use table in basic prices (Step 4). Furthermore, one would like to consider that certain types of enterprises use more (less) imports and have more (less) exports. This is achieved in Steps 5 and 6, respectively.

Step 1: Categorising enterprises and choosing the industries to be disaggregated

First, compilers must decide on the extension criterion (or criteria) for compiling their ESUTs. Chapter 1 provided guidance on this choice. The choice also depends on the characteristics of the economy; for some countries, it can be relevant to disaggregate by formal/informal economy or whether enterprises are “green” or “social”. The most frequently used criteria have so far been trading status, group affiliation and size. Chapter 2 reviewed the definitions and data sources for these three criteria.

Whatever the criterion, the available microdata should allow enterprises in the business register to be classified into categories, using a type of category variable. This is the first step according to the overview above. The relevant information for a classification may not be available for part of the enterprises in the business register. These enterprises may then either be assigned to a default category, for example, all enterprises for which there is no information on ownership may be considered as being part of the category of domestic stand-alone enterprises (see Chapter 2), or they may be classified in a separate additional category of their own, i.e. “other” or “non-classified” enterprises.

Which industries to break down?

Second, the question of which industries to disaggregate must be answered early in the construction process of ESUTs because it has implications for the entire process. Ideally, compilers should draw up a list of industries selected for disaggregation and clearly label those that have not been disaggregated when presenting their results. For practical purposes, compilers may define an additional enterprise category that includes all enterprises in industries that are not disaggregated. For example, a disaggregation by size into large enterprises and small and medium-sized enterprises (SMEs) would then be expanded to include a category of “other” enterprises. The values for this category are then: equal to zero in all industries that are disaggregated; and equal to the value for the entire industry in all industries that are not disaggregated.1 This way of proceeding avoids a “sparse” disaggregation, i.e. not having the same enterprise categories for all industries, and it makes alternative aggregations easier.

In most cases, it is neither feasible nor useful to disaggregate all industries (Ahmad, 2023[9]). In the first instance, the decision on whether to disaggregate an industry will depend on the chosen disaggregation criterion (e.g. the industries to be disaggregated might not be the same for the size criterion as for the ownership criterion). This may have to do with the inherent characteristics of the criterion, legal provisions or institutional arrangements, which imply that it is simply not meaningful to disaggregate certain industries for a criterion. For example, there is no good reason to disaggregate industries that belong to the public administration by ownership/group affiliation, or to isolate exporters in an industry when an export ban affects the main product of the industry. In this context, it is important to distinguish between industries that the compiler decides should not be disaggregated and industries where all enterprises are part of the same category. Enterprises that are part of the same category could include agriculture in countries where farming is exclusively a small-scale family business or for the extraction of crude petroleum in countries where all enterprises in that industry are exporters. Disaggregation of real estate (L in ISIC Rev. 4) deserves special attention. Part of this industry consists of “imputed rents of owner-occupied dwellings” (no enterprises involved) and part consists of “renting and operating of own or leased real estate” (enterprises involved).

The decision not to disaggregate an industry may also be determined by data availability. There may be different underlying reasons for such a lack of data. The data required for classifying enterprises may simply not be collected for an industry. For example, all enterprises in the industry may be below the size threshold for collecting ownership data. Alternatively, even when the data are collected, ESUT compilers may not have access to the required data for classifying enterprises in certain industries.

Furthermore, confidentiality issues regarding the results of the disaggregation may also play a role. This is the case when values for one category of enterprises in an industry are based on observations for very few enterprises or even a single enterprise. Then, to preserve confidentiality, the compiler may not be allowed to publish results by category of enterprises for this industry and may therefore prefer not to disaggregate the industry. Finally, compilers may also decide to restrict the number of industries to be disaggregated to reduce the scope of the exercise and keep the workload manageable, or because the research question at hand does not require the disaggregation of additional industries. For example, Michel, Hambÿe and Hertveldt (2023[10]) restricted their extension of the Belgian SUTs by exporter status to manufacturing industries so as to reduce the workload of the exercise.

The number of industries to be disaggregated also depends on the level of industry breakdown at which the disaggregation into categories of enterprises is implemented. Compilers face a trade-off when choosing the level of breakdown. On the one hand, a more detailed industry classification implies that there are fewer enterprises per category within industries. This may lead to the confidentiality issues discussed above. On the other hand, a more aggregate industry classification reduces within-industry homogeneity in terms of activity, and it is likely that categories of enterprises within more aggregate industries are also less homogenous. Hence, when working with more aggregate industries, compilers must carefully evaluate whether product similarity in more detailed industries or similarity in terms of size, ownership, exporter status or any other alternative criterion matters more for technological homogeneity.

Example for Belgium

A list of industries to be disaggregated was drawn up in the early stages of the construction process of ownership-extended SUTs for Belgium for 2015 (Hambÿe et al., 2022[11]) (Table 3.2). It was decided not to disaggregate industries for which there is no significant activity in Belgium, e.g. certain types of mining; industries for which there is no information about significant ownership links with abroad, e.g. agriculture and certain non-market service activities; industries that belong entirely to the public sector or that have a large public sector share, such as the public administration, defence or education; and financial and insurance industries, for which the relevant enterprise-level data for an industry disaggregation were not available. From the 133 industries in the regular Belgian SUTs, 110 were disaggregated.

Table 3.2. Industries to be disaggregated in Belgium according to ownership status

Section of the ISIC Rev. 4	ISIC 2-digit division	Breakdown by ownership?
Agriculture, forestry and fishing (A)	01-03
Mining and quarrying (B)	05-09	✓
Manufacturing (C)	10-33	✓
Electricity, gas, steam and air conditioning (D)	35	✓
Water supply; sewerage, waste management and remediation services (E)	36-39	✓
Construction (F)	41-43	✓
Wholesale and retail trade services; repair services of motor vehicles and motorcycles (G)	45-47	✓
Transportation and storage (H)	49-53	✓
Accommodation and food services (I)	55-56	✓
Information and communication services (J)	58-63	✓
Financial and insurance activities (K)	64-66
Real estate services (L) (except for imputed rents for owner-occupiers)	68	✓
Professional, scientific and technical services (M)	69-75	✓
Administrative and support services (N)	77-82	✓
Public administration and defence; compulsory social security (O)	84
Education (P)	85
Human health and social work activities (Q)	86-88
Arts, entertainment and recreation (R)	90-93
Other service activities (S)	94-96
Activities of households as employers (T)	97

Source: Hambÿe et al. (2022[11]).

Preparing the data

Before proceeding with the subsequent steps in the top-down method of constructing ESUTs along the lines of Table 3.1, compilers should prepare and perform controls on the data they are planning to use. This involves both the regular SUTs that serve as a starting point and the microdata that they have at their disposal, in particular the business register with the categorisation of enterprises by type and the enterprise-level data. Such controls are elaborated below.

Microdata linking

Once the ESUT compiler has chosen the industries to disaggregate and classified enterprises in the business register according to the extension criterion, the information on the category of the enterprises must be introduced into the enterprise-level databases. This is necessary in preparation for the use of these databases for the disaggregation of columns and rows in the regular SUTs. In practice, this can be achieved for each enterprise-level database by merging in the categorical variable from the business register based on a common identifier of enterprises. However, the ideal situation is one of complete microdata linking, i.e. where, based on a unique enterprise identifier, all the available enterprise-level data are merged into the business register that contains information on the categories of enterprises.

A first check should be performed on the business register extended with a classification of enterprises by category to determine how many enterprises there are for each category in each industry. In a similar vein, compilers should check the distribution over enterprise categories of the relevant variables from the enterprise-level data. This concerns, for example, turnover or sales from business surveys or tax records, exports, and imports of goods from customs data or employment from social security records.

Sample size problems

The controls may reveal sample size problems. For example, in some industries, there may be only very few foreign-owned enterprises, maybe only one or two, with none in the sample of a specific survey. Besides giving rise to confidentiality issues, this is not an ideal basis for extrapolations to estimate industry-level totals by enterprise category, and it may also lead to problems of reliability of the results.

Compilers may consider the following approaches to address such sample size problems:

• Rely on information from other sources with a larger coverage for extrapolation.

• Perform the breakdown by enterprise category at a more aggregate industry level (accepting the caveats on the homogeneity of production processes mentioned earlier).

• Proceed with the disaggregation at the most detailed industry level despite limited sample sizes but publish results only at a more aggregate level.

However, the choice should also be made with the use of the data in mind. When the use of the data is not related to the problem at hand, there will be more solutions. For example, suppose that the main goal is not to have information about foreign-owned multinational enterprises (MNEs), but about domestically owned non-multinationals, and the foreign-owned MNEs are just one of the other remaining groups of enterprises. In that particular case, these remaining groups of enterprises might be grouped together for this particular industry. If the use of the data is related to the problem at hand, for example, when the goal is to have information about foreign-owned MNEs, then one has to estimate anyway, even though the sample size is small.

When sample size is small, an alternative is Statistics Netherlands’ approach (Chong et al., 2016[6]). They compiled ESUTs where industries are split by enterprise size, using data at enterprise level from the annual business surveys. In their approach, the industries of the regular SUTs are split into even smaller parts. In total, there were 1 068 clusters/industry x size class combinations. For 77 of those clusters, there were no data in the survey sample whereas there were enterprises in the cluster. Yet it was necessary to estimate several variables for each cluster such as value added. This was accomplished in the following way, which is similar to the method described earlier in this section.

The data situation was as follows: for enterprises in the survey sample, value added was known. For each enterprise in the Netherlands, an estimate of the number of persons employed was known. First, calculate for each industry-size cluster the average of value added by person employed. Information from enterprises where this ratio is very unusual was not taken into account. Second, in a cluster with no or insufficient survey responses, use the information from the previous step of an adjacent cluster and the employment of the enterprises in the cluster itself to estimate value added for each enterprise in the cluster. Third, calculate the totals at industry level and rescale them to the regular industry totals from the annual business surveys.

Differences between microdata and national accounts

As mentioned earlier, the top-down method for constructing ESUTs takes the regular SUTs as a starting point. It relies on microdata/enterprise-level data for the disaggregation of the columns and rows according to the chosen criterion. A major caveat of this approach concerns differences between microdata and the national accounts (NA).

In general, NSOs construct NA and SUTs from available microdata (and other more aggregate data sources). This includes enterprise-level data such as the business surveys underlying (structural) business statistics or customs data underlying international trade in goods statistics. There are, however, differences between aggregates derived directly from these raw microdata/enterprise-level data and NA aggregates (or values in the SUTs). These differences are primarily due to the accounting principles of the System of National Accounts (SNA). They can be substantial. For example, Cai, Miroudot and Zürcher (2023[12]) report that, in the OECD’s AMNE data that are directly derived from enterprise-level data, output of the manufacture of chemicals and pharmaceutical products in the United Kingdom amounted to USD 46 billion in 2016, against output of USD 76 billion for that industry in the country’s IOT. Such differences are an issue for the top-down method of compiling ESUTs where columns and rows in the regular SUTs are disaggregated based on microdata. Larger differences between NA and source data should be investigated and, where possible, assigned to the correct enterprise category. Remaining minor corrections can be distributed proportionally.

There are many aspects where the rules of the SNA for the compilation of the NA and SUTs lead to changes compared to the underlying microdata. A few of the major ones are listed below (see also the relevant discussion of this point in Cadestin et al. (2018[13])):

• Universe of enterprises/activities: For the compilation of the NA, all enterprises and all activities must be considered, even illegal activities, e.g. smuggling, that are not reported in administrative data and standard surveys.

• Statistical unit: The SNA recommends compiling the NA and SUTs using information on kind-of-activity units/establishments as statistical units, while survey and administrative data are often collected at the level of legal units/enterprises. Now consider an enterprise in the automotive industry which encompasses a wide array of establishments in different industries. In the NA, production is recommended to be assigned to the various industries of the establishments. But in business statistics, it would be assigned to the automotive industry only. Whether the data are at establishment or enterprise level therefore influences the results at the industry level.

• Supply table in NA data in basic prices while business statistics data are in producers’ prices: The total output of an enterprise in business statistics contains the value of its production but may also include taxes minus subsidies on the product, trade and transport margins. However, in the supply table in the NA data, these taxes and margins are removed to arrive at a basic price. The margins are reallocated, namely to wholesale and retail trade and transport sectors and taxes minus subsidies on production included.

• Definition of output vs. turnover/sales: Turnover or sales reported by enterprises in administrative data and surveys underlying business statistics reflects the revenue of these enterprises and differs from output as defined in the NA. For example, when goods are purchased for resale without transformation (retail and wholesale activities), only the margin related to this resale activity may be considered as output in the NA. On the other hand, output in the NA should include the production of goods even when these are not sold.

• Secondary production: Secondary production can be reassigned to another industry by the NA in the SUTs.

• International trade in goods (ITG): There are significant differences between ITG and NA when it concerns trade in goods. First, according to the SNA (which is the same as the Balance of Payments Manual in this respect), the NA should only record imports and exports of goods when there is transfer of economic ownership from a non-resident to a resident, and vice versa, whereas in the customs data underlying merchandise trade statistics, goods are considered to be imported or exported when they enter or leave the economic territory (United Nations, 2011[14]), i.e. “when they cross the border”.2 The transfer of ownership does not always coincide with the crossing of a border, e.g. in the case of transit trade or when goods are sent abroad for processing or repair. Second, in the NA, exports and imports of goods are assigned to the industries that respectively produce or consume these goods, whereas merchandise trade statistics assign imports and exports to the industry of the enterprise that reports the trade. This can lead to large differences, in particular for wholesale trade. When a country has already compiled a reconciliation table between trade and SNA/Balance of Payments, this table would provide valuable input for dealing with the differences. Third, NA matches all trade to an industry, but not all trade in ITG can be matched to enterprises and subsequently to industries. This might be due to general mismatching or to the fact that the trader is unknown, e.g. because of a reporting threshold. Another explanation is that not all traders can be linked to the general business register, because they are only fiscally, but not physically, present in the country. Generally, it is advised to use as much of the microdata as possible to take the differences into account. Particularly for parts that are expected to be distributed differently among different types of enterprises, such as production abroad, processing, merchanting and re-exports. Annex 3.A provides guidance on dealing with re-exports at the micro‑level.

• Balancing of SUTs: The compilation of SUTs brings together data from many sources in a single framework. Ensuring consistency between data from various sources within this framework requires balancing, which leads to changes compared to the original data.

In certain cases, a National Statistical Office (NSO) introduces some of the changes required to conform to SNA rules directly into the microdata/enterprise-level data. For example, some of the corrections for imports and exports required by the SNA may be applied at the level of individual enterprises, e.g. the identification of transit and processing trade. The resulting adapted microdata are then more closely in line with NA aggregates/SUTs totals. If they can get access, ESUT compilers should use those adapted microdata as input for the disaggregation of industries and product categories. However, most of the corrections for conforming to SNA rules are introduced at a more aggregated level and are, therefore, not reflected in the microdata/enterprise-level data used by ESUT compilers. In particular, extrapolations are typically performed at the industry level. Naturally, the same holds for the balancing of supply and use of products as part of SUTs compilation.

ESUT compilers may try to replicate some of these corrections, e.g. extrapolations, at the level of enterprise categories within industries. For this purpose, they may rely on certain assumptions and additional data that are used in the construction of the NA and regular SUTs. Chong et al. (2018[5]) and Sallusti and Cuiccho (2023[3]) provide some examples of such assumptions and use of additional data:

• Incomplete coverage within an industry: If sources cover only part of an industry’s population of enterprises, then the missing part of activity must be estimated. This may require specific assumptions in ESUT compilation depending on the data situation. For example, if smaller size classes in holdings or group services are not covered by structural business surveys, compilers may allocate this estimated part of the industry-level activity to SMEs. Similarly, own account production activities like homebuilding or growing vegetables in allotments can be allocated to SMEs, non-multinationals and non-traders.

• Undeclared work and illegal economy: Production and costs for undeclared work and illegal activities are estimated and included in the NA. If they have access to these estimates, ESUT compilers may decide to allocate these to SMEs if they think it is appropriate. Adjusting can also be achieved during the balancing by product and with information from the labour surveys.

• Research and development (R&D) and intangible assets: Expenditure on R&D and intangible assets like software or intellectual property rights, whether purchased or internally produced, is considered as gross capital formation (investment) in the NA but is often recorded as operating costs in enterprise surveys. Therefore, NSOs apply a correction for these items when compiling the NA. If ESUT compilers have access to the survey data underlying this correction, they can allocate such expenditure to enterprise categories within industries. Otherwise, such expenditure may be allocated based on total use of the different enterprise categories within each industry.

• Financial intermediation services indirectly measured: Banks often do not charge directly for loans or deposits, instead earning through interest margins. NA make this implicit financial service on loans and deposits explicit based on interest rates. Industry and final use allocations are distributed based on production as per international agreements.

• In-kind compensation: In-kind compensation concerns goods and services provided by employers, like company cars or meals. These are recorded as operating costs but must be registered as wages in NA. Adjustments for private use of business assets by self-employed individuals are distributed proportionally.

• Expense fraud: Private expenses wrongly recorded as business expenses, like personal fuel or hotel stays, should be classified as consumption in NA. If the category of enterprises in which this type of fraud is mainly found is known, this information can be used to distribute corrections.

Nonetheless, the microdata that are used for disaggregation will generally not be fully in line with the totals in the regular SUTs. Therefore, the underlying idea of the top-down method is to derive shares from the microdata and then use proportional methods, e.g. a GRAS3 method, to match the constraints imposed by the regular SUTs. Compilers can choose to take only industry- and product-level aggregates or individual cells from regular SUTs as constraints. However, taking individual cells as constraints is much more restrictive and may be too restrictive in some cases.

In one country’s experience, there were some sources about production or compensation of employees for very small enterprises and the underground economy but no sources about costs, imports and exports. Some components were estimated by industry and type of enterprise. However, the proportionality assumption for costs, imports and exports sometimes leads to inconsistencies with production (and compensation of employees), thus generating negative entries for value added and gross operating surplus. The NSO’s solution was to assume that for problematic cells, the distribution of compensation of employees and total costs was like the distribution of production.

Step 2: Disaggregating total industry-level output and intermediate use

In this second step, total output and intermediate inputs by industry are disaggregated by type of enterprise. Industry-level gross value added by type of enterprise is derived by difference. The rows for “total output” and “total intermediate use” in the SUTs in Table 3.3 and Table 3.4 illustrate the starting point and the result of this step. It should be noted that this is an intermediate step in the disaggregation of the columns of the regular SUTs that can be implemented but need not necessarily be implemented. Compilers may decide to skip this step and proceed directly to Step 3 in Table 3.1 if they consider their data for implementing Step 3 are sufficient. They can then simply derive total output and total intermediate use for all industry-enterprise category combinations as the sum of the columns in the column-extended SUTs (Table 3.3 and Table 3.4).

In the NA and regular SUTs, output is estimated for most industries from turnover or sales, and total intermediate use is estimated from purchases. Enterprises report these variables in business surveys or administrative data such as annual accounts or tax data and this information naturally constitutes the ideal enterprise-level data for disaggregating these industry totals by enterprise category.

However, the information on turnover, sales and purchases in the enterprise-level data might not be available for all enterprises in an industry. Hence, the enterprise-level data will not cover all the industry’s total output and intermediate use in the regular SUTs. In that case, the disaggregation of total output and intermediate use by enterprise category can be achieved either by using the shares of the enterprise categories in turnover or sales and in purchases derived from the enterprise-level data or by using employment data for extrapolation if employment is available for the entire population of enterprises in the industry. This latter approach may also be applied for the disaggregation of total output and intermediate use of industries for which compilers have no microdata. For example, see the approach of Chong et al. (2018[5]) that was mentioned earlier in the chapter.

Step 3: Disaggregating the product distribution of output and intermediate use

In this third step of the top-down approach to ESUT compilation, the product structure of output and use of industries is disaggregated by type of enterprise. This comes down to disaggregating the columns of the SUTs (intermediate part). Table 3.3 and Table 3.4 illustrate, in reduced form, the starting point (regular SUTs) and the result (column-extended tables) for the supply table and the use table, respectively. They show a disaggregation into SMEs and large enterprises (large) as an example.

Table 3.3. Column-wise disaggregation of the supply table

Starting point: Regular supply table

	Industry 1		Industry 2		Total output		Total imports
Product 1	p1,1		p1,2		PP1		PM1
Product 2	p2,1		p2,2		PP2		PM2
Total output	IP1		IP2
Result: Extended supply table
	Industry 1			Industry 2		Total output		Total imports
	SME	Large		SME	Large
Product 1	p1,1,s	p1,1,l		p1,2,s	p1,2,l	PP1		PM1
Product 2	p2,1,s	p2,1,l		p2,2,s	p2,2,l	PP2		PM2
Total output	IP1,s	IP1,l		IP2,s	IP2,l

Note: SME: small and medium-sized enterprise.

Table 3.4. Column-wise disaggregation of the use table

Starting point: Regular use table at purchasers’ prices

		Industry 1			Industry 2		Total intermediate use		Final use
Product 1		u1,1			u1,2		PU1
Product 2		u2,1			u2,2		PU2
Total intermediate use		IU1			IU2
Result: Extended use table at purchasers’ prices
	Industry 1			Industry 2				Total intermediate use	Final use
	SME		Large	SME		Large
Product 1	u1,1,s		u1,1,l	u1,2,s		u1,2,l		PU1
Product 2	u2,1,s		u2,1,l	u2,2,s		u2,2,l		PU2
Total intermediate use	IU1,s		IU1,l	IU2,s		IU2,l

Note: SME: small and medium-sized enterprise.

In the construction of the regular SUTs, NSOs tend to use several enterprise-level data sources to estimate the product detail of industry-level output and intermediate use. Among these sources, there may be:

• surveys of turnover or sales by product type and of purchases for intermediate use by product type (for an example of questionnaires of such surveys, see the annexes to Chapters 5 and 6 in United Nations (2018[8]))

• a survey of industrial production with detail by product type

• customs data on exports and imports of goods (data from Intrastat and Extrastat for EU member countries)

• surveys of exports and imports of services

• administrative records of transactions under value added tax (VAT transaction data)

• other enterprise-level data sources for specific activities/industries (e.g. surveys of agricultural production) or for certain types of activities/products (e.g. surveys of R&D expenditure).

To avoid proportionality assumptions, ESUT compilers should use as many of these sources as possible. This allows them to determine, for each enterprise category in each industry that they disaggregate, a different distribution of turnover/sales by product category and a different distribution of purchases for intermediate consumption by product category, as appropriate. Combining such enterprise-level data from various sources requires correcting for inconsistencies between these sources. For example, the value of an enterprise’s exports of a good obtained from customs data may exceed the value of its production of that good reported in a survey of industrial production. Additional adjustments become necessary, for example, when the samples covered by these sources differ. These issues may already have been dealt with in the compilation process of regular SUTs. Hence, it is a definite advantage if ESUT compilers can rely on the adapted enterprise-level data that have been used for regular SUT construction.

The resulting product distribution of turnover/sales and that of purchases of intermediates – both based on the enterprise-level data sources that are available to the ESUT compiler – can serve as first estimates for the vectors of output and intermediate use of the industry-enterprise category combinations in the ESUT. For each industry, these first estimates must be adjusted to the column and row constraints which are, respectively, the totals of output and intermediate use of the enterprise categories in that industry derived in Step 2 (see above), and the industry-level product distribution (column vector) of output and intermediate use from the regular SUTs (see columns for industries in the regular supply table in Table 3.3 and in the regular use table in Table 3.4). These adjustments can be made based on proportional methods such as GRAS.

The results of this step are column-wise extended SUTs at purchasers’ prices, i.e. SUTs in which the columns for all or a selection of industries have been disaggregated by enterprise category (see the result tables of Table 3.3 and Table 3.4).

The next steps are not strictly needed to compile the ESUT framework. However, they are very useful for the EIOT, as was explained after Table 3.1.

Step 4: Deriving extended valuation tables and an extended use table at basic prices

Step 4 consists of deriving a column-extended use table at basic prices. Following the recommendation on the compilation of regular SUTs at basic prices in the Handbook on Supply and Use Tables and Input-Output Tables with Extensions and Applications (United Nations, 2018[8]), a use table at basic prices is obtained by subtracting use-side valuation tables from the use table at purchasers’ prices. These use-side valuation tables comprise information by industry and product category on trade margins, transport margins, taxes on products including non-deductible VAT, and subsidies on products. This information allows bridging between different valuation concepts in the supply and use tables framework. To derive ESUTs at basic prices, it is necessary to extend the valuation tables by enterprise category for the industries that have been selected for disaggregation.

For trade and transport margins, the usual approach to estimating use-side valuation tables for regular SUTs consists of three steps:

1. Estimate the production of margins by industry – these margins originate from the enterprises’ resale of goods that they purchase but do not transform (trade margin) and from the enterprises’ separately invoiced shipping to customers of goods they sell (transport margin).

2. Distribute these supply-side margins over product categories and then sum by product category to obtain a vector of product-level margins.

3. Allocate the total product-level margins to users of the products in the use table.

This approach is illustrated in the Handbook on Supply and Use Tables and Input-Output Tables with Extensions and Applications (see Box 7.1 in United Nations (2018[8]). The estimation of the production of margins by industry is generally based on enterprise-level data from business surveys and surveys of industrial production on purchases for resale and transportation invoiced separately. These data sources may also contain information that is relevant for the distribution of margins over product categories. Since data for the allocation to users are rarely available, this allocation is mostly based on assumptions.

For taxes and subsidies on products, the approach to estimating valuation tables is similar to the approach for margins but shorter as it comprises only two steps: 1) estimate the value of these taxes and subsidies by product category; and 2) allocate these product-level totals to users of the products in the use table. The estimation of the amounts of the taxes and subsidies on products and their allocation to specific product categories in the SUTs are usually based on government revenue accounts and assumptions. For the allocation of these product-level totals to users, information from specific tax legislation is combined with assumptions. The approach for non-deductible VAT is specific and so are the data used for this purpose. The overwhelming part of non-deductible VAT is recorded as being levied on household final consumption expenditure.

In the top-down method for ESUT compilation, the valuation tables of the regular SUTs are the starting point for deriving extended valuation tables. The columns of the industries selected for disaggregation must then be disaggregated in these regular valuation tables. As can be easily understood from the description above of the construction of regular use-side valuation tables, there is limited enterprise-level data involved in the allocation of supply-side margins and taxes and subsidies on products to users, i.e. to the entries in the use table. Hence, there is little information that ESUT compilers could use for disaggregating the industry columns in the regular valuation tables. In these circumstances, the best they can do is reproduce and extend the assumptions made in the allocation of margins and taxes and subsidies on products to users in the regular valuation tables. For taxes and subsidies on products, they may also re-examine the relevant legislation for a specific allocation, for example, the use of VAT rates. However, for most industries, it is unlikely that the results will be substantially different from a disaggregation of the industry column in proportion to the use of the enterprise categories. Therefore, ESUT compilers may choose such a use-based proportional disaggregation by default. This is what has been done in work on ESUTs by exporter status for Belgium (Michel, Hambÿe and Hertveldt, 2023[10]).

However, sometimes information is available to do more than proportional distributions. For margins, this could be knowledge about the distribution channels relevant for each cell in the use table or even survey data from wholesalers and retailers covering the product sales and purchases of goods for resale with further processing. With information about the share of a product flowing through each relevant distribution channel, an effective margin rate can be estimated for each cell. In the case of the extended table, this process can be carried out independently for each enterprise type within each industry, which allows information about inter-industry relationships specific to a given enterprise type to be leveraged during the estimation process. For example, if it is known that most foreign MNEs operating within a certain wholesale category primarily serve manufacturing units also within the foreign MNE enterprise type, then this information should be used to direct the allocation of these margins to industries within the specified enterprise type. This information must be constrained to the control totals from the regular SUTs. For taxes and subsidies, the ability to distinguish them by underlying legislation may be helpful to the extent that different taxes or subsidies are applicable in different ways to different enterprise types. In that case, taxes and subsidies may be allocated to specific enterprise types as appropriate.

Once the use-side valuation tables have been extended column-wise, they can be subtracted from the column-extended use table at purchasers’ prices (Table 3.4) to obtain a column-extended use table at basic prices.

Step 5: Disaggregating use into use of imports and of domestic production

The next step is to split the column-extended use table at basic prices into a column-extended use table of domestic production (domestic use table) and a column-extended use table of imports (import use table), both also at basic prices. The way of proceeding is to first compile the extended use table of imports from the use table of imports in the regular SUT by disaggregating the columns of the industries selected for disaggregation, then derive the extended domestic use table by subtracting the extended import use table from the extended use table at basic prices obtained in the previous step (Step 4) of this top-down method of ESUT compilation. Table 3.5 and Table 3.6 illustrate the starting points and results of these operations.

Table 3.5. Disaggregation of the import use table

Starting point: Regular import use table

		Industry 1		Industry 2			Total imports
Product 1		m1,1		m1,2			PM1
Product 2		m2,1		m2,2			PM2
Total imports		IM1		IM2
Result: Extended import use table
	Industry 1				Industry 2			Total imports
	SME		Large		SME	Large
Product 1	m1,1,s		m1,1,l		m1,2,s	m1,2,l		PM1
Product 2	m2,1,s		m2,1,l		m2,2,s	m2,2,l		PM2
Total imports	IM1,s		IM1,l		IM2,s	IM2,l

Note: SME: small and medium-sized enterprise.

Table 3.6. Derivation of the domestic use table

Starting point: Extended use table at basic prices

		Industry 1				Industry 2				Total Intermediate use		Final use
		SME		Large		SME		Large
Product 1		p1,1,s		p1,1,l		p1,2,s		p1,2,l		PU1
Product 2		p2,1,s		p2,1,l		p2,2,s		p2,2,l		PU2
Total intermediate use		IU1,s		IU1,l		IU2,s		IU2,l
Result: Extended domestic use table
			Industry 1				Industry 2				Total Intermediate use	Final use
			SME		Large		SME		Large
Domestically produced	Product 1		d1,1,s		d1,1,l		d1,2,s		d1,2,l		PD1
	Product 2		d2,1,s		d2,1,l		d2,2,s		d2,2,l		PD2
	Total intermediate use of domestic output		ID1,s		ID1,l		ID2,s		ID2,l
	Total imports for intermediate use		IM1,s		IM1,l		IM2,s		IM2,l
	Total intermediate use		IU1,s		IU1,l		IU2,s		IU2,l

Note: SME: small and medium-sized enterprise.

In the compilation of import use tables for regular SUTs, NSOs distribute the imports of each product category (from the supply table) over intermediate and final use, i.e. they determine how much of the consumption of a product category by industries and final users is imported. Ideally, they do so based on import data by enterprise and product category – for both goods and services – that are corrected to conform to SNA rules (corrections for transit trade, merchanting, processing trade, etc.; see the discussion on the preparation of enterprise-level data above). These data allow for an estimation of industry-level imports for each product category. Nonetheless, the compilation of the import use table generally also involves some form of proportional allocation, for example, for allocating imports to categories of final use and ensuring that the row totals in the import use table match imports by product category from the supply table. A central issue in the compilation method based on enterprise-level import data is that any entry in the import use table may not exceed the corresponding entry in the use table at basic prices.

For the construction of the extended import use table according to the top-down method, the ideal situation is that compilers have access to the corrected enterprise-level import data that have been used in the construction of the regular import use tables. This enables them to disaggregate the industry’s use of imported products by enterprise category and to restrict proportional disaggregations to entries in the import use table for which no underlying data are available. Ideally, there is also information available about re-exports (see also Annex 3.A) and production abroad4 if they are relevant trade flows for the country under concern. It is recommended to place such imports and exports that are not related to production in the domestic economy apart. This procedure has two types of constraints: 1) the imports of a product by an industry-enterprise-category combination may not exceed total intermediate use of that product by that industry-enterprise-category combination, i.e. the value of an entry in the extended import use table may not exceed the value of the corresponding entry in the extended use table at basic prices; and 2) the sum of the imports of a product by all enterprise categories in an industry may not exceed the industry’s value of imports of that product in the regular import use table. These constraints may turn out to be conflicting, which requires compilers to check the plausibility of their disaggregation of the use table and/or to release one of the constraints. Hambÿe, Michel and Trachez (2023[15]) provide an example of the construction of an extended import use table along these lines in their work on an ESUTs by group affiliation for Belgium for the year 2019.

There are sometimes large differences between NA and ITG when it concerns imports and exports by industry. As explained earlier in this chapter, ITG might see more imports and exports by wholesalers and transporters whereas NA might see more imports and exports by other industries. Clearly, this is the case when these industries trade significantly according to ITG but the allocations need to be reviewed for coherence and plausibility.

Chong et al. (2016[6]) describe one way for dealing with this. They assume that exports of goods from industries other than wholesale, transport and storage are produced by the industries themselves. Exports of domestically produced goods by the wholesale, transport and storage industries are mostly produced by other industries and tend to be very small. These exports have, just as the exports that cannot be matched to enterprises, to be assigned to the producing industries, by enterprise type. Similarly, many imports of goods by wholesale, transport and storage must be assigned to other industries and to final use. A way to do this for imports is described below; the method for exports is similar.

1. Comparison of imports (in ITG) and use (in SUTs) in each industry by enterprise type:

a. If use exceeds imports, the assumption is that all imports by this industry by enterprise type are being used by themselves.

b. If imports exceed use, the assumption is that all imports by the industry by enterprise type are being used by that industry and that the additional imports are used by other industry by enterprise types. This yields a remainder of imports that needs to be redistributed over industry by enterprise types where use exceeds imports.

2. Remaining use by industry by enterprise type is calculated, i.e. the part of use that is not imported.

3. The remaining imports from Step 2 and unassigned trade are proportionally redistributed over industry by enterprise types with use exceeding imports based on their remaining use as calculated in Step 2.

4. Total imports by industry by enterprise type are calculated by adding imports from Steps 1 and 3.

This is illustrated in Table 3.7. For simplicity, the enterprise category is not taken into account.

Table 3.7. Example of the distribution of imports of a product by destination industry

	Imports in ITG	Use in SUT	Step 1	Step 2	Step 3	Step 4
Furniture industry	20	40	20	20	5	25
Wholesale	40	10	10	0	0	10
Household consumption	0	20	0	20	5	5
Subtotal (excluding re-exports)	60	70		40
Re-exports	0	20	20	0		20
Total	60	90	50	40	0	60
Redistributed	-	-	10	-	-	-

Note: ITG: international trade in goods; SUT: supply and use table.

Source: Chong et al. (2016[6]).

One alternative was to distribute the excess imports proportionally by total use of the product by industry by enterprise type or to distribute proportionally by imports that are already distributed. However, it is thought that the method of Chong et al. is superior. For example, it assigns imports to enterprise types that have relatively few imports in the linked business-trade data. Since larger traders usually link well to enterprises, due to the efforts of the NSOs, the larger traders are overrepresented. Another alternative is to use a mix of the method of Chong et al. and a proportionality method.

If they do not have access to relevant enterprise-level import data, compilers may rely on a proportional attribution of import use to enterprise categories, although this fails to reflect the fact that certain categories of enterprises, e.g. large ones or affiliates of MNEs, tend to import a larger share of the intermediates they use.

The resulting extended import use table must then be subtracted from the extended use table at basic prices to obtain the extended domestic use table. All these tables are extended column-wise (see Tables 3.5 and 3.6). In line with the recommendations in the Handbook on Supply and Use Tables and Input-Output Tables with Extensions and Applications (United Nations, 2018[8]), the extended domestic use table comprises a row reporting total imports by enterprise categories within industries and by final users.

Step 6: Disaggregating the use of domestic output

In this final step, one can use two approaches. The first is to disaggregate domestic output by producing enterprise category. This entails compiling fully extended SUTs, where fully extended means a column-wise and row-wise disaggregation by enterprise category. After Step 5, there is only a column-wise disaggregation by enterprise category. The second approach is to disaggregate domestic output by producing enterprise category and industry. This entails compiling an extended supply table for domestic use table by removing supply for exports. Both approaches will provide the necessary data to compile an EIOT. The second approach is theoretically more appealing, since it uses the additional information which type of enterprise in which industry is exporting which product. The first approach does not use the industry detail. However, the second approach needs more data and it might put some constraints on the data in practice. It could be a worthwhile avenue when the export data in the NA and ITG do not differ too much by product and industry. When each product is almost completely produced by one industry (in other words, there is not much secondary production), both approaches will lead to similar results.

Disaggregating by producing enterprise category: Fully extended SUTs

In this approach, the rows of the tables – relating to the product categories of domestic production – are disaggregated. For the supply table, this requires no further data or calculations and comes down to a simple expansion of the column-extended supply table derived in Step 3. For the use table, this consists of the disaggregation of the rows for product categories in the column-extended domestic use table at basic prices derived in Step 5.

Fully extended supply table

The fully extended supply table can be directly derived from the column-extended supply table as illustrated in Table 3.8. The underlying disaggregation of the rows concerns only domestic production, not imports. Hence, the table is restricted to domestic production. The starting point is the column-extended supply table that was the result of Step 3 (see Table 3.3), in which the industry columns are disaggregated by enterprise category. Without any further calculations, this table can be rewritten as a fully extended supply table in which the product categories are also disaggregated by enterprise category. For any product (or row) in the column-extended supply table, entries indicate by which type of enterprise the product has been produced. So, the rows can simply be expanded to show the producing enterprise category, e.g. SME and large enterprises as in Table 3.8 (result table). In the rows for an enterprise category (e.g. SME), the entries in the columns corresponding to other enterprise categories (large enterprises) must be filled with zeros. Summing along the rows yields total (domestic) output for product categories by enterprise type.

Table 3.8. Disaggregating the rows in the supply table

Starting point: Column-extended supply table (make table)

		Industry 1			Industry 2				Total output
		SME		Large	SME			Large
Product 1		p1,1,s		p1,1,l	p1,2,s			p1,2,l	PP1
Product 2		p2,1,s		p2,1,l	p2,2,s			p2,2,l	PP2
Total output		IP1,s		IP1,l	IP2,s			IP2,l
Result: Fully extended supply table (make table)
		Industry 1				Industry 2			Total output
		SME	Large			SME	Large
Product 1	SME	p1,1,s	0			p1,2,s	0		PP1,s
	Large	0	p1,1,l			0	p1,2,l		PP1,l
Product 2	SME	p2,1,s	0			p2,2,s	0		PP2,s
	Large	0	p2,1,l			0	p2,2,l		PP2,l
Total output		IP1,s	IP1,l			IP2,s	IP2,l

Note: SME: small and medium-sized enterprise.

Fully extended use table

The domestic use table resulting from Step 5 of the top-down method of ESUT compilation is extended column-wise (see Table 3.6). In other words, the information in the rows of this table on the use of products, which comes from domestic production, has not yet been disaggregated in terms of the producing enterprise category. This needs to be done to obtain a fully extended use table (see result table in Table 3.9). Determining which type of enterprise has produced the goods and services purchased for intermediate and final use is a complex task. ESUT compilers may rely on various sources of information to accomplish this.

First, the proportions of enterprise categories in total domestic output of each product category are known from the fully extended supply table “total output” column in the result table of Table 3.8). Any proportional disaggregation of the rows for product categories in the column-extended domestic use table should be based on these proportions. With such a proportional disaggregation, the row totals in the fully extended domestic use table automatically respect the values of total domestic output of each product by enterprise category (“total output” column in the fully extended use table of Table 3.9 matching the “total output” column in the fully extended supply table of Table 3.8). When compilers use additional enterprise-level data for a non-proportional disaggregation of the rows in the column-extended use table, the values of total domestic output of products by enterprise category from the fully extended supply table become a constraint in terms of the row totals for the resulting fully extended use table (last column with “total output” in the fully extended domestic use table in Table 3.9).

Second, for disaggregating exports of products by producing enterprise category, compilers should ideally rely on enterprise-level export data that are corrected to conform to SNA rules and that contain information on exported products (goods and services). This is of relevance since it has been shown that certain enterprise categories are more export-oriented than others, e.g. affiliates of MNEs compared to domestic enterprises. The way of proceeding is to cross-tabulate these enterprise-level export data by product and enterprise category to calculate shares that can be used to distribute total product-level exports in the domestic use table over enterprise categories. The result is a disaggregation of the rows in the export column of the domestic use table (see the column “exports” in Table 3.9) by enterprise category.

Third, enterprise-level data for disaggregating the remaining parts of the rows in the column-extended domestic use table are harder to come by. Such disaggregations require data that record domestic transactions between enterprises and between enterprises and final consumers. This is the case in data on transactions under the VAT regime, so-called VAT transaction data. If available, such data may indeed be used for disaggregating the rows of the column-extended domestic use table, except for the entries for exports. However, such data are not collected in many countries, and it may prove difficult for ESUT compilers to get access to these data. So far, there is no example of ESUT compilation in which VAT transaction data have been used for the purpose of disaggregating the rows of the use table by enterprise category. Clearly, this is a possibility that should be further explored. Potential issues in the use of VAT transaction data for row disaggregation includes limited coverage, particularly for transactions with final consumers, and the identification of the products that are delivered as the data record transactions between enterprises or enterprises and final consumers. In the absence of such enterprise-level data, compilers must rely on a proportional disaggregation of the rows as described above.

Table 3.9. Disaggregating the rows in the domestic use table

Starting point: Column-extended domestic use table

					Industry 1				Industry 2			Total intermediate use		Domestic final use		Exports
					SME		Large		SME		Large
Domestically produced		Product 1			d1,1,s		d1,1,l		d1,2,s		d1,2,l	PD1		PF1		PX1
		Product 2			d2,1,s		d2,1,l		d2,2,s		d2,2,l	PD2		PF2		PX2
		Total intermediate use of domestic output			ID1,s		ID1,l		ID2,s		ID2,l
		Total imports for intermediate use			IM1,s		IM1,l		IM2,s		IM2,l
		Total intermediate use			IU1,s		IU1,l		IU2,s		IU2,l
Result: Fully extended domestic use table
			Industry 1			Industry 2				Total intermediate use			Domestic final use		Exports	Total output
			SME	Large		SME		Large
Product 1	SME		d1,1,s,s	d1,1,s,l		d1,2,s,s		d1,2,s,l		PD1,s			PF1,s		PX1,s	PP1,s
	Large		d1,1,l,s	d1,1,l,l		d1,2,l,s		d1,2,l,l		PD1,l			PF1,l		PX1,l	PP1,l
Product 2	SME		d2,1,s,s	d2,1,s,l		d2,2,s,s		d2,2,s,l		PD2,s			PF2,s		PX2,s	PP2,s
	Large		d2,1,s	d2,1,l,l		d2,2,l,s		d2,2,l,l		PD2,l			PF2,l		PX2,l	PP2,l
Total intermediate use of domestic output			ID1,s	ID1,l		ID2,s		ID2,l
Total imports			IM1,s	IM1,l		IM2,s		IM2,l
Total intermediate use			IU1,s	IU1,l		IU2,s		IU2,l

Note: SME: small and medium-sized enterprise.

The result of this approach is fully extended SUTs at basic prices. These tables show industry-enterprise-category combinations in the columns and product-enterprise-category combinations in the rows (see result tables in Table 3.8 and Table 3.9). They may be used to derive an EIOT according to one of the methods described in the Handbook on Supply and Use Tables and Input-Output Tables with Extensions and Applications (United Nations, 2018[8]). These methods are further elaborated in Chapter 6.

Disaggregating the use of domestic output by producing enterprise category and industry

In this approach, the goal is to obtain an extended table with supply of domestically produced goods and services to be used in the domestic economy. Table 3.10 illustrates the data at start and finish.

Table 3.10. Disaggregating supply of domestic production

Starting situation

 

Extended supply table (Table 3.3)

	Industry 1						Industry 2							Total output				Total imports
	SME			Large			SME				Large
Product 1	p1,1,s			p1,1,l			p1,2,s				p1,2,l			PP1				PM1
Product 2	p2,1,s			p2,1,l			p2,2,s				p2,2,l			PP2				PM2
Total output	IP1,s			IP1,l			IP2,s				IP2,l
Regular export table
		Industry 1							Industry 2							Total exports
Product 1		X1,1							X1,2							XP1
Product 2		X2,1							X2,2							XP2
Total exports		XI1							XI2
Extended export table based on enterprise-level data
			Industry 1							Industry 2							Total exports
			SME			Large				SME			Large
Product 1			XF1,1,s			XF1,1,l				XF1,2,s			XF1,2,l				XP1
Product 2			XF2,1,s			XF2,1,l				XF2,2,s			XF2,2,l				XP2
Total exports			XF1,s			XF1,l				XF2,s			XF2,l
Result
Extended export table based on enterprise-level data, matching the regular export table
			Industry 1							Industry 2							Total exports
			SME			Large				SME			Large
Product 1			X1,1,s			X1,1,l				X1,2,s			X1,2,l				XP1
Product 2			X2,1,s			X2,1,l				X2,2,s			X2,2,l				XP2
Total exports			XI1,s			XI1,l				XI2,s			XI2,l
Extended supply table for domestic use
		Industry 1						Industry 2							Total domestic production for domestic use
		SME			Large			SME				Large
Product 1		DU1,1,s			DU1,1,l			DU1,2,s				DU1,2,l			DU1
Product 2		DU2,1,s			DU2,1,l			DU2,2,s				DU2,2,l			DU2
Total supply for domestic use		DU1,s			DU1,l			DU2,s				DU2,l

Note: SME: small and medium-sized enterprise.

For example, the element DU_1,1,s is equal to p_1,1,2 minus X_1,1,s.

In this process, it is assumed that one has a regular export table, which shows how much each industry exports each product. This table is combined with enterprise-level export data by product category. These export data are corrected for re-exports (see Annex 3.A) and ideally also for production abroad. The reason is that only domestically produced goods and services are needed. Subsequently, the enterprise-level data are aggregated by product and by industry and enterprise type. Just as in the imports case before, aggregating the results will yield something that is different from the regular export table. Therefore, the data have to be rearranged. The method that was used to rearrange the import data can be applied again.

After rearranging the data, aggregating the extended export table to an export table will still lead to a different table than the regular export table. One can choose to ignore this or force the extended flow table to add up to the regular export table by applying a GRAS procedure again. The data would have to suffice two constraints. The first is that at product level, total exports would be the same as that in the regular export table. The second is that at industry level, total exports are the same as that as in the regular export table. One can choose to put additional constraints, namely that aggregating the extended export table yields the regular export table, but that condition might be too strong. Now subtract the extended export table from the total extended supply table to obtain the extended supply table with supply of domestic goods and services to the domestic market.

The result of this step is an extended domestic supply for domestic use table. This, together with the extended supply table from Step 3, may be used to derive an EIOT according to one of the methods described in the Handbook on Supply and Use Tables and Input-Output Tables with Extensions and Applications (United Nations, 2018[8]). These methods are further elaborated in Chapter 6.

References

[9] Ahmad, N. (2023), “Accounting frameworks for global value chains”, in Ahmad, N. et al. (eds.), Challenges of Globalization in the Measurement of National Accounts, Studies in Income and Wealth, 81, University of Chicago Press, New York, NY.

[13] Cadestin, C. et al. (2018), “Multinational enterprises and global value chains: the OECD analytical AMNE database”, OECD Trade Policy Papers, No. 211, OECD Publishing, Paris, https://doi.org/10.1787/d9de288d-en.

[12] Cai, M., S. Miroudot and C. Zürcher (2023), The 2023 Edition of the OECD Analytical AMNE Database: Methodology and New Evidence on the Role of Multinational Production in Global Value Chains, OECD, Paper presented at the 29th IIOA conference in Alghero, https://www.iioa.org/conferences/29th/papers/files/4568_AAMNE2023_IIOA_paper_final.pdf.

[5] Chong, S. et al. (2018), The Role of Small and Medium-sized Enterprises in the Dutch Economy: An Analysis Using an Extended Supply and Use Table, Statistics Netherlands, The Hague/Heerlen/Bonaire, https://www.cbs.nl/-/media/_pdf/2018/34/2018ep39-the-role-of-small-and-mediumsized-enterprices-in-the-dutch-eco-1.pdf.

[6] Chong, S. et al. (2016), Een IO-tabel voor het MKB en grootbedrijf [An IO Table for SMEs and Large Companies], Statistics Netherlands, in Dutch.

[19] Fiallos, A., A. Liberatore and S. Cassimon (2024), “CIF/FOB margins: Insights on global transport and insurance costs of merchandise trade”, OECD Statistics Working Papers, No. 2024/05, OECD Publishing, Paris, https://doi.org/10.1787/469123ab-en.

[1] Hambÿe, C., B. Hertveldt and B. Michel (2018), “Value chain integration of export-oriented and domestic market manufacturing enterprises: An analysis based on a heterogeneous input-output table for Belgium”, Working Paper 11-18, Federal Planning Bureau, Brussels, https://www.plan.be/publications/publication-1826-en-value_chain_integration_of_export_oriented_domestic_market_manufacturing_firms_an_analysis_based_on_a.

[15] Hambÿe, C., B. Michel and G. Trachez (2023), Ownership and Size Extended Supply and Use and Input-output Tables for Belgium: Final Report, Eurostat Grant Agreement 101055571-2021-BE-NA, Belgian Federal Planning Bureau.

[11] Hambÿe, C. et al. (2022), “Les groupes multinationaux en Belgique: Structure et activité économique”, Working Paper 5-22, Federal Planning Bureau, Brussels, in French, https://www.plan.be/sites/default/files/documents/WP_202205_FR.pdf.

[4] INEGI (2023), Extended Supply and Use Tables: Base Year 2018, National Institute of Statistics and Geography, Mexico, https://en.www.inegi.org.mx/programas/coue/2018.

[17] Lemmers, O. and K. Wong (2019), “Distinguishing between imports for domestic use and for re-exports: A novel method illustrated for the Netherlands”, National Institute Economic Review, Vol. 249, pp. R59-R67, https://doi.org/10.1177/002795011924900115.

[10] Michel, B., C. Hambÿe and B. Hertveldt (2023), “The role of exporters and domestic producers in GVCs: Evidence for Belgium based on extended national supply-and-use tables integrated into a global multiregional input-output table”, in Ahmad, N. et al. (eds.), Challenges of Globalization in the Measurement of National Accounts, University of Chicago Press.

[7] Michel, B. et al. (2019), “Extended supply and use tables for Belgium: Where do we stand?”, Eurostat Review on National Accounts and Macroeconomic Indicators (Eurona), Vol. 2/2019, pp. 51-71.

[2] Nilsson, M., P. Rørmose Jensen and N. Holst Jensen (2018), Compilation of Extended Supply and Use Tables in Denmark and Possible Applications in Input-Output Analyses, Paper presented at the 2018 IARIW conference, http://old.iariw.org/copenhagen/jensen.pdf.

[18] OECD (2024), International Transport and Insurance Costs of Merchandise Trade (ITIC), http://www.oecd.org/en/data/datasets/international-transport-and-insurance-costs-of-merchandise-trade-itic.html.

[16] Roos, J. (2006), Identifying and Measuring Re-exports and Re-imports, Paper presented at the 7th OECD International Trade Statistics Expert Meeting, Paris, STD/NAES/TASS/ITS(2006)18.

[3] Sallusti, F. and S. Cuicchio (2023), Towards the Compilation of eSUT for Italian Economy, Paper presented at the 2023 IIOA conference, Alghero, https://www.iioa.org/conferences/29th/papers/files/4861_sallusti_cuicchio_ISTAT_eSUT.pdf.

[8] United Nations (2018), Handbook on Supply and Use Tables and Input-Output Tables with Extensions and Applications, United Nations, New York, NY, https://doi.org/10.18356/9789213582794.

[14] United Nations (2011), International Merchandise Trade Statistics: Concepts and Definitions 2010, United Nations, New York, NY, https://unstats.un.org/unsd/trade/eg-imts/IMTS%202010%20(English).pdf.

Annex 3.A. Dealing with re-exports

Re-exports are “transactions of goods which were previously imported with a change in economic ownership and then exported without any substantial transformation” (United Nations, 2018[8]). For example, imports of laptops from the People’s Republic of China via the port of Rotterdam by a Dutch merchant who subsequently exports the laptops to the European hinterland. Since re-exports are often conducted by foreign multinationals, who use a country as a conduit to reach other markets, the subject is relevant in the ESUT and EIOT framework as well. One cannot assume that each type of enterprise has the same share of re-exports in total exports for a given product. Bias in re-exports leads to bias in exports of domestic production, which leads to biased estimates of domestic dependencies from foreign markets for each type of enterprise. Since information on determining re-exports is generally scattered in the literature, this annex provides guidance.

Data sources

Generally, the data sources are the following:

• National accounts data, with imports, exports, domestic supply and domestic use at product level.

• An import use table with estimates of how much imports are used to export directly, at product level.

• Trade in goods statistics, with imports and exports at product level for each trader.

• A bridge table that links traders to enterprises in the general business register. The general business register contains information about industry, size and possibly other relevant information about enterprises.

Methods

There are various ways to delineate the export side of re-exports; see Roos (2006[16]) for an overview. Examples are:

• In a SUT framework, identifying when exports of a given product exceed the domestic production of this product.

• Using trade in goods surveys that explicitly ask whether exports are re-exports. However, it is known that the concept of re-exports is difficult for traders. This may lead to underreporting of re‑exports.

• Using information from large case units, who are in close contact with major enterprises, or information obtained via another contact with the relevant enterprises.

• Analysing product information, e.g. fruits that are typically not domestically produced in large quantities.

• Assuming that when imports are at least half of exports at product by trader level, the exports of this product by this trader are re-exports.

Note that information is available at the trader level for all the above methods, except for the first one (SUT framework). Matching traders to the general business register, adding industry and type of enterprise (e.g. small and medium-sized enterprises) yields information at the desired level. It may also allow one to match the information of the re-exporter to information from other surveys, ideally at the level of the enterprise, and if not, at the level of the industry.

Information on the import side might be useful to obtain information about the country of origin; see Lemmers and Wong (2019[17]). One needs various data to arrive at an estimate of the value of imports for re-exports, for example, transport margins and taxes less subsidies on products such as import tariffs, ideally at product level. One also needs more specific information, such as distribution margins, ideally at product level, of the enterprise involved. The latter information can be obtained from surveys. In the European Union, it would be the Structural Business Statistics.

In the ideal case, one has complete information at enterprise level. Then one can obtain the value of imports for re-exports at product level by peeling off the re-export value. Removing the transport margins, taxes less subsidies on products and the distribution margins yields an estimate of the import value. Ideally, this process takes into account that tariffs will be different across countries of origin. The OECD regularly estimates and publishes such information (OECD, 2024[18]); see Fiallos et al. (2024[19]) for the methodology.

Notes

← 1.  This category of “other” enterprises may also include enterprises which cannot be classified for the chosen criterion but belong to industries that the compiler wants to disaggregate. In that case, there will also be values for this category of “other” enterprises in industries that are disaggregated.

← 2.  Note that this is not an issue for international trade in services, since in this statistic there is always transfer of ownership.

← 3.  See Chapter 5 for a more detailed discussion of the GRAS scaling method.

← 4.  Production abroad refers to the economic activities related to the production of goods and services owned by a country’s residents that take place outside of the country’s economic territory. Following the economic ownership principle, when the residents buy something from non-residents these are imports; when they sell something to non‑residents these are exports.