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

4. Compiling extended supply and use tables and extended input-output tables in a data-scarce environment

Abstract

This chapter describes how to compile extended supply and use tables (ESUTs) and extended input-output tables (EIOTs) in a data-scarce environment. After a short introduction, it begins by explaining the method used to compile an EIOT. It then examines how to obtain shares for value added and trade to disaggregate the input-output table. The next section gives a step-by-step guide to disaggregating the input-output table into an EIOT. The chapter also considers how to compile extended supply and use tables when extended input-output and supply and use tables are available. The chapter concludes with a comparison of the compilation of an EIOT in a data-rich and in a data-scare environment.

Introduction

One might not have the time nor (access to) the data to compile an extended supply and use table (ESUT) and subsequently an extended input-output table (EIOT) using all possible microdata. This chapter explains how to proceed in such cases: how to compile an EIOT with less data or how to compile an ESUT from an EIOT. First, it points out that some of the recommendations made in Chapter 3 might be followed up anyway. Subsequently, the chapter describes a method to compile an EIOT step by step. This approach requires the share of each enterprise type in output, value added, imports and exports by industry. The chapter thus explains how to obtain that information in a robust way. Since some countries have an EIOT but not an ESUT, the chapter then details how to compile an ESUT in that specific situation. Finally, it compares the pros, cons and robustness of the data-rich and data-scarce approaches.

General considerations

Even without all the data described in Chapter 3, an EIOT can be compiled. It is still advisable to use as much information as possible, following the guidelines provided in Chapter 3. For example, it is recommended to compare the industry totals derived from enterprise data and national accounts and investigate possible large differences. Furthermore, some information might still be available about undeclared work and/or illegal activities and implicit financial services. In addition, in real estate, it is either necessary to distinguish between “imputed rents of owner-occupied dwellings” (no enterprises involved) and “renting and operating of own or leased real estate” (enterprises involved) or making no disaggregation at all.

It is always necessary to shift some imports from wholesalers (and possibly other industries such as transport) to the intermediate and final users. Similarly, some exports need to be shifted from wholesalers (and possibly other industries) to goods-producing industries. Chapter 3 explained the reason why much trade goes via wholesalers but trade statistics assign it to the wholesalers and national accounts assign it to the users and producers. Chapter 3 recommended shifting imports and exports using the method of Chong et al. (2016[1]). Since this method uses information at the product level, which in this chapter is assumed to be unavailable, the method of Cadestin et al. (2018[2]) is proposed. This method is explained later in the chapter.

A method to compile an extended input-output table

This section describes the method in OECD (2015[3]), which was later elaborated and extended in OECD (2018[4]). Assuming the desired breakdown of an industry is into small and medium-sized enterprises (SMEs) and large enterprises, the following data are needed:

• share in gross output by industry by type of enterprise

• share in value added by industry by type of enterprise

• share in imports by industry by type of enterprise

• share in exports by industry by type of enterprise

• the input-output table (IOT) that is to be broken down by type of enterprise.

Chapter 2 described possible data sources for this type of breakdown and for the other main breakdowns (by ownership or by trading status). The share of SMEs and the share of large enterprises in each industry that will be broken down must add up to 100%. The main hurdle is expected to be imports and exports of services. Compared to trade in goods, it is not so common for these to be broken down by enterprise type. In these cases, researchers have used the share of a type of enterprise in the imports (exports) of goods in the industry as a proxy for their share in total imports (exports).

The method consists of the following steps:

1. disaggregating the columns in the IOT: value added, gross output and imports by industry

2. disaggregating the columns in the IOT: domestic intermediate use

3. disaggregating the rows in the IOT: exports

4. disaggregating the rows in the IOT: domestic intermediate and final use.

The steps are explained in detail with basic numerical examples below, using the fictitious data in Table 4.1 and Table 4.2 as a starting point.

Table 4.1. Shares of small and medium-sized enterprises and large enterprises in key macroeconomic variables, by industry

	SMEs		Large enterprises
	Industry 1	Industry 2	Industry 1	Industry 2
Gross output	30%	60%	70%	40%
Value added	25%	55%	75%	45%
Imports	20%	50%	80%	50%
Exports	25%	45%	75%	55%

Note: SME: small and medium-sized enterprise.

Table 4.2. Input-output table

	Industry 1	Industry 2	Domestic final use	Exports	Gross output
Industry 1	40	10	50	100	200
Industry 2	30	20	100	30	180
Value added	50	100
Taxes less subsidies on production	20	10
Imports	60	40
Gross output	200	180

Obtaining the shares for value added and trade to disaggregate the input-output table

Although the shares of SMEs and large enterprises that are observed in the data directly can be used, it is advised to use the method of Cadestin et al. (2018[2]) instead. The advantage of this method is that it can easily deal with missing values while easily enabling sanity checks of the parameters. Furthermore, it can be used to obtain a higher quality when compiling time series of EIOTs.

The method deals with differences between microdata and national accounts. As explained in Chapter 3, under “Differences between microdata and National Accounts”, these two sources yield different industry totals for output, value added, imports and exports. Therefore, the data have to be reconciled. The following paragraphs explain how to do this for value added; the methodology for trade is the same.

Define $v$ as the value added at the industry level, $v_s$ as value added at SMEs in this industry and $v_l$ as value added at large enterprises in this industry. Then $v=v_s+v_l$. For output at industry level, similar variables $x, x_s$ and $x_l$ are defined. Since value added is equal to the value added per unit of output times output, the following equation holds:

$v=\frac{v_s}{x_s}.x_s+\frac{v_l}{x_l}.x_l$

Define parameter $p$ as the ratio between large enterprise value-added intensity and SME value-added intensity:

$p=\frac{\frac{v_l}{x_l}}{\frac{v_s}{x_s}}$

Integrating $p$ into the earlier equation leads to:

$x_s+p.x_l=v.\left(\frac{x_s}{v_s}\right)$

Value added at SMEs is then equal to:

$v_s=\frac{v}{1+p.\left(\frac{x_l}{x_s}\right)}$

Value added at large enterprises is defined as $v_l=v-v_s$.

This reconciled the value added by industry from national accounts with that in the enterprise data: value added of SMEs plus value added of large enterprises is now equal to total value added from national accounts, for every industry. The methodology facilitates the estimation of missing values, since the only information that is required from the enterprise data is the ratio $p$, which represents the difference in the value added to output ratio between large enterprises and SMEs. When $p$ is missing, one can use information about the “upstreamness” in value chains of the different types of enterprises if this information is available. If it is, for example, known that SMEs mainly focus on turning ‘almost-finished’ products into products ready for final use, while large enterprises take care of more upstream activities as well, the value added to gross output ratios for SMEs tend to be smaller than those of large enterprises. Alternatively, one can use the value of the ratio $p$ from a more aggregate level, similar industries, comparable countries or a different year.

This approach might lead to value added that is higher than output. First, when $p<\frac{v-x_s}{x_l}$, value added of SMEs will be larger than their output. Second, when $p>\frac{x_s}{v-x_l}$, the value added of large enterprises will be higher than their output. When this happens, it is recommended to have a close look at the data. Preferably, speak to experts. Otherwise, use estimates from a different year, a similar industry or a similar country. Only for cases that cover a very small part of the economy, is it recommended to take the closest value of $p$, such that value added is lower than output. The advantage is that this is easy to program.

The same methodology is employed for trade, based on differences in export intensity and import intensity among SMEs and large enterprises.

When compiling time series of EIOTs, it is recommended to monitor the value-added intensities and trade intensities by type of enterprise by industry over time. Furthermore, it is recommended to keep an eye on the ratio $p$ that represents the intensity of large enterprises divided by the intensity of SMEs.

Disaggregating the input-output table to an extended input-output table

This section provides a step-by-step explanation on how to disaggregate an IOT into an EIOT.

Step 1. Disaggregating the columns in the IOT: Value added, gross output and imports by industry

If no product detail information is available on the imports by SMEs or by large enterprises, one has to assume that in an industry, the SMEs import the same products in the same proportions as large enterprises. Combining the shares of SMEs and large enterprises in value added, gross output and imports with these variables in the IOT allows disaggregating these data. Table 4.3 shows the result of this step.

Table 4.3. Result of step 1

	SMEs		Large enterprises
	Industry 1	Industry 2	Industry 1	Industry 2
Industry 1
Industry 2
Value added	12.5	55	37.5	45
Taxes less subsidies on production	5	5.5	15	4.5
Imports	12	20	48	20
Gross output	60	108	140	72

Note: SME: small and medium-sized enterprise.

Step 2. Disaggregating the columns in the IOT: Domestic intermediate use

Total domestic intermediate use is equal to gross output – value added – taxes less subsidies on production – imports. Applying this to Table 4.3 yields, for example, that the total intermediate use of SMEs and large enterprises in industry 1 is equal to 30.5 and 39.5, respectively. This implies that the share of SMEs in total intermediate use of industry 1 is equal to 100*30.5/(30.5+39.5) = 43.6%. Combining this share with the intermediate use of industry 1 obtained from industry 1 (40, see Table 4.2) yields that the intermediate inputs of SMEs in industry 1 obtained from industry 1 are equal to 43.6% * 40 = 17.4. Performing similar calculations for the other cells related to intermediate use.

In this process, a proportionality assumption similar to that for imports is used: SMEs and large enterprises obtain their intermediate goods and services from the same industries, proportional to their total intermediate use.

Table 4.4. Result of step 2

	SMEs		Large enterprises
	Industry 1	Industry 2	Industry 1	Industry 2
Industry 1	17.4	9.2	22.6	0.8
Industry 2	13.1	18.3	16.9	1.7
Value added	12.5	55	37.5	45
Taxes less subsidies on production	5	5.5	15	4.5
Imports	12	20	48	20
Gross output	60	108	140	72

Note: SME: small and medium-sized enterprise.

Step 3. Disaggregating the rows in the IOT: Exports

Combining the shares of SMEs and large enterprises in total exports by industry with the total exports in the IOT yields exports by SMEs and large enterprises by industry. For example, the share of SMEs in exports of industry 1 is 25% while these exports are equal to 100. Therefore, exports by SMEs in industry 1 equal 25.

Table 4.5. Result of step 3

		SMEs		Large enterprises		Domestic	Exports	Gross output
		Industry 1	Industry 2	Industry 1	Industry 2	final use
SMEs	Industry 1						25	60
	Industry 2						13.5	108
Large enterprises	Industry 1						75	140
	Industry 2						16.5	72
Value added		12.5	55	37.5	45
Taxes less subsidies on production		5	5.5	15	4.5
Imports		12	20	48	20
Gross output		60	108	140	72

Note: SME: small and medium-sized enterprise.

Step 4. Disaggregating the rows in the IOT: Domestic intermediate and final use

Again, if no extra information is available (such as value-added tax or payments information), one has to rely on proportionality assumptions. In this step, it would be assumed that SMEs and large enterprises supply the same intermediate and final users proportionally to their total domestic sales. This is implemented as follows: total domestic sales are equal to gross output – exports. Hence, the total domestic sales by SMEs and large enterprises in industry 1 are 35 and 65, respectively. The share of SMEs in domestic sales by industry 1 is 100*35/(35 + 65) = 35%. Then it is assumed that the share of sales by SMEs in industry 1 to SMEs in industry 1, the share of sales by SMEs in industry 1 to domestic final use and so on is equal to 35% as well. For example, SMEs’ sales in industry 1 to domestic final use are equal to 35% * 50 = 17.5. The result of Step 4, Table 4.6, is an EIOT.

Table 4.6. Result of step 4

		SMEs		Large enterprises		Domestic	Exports	Gross output
		Industry 1	Industry 2	Industry 1	Industry 2	final use
SMEs	Industry 1	6.1	3.2	7.9	0.3	17.5	25	60
	Industry 2	8.2	11.6	10.7	1.1	63.0	13.5	108
Large enterprises	Industry 1	11.3	6.0	14.7	0.5	32.5	75	140
	Industry 2	4.8	6.8	6.3	0.6	37.0	16.5	72
Value added		12.5	55	37.5	45
Taxes less subsidies on production		5	5.5	15	4.5
Imports		12	20	48	20
Gross output		60	108	140	72

Note: SME: small and medium-sized enterprise.

Compiling an extended supply and use table when an extended input-output table and a supply and use table are available

It is possible that an EIOT is available but an ESUT is not. For example, if an EIOT has been compiled using the previously described method or using value-added tax and/or payments data. Or a country produces the IOT first and only then compiles the supply and use table (SUT). An example is the People’s Republic of China (hereafter “China”). Due to constraints in basic statistical units and data availability, the National Bureau of Statistics’ (NBS) routine practice in China in compiling SUTs is different from the System of National Accounts recommendation (Yang, Wei and Zhu, 2016[5]). The NBS first compiles the IOTs and supply tables based on special input-output surveys every five years, then estimates the use tables based on IOTs, supply tables, annual national accounts data and technology assumptions.

The following country example by Yang et al. (2022[6]) applies the same practice to compile an ESUT. Among others, it uses the regular SUTs and EIOTs capturing enterprise ownership that were compiled by Duan et al. (2012[7]) and Ma, Wang and Zhu (2015[8]). First, it constructs an extended supply table, then an extended use table capturing enterprise ownership. The reason for using this extension of an SUT is that foreign-invested enterprises (FIEs) and domestically owned enterprises (DOEs) are different in many aspects, such as production input, export pattern and impacts on the local economy. Furthermore, FIEs and DOEs play different roles in generating local value-added, and they might have different effects on technology dissemination and skill building as well.

Data sources

The ESUT compilation process uses the regular SUTs compiled by the NBS for 2012, with 62 industries and 96 products. In addition, other data included product-by-product IOTs with 139 products from the NBS, output-related data by ownership for various sectors from the China Industry Statistical Yearbook 2013, China Statistical Yearbook 2013, China Economic Census Yearbook 2013, other statistical yearbooks about related service sectors and the CEInet Statistics Database (China Economic Information network).

Estimation of the extended supply table

The extended supply table can be estimated by splitting the make matrix in the supply table into two parts with FIEs and DOEs (F and D in Table 4.7).

Table 4.7. Framework of the extended supply table

		Industry		Valuation			Total supply at purchasers’ prices
		DOEs	FIEs	Import	Margin	Taxes on product
Product	Domestically owned enterprises (DOEs)	D
	Foreign-invested enterprises (FIEs)		F
Total output

Specifically, set $s_{ij}$ as the element of the make matrix of the regular supply table, denoting the supply of product i by industry j. Let $x_j$, $x_j^F$ and $x_j^D$ denote the output of industry j, at total level, of FIEs and DOEs, respectively. Now set

$s_{ij}^F=s_{ij}*(x_j^F/x_j)$，where $s_{ij}^F$ denotes the supply of product $i$ by FIEs in industry j

$s_{ij}^D=s_{ij}*(x_j^D/x_j)$, where $s_{ij}^D$ denotes the supply of product $i$ by DOEs in industry j

This method has the implicit assumption that in each industry, the product structure of FIEs and DOEs is identical.

The key question is how to estimate $x_j^F/x_j$ or $x_j^D/x_j$. First, due to data availability, the NBS’ regular SUT was aggregated to 84 commodities by 52 industries. The subsequent approach to obtain the shares of FIEs and DOEs in output differs by industry. It is a nice example of “use what you have”:

• Farming, forestry, animal production, fishery and related support services: take the ratio of “foreign direct investment” (cumulative amount for 2003-12) and “gross fixed capital formation” (cumulative amount for 2003-12) as the share of FIEs in output of the industry.

• Thirty-four mining/manufacturing industries: use “sales of enterprises above designated size”1 of FIEs and DOEs, by industry.

• Construction: use “gross output” by FIEs and DOEs.

• Wholesale and retail trade: use “output” of FIEs and DOEs; alternatively, for larger FIEs and DOEs (with a revenue higher than the threshold of designated size for the industry), “revenue from principal business” minus “cost of principal business” can be used, or “total sales value” minus “total purchases value”.

• Accommodation: use “revenue from principal business above designated size” by FIEs and DOEs.

• Food and beverage services: use “revenue from principal business above designated size” by FIEs and DOEs.

• Finance (exclude insurance): use the “paid-in capital” share of foreign-funded banks and domestic-funded banks.

• Insurance: take the ratio of “premium of primary insurance between FIEs and the total industry” as the share of FIEs in output of the total industry.

• Real estate: use the “completed investment of real estate development companies” by FIEs and DOEs.

• Other ten service sectors: take the ratio of “foreign direct investment” (cumulative amount for 2003‑12) and “gross fixed capital formation” (cumulative amount for 2003-12) as the share of FIEs in output of the industry. Alternatively, use the share of FIEs in “business revenue”.

Estimation of the extended use table

This table is constructed in two steps. First, as an unbalanced table, then as a balanced table.

In the previous step, the extended supply table was constructed. Using this table and the extended product-by-product IOT, one can now derive the extended use table under the product technology assumption.2

Table 4.8. The framework of the extended use table

		Industries		Final use	Total use at basic prices
		DOEs	FIEs
Product	Domestically owned enterprises (DOEs)
	Foreign-invested enterprises (FIEs)
	Import
Taxes less subsidies on products
Value added
Total output

The extended product-by-product IOT and the extended supply table contain the following information: the output coefficients matrix by industry ($T$), the output vector by industry $X_I$, the total use of domestic product $X$, the total use of import product $X_m$, the domestic intermediate input coefficients matrix by product $A$, the import intermediate input coefficients matrix by product $A_m$, the value-added matrix by product $A_v$, the final use matrix for domestic product $F$ and the final use matrix for import product $F_m$.

Table 4.9. Calculation of the unbalanced extended use table by enterprise heterogeneity

		Industries		Final use	Total use at basic prices
		DOEs	FIEs
Product	Domestically owned enterprises (DOEs)	U=ATXI^		F	X
	Foreign-invested enterprises (FIEs)
	Import	M=AmTXI^		Fm	Xm
Value added		VI=AvTXI^
Total output		XI

Using the product technology assumption, the domestic intermediate input matrix $U$ in the extended use table can be calculated as$\mathrm{ }U=AT\widehat{X_I}$. Similarly, the import intermediate input matrix $M$ ($M=A_{m}T\widehat{X_I}$) and the value-added matrix $V_I$ ($V_I=A_{v}T\widehat{X_I}$) can be derived. Other data can be obtained directly from the EIOT and the extended supply table, such as the total input by industry${ X}_I$, the final use matrix for domestic product $F$ and import product $F_m$.

This extended use table is still unbalanced. Yang et al. (2022[6]) used a Quadratic Programming Model to balance the table. The table had to suffice several constraints. Namely, when the DOEs and the FIEs in the extended use table were added up, the value of each cell needed to match the value of the corresponding cell in the NBS’ regular use table. Furthermore, the table also needed to satisfy the row and column constraints. In more detail, minimise S where S is:

$S=\sum _{j=1}^{84}\sum _{i=1}^{52}\sum _l\sum _k\left[{\left(u_{ij}^{lk}-{u0}_{ij}^{lk}\right)}^2/{u0}_{ij}^{lk}\right]+\sum _{j=1}^{84}\sum _{i=1}^{52}\sum _k\left[{\left(m_{ij}^k-{m0}_{ij}^k\right)}^2/{m0}_{ij}^k\right]+\sum _{s2=1}^4\sum _{i=1}^{52}\sum _k\left[{\left(v_{s2,j}^k-{v0}_{s2,j}^k\right)}^2/{v0}_{s2,j}^k\right]+\sum _{j=1}^{84}\sum _{s1=1}^3\sum _l\left[{\left(f_{i,s1}^l-{f0}_{i,s1}^l\right)}^2/{f0}_{i,s1}^l\right]+\sum _{j=1}^{84}\sum _{s1=1}^3\left[{\left({fm}_{i,s1}^l-{fm0}_{i,s1}^l\right)}^2/{fm0}_{i,s1}^l\right]$

such that the following identities and constraints hold:

• Row sum identities required by the use table

$\sum _{j=1}^{52}\sum _k{u}_{ij}^{lk}+\sum _{s1=1}^3{f}_{i,s1}^l=x_i^l, i=\mathrm{1,2},\cdots , 84;l=DOEs, FIEs$

$\sum _{j=1}^{52}\sum _k{m}_{ij}^k+\sum _{s1=1}^3{fm}_{i,s1}={xm}_i, i=\mathrm{1,2},\cdots , 84$

• Column sum identities required by the use table

$\sum _{i=1}^{84}\sum _l{u}_{ij}^{lk}+\sum _{i=1}^{84}{m}_{i,j}^k+\sum _{s2=1}^4{v}_{s2,j}^k={xI}_j^k, j=\mathrm{1,2},\cdots , 52;k=DOEs, FIEs$

• Adding up conditions for intermediate inputs

$\sum _l\sum _k{u}_{ij}^{lk}+\sum _k{m}_{i,j}^k=u_{ij}, i=\mathrm{1,2},\cdots , 84; j=\mathrm{1,2},\cdots , 52;l,k=DOEs, FIEs$

• Adding up conditions for product outputs

$\sum _k{v}_{s2,j}^k=v_{s2,j}, s2=\mathrm{1,2},\mathrm{3,4}; j=\mathrm{1,2},\cdots , 52; k=DOEs, FIEs$

• Non-negative constraints

$u_{ij}^{lk},m_{i,j}^k, {fm}_{i,s1} \ge 0$

where:

• The initial values ${u0}_{ij}^{lk},{m0}_{ij}^k,\mathrm{ }{v0}_{s2,j}^k,\mathrm{ }{f0}_{i,s1}^l,\mathrm{ }{fm0}_{i,s1}^l$ are from the unbalanced extended use table, corresponding to $U,M,V,F,F_m$ in Table 4.9.

• The constraints $x_i^l,{xm}_i,\mathrm{ }{xI}_j^k$ are from the unbalanced extended use table, corresponding to $X,X_m,X_I$ in Table 4.9.

• $k=DOEs,\mathrm{ }FIEs$ means the industry is DOEs or FIEs while $l=DOEs,\mathrm{ }FIEs$ means the product is from DOEs or FIEs; $s1=\mathrm{1,2},3$ represents consumption, capital formation and export in final use; $s2=\mathrm{1,2},\mathrm{3,4}$ represents the compensation of employees, net taxes on production, depreciation of fixed assets and operating surplus in value added, respectively.

• $u_{ij}$ is the intermediate input matrix and $v_{s2,j}$ is the value-added matrix in the use table provided by the NBS.

This process yields the balanced extended use table.

Comparing the compilation of an extended input-output table in data-rich and data-scarce environments

This chapter and Chapter 3 showed that, generally, there are two ways to compile ESUTs and EIOTs. Namely, in a data-rich environment, using as much data as possible, and in a data-scarce environment, using less data and more assumptions. Both methods are discussed below and Table 4.10 summarises both methods, indicating their stronger and weaker points. Work by the OECD and Statistics Netherlands shows that the results in their specific cases are robust under different approaches. However, note that these findings were obtained in specific contexts, concerning the countries, years, and dimensions of enterprise heterogeneity studied. Therefore, one should not draw general conclusions from the two cases discussed.

Data-rich environment

Ideally, one reuses as much as possible the existing data and software infrastructure that is used for the compilation of regular SUTs. For example, if the infrastructure is flexible with respect to industries, one might feed the system with value added and production by industry by enterprise type. The “industry by enterprise type” is then a quasi-industry. This was the approach used by Chong et al. (2016[1]; 2019[9]). They used as much as possible the existing microdata, infrastructure and methods then compiled an ESUT and subsequently an EIOT.

This approach requires a lot of data and a lot of time to compile and adjust the data. It is not feasible for external researchers unless they have access to microdata, which is often not the case. The advantage of this approach is that it maximally capitalises on what is used in the regular process, staying as close as possible to that process. It allows integrating supply and use at a detailed level as well as for different input structures for different types of enterprises in the same industry, both for foreign/domestic inputs and for production technology. Furthermore, the approach allows for flexibility, since one can enter detailed information when desired.

Data-scarce environment

Since the necessary detailed data are generally not accessible for external researchers, other methods have been devised, as was described earlier in this chapter. In this approach, data consumption and time consumption are limited, which makes it accessible for external researchers. Obviously, this comes at a cost. There is no integration of supply and use at a detailed level. The method does not allow for differences in production technology for different types of enterprises in the same industry. The method has limited flexibility; one cannot adapt detailed items since one works at an aggregated level. However, some adaptions will still be possible, for example, assigning illegal production in an industry non-proportional over the different types of enterprises. Or using information about production, imports and exports at product level. Since there is no integration at supply and use of products, this method must make more assumptions. For example, a proportionality assumption is used to assign the intermediate supply of a type of enterprise in an industry to the users (enterprise types by industry). Still, the method yields extra information compared to a completely proportional disaggregation. First, since it allows for differences in foreign/domestic inputs. Second, because imports and value added form a sizeable part of inputs of an industry, and these numbers are set correctly by construction, only the intermediate part of the IOT can be partly wrong.

Table 4.10. Comparison of the data-rich and data-scarce methods

	Data-rich	Data-scarce
Data consumption	Very high	Limited
Time consumption	Very high	Limited
Quality	High	Average
Possible for external researchers	Not often	Yes
Assumptions	Some	More
Trade in services*	Problematic	Problematic
Flexible	Yes	Average

Note: * This is a general data gap. Many countries do not have trade by services by industry split by type of enterprise as well.

Robustness of the data-scarce method

In practice, most researchers will have to use the data-scarce approach since they do not have access to the data needed for the data-rich approach. Therefore, it is necessary to assess its quality by checking the robustness of the results. Both OECD (2018[4]) and Lemmers (2020[10]) do exactly that. OECD (2018[4]) point out that the data-scarce method has three important assumptions:

• There is no substitution between imports and domestically purchased products among small and large enterprises. The share of imports in purchases will differ by enterprise type, but the product baskets for imports and domestic purchases are the same within each enterprise type.

• There are no differences in the use of the production of small and large enterprises. Their production is proportionally distributed to domestic intermediate and final use.

• Enterprises have no preference for purchasing from either small or large enterprises.

They find that indicators such as imports embedded in exports hardly differ under different scenarios, but that the domestic intermediate relations between different types of enterprises can be different.

Lemmers uses the EIOT by Chong et al. (2016[1]) that was compiled using a data-rich approach. He then aggregates to a usual IOT without types of enterprises and calculates the shares of each type of enterprise in an industry. This information is subsequently used to compile a top-down EIOT. That table is compared by calculating several indicators to the table of Chong et al. These indicators are domestic value added embodied in exports, value added embodied in exports as share of total value added and export spillovers (value added at suppliers/value added at exporting industry). They are calculated at the level of industry by enterprise type. Generally, the differences between indicators based on the two EIOTs are small. There are, however, some exceptions, i.e. the beverages and tobacco industry. However, there are usually clear causes for the differences; for example, in the aforementioned industry, a sizeable part of the SME production is the illegal production of cannabis. Large enterprises do not produce this. Hence the input-output structure of the different types of enterprises in the same industry will be very different.

As mentioned earlier, these two findings were obtained in specific contexts. Therefore, one should not draw general conclusions from these cases.

References

[2] 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.

[9] Chong, S. et al. (2019), “The role of small-and medium-sized enterprises in the Dutch economy: An analysis using an extended supply and use table”, Journal of Economic Structures, Vol. 8/1, pp. 1-24, https://doi.org/10.1186/s40008-019-0139-1.

[1] Chong, S. et al. (2016), Een IO-tabel voor het MKB en grootbedrijf (in Dutch), Statistics Netherlands.

[7] Duan, Y. et al. (2012), “Does domestic value added induced by China’s exports really belong to China?”, China & World Economy, Vol. 20/5, pp. 83-102, https://doi.org/10.1111/j.1749-124X.2012.01303.x.

[10] Lemmers, O. (2020), Capturing Heterogeneity in Global Value Chains: How to Slice and Dice?, Paper presented at the ESCoE conference on Economic Measurement 2020.

[8] Ma, H., Z. Wang and K. Zhu (2015), “Domestic value-added in China’s exports and its distribution by firm ownership”, Journal of Comparative Economics, Vol. 43/1, pp. 3-18, https://doi.org/10.1016/j.jce.2014.11.006.

[4] OECD (2018), Accounting for Firm Heterogeneity in Global Value Chains: The Role of Small and Medium-sized Enterprises, STD/CSSP/WPTGS(2018)5, OECD, Paris, https://one.oecd.org/document/STD/CSSP/WPTGS(2018)5/en/pdf.

[3] OECD (2015), Firm Heterogeneity and Trade in Value Added, STD/CSSP/WPTGS(2015)23, OECD, Paris, https://one.oecd.org/document/STD/CSSP/WPTGS(2015)23/en/pdf.

[6] Yang, C. et al. (2022), Extended Supply-Use Tables by Firm Heterogeneity for China, Paper presented at the 2023 IIOA Conference in Alghero, https://iioa.org/conferences/29th/papers/files/4561.pdf.

[5] Yang, C., R. Wei and K. Zhu (2016), Compiling SUTs with Firm Heterogeneity: Methods and Some Empirical Results for the Case of China, Paper presented at the 2016 IIOA conference in Seoul.

Notes

← 1.  The term “industrial enterprises above designated size” refers to enterprises with an annual main business revenue that exceeds a certain threshold set by the government. This threshold can differ by industry and by year.

← 2.  This assumes that a product has the same input structure regardless of the industry in which it is produced.