«UK Economic Performance: How Far Do Intangibles Count? Rebecca Riley and Catherine Robinson March 2011 INNODRIVE Working Paper No 14. The research ...»
It is important to be clear about what we are trying to capture with respect to firm level intangibles. In this paper we focus on own account production of intangibles and not those purchased by firms. Görzig et al (2011) highlight a number of reasons why this component is likely to be substantial and generally poorly measured, not least with respect to ICT assets because of accounting practices that make it more sensible to record ICT investment as an intermediate expenditure (c.f. Lequiller and Blades, 2006).
3. UK DATA SOURCES Measures of UK firms‟ expenditures on intangible capital goods may be derived from a number of data sources.1 For the purposes of developing and analyzing occupationally based measures of these expenditures we require information on the occupational distribution of the firm‟s workforce in addition to standard financial information on the operations of the firm.2 Details of UK firms‟ employees, their occupations, earnings and hours worked are available from the Annual Survey of Hours and Earnings (ASHE). These employee data can be linked via the ONS‟s Inter-Departmental Business Register (IDBR) to firms in the Annual Business Inquiry (ABI), which holds information on firms‟ labour costs, output, capital investment, intermediate expenditures, and employment. However, because the ASHE is but a 1% sample of employees in UK businesses, we are only able to construct adequate occupational measures for the small sample of (very large) UK businesses that have sufficient employees included in the ASHE. For this reason we match the employee data to firms by detailed industry and size group, rather than by firm identifier. Here we outline the two business surveys we use, the properties of the linked and matched samples, coverage, limitations and other issues.
3.1 Business surveys The ASHE is a 1% random sample of employee jobs on the PAYE register held by the UK tax authorities, and contains detailed information on approximately 160,000 employees every year. Sample selection occurs on the basis of National Insurance numbers and is maintained over time, thus the ASHE contains longitudinal information on UK employees. The survey 1 For example, the Community Innovation Surveys (CIS) include information on firms‟ expenditures on R&D (intra- and extra-mural), knowledge acquisition (purchases external to the firm), personnel training, as well as expenditures on design, market research and advertising associated with the development of new or significantly improved goods and services. The Business and Enterprise Research & Development Inquiry contains information on scientific R&D. Limitations of these surveys are small sample sizes and relatively narrow focus (e.g. focus on scientific R&D and lack of information on expenditures on organizational structures). Where the CIS is concerned a further constraint is intermittent availability and non-response bias in earlier surveys (see Criscuolo and Haskel, 2002).
2 Linked employer-employee data (LEED) of this kind have been used elsewhere in the labour economics literature to explore the nature of human capital formation (cf. Abowd et al., 1999).
8 covers all sectors of the UK economy. Detailed information on pay and hours worked are collected from employers, as well as detailed occupation and industry category. It contains no information on employees‟ qualifications. Low-paid workers are under-recorded in the ASHE; weighting procedures allow this to be taken into account.
The ABI is a census of UK businesses with more than 250 employees and a stratified (by industry, region and employment size) sample of smaller tax-registered businesses.3 On average response rates are 82%.4 Sector coverage is almost complete; however there are a number of omissions and also a number of sectors where inputs are not thought to be directly comparable to the measured outputs. Typically, the latter consist of public sectors, such as education and health. Sectors that are not covered include certain industries within agriculture, public administration and defence, and, particularly pertinent to this paper, the financial services sector is omitted completely. Together, industries included in the ABI account for approximately two thirds of the UK economy.5 The ABI contains employment and financial information on approximately 50 thousand UK enterprises every year since 1998. Although the information in the ABI yields annual longitudinal information for larger firms, there are large gaps in the data for smaller firms because of the rotating sampling strategy (smaller firms cannot be included in the sample in consecutive years). This means that when we capitalize investment flows we need to interpolate data for smaller firms.
Data in the ABI are collected at the level of the Reporting Unit. This is an administrative unit within a firm, decided upon entirely by the firm for the sake of convenience. One problem with using reporting unit data as the unit of analysis is that the firm does not necessarily use a consistent definition over time (Harris, 2002). An alternative is to disaggregate the data back down to the plant level, but this involves the assumption that all plants within a ReportFor a full description of the ABI see Barnes & Martin (2002) and Criscuolo et al. (2003).
ing Unit are identical. We use ABI information aggregated to the Enterprise reference (the smallest legally distinguished production unit). This is because it is at this level that we can link the ABI to the ASHE, although postcodes can be used to extend the link to more disaggregated units in some instances, and the enterprise is perhaps most akin to the concept of a firm.6 Information on firms‟ stocks of plant, machinery and equipment can be constructed from the ABI information on investment.7 We ignore expenditure on land and buildings. These are quite volatile over time, and one might argue these fluctuations have little to do with the productive capacity of a plant. The capital stock items that can be constructed are less reliable for Utilities and Construction firms and for the public sectors, and hence we exclude these sectors from our sample.
3.2 Linked and matched samples We link the information on the occupational distribution of labour costs, hours worked and employment from the ASHE to the financial information of firms in the ABI via the Enterprise reference available in both datasets.8 In this way we generate estimates of firms‟ expenditures and use of workers in particular occupation categories. Firm employment and 6 The vast majority of Enterprise references are associated with a single Reporting Unit (and single plant). We exclude Enterprises references for which we do not have full financial information on all Reporting Units. We allocate enterprise industry and City Region on the basis of the industry and City Region of the majority of employees within the enterprise. In the few cases where we observe ties these characteristics are allocated randomly.
7 We use plant, machinery and equipment capital stock data provided by Richard Harris augmented with firms‟ leasing of these assets. These were constructed using starting stocks at the 3-digit level of industry disaggregation provided by NIESR/Mary O‟Mahony. For manufacturing, these starting stocks run from 1969 and are subsequently built-up using the Perpetual Inventory Method of calculation, using real net capital expenditures year on year. The methodology underlying the construction of these data is fully described in Harris & Drinkwater (2002).
8 Enterprise references are not available in the ASHE before 2002 and need to be linked in via the PAYE reference. We thank Richard Upward (personal correspondence, 2007) for his advice in using 1997 and 2004 lookup tables for ENTREF-PAYEREF in order to add in Enterprise references in the ASHE 1998-2001. Where there is conflict between the ENTREF in the 1997 and 2004 tables we allocate the ENTREF that corresponds to the same 4/5 digit industry.
10 labour cost totals are constrained to be as in the ABI. However, this procedure is only meaningful for firms where we have a sufficient number of ASHE employees. Because of the restrictive ASHE sample of firms‟ employees, our firm-sample becomes very limited in coverage, highly skewed towards very large firms. For instance, focusing on linked firms where we have at least 10 ASHE employees and for which we have full financial information, we achieve a sample of approximately 400 firms per annum. For these reasons we match in the occupational information via 3-digit SIC category and 4 employment size bands by year.
Where this leads to cell sizes smaller than 30 ASHE employees we merge employment size groups, and, in some cases, move to the 2-digit SIC category. This procedure leaves us with on average 270 linking cells per year, distinguished by 163 industry categories (see Table 1).
For the small sample of large firms for which it makes sense to link the ASHE occupational information by the enterprise identifier we check the coherence of the occupational distribution between the two measures (note that both are only approximations of the true data).
Table 2 shows the correlation across firm-years of the share of employees (hours, or labour costs) in a particular occupation group as measured by the 1% sample of the firm‟s workforce and as measured by the 1% sample of employees in the same industry/firm-size category. The correlations for the three occupation groups that define our intangibles are positive and highly statistically significant. For R&D and ICT workers these correlations are higher (73% for labour cost shares) than for organizational workers (50% for labour cost shares); R&D and ICT tend to be more concentrated in particular industries.
3.3 Coverage and other issues As our approach is harmonsied with a number of other European countries we make a number of restrictions on the data. We exclude firms with turnover less than €2million (averaged at 2000 prices) and/or employing less than 30 employees. In addition, we exclude firms in the agricultural and public sectors. We have already alluded to further constraints on the basis of the British data specifically. There is poor or incomplete coverage of the following sectors: mining and quarrying of energy producing materials; manufacturing of coke; refined petroleum products and nuclear fuel; electricity, gas and water supply; construction;
financial intermediation; health and social work. In total, this leaves us with a sample of approximately 11,000 enterprises per annum 1998-2006.
11 Our data covers only Great Britain; i.e. the UK excluding Northern Ireland. However, when we weight up the firm-level data to be nationally representative of the industries in our sample, we weight to a published UK total. Sample coverage is indicated in Table 3, which shows, for the industries we consider, the share of UK total gross value added and employment accounted for by the firms in our sample. Coverage averages a third of GVA and is a bit higher measured on employment. The industries we consider account for approximately 55% of whole economy GVA (calculated using EU KLEMS).
In weighting up the firm-level data to be nationally representative (as we do in what follows) we aggregate to ABI broad industry totals published by the ONS (gross value added for financial items, employment for hours worked and employment items, labour costs for intangible items). We make an adjustment for differences in labour use between the smaller firms that are excluded from our sample and the larger firms included in the sample on the basis of the ASHE. We use the Business Structure Database, which holds information on turnover and employment for all UK tax registered businesses, in order to derive within industry weights by firm-size category.
The ABI financial information is published in current values. ONS input and output deflators are used to construct real values; these are typically available at the 2-digit sector level.
Intangible investment is deflated using the average earnings index.
Our classification of workers into “intangible” producing occupations is constructed on the basis of detailed occupational classification and information on workers‟ qualifications. We base our grouping of occupations on ISCO88 as listed in Görzig et al. (2011), facilitating international harmonization. Look-up tables to the UK Standard Occupational Classification are available from the ONS. In the absence of information on workers‟ qualifications in the ASHE we evaluate the average skill content of individual occupations using the Labour Force Survey (LFS) and classify occupations accordingly. The change in UK occupational classifications between SOC90 and SOC2000 causes a discontinuity in our data between 2001 and 2002; given this, the data are not strictly comparable between the first 4 years and 12 the latter 5 years of our sample. Using the LFS, which is coded to both SOC90 and SOC2000 in some years, we attempt to minimize this discontinuity.
4. MEASURING INTANGIBLE CAPITALWe measure intangible investment and capital following the methodology adopted in INNODRIVE (European Commission FP7 project), which is described in full in Görzig et al.
(2011). Here we outline briefly the methodology.
4.1 Basic approach Crucially, a firm‟s investments in intangible assets are assumed to be proportional to the firm‟s labour costs associated with workers in intangible occupations (i.e. involved in the creation of intangible capital goods). The proportionality factor is a multiple of three separate parameters: the share of intangible workers‟ time that contributes to future production,, a scaling factor to account for other (non-labour) inputs associated with the production of intangible capital goods,, and the ratio of the marginal product of intangible workers to the wages they are paid, ; IC=R&D, ICT, ORG. Thus, a firm‟s investment in intangible
assets is derived as: