Avoiding double counting or omission when measuring R&D output and capital formation

Joint meeting of Canberra II and NESTI Groups

May 31, 1 June 2006

Charles Aspden (OECD) and Mark de Haan (Netherlands CBS)

Introduction

  1. This note is based to a large extent on the papers prepared by Peleg and Mandel (2003), de Haan (2003) and Aspden (2004) for the Canberra II Group, and is for the most part an update of the information and an attempt at reconciling the different views expressed in them.
  2. At present, expenditure on R&D is treated as consumption, but if the UNSC accepts the ISWGNA/AEG recommendation to include R&D in the asset boundary of the SNA then R&D output that meets the general definition of an asset and those of afixed asset (i.e. produced assets that are used repeatedly, or continuously, in processes of production for more than one year) one of the problems national accounts compilers will face is the possibility of double counting or omissionof capital formation. The problem lies with own account capital formation (i.e. assets produced for own use).
  3. In general, non-market production, including own account capital formation, is estimated by summing costs. This means the costs of labour, capital and intermediate inputs and other taxes (less subsidies) on production need to be added together. The cost of capital should be the value of the capital services provided by all the fixed assets used during the period in producing the asset. In some cases national accountants do the estimation themselves and in other cases they rely on survey respondents to do it for them. Who ever does it, it is important that all the appropriate costs are included when deriving the estimates of capital formation and that costs are used only once.
  4. In a sense, all capital formation involves a form of double (or multiple) counting in the national accounts. The production of an asset is recorded ingross value added and GDP. In subsequent periods it can provide capital services which contribute to the production of goods and services, including other assets. Hence, over time there is a multiple counting. This is why the inclusion of R&D in the asset boundary will raise the level of GDP.
  5. This kind of temporal double counting is an accepted feature of the national accounts and is different from the issue addressed in this paper, which is concerned with over and under-estimating the value asset production in a period. This can happen if the costs of production are used more than once to derive estimates of own account capital formation in the same period, typically for different types of assets.

R&D and software

  1. R&D surveys conducted as per the Frascati Manual (FM) are the best available source of data for compiling estimates of R&D capital formation. Most R&D capital formation is produced on own account and in the absence of being able to observe prices for equivalent assets its value should be obtained by summing its cost of production. The expenditure data collected by R&D surveys require some adjustment for national accounts purposes, but having made them double counting can arise if the data are summed to obtain estimates of own account capital formation of R&D, while at the same time some, or all, of the same data are also summed to obtain own account capital formation of some other asset. For example, in the case of R&D and software some producers will have either:
  2. undertaken R&D in the course of producing software, or
  3. produced software in the course of undertaking R&D
  4. An R&D survey adhering to the FM would record either some or all of the expenditure incase (a)and all of the expenditure in case (b) as expenditure on R&D.At present, a separate survey seeking data on gross fixed capital formation of software is likely to obtain the total value of the expenditure in case (a) and that part of the software component not expensed in case (b). Hence, if the data from the two surveys were added together there would beeither a partial or completedouble count of capital formation of R&Dand software inboth cases.
  5. The FM includes capital expenditures in deriving expenditures on R&D. For national accounts purposes the former should be replaced by estimates of capital services from the fixed assets used in undertaking the R&D. The necessary adjustments should be relatively straightforward to accomplish for purchases of capital goods, but more difficult to deal with if the fixed assets have been produced on own account. Own account production of software is ubiquitous and so the problem of double counting software GFCF could be significant (see de Haan and van Rooijen-Horsten, 2004). The problem will be less severe for most other forms of fixed assets, such as computers and buildings, which are very likely to be purchased.
  6. In October 2001 an OECD task force was set up to address the great disparity in the estimates of software GFCF compiled by member countries. One of its first actions was to conduct a country survey. The survey found that own account software accounted for about a third of total software GFCF. Some countries reported that they obtained their estimates from business surveys, while others found that their survey data appeared to grossly understate GFCF and as a consequence they used macro methods to estimate own account GFCF.
  7. The task force concluded that in the absence of reliable survey-based data, countries should adopt a macro approach. In summary, the macro-approach values own-account software capital formation as the sum of its inputs. As R&D is not currently treated as fixed capital in the SNA, expenditure on R&D is treated as an input cost, and, so, a critical feature of this approach is the implicit inclusion of software R&D in fixed capital formation of software. Thus, those countries following the task force’s recommendations are already including software R&D in their estimates of software GFCF – or at least doing their best to do so. It is also very likely that those countries relying on businesses to estimate their own account GFCF expect them to include expenditure on software R&D.
  8. Therefore, it is proposed that the advice to NSOs who use macro methods to estimate software GFCF be amended along the following lines:
  9. The Frascati guidelines clearly state that certain software development projects may entirely fall under the Frascati definition of R&D (cf. paragraphs 135 and further). “For a software development to be classified as R&D, its completion must be dependent on a scientific and/or technological advance, and the aim of the project must be the systematic resolution of a scientific and/or technological uncertainty”, and, “The nature of software development is such as to make identifying its R&D component, if any, difficult”. When no reasonable distinction can be made between the software or R&D constituents of an activity, the concomitant current expenditure should either be categorised as software or R&D. Since the main outcomes of these projects are software originals, it seems most obvious to categorise these projects as software related.
  10. However, if the expenditure on R&D is a substantial part of the total cost of undertaking the development of the software and if the capital services of the R&D are unlikely to be consumed either largely or entirely in the development of a single software product then the capital formation of the software and the R&D undertaken to develop it should be recorded separately and the capital services from the R&D should be used when summing costs to value the software GFCF. This could be implemented as follows:
  11. Subtract software R&D (collected as per the FM) from the macro estimate of own account software GFCF, and
  12. Estimate the value of capital services produced by software R&D fixed assets and add this back to the estimates of own account software GFCF.
  13. This still leaves a possible double count of software GFCF undertaken as part of non-software R&D. One possible way of overcoming this would be to ask respondents to R&D surveys to separately identify expenditure on software development. Such estimates could then be deducted from the macro estimates of software GFCF.

R&D and mineral exploration

  1. Another type of intellectual property asset that may be the subject of double counting is mineral exploration. GFCF of mineral exploration, like R&D, is likely candidate to be estimated using a special survey, and although the FM indicates that “…surveying and prospecting activities of commercial companies will be almost entirely excluded from R&D”, there is a possibility of overlap. According to the FM, R&D involved in “mining and prospecting” is restricted to the following two activities:
  2. The development of new or substantially improved methods and equipment for data acquisition, processing and study of the data collected;
  3. Surveying undertaken as an integral part of R&D project on geological phenomena per se.
  4. These two activities may overlap with mineral exploration as defined in the SNA. In paragraph 10.91, the 1993 SNA indicates that mineral exploration costs “..include not only the costs of actual drilling and boring, but also the costs incurred to make it possible to carry out tests, for example, the costs of aerial or other surveys…”. These additional costs seem to be related to the two activities described by the FM, above.
  5. As for software, if mineral exploration R&D is considered to provide on-going capital services then only these capital services should be included in deriving estimates of own account GFCF of mineral exploration. If, however, it is considered that mineral exploration R&D is used up rapidly it would be best to record it as intermediate consumption when estimating own account mineral exploration GFCF and exclude it from R&D GFCF.

R&D and estimates of GFCF obtained directly from surveys

  1. Capital expenditure or more general surveys are commonly used to obtain estimates of GFCF directly from units. Respondents are asked to report their purchasers (and may be sales) of fixed assets and the value of own account capital formation - estimated by summing costs. Are the costs of undertaking R&D and the costs of purchased R&D included in the estimates? With the recognition of R&D assets, only the cost of the services provided in the period should be included in the estimates of own account capital formation. In this respect, R&D assets are no different from any other fixed assets used in own account capital formation, but respondents will need to be provided with advice on how to deal with R&D when summing costs to estimate own account capital formation.
  2. There is inherently a risk of double counting or omission when obtaining estimates of GFCF for different types of assets from different sources. Given the nature of R&D, the fact that most GFCF is own account and the fact that business accounting standards do not recognise R&D output as an asset in the way the SNA is expected to, then the risk of omission or double counting may be more acute for R&D than other types of asset. Adding an R&D category to the general survey form would, in concept, avoid double counting, as businesses are unlikely to record the same expenditure twice and overstate the total.The estimates of expenditure on R&D obtained from a general survey are almost certain to be inferior and therefore different, possibly to a substantial degree, from those of a specialised R&D survey of the kind undertaken by many countries as per the FM. This raises the difficulty of melding data from the two types of survey. One possible way of overcoming the difficulty would be to coordinate the R&D surveys with the business surveys used to obtain GFCF data for other types of asset. This would entail sending the R&D and capital expenditure, or more general, survey forms to a common set of units in respect of the same accounting period. If the expenditures on R&D reported in the two surveys differed the respondent would be asked to resolve the discrepancy.
  3. This possible solution does have problems. First, it would increase the reporting burden of the units selected. Second, the optimum sample designs for the R&D and general surveys may be substantially different, which is likely to result in the need for larger samples to avoid a reduction in estimate quality. Larger samples would, of course, increase the costsfor the NSO and increaserespondent burden.This suggests that before embarking on such a strategy, NSOs should undertake an investigation to determine the magnitude of the problem and whether it is concentrated in certain industries or types of business (e.g. vary large ones). With this knowledge a partial sample overlap could be devised, if indeed one is required at all.

Conclusions

  1. There is a potential problem of double counting GFCF with the inclusion of R&D in the asset boundary. This problem arises when summing costs to estimate own account capital formation: the capital cost should be the cost of the capital services provided and not the cost of assets acquired in the period. Essentially, the problem is one of recording the using up of assets at the appropriate time.
  2. In the case of R&D GFCF based on data from R&D surveys the problem can be overcome by estimating the flow of capital services from the expenditures of fixed assets recorded in the surveys. This should be straightforward for purchased assets, but is likely to be problematical for own account capital formation. In the case of fixed assets whose estimates of GFCF are obtained from survey respondents, the respondents need to be advised as to the appropriate way of estimating own account capital formation.
  3. Using the R&D survey estimates of R&D GFCF in conjunction with the survey estimates for other asset types would probably lead to under- or over-statement of total GFCF. The solution of having a common sample for both types of survey could be expensive and investigations may be required so as to minimise the overlap needed.
  4. The current recommendations on estimation of software GFCF using a macro method will need to be amended along the lines indicated.

References

Aspden C. (2004) The demarcation between GFCF of software and R&D, paper presented at the fourth meeting of the Canberra II Group, September, 2004

Haan, M. de (2003) Capitalization of R&D – some comments, Canberra II Group paper

Haan, M. and M. van Rooijen-Horsten, Measuring R&D output and knowledge capital formation in open economies, paper prepared for 28th IARIW conference, 2004

Mandler, P. and S. Peleg (2003) R&D and software, paper presented at the second meeting of the Canberra II Group, October 2003

OECD Software Task Force (2002), Report of the OECD Task Force on Software Measurement in the National Accounts, paper presented to the OECD Meeting of National Accounts Experts, Paris, October, 2002.