Data sources for other research and development components

View the most recent version.

39. One of the long-standing issues with adopting and implementing the treatment of R&D as capital in the national accounts has been the perceived problem of data availability. Fortunately, an internationally recognized information set estimating R&D expenditure already exists. The compilation of these data, as detailed by the Frascati Manual, was initiated by the National Experts on Science and Technology Indicators (NESTI), a working group of the OECD. The Frascati expenditures provide information on R&D expenditures, organized by funding and performing sector.

Basic data sources in Canada

40. Statistics Canada's Science, Innovation and Electronic Information Division has produced data based on the FM for the years 1963 to the present.¹² The data are referred to as Gross Domestic Expenditure on Research and Development (GERD). These expenditures provide a solid basis for measuring R&D output and investment. Information is collected using both surveys and administrative sources. Surveys cover the business sector units with over $1 million of R&D expenditures, whereas tax data are used for firms below this threshold. Business sector data are available at a detailed industry level for basic, applied and developmental research. A combination of survey and administrative data are also used to compile data for government and its sub-sectors (which includes higher education).

Sectors and industries

41. An issue that has been noted in several international studies concerning R&D capital formation has been that of linking the FM sectors with SNA sectors (Table 2).¹³ This sectoring issue does not present a complicated problem in the Canadian case. For business enterprises, the FM data is available at a detailed industry level and therefore a split between financial and non-financial industries-sectors is not difficult. Source of funding information is available for the private non-profit sector.¹⁴ Higher education is assigned to the general government sector in Canada.

Table 2
Linking Frascati Manual sectors to System of National Accounts sectors

42. In calculating the production account in the RDSA in Canada, a similar classification issue arises. It involves mapping the industry FM data to the industry detail used in the Input-Output Tables (IOT) in the SNA. This process does not change the levels of expenditure. Both FM and SNA are based on NAICS, however the SNA tables introduce some aggregations not found in NAICS. A well-defined concordance between NAICS and the IOT allows for a relative easy conversion of the industry data.

Other System of National Accounts adjustments

43. Additional adjustments are required so that the FM expenditures conform to the framework of the national accounts. These adjustments transform the expenditure-based (or cost-based)¹⁵FM data to an output measure as defined in the SNA. Most of the adjustments for Canada are based on data from within the SNA, and are as follows:

• Data from the Input-Output Tables (IOT) are used to estimate other taxes on production and net operating surplus.
• Data for subsidies are found within the FM framework.
• Export and import data are available from the FM accounts but are supplemented with data from the Balance of Payments (BOP) for exports and imports of R&D services. Transactions in existing R&D assets are not considered in this study.
• Price data were available from the IOT and from Prices Division in Statistics Canada for inputs of labour and other expenses.

44. The FM data overlap with capital expenditures on computer software that are already in the CSNA. Because of the growing importance of other intangible assets to the economy, keeping software expenditure separate from other R&D expenditure seems the preferred treatment. This implies that other R&D would be its own commodity in the IOT, and distinguished from software development. Mineral exploration, another intangible asset, also remains separate from R&D. However R&D done by the mining sector, which is not included in the exploration costs, is part of R&D expenditures. By removing software from the FM based R&D expenditures, the total FM R&D estimates for Canada are reduced by 4.2% in the 2004 reference year.

Data issues

Required input-output system detail

45. Several data issues arise with the use of Frascati data in Canada. One involves the "other current costs" data in the FM database. Total FM expenses are split into expenditure on wages and salaries, capital spending and other current costs. Information is not available on what these other current costs are but this more detailed data is required to allocate inputs in the Input-Output (IO) system. The other current costs comprise approximately onethird of total costs. One option would be to consider expanding the R&D surveys to include questions relating to these costs.

Imported research and development

46. One data concern relates to the scope of the R&D data from FM sources. The FM surveys include only producers or funders of domestic R&D. However, when non-producing units import R&D, these expenditures are not included in the FM data. For example, a company may not do any R&D of its own in Canada but may contract for R&D activities from another country. Since they are not producers of R&D they would not be in the scope of the Canadian FM-based surveys. However, these cross border purchases are collected through the Balance of Payments surveys and added to the import totals.

Multinational enterprises and exported research and development

47. A complicating issue involves data from multinational enterprises (MNEs). These companies can distribute the findings of their research across borders with little or no indication as to where the R&D was initially undertaken and where the R&D is eventually used. The Canadian practice for measuring R&D in international trade data include the use of tax records to help discern the flow of R&D across borders. Canadian firms have an incentive to report their R&D spending, since they receive tax benefits for these expenditures. The impact of these flows of international transactions on the level of the capital stock is an issue which still requires further investigation.

Multi-provincial corporations

48. A similar issue also exists in measuring regional data in Canada. Multi-provincial corporations, with R&D activity in several provinces or production units in other Canadian locations than their R&D unit, create measurement issues across provincial boundaries. Provincial data in this study are not adjusted for these trade flows. Further research is required to determine whether data is available for an adjustment to the data for these flows.

Post-secondary research and development estimates

49. One possible weakness of the Canadian FM information is that the higher education data is collected through a combination of administrative information and an estimation model. The model, necessarily, requires several assumptions in its estimation process. An example of this is data carried forward from a faculty time use survey undertaken in 1998-1999. Whether the estimates for time use are still valid is not known. An updated study on faculty time use would answer the question regarding the quality of this data.¹⁶

Producers and funders of research and development

50. The FM data available in Canada provides information on both the funder and performer (or producer) of the R&D. General government, not including universities, funds more research than it undertakes and therefore the distribution of ownership affects flows and stocks within the accounts. The allocation of the data amongst sectors could also affect productivity since, currently in the CSNA, government output does not include a full return to capital.

51. The performer (intramural) estimates in the Canadian FM data are considered of higher quality than the funder statistics (extramural expenditures)¹⁷. This is because the performer is the unit that is surveyed. The performer also has more precise information of how and when the money was used. For example, the performer may not use the full amount of funding in a given year or may have a different financial year than the funder.

52. In this study, calculations are based on the performer (producer) of the R&D for the production account. The data were then adjusted for exports, domestic sales, imports and domestic purchases so that in the sector accounts, the final owner of the R&D was attributed with the investment flow. In most cases, the performer of the R&D is also the owner. However, depending on the nature of the contract between the performer and funder, ownership rights may be shared between the two parties. This is an important distinction since in the National Accounts an asset is allocated to the sector of the owner.

Measurement issues and methodology

53. There are several measurement issues that surface in the creation of an RDSA. These include calculating output (valuation), inventory accumulation (allocation), depreciation (measuring the service lives of R&D) and deflation (calculating real estimates). Each will be discussed in turn.

Calculating output (valuation)

54. Estimates of output can be established from the FM data that would be measured at basic prices. However, since the FM captures only R&D expenditures (or at cost as denoted by the SNA93 Rev1), a method has to be developed to estimate the output of R&D.

55. The issue of valuation is not straightforward since the majority of R&D is undertaken on own-account (inhouse) and therefore is never directly sold and no market price exists. One step used to transform the FM data to an output concept involves estimating a rate of return which would typically be reflected in a market sale of an asset. The return is also known as the net operating surplus.

56. An alternative, as suggested in the SNA93 Rev1, would be to measure R&D at cost. Net operating surplus would then not need to be calculated. However, the value of R&D would not be comparable with other assets in the SNA that are based on output values at basic prices. It would also not quantify the value of transacted R&D sold.

57. Is adding operating surplus to the expenditure data the correct treatment for own-account R&D which is not sold? It could be argued that the unit collects the benefits from this R&D through the prices received on the market for the products the related R&D was instrumental in producing. It would then follow that no rate of return should be added to the R&D expenditures as this would lead to a "double-count". However, the production of the R&D and the use of the R&D asset to produce goods or services are in different periods. Also, if no return was directly added, the valuation of the R&D would change depending on whether the R&D was performed on ownaccount or whether it was bought from another unit. This could imply changes to the size of an economy if a structural shift away or towards own-account R&D occurred, which is not a desirable feature of measures of economic activity.

58. Calculating the output of R&D at cost for own-account producers would suggest including labour costs and other intermediate expenses as well as capital costs but excluding net operating surplus (and thereby excluding business profits). The cost data are known and therefore relatively straightforward to calculate. On the other hand, the data for a rate of return are not available from the FM data. A similar issue is encountered in the CSNA with own-account construction. The treatment for construction has been to value the investment at cost.

59. A further consideration is that the inclusion of a return for R&D would result in the reallocation of surplus from the activity related to the R&D to the actual R&D activity. This, in turn, would affect productivity measures since an activity may have a higher productivity measure because it includes a return that rightfully should be allocated to R&D.

Rates of return

60. If operating surplus is deemed essential for R&D, the issue then becomes one of how to establish a rate of return. In a perfect market, R&D would trade at the price that approximates the discounted value of the future stream of income it would generate. However, this information on market prices and therefore on rates of return is not available and therefore an estimation method is required.

61. The U.S. satellite account used a return of 15% based on the average of many different studies done in the U.S. This rate was higher than that of other assets. In a previous study completed by Statistics Canada¹⁸, the rate of return used was equal to that in the R&D industry North American Industrial Classification Systems (NAICS) 5415, Computer Systems Design and Related Services and 5417; Scientific Research and Development Services on R&D sold.

62. This study calculated four scenarios for estimating the rates of return. One scenario assumed no operating surplus, except for consumption of fixed capital (CFC), for own-account R&D. ¹⁹ Further, it used the rate of return of the industry in which the R&D was performed for R&D sold to other units. This method is similar to the one used in the first Statistics Canada study but differs in that the rates of return used are specific to each industry rather than the rate of return for the R&D industry. This scenario provides the base case for this study and all results refer to this case, unless otherwise specified. It was chosen as the base case since it is an approach similar to the U.S. method and corresponds to the "cost approach" as recommended by the SNA Rev1 and the data is available from the FM source.

63. Based on the work in the U.S.²⁰ and elsewhere, three other scenarios were established. In these cases, a return was added whether the R&D had been sold on the market or if it was used in house by the producer. One scenario used the rate of return of the industry in which the R&D was performed for all R&D. The other two scenarios added a premium to this rate of return to reflect the higher profits generally received by R&D intensive companies. The two premiums added an additional 5% and 10% to the industry return. The addition of a rate of return was also assumed by Mandler and Peleg²¹ in their work on bridge tables.

64. The calculation of net operating surplus is restricted to the business sector. For the government and the non-profit sectors, output equals operating costs (including CFC), following the National Accounts' convention.²²

Other adjustments to calculate output

65. Transforming FM data to SNA output requires several steps beyond that of adding a rate of return.²³ Table 3 illustrates the adjustments required and their magnitude in the Canadian SNA.

Table 3
Reconciliation between FM data and Research and Development Satellite Account, 2004

66. Output includes labour and other current costs, and these expenses are defined in the same way in the National Accounts as in the FM estimates. The FM data also include capital expenditure on land, buildings and machinery and equipment. To calculate total output, estimates of depreciation of this capital are required.

67. To depreciate the capital costs, a Perpetual Inventory Method (PIM) was used. Land was excluded from the calculation. Since detailed information of the capital used in the production of R&D was not available, certain assumptions were made. Asset lives for government and non-profit capital stock were set at ten years. Asset lives for business capital were set equal to the asset lives of the industry in which the R&D was performed. Since the consumption of fixed capital estimates are included as costs in the production of R&D, they are included as part of the investment on R&D.

68. An adjustment also was made for subsidies paid by the government for R&D. A subsidy, as defined in the SNA, is an unreturned payment given by a government unit to a business for production based on the levels they produce, sell or import. This is different than a government paying for R&D work on a contract basis. The FM data provide the details to remove subsidies from the expenditure estimates by industry.

69. The FM estimates do not include other taxes on production. Therefore an adjustment is required to value R&D at basic prices. Taxes were calculated by using the tax rate for each Input-Output (IO) industry and multiplying this by the R&D output in that industry.

70. Two sources of data, those published by Balance of Payments (BOP) and those produced by the FM system, are available for the measurement of international trade of R&D. These estimates are collected from different sources and measure theoretically different values and therefore are not equal. FM data measures the expenditures in creating R&D while BOP export data may include a margin. A reconciliation procedure was undertaken to match trade data for all industries. FM expense data were increased to reflect these new totals as the difference was considered under-coverage in output. Thus, production, investment and exports were all increased as a result of this export adjustment (see Table 3). It should be noted that the BOP and FM international trade data did differ in level; however both sources indicated the same trend as well as a net trade surplus for R&D expenditure for all years in the study.

Supply and use of research and development

71. To implement the change in treatment of R&D in the SNA, a full supply and use framework must be developed. The supply of R&D originates either from domestic production, R&D purchased in Canada or from imports of R&D services.

Table 4
Research and development supply and use, 2004

72. Once the supply of R&D produced is established, the uses of R&D can be considered. R&D can be exported, sold within Canada or used as capital in domestic production. The domestic capital or gross fixed capital formation (FCF) needs to be valued at market prices, thereby matching the valuation of other FCF estimates.

73. A reconciliation process was undertaken for domestic sales of R&D between sectors in the economy. By definition, total domestic sales and purchases from all sectors in the economy should be equal. However these two data series are not reconciled within the FM data system at the industry level. The reconciliation process was done for each industry within the business sector, for all government sectors, including the federal and provincial government and universities, as well as the non-profit sector in the RDSA. The FM "performing" data are considered more accurate than the "funding" data and therefore were used as the benchmark.

74. In the present study all purchases of R&D, whether from a domestic source or internationally imported, are considered as final R&D expenditure and not an input into other R&D work. Therefore, by capitalizing these R&D expenses, intermediate expenses will decrease by the total of these purchases. No information regarding the use of the purchased R&D is available on these transactions however this treatment mirrors what was done in the CSNA for software capitalization. This assumption does not affect the level of R&D investment.

Calculating gross domestic product

75. The previous steps calculated the supply and use of R&D including the amount of R&D investment or FCF to be added to the final expenditure. Three more steps are required to calculate the adjustment to GDP in the core accounts.

76. The first step involves an adjustment for the spending of government and non-profit institutions. R&D expenses for these sectors are already included as current expenditure in final expenditure. Therefore the change in treatment of R&D implies reclassifying the current expenditure to investment. This results in no additional impact on GDP but rather a re-allocation of expenditure.

77. The second step is then to calculate consumption of fixed capital (CFC) or depreciation using the FCF. CFC for R&D capital needs to be calculated and included in the accounts. The result is that government and nonprofit expenditure increase by the estimate of CFC resulting in a similar increase to GDP.

78. The third step is to move the valuation to market prices. This requires the addition of taxes on products and the subtraction of subsidies on products. However, since these taxes and subsidies are already accounted for in the CSNA, there is no additional impact on GDP.

Quarterly data

79. Quarterly estimates of R&D components are required for full integration in the CSNA. The FM data published by Statistics Canada include intentions on R&D expenditure up to the year 2007. As a result, timely data are available for the estimation of R&D capitalization. However, these data are only provided on an annual basis. To create sub-annual estimates, a quarterly indicator series, based on R&D employment was created. This series was derived from occupational data that had high R&D concentrations (e.g., scientist, engineers, and university professors). The list of occupations used is provided in Appendix 12. These data, gleaned from the Statistics Canada's Labour Force Survey (LFS), were initially examined on an annual basis to ensure that they provided a good indication of R&D output. Quarterly results can be found in Appendix 1.

Regional data

80. Integration into the CSNA also requires annual regional R&D data. The FM data provide a regional picture of R&D expenditure in Canada. Funder and performer data are available by province and territory. Industry breakdowns can also be calculated from the survey data. Thus the same methodology used to calculate national R&D estimates can also be used by province since complete regional Input-Output (IO) tables are also available. However, inter-provincial or inter-regional trade data for R&D are not available and no estimates were calculated for this study. Availability of this data could affect the level of R&D investment in a region. For example, if a region is a net importer of R&D, they would have more investment to capitalize. Regional results are presented in Appendix 2.

 

¹² Survey 4201, Research and Development in Canadian Industry is an example of one of several surveys undertaken to gather R&D statistics at Statistics Canada.

¹³ Carol Robbins, Linking Frascati-based R&D Spending to the System of National Accounts, 2006.

¹⁴ In the CSNA, the private non-profit sector is entitled persons and unincorporated business, and is not split any further.

¹⁵ . As denoted by SNA93 Rev1.

¹⁶ . Canadian R&D data are generally divided into natural sciences and engineering effort and social sciences. For the business sector, R&D data is only collected for natural sciences and engineering. R&D for social sciences undertaken by the business sector is therefore missing. No adjustment is made in the study for this data gap.

¹⁷ . The FM data is organized by intramural and extramural expenditures. Intramural expenditures are those performed within a unit no matter the source of funding. Extramural expenditures are those paid to another unit for R&D services.

¹⁸ . Siddiqi and Salem, A Proposal for Treating Research and Development as Capital Expenditures in the Canadian SNA, Statistics Canada, June 2006.

¹⁹ . CFC is known as capital consumption allowance (CCA) in the Canadian Accounts. CCA is dominated by consumption of fixed capital (depreciation) as well as some small miscellaneous valuation adjustments.

²⁰ . Bureau of Economic Analysis, R&D Satellite Account: Preliminary Estimates, September 2006. See also Ram Acharya, Own and Total Economy Returns to R&D: How Different Are They Across Industries?, Industry Canada, December 2006.

²¹ . Mandler and Peleg, Background and Issues Paper for the R&D-SNA Taskforce, Voorburg, April 2003.

²² . The new SNA guidelines may recommend that a return to capital also be added to government activity.

²³ . Considerable work has been done in creating bridging tables for R&D capitalization. Carol A. Robbins' paper Linking Frascati-based R&D spending to the System of National Accounts March 2006 provides a detailed account of the bridging procedure.