SNA/M1.20/5.5 - Price And Volume Measurement Of Goods . - United Nations

2. Sourcing current price output data on digital products2.1. Existing MaterialThe ability of digital services to ‘go viral’ raises significant questions about how statisticianstrack current price output data, particularly when sales may be through discreet websites / online stores which may be too small, until the product gains rapid market-share to be capturedin a survey, or where the provider may not be domestic. A classic example is the Pokemon Gophenomenon where sales of a particular computer game devised in one country exploded in ashort time period around the world in 2016, Whilst this individual product would obviously begrouped within software, this example signposts the key challenges which statisticianscurrently face:By going from being irrelevant in price collection terms to becoming of noticeably moresignificant weight, if only for a short period, there is an obvious question of how tocapture this firm in surveys and how to update weights for aggregation to the wholeeconomy level. By being sold via ‘appstores’, which themselves might not necessarily be domestic innature, it is obvious that the sourcing of data on these rapidly changing sales numbersis a challenge which requires statistical agencies to actively interact with thesealternative ‘market-makers’. 8In line with other products, given the preferred method to obtain volume measures of digitalproducts is by deflating their current-price output by appropriate price indices, it is essential todiscuss how to obtain the current-price output of these products. 2.1.1. Cloud computing – an exampleA core example here is cloud computing, where available estimates indicate a dramatic rise innominal output, which is forecast to continue; see Figure Two. While there are differences inthe extent to which adoption is taking place across countries, the percentage of businesses thatpurchase any cloud services can be above 40% for some countries and is above 20% for theEU-28 countries, as shown in Figure Three.As much of cloud computing is an intermediate input to production, it is hard to track in thestatistical system. Specifically, the data do not typically distinguish between cloud services andtraditional services and whether services are produced internally or purchased, or generated atthe “edge” (Byrne, Corrado and Sichel 2018). 9See also the parallel papers from the Globalisation task team.A large increase in the volume of data means that it is not feasible to transmit all of it to the cloud forprocessing in real time. Hence, businesses and governments locate the processing and storage data collectionsclose to internet providers networks, giving rise to the terminology “edge computing”, reflecting this proximity.This data streamlining solution allows the transmission of only higher-value data to a cloud centre for furtheruse.897

Figure Two: Global cloud market revenue forecast, 2017-2021.Source: Gartner (2018). Reproduced from Coyle and Nguyen (2018, p. 3)It is clear rapid product turnover and increasing product varieties are a feature of cloudcomputing services. These features cause measurement problems for even for regular products,but here there is also an increase in the use of such services, highlighting the need to focusattention on improving measurement.Cloud computing services can be thought of as a substitute for investment in computer andcommunications hardware by firms, as well as the development of own-account software.Essentially, fixed capital investment is replaced by the purchase of an intermediate input, cloudcomputing services.There are a diverse range of services provided, which can be categorized into the followingproduct classes (Byrne, Corrado and Sichel 2018; p. 6): Infrastructure as a Service (IaaS) – provides processing, storage, networks, and otherfundamental computing services, where the consumer can deploy and run arbitrarysoftware, including operating systems as well as applications. The consumer neithermanages nor controls the underlying cloud infrastructure but has control overoperating systems, storage and deployed applications, and possibly some control ofselect networking components.Platform as a Service (PaaS) – provides ability to deploy consumer-created applicationscreated using programming languages, libraries, services, and tools. The consumerneither manages nor controls the underlying cloud infrastructure including network,servers, operating systems, or storage but has control over the deployed applications8

Software as a Service (SaaS) – provides the capability of running providers’ applicationon a cloud infrastructure. The applications are accessible from various client devicesthrough either a thin-client interface (e.g. web browser) or a programme interface. Theconsumer neither manages nor controls the underlying cloud infrastructure includingnetwork, servers, operating system, storage, or even individual application capabilities,apart from limited user-specific application configuration settings.Function as a Service (FaaS) – Provides the capability of deploying functions (code) ona cloud infrastructure where an Application Programme Interface (API) gatewaycontrols all aspect of execution. The consumer (who would be a software developer) nolonger manages nor controls the underlying cloud infrastructure including networks,servers, operating systems, storage or the computing programme.Figure Three: Percentage of enterprises that buy any cloud service, comparison by EUcountries, 2015.Source: Eurostat. Reproduced from Coyle and Nguyen (2018; p. 10).9

There are a huge range of options available to consumers for each of the categories. Forexample, Amazon Web Services (AWS) provides a range of services across four regions in theU.S., with different prices by region. Their services include EC2 – Elastic Compute Cloud(renting a virtual machine from AWS), RDS – Relational Database Service (renting databasesoftware with a virtual machine) and S3 – Simple Storage Solution (renting hard disk space).Various services and pricing options are available within each. As an example, Coyle andNguyen (2018, p. 24) provide the information for EC2 compute products from AWS in Table1, where compute products are called “instances”:Table 1: Overview of AWS EC2 General Purpose Instance TypesSource: AWS press releases.AS such it is clear that before we can consider how to effectively tackle prices and volumes, itis key for statisticians to take care to address how they assemble high quality nominal outputdata, or in the absence of this, how they extract sufficient insights from their existing data tobe able to tackle measurement of the digital sector. Examples of country practices are presentedbelow, together with a discussion of the conceptual and empirical issues which need to beaddressed. Some of the country examples also discuss the methods used to obtain the current-10

price output of new digital products, such as those provided by digital intermediaries. Aproposed strategy to address these issues is also outlined.2.1.2. United States of AmericaThe US Bureau of Economic Analysis (BEA) 10 has constructed estimates of the US digitaleconomy within a supply‐use framework following a three step process. First, BEA developeda conceptual definition of the digital economy. Second, BEA identified specific goods andservices categories within BEA’s supply‐use framework relevant to measuring the digitaleconomy. Third, BEA used the supply‐use framework to identify the industries responsible forproducing these goods and services, and estimated output, value added, employment andcompensation for these industries. The BEA included in its definition of the digital economy(1) the digital‐enabling infrastructure needed for a computer network to exist and operate, (2)the digital transactions that take place using that system (“e‐commerce”), and (3) the contentthat digital economy users create and access (“digital media”).The data sources for the current-price output depend on whether the year in question in abenchmark year for the input-output tables. In benchmark years, the key data source is the U.S.Census Bureau’s Economic Census, which is conducted once every five years. In nonbenchmark years, the key data source is the annual surveys that cover selected industries, suchas manufacturing and services.2.1.3. AustraliaThe Australian Bureau of Statistics (ABS) has developed preliminary experimental estimatesof digital activity in the Australian economy using the US Bureau of Economic Analysisapproach. Under this approach, digital products were selected from ABS supply-use tables forthree broad digital activities (digital enabling infrastructure, digital media and E-commerce),after which the industry sources were identified for these digital products.The current price estimation of the digital activities involved modelling the relevant digitalproducts; compiling the gross output, intermediate consumption and value added of the digitalactivities within the source supply-use industry classifications (SUICs); and aggregating theestimated inputs and outputs across the industries.Within some of the selected supply-use product classifications (SUPCs), only certaincomponents were relevant to the conceptual measurement of digital activities. To ensureconsistency within the established scope, the supply-use gross output of such components weremodelled using other data sources such as ABS input-output table, NAB online retail salesindex (NORSI), retail turnover by industry group and ABS Business Characteristics Survey(BCS).Next, the digital activities were estimated in each of the identified primary and secondaryindustries. In a particular SUIC, digital gross output was estimated as the sum of the selectedand modelled products. Due to the lack of information on the production processes, the industryproduction function for the digital products was assumed to be identical to the “non-digital”10See Barefoot et al (2018)11

counterpart. Hence, the associated digital value added was estimated as the total value addedweighted by the share of the estimated digital output in total output. Total intermediateconsumption for the production of the digital products was calculated as the difference betweenthe digital gross output and digital value added, which was then proportionately split amongstSUPCs.Gross output, intermediate consumption and value added by digital activity were estimated asthe sum of the relevant products across the source SUICs. 112.1.4. CanadaIn 2019, Statistics Canada released the results of a study on the provisional estimates of privateshort-term accommodation in Canada. Private short-term accommodation is defined as thelisting and rental of privately-owned dwellings on a short-term basis via an intermediary digitalplatform. The input data for the study were acquired from a third-party market research firmthat specializes in providing data analytics for private short-term accommodation rentalplatforms. The acquired data included public information, such as the listing type and rentalprice, that the firm collects, via web scraping, from various short-term rental platforms. Thethird-party firm also provided additional market information, such as estimated occupancyrates and earned revenue, that they derived using their own proprietary methods. 12 Theexperience of Statistics Canada suggests that, besides using traditional surveys, nationalstatistical offices should explore alternative data sources such as third-party firms to collectinput data to estimate the current-price output of digital products.2.1.5. JapanJapan has conducted a study to construct a digital supply and use table (SUT) for 2015 usingthe framework for the digital SUT which was developed by the OECD. From the nationalbenchmark SUTs and input-output tables, the current-price output of industries operating inthe digital economy were identified and estimated.A number of approaches were used to measure the current-price output of the industries andproducts in the digital economy, with each approach dependent on the industry or productconcerned. For digitally enabling industries, the standard SUT (which was constructed usingeconomic censuses as one of the data sources) was used to identify these industries and thecorresponding current-price output. In the case of e-tailers (i.e., retail establishment with onlinesales ratio of 50% or more) and firms dependent on intermediary platforms, economic censusdata were used to identify these units using indicators such as the share of Internet transactionsof retailers and share of e-commerce before their current-price output was calculated. Thecurrent-price output of e-tailers was obtained by multiplying their margin rate by their salesrevenue at the most detailed level possible. The current-price output of firms dependent onintermediary platforms was obtained by summing up the current-price output of the individualfirms. In the case of digital platforms and digital only firms providing finance/insuranceservices, additional survey results were used to identify these firms and obtain theirMore information on the methods used is available from “Measuring digital activity”.More information on the detailed methodology and limitations encountered is available 9001/article/00001-eng.pdf.111212

corresponding current-price output. In addition, ICT services were split into (1) cloudcomputing service (paid), (2) digital intermediary service (paid), and (3) internet-ad spaceprovision service by dividing and re-organizing Internet-related services. For these services,the sub-divided Benchmark Make and Use tables were used as the basis for estimation. For thethird service, the ratio of ad revenues estimated from the “basic survey of telecommunicationindustry” was used as the basis for estimation. With regard to digital intermediary service(paid), the amount was based on the report on sharing economy services by the Cabinet Office.The Japanese experience shows how to some extent the current-price output of the digitaleconomy can be obtained using existing data sources.2.2. Options ConsideredThere are at least three key challenges in tracking and measuring the current-price output ofdigital products:Most of the businesses which produce digital products have complex and nontraditional legal structures and business models. This makes classifying the economicactivity and agents very challenging. For example, digital intermediaries perform animportant facilitation process which must be appropriately accounted for within theSNA. The facilitation activity is distinct from the actual goods or services exchangedbetween producers and consumers. As a result, even where the national accounts maycapture the economic activity, care must be taken to ensure it is classifiedappropriately. There are important questions about where some of the economic agents providing thefinal services which may be ‘digitally-enabled’ 13 are located vis-à-vis the productionboundary. Household production of such services would not be identified in traditionalbusiness surveys, but it is an open matter of debate whether such activity should becaptured in the household sector within the national accounts and how best to do this.Nevertheless, since most of this production is d

Manual on Producer Price Indices ' prepared by the IMF. 3, the Eurostat -OECD ' Methodological Guide for Developing Producer Price Indices for Services ' 4, and the ' Manual on Consumer Price Indices ' prepared by the ILO. 5 6the ' Export and Import Price Index Manual ' prepared by the IMF