Old Wine in New Digital Bottles: The Challenges of Measuring the Digital Economy
梳理了数字经济时代经济测量面临的挑战，包括价格指数构建、部门分类、贸易流量和总量经济规模等，指出数字技术使标准宏观经济统计与福利之间的差距扩大，并探讨了如何改进测量以反映经济进步。
A quarter of a century after the era of the digital economy began, if this is dated to the productivity advances of the mid-1990s and the dot com stock market bubble, its measurement still presents unresolved challenges to economists and statisticians. As digital is a general-purpose technology with pervasive effects, these challenges occur across many economic measurement domains, ranging from the construction of price indices to sectoral classifications, trade flows, and measurement of the size and structure of the aggregate economy. Since the mid-1990s, and particularly since the arrival of smartphones and 3G and beyond mobile networks in 2007/8, consumption and production have been transformed. Consumers in the UK were by late 2021 spending 28 h out of the 168 h in a week online (Ofcom 2022), while the US had reached an average 23 h a week by mid-2018 (Center for the Digital Future 2018). All this activity, whether for work, home production or leisure, involves the flow of vast quantities of data. Electronic commerce has grown consistently, such that at the retail level for example, e-commerce accounts for 24% of UK retail sales (ONS 2022a), and 14.5% in the US (US Census Bureau, August 2022). Production activity has been shaped by the scope for extending supply chains (including across borders) to attain gains from specialization, involving sophisticated logistics and data flows. New digital platform business models have become widespread, including those “free” services funded by advertising that almost all of us use daily. One summary indicator of the structural transformation of the economy brought about by information and communications technologies (“digital” for short) is the increasing “weightlessness” of the economy (Coyle 1997), and the corresponding importance of intangible activities and assets (Haskel and Westlake 2017, 2022). The shift toward services, e-commerce, platform models, and the dematerialization of many products such as CDs to streamed music, or the combination of many physical goods into one mobile handset is reflected in the shrinking ratio of domestic material footprint to GDP (Figure 1). It is a useful metaphor for the transition to digital. Source: Author's construction based on ONS (2005) for 1970–1990, https://circabc.europa.eu/sd/a/2e3e7aa5-3826-40dc-a6c8-2da986b30a27/UK%20material%20flows%20review%20-%20final%20report.pdf. For 1990 on, https://www.ons.gov.uk/economy/environmentalaccounts/datasets/ukenvironmentalaccountsmaterialflowsaccountunitedkingdom. Spliced in 1990. Rebased to 1970 = 100. The implications of structural shifts in the economy for the difficulty of measuring output and productivity were noted early by Zvi Griliches (1992, 1994). He considered about 70% of US economic output fell into “hard to measure” categories by 1990, and the proportion accounted for by these categories had risen to 76% by 2019 (Coyle 2024 forthcoming)—see Table 1. He noted that while work had begun to adjust computer and semiconductor prices for the evident rapid technological progress, data collection efforts had not kept pace with rapid change elsewhere in the economy. Moreover, conceptual challenges remained in key sectors such as finance, health and professional services: “Our ability to interpret changes in aggregate total factor productivity has declined, and major portions of actual technical change have eluded our measurement framework entirely,” (Griliches 1994, 10). Thus (especially if information and communications are included), the “hard to measure” categories account for about four in every five dollars in the US economy. One way to categorize the specific measurement challenges is to consider the standard intuition that nominal output equals volume times price. Nominal output can generally be most easily measured. For some goods, volume figures may be available. More often, statistical agencies measure prices and construct price indices to divide into revenues to get “real” or volume-terms output. For some categories of items, the importance of quality change may be acknowledged and reflected in the price indices. However, the nominal = real*(quality-adjusted) price calculation is not straightforward for many activities in today's economies. Table 2 illustrates. For some digital activities, none of price, quantity or quality may be observed, or easy to define; and even relevant revenue data (e.g., at the national level) may be unobtainable. Most of the digital economy measurement challenges reflect longstanding economic questions—hence the “old wine” of the title. There is a vast literature on price indices and quality change, for example, and similarly the gap between GDP and economic welfare has been debated for decades. But the scope of the challenges is extensive given how pervasive digital technology has become, both in consumption and production. This paper does not attempt to be comprehensive in assessing these challenges. In particular, it does not address questions about how to estimate the “digital economy” as a whole, which has been piloted by national statistical offices among others (see e.g. Ahmad et al. 2017; Nicholson 2020; ONS 2022b). Nor is it a comprehensive survey of the (vast) existing literature on the digital economy. Rather, I highlight some key areas of measurement relevant to areas of interest to policymakers but principally motivated by the underlying question of how to assess economic progress in an era of significant structural change. The fundamental issue is the wedge digital is driving between standard macroeconomic statistics and aggregate economic welfare (Coyle 2014, Chapter 6). In this way, the discussion here is related to the broader “Beyond GDP” agenda, although this more often focuses on other missing welfare elements such as natural capital and human capital. These measurement issues are broad in scope but the paper divides these areas into those concerning prices, inputs, and outputs. The purpose of a price index is to aggregate many individual item prices into a single index number to deflate current-price output. For components of final consumer demand, this is done by measuring what change in prices would have kept utility constant. While it is sometimes said that real GDP is no more than a measure of aggregate economic activity, this is a fully correct statement only for nominal GDP: the deflation of the nominal aggregate by a constant-utility deflator creates a partial welfare metric in real GDP. What is more, as Thomas Schelling strikingly put it: “What we call real magnitudes are not completely real; only the money magnitudes are real. The “real” ones are hypothetical… Deflated values are inherently incommensurable, being physical volumes or index approximations to them,” (Schelling 1958, 332). In other words, real output has no natural units; when aggregated it is a representation of adding up apples, cars, haircuts, insurance, management consultancy, aspirin and all the myriad products and services in the economy, converted to a monetary metric of “real” or “constant price” dollars. The use of hedonic price indices to quality adjust some prices underlines this point about real GDP being a partial welfare aggregate. Hedonic regressions are a response to the rapid change in the quality of some goods (such as mobile phones or computers, among others) during periods of rapid technical change. As originally introduced by Griliches (1961), behind them lies the notion that a 2023 laptop is far better quality and thus delivers far more utility to the consumer from its characteristics such as processing speed or memory—than a 2003 laptop. The hedonic adjustment thus adds part of consumer surplus (linked to the identified characteristics) to the real output measure. Rosen (1974) extended this to consider the price-quality outcome from introducing production, while Triplett (1987, 2006) argued that in imperfectly competitive markets a hedonic regression would capture producer decisions about the selection and hence implicit price of quality characteristics, and also reflect producer heterogeneity. In general, both quality change and other surplus-generating aspects such as new goods, or greater variety of goods, present well-known challenges for the construction of price indices (Coyle 2024 forthcoming). A number of reports over the years looking in detail at US price indices (Stigler 1961; Boskin et al. 1996; Moulton 2018) have concluded that there is some upward bias to price indices and hence downward bias to real output due to incomplete adjustments for these phenomena. In the digital economy, rapid technological improvements, and greatly increased variety/new goods/personalization have expanded the scope of the measurement challenge. More recent studies focusing on digital goods and services specifically suggest that there is some modest upward bias in price indices (Ahmad et al. 2017; Aghion et al. 2019; Byrne et al. 2018; Reinsdorf and Schreyer 2019) but also conclude the extent of the bias has probably declined over recent years. However, investigation of specific prices affected by digital transformation suggests that it is important to consider in more detail the challenge of index construction. One illustration is the construction of an output price index for telecommunications services. Abdirahman et al. (2020, 2022) constructed a range of possible indices for this sector for the UK, motivated by the fact that the then-published official index indicated that its output price had been flat throughout the 2010s whereas the decade had seen substantial technological progress in terms of compression, speed, and so on, while data usage had soared. A unit value index of sector revenues divided by bytes of data transmitted, on the other hand, declined by about 90% over the same period. In between lay several alternative indices varying according to whether the prices for components of the services are weighted together using revenue or volume weights. The volume weighted sector index declined more because of the rapid increase in volumes, whereas the revenue weighted sector index declined less (although still substantially) because telecoms providers charge far more per byte for legacy services such as voice calls and SMS messages than for newer online services (Figure 2). Source: Abdirahman et al. 2022; Option A adds in new data services; Option B is a unit value index, £/byte; Option A.1 uses revenue weights and A3 uses volume weights. Treatment of access charges differs across A.1–A.3. Which of these is “correct”? Any of the alternatives makes inflation in the services lower and output growth higher than in the status quo, and a revenue weighted index has been incorporated into the newly double deflated UK national accounts statistics. The selection of revenue weights, the less steeply declining indices in the chart, assigns the economic value to telecoms carriers; but although they have innovated in their networks, this is surely only part of the increase in utility for users who are interested in the content carried rather than the carrier. The use of volume weights, the more steeply declining alternatives, on the other hand similarly assigns too much value to the data per se, the sheer number of bytes, rather than the informational content (and anyway, surely there are diminishing returns to bytes given the hard constraint on time available). Users get utility from the downstream services carried by the bytes and the networks over which the bytes flow. And while price per byte is still higher for legacy services, consumers are steadily switching to the low-price online services. How much economic credit is due to telecoms providers and how much to adjacent sectors of the economy? This is an open question involving significant economic issues. How should we think about price indexes for complementary digital goods? Do we need a price index for unpriced services consumed via communications networks or should their use in fact be reflected in the price index for complementary services? How is utility related to data volume and use, and is a time use approach fruitful? There are other issues related to prices affected by digital technology, reflecting the need to take careful account of how behavior has been changing. For example, the price of cloud services has fallen since their arrival (Coyle and 2019; Byrne and One issue is the shift from the capital and for to spending on cloud identified in statistics on there be by cloud including some in the same national this is to a downward on productivity during the of the shift et al. the rapid growth in use of cloud services (and data the to capture the significant price in the existing sectoral producer price index and the question of to credit the economic data in not but rather it is the activities they that A issue of weights and of economic value A UK productivity growth into sectoral components by and to the of price shifts between sectors suggests that the information and services sector is one of the sectors for a significant part of the aggregate productivity What is more, the suggests that in this sector a price shift after a substantial indices for this sector investigation in to the productivity sectors that the as the most to the aggregate are finance, some of as as sectors (Coyle and 2022; and 2022). In there has been substantial change in production and or there are challenges in measuring A number of recent on productivity among highlight the importance of both those the production and as and are not et al. 2022). For et al. the use of data in productivity in US with a significant from the use of relevant and et al. a between digital technology and level in with across digital and in UK level data that the of not is related to and with a to for those more than one digital to use data and digital to be of those that et al. There is much to about the between specific digital and level that out are and and data is as an important economic how to value it or the services it an open The most approach in is the of for activities (e.g., and the measurement challenge is that data markets are and not prices However, is to the economic value by use of if the data is use, but a significant if the is to use the data in of its from supply management to production to and in the et al. For example, accounted for a part of the increase in US which by an in the sector to its logistics and A number of other to at individual level and at more aggregate are in and ranging from the use of market statistics et al. 2019) to (Coyle and and and There is no about the to use, this with many open at given the in many about data The use of data and related digital in production to in some The importance of management for has been (e.g., et al. et al. et al. 2022). are specific also with who are more than in less even in the same 2022). In to the collection of new survey data on management (e.g., ONS 2018) and on the use of digital there questions about the of capital and it to be among rather than increasing use of digital some digital is also production The of production chains and from the on by the new information and technologies including has become in some trade flows. The recent of trade value data in the is into how production chains trade data (Ahmad and and However, trade statistics are not the only of measurement challenges. national with shifts of activity and value between It is to the extent of out of production, such that well-known (such as or not in fact or of the related of much of the value by a is by to the (e.g., In one and that of in the and sectors that they In the a statement to this most in and and and of the extent that not that they activities, these be there is a conceptual between and components for out the range of business models by In some sectors are in final goods and services may be into sectors on what they as their Source: and of a sector such shifts are way is particularly in the of major As and and has been steadily by digital as and is often to lower such as or activities are to by major and The use management such as models that capture data from construction to are sometimes as “digital some of the are components quality and for The are also with that can data to the “digital the of services to the and the is in This a quality in the services using the such as services using and the activities the supply are in sectors and so there may be some shifts among the price of services is not there be an upward bias in the price in and services, which use services or in construction when these services are by the major The issues in terms of the of economic value are to those in the of telecommunications services: digital is both the between complementary activities and the quality of related As noted real GDP is an economic welfare but an The broad scope of digital change is driving a wedge between real GDP and what be considered to be a welfare et al. have the scope of welfare as a of with market sector GDP and total GDP on but to missing capital and the production and quality adjustment for services, all the way to a broad et al. have piloted of of the on existing official statistics. the literature on expanded of aggregate activity the is to account more for the welfare of digital and thus to at the of the of For many the point has been the national For example, et al. and Ahmad et al. have digital services as part of an implicit of and data in for the on of production for activities the production such as online and and production such as open or that this shift between the market and the would the of productivity based on GDP. A number of to and measuring a welfare have or This by et al. and et al. The is to estimate the price and hence the welfare from the as in the of hedonic part of the consumer surplus is and to GDP. The of the welfare from goods is generally in this The approach has been in and for of digital goods (e.g., et al. 2020; et al. the measure of consumer surplus that can be from such and adding it to to a measure and values for a range of digital goods, with some but what is the production and should the consumer surplus from goods not be And how is the time constraint to be in using such A difficulty is the of with the of statistical production. of the existing in the literature are or not Schreyer an to GDP that the national accounts framework using the from the in et al. and this of to the of The value of this is as the value of production of a the is seen as the of value which is an to consumer this activity lies the production it is possible to a unit index for this production, quality by the scope of alternative approach is by and who the of digital as a consumption technology, in the framework of One is the of the welfare from the as to their to would be the of access to the via a mobile to the to GDP from the and data. The framework per unit of GDP to change over as they the of aggregate activity in the flow and are in in nominal terms but not when price are for As noted real GDP is an welfare so the challenge is to in the framework to welfare gains from digital. In this paper the is to the standard measure of GDP into an by adding a shift that is the standard or when prices change. the consumer technology shift would need to be to the While many of the to measuring aggregate welfare to the of the national there have been some that take a and suggest the approach of to estimate that is the of utility in monetary terms across all uses of As time use is a we our h to some assigns to work and time as an into a account work and time to The challenge is the utility from This measurement of of a literature and for a the of a monetary to The issues here are but they have not become in measurement the digital transformation to The scope of the issues is pervasive at in the economies. a number of the studies here have that has not a to the pace of price and hence to real output and this at with This paper has that the of with the of transformation lies in the welfare real GDP as is missing at part of the of digital on for work have been One the and of and output price and particularly the issue of of economic value across complementary The most areas of are the most or and sectors or those are most of their to the and use of specific One challenge in this of is access to for example, data on the price of cloud services in the here other data collection the of by statistical across standard sector is challenge. However, there are also some conceptual issues. The to the the use of to “real” whether and which broader of economic welfare are for the digital economy. The behind “Beyond GDP” statistics is and there to be that some of digital to economic welfare are not there is a broader about digital not as part of the economic There as no about how to digital welfare it is that the of of the a for welfare or The often in other areas of and there is a wedge between and welfare is a economists are with and it is not easily to statistical production. here is of a quarter century into the digital the of what statistics we need to economic and measure progress is to of the conceptual issues are new The same driving need to statistics to progress evident in the about the statistical framework in the early the extent to which aggregate statistics and should be a welfare measure. The of that of statistical can us approach the measurement challenges of today's digital economy. over the years for on digital measurement from the for the of the number the the and the is I have from from many including at the in this paper I would also particularly to on I also to for The no of