Economics of data (Part 4: The data economy)

Economics of data (Part 4: The data economy)

Phuah Eng Chye (29 August 2020)

Does the new information economy have new economics, in the long run? When the economy shifted from agrarian to industrial, economists focused on capital accumulation and removed land from production functions. As we shift from an industrial to a knowledge economy, the nature of inputs is changing again. In the information age, production increasingly revolves around information and, specifically, data. Many firms are valued primarily for the data they have accumulated. As of 2015, global production of information and communications technology (ICT) goods and services was responsible for 6.5% of global GDP, and 100 million jobs…How will this new data economy evolve? Because data is non-rival, increases productivity and is freely replicable (has returns to scale), current thinking equates data growth with idea or technological growth. This article uses a simple framework to argue the contrary, that data accumulation is more like capital accumulation, which, by itself, cannot propel growth in the long run. Maryam Farboodi, Laura Veldkamp (2019) “A growth model of the data economy”.

“The rise of information technology and big data analytics has given rise to the new economy. But are its economics new?” Laura Veldkamp and Cindy Chung suggest there is “evidence of all these changes in the literature…But to predict how far-reaching these changes will be and how large they will eventually become, requires a framework for prediction. In the midst of structural transformation, projection of past trends into the future is not a reliable guide. Instead, we need structural models and theories to guide our thinking about the nature of this change…While data is affecting every corner of the economy, and evidence is accumulating that its effects are transformative, theories exploring the role and value of data are only just emerging”.

Data and technology are not the same

The contribution of technology to value is recognised in economic models but this cannot simply be applied to data. However, Maryam Farboodi and Laura Veldkamp makes the important distinction that “data and technology are not the same. They entail different benefits and different costs…The benefits of forecasting[1] and inventing technologies differ: Data and technology also have different production costs. Producing new technology requires resources: skilled labor, a laboratory and prototypes. In contrast, data is a by-product of economic activity…data itself is not produced in a lab. More data comes from more economic activity. This difference in production matters. One of the fundamental insights[2]…is that monopolies are necessary to incentivize idea production. This is not true of data production. Because data is a by-product of economic transactions and data is less prone to leakage, no extra incentives are needed for its production”.

Laura Veldkamp and Cindy Chung note “ideas or technologies leak…When they are hired away by competitors, Silicon Valley workers take their technological knowledge with them. But data is not embodied in one’s mind…A worker might steal data from their firm. But that is a crime. The data is not embodied in their human capital. This feature is also what distinguishes data from human capital or forms of learning-by-doing. Conversely, some ideas are illegal to take from one firm to another. These are ideas that are protected by patents. But patents do not protect data. There are no legal institutions designed specifically to ensure the exclusive right of one entity to use a particular set of data. At the same time, because data does not easily leak, and because protection is not needed to incentivize the creation of data, such an institution is probably not needed. Lower leakage of knowledge encoded as data is important for growth”. “If data lends itself to less diffusion than traditional technologies, then the growing data economy could be responsible for the decline in firm dynamism”.

Laura Veldkamp and Cindy Chung adds “the ability to monetize data and the widespread sale of data also distinguishes it from technology. A few features of data lend themselves well to such transactions. First, a seller can clearly describe the contents of a data set, without revealing its information content[3]. Second, data can be easily split[4]…data is less likely to leak…data can be sold both directly and indirectly[5]…Such services monetize data, without transferring the underlying data used in the service”.

Firm level effects

These distinctions are evident at the firm level. First, Maryam Farboodi and Laura Veldkamp point out “the way in which data is produced means that, at low levels, data has increasing returns…More data makes a firm more productive, which results in more production and more transactions, which generates more data, and further increases productivity and data generation…It is the dominant force when data is scarce, before the diminishing returns to forecasting set in and overwhelm it. This increasing-returns force can generate a growth poverty trap. Firms, industries, or countries may have low levels of data, which confine them to low levels of production and transactions, which make adopting new data technologies not worth the cost. Thus, data could explain some of the increase in inequality, across firms, industries, or even countries that adopt data technologies at different rates. While the increasing returns to data can generate firm size inequality, eventually the decreasing returns to data cause firms to converge. Thus, the properties of data suggest that the part of firm size dispersion that is due to data is a transitory phenomenon”.

Second, Maryam Farboodi and Laura Veldkamp argue “when data is scarce, little is lost due to depreciation. As data stocks grow large, depreciation losses are substantial. The point at which data depreciation equals the inflow of new data is a data steady state. Firms with less data than their steady state grow in data, and therefore in productivity and investment. If a firm ever had more data than its steady state level, it should shrink. But without any other source of growth in the model, data-driven growth, like capital-driven growth eventually grinds to a halt”.

On this point, Wendy C.Y. Li, Makoto Nirei and Kazufumi Yamana note that “unlike R&D assets that may depreciate due to obsolescence, data can create new value through data fusion and innovations in data-driven business models – unique features that create unprecedented challenges in measurement”. In addition, they suggest that “as AI is becoming cheap, data decide the overall power and accuracy of an algorithm, and so are vital for firm competitiveness”.

Overall, Maryam Farboodi and Laura Veldkamp conclude “the economics of transactions data bears some resemblance to technology and some to capital. It is not identical to either. Yet, when economies accumulate data alone, the aggregate growth economics are similar to an economy that accumulates capital alone. Diminishing returns set in and the gains are bounded. Yet, the transition paths differ. There can be regions of increasing returns that create possible poverty traps. Such traps arise with capital externalities as well. Data’s production process, with its feedback loop from data to production and back to data, makes such increasing returns a natural outcome. When markets for data exist, some of the effects are mitigated, but the diminishing returns persist. Even if data does not increase output at all, but is only a form of business stealing, the dynamics are unchanged. Thus, while the accumulation and analysis of data may be the hallmark of the new economy, this new economy has many economic forces at work that are old and familiar”.

Intangible assets and economic measurement

Laura Veldkamp and Cindy Chung notes “data is also a form of intangible capital, though it is distinct from other forms of intangible capital because it poses different measurement issues. Intangible capital measures generally use the cost of investment to value the intangible capital stock. But data is a by-product of activity, with little or no creation cost…so it is often not counted as having positive value in the intangible capital stock…If a firm buys data, it is clear that it should be valued at its market price. But if the firm produces its own data, its value may be determined at the firm’s discretion”. Depending on whether data is a service flow or capital transaction, this can “change the timing of the measurement of economic value” in calculating GDP.

Laura Veldkamp and Cindy Chung highlights “not all the gains in well-being arising from digital goods and services are captured by measures of gross domestic product (GDP). One missing component from national accounts is the value of zero-price goods, which are prevalent in the digital economy…Often, such zero-price digital goods are not truly free…service is being bartered for personal data. Barter is not measured by GDP. This is reaffirmed by various studies cited in their analysis that reflect the increased consumer surpluses or output-savings not captured in GDP measurement”. “Since these goods reduce our need for value that would otherwise be measured in GDP and are not themselves counted, they bypass GDP and create consumer surplus directly”. In relation to this, there has been attempts[6] to construct new metrics to capture their impact on GDP.

Overall, measurement constitutes a major challenge to modelling the data economy. Higher quantities of data do not equal to higher output. Effects such as Baumol’s cost disease and zero-pricing confound attempts to decompose nominal value into inflation and output. In a data economy, the ambiguity differentiating inflation and output growth is not easily resolved.

Long-term economic growth models and Baumol’s cost disease.

Estimates for the value of firm-level data provide a basis to modify variables for productivity, capital, human capital and investment to feed into long-term economic growth models. In this regard, labour or tangible capital is substituted by intangible capital to assess the consequential effects; particularly on the capital-labour share of income.

While data appears to have a significant impact on economic growth, there are constraints. Laura Veldkamp and Cindy Chung suggest “infinite data brings perfect foresight. Foresight makes firms more profitable…But it does not allow a firm, and certainly not the aggregate economy, to produce arbitrarily large quantities or value of real goods. Only innovation can do that”. “The bounded benefits to data imply that there must be decreasing returns at some point. This does not mean that data will not facilitate growth. It may be that data is a complement to innovation. But, just like capital, data alone can only achieve a level improvement in GDP, not a sustained increase in the rate of growth. Whether data complements or substitutes for innovation is an important open question for future theoretical and empirical research…New work is needed to bring the richness of the growth approach together with the acknowledgment that data is information and to deliver a fuller picture of the reality and the potential of the new data economy”.

The role of data in raising economic growth is potentially elevated by AI. In tandem with this, Philippe Aghion, Benjamin F. Jones and Charles I. Jones note “models that conceptualize AI as a force of increasing automation suggest that an upswing in automation may be seen in the factor payments going to capital – the capital share”; and that the correlation between the rise in capital share of GDP with the adoption of robots “has been a central topic of research”.

Philippe Aghion, Benjamin F. Jones and Charles I. Jones argue “Baumol’s cost disease[7] leads to the traditional industries’ declining share of GDP as they become automated, which is offset by the growing fraction of automated industry. This explains the observed stability in the capital share and per capita GDP growth over the past century despite evolving automation. In their model, labor share remains substantial because labor represents a bottleneck for growth…AI can increase growth of new ideas and potentially obviate the role of population growth in generating exponential economic growth. Nevertheless, even though AI can theoretically generate singularity – a notion that artificial super intelligence could trigger runaway technological growth and propel infinite income in finite time – growth may remain limited due to essential areas of production that are hard to improve. Moreover, AI may discourage future innovation for fear of imitation, undermining incentives to innovate in the first place”.

Technological singularity

It has been anticipated “that rapid growth in computation and artificial intelligence will cross some boundary or Singularity, after which economic growth will accelerate sharply as an ever-increasing pace of improvements cascade through the economy”.  William D. Nordhaus notes “at the point where computers have achieved superintelligence, we have reached the Singularity where humans become economically superfluous in the sense that they make no difference to economic performance”.

William D. Nordhaus however emphasises that “rapid growth in the productivity of computers or information technology…has no necessary implication for aggregate economic growth…Consumers may love their iPhones, but they cannot eat the electronic output. Similarly, at least with today’s technologies, production requires scarce inputs (stuff) in the form of labor, energy, and natural resources as well as information for most goods and services”.

William D. Nordhaus highlights “the empirical question is the degree of substitutability between information and human labor. His analysis is “that information and computers will come to dominate the economy only if the information inputs or outputs take a rising share of consumption or inputs. This requires that the expenditure shares or input cost shares of information rise over time, which in turn requires that the volume of expenditures or inputs rise more rapidly than the relative prices fall”. He generally found growth trajectories for the share of capital in total income and the share of informational capital in total capital to be relatively slow. He tentatively concludes based on “projecting the trends of the last decade or more, it would be in the order of a century before these variables would reach the level associated with the growth Singularity”.

Conclusion: The government’s role in shaping the data economy

The data economy is decidedly different from the industrial economy. Blayne Haggart argues governments must recognise “the data-driven economy runs according to a different logic than one that prioritizes finance or production. Consequently, policy making designed to maximize employment and economic activity in a production-based economy will not necessarily have the same effects when targeting the data-intensive giants of the information age”. In this regard, governments will play a major role in shaping outcomes as the degree of government involvement or inaction will determine whether “this market functions in a socially optimal manner rather than in the interests of its most powerful actors”.

At the core is the notion that “a data-driven economy is founded on the ability to control data. Who controls data, who decides what data is worth collecting and how data is used are therefore key political questions with society-wide ramifications…Such control over data can also be used to create relations of economic dependency that more closely resemble feudal economies than free markets…Resolving these issues will necessarily create winners and losers”.

There are spillover effects into other policy areas. Blayne Haggart notes while free trade policies may make sense for a manufacturing-based economy, allowing data and intellectual property to flow freely across borders raises privacy and global anti-competitive network concerns. Hence, one should not rely on inappropriate analogies (i.e. free trade) and the consideration of “some restrictions on cross-border data flows may make economic sense”.

An unfortunate observation is that “constant surveillance is also fundamental to the functioning of internet-connected devices that work only with a constant data stream. A data-driven economy, in other words, is also a surveillance economy”. Blayne Haggart explains “the logic in the security and economic cases is the same: in a knowledge economy, anything less than total surveillance is seen as a potential threat or economic loss. In a surveillance economy and society, therefore, effective democratic oversight of both the state and economic actors is essential to resolving the tension between the threats posed by such surveillance and the necessary role of surveillance in enabling the data-driven economy”.

The fallacy of relying on traditional economic models to extrapolate the future and to guide policy-making is thus exposed. A broad range of policy responses to challenges such as globalisation and ideologies are pending recalibration but are handicapped by the fact that traditional policy metrics such as physical output and trade are becoming less meaningful. There is also a need to more closely examine the macroeconomic role of data in allocating resources and in creating and distributing value. In particular, there is a need to grapple with the challenge posed by the corrosive effect of datafication on fiscal revenues.


Blayne Haggart (5 March 2018) “The government’s role in constructing the data-driven economy”. Centre for International Governance Innovation.

Laura Veldkamp, Cindy Chung (30 October 2019) “Data and the aggregate economy”. Journal of Economic Literature.

Maryam Farboodi, Laura Veldkamp (29 October 2019) “A growth model of the data economy”.  Working Paper, MIT.

Philippe Aghion, Benjamin F. Jones, Charles I. Jones (October 2017) “Artificial intelligence and economic growth”. NBER.

Phuah Eng Chye (2015) Policy paradigms for the anorexic and financialised economy: Managing the transition to an information society.

Phuah Eng Chye (26 August 2017) “The services economy: Revisiting Baumol’s cost disease”.

Phuah Eng Chye (7 July 2018) “Labour share of income (Part 7 – The role of wages and profits)”.

Phuah Eng Chye (18 July 2020) “Economics of data (Part 1: What is data?)”.

Phuah Eng Chye (1 August 2020) “Economics of data (Part 2: Market approach to valuing data)”.

Phuah Eng Chye (15 August 2020) “Economics of data (Part 3: Relationship between data and value and the monetisation framework)”.

Wendy C.Y. Li, Makoto Nirei, Kazufumi Yamana (23 July 2019) “Value of data: There’s no such thing as a free lunch in the digital economy”.

William D. Nordhaus (9 September 2015) “Are we approaching an economic singularity? Information technology and the future of economic growth”. Cowles Foundation Discussion Paper.

[1] The use of data to forecast have finite gains and therefore yield diminishing returns.

[2] Attributed to Paul Romer.

[3] “A buyer can know exactly how many users or clicks or transactions…and still not know what the data will say”. See Laura Veldkamp and Cindy Chung.

[4] Data quantities can be adjusted depending on how much buyers are willing to pay.

[5] The data can be sold directly or as part of a service.

[6] See Laura Veldkamp and Cindy Chung.

[7] See “The services economy: Revisiting Baumol’s cost disease”.