Realising the Productivity Potential of ICTs

Ilkka Tuomi, IPTS

Issue: Our current understanding of ICT productivity and growth effects is based on models that are not well suited to knowledge- and innovation-based economies.

Relevance: Although it is widely believed that ICT productivity impacts became clearly visible in the second half of the 1990s in many developed countries, and that EU has been lagging the US in appropriating ICTs, such beliefs may be unsubstantiated. Effective policy-making may require new frameworks that link growth and development in order to conceptually grasp and foster productivity.

Introduction

In recent years numerous influential studies have focused on the importance of ICTs for economic growth and improvements in productivity. These studies often started out from Robert Solow's famous observation, known as the Solow paradox, that despite the extensive use of ICTs, up until the mid-1990s they do not seem to have made a noticeable impact on productivity. Recent studies have claimed that the paradox has now been solved. According to these studies, ICTs started to become visible in the second half of the 1990s, and ICT was the most important source of productivity growth in many developed countries. It has also been argued that in comparison with the U.S., Europe has been slow to appropriate the productivity benefits associated with ICTs.

For policy-makers, the central role of ICT in the modern economy means that it is important to understand the links between ICT, productivity growth, and economic development. It is therefore important to have a clear picture of what, exactly, we know about ICT productivity impacts.

A closer look at the assumptions of the econometric models that underlie our current knowledge about ICT productivity impacts reveals some interesting open issues. Below we discuss these, and argue that a broader focus on ICTs as enablers of economic development is needed to understand their growth and productivity impacts.

Information and Communication Technology in the neoclassical productivity framework

ICTs can influence productivity through three different mechanisms. Firstly, when ICT producers learn to create more output without increasing their inputs, the efficiency of the ICT producing sector increases. This improvement may appear as an increase in overall economic efficiency and so be recorded as total factor productivity growth. Secondly, when ICT using sectors invest in ICT, their labour productivity typically increases. This is because of "capital deepening" which lowers the relative amount of labour needed to produce a given output. Thirdly, the use of ICTs can make the user industries more efficient, thus increasing their total factor productivity.

In the 1990s, ICTs started to become visible in economic statistics. ICT investments increased rapidly and ICTs became a substantial part of total fixed investment in many countries. In industries that were heavy users of ICTs, capital deepening increased labour productivity, and much of labour productivity growth could be associated with these investments. In ICT producing industries, rapid technical advances become recorded as increases in total factor productivity, and much of the overall total factor productivity growth can be traced back to these advances. In this sense, ICTs become the drivers of productivity growth in the 1990s, with the U.S. leading the way.

This interpretation may, however, produce a rather misleading picture of the role of ICTs in productivity improvements and economic growth.

Conceptually, most influential studies on ICT growth and productivity impacts start from the neoclassical growth accounting framework.¹ In this framework, the growth rate of total output is shown to be a weighted sum of the growth rates of the inputs, plus a residual factor that equals the growth rate of total factor productivity. In ICT productivity studies the inputs are typically decomposed into labour, ICT-related capital, and non-ICT capital. The residual factor is often called the Solow residual. It represents growth that remains unexplained after the impact of labour and capital inputs on growth are taken into account.

To understand the essence of the neoclassical growth accounting framework, it is useful first to clarify the nature of the residual.

Technical advance, the Solow residual, and total factor productivity

Historically, the Solow residual has been associated with technical progress. The famous "productivity paradox" was associated with the fact that despite the rapid diffusion of ICTs in the 1980s, the Solow residual more or less disappeared after 1973 in the observed growth data. In other words, since the early 1970s, ICTs did not seem to have any noticeable impact on economic efficiency. In the second half of the 1990s the paradox, however, seemed to go away. To understand why this happened, we have to understand what the Solow residual actually measures.

In the neoclassical productivity framework the Solow residual is directly associated with the rate of total factor productivity growth. Total factor productivity-also known as multi-factor productivity-gives the overall efficiency of using productive inputs, most importantly labour and capital services, and-depending on the exact formulation used-land, energy and intermediate inputs.

It would be natural to expect that ICTs would reveal their productivity impact on the overall economic efficiency, and become visible in the Solow residual. The rapid measured growth in total factor productivity and its concentration in the ICT-producing sectors in the second half of the 1990s, indeed, has often been interpreted this way. Whereas the productivity paradox of the 1980s demonstrated itself in the dismal improvements in total factor productivity and the disappearance of the Solow residual, in the second half of the 1990s total factor productivity grew rapidly in the U.S. and in some other ICT-intensive economies, and the residual became visible again. This was often interpreted as the impact of ICTs.

Strictly speaking, this interpretation is not correct. When total factor productivity is consistently measured using the neoclassical productivity framework, total factor productivity improvements by definition remain unexplained "manna from heaven." In the neoclassical productivity framework, total factor productivity is not due to improvements that are paid for; instead, in this framework "technological advances" are unplanned costless improvements that are exogenous to the economic system. It is therefore important to realise that total factor productivity does not measure technical progress in any normal sense. Instead, total factor productivity measures unknown factors exogenous to the economic system, but which inherently remain beyond policy-implications-oriented frameworks used to understand growth and productivity.

In fact, in the 1950s, Abramowitz famously called the total factor productivity residual "our measure of ignorance." In the standard growth accounting framework, total factor productivity can measure earthquakes, good weather, wars, changing terms of international trade and global outsourcing, firm-level and inter-industry competition, stock-based labour compensation schemes, mismeasured working hours, fluctuations in capacity utilisation, changes in tax structure, and all other factors that are not explicitly measured. For example, in ICT industries where labour has extensively been compensated with employee stock option grants, total factor productivity will noticeably diminish in the next couple of years, as the new international accounting rules make options accountable as normal labour costs.

"Technical advance" in neoclassical productivity studies, in other words, typically means "everything that is not measured as labour and capital services." If all productive factors were to be included accurately in the neoclassical equations that describe how economic inputs are translated into economic output, total factor productivity would become a constant and its growth rate, the Solow residual, would become a stochastic error term in those equations. The policy implications of studies that show that some countries have slower total factor productivity growth than other countries, therefore, are not conceptually clear.

At present we know, however, that the measured total factor productivity growth has been strongly concentrated in ICT equipment manufacturing sectors. As total factor productivity growth has been slow outside these sectors, many researchers have argued that ICT use has not increased economic efficiency. In this sense, as Robert Gordon² puts it, "the Solow computer paradox survives intact for most of the economy." Others³ have argued that industries that invest in ICTs extensively, in fact, have shown total factor productivity improvements in recent years. Such conflicting views typically reflect differences in the data used, adjustments for cyclical factors, and variations in research methodology. In general, these conflicting interpretations, however, build on shared basic assumptions of the growth accounting framework (an issue which cannot be explored further here for space reasons).

The sources of productivity growth in the 1990s

Assuming, as ICT productivity studies normally do, that the neoclassical framework works, it is interesting to understand why, exactly, ICT in these studies seems to be a key source of growth and productivity improvement. Why did ICTs become such an important factor in the 1990s? To understand the reasons for this, we have to find the mechanisms that produce growth in these productivity studies. A somewhat surprising result is that most of the ICT-related growth is produced by researchers who put growth were they believe it should be found.

The growth accounting framework separates the contributions made by labour, ICT capital and non-ICT capital to economic output growth by generating time-series data of these different inputs. It then weights the growth rates of the inputs to derive the overall growth rate of the economy, typically measured as total value added. Using this procedure, productivity researchers can say how much the different potential growth sources actually contributed to growth. In a similar way, researchers can separate different industries and study productivity developments within industries and economic sectors.

A central question in growth accounting is how to choose the appropriate weights for the different inputs. This is the point where the neoclassical theoretical assumptions enter the picture. Theoretically, if all economic actors allocate their resources rationally and the economy is in equilibrium, producers use different inputs in ratios that reflect the marginal productivities of these inputs. Furthermore, if the economy is perfectly competitive, in theory the prices of the different inputs also equal their marginal productivities.⁴ For labour, the price for labour services equals the wage, and for capital it equals the market rental price. One possible way to weight the different inputs to the production process is therefore to multiply the working hours by the wage rate and the amount of capital services by their current market price. In this way, the economic value of inputs can be added to get a number that represents the total value of inputs.

In practice, producers often own most of the capital goods that they use in production. Capital services generated by these goods therefore do not necessarily have market prices, and productivity researchers have to estimate them. This is done using the concept of "user costs." The cost to the user is the "rental cost" that the capital owner "pays" for using the capital good.

The user cost consists of gross rate of return multiplied by the current remaining value of the invested amount. One factor in the gross rate of return is the net rate of return that the invested amount would earn if it were producing income in the overall economy. In ICT productivity studies, this net rate of return is often assumed to be about 4 percent. In addition, the gross rate of return includes depreciation that accounts for wear, tear, and other losses of productive efficiency, and a factor that accounts for the revaluation of the price of the invested capital good. Although different studies use different methods to calculate these different components of gross rate of return, for computers the annual depreciation is often calculated to be about 30 percent, and the price decline is estimated to be in the same range, leading to gross rate of return of some 65 percent.

The neoclassical productivity framework proceeds from these starting points in a relatively straightforward way. It develops time series data that describe the evolution of productive stocks of different types of capital assets and labour, multiplies these with user costs and wages, and compares the time series of economic output with the inputs. By looking at the unexplained difference between output growth and the growth of combined labour and capital inputs, it arrives at numbers that represent total factor productivity growth. By comparing the growth rate of output with growth rate of labour inputs, it arrives at numbers that represent labour productivity growth.

One particularly interesting theoretical issue has been underlying most of the results of ICT productivity studies, however. This is the way that ICT output, investments and capital are measured. It seems possible that we have considerably overestimated the growth and productivity impacts of ICT in the 1990s.

Computer price indices as the source of growth

The basic problem in measuring computer productivity impacts is that we need a good estimate of the computing services generated by computers. To isolate the impact of computer production and investments, we have to multiply the user costs of computing investments by the volume of computing. But what could be the appropriate way to measure the "volume of computing?" How can we measure the flow of services generated by computers? Should we use cubic meters, tons, electricity consumed or the number of computer boxes shipped?

Productivity researchers typically solve this problem by measuring the economic value of accumulated investments and correcting for price changes across the different years when investments are accumulated. The current stock that generates services would then equal the accumulated investments, corrected for price changes, minus depreciation of assets through wear, tear, and obsolescence.

For computers, simple price changes, however, are not enough. A typical desktop PC may cost 5 percent less this year than last, but it may also have double the hard disk capacity and a processor that is twice as fast. Computer price indices, therefore, need to be "quality adjusted." In fact, in nominal terms the median desktop computer prices have been quite stable during the last three decades, although in recent years they have dropped from about 2000 USD to about 1000 USD in the U.S.

The "volume" of computing services is calculated by accumulating "productive stocks" of computing, and assuming that the stream of computing services is proportional to the size of the productive asset. Whereas national accounts and business firms normally calculate their assets based on their current market value or historical investment value after depreciation, productivity researchers are interested in productive assets that reflect their ability to produce services.

Productive assets, therefore, become different from conventional economic assets. When researchers make adjustments that change the economic market value of computers into productive value, they actually generate most of the growth that appears in productivity statistics in the 1990s. This is illustrated in Figure 1, which shows the evolution of computer assets in the U.S., both for their current cost value that is supposed to measure the replacement value of these stocks, and for productive value, which is supposed to measure the volume of computing assets.

As can be seen in Figure 1, the value of U.S. computing assets has roughly doubled over the two decades since the 1980s, while growth in the 1990s was relatively modest. The estimated value of productive assets that generate computing services, however, grew extremely rapidly in the second half of the 1990s. This rapid growth, in fact, has been the main source of research results that show that ICTs became important for economic growth and productivity improvements in the 1990s. As the neoclassical growth accounting framework multiplies the growth rate of productive stocks with their corresponding user costs, which for computers are extremely high due to the rapid decay of computer investments, studies which include a separate ICT-capital term (i.e. breaking capital down into ICT and non-ICT) point to computer investments as the main source of growth.

One may, however, wonder whether the market really measures the value of computing as badly as implied by Figure 1. If we used the market value of computing assets instead of the estimated productive value, the neoclassical growth accounting framework would show that ICTs had a negligible impact on economic growth and productivity improvements in the 1990s.

To understand this issue, one needs to note that the difference between the two curves in Figure 1 is created mainly by price index adjustments that try to account for technical improvements in computing. The U.S. Bureau of Economic Advisors calculates these quality adjusted price indices, which are also widely used in European and international productivity studies. These price indices are "hedonic" indices that estimate price changes across time for constant quality computing products. In effect, they statistically fit dollar values for different technical characteristics of computers, such as processor speed, bus bandwidth, and hard disk size, and use these estimated parameters to calculate the price change of a bundle of technical characteristics from one year to the next. These indices are then used to adjust the market value of computers so that today's prices become comparable with yesterday's prices and can be added to get an estimate of the volume of accumulated productive stocks of computing.

Computers have been important for measured growth because computer prices have been aggressively adjusted for quality improvements. In other ICT products and services the adjustments have been much less prominent. This can be seen in Figure 2, which shows the price indices for computers, communications, software, and other products using the year 1996 as the base year. Whereas in the case of software and telecommunications equipment the productive stocks have grown almost exactly at the speed of net investments, in that of computing equipment the rates of growth have diverged radically. The reason for the rapid growth of productive computing stocks is the rapid decline in computer price indices. In neoclassical productivity studies, this decline becomes doubly influential as it affects both the size of productive assets and the user costs that multiply the growth speed of these assets.

Most European countries do not use hedonic price indices in their national accounts. As a consequence, their computer price indices decline much more slowly, in some cases showing price increases instead of declines. International studies therefore typically use the U.S. hedonic price indices to derive estimates of productive ICT assets in different countries, assuming that national statistics do not give a correct picture. These studies, however, typically do not correct the output, which usually is taken to be the GDP or industry value added as it is recorded in national accounts.

A fundamental question is whether the quality adjusted price indices lead to correct estimates of "computing volume." One may argue both from theoretical and empirical points of view that this is not the case. Theoretically correct price indices have to be "chained" within product categories across time, leading to product specific valuations of economic services, at the same time splitting the economy into numerous incommensurable "economies" where money can no longer be added. In this world, car money and computer money have different colours. This has profound implications for the economic theory of value.

Hedonic price adjustments also make the value of money dependent on technical change and rapidly changing technical characteristics. This blurs the boundaries between technical and economic worlds. This is a fundamental challenge, as economic theory was supposed to generate a theoretical system that can be studied autonomously, treating considerations about social, mental, ethical, or technical sources of values as exogenous. Such external considerations appear in Figure 1 as the difference between those productive assets whose value the analyst imputes , and the assets that the market perceives and values. Furthermore, the extremely rapid technical change in computing in effect means that these products live in a world of hyper-deflation, where conventional growth accounting methods are known to break down.

Empirically, the hedonic computer price indices most probably exaggerate the growth of computing assets. This is because they assume that improvements in technical parameters directly translate into increased computing services. This logic would mean, for example, that we are now roughly a thousand times more effective word processors than twenty years ago. An alternative explanation is that a considerable part of the decay in computer prices is in fact generated by decay. As ICT industry people sometimes say, they are in a fish business where goods start to stink if they stay on the shelves. The value of old technologies is creatively destroyed in a somewhat similar way as the latest fashion products destroy the value of yesterday's fashion. In this sense, modern ICTs are products that can simultaneously be described as durable goods and consumption goods.

Although computers have become technically much more advanced over the years, much of this progress has been consumed by increasingly complex software, and it is not clear what the net effect has been. In the networked computing world, all computing does not necessarily represent productive use. Firewalls, virus protection software and spam filters create growth and drive computer users towards faster computers, but it is not clear that these advances should be interpreted as growth of productive ICT stocks. The situation is analogous to the problems of GDP measurement, where crime, pollution, and other defensive costs become recorded as economic growth. It therefore appears that we need more research on the actual productivity impacts of ICTs.

Conclusion

ICTs are composite goods that consist of hardware, software, skills, systems integration, operational support, and infrastructure. The productive use of ICTs often requires organisational and working practice changes, and depends on contextual factors, such as transport infrastructure, cultural values, and the routines organising everyday life. It is therefore difficult to isolate ICT productivity impacts using the traditional productivity frameworks that allocate productivity improvements to specific investments. ICT investments become productive in combination with other investments and often through recombination of existing assets for new uses. This does not mean that ICTs would be irrelevant for economic growth and productivity. ICTs became a fundamental element of the economy and society in the 1990s. However, a closer study of ICT productivity impacts also reveals that our current concepts of economic growth and productivity perhaps address the economic impact of ICTs only in a somewhat limited sense. We therefore may need to rethink why productivity was understood to be such a central concept for policy and what, exactly, we mean by productivity and growth in the knowledge economy.

One way to move towards a new paradigm of productivity could be found, for example, by studying the growth impacts of ICT using Amartya Sen's capability-based model of economic development. This framework could help policy-markers to describe what types of technical change could reasonably be called development. It could also allow us to describe how ICTs augment and enhance those basic capabilities that are fundamental for economic development.⁵

Keywords

ICT, productivity, Solow paradox, growth accounting, classical model

Notes

1. Cf., Schreyer, 2001.

2. Gordon, 2000:57.

3. E.g. Van Ark, Melka, et al., 2002.

4. For a clear and compact elaboration of these mathematical relations, see Schreyer, Bignon, & Dupont, 2003.

5. This possibility has been discussed in more detail in Tuomi, 2004.

References

• Gordon, R.J. Does the "New Economy" measure up to the great inventions of the past? Journal of Economic Perspectives, 14 (4), 2000 pp.49-74.

• Schreyer, P. OECD Productivity Manual: A Guide to the Measurement of Industry-Level and Aggregate Productivity Growth. Paris: OECD Statistics Directorate, National Accounts Division, 2001.

• Schreyer, P., Bignon, P.-E., & Dupont, J. OECD capital services estimates: methodology and a first set of results. OECD Statistics Working Paper 2003/6, published 2003. http://www.oecd.org/statistics/productivity

• Tuomi, I. Knowledge society and the new productivity paradigm: a critical review of productivity theory and the impacts of ICT. IPTS Working Paper, February 2004. http://www.cordis.lu/ist/about/socio-eco.htm

• Van Ark, B., Melka, J., Mulder, N., Timmer, M., & Ypma, G. ICT Investment and Growth Accounts for the European Union, 1980-2000: Final Report on "ICT and Growth Accounting" for the DG Economics and Finance of the European Commission, Brussels, 2002. http://www.eco.rug.nl/GGDC/dseries/Data/ICT/euictgrowth.pdf

Contact

Ilkka Tuomi, IPTS

, :. E-mail: Contact Form

About the author

• Ilkka Tuomi graduated in Theoretical Physics at the University of Helsinki, and has a Ph.D. on Adult Education from the same university. His recent research has focused on innovation, open source, ISTs, and the knowledge society. Before joining the IPTS as a visiting scientist he was a visiting scholar at the University of California, Berkeley, and from 1987 to 2001 with Nokia Research Center, most recently as Principal Scientist, Information Society and Knowledge Management.

Contents of issue 85