# Archived — ICT and Total Factor Productivity Growth: Intangible Capital or Productive Externalities?

## Archived Information

Information identified as archived on the Web is for reference, research or recordkeeping purposes. It has not been altered or updated after the date of archiving. Web pages that are archived on the Web are not subject to the Government of Canada Web Standards. As per the Communications Policy of the Government of Canada, you can request alternate formats on the "Contact Us" page.

The views and opinions expressed in the report are those of the author alone and do not represent, in any way, the views or opinions of the Department of Industry or of the Government of Canada.

## Abstract

What accounts for the exceptional total factor productivity (TFP) growth performance of the United States and to some extent some of the other OECD countries after the mid-1990s? Most commentators have pointed to enormous productivity gains in the production of Information and Communications Technology (ICT) as the answer. But according to standard neoclassical theory, technical progress in one industry should not raise TFP growth in other industries. Yet the TFP acceleration is due mostly to industries that use, but do not produce, ICT capital. This paper investigates two explanations for this apparent puzzle, one based on the existence of intangible capital that is not measured in the National Income Accounts, and the other based on productive externalities.  While both explanations can match the observed behavior of TFP growth, the two have very different implications for economic policy and welfare. We show that the two explanations can be distinguished using a cross-country, cross-industry data set. Using newly constructed very comprehensive data covering 16 OECD countries for 24 industries for a period of 32 years, we find evidence of intangible capital accumulation, but no evidence of positive spillovers to ICT investment. These results are robust across different estimation techniques and under different assumptions regarding market structure.

## I. Introduction

After the mid-1990s, both labor and total factor productivity (TFP) accelerated in the United States and, to a lesser extent, in some other OECD countries.  A large body of work has explored the sources and breadth of the U.S. acceleration and suggested reasons for the exceptionally good performance of the U.S. economy.  Much of this research focuses on the role of information and communications technology (ICT)Footnote 1  Jorgenson, Ho and Stiroh (2006), for example, argue that the overall increase in the U.S. "speed limit" for growth is due to ICT. Growth accounting at an industry level for data from 1987-2004 shows that the simple explanation based on only ICT-producing industries for the U.S. TFP acceleration is incomplete at best. A growing body of literature finds that the TFP acceleration was, in fact, broad-based—not narrowly located in ICT production.  Using industry-level data for the United States, Corrado et al (2006), and Bosworth and Triplett (2006) find that non-ICT-producing sectors saw a sizeable acceleration in TFP in the 2000s, whereas TFP growth slowed in ICT-producing sectors in the 2000s.

This finding is a puzzle.  From the perspective of neoclassical economics, which underlies almost all the recent discussions of this issue, there is no reason to expect acceleration in the pace of TFP growth outside of ICT production. According to this theory, the fall in input prices do not shift production functions of the output sector. Of course, the fall in price leads to ICT capital deepening throughout the economy boosting labor productivity in ICT-using sectors—but does not change TFP in sectors that only use but do not produce ICT.

However, there are potentially two channels that the fall in the price of ICT could affect TFP of ICT-using industries. The first channel is that the resulting ICT deepening may lead to more use of complementary intangible capital, and the second channel is that there might be presence of positive externality of ICT use.Footnote 2 Let us discuss both lines of argument further.

Firm-level studies suggest that benefiting from ICT investments requires substantial and costly co-investments in complementary capital, with long and variable lags.Footnote 3  For example, firms that use computers more intensively may reorganize production, thereby creating 'intangible capital' in the form of organizational knowledge.  The use of ICT may also prompt for more R&D investment some part of which might be intangible as well. The resulting "organizational capital" is analogous to physical capital that companies accumulate in a purposeful way.  Conceptually, we think of this unobserved complementary capital as an additional input into a standard neoclassical production function.Footnote 4  In addition to the firm-level studies cited above, macro studies also argue that complementary investment is quantitatively significant (see, e.g., Laitner and Stolyarov, 2003).

The literature also suggests the likelihood of sizeable externalities to ICT.  For example, successful new managerial ideas—including those that take advantage of ICT, such as the use of a new business information system—seem likely to diffuse to other firms.  Imitation may be easier and less costly than the initial co-invention of, say, a new organization change, because one learns by watching and analyzing the experimentation, the successes and, importantly, the mistakes made by others.Footnote 5

The first set of considerations is completely consistent with the growth accounting framework but suggest that the production function is mismeasured because we don't observe all inputs (the service flow from complementary, intangible capital) or all outputs (the investment in complementary capital).  Hence, TFP is mismeasured. The second set of ideas, related to externalities, suggests that ICT might also explain "true" technology change (although the change would be endogenous, not exogenous).  Empirically, the challenge is to infer the presence of ICT externalities while allowing for the existence of unobserved complementary investments.

This paper suggests a method to distinguish between these two explanations for the speedup of TFP in ICT-using industries. For that, the paper develops an estimating equation based on the model initiated by Basu, Fernald, Oulton and Srinivasan (2003; henceforth BFOS). In the BFOS model, reaping the full benefits of ICT requires firms to accumulate a stock of intangible knowledge capital.  And, according to this model observed investments in ICT are a proxy for unobserved investments in reorganization or other intangible knowledge.

Note that the BFOS story hews as closely as possible to neoclassical assumptions while explaining the puzzle of TFP growth in ICT-using industries.  If growth accounting could include intangible capital as an input to production then it would show no technical change in ICT-using industries.  (Of course, measuring intangible capital directly is very difficult at best; see Corrado, Hulten and Sichel (2006).)  But the story can easily be extended to include non-neoclassical features that would explain true technical progress in ICT-using industries via other mechanisms, such as externalities.  Indeed, to the extent that much of the intangible capital accumulated by ICT users is knowledge, which is a non-rival good, it would be natural to expect externalities.  For example, the innovations that have made Amazon.com and Wal-Mart market leaders could presumably be imitated at a fraction of the cost it took to develop these new ideas in the first place, at least in the long run.

Unfortunately, once one allows for the existence of intangible capital, accumulated in proportion to investment in ICT, it is very difficult to detect externalities using conventional techniques.  The paper documents this basic identification problem.  But allowing for intangibles is important, as a variety of papers have made a strong case for its existence.  The paper then suggests that international industry data can help distinguish the effects of intangibles from true externalities.  This method is particularly appropriate for OECD countries which are more trade oriented and are leaders in the business applications of ICT. Accordingly, we use data for 16 OECD countries () for 24 industries covering the period 1973 to 2004.Footnote 6

Indeed, we find evidence that firms accumulate intangible capital in the way our theory predicts.  We find evidence of positive spillovers to ICT investment neither within a country nor across national boundaries.  Reassuringly, we do find positive and statistically significant spillover effects of R&D investment at the domestic level and also at the foreign level in recent decades.  These findings are robust with different estimation techniques. The result that there is no ICT spillover remains unchanged whether we impose or relax the assumptions of perfect competition and constant return to scale. However, the presence of intangible capital story survives only if we do not impose the assumptions of CRS and perfect competition.

The rest of the paper is structured as follows.  First, we review the basic intangible-capital model presented in BFOS, and derive an estimating equation.  In Section III, we show that it is difficult to use the basic BFOS framework to identify externalities—there is an identification problem.  We then show that the problem can be solved by using cross-country, cross-industry data.  Section IV gives an overview of the data we use.  Section V discusses the results. Section VI concludes.

top of page

## II. The Basic Model

We now turn to a formal model which is based on the model by Basu, Fernald, Oulton and Srinivasan (2003; henceforth BFOS). In the BFOS model, capturing the full benefits of ICT requires firms to make intangible investment. The assumption that complementary investments are needed to derive the full benefits of ICT is supported both by GPT theory and by firm-level evidence.Footnote 7 These investments may include resources diverted to learning; they may involve purposeful innovation arising from R&D. Since (intangible) capital accumulation is a slow process, the full benefits of the ICT revolution show up in the ICT-using sectors with significant lags.

Formally, value added in industries $i$ at time $t$ that use, but do not produce, ICT is given by (we suppress the country subscript):

(1) ${Q}_{it}\equiv {Y}_{it}+{A}_{it}=F\left({Z}_{t}G\left({K}_{it}^{IT},{C}_{it}\right),{K}_{it}^{NT},{L}_{it}\right),i=1,\dots ,N$

where $Q$ is total output; the difference in terms of measurement is that $Y$ is observable output by national accountants, and $A$ is the investment flow that is not measured.Footnote 8 It is the time and resource cost of training and creating new business structures.Footnote 9 Each industry hires labor L and rents ICT capital ${K}^{IT}$ and non-ICT capital ${K}^{NT}$ in competitive, economy-wide markets. $Z$ is a technology term that each industry takes as exogenous. $F$ and $G$ are homogeneous of degree 1 in their arguments. For simplicity, we ignore materials input, imperfect competition, increasing returns, and capital adjustment costs. All could be added, at the cost of considerable notation. But it is straightforward to include all of these features in the empirical work, and we do.

Industries forego producing market output $Y$ to accumulate complementary $C$ capital as follows (for the rest of the theoretical model, we suppress industry subscript $i$):

(2) ${C}_{t}={A}_{t}+\left(1-{\delta }_{c}\right){C}_{t-1}$,

where ${\delta }_{C}$ is the depreciation rate of investment $A$. The economic difference between $A$ and $NT$ capital is that they interact in different ways with ICT capital. The main economic implication of the separability assumption built into is that the marginal productivities of ${K}^{IT}$ and $C$ are closely tied to one another. We assume that the elasticity of substitution between the two inputs in the production of $G$ is relatively small. We also assume Inada-like conditions to the effect that the marginal productivity of each input is very low if the level of the other is close to zero. Thus, when the GPT arrives and ICT capital starts getting cheap, the incentive to also accumulate $C$ is very strong.

Note that conceptually, innovation as traditionally construed can take two forms. First, we lump purposeful unmeasured innovations into $C$ (indeed, we have assumed that all purposeful innovation is closely linked to ICT). Second, we interpret $Z$ as all exogenous increases in technology, including the component of organizational change that spills over as an externality from the sector of origin for example, the idea of using individual electric motors at each workstation in a factory, rather than relying on the single drive train of a steam engine.

Differentiating (1), imposing constant returns to scale and perfect competition assumptions, and manipulating the expression algebraically, we have an expression for the measured Solow residual:

(3) $\Delta {y}^{NT}-\frac{{P}_{{K}^{IT}}{K}^{IT}}{PY}\Delta {k}^{IT}-\frac{{P}_{{K}^{IT}}{K}^{NT}}{PY}\Delta {k}^{NT}-\frac{WL}{PY}\Delta l\equiv \Delta TPF=\frac{{F}_{C}C}{Y}\mathrm{\Delta c}-\frac{A}{Y}\Delta a+{s}_{Z}\Delta z$,

where $P$, $P,{P}_{{K}^{IT}},{P}_{{K}^{NT}}$ are prices of output, ICT capital and non-ICT capital respectively; $\Delta x=d\mathrm{log}X}{dt}$, and ${s}_{Z}\equiv \left({F}_{Z}Z}{Y}\right)$. Note that in this model, TFP growth is not equal to technological change (as in traditional neoclassical model), that is, $\Delta T\mathrm{FP}\ne {s}_{Z}\Delta z$ . Hence, omission of complementary inputs can cause either overestimate or underestimate of TFP growth. When unmeasured output is growing $\left(\Delta a>0\right)$, TFP growth is underestimated (the 1974 story) as resources are diverted to investment. When unmeasured input is growing $\left(\mathrm{\Delta c}>0\right)$, TFP growth is overestimated. This point is simple but important. Of course, if one corrects only output mismeasurement $\left(\Delta a\right)$, then ICT will appear fantastically productive, far beyond what is ordinarily measured. But firms divert resources to unobserved investment $\Delta a$ in order to create an intangible capital stock, which contributes to future production. The resulting unmeasured flow of capital services implies a bias in the other direction.

The net bias may be either positive or negative at a point in time, but in a dynamically efficient economy the mismeasurement is necessarily positive: True steady-state TFP growth is lower than measured, not higher.Footnote 10 In steady state, of course, the accumulation equation implies that $\Delta c=\Delta a$. Hence, steady-state mismeasurement depends on $g$, the steady-state rate of growth, and ${r}^{*}$, the steady-state real interest rate.

We now seek an observable proxy for unobserved investment in, and growth in the stock of, complementary capital. As shown in Appendix B, with the assumption that the production function of $G$ is CES in inputs $C$ and ICT, equation (3) becomes:

(4) $\Delta TF{P}_{t}=\left[{F}_{C}-1\right]\beta {\stackrel{˜}{k}}_{t}^{IT}+\left[\frac{\left(1-{\delta }_{C}\right)}{1+g}\right]\beta {\stackrel{˜}{k}}_{t-1}^{IT}+{s}_{G}\Delta {z}_{t}$

where $\beta ={\left(\frac{1-\alpha }{\alpha }\right)}^{\sigma }\left(\frac{P}{{P}_{C}}\right){\left(\frac{{P}_{{K}^{II}}}{{P}_{C}}\right)}^{\alpha -1};{\stackrel{˜}{k}}_{t}^{IT}={s}_{{K}^{IT}}\left[{k}_{t}^{IT}+\sigma \Delta \mathrm{ln}{\left(\frac{{P}_{{K}^{IT}}}{{P}_{C}}\right)}_{t}\right]$, and ${s}_{{K}^{IT}}={P}_{{K}^{IT}}{K}^{IT}}{PY}$

So ceteris paribus the mismeasurement of complementary capital is more important in those industries where ${s}_{{K}^{IT}}$, the share of ICT in revenue, is high.

Finally, the TFP growth represents more than just pure technological change, including positive externality created by some factors. In accordance with this fact, we decompose the $\Delta z$ term to represent externalities, $\Delta e$, as well as exogenous technical change, $\Delta t$. So far we have suppressed industry subscript $i$, but now need to introduce them for estimation using a panel of industries (we still continue to suppress country subscript). Hence,

(5) $\Delta TF{P}_{it}=\left[{F}_{C}-1\right]\beta {\stackrel{˜}{k}}_{it}^{IT}+\left[\frac{\left(1-{\delta }_{c}\right)}{\left(1+g\right)}\right]\beta {\stackrel{˜}{k}}_{i,t-1}^{IT}+{s}_{G}\Delta {e}_{it}+{s}_{G}\Delta {t}_{it}$

This model has several general implications. First, one might find a link between ICT use and measured TFP even if there are no externalities to ICT use. Second, the correct proxy for ICT use involves the interaction of ICT-intensity (the ICT share) and the growth rate. Third, one needs to control for both current and lagged $\stackrel{˜}{k}$ Since these values are correlated in the data, if one omits one of them, then the regression has an omitted variable problem. Fourth, the first term on the right-hand side of (5) is proportional to $\left({r}^{*}+{\delta }_{c}-1\right)$, so under reasonable circumstances it is negative. The second term, on the other hand, is clearly positive. Hence, other things equal, the productivity acceleration should be positively correlated with lagged ICT capital growth but negatively correlated with current ICT capital growth (with these growth rates scaled by the share of ICT capital in output).Footnote 11

top of page

## III. A Basic Identification Problem and its Solution

There are a number of challenges in implementing this framework empirically. First, it is unclear how long the lags are between ICT investment and complementary investment. In other words, the length of a period is a free parameter, and theory gives little guidance. The lagged $\stackrel{˜}{k}$ may be last years ICT capital accumulation, or the last decades. Furthermore, the equation for the accumulation of complementary capital has no adjustment costs, or time-to-build or time-to-plan lags in the accumulation of $C$. But such frictions and lags are likely to be important in practice, making it even harder to uncover the link between ICT and measured TFP.

Second, if we estimate the coefficient of $\stackrel{˜}{k}$ (our main variable of interest) to be positively significant, interpretation of the results may be clouded by uncertainty of whether our proxies are capturing only unobserved organizational capital, or whether the proxies are affecting TFP directly through spillovers. Since the coefficient of $\stackrel{˜}{k}$ represents the impact of the product of ICT share and ICT growth on same industry's TFP growth, its positive significance may imply that there are intra-industry (within firms) ICT spillovers. That is, the firm cannot reap all the benefits of its ICT use and part of it is spilled to firms within the same industry. Alternatively, its significance might be an outcome of unmeasured intangible capital that went along with ICT use. In other words, one can no longer tell whether the $\stackrel{˜}{k}$ terms represent intra-industry externalities that are internalized within the industry or accumulation of other private capital stock. Similarly, if we find that lagged $\stackrel{˜}{k}$ is important for explaining current productivity growth we do not know whether that finding supports the theory we have outlined, or whether it indicates that the externality is a function of lagged capital.

Third and more fundamentally, consider the problem of estimating the externalities represented by $\Delta e$ in equation (5). Besides the possibility of intra-industry ICT spillovers, there could be presence of inter-industry spillovers (firms in one industry benefits from ICT use by firms in other industries) as well. We can estimate this inter-industry ICT spillover by using aggregate ICT capital in the economy (more precisely ICT of all other industries in the economy). However, there are other factors that are considered to have spillovers to TFP, and we need to control these factors so that their effects are not erroneously captured by ICT variables. R&D is considered one such variable that spills positive affects to TFP. Hence, we can measure the $\Delta e$ by the growth rate of economy-wide aggregate ICT $\left(\Delta {k}_{t}\right)$, own industry R&D growth $\left(\Delta {r}_{it}\right)$, and all other industries (within a country) aggregate R&D growth $\left(\Delta {r}_{t}\right)$ such that $\Delta {e}_{it}=\gamma \Delta {k}_{t}^{IT}+{\lambda }_{1}\Delta {r}_{it}+{\lambda }_{2}\Delta {r}_{t}$ . In this case, equation (5) can be written as:

(6) $\Delta TF{P}_{it}=\alpha +{\beta }_{1}{\stackrel{˜}{k}}_{it}^{IT}+{\beta }_{2}{\stackrel{˜}{k}}_{i,t-1}^{IT}+\lambda \Delta {k}_{t}^{IT}+{\lambda }_{1}\Delta {R}_{it}+{\lambda }_{2}\Delta {r}_{t}+{s}_{G}\Delta {t}_{it}+{u}_{ijt}$

So, if coefficient $\gamma$ is significant then it would mean the presence of domestic inter-industry ICT spillover. In that case, if both or one $\beta$ is also significant, we would have reason to believe that $\stackrel{˜}{k}$ more likely captures intra-industry ICT spillover rather than intangible, as firms potentially learn more from the activities of other firms in their own industry than from other random firms in the economy. If $\gamma$ is not significant but $\beta$ is, we still face the same problem of whether $\stackrel{˜}{k}$ captures the effect of intangible capital or intra-industry ICT spillover. In this case, the country dimension of the data allows us to cut through the confusion so that we can estimate externalities that is robust to the existence of intangible capital. If there are industry-level externalities at the domestic level (the possible explanation of the positive $\beta$), it is reasonable to expect that they do not stop at national boundaries. Indeed, Stockman (1988) defines technology change (including the effects of possible externalities) as shocks that change output in the same industry across a group of countries. Hence for positive $\beta$ to imply intra-industry ICT spillover, one would like to see positive impact of foreign intra-industry ICT. If $\beta$ is significant but the coefficient on foreign intra-industry is not, then the story is more in line with intangible capital.

Hence, we estimate the equation by adding ICT growth in the same industry in a foreign country. Besides, we also control for foreign intra-industry R&D $\left(\Delta {r}_{it}^{*}\right)$ R&D in all foreign countries in the same industry, foreign aggregate ICT in all other industries $\left(\Delta {r}_{t}^{*}\right)$ and foreign aggregate R&D in all other industries $\left(\Delta {r}_{t}^{*}\right)$. Note that the variables with "star" as superscript are respective foreign variables. Thus the full form estimating equation (with country subscript j added) would be

(7) $\Delta TF{P}_{ijt}=\alpha +{\beta }_{1}{\stackrel{˜}{k}}_{ijt}^{IT}+{\beta }_{2}{\stackrel{˜}{k}}_{ijt-1}^{IT}+{\gamma }_{1}\Delta {k}_{jt}^{IT}+{\lambda }_{1}\Delta {r}_{ijt}+{\lambda }_{2}\Delta {r}_{jt}$
$+{\gamma }_{1}^{*}\Delta {k}_{ijt}^{IT*}+{\gamma }_{2}^{*}\Delta {k}_{jt}^{IT*}+{\lambda }_{1}^{*}\Delta {r}_{ijt}^{*}+{\lambda }_{2}^{*}\Delta {r}_{jt}^{*}+{u}_{ijt}$

Intuitively, the reason that equation (7) is robust to intangible capital accumulation is that intangible assets should be accumulated in proportion to ones own investment, but not to the investment of an industry in another country. Thus, cross-country data can solve the identification problem. Of course, one might argue that the externality from foreign capital is not quite as large as the externality to investment in ones own home industry. To that extent, estimates of (7) will represent a conservative lower bound to the external effects of domestic ICT investment.

We run regression of equation (5) as a first step. Then we estimate equation (6) using both aggregate and industry-level data for the countries in our sample (note that $\Delta {k}_{t}$ and $\Delta {r}_{t}$ appear without industry subscript). Finally, we estimate the full equation (7) with foreign variables included.

top of page

## IV. Data

The data used in this paper are from the EUKLEMS database compiled by the Groningen Growth and Development Centre (GGDC), the OECD's Analytical Business Expenditure for Research and Development (ANBERD) database. For few countries, several data series have been supplemented from national statistical agencies. We use data for 16 OECD countries (Australia, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, South Korea, the Netherlands, Spain, Sweden, the United Kingdom and the United States) and 24 industries covering the period 1973 to 2004. Among the 24 industries, 13 are manufacturing, nine are services industries, one is electricity, gas and water supply industry and the remaining one is construction industry. The only industries that are missing from the study are agriculture, fishing and forestry industries, as they do not have R&D data available (the industry list is given in Appendix A, Table A1).

Data on gross output, value added, intermediate input, hours of work, ICT capital, non-ICT capital are taken from the EUKLEMS database. The detailed description of the EUKLEMS data is available in Inklaar, Timmer and van Ark (2006). Data on R&D stock is taken from OECD's ANBERD database. Again, in several cases, both these data have been supplemented by national statistical agencies (see Appendix B for data description). In the database, output is defined at constant price. Labor input is measured as composition-adjusted hours worked. This is the product of total hours worked and an adjustment for differences in the marginal product of heterogeneous workers based on their relative wages.

Instead of using data on capital services we use data on capital stock. Capital services data are available only in index form and capital stock data are available in levels. Since we estimate production function also at the level form and it is not straight forward to derive level from capital services in index, we prefer to use capital stock (rather than capital services). However since ICT capital stock was computed by adding three separate assets and non-ICT capital was obtained by adding five different assets (with different depreciation rates), there is no additional benefit of data precision in using capital services over capital stock.

We use three types of ICT capital stock both individually and also by aggregating them into one series. The three types of ICT capital are computers, communication equipment and software. The non-ICT capital is used only as an aggregate series. For detail on how these capital stock data are estimated, see Timmer et al (2007).

The R&D expenditure data are in national currency current price which were converted to constant price using industry value added deflators then transferred to PPP at 1997 rate (using industry level PPP for each country) and constructed capital stock with 15% depreciation rates.

Appendix A presents average data across time (1973–2004) for 16 countries and 24 industries, with data availability information in Table A1. With 24 industries and 32 years of data, there are maximum of 768 observations for each country. Similarly with 16 countries, for each industry, the maximum number of observation is 512. Among the 24 industries, 23 are ICT-using industries and electrical and optical equipment industry is the only one that is ICT-producing. As our focus is on ICT-using industries, we will exclude ICT-producing industry from our estimation.

Looking at the upper panel, the data on value added and labor are complete for all countries. The data availability is quite good for both types of capital for all countries except for Ireland and Sweden. For Ireland, the capital stock data starts only from 1995 and those for Sweden only from 1993. On R&D variable, Korea has the smallest number of observation, as its data start only from 1995. By industry, the data points for value added, labor, ICT and non-ICT capital are almost similar (although not complete for capital stock as some countries' data are missing for the earlier years as mentioned above). The data availability on R&D, however, varies by industry, with minimum observation of 204 for hotel and restaurant industry.

An inspection at the data shows that there is quite a variation across industries and countries in value added (Table A2), hours of work (Table A3), R&D capital stock (Table A4), ICT capital stock (Table A5), non-ICT capital stock (Table A6), share of ICT to total capital stock (Table A7) ICT intensity in terms of value added, (Table A8) and R&D intensity in terms of value added (Table A9). The share of ICT to total capital stock ranges from 0.1% in real estate activities to 28% in post and telecommunications (Table A6). The industry mean of share of ICT to total capital stock is 3.6% with median of 4.5%. Taking industry median value as a cut off point, 11 industries have the share higher than median and 12 industries have share lower than median. Only four manufacturing industries have ICT to capital share higher that that of industry median. The country median ranges from a low of 2% for Canada and Belgium to a high of 8.2% for Sweden.

Looking at ICT intensity (ICT to GDP share) in Table A8, we see that it ranges from as low as 1.2% for real state activities industry to as high as 147% to post and telecommunication industry in the US. Almost similar variation is noticed across countries in the same industry. For example, in renting of M&E and other business industry, (the industry with second highest ICT share after post and telecommunication industry) the ICT intensity varies from low of 6% in Belgium to high of 51% in Korea. Most of the industries that have higher share of ICT in total capital stock are also the industries which have higher share of ICT capital stock to value added. There are few exceptions. For example, the electricity, gas and water supply is highly capital-intensive in terms of value added (11%) but has a very low share of ICT in total capital stock (1.8%). A more pronounced difference in two shares is found for transport storage (19% vs. 5%). It implies that both these industries have very high capital to value added ratio. There are only two industries whose ICT share in total capital is higher than their respective capital to value added shares. They are construction and wholesale trade. As a result, they have the lowest capital to value added shares.

Finally, one might argue that using firm-level data would allow us to avoid the complexity of using cross-country data sets. Returning to equation (7), the beneficial externalities to a firm presumably come from the investments of other firms within the same industry. So why not estimate (7), with i indexing firms, and the externality coming from industry-level ICT investment?

The main reason why firm-level data are not suitable for the purpose of estimating externalities in our framework is the lack of true price deflators for firm-level output. In the realistic case where firms within an industry produce heterogeneous output, Klette and Griliches (1996) have shown that the procedure of replacing the true firm-level deflator with an aggregate industry deflator leads to an omitted variable problem that is particularly worrisome for the purpose at hand. They show that the omitted variable is correlated with industry inputs, even controlling for firm inputs that is, it has exactly the characteristics of an externality! But the source of the correlation with industry inputs is the mismeasurement of real output, not a true productive externality. Thus, using firm-level data which would be preferred according to theory, will not give useful answers as long as one needs to deflate firm revenues with an aggregate industry price index.

Although industry data have other problems, including the problem of distinguishing own unobserved investment from a productive externality, there are true price deflators for industry output, and thus the Klette-Griliches problem does not apply. Furthermore, the industry level data makes it possible to use data for so many countries, industries and across years.

top of page

## V. Results

At the beginning, we omit controls for intangible capital; those will be added later. For most part of the analysis, we allow for non-constant returns to scale and perfect competition. Thus, our dependent variable is output rather than TFP, and we enter the inputs separately on the right-hand side. Towards the end of the paper, we impose CRS and perfect competition and use TFP as dependent variable. To begin with we estimate level equation and revert to growth one subsequently. We use OLS for most of the estimation and use GMM as robustness check. We also decompose the most preferred estimation in two ways: first, into decade-long three sub-periods and second aggregate ICT into three components: software, information technology and communication technology.

The results for level equation are presented in Table 1, using all industries (both ICT-producing and ICT-using) in specifications 1–6 and using only ICT-using industries (dropping one industry ICT-producing, electrical and optical equipment industry) in specifications 7 and 8. For the first seven specifications, we use gross output as the dependent variable and value added for the last specification. We also augment our basic estimating equation with own industry domestic R&D (specifications 6-8).

We estimate fairly standard coefficients for all the inputs except labor, whose coefficient seems too low. The coefficient for non-ICT capital is also implausibly low once we introduce industry fixed effects. Since the variation in capital and especially in non-ICT capital is across industries (cross-section) rather than across time within the same industry (time series), the use of industry fixed effect as a control lowers the magnitude of its coefficient. The estimation of low coefficients, especially on capital, is a common outcome from fixed effects regression (see Griliches and Mairesse 1998). On the other hand, the coefficient of intermediate input is quite high once the industry fixed effect is introduced.

Overall, the labor and capital coefficients are generally smaller than their respective average factor shares, whereas the coefficient of intermediate input is larger than its average factor share in the raw data. The sum of coefficients is very close to unity in all specifications except for specifications when we start having all three (country, industry and time) effects (in specifications 5). The sum of coefficient falls further with the introduction of R&D variable (specification 6), with the dropping of ICT-producing industry (specification 7) and value added as dependent variable (specification 8). The data reject the null hypothesis that there is constant return to scale, i.e. the sum of coefficients (labor, capital and intermediate input) is unity in all specifications. All the coefficients, including those on R&D capital, are significant at the 1% level.

Results in Table 2 are in some way repetition of the specifications of Table 1, but now estimated using growth rates and considering only ICT-using 23 industries (dropping ICT-producing, electrical and optical equipment, industry from the estimation sample). Here, as expected, the estimated coefficients on the capital input variables fall and are often close to zero. This is particularly true for ICT capital, which of course has a smaller share to start with. The coefficient of non-ICT capital is significant once we introduce time fixed effect and insignificant when industry or country fixed effect is introduced. Domestic R&D continues to be significant in most specifications.Footnote 12 Even though industry and country fixed effects are wiped out in first difference equation, we still carry them in some specifications in this table. But our preferred specification is (3) with only time fixed effect and we continue this method for the rest of the estimations.

In Table 3, we introduce the variables that should proxy for intangible capital accumulation, as required by the model in Section II. We extend the growth equations that provided results for Table 2 by adding first the ICT capital share in gross output (specification 1) and then the product of the ICT ratio to gross output and ICT capital growth (specification 2). Then we check the impact of this product variable in one period lag (spec. 3). Although the theory we have developed is not a guide on how far the lags should, it suggests that to control for intangible capital accumulation we need enter the variable for two time periods. Accordingly, we extend the estimation to include contemporaneous and one-period lags (spec. 4). As the theory suggests, regarding the control for intangible, the more recent variable should have a negative sign and the longer lag should have a positive sign. This is exactly the sign pattern we find in specification 5, although only the longer lag is statistically significant. Specification (6) shows that the data do not advise further longer lags.

Number 5 is our preferred specification, where first term enters with no effect (negative coefficient) and second term implies positive impact (as suggested by the theory) at 10% level. For the rest of the analysis, we will extend this specification to the full form as given in equation (7). Without further control, the difficulty with this estimation is that there is no way of knowing whether the significance of this variable indicates the impact of intangible or the benefits of ICT use of one firms spills to other firms in the same industry (called within-industry ICT spillover). If there are spillovers at the industry level (if the full reward of ICT use is not captured at the firm level), then the coefficient will show up with positive sign indicating industry level ICT use efficiency.

Besides, there is also the case that the ICT spillovers might be across industries. That is, a firm in an industry may benefit not only from the ICT use of firms within the same industry, but also from ICT use of firms in other industries (called inter-industry spillover). We want to estimate this channel of ICT spillover as well. Furthermore, along with intra-industry R&D spillover, we would like to control for inter-industry R&D spillover.

We begin to look for external effects, using as a baseline our results from Table 3, where we allow for intangible capital accumulation. We take the specification 5 in Table 3 and estimate equation (7) in its full form except for the fact that the different price levels that appear in the equation are not considered. However, as one does not expect the relative price of ICT in the sample countries to be very different, their exclusion should not bias our results. We begin by looking for within-country spillovers by adding aggregate ICT which is the domestic total ICT of all other industries (named inter-industry aggregate ICT), and aggregate R&D in specification 1. These two variables are significant at the 10% and the 1% levels, respectively, implying that there are spillovers from other industries' use of ICT and other industries' R&D capital.

Furthermore, even in the presence of these two variables, the ICT ratio and ICT growth (product) term is negative in one year lag and positively significant (at 10% level) in 2 years lag. The significance of this variable may indicate either intra-industry ICT spillovers or presence of intangible. More importantly, in the presence of the positive inter-industry ICT spillovers that we have estimated, it is hard to believe that there is no intra-industry spillover, as firms generally learn from firms in the same industry than firms in other industries. Hence, the significant result on the coefficient of the product of ICT ratio and ICT growth in specification 1 is more in line with domestic intra-industry ICT spillover rather than intangible capital story.

To explore further, we introduce foreign intra-industry ICT in the estimation. As explained above, our line of reasoning is that if the spillovers indicate domestic intra-industry efficiency, then foreign intra-industry variable should also be significant. The notion is that unless there is a convincing case that spillovers stop at the border (which is not the case in R&D as several studies have shown) then foreign intra-industry spillover should be present (may be in lower magnitude) if there is domestic intra-industry ICT spillover.

When we add foreign intra-industry ICT growth variable (spec. 2) ICT capital growth in the same industry in all foreign countries the domestic aggregate ICT variable loses its significance. Next we add foreign intra-industry R&D, foreign inter-industry ICT (ICT in all other industries in all foreign countries) and foreign inter-industry R&D (spec. 3). None of them are significant; all of them have wrong sign except for inter-industry R&D and oddly intra-industry R&D variable is statistically significant. In this full specification 3, the two-period lag of the product of ICT ratio and ICT growth is significant but neither domestic inter nor foreign intra and inter industry ICT variables are significant. Since there is no trace of domestic inter and foreign intra- and inter-industry ICT, the positive impact of the variable product of ICT ratio and ICT growth implies the impact of intangible capital that goes along with the ICT investment.

In specification (4), we remove all R&D related variables (both domestic and foreign) and estimate the impact of ICT. Interestingly, the product term loses its significance. Even more interesting is the case that the aggregate domestic ICT variable is quite significant, whereas the foreign ICT variables are not significant. This is a perfect case of domestic inter-industry ICT spillovers, and no trace of intangible capital impact. However, this is an outcome of model misspecification, as R&D, the TFP spillers, is taken out of the estimation.

Thus, we find some qualified support for positive externalities from aggregate domestic ICT growth (specification 1). That support vanishes, though, while we introduce the foreign ICT variables, which themselves are not significant either. Hence there is no support for the proposition that foreign within-industry and foreign inter-industry ICT usage has positive externalities (specifications 2 & 3). The significance of the ICT variable at the domestic level but insignificance at the foreign level indicates that the positive impact on productivity might be generated by intangible capital than by ICT spillovers. Another finding is that once we remove R&D variable, the aggregate ICT growth spillover becomes stronger (specification 4) suggesting that while estimating the impact of ICT on TFP, the control of R&D is essential otherwise the result would be biased. Furthermore, the models that do not control for R&D while measuring impact of ICT would be wrongly ascribing R&D spillovers to ICT spillovers.

The domestic aggregate R&D is consistently significant in all specifications, whereas own-industry R&D is positively significant only if domestic aggregate R&D is taken out from the estimation. While introduced together, the inter-industry R&D growth nullifies the impact of own industry R&D growth as these two variables own industry R&D and domestic inter-industry R&D are positively related. A higher inter-industry R&D growth rate leads to higher TFP growth, whereas own industry's higher R&D growth rate does not reflect into higher TFP growth.

An important question is whether the nature of ICT spillover has changed over time. To test this, we decompose the estimation in specification (3) into three sub-periods with 10 years each (1975–1984, 1985–1994 and 1995–2004; the samples from 1973 and 1974 are excluded) and present results in specifications 5 through 7. This sub-division allows us to evaluate whether there was anything different in the post 1995 period the time when both the US average labor productivity growth and TFP growth accelerated. In none of the three decades domestic aggregate (inter-industry) ICT variable is significant. The foreign intra-industry ICT is positive (insignificant) for the first decade, negative (insignificant) for the second decade and negatively significant at the 10% level for the last decade (1995–2004). The foreign inter-industry ICT coefficients are negative but insignificant in all three decades. For R&D, however, the situation is different. The coefficient on aggregate domestic R&D is significant for the last two decades; foreign intra-industry R&D is negative throughout and (wrongly) significant in the first decade, whereas foreign inter-industry R&D is positively significant for the recent decade.

As far as the ICT story is concerned, this decomposition confirms the finding in previous specification that there is no ICT spillovers, neither inter nor intra; neither domestic nor foreign. Any indication of positive ICT spillovers can occur due to either misspecification of the model or due to missing measurement of intangible capital.

What is somewhat surprising is the negatively significant coefficients of ICT capital in specifications 1 through 3 (at 10% level) of Table 4 (spec. 4 is not the preferred model as R&D variables are excluded). Taken literally, this could mean that the ICT capital is unproductive. This is not the first time the own industry ICT capital growth is found to be negatively correlated with its own TFP growth. In the US data from 1984 to 1999, Stiroh (2002), under different estimation techniques, finds that ICT growth is negatively related with TFP growth even at the lower level of significance. In his study, the strong negative effect is driven by telecom capital, as the coefficient of computer capital is negative but not significant. In our case, the decomposition shows that this negative coefficient was caused by the situation in the first decade of the study (Spec 5); in the last two decades, the impact of ICT capital growth on TFP growth was nil (Specs 6 and 7).

To understand if any of the three ICT assets have spillover and which asset types caused this negative impact we decompose the ICT capital into: information technology (IT), communication technology (CT) and software.Footnote 13 Results in Table 5 show that the negative coefficient on ICT in Table 4 was driven only by CT whose coefficient is negatively significant at the 5% level for the sample of entire study period. Furthermore, the regression results for three sub-periods show that the coefficient was negative only for the first sub-period, 1975–1984 (results not reported to avoid clutter) and that too only for CT. During the periods 1985-1994 (again results not reported) and 1995–2004 (reported in Table 5), none of the three ICT assets were significant. To sum up, the ICT capital is not a drag in TFP growth especially not so in the more recent years, but the industries with higher ICT growth may not be the ones which necessarily acquire higher TFP growth. The results at the domestic aggregate category show that none of the three assets generated spillover type effects.

In Table 6, we conduct a robustness check using generalized method of moment (GMM) system estimation to address the potential issue of endogeneity in the above panel (within) estimation. The first specification is copied from specification (3) of Table 4. Specifications 2-4 use system GMM and treat different variables as endogenous. In specification (2), we treat two variables, intermediate input and labor hour, as endogenous and use their lags of two period and back as GMM-type instruments. All other variables are used as exogenous and the difference of each of these variables (difference of the difference as these variables are already in difference form) is used as standard instrument. All of these are instruments for difference equation. For the level equation, we use the second lag of the difference of all endogenous and exogenous variables as instruments. Only the second lags of the variables are used because the moment conditions using the higher lags are redundant (Blundell and Bund, 1998). The model is estimated using two-step GMM and the standard errors are robust. The p-value for AR(1) shows that, as required by theory, there is first order serial correlation (the null is rejected), but the null that there is no second order correlation is not rejected at 10% and higher level, contrary to what theory requires us to do.

In specification (3), we use only intermediate input as endogenous. All instruments are as explained above. The non-rejection of AR(2) at 7% level and above strongly rejects the model. In Specification (4) we use only labor hour as endogenous. In terms of AR(2), this specification is better because as required by theory we cannot reject the null that there is no second order serial correlation at least for the 11% level. Among the three specifications in GMM, our preferred and theoretically sound model is specification (4).

The major difference in the results between OLS and GMM estimations are the following: (i) the coefficients on non-ICT is significant in OLS and turns insignificant in GMM, (ii) the lag effects strengthens (iii) the strong positive coefficient on domestic aggregate R&D in OLS either loses strength or becomes nil in GMM, (iv) the foreign intra-industry ICT which was insignificant in OLS turns significant in spec. (4) and (v) the negatively significant coefficient on foreign intra-industry R&D in OLS turns insignificant in GMM. In terms of the impact of ICT, the prime interest for this paper, none of these differences qualitatively cast doubt on the previous finding that any potential spillover impacts of ICT abode well with the intangible capital story but not with the presence of its spillovers. Overall, the GMM estimations suggest that the OLS results were not driven by simultaneity issue; even when we control for potential endogeneity, the results qualitatively remain the same.

Finally, in Table 7, we run regressions similar to those of Table 4, but now using TFP growth rather than output growth as the dependent variable (and removing own-inputs from the right-hand side of all the specifications). Using TFP rather than estimating the coefficients on the inputs is an attempt to gain power by imposing the conditions for cost minimization (although, since we do not include an input aggregate on the right-hand side, as in Hall (1990), we are also imposing constant returns to scale). The qualitatively new result is that when we estimate using TFP, the share-weighted lags of own ICT investment retain the negative and positive signs predicted by theory, although, none of the lags is significant. Second, looking at the full specification (4), we see that foreign intra-industry ICT and foreign inter-industry R&D which were insignificant in Table 4, turns negatively significant with TFP as dependent variable which is difficult to rationalize. As before within a country, R&D has spillovers that are statistically significant. Impact of foreign aggregate ICT growth is insignificant. In specification (5), we introduced cross country aggregate R&D and ICT as an interaction term, and as a result, the oddly negative coefficient (inter-industry aggregate R&D) turns insignificant. Among the four foreign variables only intra-industry ICT is (negatively) significant. In the last specification, we take out all R&D related variables and as a result both domestic and foreign aggregate ICT variables become significant. This result is capturing the positive impact of excluded variable R&D. Overall, the story holds; there is no evidence of aggregate ICT positive externalities, either within or across countries.

top of page

## VI. Conclusions

Using a simple model, we show that it is difficult to estimate ICT spillovers when one also allows for intangible investment that is complementary to ICT. Yet a variety of evidence, both micro and macro in nature, suggest that intangible investment is very important, and is particularly strongly associated with ICT investment. We then propose a solution to this problem, if ICT spillovers operate across borders. Since intangible capital investment is confined within country boundaries, using cross-country, cross-industry data allows us to solve this fundamental identification problem.

We assemble a large data set for 24 industries in 16 OECD countries over 32 years. This rich data set allows us to test for externalities even if there is intangible capital accumulation. Indeed, we find evidence that firms accumulate intangible capital in the way our theory predicts. Allowing for the intangible capital accumulation—which might otherwise be mistaken for positive externalities to lagged ICT capital investment—we find no evidence of positive spillovers to ICT investment across national boundaries and within countries. This is true across two estimation techniques we have used. This is also true whether we use gross output or TFP growth (imposing the conditions for cost minimization) as our dependent variable.  Reassuringly, we do find positive and statistically significant effects of R&D investment at the domestic level and at the international level in more recent period. This finding, in keeping with the large literature on R&D spillovers, suggests that our failure to find ICT externalities is not due to some quirk of our data or specification.

## References

• Bakhshi, H., N. Oulton and J. Thompson. (2003). Modelling investment when relative prices are trending: theory and evidence for the United Kingdom. Bank of England Working Paper 189.
• Basu, S., J.G. Fernald, N. Oulton, and S. Srinivasan (2003). The case of the missing productivity growth. NBER Macroeconomics Annual 2003, M. Gertler and K. Rogoff, eds. Cambridge, MA: MIT Press.
• Bloom, N., R. Sadun and J. van Reenen (2005). It ain't what you do it's the way that you do It - Testing explanations of productivity growth using U.S. transplants. Unpublished, London School of Economics.
• Blundell, R. and S. Bond (1998). "Initial Conditions and Moment Restrictions in Dynamic Panel Data Models. Journal of Econometrics, 87: 115–143.
• Bosworth, B. P. and J. E. Triplett. (2006). Is the 21st Century Productivity Expansion Still in Services? And What Should Be Done about It? Manuscript (July).
• Bresnahan, T. F. and M. Trajtenberg. (1995). General purpose technologies: Engines of growth? Journal of Econometrics 65 (Special Issue, January): 83–108.
• Bresnahan, Timothy, E. Brynjolfsson and L. M. Hitt (2002).  "Information Technology, Workplace Organization and the Demand for Skilled Labor:  Firm-level Evidence,"  Quarterly Journal of Economics, 117(1): 339–376.
• Brynjolfsson, E. and L. M. Hitt. (2000). Beyond computation: Information technology, organizational transformation and business performance. Journal of Economic Perspectives 14(4): 23–48.
• _______. (2002). Computing productivity: Firm-level evidence, MIT Working Paper 4210–01.
• Chandler, A. D. Jr. (1977). The Visible Hand. Cambridge: Harvard University Press.
• Corrado, Carol, Paul Lengermann, Eric J. Bartelsman, and J. Joseph Beaulieu (2006). Modeling aggregate productivity at a disaggregate level: New results for U.S. sectors and industries. Manuscript.
• Corrado, Carol, Charles Hulten, and Dan Sichel (2006). Intangible capital and economic growth. NBER working paper No.: 11948.
• Greenwood, J. and M. Yorokoglu. (1997). Carnegie-Rochester Conference Series on Public Policy 46: 49–95.
• Griliches, Z. and J. Mairesse (1998) "Production Functions: The Search for Identification," in Z. Griliches, Practicing Econometrics, Elgar, 383–411.
• Hall, R. E. (2001). The stock market and capital accumulation. American Economic Review 91 (December):1185–1202.
• _______(1990). "Invarinence Properties of Solow's Productivity Residual," in Peter Diamond (ed.), Growth, Productivity, Unemployment, the MIT Press, Cambridge, MA.
• Hornstein, A. and P. Krusell. (1996). Can technology improvements cause productivity slowdowns? In NBER Macroeconomics Annual edited by Bernanke and Rotemberg. Cambridge, MA: National Bureau of Economic Research.
• Inklaar, Robert, Marcel P. Timmer and Bart van Ark (2006). Mind the gap! International comparisons of productivity in services and goods production. University of Groningen GGDC Research Memorandum GD-89, August.
• Jorgenson, D. W. (2001). Information technology and the U.S. economy. American Economic Review 91(March): 1–32.
• Jorgenson, D.W., Mun Ho, and Kevin Stiroh (2006). The Sources of the Second Surge of U.S. Productivity and Implications for the Future. Manuscript.
• Klette, Tor Jakob & Griliches, Zvi (1996). "The Inconsistency of Common Scale Estimators When Output Prices Are Unobserved and Endogenous," Journal of Applied Econometrics, July–Aug, vol. 11(4): 343–61.
• Laitner, J. and D. Stolyarov. (2003). Technological change and the stock market. American Economic Review 93 (September): 1240–1267.
• Oliner, S.D. and D. E. Sichel. (2000). The resurgence of growth in the late 1990s: Is information technology the story? Journal of Economic Perspectives 14 (Fall):3–22.
• Oliner, S.D. and D. E. Sichel and Kevin J Stiroh. (2007). "Explaining a Productive Decade", in William C. Brainard and Geroge L. Perry (eds.) Brookings Papers on Economic Activity, 1: 81–152.
• Stiroh, K. J. (2002). Are ICT Spillovers Driving the New Economy? Review of Income and Wealth, Series 48 (1) March.
• Stiroh, K. J. and Matthew Botsch (2006). Information Technology and Productivity Growth in the 2000s. Federal Reserve Bank of New York, Mimeo, August.
• Stockman, Alan C. (1988). "Sectoral and National Aggregate Disturbances to Industrial Output in Seven European Countries", Journal of Monetary Economics.
• Tevlin, S. and Whelan, K. Explaining the Investment Boom of the 1990s. Journal of Money, Credit and Banking (forthcoming).
• Timmer, Marcel; Moergastel, Ton van; Stuivenwold, Edwin; Ypma, Gerard; O'Mahony and Kangasniemi, Mari. "EU KLEMS Growth and Productivity Accounts: Version 1, Part I Methodology, March 2007.

top of page

## Tables

Table 1. Level Equation, 1973–2004
All Industries Only ICT-using industries
Gross output as dependent variable Gross output as dependent variable Value added as dependent variable
(1) (2) (3) (4) (5) (6) (7) (7)
Employment 0.120
(0.003)Footnote a
0.134
(0.003)Footnote a
0.131
(0.004)Footnote a
0.115
(0.003)Footnote a
0.106
(0.006)Footnote a
0.066
(0.008)Footnote a
0.069
(0.008)Footnote a
0.332
(0.021)Footnote a
ICT 0.031
(0.002)Footnote a
0.030
(0.002)Footnote a
0.026
(0.002)Footnote a
0.037
(0.002)Footnote a
0.031
(0.002)Footnote a
0.034
(0.002)Footnote a
0.036
(0.002)Footnote a
0.080
(0.006)Footnote a
Non-ICT 0.201
(0.003)Footnote a
0.204
(0.003)Footnote a
0.081
(0.005)Footnote a
0.198
(0.003)Footnote a
0.071
(0.005)Footnote a
0.062
(0.009)Footnote a
0.064
(0.006)Footnote a
0.233
(0.009)Footnote a
Intermediate inputs 0.668
(0.004)Footnote a
0.654
(0.004)Footnote a
0.767
(0.007)Footnote a
0.671
(0.004)Footnote a
0.715
(0.007)Footnote a
0.724
(0.006)Footnote a
0.705
(0.008)Footnote a

Own industry R&D           0.016
(0.001)Footnote a
0.016
(0.001)Footnote a
0.079
(0.005)Footnote a
Country FE NO YES NO NO YES YES YES YES
Industry FE NO NO YES NO YES YES YES YES
Time FE NO NO NO YES YES YES YES YES
Constant 0.907
(0.018Footnote a
0.939
(0.023)Footnote a
0.954
(0.023)Footnote a
1.55
(0.050)Footnote a
1.65
(0.046)Footnote a
1.79
(0.052)Footnote a
1.94
(0.054)Footnote a
3.73
(0.124)Footnote a
R2 0.98 0.98 0.99 0.98 0.99 0.99 0.99 0.95
N 11012 11012 11012 11012 11012 9352 8927 8927
Sum of coefficients 1.02 1.02 1.01 1.02 0.92 0.89 0.87 0.65

All regression variables are in log level.

Standard errors are in parentheses, and all are robust standard errors.

Employment is measured in millions of hours worked by persons engaged. The variable ICT is information and communication technology capital stock in real prices which is the sum of IT (information technology), CT (communication technology) and Software capital stocks. The sum of ICT and NICT is total capital stocks. The data on gross output, intermediate inputs, value added, ICT and NICT are in millions of US $1997 PPP. The data are also in millions of US$ 1997 PPP.

The "sum of coefficient" includes the coefficients of only employment, ICT capital, non-ICT capital and intermediate inputs and not that on R&D capital. The same is true for the test of constant return to scale (CRS).

Table 2. Annual Growth Regression, 1973–2004
Gross output as dependent variable Value added as dependent variable
(1) (2) (3) (4) (5) (6) (7)
Employment 0.157
(0.013)Footnote a
0.164
(0.014)Footnote a
0.141
(0.013)Footnote a
0.146
(0.013)Footnote a
0.141
(0.013)Footnote a
0.163
(0.014)Footnote a
0.145
(0.013)Footnote a
ICT −0.001
(0.003)
0.001
(0.003)
−0.003
(0.003)
−0.002
(0.003)
−0.003
(0.003)
0.001
(0.003)
−0.002
(0.011)
Non-ICT 0.020
(0.012)
0.018
(0.012)
0.039
(0.012)Footnote a
0.029
(0.012)Footnote b
0.030
(0.013)Footnote b
0.010
(0.012)
0.098
(0.039)Footnote b
Intermediate inputs 0.647
(0.016)Footnote a
0.643
(0.016)Footnote a
0.638
(0.016)Footnote a
0.633
(0.016)Footnote a
0.638
(0.016)Footnote a
0.643
(0.016)Footnote a
0.436
(0.034)Footnote a
Own industry R&D 0.002
(0.001)
0.003
(0.001)Footnote b
0.002
(0.001)Footnote c
0.003
(0.001)Footnote b
0.002
(0.001)
0.003
(0.001)Footnote b
0.011
(0.003)Footnote a
Country FE YES NO NO NO YES YES NO
Industry FE NO YES NO YES NO YES NO
Time FE NO NO YES YES YES NO YES
Constant 0.320
(0.174Footnote c
0.596
(0.079)Footnote a
1.75
(0.223)Footnote a
1.70
(0.227)Footnote a
1.53
(0.270)Footnote a
0.303
(0.172)Footnote c
4.61
(0.742)Footnote a
R2 0.80 0.80 0.80 0.81 0.81 0.80 0.10
N 8571 8571 8571 8571 8571 8571 8571

All regression variables are in annual log difference multiplied by 100.

Standard errors are in parentheses, and all are robust standard errors.

Table 3. Introducing ICT Share and its Interaction in Growth Equation, 1973–2004
(1) (2) (3) (4) (5) (6)
Employment 0.138
(0.013)Footnote a
0.139
(0.013)Footnote a
0.141
(0.013)Footnote a
0.141
(0.013)Footnote a
0.141
(0.013)Footnote a
0.141
(0.014)Footnote a
ICT −0.004
(0.003)
−0.007
(0.004)Footnote c
−0.007
(0.003)Footnote b
−0.008
(0.004Footnote c
−0.005
(0.003)
−0.004
(0.003)
Non-ICT 0.039
(0.012)Footnote a
0.040
(0.013)Footnote a
0.032
(0.013Footnote b
0.035
(0.013)Footnote a
0.028
(0.013)Footnote b
0.030
(0.014)Footnote b
Intermediate inputs 0.633
(0.016)Footnote a
0.634
(0.016)Footnote a
0.632
(0.017)Footnote a
0.629
(0.017)Footnote a
0.626
(0.018)Footnote a
0.619
(0.018)Footnote a
Own industry R&D 0.002
(0.001)Footnote c
0.002
(0.001)Footnote c
0.002
(0.001)Footnote c
0.002
(0.001)
0.002
(0.001)
0.002
(0.001)Footnote c
Incorporating ICT share and its interaction with ICT growth
ICT share (in percent) 0.061
(0.018)Footnote a

ICT Ratio × ICT   0.230
(0.165)
0.089
(0.297)

ICT Ratio × ICT one year LAG     0.263
(0.157)Footnote c
0.200
(0.280)
−0.049
(0.218)

ICT Ratio × ICT two year LAG         0.313
(0.178)Footnote c
0.188
(0.220)
ICT Ratio × ICT three year LAG           0.128
(0.224)
Constant −0.461
(0.264)Footnote c
−0.404
(0.264)
1.84
(0.235)Footnote a
−0.408
(0.269)
1.87
(0.239)Footnote a
1.11
(0.218)Footnote a
R2 0.80 0.80 0.80 0.80 0.79 0.79
N 8395 8395 8136 8111 7853 7596

The dependent variable is gross output. The dependent variable and independent variables (employment, ICT, non-ICT and domestic R&D) are in annual log difference multiplied by 100. Standard errors are in parentheses, and all are robust standard errors.

The variable "ICT Ratio" is the ratio of ICT compensation to gross output, whereas "ICT Share" is ICT Ratio multiplied by 100. The variable "ICT Ratio x ICT" is the product of ICT Ratio and first log difference of ICT capital. We report the results up to lag of three year. Introducing longer lag reduces the preciseness of the estimation.

All regressions include only year fixed effects. Since we are estimating first difference equation, the cross-section fixed effect (country and/or industry) wipes out. We could use industry/country fixed effect only under the assumption that the original level equations have country and/or industry specific trends. And allowing there to be permanent country and/or industry effects in the differenced equation is incoherent in the long run.

Table 4. Introducing Domestic and Foreign ICT and R&D, 1973–2004
All years 1975-84
(5)
1985-94
(6)
1995-04
(7)
(1) (2) (3) (4)
Employment 0.141
(0.013)Footnote a
0.143
(0.013)Footnote a
0.144
(0.014)Footnote a
0.139
(0.016)Footnote a
0.166
(0.033)Footnote a
0.146
(0.027)Footnote a
0.131
(0.015)Footnote a
ICT −0.007
(0.004)Footnote c
−0.007
(0.004)Footnote c
−0.007
(0.004Footnote c
−0.014
(0.004)Footnote a
−0.017
(0.006)Footnote a
0.003
(0.007)
−0.002
(0.005)
Non-ICT 0.026
(0.013)Footnote c
0.028
(0.014)Footnote b
0.028
(0.014)Footnote b
0.070
(0.014)Footnote a
−0.005
(0.031)
0.017
(0.020)
0.050
(0.023)Footnote b
Intermediate inputs 0.625
(0.018)Footnote a
0.625
(0.018)Footnote a
0.625
(0.018)Footnote a
0.639
(0.016)Footnote a
0.645
(0.045)Footnote a
0.592
(0.028)Footnote a
0.638
(0.019)Footnote a
Own industry R&D 0.001
(0.001)
0.001
(0.001)
0.002
(0.001)
0.002
(0.001)
0.001
(0.007)
0.001
(0.002)
0.003
(0.002)
ICT Ratio X ICT
one year LAG
−0.058
(0.220)
−0.038
(0.221)
−0.023
(0.223)
0.007
(0.209)
−0.138
(0.702)
0.523
(0.375)
−0.126
(0.271)
ICT Ratio X ICT
two year LAG
0.306
(0.177)Footnote c
0.311
(0.178)Footnote c
0.329
(0.179)Footnote c
0.106
(0.191)
0.788
(0.660)
−0.397
(0.381)
0.491
(0.200)Footnote b
Domestic aggregate variables
ICT—Inter-industry 0.015
(0.009)Footnote c
0.014
(0.009)
0.013
(0.010)
0.045
(0.010)Footnote a
0.030
(0.022)
−0.015
(0.019)
0.010
(0.013)
R&D— Inter-industry 0.023
(0.006)Footnote a
0.022
(0.006)Footnote a
0.022
(0.007)Footnote a
0.022
(0.037)
0.025
(0.010)Footnote b
0.024
(0.008)Footnote a
Foreign variables
ICT— Intra-industry   −0.014
(0.010)
−0.014
(0.011)
−0.016
(0.010)
0.008
(0.018)
−0.014
(0.020)
−0.031
(0.017)Footnote c
ICT— Inter-industry     −0.025
(0.055)
0.028
(0.052)
−0.021
(0.135)
−0.095
(0.105)
−0.008
(0.076)
R&D — Intra-industry     −0.008
(0.004)Footnote b
−0.023
(0.010)Footnote b
−0.007
(0.005)
−0.004
(0.009)
R&D — Inter-industry     0.019
(0.064)
−0.058
(0.128)
−0.117
(0.103)
0.359
(0.126)Footnote a
R2 0.79 0.79 0.79 0.73 0.8 0.78 0.81
N 7853 7853 7853 9185 1921 2614 3318

The dependent variable is gross output. The dependent variable and independent variables (employment, ICT, non-ICT and domestic R&D) are in annual log difference multiplied by 100. Standard errors are in parentheses, and all are robust standard errors.

The variable "inter-industry aggregate ICT (R&D) is the domestic total ICT (R&D) capital of all other industries. Among foreign variables, "intra-industry ICT (R&D)" are the sum of the same industry ICT(R&D) across all foreign countries. The share and lag variables are as defined in Table 3.  All regressions include only year fixed effects but not country/industry fixed effects.

Table 5. Decomposing ICT into Three Components, 1973–2004
All years Years 1995-2004
Coefficients Standard error Coefficients Standard error
Employment 0.142 (0.019)Footnote a 0.132 (0.017)Footnote a
Information technology (IT) −0.001 −0.003 −0.011 −0.005
Communication technology (CT) −0.005 (0.002)Footnote b 0.007 −0.006
Software −0.001 −0.004 0.004 −0.007
Non-ICT  0.03 (0.015)Footnote b 0.049 (0.048)Footnote c
Intermediate inputs 0.624 (0.019)Footnote a 0.634 (0.026)Footnote c
Own industry R&D 0.002 −0.001 0.003 −0.002
ICT Ratio X ICT — one year LAG 0.031 −0.244 −0.055 −0.307
ICT Ratio X ICT — one year LAG 0.367 (0.191)Footnote c 0.522 (0.213)Footnote b
Domestic aggregate variables
IT — Inter-industry 0.006 −0.006 0.011 −0.01
CT — Inter-industry 0.007 −0.008 −0.004 −0.009
Software — Inter-industry −0.004 −0.008 0.016 −0.017
R&D — Inter-industry 0.023 (0.009)Footnote b 0.024 (0.010)Footnote b
Foreign variables
ICT — Intra-industry −0.014 −0.012 −0.038 (0.020)Footnote c
ICT — Inter-industry −0.015 −0.058 0.027 −0.085
R&D — Intra-industry −0.009 −0.004 −0.003 −0.01
R&D — Inter-industry 0.001 −0.067 0.391 (0.134)Footnote a
R2 0.79   0.8
N 6949   2840

The dependent variable is gross output. The dependent variable and independent variables (employment, IT, CT software, ICT, non-ICT and domestic R&D) are in annual log difference multiplied by 100. Standard errors are in parentheses, and all are robust standard errors.

All IT, CT and Software capital stocks are in real prices. The sum of these three is ICT. When we decompose the ICT data into three types of capital, we lose quite a large number of observations, as ICT data on Belgium and Canada are not available by three asset types. All other variables are as defined in Tables 3 and 4.

Table 6. Robustness Check Using System GMM Estimations, 1973–2004
OLS System GMM
(1) (2) (3) (4)
Employment 0.144
(0.014)Footnote a
0.163
(0.022)Footnote a
0.124
(0.028)Footnote a
0.179
(0.032)Footnote a
ICT −0.007
(0.004)Footnote c
−0.004
(0.008)
−0.003
(0.007)
−0.006
(0.006)Footnote a
Non-ICT  0.028
(0.014)Footnote b
−0.008
(0.025)
0.004
(0.036)
−0.002
(0.017)
Intermediate inputs 0.625
(0.018)Footnote a
0.637
(0.036)Footnote a
0.632
(0.052)Footnote a
0.624
(0.035)Footnote a
Own industry R&D 0.002
(0.001)
0.003
(0.002)
0.002
(0.001)
0.003
(0.002)Footnote c
ICT Ratio +  ICT — one year LAG −0.023
(0.223)
−0.125
(0.166)
0.025
(0.164)
−0.128
(0.12)
ICT Ratio +  ICT — two year LAG 0.329
(0.179)Footnote c
0.437
(0.150)Footnote a
0.384
(0.157)Footnote b
0.329
(0.150)Footnote b
Domestic aggregate variables
ICT— Inter-industry 0.013
(0.01)
0.002
(0.017)
0.006
(0.015)
−0.006
(0.016)
R&D— Inter-industry 0.022
(0.007)Footnote a
0.014
(0.011)
0.017
(0.013)
0.018
(0.010)Footnote c
Foreign variables
ICT— Intra-industry −0.014
(0.011)
−0.036
(0.022)
−0.032
(0.02)
−0.049
(0.022)Footnote b
ICT— Inter-industry −0.025
(0.055)
−0.097
(0.132)
−0.042
(0.152)
−0.203
(0.130)
R&D — Intra-industry −0.008
(0.004)Footnote b
−0.005
(0.005)
−0.005
(0.006)
−0.005
(0.006)
R&D — Inter-industry 0.019
(0.064)
−0.233
(0.133)
−0.314
(0.178)Footnote c
−0.222
(0.146)
AR(1)  p-value   0.00 0.00 0.00
AR(2)  p-value   0.10 0.07 0.11

The dependent variable is gross output. The dependent variable and independent variables (employment, ICT, non-ICT and domestic R&D) are in annual log difference multiplied by 100. The System GMM is estimated using two-step procedure. Standard errors are in parentheses, and all are robust standard errors.

The share and lag variables are as defined in Table 3.The variables "inter-industry" and "intra-industry" both domestic and foreign are as defined in Table 4. All regressions include only year fixed.

In specification (2) employment and intermediate inputs are treated as endogenous, in specification (3), intermediate input treated as endogenous and in (4) employment is treated as endogenous.

Table 7: TFP Growth as Dependent Variable, 1973–2004
(1) (2) (3) (4) (5) (6)
Own industry R&D 0.003
(0.001)Footnote a
0.003
(0.001)Footnote a
0.003
(0.001)Footnote a
0.003
(0.001)Footnote a
0.003
(0.001)Footnote a

ICT Ratio × ICT — one year LAG −0.306
(0.207)
−0.260
(0.209)
−0.259
(0.210)
−0.265
(0.211)
−0.267
(0.210)
−0.206
(0.172)
ICT Ratio × ICT —  two year LAG 0.239
(0.234)
0.236
(0.237)
0.229
(0.239)
0.213
(0.241)
0.214
(0.241)
0.309
(0.194)
Domestic aggregate variables
ICT— Inter-industry 0.001
(0.008)
−0.002
(0.008)
0.003
(0.009)
0.003
(0.009)
0.003
(0.009)
0.029
(0.009)Footnote a
R&D— Inter-industry 0.014
(0.005)Footnote a
0.015
(0.005)Footnote a
0.014
(0.005)Footnote a
0.009
(0.005))Footnote c
0.009
(0.005))Footnote c
0.009
(0.005))Footnote c
Foreign variables
ICT— Intra-industry   −0.034
(0.008)Footnote a
−0.030
(0.009)Footnote a
−0.030
(0.009)Footnote a
−0.031
(0.009)Footnote a
0.028
(0.009)Footnote a
ICT— Inter-industry     0.095
(0.062)
0.083
(0.063)
0.137
(0.100)
0.250
(0.057)Footnote a
R&D — Intra-industry   0.007
(0.003)Footnote b
0.007
(0.003)Footnote a
0.004
(0.003)
0.004
(0.003)

R&D — Inter-industry       −0.246
(0.053)Footnote a
−0.146
(0.153)

ICT— Inter-industry × R&D — Inter-industry         −0.009
(0.013)

R-Square 0.03 0.03 0.03 0.03 0.03 0.03
N 8201 8201 8201 8201 8201 9151

The dependent variable is the TFP log difference, where TFP is computed based on gross output. All other variables are as defined in Tables 3 and 4.

All regressions include only year fixed effects.

top of page

## Appendix A: Data summary

Table A1: Data availability (Number of Observations)
Country Name Value added Labor ICT Capital Non-ICT Capital R&D
1 Australia (AUS) 768 768 768 768 672
2 Belgium (BEL) 768 768 768 768 411
3 Canada (CAN) 768 768 600 600 718
4 Denmark (DNK) 768 768 768 768 640
5 Finland (FIN) 768 768 764 768 664
6 France (FRA 768 768 768 768 562
7 Germany (GER 768 768 768 768 665
8 Ireland (IRL 768 768 240 240 527
9 Italy (ITA) 768 768 768 768 700
10 Japan (JPN) 768 768 768 768 545
11 S. Korea (KOR 768 768 672 672 201
12 Netherlands (NLD) 768 768 768 768 661
13 Spain (SPN) 768 768 768 768 715
14 Sweden (SWE) 768 768 288 288 536
15 UK 768 768 768 768 574
16 US 768 768 768 768 736
International Standard Industrial Classification -3
1 15-16 512 512 459 459 476
2 17-19 512 512 457 459 272
3 20 512 512 459 459 438
4 21-22 512 512 459 459 458
5 23 512 512 459 459 412
6 24 512 512 459 459 476
7 25 512 512 459 459 475
8 26 512 512 459 459 476
9 27-28 512 512 459 459 476
10 29 512 512 459 459 476
11 30-33 512 512 459 459 476
12 34-35 512 512 459 459 475
13 36-37 512 512 459 459 450
14 E 512 512 459 459 423
15 F 512 512 459 459 429
16 51 512 512 459 459 305
17 50, 52 512 512 459 459 305
18 H 512 512 459 459 204
19 60 to 63 512 512 459 459 330
20 64 512 512 459 459 345
21 J 512 512 459 459 266
22 70 512 512 457 459 406
23 71 to 74 512 512 459 459 406
24 L to Q 512 512 459 459 272
In this database the last category, "community, social and personal services" includes public administration and defence (ISIC 75), education (ISIC 80), health and social work (ISIC 85), other community, social and personal services (ISIC 90-93), private households with employed persons (95-97) and extra-territorial organizations and bodies (ISIC 99). The group of all these services can be considered as non-market services.

1. Food, beverages and tobacco

2. Textiles, leather and footwear

3. Wood and products of wood and cork

4. Pulp, paper, printing & publishing

5. Coke, refined petroleum & nuclear fuel

6. Chemicals and chemical products

7. Rubber and plastics

8. Other non-metallic mineral

9. Basic metals & fabricated metal

10. Machinery, nec

11. Electrical and optical equip.

12. Transport equipment

13. Manufacturing nec; recycling

14. Electricity, gas & water supply

15. Construction

18. Hotels and restaurants

19. Transport and storage

20. Post and telecommunications

21. Financial intermediation

22. Real estate activities

23. Renting of M&E & other businesses

24. Community, social and personal services

Table A2: Value added (in billions of US $PPP), 1973–2004 Country AUS BEL CAN DNK FIN FRA GER IRL ITA JPN KOR NLD SPN SWE UK US Industry share Industry 1 8.8 4.9 13.2 3.9 1.8 20.8 39.5 2.6 17.3 44.5 3.8 9.0 14.2 2.9 22.6 122.6 2.2 2 2.3 2.3 3.9 0.6 0.6 8.9 9.7 0.4 22.0 27.8 6.0 1.4 8.7 0.4 9.4 48.1 1.0 3 1.7 0.4 6.0 0.5 1.3 3.3 7.1 0.2 5.6 7.0 0.8 0.4 3.4 1.8 1.7 38.8 0.5 4 4.8 2.7 16.1 2.3 5.7 16.9 29.9 1.8 11.8 43.7 3.0 4.5 8.2 8.2 18.7 160.9 2.2 5 0.4 0.7 1.0 0.0 0.2 1.6 18.2 0.0 2.9 21.1 2.0 0.7 2.3 0.1 4.0 14.9 0.5 6 2.6 6.7 9.4 1.8 1.8 15.7 39.1 6.3 17.1 40.5 15.5 10.4 10.2 4.5 19.1 141.4 2.3 7 1.3 1.9 3.5 0.9 0.7 11.7 20.5 0.4 15.8 22.1 3.5 1.8 6.2 1.0 11.7 47.6 1.0 8 2.7 2.6 3.0 1.2 0.8 9.2 20.5 0.6 18.4 20.8 4.6 2.3 10.9 0.9 8.5 35.9 0.9 9 7.8 6.2 12.3 1.6 2.4 46.7 49.8 0.6 42.3 89.0 8.8 6.1 18.5 5.5 22.5 133.5 3.0 10 2.1 2.5 7.3 2.5 2.2 12.5 56.2 0.4 29.2 59.7 5.1 2.7 6.6 4.0 21.4 117.8 2.2 11 1.7 3.3 5.4 1.1 2.7 17.9 44.7 2.0 13.2 86.8 19.2 3.3 5.4 5.6 18.3 239.5 3.1 12 3.4 2.5 14.1 0.4 0.5 13.7 37.7 0.2 10.6 62.1 8.9 1.6 9.0 3.0 13.7 143.4 2.1 13 1.8 1.6 3.8 1.6 0.7 6.0 12.8 0.3 13.1 14.4 2.2 2.7 5.1 0.8 12.7 42.0 0.8 14 7.5 3.4 17.6 1.4 2.0 13.7 22.0 0.4 16.3 35.7 4.9 3.1 8.0 5.7 15.6 174.6 2.2 15 23.0 9.8 38.3 5.0 7.8 47.4 87.7 2.9 62.8 191.9 34.4 13.1 39.9 6.9 45.0 315.5 6.2 16 12.0 14.6 25.4 7.6 5.9 40.8 100.0 1.9 34.3 89.9 9.4 22.0 20.6 7.1 30.3 322.0 4.9 17 20.8 11.8 30.2 8.8 4.6 63.7 88.3 3.8 64.0 118.6 11.2 17.6 38.0 9.5 60.6 359.3 6.0 18 5.4 2.3 11.8 1.4 0.8 22.0 21.3 1.0 21.7 41.9 5.9 4.3 23.1 1.1 15.0 144.5 2.1 19 16.8 5.7 24.8 4.2 4.9 30.0 32.7 1.3 28.0 91.6 13.6 17.8 17.1 6.2 33.2 192.6 3.4 20 7.5 4.5 17.5 2.4 1.7 26.8 39.1 1.0 17.0 36.6 6.6 5.5 8.5 8.6 52.5 134.0 2.4 21 27.7 13.1 32.2 5.8 2.9 54.6 67.4 4.2 51.0 119.3 21.5 20.7 29.4 6.4 70.0 376.9 6.0 22 27.2 20.0 53.9 12.7 6.2 95.2 148.1 3.4 128.1 154.2 21.1 19.5 47.8 15.1 67.4 744.9 10.4 23 17.1 11.5 22.4 5.1 2.9 66.3 103.5 3.2 62.3 69.7 12.6 18.3 20.0 9.2 66.1 587.5 7.1 24 75.4 51.6 162.4 33.2 22.5 291.0 383.1 11.9 223.4 548.7 93.0 92.3 136.9 48.0 265.8 1685.3 27.3 C-avg. 11.7 7.8 22.3 4.4 3.5 39.0 61.6 2.1 38.7 84.9 13.2 11.7 20.7 6.8 37.7 263.5 C- share 1.9 1.2 3.5 0.7 0.6 6.2 9.8 0.3 6.1 13.5 2.1 1.9 3.3 1.1 6.0 41.8 100 1. Food, beverages and tobacco 2. Textiles, leather and footwear 3. Wood and products of wood and cork 4. Pulp, paper, printing & publishing 5. Coke, refined petroleum & nuclear fuel 6. Chemicals and chemical products 7. Rubber and plastics 8. Other non-metallic mineral 9. Basic metals & fabricated metal 10. Machinery, nec 11. Electrical and optical equip. 12. Transport equipment 13. Manufacturing nec; recycling 14. Electricity, gas & water supply 15. Construction 16. Wholesale trade 17. Retail trade 18. Hotels and restaurants 19. Transport and storage 20. Post and telecommunications 21. Financial intermediation 22. Real estate activities 23. Renting of M&E & other businesses 24. Community, social and personal services Table A3: Employment in hours worked by persons engaged (in millions), 1973–2004 Country AUS BEL CAN DNK FIN FRA GER IRL ITA JPN KOR NLD SPN SWE UK US Industry share Industry 1 352 164 515 140 96 987 1768 110 888 3032 803 277 739 134 1070 3442 2.4 2 214 150 339 49 80 842 1095 40 1980 3224 2930 77 720 48 1045 3384 2.7 3 102 24 242 24 67 197 409 13 429 810 184 75 191 83 208 1490 0.8 4 235 90 487 86 140 580 1176 44 508 2334 514 220 309 196 927 4074 2.0 5 13 13 37 1 6 72 96 1 51 84 38 11 22 6 53 338 0.1 6 106 127 201 42 33 298 1026 40 447 914 466 134 289 64 621 2149 1.2 7 88 43 172 31 28 337 567 20 315 1403 490 50 158 47 533 1697 1.0 8 127 72 117 40 34 319 658 23 515 1250 456 75 348 43 418 1145 0.9 9 399 233 514 88 89 1129 2018 35 1531 3679 996 230 685 223 1515 4484 3.0 10 177 86 268 115 106 728 2124 30 1040 2801 833 149 290 197 1019 3697 2.3 11 135 121 264 76 78 784 2091 90 873 4334 1535 199 300 162 1205 5666 3.0 12 232 107 414 38 50 750 1414 17 615 2242 885 113 475 181 1113 3670 2.0 13 156 69 247 55 37 386 657 24 581 1533 473 150 352 81 478 1741 1.2 14 183 51 159 26 38 269 603 29 258 544 136 65 137 53 389 1574 0.7 15 1137 390 1541 282 368 3320 4835 201 2840 12459 3051 726 2409 479 3874 12577 8.4 16 819 329 1240 253 178 1543 2897 83 1885 7360 3893 662 803 363 2242 11627 6.0 17 1846 514 2740 390 365 3689 5504 259 5142 14661 3081 993 2966 528 5237 25168 12.1 18 518 147 1337 100 123 1331 1928 135 2000 7881 3059 308 1473 176 1825 11571 5.6 19 753 326 1015 225 236 1605 2926 93 1995 6419 1994 520 1146 364 2212 7193 4.8 20 274 140 389 81 71 611 807 45 545 1029 259 143 294 112 965 2470 1.4 21 524 218 937 126 86 1141 1847 76 886 3341 1010 364 562 122 1603 9289 3.7 22 154 21 153 49 55 397 405 12 127 1467 481 90 121 98 432 2722 1.1 23 989 469 1251 256 190 3553 3621 137 2289 6792 1165 1037 1132 427 4717 19056 7.8 24 3193 1713 6488 1315 1018 10047 14222 551 8491 20687 6049 2685 5311 2243 10726 61416 25.9 C-avg. 530 234 878 162 149 1455 2279 88 1510 4595 1449 390 885 268 1851 8402 C- share 2.1 0.93 3.5 0.64 0.59 5.8 9.1 0.35 6.0 18.3 5.8 1.6 3.5 1.1 7.4 33.4 100 1. Food, beverages and tobacco 2. Textiles, leather and footwear 3. Wood and products of wood and cork 4. Pulp, paper, printing & publishing 5. Coke, refined petroleum & nuclear fuel 6. Chemicals and chemical products 7. Rubber and plastics 8. Other non-metallic mineral 9. Basic metals & fabricated metal 10. Machinery, nec 11. Electrical and optical equip. 12. Transport equipment 13. Manufacturing nec; recycling 14. Electricity, gas & water supply 15. Construction 16. Wholesale trade 17. Retail trade 18. Hotels and restaurants 19. Transport and storage 20. Post and telecommunications 21. Financial intermediation 22. Real estate activities 23. Renting of M&E & other businesses 24. Community, social and personal services Table A4: R&D stock (in millions of US$ PPP), 1973–2004
Country AUS BEL CAN DNK FIN FRA GER IRL ITA JPN KOR NLD SPN SWE UK US Industry share
Industry
1 416 273 444 298 165 1043 1275 137 219 3319 253 1004 247 347 1713 7730 1.5
2 56 186 187 12 29 277 491 29 90 2415 155 52 131 23 139 1525 0.5
3 35 18 115 13 53 58 332 6 20 575 12 7 8 40 21 1705 0.2
4 159 161 673 24 440 315 426 14 38 2429 125 51 83 965 214 7911 1.1
5 24 99 586   39 353 2619   354 1923 301 516 215 21 2582 7021 1.6
6 755 5273 2017 1363 886 12507 27739 333 5067 28822 4612 6361 1464 3551 14409 81704 16.1
7 84 413 130 81 117 1999 2497 40 1217 5455 501 171 396 195 652 7038 1.7
8 161 286 89 112 92 1429 1649 29 167 4965 394 77 253 143 823 4197 1.2
9 593 730 1008 87 300 4602 4525 37 1120 14580 723 454 350 949 1533 12213 3.6
10 338 636 684 602 643 2268 15698 39 1752 20128 1350 498 471 1926 5298 22710 6.1
11 969 3006 5695 601 1787 16915 29501 408 4637 66696 19245 4554 1329 5773 13132 156581 27.0
12 659 430 3036 79 80 17242 26739 16 6193 28880 7540 510 1440 2430 11933 194897 24.7
13 46 122 137 480 33 261 286 18 157 1593 163 49 70 46 1099 3530 0.7
14 134 55 1044 7 115 1308 469 4 958 1726 521 31 172 389 757 1086 0.7
15 74 173 65 30 69 500 391 1 331 5213 1708 135 110 78 142 1187 0.8
16 445 94 953 367 51 - 582 5 96 41 43 302 30 29 129 42932 4.0
17 33 5 55 37 3 - 36 0 9 4 3 18 2 2 14 2469 0.2
18 7 4 3 3 0 - 6 0 1 4 - 1 0 - 1 553 0.1
19 71 51 155   13 118 636 5 10 979 47 37 39 5 18 1651 0.3
20 383 122 603 175 178 2921 - 50 161 - 1555 256 318 1450 4547 7428 1.9
21 445 117 533 87 6 - 77 7 115 4 10 105 21 165 25 5812 0.7
22 83 94 232 83 11 152 209 14 201 519 150 49 90 131 505 3373 0.5
23 751 549 2246 571 83 1233 1780 108 1946 4251 1437 532 609 1007 4376 35125 4.6
24 48 35 4 4 19 - 24 1 30 0 157 261 53 34 96 57 0.1
C-avg. 282 539 862 232 217 3447 5130 57 1037 8457 1783 668 329 857 2673 25435
C- share 0.54 1.0 1.7 0.45 0.42 6.6 9.9 0.11 2.0 16.3 3.4 1.3 0.63 1.6 5.1 48.9 100

1. Food, beverages and tobacco

2. Textiles, leather and footwear

3. Wood and products of wood and cork

4. Pulp, paper, printing & publishing

5. Coke, refined petroleum & nuclear fuel

6. Chemicals and chemical products

7. Rubber and plastics

8. Other non-metallic mineral

9. Basic metals & fabricated metal

10. Machinery, nec

11. Electrical and optical equip.

12. Transport equipment

13. Manufacturing nec; recycling

14. Electricity, gas & water supply

15. Construction

18. Hotels and restaurants

19. Transport and storage

20. Post and telecommunications

21. Financial intermediation

22. Real estate activities

23. Renting of M&E & other businesses

24. Community, social and personal services

Table A5: Information and communication technology (ICT) capital stock in real price (in millions of US $PPP), 1973–2004 Country AUS BEL CAN DNK FIN FRA GER IRL ITA JPN KOR NLD SPN SWE UK US Industry share Industry 1 1038 157 245 169 113 2075 1428 272 576 753 276 514 1123 556 1141 5772 1.0 2 98 75 37 24 13 325 323 22 948 428 181 53 434 11 350 1336 0.3 3 46 13 75 29 53 136 197 10 926 61 11 14 147 212 92 656 0.2 4 573 226 287 237 409 1233 3125 253 1015 1621 143 435 783 1445 1440 9332 1.4 5 49 38 70 18 15 550 361 4 143 220 184 528 130 132 104 2370 0.3 6 132 309 211 124 105 1418 2497 287 3900 2770 729 650 854 529 1628 11995 1.8 7 52 119 71 50 25 689 810 52 694 435 281 120 461 85 868 1307 0.4 8 331 67 56 49 44 629 1117 66 633 452 246 102 1148 30 382 1868 0.5 9 534 240 248 94 118 1384 1971 55 1577 3612 508 384 1254 933 743 6402 1.3 10 114 100 89 200 162 730 2912 36 1765 4200 376 233 317 590 1432 10722 1.5 11 204 257 217 219 202 3757 3989 315 2105 12199 2065 707 854 809 2196 22271 3.4 12 247 59 332 26 20 1858 3030 18 777 3287 903 139 635 1220 967 11914 1.6 13 88 44 44 128 30 329 491 41 1252 572 88 84 260 75 491 2804 0.4 14 2255 160 1431 57 161 3924 2987 140 507 3935 162 291 1126 608 2319 15962 2.3 15 1247 95 481 170 190 1162 1831 130 1683 3146 881 349 506 255 724 9553 1.4 16 1324 905 1341 868 610 2626 7958 224 1394 3717 122 2270 999 1365 4481 34563 4.2 17 2172 224 1085 968 398 1749 6585 674 2404 7756 358 1139 1815 1827 6226 28990 4.1 18 445 67 273 158 23 645 1468 441 489 419 65 91 405 72 751 4087 0.6 19 3300 1073 1440 562 529 2004 6231 357 15292 5001 290 2147 9071 1043 2541 47352 6.3 20 3226 1947 24446 485 833 5577 17811 418 8784 34550 4264 4371 3808 12322 42039 197116 23.2 21 5998 1496 4293 924 633 6093 10959 334 6956 13676 3509 4843 4635 1746 14019 97597 11.4 22 911 1093 1512 449 33 607 1153 268 707 424 17 251 817 11 1839 8671 1.2 23 2414 718 2336 1900 511 8995 21835 566 6112 23721 6377 2538 2565 4129 13734 116232 13.8 24 9354 1073 8216 2395 1738 10932 22930 2834 9435 48643 7471 7886 10614 3395 18782 102779 17.2 C-avg. 1506 440 2035 429 290 2476 5167 326 2920 7316 1229 1256 1865 1392 4970 31319 C- share 2.3 0.7 3.1 0.7 0.4 3.8 8.0 0.5 4.5 11.3 1.9 1.9 2.9 2.1 7.7 48.2 100 1. Food, beverages and tobacco 2. Textiles, leather and footwear 3. Wood and products of wood and cork 4. Pulp, paper, printing & publishing 5. Coke, refined petroleum & nuclear fuel 6. Chemicals and chemical products 7. Rubber and plastics 8. Other non-metallic mineral 9. Basic metals & fabricated metal 10. Machinery, nec 11. Electrical and optical equip. 12. Transport equipment 13. Manufacturing nec; recycling 14. Electricity, gas & water supply 15. Construction 16. Wholesale trade 17. Retail trade 18. Hotels and restaurants 19. Transport and storage 20. Post and telecommunications 21. Financial intermediation 22. Real estate activities 23. Renting of M&E & other businesses 24. Community, social and personal services Table A6: Non-information and communication technology (NICT) capital stock in real price (in billions of US$ PPP), 1973–2004
Country AUS BEL CAN DNK FIN FRA GER IRL ITA JPN KOR NLD SPN SWE UK US Industry share
Industry
1 12.0 10.0 13.3 6.6 3.3 44.5 60.9 4.2 37.2 43.1 6.9 16.7 31.1 4.9 30.1 142.9 1.1
2 1.8 4.6 3.2 1.2 0.8 7.0 9.6 0.3 40.7 52.0 10.5 2.2 12.1 0.6 9.4 53.8 0.5
3 1.9 1.7 7.5 0.7 2.1 2.9 8.1 0.3 17.6 9.7 1.4 0.5 4.5 3.9 2.8 30.4 0.2
4 4.5 4.2 24.2 2.7 9.8 16.9 32.8 0.9 20.0 62.5 5.2 5.9 12.9 14.3 18.1 139.7 0.9
5 1.9 1.6 11.3 0.6 1.0 9.1 11.6 0.0 4.8 19.0 4.2 2.6 3.9 0.6 8.8 73.0 0.4
6 5.5 9.1 16.6 3.7 3.9 27.3 66.4 6.7 59.9 108.9 25.9 22.5 22.3 6.7 39.4 159.8 1.4
7 2.2 5.4 3.5 1.3 0.9 11.5 23.8 0.9 40.0 43.8 7.3 2.9 11.0 2.0 18.7 50.5 0.5
8 4.3 6.3 4.5 1.7 1.4 12.3 29.8 2.1 37.2 34.9 14.6 4.0 28.0 1.6 8.0 45.5 0.6
9 14.6 11.2 23.2 2.1 4.1 37.9 63.8 0.7 107.2 243.8 25.5 10.0 45.1 8.3 28.5 181.0 1.9
10 1.7 2.7 3.6 2.6 2.1 13.4 44.5 0.5 62.9 102.6 7.8 2.8 6.0 5.6 21.0 76.8 0.8
11 2.3 3.3 4.1 1.0 1.3 22.3 44.3 4.3 17.6 162.0 33.0 6.2 7.7 4.3 16.7 165.2 1.2
12 4.6 3.1 15.5 0.4 0.9 31.4 49.8 0.4 27.3 173.1 19.3 3.4 12.2 5.7 18.8 128.8 1.2
13 1.0 4.4 1.7 2.0 0.9 5.9 13.6 0.4 28.0 21.2 2.3 3.5 6.3 1.3 8.0 48.0 0.4
14 76.8 19.2 168.4 10.4 13.2 92.2 104.0 11.6 97.8 318.1 52.2 29.0 39.8 53.0 65.3 789.7 4.6
15 12.2 8.4 14.6 3.2 3.8 23.9 33.6 1.6 49.1 90.9 53.4 6.8 38.7 6.0 13.7 110.8 1.1
16 14.2 23.7 9.4 7.5 7.1 37.5 53.7 1.9 28.1 105.9 7.3 24.3 30.3 8.0 19.4 189.7 1.4
17 19.0 5.9 21.5 9.3 5.0 88.6 54.2 6.6 53.3 180.8 17.2 21.1 59.7 10.8 62.0 520.5 2.7
18 9.9 5.4 14.5 1.8 0.9 30.5 26.1 4.0 29.2 89.7 11.0 4.7 19.4 1.8 23.6 222.4 1.2
19 115.4 24.8 112.1 16.0 26.4 89.3 118.5 3.4 74.6 435.6 79.9 65.5 62.2 31.4 83.7 531.9 4.5
20 23.0 45.0 38.2 8.6 4.3 65.2 86.8 5.0 36.9 120.6 11.3 10.6 9.1 45.1 77.6 366.2 2.3
21 40.7 23.1 26.1 5.3 1.4 55.4 92.2 3.0 105.2 73.2 16.7 36.9 19.5 4.9 63.2 329.0 2.1
22 322.9 196.9 70.6 324.7 111.6 2491.8 2769.6 99.1 130.8 1484.7 423.3 502.0 1292.8 146.0 1031.4 7383.4 44.7
23 15.4 129.3 3.0 3.6 1.5 52.8 89.9 4.6 762.6 92.9 242.4 12.3 8.7 8.5 32.0 201.6 4.0
24 148.8 94.2 365.8 75.2 40.5 897.9 1119.8 36.0 545.4 1254.2 338.0 335.9 295.1 83.0 540.6 2428.5 20.5
C-avg. 35.7 26.8 40.7 20.5 10.3 173.6 208.6 8.3 100.6 221.8 59.0 47.2 86.6 19.1 93.4 598.7
C- share 2.0 1.5 2.3 1.2 0.6 9.9 11.9 0.5 5.7 12.7 3.4 2.7 4.9 1.1 5.3 34.2 100

1. Food, beverages and tobacco

2. Textiles, leather and footwear

3. Wood and products of wood and cork

4. Pulp, paper, printing & publishing

5. Coke, refined petroleum & nuclear fuel

6. Chemicals and chemical products

7. Rubber and plastics

8. Other non-metallic mineral

9. Basic metals & fabricated metal

10. Machinery, nec

11. Electrical and optical equip.

12. Transport equipment

13. Manufacturing nec; recycling

14. Electricity, gas & water supply

15. Construction

18. Hotels and restaurants

19. Transport and storage

20. Post and telecommunications

21. Financial intermediation

22. Real estate activities

23. Renting of M&E & other businesses

24. Community, social and personal services

Table A7: Share of information and communication technology (ICT) in total capital stock, 1973–2004
Country AUS BEL CAN DNK FIN FRA GER IRL ITA JPN KOR NLD SPN SWE UK US Industry share
Industry
1 8.0 1.5 1.8 2.5 3.3 4.5 2.3 6.1 1.5 1.7 3.8 3.0 3.5 10.2 3.7 3.9 3.3
2 5.2 1.6 1.1 2.0 1.6 4.4 3.3 6.8 2.3 0.8 1.7 2.4 3.5 1.8 3.6 2.4 2.2
3 2.4 0.8 1.0 4.0 2.5 4.5 2.4 3.2 5.0 0.6 0.8 2.7 3.2 5.2 3.2 2.1 2.7
4 11.3 5.1 1.2 8.1 4.0 6.8 8.7 21.9 4.8 2.5 2.7 6.9 5.7 9.2 7.4 6.3 5.7
5 2.5 2.3 0.6 2.9 1.5 5.7 3.0 100.0 2.9 1.1 4.2 16.9 3.2 18.0 1.2 3.1 3.1
6 2.3 3.3 1.3 3.2 2.6 4.9 3.6 4.1 6.1 2.5 2.7 2.8 3.7 7.3 4.0 7.0 4.6
7 2.3 2.2 2.0 3.7 2.7 5.7 3.3 5.5 1.7 1.0 3.7 4.0 4.0 4.1 4.4 2.5 2.6
8 7.1 1.1 1.2 2.8 3.0 4.9 3.6 3.0 1.7 1.3 1.7 2.5 3.9 1.8 4.6 3.9 3.0
9 3.5 2.1 1.1 4.3 2.8 3.5 3.0 7.3 1.4 1.5 2.0 3.7 2.7 10.1 2.5 3.4 2.4
10 6.3 3.6 2.4 7.1 7.2 5.2 6.1 6.7 2.7 3.9 4.6 7.7 5.0 9.5 6.4 12.3 6.3
11 8.1 7.2 5.0 18.0 13.4 14.4 8.3 6.8 10.7 7.0 5.9 10.2 10.0 15.8 11.6 11.9 9.6
12 5.1 1.9 2.1 6.1 2.2 5.6 5.7 4.3 2.8 1.9 4.5 3.9 4.9 17.6 4.9 8.5 4.9
13 8.1 1.0 2.5 6.0 3.2 5.3 3.5 9.3 4.3 2.6 3.7 2.3 4.0 5.5 5.8 5.5 4.4
14 2.9 0.8 0.8 0.5 1.2 4.1 2.8 1.2 0.5 1.2 0.3 1.0 2.8 1.1 3.4 2.0 1.8
15 9.3 1.1 3.2 5.0 4.8 4.6 5.2 7.5 3.3 3.3 1.6 4.9 1.3 4.1 5.0 7.9 4.5
16 8.5 3.7 12.5 10.4 7.9 6.5 12.9 10.5 4.7 3.4 1.6 8.5 3.2 14.6 18.8 15.4 10.2
17 10.3 3.7 4.8 9.4 7.4 1.9 10.8 9.3 4.3 4.1 2.0 5.1 3.0 14.5 9.1 5.3 5.4
18 4.3 1.2 1.8 8.1 2.5 2.1 5.3 9.9 1.6 0.5 0.6 1.9 2.0 3.8 3.1 1.8 2.0
19 2.8 4.1 1.3 3.4 2.0 2.2 5.0 9.5 17.0 1.1 0.4 3.2 12.7 3.2 2.9 8.2 5.0
20 12.3 4.1 39.0 5.3 16.2 7.9 17.0 7.7 19.2 22.3 27.4 29.2 29.5 21.5 35.1 35.0 27.5
21 12.8 6.1 14.1 14.8 31.1 9.9 10.6 10.0 6.2 15.7 17.4 11.6 19.2 26.3 18.2 22.9 16.6
22 0.3 0.6 2.1 0.1 0.0 0.0 0.0 0.3 0.5 0.0 0.0 0.0 0.1 0.0 0.2 0.1 0.1
23 13.6 0.6 43.8 34.5 25.4 14.6 19.5 11.0 0.8 20.3 2.6 17.1 22.8 32.7 30.0 36.6 11.4
24 5.9 1.1 2.2 3.1 4.1 1.2 2.0 7.3 1.7 3.7 2.2 2.3 3.5 3.9 3.4 4.1 3.0
Mean 4.0 1.6 4.8 2.1 2.7 1.4 2.4 3.8 2.8 3.2 2.0 2.6 2.1 6.8 5.1 5.0 3.6
Median 6.1 2.0 2.0 4.7 3.1 4.9 4.3 7.3 2.8 2.2 2.4 3.8 3.6 8.2 4.5 5.4 4.5

1. Food, beverages and tobacco

2. Textiles, leather and footwear

3. Wood and products of wood and cork

4. Pulp, paper, printing & publishing

5. Coke, refined petroleum & nuclear fuel

6. Chemicals and chemical products

7. Rubber and plastics

8. Other non-metallic mineral

9. Basic metals & fabricated metal

10. Machinery, nec

11. Electrical and optical equip.

12. Transport equipment

13. Manufacturing nec; recycling

14. Electricity, gas & water supply

15. Construction

18. Hotels and restaurants

19. Transport and storage

20. Post and telecommunications

21. Financial intermediation

22. Real estate activities

23. Renting of M&E & other businesses

24. Community, social and personal services

Table A8: ICT capital stock-value added ratio (in percent), 1973–2004
Country AUS BEL CAN DNK FIN FRA GER IRL ITA JPN KOR NLD SPN SWE UK US Industry share
Industry
1 11.8 3.2 1.9 4.3 6.3 10.0 3.6 10.4 3.3 1.7 7.3 5.7 7.9 19.4 5.1 4.7 4.9
2 4.3 3.3 1.0 3.7 2.2 3.7 3.3 5.6 4.3 1.5 3.0 3.7 5.0 2.8 3.7 2.8 3.1
3 2.7 3.4 1.3 6.0 4.1 4.2 2.8 5.9 16.7 0.9 1.4 3.9 4.4 12.0 5.4 1.7 3.4
4 12.0 8.2 1.8 10.3 7.2 7.3 10.5 14.4 8.6 3.7 4.7 9.6 9.6 17.6 7.7 5.8 6.7
5 11.2 5.4 6.8 40.4 7.9 34.4 2.0 15.8 4.9 1.0 9.0 80.0 5.8 103.0 2.6 15.9 7.0
6 5.0 4.6 2.2 6.9 6.0 9.0 6.4 4.6 22.9 6.8 4.7 6.2 8.4 11.7 8.5 8.5 8.2
7 4.0 6.2 2.0 5.3 3.8 5.9 3.9 12.6 4.4 2.0 8.1 6.5 7.5 8.4 7.4 2.7 4.1
8 12.1 2.6 1.9 4.2 5.7 6.8 5.5 11.2 3.4 2.2 5.4 4.5 10.6 3.2 4.5 5.2 5.1
9 6.9 3.9 2.0 5.8 5.0 3.0 4.0 9.7 3.7 4.1 5.8 6.3 6.8 17.1 3.3 4.8 4.4
10 5.5 4.0 1.2 8.1 7.2 5.8 5.2 9.0 6.0 7.0 7.4 8.7 4.8 14.9 6.7 9.1 7.2
11 12.1 7.7 4.0 19.8 7.4 21.0 8.9 16.1 15.9 14.1 10.7 21.3 15.9 14.6 12.0 9.3 11.1
12 7.4 2.4 2.4 6.9 3.8 13.6 8.0 11.2 7.3 5.3 10.2 8.7 7.1 40.0 7.1 8.3 7.8
13 4.9 2.7 1.2 7.8 4.1 5.5 3.8 15.4 9.5 4.0 4.0 3.2 5.1 9.4 3.9 6.7 5.6
14 29.9 4.7 8.1 4.1 8.1 28.5 13.6 31.2 3.1 11.0 3.3 9.5 14.0 10.6 14.9 9.1 10.9
15 5.4 1.0 1.3 3.4 2.4 2.5 2.1 4.5 2.7 1.6 2.6 2.7 1.3 3.7 1.6 3.0 2.4
16 11.0 6.2 5.3 11.4 10.3 6.4 8.0 11.5 4.1 4.1 1.3 10.3 4.8 19.3 14.8 10.7 8.7
17 10.4 1.9 3.6 11.0 8.6 2.7 7.5 17.9 3.8 6.5 3.2 6.5 4.8 19.3 10.3 8.1 7.1
18 8.3 2.9 2.3 11.3 2.9 2.9 6.9 45.6 2.3 1.0 1.1 2.1 1.8 6.4 5.0 2.8 3.1
19 19.6 18.7 5.8 13.2 10.8 6.7 19.1 27.2 54.5 5.5 2.1 12.0 53.1 16.7 7.6 24.6 18.9
20 43.0 43.0 139.8 20.5 49.3 20.8 45.5 41.8 51.6 94.4 64.7 80.2 44.8 142.7 80.1 147.1 97.9
21 21.6 11.4 13.3 16.0 21.7 11.2 16.3 8.0 13.6 11.5 16.3 23.4 15.7 27.4 20.0 25.9 19.7
22 3.3 5.5 2.8 3.5 0.5 0.6 0.8 8.0 0.6 0.3 0.1 1.3 1.7 0.1 2.7 1.2 1.2
23 14.1 6.2 10.4 37.5 17.7 13.6 21.1 17.7 9.8 34.0 50.8 13.9 12.9 45.0 20.8 19.8 19.9
24 12.4 2.1 5.1 7.2 7.7 3.8 6.0 23.8 4.2 8.9 8.0 8.5 7.8 7.1 7.1 6.1 6.5
C-avg. 12.8 5.7 9.1 9.7 8.3 6.3 8.4 15.5 7.5 8.6 9.3 10.7 9.0 20.6 13.2 11.9

1. Food, beverages and tobacco

2. Textiles, leather and footwear

3. Wood and products of wood and cork

4. Pulp, paper, printing & publishing

5. Coke, refined petroleum & nuclear fuel

6. Chemicals and chemical products

7. Rubber and plastics

8. Other non-metallic mineral

9. Basic metals & fabricated metal

10. Machinery, nec

11. Electrical and optical equip.

12. Transport equipment

13. Manufacturing nec; recycling

14. Electricity, gas & water supply

15. Construction

18. Hotels and restaurants

19. Transport and storage

20. Post and telecommunications

21. Financial intermediation

22. Real estate activities

23. Renting of M&E & other businesses

24. Community, social and personal services

Table A9: R&D expenditure intensity in value added (in percent), 1973–2004
Country AUS BEL CAN DNK FIN FRA GER IRL ITA JPN KOR NLD SPN SWE UK US Industry share
Industry
1 0.98 1.41 0.54 1.59 1.87 1.00 0.59 1.04 0.28 1.45 2.15 2.02 0.42 1.87 1.21 1.09 1.04
2 0.58 1.94 1.16 0.45 1.14 0.76 1.28 1.56 0.14 1.71 0.85 0.89 0.47 1.17 0.31 0.70 0.85
3 0.37 0.92 0.41 0.48 0.88 0.35 0.80 1.58 0.08 1.59 0.53 0.28 0.09 0.52 0.32 0.71 0.67
4 0.75 1.23 0.87 0.29 1.43 0.29 0.28 0.15 0.07 0.86 1.09 0.24 0.24 2.02 0.28 0.99 0.80
5 1.48 2.87 8.92 0.00 4.08 6.17 2.05 0.00 1.69 1.34 4.44 9.31 1.17 3.64 9.40 8.42 4.47
6 4.99 19.34 5.09 19.05 11.62 16.50 13.22 1.46 5.79 16.35 10.25 11.50 3.56 19.17 16.29 10.31 11.58
7 1.21 5.61 0.79 1.85 4.01 5.05 2.45 2.29 1.42 4.88 4.58 1.86 1.36 3.44 0.90 2.56 2.81
8 1.13 2.50 0.44 1.41 2.07 2.39 1.58 1.02 0.20 4.30 2.52 0.68 0.48 2.13 1.30 1.81 1.81
9 1.53 2.85 1.55 0.96 2.81 1.41 1.59 1.29 0.47 2.93 2.48 1.44 0.44 2.80 0.97 1.46 1.66
10 3.06 5.92 1.95 5.01 5.79 4.20 4.89 2.50 1.33 7.23 9.35 4.72 1.85 9.01 4.42 3.32 4.38
11 11.45 21.04 25.05 12.77 23.11 19.14 12.14 5.79 7.04 21.14 41.42 26.56 5.34 37.11 12.75 18.14 18.77
12 4.69 4.26 4.46 2.47 3.15 21.95 14.33 1.90 10.01 10.47 26.78 5.94 3.67 18.99 14.08 22.10 16.60
13 0.52 1.71 0.80 4.47 1.31 1.13 0.51 1.22 0.19 2.12 2.32 0.51 0.39 1.11 1.17 1.52 1.23
14 0.32 0.46 0.95 0.17 1.20 1.80 0.37 0.31 0.81 0.98 2.96 0.22 0.48 1.33 0.88 0.10 0.49
15 0.10 0.42 0.04 0.11 0.23 0.16 0.08 0.01 0.06 0.44 1.50 0.17 0.06 0.28 0.06 0.07 0.21
16 0.89 0.17 0.94 0.94 0.20 0.00 0.10 0.11 0.10 0.16 0.17 0.49 0.08 0.09 0.41 3.23 1.63
17 0.04 0.01 0.04 0.07 0.01 0.00 0.01 0.00 0.00 0.01 0.01 0.03 0.00 0.01 0.02 0.16 0.08
18 0.05 0.08 0.01 0.06 0.01 0.00 0.01 0.00 0.00 0.01 0.00 0.01 0.00 0.00 0.02 0.09 0.06
19 0.12 0.24 0.10 0.00 0.07 0.09 0.46 0.10 0.02 0.14 0.10 0.07 0.05 0.03 0.05 0.18 0.16
20 1.58 1.06 0.90 2.22 3.30 3.32 0.00 1.19 0.21 0.00 8.35 1.00 1.09 4.04 2.15 1.05 1.53
21 0.53 0.22 0.37 0.87 0.07 0.00 0.02 0.10 0.12 0.01 0.02 0.19 0.08 1.52 0.06 0.33 0.22
22 0.11 0.13 0.13 0.18 0.09 0.04 0.05 0.17 0.04 0.22 0.29 0.08 0.05 0.22 0.17 0.14 0.12
23 1.52 1.33 2.75 3.07 1.47 0.40 0.50 1.38 0.67 4.15 4.57 0.67 0.72 2.59 1.35 1.51 1.51
24 0.02 0.02 0.00 0.00 0.04 0.00 0.00 0.01 0.00 0.00 0.06 0.05 0.02 0.05 0.01 0.00 0.01
C-avg. 0.58 1.69 0.87 1.22 1.69 1.76 1.58 0.73 0.51 2.42 4.98 1.12 0.38 3.32 1.34 2.01

1. Food, beverages and tobacco

2. Textiles, leather and footwear

3. Wood and products of wood and cork

4. Pulp, paper, printing & publishing

5. Coke, refined petroleum & nuclear fuel

6. Chemicals and chemical products

7. Rubber and plastics

8. Other non-metallic mineral

9. Basic metals & fabricated metal

10. Machinery, nec

11. Electrical and optical equip.

12. Transport equipment

13. Manufacturing nec; recycling

14. Electricity, gas & water supply

15. Construction

18. Hotels and restaurants

19. Transport and storage

20. Post and telecommunications

21. Financial intermediation

22. Real estate activities

23. Renting of M&E & other businesses

24. Community, social and personal services

top of page

## Appendix B: Derivation of the Estimating Equation

We can write the production function (equation 1) in growth rates as:
(B1) $\Delta q=\frac{{F}_{{K}^{IT}}{K}^{IT}}{Q}\Delta {k}^{IT}+\frac{{F}_{C}C}{Q}\Delta c+\frac{{F}_{{K}^{NT}}{K}^{NT}}{Q}\Delta {k}^{NT}+\frac{{F}_{L}L}{Q}\Delta l+\frac{{F}_{Z}Z}{Q}\Delta z$

where $\Delta x=d\mathrm{log}X}{dt}$. If one assumes constant returns to scale and perfect competition, we will have

(B2) $\frac{{F}_{C}C}{Q}+\frac{{F}_{{K}^{IT}}{K}^{IT}}{Q}+\frac{{F}_{{K}^{NT}}{K}^{NT}}{Q}+\frac{{F}_{L}L}{Q}=\frac{{F}_{L}C}{Q}+\frac{{P}_{K}^{IT}{K}^{IT}}{PQ}+\frac{{P}_{K}^{NT}{K}^{NT}}{PQ}+\frac{WL}{PQ}=1$

If we observed total output $Q$, and knew the required rates of return to capital, we could back out the elasticity of output with respect to complementary capital, $C$ as follow:

(B3) $\frac{{F}_{C}C}{Q}=1-\frac{WL}{PQ}-\frac{{P}_{K}^{IT}{K}^{IT}}{PQ}-\frac{{P}_{K}^{NT}{K}^{NT}}{PQ}$

Without independent information on the flow of $A$ or the stock of $C$ (perhaps from stock market valuations), one cannot implement this procedure using measured output, $Y$. Rewrite the equation (B3) as:

$\frac{{F}_{C}C}{Y}=\frac{Q}{Y}-\frac{WL}{PY}-\frac{{P}_{K}^{IT}{K}^{IT}}{PY}-\frac{{P}_{K}^{NT}{K}^{NT}}{PY}$

Since $Q}{Y}$ is not observed, we cannot get away with this approach and hence we need to make an assumption regarding the use of intangible capital. We do that by assuming that observed growth in ICT capital provides a reasonable proxy for unobserved investment in, and growth in the stock of, complementary capital. Suppose $G$ takes a CES form:Footnote 14

$G={\left[\alpha {K}^{I{T}^{\frac{\sigma -1}{\sigma }}}+\left(1-\alpha \right){C}^{\frac{\alpha -1}{\alpha }}\right]}^{\frac{\sigma }{\sigma -1}}$

Consider the optimization sub-problem of producing $G$ at minimum cost each period. Let ${P}_{{K}^{IT}}}{{P}_{C}}$ be the relative rental rate of ICT capital to C-capital. The first order condition of cost minimization implies that:
(B4) ${\left(C}{{K}^{IT}}\right)}_{t}={\left(\frac{1-\alpha }{\alpha }\right)}^{\sigma }{\left({P}_{{K}^{IT}}}{{P}_{C}}\right)}_{t}^{\sigma }$,

(B4') $\Delta {c}_{t}=\Delta {k}_{t}^{IT}+\sigma \Delta \mathrm{ln}{\left({P}_{{K}^{IT}}}{{P}_{C}}\right)}_{t}$

This equation links growth in complementary capital and growth of observed ICT capital.

We can use the accumulation equation to express unobserved investment $\Delta a$ in terms of current and lagged growth in unobserved capital $\Delta c$: $\Delta {a}_{t}=\left(C}{A}\right)\left[\Delta {c}_{t}-\left(\left(1-{\delta }_{C}\right)}{\left(1+g\right)}\right)\Delta {c}_{t-1}\right]$. Substituting this expression in equation (3) in the text, we have $\Delta TFP=\frac{{F}_{C}C}{Y}\Delta {c}_{t}-\frac{C}{Y}\left[\Delta {c}_{t}-\frac{\left(1-{\delta }_{C}\right)}{\left(1+g\right)}\Delta {c}_{t-1}\right]+{s}_{z}\Delta z$. Substituting equation (B4') for $\Delta c$ into this expression, we have in principle an equation for TFP growth that indicates the importance of complementary capital accumulation:

(B5) $\Delta TF{P}_{t}=\left[\frac{{F}_{c}C}{Y}-\frac{C}{Y}\right]\left[\Delta {k}_{t}^{IT}+\sigma \Delta \mathrm{ln}{\left(\frac{{P}_{{K}^{IT}}}{{P}_{C}}\right)}_{t}\right]+\left[\frac{C}{Y}\frac{\left(1-{\delta }_{C}\right)}{\left(1+g\right)}\Delta {c}_{t-1}\right]+{s}_{z}\Delta z$

As an estimating equation, (B5) has the difficulty in a sense that industries are likely to differ in their longrun $C}{Y}$ ratios. Using equation (B4) and dividing both sides by $Y$ and multiplying and dividing the right-hand side by ${P}_{{K}^{IT}}}{{P}_{C}}$ and $P$, we have

$\frac{C}{Y}={\left(\frac{1-\alpha }{\alpha }\right)}^{\sigma }\left(\frac{P}{{P}_{C}}\right){\left(\frac{{P}_{{K}^{IT}}}{{P}_{C}}\right)}^{\sigma -1}\frac{{P}_{{K}^{TT}}{K}^{IT}}{PY}=\beta {s}_{{K}^{IT}}$, where $\beta ={\left(\frac{1-\alpha }{\alpha }\right)}^{\sigma }\left(\frac{P}{{P}_{C}}\right){\left(\frac{{P}_{{K}^{IT}}}{{P}_{C}}\right)}^{\sigma -1}$; ${s}_{{K}^{IT}}$ is the revenue share of ICT capital. Substituting this expression in equation (B5), we have

(B6) $\Delta TF{P}_{t}=\beta {s}_{{K}^{IT}}\left[{F}_{c}-1\right]\left[\Delta {k}_{t}^{IT}+\sigma \Delta \mathrm{ln}{\left(\frac{{P}_{{K}^{IT}}}{{P}_{C}}\right)}_{t}\right]+\beta {s}_{{K}^{IT}}\left[\frac{\left(1-{\delta }_{c}\right)}{\left(1+g\right)}\right]\left[\Delta {k}_{t-1}^{IT}+\sigma {\left(\frac{{P}_{{K}^{IT}}}{{P}_{C}}\right)}_{t-1}\right]+{s}_{z}\Delta {z}_{t}$

This is the equation reproduced as equation (4) in the text.

top of page

## Appendix C: Data Description

### 1. Converting current price data into constant price

From the EUKLEMS database we use data on industry gross output, industry value added, intermediate input, labor compensation, total capital compensation and both ICT and non-ICT capital compensation, all at current price. This database also have volume indices (1995 = 100) on gross output, value added, intermediate input, labor services, total capital services, ICT capital services and non-ICT capital services. Using their respective volume indices, we convert current price gross output, value added and intermediate input into constant price. For labor compensation, total capital compensation, ICT capital compensation and non-ICT capital compensation we use the indices of respective services to convert into constant price. For example, capital compensation at current price is converted into constant price using capital services.

We do the conversion from current price to constant price in the following way. Let us suppose that there are only three years 1990, 1995 and 2005 and we have variable called py (the product of price and volume). Then the time series of the variable at current price will be given by ${p}_{90}{y}_{90}$, ${p}_{95}{y}_{95}$ and ${p}_{2000}{y}_{2000}$. In that case, the data on volume indices (1995 = 100) were computed using following mechanism at the first place.

${i}_{1990}=\frac{{p}_{95}{y}_{90}}{{p}_{95}{y}_{95}}×100$; ${i}_{1995}=\frac{{p}_{95}{y}_{95}}{{p}_{95}{y}_{95}}×100$; ${i}_{2000}=\frac{{p}_{95}{y}_{2000}}{{p}_{95}{y}_{95}}×100$

Thus we have

 (C1) ${i}_{t}=\left({p}_{95}}{\left({p}_{95}{y}_{95}\right)}\right)×100×{y}_{t}$ t = 1990, 1995, 2000

Using this relation, the value at constant price can be computed as follows:

(C2) ${p}_{95}{y}_{t}=\left({i}_{t}×{p}_{95}{y}_{95}\right)}{100}$. Note that for ${p}_{95}{y}_{95}$, we can use the current value series for year 1995. So, multiplication of year specific indices with base year current price value (which would be the same at constant price as well) and division by 100 would provide the value at constant price.

With two sets of data on both current and constant prices, the value added deflator can be computed as:

${\mathrm{deflator}}_{t}={p}_{t}{y}_{t}}{{p}_{95}{y}_{t}}$

It will be equal to 1 for the reference year 1995. We have used this deflator to deflate the R&D data.

### 2. Capital Input Data

The EUKLEMS database has data on capital inputs by eight asset types, of which three are information and communication technology (ICT), and five are non-ICT assets. The three ICT assets are: (1) computing equipment (IT), (2) communications equipment (CT) and software. For detail on how these capital stock data are estimated, see Timmer et al (2007). In the EUKLEMS database, these capital stock data are available only for 13 countries: Australia, Austria, Denmark, Finland, Germany, Italy, Japan, Korea, the Netherlands, Portugal, Sweden, UK and the US. Among the remaining five countries in the sample, data for four countries (Belgium, France, Ireland, and Spain) were obtained through respective national offices of EUKLEMS consortium. For Canada, the data are from Statistics Canada. Note that for Belgium and Canada, there are only two types of assets available: ICT and non-ICT.

In terms of data availability, since the data on Germany were available only from 1991, the data for West Germany has been used till 1990. For Korea, the capital stock data start from 1977, and for Canada they start only from 1980. Data on Sweden start at 1993. The country with the least number of data points is Ireland, whose capital stock data are available only from 1995.

For all 10 Euro zone countries (Austria, Belgium, Finland, France, Germany, Ireland, Italy, the Netherlands, Portugal and Spain), the data are in Euros, whereas for the remaining eight sample countries they are in respective national currencies at constant price for year 1995. These data at Euros and other national currencies were converted to US$PPP using the industry level PPP data provided in the EUKLEMS website. In the EUKLEMS database, data are provided for 1997 PPP (in national currency per German Euro). These data have been converted to national currency per US$ using national currency per German Euro / US $per German Euro. Then the real variables on gross output, value added, labor compensation, total capital services, ICT capital services, NICT capital services, R&D stock at national currency were converted into 1997 US$ based PPP by dividing the national currency variable by PPP.

As for the EUKLEMS database, the data on capital compensation and share of ICT capital and non-ICT capital are given. We multiply share of ICT and non-ICT capital by value of capital compensation to decompose total value of capital compensation into value of ICT and non-ICT capital. Once these compensations for ICT and non-ICT capital are computed we use the indices of ICT capital service and non-ICT capital services to have constant price data.

### 3. R&D data

R&D data are available from the STAN ANBERD 2 and ANBERD 3 databases. The R&D data generally extends from 1973 to 2004 with few exceptions. For Belgium, the data are available from only 1987 and Korea only from 1995. The data for 2004 are missing for Australia, France, Japan, Sweden and the US. However, for these countries we have estimated the R&D investment for 2004 using the growth rate of R&D investment in 2003 over that in 2002.

The data are taken in national currency at current price. First, the national currency data are converted to national currency constant price using GDP deflator based on EUKLEMS mentioned above. Then the constant price data are converted to PPP US$using 1997 PPP exchange rate of national currency per US$ at PPP. Then these PPP US\$ adjusted constant price R&D expenditure data are used to compute R&D capital stock using perpetual inventory method with annual discount rate of 15%. To establish capital stock for the first year, 1973, we use average annual growth rate (across time from 1973 to 2004) which varies by both country and industry. In very few cases, the average annual growth rates were slightly negative; in that case, we use the growth rate of 3% instead of using average growth rate of across countries and industries in the sample.

Date modified: