Japan and the World Economy 28 (2013) 60–71
Contents lists available at ScienceDirect
Japan and the World Economy journal homepage: www.elsevier.com/locate/jwe
Growth and technological progress in selected Pacific countries Benedetto Molinari a, Jesu´s Rodrı´guez a, Jose´ L. Torres b,* a b
Department of Economics, University Pablo de Olavide, Spain Department of Economics, University of Ma´laga, Spain
A R T I C L E I N F O
A B S T R A C T
Article history: Received 21 June 2012 Received in revised form 15 April 2013 Accepted 11 August 2013 Available online 19 August 2013
This paper studies the sources of technological progress that determined output and labor productivity growth across a group of leading Pacific economies – Australia, Japan, South Korea, and the U.S. – in the period 1980–2006. We consider three alternative sources of technological progress: disembodied and factor-embodied technical change both to capital and labor. The contribution to growth of each of these sources is evaluated using both traditional and equilibrium growth accounting techniques. We find that capital accumulation is the main determinant of GDP growth in Australia, Japan and the U.S., whereas the main contribution in South Korea is given by Total Factor Productivity (neutral technology). In general, about a half of the contribution to growth of capital-embodied technical change comes from Information and Communication Technology in all the considered economies. We conclude that the higher growth of South Korea, due to Total Factor Productivity change, can be explained by changes in the intensity in the capital/labor use. ß 2013 Elsevier B.V. All rights reserved.
JEL classification: O3 O4 Keywords: Productivity growth Investment-specific technological change Neutral technical change Human capital accumulation
1. Introduction This paper studies the sources of technological progress that determined GDP and labor productivity growth across a group of leading Pacific economies – Australia, Japan, South Korea, and the U.S. – during the period 1980–2006. Technological progress is the key factor driving output and productivity growth in the long-run. From the seminal neoclassical growth theory, a large branch of theoretical and empirical literature investigated technical change as determinant of economic growth in modern economies, reaching a wide consensus about its primary role. This paper analyzes in detail the relationship between technology and growth by studying among the sources of technological progress which is the one that mostly affected and determined the growth rate of output and labor productivity. The sources of technological progress are various and heterogenous. Technical change may occur as neutral progress, as investment-specific progress, or as labor efficiency progress. While the first is typically associated with multifactor productivity,1 e.g.,
* Corresponding author. Tel.: +34 952131296; fax: +34 952131299. E-mail address:
[email protected] (J.L. Torres). 1 In the rest of the paper we refer to this source indistinctly as disembodied or neutral technical change. 0922-1425/$ – see front matter ß 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.japwor.2013.08.005
improvement of business organization or institutional factors, the second represents the improvement of capital efficiency in production due to technical change embodied in capital assets. The last form of progress refers to every possible source of improvement in labor efficiency, i.e., higher fraction of skilled workers, learning-by-doing tasks, or enhanced accumulation of common productive knowledge spread in the society. We group all these types of progress as human capital accumulation and incorporate it as a factor-embodied technical change associated with labor. Although a large branch of literature emphasized the importance of human capital as a source of output and productivity growth,2 still growth decomposition exercises rarely account for the effect of human capital [henceforth, HC], thus imputing to Total Factor Productivity [henceforth, TFP] its contribution to growth. In our opinion, such miscalculation is not a minor issue and it may have costly implications for the policy maker insofar as policies targeted on TFP are different from the ones meant to enhance human capital in the society. Similarly, most growth decomposition exercises do not account for capital embodied technical
2 There is a countless literature on the importance of human capital as determinant of growth, starting with Schultz (1961) and Denison (1962), among others.
B. Molinari et al. / Japan and the World Economy 28 (2013) 60–71
change as a technological factor different from disembodied technical change. We identify investment-specific technical change [henceforth, ISTC] using the series of quality-adjusted prices of investment. These prices are constructed combining the information contained in EU KLEMS database3 with the quality-adjusted investment prices for the U.S. estimated by Gordon (1990) and extended by Cummins and Violante (2002) [henceforth, the GCV]. These GCV prices refer to the U.S. economy. For the analysis presented in this paper, we extend the GCV database to the period 2000–2006 and use the methodology proposed by Schreyer (2002) to obtain quality-adjusted series for Australia, Japan and South Korea.4 We disaggregate the measures of capital to scrutinize the marginal effect of each investment asset. Particularly, we compute ISTC for each asset category given in the EU KLEMS database, namely: (i) hardware, (ii) software, (iii) communication equipment – these three typically referred to as Information and Communication Technology [henceforth, ICT] equipment – (iv) transport, (v) machinery and (vi) other equipment – or non-ICT equipment – and (vii) structures. The quality and efficiency improvements widely differ among different assets: ICT assets have bolstered productivity more effectively than earlier technologies, and have had a definite impact on the economy. Numerous studies have pointed to the special role played by these technologies in the recovery of productivity growth since the mid-1990s in the United States and some European countries (see among others, Colecchia and Schreyer, 2002; Stiroh, 2002; Timmer et al., 2003; Timmer and van Ark, 2005). Once we identify the technological progresses embodied in factors (ISTC and human capital improvement), we use two different approaches to estimate neutral progress: (i) the traditional growth accounting decomposition and (ii) a calibrated general equilibrium model. There is a long-standing debate in the literature about which of these two approaches better identifies the determinants of growth, e.g., Greenwood et al. (1997), Hulten (1992), Oulton (2007) and Greenwood and Krusell (2007). We take a neutral stance in this debate, thus performing our analysis using both approaches. In turn, for the traditional growth accounting, we report the results using three versions of it: the traditional one proposed by Solow (1956), the one suggested by Jorgenson (1966) and the one of Hulten (1992). Regarding the general equilibrium approach, we use an extension of the Greenwood et al. (1997) model, developed in Rodrı´guez-Lo´pez and Torres (2012) augmented with endogenous human capital accumulation. Our main finding is that growth has a similar composition in Australia, Japan and the U.S., while it has an opposite pattern in South Korea. We find that factors accumulation explains about 60–70% of output growth in Australia, Japan and the U.S., whereas technological progress explains about 30–40%. In the case of South Korea, factors accumulation only explains approximately 36% of output growth, while technology explains 64%. We show that this difference is explained by the neutral technical change. Whereas in the first three countries, TFP contribution to output growth is close to zero or even negative according to Hulten’s decomposition, in South Korea neutral technology alone explains approximately 40% of output growth, being the most important determinant of growth in that country. Similar results are obtained in terms of labor
61
productivity growth. The general equilibrium approach confirms these results, showing that in Australia, Japan and the U.S. the main factor of labor productivity growth in the long-run is ISTC. In the case of South Korea the main contribution comes from TFP, followed by human capital. Additionally, we obtain that TFP contribution to productivity is positive for South Korea and Japan, but negative for the U.S. and Australia. Getting to a more detailed analysis, we show that ISTC contribution to growth is similar in Japan and South Korea, 0.79 and 0.70 percentage points, respectively, 0.96 percentage points in Australia and 1.09 percentage points in the U.S. However, using the Jorgenson (1966) approach, the larger contribution to output growth from ISTC corresponds to South Korea (0.71 percentage points), whereas for the other three countries the contributions is lower (0.58 for Australia, 0.46 for Japan and 0.49 for the U.S.). The differences between the two approaches are explained by the much higher capital investment process in South Korea compared to the other countries. Our results show that the large output and productivity growth observed for South Korea, due to TFP or neutral technological change, can be explained by changes in the intensity in the capital/ labor use. Efficiency gains by increasing the capital/labor ratio drives Total Factor Productivity of the Korean economy. In fact, capital factor shares of South Korea are lower than the observed for Australia, Japan and the U.S. Moreover, the capital share has increased over the selected period for the Korean economy, evincing a pattern consistent with an economy in transition to a balanced growth path corresponding to more advanced economies. The rest of the paper is structured as follows. Section 2 describes the data set and the logic of the calibration. Section 3 introduces the different growth accounting approaches: the statistical growth accounting and the equilibrium growth accounting. Estimates of the contribution to output and labor productivity growth are presented in Section 4. Finally, Section 5 summarizes and concludes. 2. Data Although originally created to keep track of economic growth in European countries, the EU KLEMS Database also reports data on some non-European countries. We use it to collect Australian, Japanese, Korean, and U.S. data on nominal and real output and productive factors compensations, on the amount (measured as total worked hours) and quality of labor services, and finally on nominal and real investment in physical capital break up in seven categories: (i) hardware and office equipment, (ii) communication equipment, (iii) software, (iv) transport equipment, (v) machinery, (vi) other equipment, (vii) structures. Data on investment are then used to construct the series of capital stock by mean of the permanent inventory method. The upper panel of Table 1 reports the mean value over the period 1980–2006 of annual growth rates for the set of considered variables. The average growth rate of real output has been fairly similar across Australia, Japan, and the U.S. (around 3%), while it has been sensibly higher in the case of South Korea (approximately Table 1 Average annual growth rates, 1980–2006. Australia
Japan
Korea
U.S.
3.50 1.84 1.66 3.58
2.29 2.55 0.26 3.79
7.70 6.33 1.37 8.19
2.92 1.64 1.28 3.51
0.64 3.20
0.69 1.95
1.53 2.39
0.30 3.27
3
All details about EU KLEMS project can be found at http://www.euklems.net. 4 In fact, the EU KLEMS database uses Schreyer’s methodology to quality-adjust the prices of ICT equipment starting from the corresponding NIPA prices. GCV prices are used because non-ICT equipment prices are not quality-adjusted in EU KLEMS. The updated series of quality-adjusted prices for all asset categories is a key contribution of this paper. It is worth noting that if only quality-adjusted ICT prices are used, then growth accounting exercises tend to overweight the importance of ICT as a factor of growth behind the 1995 U.S. productivity growth upsurge (see, for example, Collechia and Schreyer, 2001; Jorgenson and Stiroh, 2000).
Variables Output Labor productivity Worked hours Capital Technology HCI ISTC
62
B. Molinari et al. / Japan and the World Economy 28 (2013) 60–71
7.70%). This higher rate in South Korea went along with an enhanced accumulation of physical capital, which has grown almost three times faster than in the other three countries, i.e., 8.2% on annual basis against around 3.5%. The growth of labor appears more stable over the period, with a similar 1.5% annual increase of worked hours in U.S., Australia, and South Korea. Japan constitutes an exception, showing a negative average growth rate in the considered period (column 3).5 The growth rate of labor productivity has been similar in Australia and U.S., but relatively higher in South Korea and Japan, in the first case clearly due to the enhanced growth of output, and in the second case to the negative growth of labor. The lower panel of Table 1 reports the averaged annual growth rate of technological progress in physical capital and of Human Capital. The index of Human Capital Improvement [henceforth, HCI] is constructed using the Jorgenson et al. (1987) approach which explicitly takes account several features of the labor force, namely: educational attainment, age, industry, and gender. These four features are provided by the EU KLEMS database. The average annual growth rate of HCI in South Korea has been of 1.53%, indicating a strong rise in labor efficiency. Australia and Japan show a similar pattern of HCI, with an average annual growth rate of 0.70%, while U.S. shows a significantly lower growth rate (0.30%), which is possibly explained by higher initial level of HCI to which the other countries have been converging. To pin down the growth rate of technical change embodied in capital, we construct a series of ISTC using data on quality-adjusted prices of investment. As in Greenwood et al. (1997), a variation in the real price of quality-adjusted investment goods is interpreted as a technology improvement specific to that good, which affects its average cost of production. EU KLEMS does not provide all the information to construct the ISTC because it only reports data on quality-adjusted prices for ICT assets but not for non-ICT assets. To overcome this issue, we combine EU KLEMS data with the series of quality-adjusted price of equipment and machinery provided by Gordon (1990) and later extended by Cummins and Violante (2002) for the U.S. [henceforth, GCV]. Using a To¨rnqvist index weighted with nominal investment shares, GCV data are used to build U.S. category-specific annual deflator indexes, qUS i;t , one for each EU KLEMS category of nominal investment i. Then, similar harmonized deflator indexes for Australia, Japan, and South Korea are obtained applying the Schreyer’s (2002) methodology to the U.S. data.6 j ISTC are finally obtained using the expression Qi;tj ¼ PC t =qi;t for i 2 {1, . . ., 7} and j 2 {Australia, Japan, South, Korea, the U . S . }, where PCt is the price index for consumption of non-durables and j services less housing, and qi;t are the quality-adjusted price indexes. This expression represents the amount of capital that can be purchased by one unit of output at time t, and we interpret an increase of Qi;tj as a positive technology innovation that reduces the average cost of production of investment good i expressed in units of consumption good. We assume that all assets but structures are j subject to efficiency improvements from ISTC, i.e., Qstr;t ¼ 1 for 8t, j. As in the case of HCI, the annual growth rate of ISTC also exhibits significant differences across countries. It is higher in U.S. and Australia (above 3% on average), and lower in South Korea and Japan (2.4% and 1.95%, respectively). While it is reasonable to expect a lower figure in South Korea as compared to the U.S. one, such low number for Japan comes at a surprise. By analyzing the
5
A detailed analysis about the behavior of worked hours in Japan can be found in Hayashi and Prescott (2002). 6 A detailed explanation of the construction of series can be found in the technical Appendix of this paper. A similar application of the Schreyer’s methodology to obtain harmonized deflators is given in Basu et al. (2003), who compare the evolution of productivity in U.K. and U.S.
composition of capital, however, we find that the fraction of traditional capital assets (non-ICT) over total assets is sensibly larger in Japan than in the other countries, which could explain the lower aggregate figure, given that the technical change associated with non-ICT assets is typically lower than that of ICT. 3. Growth decomposition The first and still widely employed methodology to study the determinants of output growth is the traditional or statistical growth accounting.7 This approach identifies the unobservable contribution of technological progress to growth as the unexplained residual of actual output growth rate after controlling for the growth rates of production factors. In other words, the observed growth of output is either imputed to the increasing amount of factors employed in production, or to the development of better technologies to produce. Starting with the seminal paper of Greenwood et al. (1997), a subsequent literature proposed to use a general equilibrium growth model to obtain a structural decomposition of growth. There are several advantages in pursuing this strategy against the traditional growth accounting. First, the general equilibrium growth model is the workhorse of modern economics and the accepted paradigm for studying most macroeconomic phenomena, including business cycles tax policy, and monetary policy. Second, it is consistent with an agents’ optimization based framework. Third, it features several desirable properties in accordance with the predictions of economic theory, as the balanced growth path and the fact that only technological progress can explain labor productivity growth in the long run. As pointed out by Cummins and Violante (2002), this last property better fits the determinants of productivity growth in the long-run, given that part of the observed growth in capital is in fact generated by technical change itself. 3.1. Statistical growth accounting In this paper we employ three versions of the traditional or statistical growth accounting approach: (i) the original one developed by Solow (1956), (ii) the refinement of that theory proposed by Jorgenson (1966), and (iii) the subsequent refinement of Hulten (1992). We extend the last two methods by including human capital as an additional embodied technical change to the labor input. Jorgenson criticized Solow’s theory on the ground that it considers only TFP as source of technical change, thus neglecting efficiency improvements in capital assets due to investmentspecific technical change. He argues that Solow erroneously accounted ISTC as neutral technology, thus distorting the true dynamics of TFP. Later on, Hulten (1992) proposed a refinement of Jorgenson’s approach to identify the contribution of ISTC to growth. We employ two extended versions of Jorgenson’s and Hulten’s theories that incorporate the observed progress in human capital as additional source of growth. Specifically, we add an extra term to their growth accounting equations to capture the efficiency improvement of labor services generated by the accumulation of human capital embedded in workers. We leave the original Solow’s method untouched, and we use it as benchmark to identify the effect on growth of factors-embodied technological progress. According to Solow (1956), the actual growth rate of output, labeled gY, can be decomposed into:
g Y ¼ g SA þ |{z}
Neutral
X
vi g K i
i |fflfflfflfflffl{zfflfflfflfflffl}
þ
vg
l L |ffl{zffl}
(1)
Labor accumulation
Capital accumulation
7 The term traditional or statistical to refer to Solow-alike Growth Accounting method was firstly used in Cummins and Violante (2002).
B. Molinari et al. / Japan and the World Economy 28 (2013) 60–71
where g K i is the growth rate of capital Ki, gL is the growth rate of labor input (hours) Lt, and At is the Total Factor Productivity or Solow’s residual. The coefficient g SA thus measures the contribution of neutral technology to output growth as identified by Solow. The weights vi are the elasticities of output with respect to capital asset i, usually measured as the ratio of the marginal to the average product of capital, and vl is the elasticity of output with respect to labor measured as the ratio of the marginal to the average product of labor. As it is standard in the literature, we assume a CobbDouglas production function and therefore, elasticities of output with respect to inputs can be measured as the input-income shares. Using the same approach, the growth of labor productivity, gY/L labeled g, can be decomposed as follows: X g Y=L g Y g L ¼ g SA þ vi ðg K i g L Þ (2) |{z} i |fflfflfflfflfflfflfflfflfflfflfflffl ffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} Neutral Capital deepening
P where the second equality uses the fact that vl ¼ 1 i vi .8 Our version of Jorgenson’s (1966) approach explicitly takes into account the existence of both ISTC and enhanced human capital [HCI] that improve the efficiency of, respectively, capital and labor services. Accordingly, output growth is decomposed into: X X g Y ¼ g JA þ vi g K i þ zi g i þ vl g L þ vl g H |ffl{zffl} |ffl{zffl} |{z} i i HCI |fflfflfflfflffl{zfflfflfflfflffl} |fflfflfflffl{zfflfflffl ffl} Labor accumulation Neutral Capital accumulation
ISTC
(3) J A
where g is the neutral technological progress as defined by Jorgenson and gi is country-specific growth rate of ISTC in capital i. The weights zi measure the ratio of nominal investment in asset i to nominal GDP. Similarly, labor productivity growth is decomposed as follows: X X g Y=L g Y g L ¼ g JA þ vi ðg K i g L Þ þ zi g i þ vl g H (4) |ffl{zffl} |{z} i i HCI |fflfflfflfflfflfflfflfflfflfflfflffl ffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflffl{zfflfflffl ffl} Neutral Capital deepening
ISTC
Finally, Hulten (1992) argued that technical innovations as the ones which are usually referred to as ISTC in the literature, would ideally affect the whole stock of capital and not just new investments, and their contributions to growth should therefore be weighted accordingly. As a result, in our version of Hulten approach output growth is decomposed into: X X g Y ¼ g UA þ vi g K i þ vi g i þ vl g L þ vl g H |ffl{zffl} |ffl{zffl} |{z} i i HCI |fflfflfflfflffl{zfflfflfflfflffl} |fflfflfflffl{zfflfflfflffl} Labor accumulation Neutral Capital accumulation
ISTC
(5) and labor productivity as follows: X X g Y=L g Y g L ¼ g UA þ vi ðg K i g L Þ þ vi g i þ vl g H |ffl{zffl} |{z} i i HCI |fflfflfflfflfflfflfflfflfflfflfflffl ffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflffl{zfflfflfflffl} Neutral Capital deepening
3.2. Equilibrium growth accounting We build on Rodrı´guez-Lo´pez and Torres (2012) to provide a structural decomposition of output growth developing a general equilibrium growth model in which three key elements are present: (i) the existence of different types of capital, (ii) the presence of technical change specific to each capital equipment, (iii) the presence of Human Capital that enhances the efficiency of labor services in the production function. Output is accordingly produced as a combination of eight productive factors: worked hours Lt and the seven different capital types Ki for i = {1, . . ., 7} that, consistently with our data set, are grouped into three broad categories: non-residential structures Kstr,t, non-ICT equipment and machinery Knict,t, and ICT equipment Kict,t. According to EU KLEMS statistics, this last category comprises hardware, software, and communication networks, while non-ICT equipment comprises transport equipment, machinery, and other equipment. We assume that ISTC and HCI affect only the productivity of the corresponding factor, and that both are exogenous to agents’ optimal decisions. In the model, the representative agent allocates non-leisure time between production and learning. In this context, the steadystate equilibrium rate of growth of the economy depends on the allocation of time to acquiring education. Two goods are produced in the economy: a final good and a human capital good. The final good can be used for three purposes: consumption, physical capital investment and education (or human capital investment). Household. The model economy is populated by an infinitely lived representative agent who maximizes the expected value of her lifetime utility: E0
1 X
bt UðC t ; Ot Þ
ISTC
From a direct comparison of Eqs. (3) and (5), it appears clear how Hulten (1992) differs from Jorgenson (1966). While he uses capital shares vi to weigh for the contributions of ISTC, the second uses investment ratios zi, and the difference between g JA and g U A can thus be used as a measure of the effect of different weights of ISTC on the implied contribution of neutral technology to growth.
8 We restrict our analysis to the case of constant returns to scale which implies that the sum of ratios of the marginal to the average product of capitals plus the one P of labor must sum to one, i.e., vl þ i vi ¼ 1.
(7)
t¼0
where Ct and 0 < Ot < 1, represent consumption and leisure, perspectively, b is the agent’s subjective discount factor and E0 is the conditional expectation operator at time 0. Following DeJong and Ingram (2001), we assume that the agent supplies two types of labor inputs to the market: raw labor and human capital. Raw labor is the constant labor input that the agent was born with, while human capital is the skills that are acquired by the agent either through formal schooling or through on the job training. This formulation of labor inputs allows us to discuss skills and human capital formation without having to introduce different types of agents, e.g., high-skilled and low-skilled. Non-leisure time is split between time on the job (total number of worked hours producing output), Lt, and time in education (producing human capital), St. The household time restriction is defined as Ot þ Lt þ St ¼ 1
(6)
63
(8)
where total number of effective hours have been normalized to one. The agent’s utility function is given by UðC t ; Ot Þ ¼ g lnC t þ ð1 g Þlnð1 Lt St Þ
(9)
where 0 < g < 1 is the elasticity of substitution between consumption and leisure. Each household saves in the form of investment, It, and receives capital interest income RtKt where Rt is the return to capital and Kt is the physical capital stock. Total labor earnings are given by WtHtLt where Ht is the households stock of human capital and Wt is the wage. Note that human capital only receives income if associated with working time. The household budget constraint is defined as: X X Ii;t ¼ Ri;t K i;t þ W t Ht Lt (10) Ct þ i
i
B. Molinari et al. / Japan and the World Economy 28 (2013) 60–71
64
where Ii,t is investment is asset type i, Ri,t is the rental price of asset type i, and Ki,t is capital stock of asset Ki,t. Capital holdings evolve according to: K i;tþ1 ¼ ð1 di ÞK i;t þ Q i;t Ii;t
(11)
where di is the depreciation rate. As mentioned in Section 2, Qi,t represents the amount of asset i than can be purchased with one unit of consumption good. We interpret this relative price as a measure of the current state of technology to produce i. In fact, when a technical innovation in the production of asset i occurs it either increases the quality of the good, adds new characteristics or lowers its average cost of production. In either case, such innovation is reflected in a variation of the quality-adjusted relative price Qi,t, which therefore can be used to represent the technical change specific to the investment good i. In general, Qi,t may increase or decrease over time according to the technical change actually occurred. In this paper, however, we neglect the cyclical behavior of ISTC, and we only focus on its long run trend by assuming that ISTC evolves according to: Q i;t ¼ ð1 þ g Q ;i ÞQ i;t1
(12)
where gQ,i > 0 is the exogenous growth rate specific to asset i. Finally, we incorporate in the model the standard assumption of no investment-specific technical change in structures, i.e., Qstr,t = 1 8 t.9 The stock of human capital evolves according to Htþ1 ¼ ð1 dH ÞHt þ IH;t
(14)
where 0 < u < 1. New investments in human capital are producing by combining the existing stock of human capital with the available time spent investing in education. The efficiency of new human capital production is governed by Bt and u. Human capital depreciation, 0 < dH < 1, reflects the aging and replacement of the population. That is, we have to continually train new cohorts in order to maintain the stock of human capital. One can also see this model as one with vintage human capital. New skills are needed to design, introduce and/or use the new, more efficient capital equipment, while some skills become obsolete as older vintages of capital become obsolete. As far as u is positive but smaller than one, expression (14) preserves the law of diminishing returns to education. The problem faced by the household is to choose a sequence fC t ; Lt ; St ; ðIi;t Þi g1 t¼0 for to maximize utility (9), subject to the budget constraints (10) and the laws of motion (11) and (13), given initial conditions Ki,0, H0. Firms. There exists a single representative firm in the economy, who produces and sells whichever quantity of output demanded by the household in a perfectly competitive goods market. The problem of this firm is to find the optimal values for the utilization of labor and of the different types of capital assets to maximize its profits at each period t. Both capital and labor services are hired in perfectly competitive factors’ markets. The technology to produce is given by a constant return to scale Cobb–Douglas production function, Y t ¼ At
Y a Ki;ti ðHt Lt Þal
(15)
i
9
i
Notice that both output and investment are measured in units of consumption. The equilibrium outcome for this model economy is characterized by the law of motion of capital assets (11), the production function (15), the market clearing condition (16) and the following first-order conditions: 1g
g
C t ¼ W t Ht ð1 Lt St Þ
(17)
1g bg W tþ1 Ltþ1 ¼ Et u C tþ1 ð1 Lt St Þð1 uÞBt Htu S t þ bEt
u 1u ð1 g Þ½1 dH þ uBtþ1 Htþ1 Stþ1
u ð1 Lt St Þð1 uÞBt Htu S t
(18)
(13)
where IH,t is the investment in skill formation. Both time in education and goods are needed for skill acquisition activities, u IH;t ¼ Bt Htu S1 t
where At is Total Factor Productivity and 0 < al, : ai < 1. In Eq. (15), the services from labor are expressed in efficiency units. This last is controlled by the amount of human capital embedded in workers, with the standard assumption that a higher level of human capital entails a better ability to workers and, therefore, a higher quality of labor. Equilibrium. A standard market clearing condition is used to close the model. Output Yt can be purchased to consume or to invest in each capital asset, X Ii;t (16) Y t ¼ Ct þ
In the literature on ISTC, structures are typically used as the benchmark capital where no technical change occurs. Gort et al. (1999), however, estimated that NIPA prices for non-residential structures should be quality-adjusted by a 1% on annual basis.
1 ¼ bEt
C t Q i;t ðQ R þ 1 di Þ C tþ1 Q i;tþ1 i;tþ1 i;tþ1
(19)
Eq. (17) is the intratemporal condition that equates the marginal rate of substitution between consumption and leisure representing the opportunity cost of one additional unit of leisure. Eq. (18) is the intertemporal condition that equates the marginal cost of human capital with the discounted expected marginal benefit of human capital. Finally, the system given by (19) is the Euler-equations system for capital investment in each asset, where the marginal cost of each capital asset is equated to its discounted expected marginal benefit. The balanced growth path and growth decomposition. We focus our attention to the balanced growth path equilibrium where all variables grow at constant rates. According to the model, output, consumption and investment all share the same long run growth rate g Y ¼ g C ¼ fg I;i gi¼1 . In this model, hours worked grow by the population growth rate, which is normalized to zero. Similarly, in the balance growth path the growth rate of education is zero, that is, gL = gS = 0. From the law of motion (11), the balanced growth rates of the different types of capital are ð1 þ g K;i Þ ¼ ð1 þ g Q;i Þð1 þ g Y Þ
(20)
Expression (20) states that the stock of capital i grows by the common growth rate of the economy times the rate of its progress of ISTC. Combining the production function (15) and the growth rates in (20), the log-linear expression for the balanced growth path is X ai g g Y=L ¼ A þ g þ gH (21) al al Q ;i |{z} i |{z} |fflfflfflfflfflffl ffl{zfflfflfflfflfflfflffl} HCI Neutral
ISTC
where gA is the growth of TFP. Expression (21) states that the long run growth rate of labor productivity can be decomposed into a combination of (i) the growth rate of neutral technology, (ii) the growth rates of ISTC, (iii)
B. Molinari et al. / Japan and the World Economy 28 (2013) 60–71
and the growth rate of HCI. Note that in the long run the growth rate of output coincides with the growth rate of labor productivity. Hence, the model predicts that only technological progress explains output growth in the long run. This property characterizes the most remarkable difference between equilibrium and traditional growth accounting procedures. As pointed out by Greenwood et al. (1997) and Cummins and Violante (2002), it also better fits the determinants of productivity growth in the long-run, given that part of the observed variations in capital deepening are the results of technical change and therefore endogenous.
65
shares appear fairly similar in the case of South Korea, while there are significant differences in Australia, Japan, and the U.S. In general, capital income shares are higher than investment shares as expected, and accordingly Hulten’s approach will assign a higher weight to the contribution of ISTC to growth than that of Jorgenson. Regarding the equilibrium growth accounting, we calibrate the coefficients a’s in Eq. (21) exploiting the steady state relationship of the model. In equilibrium,
ai ¼
Ri K i Y
(23)
al ¼
W i Li Y
(24)
3.3. Calibration The parameters that appear in the growth accounting approaches in Sections 3.1 and 3.2 are calibrated targeting some long run empirical facts of our set of countries. In particular, the elasticities of output with respect to capital i and labor, fvi ; vl g, are measured with the ratio of the marginal to the average product of the corresponding factor. Given the national accounting rules and under the hypothesis of perfectly competitive factor markets, these elasticities coincide with the income shares per factor. As regards the cost shares of capital, we follow the recommendations of OECD (2001) to construct the series of capital, which are based on the concept of capital services. The idea is to capture the productive services embedded in the stock of capital. This concept of productive capital can be seen as a volume index of capital services. The expression driving the concept of capital services for the asset i is as follows: VCSit ¼ mit K it
(22)
where mit is, in turn, the nominal usage cost of capital. Let REt denote the total compensation of employees. Then, income shares are given by the following expressions:
vi;t ¼
VCSit P REt þ i VCSit
For calibration purposes, we use the average values of cost shares over the period 1980–2006 as estimators for the weights vi , P i.e., vi ¼ T 1 t vi;t . The labor cost share is then obtained residually P using the constant return to scale condition, i.e., vl ¼ 1 i vi . Table 2 shows the income shares of capital together with the investment shares, which replaces the capital shares in Hulten’s approach as weights for ISTC. Investment and capital income Table 2 Cross-country empirical evidence, 1980–2006.
Capital-income shares (vi ) ICT equipment (i + ii + iii) Hardware (i) Software (ii) Communication equipment (iii) Non-ICT equipment (iv + v + vi) Transport equipment (iv) Machinery (v) Other equipment (vi) Structures (vii) Investment-income shares (zi) ICT equipment (i + ii + iii) Hardware (i) Software (ii) Communication equipment (iii) Non-ICT equipment (iv + v + vi) Transport equipment (iv) Machinery (v) Other equipment (vi) Structures (vii)
Australia
Japan
Korea
U.S.
0.046 0.021 0.015 0.010 0.141 0.048 0.073 0.020 0.164
0.043 0.019 0.013 0.011 0.181 0.031 0.101 0.049 0.180
0.027 0.011 0.008 0.008 0.108 0.030 0.079 0.000 0.069
0.062 0.020 0.024 0.018 0.142 0.035 0.098 0.008 0.147
0.028 0.012 0.010 0.006 0.085 0.031 0.044 0.010 0.082
0.026 0.011 0.008 0.006 0.098 0.021 0.051 0.026 0.081
0.028 0.007 0.011 0.010 0.119 0.037 0.083 0.000 0.145
0.031 0.009 0.013 0.009 0.057 0.016 0.038 0.003 0.055
According to Eq. (23), ai corresponds to capital-income share, and therefore ai ¼ vi . Consistently with this measure, we calibrate al ¼ vl . Finally, depreciation rates are calibrated taking the ratio of EU KLEMS estimates of the stock of capital i and of the gross formation of fixed capital. These estimates are stable across years and similar across countries. Across all considered countries, structures depreciate by 2.8% on annual basis, while ICT equipment depreciates much faster. For instance, a software license fully depreciates in about two years, implying a depreciation rates of 42%. This time length is four years for hardware equipment.10 4. Results We begin the quantitative analysis by providing first output growth decomposition under the traditional growth accounting. The contributions of both types of embodied technical progresses (ISTC and HCI), disembodied technical progress, and capital and hours accumulation in the four selected countries are reported in Table 3 for each of the three alternative methods. Several results are worth noticing. First, the three alternative traditional growth accounting approaches all assign the same fraction of growth to technological sources. Hence, the contribution of overall technological progress to growth is not sensitive to the approach used. As apparent inspecting expressions (1), (3), and (5), this result is due to the same coefficients fvi ; : vl gi used in the different models to weight for the relative contributions of factors accumulation. Important differences, however, exist in the relative contribution to growth of the various sources of technological progress. As expected, the impact of neutral technical change is highest in Solow’s than in the other two models, given that Solow’s TFP incorporates the effects of ISTC and HCI thus biasing the results upward. In general, neutral technical change contributes more to output growth in Jorgenson’s than in Hulten’s approach. This is a reasonable finding because Jorgenson only recognizes the existence of embedded technological progress in new capital assets (investment), while Hulten considers that ISTC affect the productivity of the whole stock of existing capital, thus shifting production more heavily. Second, we find that factors accumulation is the main driving force, in relative terms, of output growth in Australia, Japan and the U.S., explaining more than two-third of output growth, while technological progress explains the remaining one-third. In detail, capital accumulation explains approximately 42% of total output growth in the U.S., 36% in Australia, and 67% in Japan, while the growth of hours contributes to approximately 28% of output growth in the U.S., 30% in Australia but has a negative contribution in Japan (6.8%), being possibly the responsible for the lower output growth in this country with respect to U.S. and Australia. In 10 A more detailed description of depreciation rates for each capital asset can be found in the Data Transformation Appendix of the paper.
B. Molinari et al. / Japan and the World Economy 28 (2013) 60–71
66 Table 3 Contribution to GDP growth, 1980–2006. GDP growth (gY)
Australia
Japan
3.50 Sol. Contribution Capital Labor TFP ISTC HCI Decomposition (%) Capital (i) Labor (ii) Factors (i + ii) TFP (iii) ISTC (iv) HCI (v) Technology (iii + iv + v)
1.26 1.08 1.17 – – 35.9 30.8 33.3 – –
Korea
2.29 Jorg.
U.S.
7.70
Hul.
Sol.
Jorg.
1.26 1.08 0.17 0.58 0.42
1.26 1.08 0.20 0.96 0.42
1.53 0.15 0.91 – –
1.53 0.15 0.04 0.46 0.41
1.53 0.15 0.29 0.79 0.41
35.9 30.8 66.7% 5.0 16.5 11.9 33.3%
35.9 30.8
67.0 6.8
67.0 6.8
21.8 14.1
5.9 27.3 11.9
39.8 – –
67.0 6.8 60.2% 2.0 19.9 18.0 39.8%
12.6 34.4 18.0
64.1 – –
the case of South Korea the results are the opposite. Approximately 64% of output growth is accounted for by technological progress and only 35% by factors accumulation. This finding broadly confirms that Korean economy is still at different stage of economic development than the other three countries. Nevertheless, absolute contributions to output growth from ISTC and capital accumulation are similar across the four selected countries. In fact, capital contribution to output growth ranges from 1.23 percentage points for the U.S. to 1.68 percentage points for South Korea. Moreover, labor input contribution is also similar across countries (around 1.2 percentage points for Australia and the U.S. and around 1.7 percentage points for South Korea) but for Japan, in which case the contribution from hours is negative as shown before. Although contribution from capital accumulation is similar across the four selected countries, behind this result important differences exist between South Korea and the other three countries. The higher capital accumulation process in South Korea (see Table 1) is compensated by a lower capital income share for this economy which results in a similar contribution to output growth from this factor similar to the other three countries, that in spite of a lower rate of capital accumulation across time their capital income shares are larger. Jorgenson’s and Hulten’s approaches provide a closer look at the actual sources of technology, revealing our third result. The contribution of neutral technology in Australia, Japan and the U.S. is extremely small using Jorgenson (from 0.04 to 0.17 percentage points) and even negative using Hulten (from 0.2 to 0.43 percentage points). Only in South Korea TFP plays an important role whichever is the model used to account for it, with an annual contribution to output growth that goes from about 5% with Solow to approximately 3% with Jorgenson and Hulten. This finding can be explained as empirical counterpart of the theories of TFP as an indicator of institutional factors, e.g., laws and tribunals, institutions, property rights, patent protection, trade barriers, regulatory practices in the goods and the input markets, etc. In the three developed economies considered, in fact, the space for growth from an improvement of TFP seems barely null, while in South Korea those institutional factors may be still playing a crucial role on economic growth. Moreover, TFP contribution is negative for Australia, Japan and the U.S. under the Hulten approach. In this view, our fourth result indicates that the main contribution to output growth of technology in Australia, Japan, and U.S. arrives from ISTC being, respectively, 27%, 34%, and 37% using Hulten’s model. It is worth noting that these numbers exceed the overall contribution of technology. This is due to the negative contribution of TFP that partially offset that of ISTC and HCI in the
Hul.
Sol. 1.68 1.09 4.93 – –
2.92 Jorg. 1.68 1.09 3.00 0.71 1.22 21.8 14.1 35.9% 39.0 9.3 15.8 64.1%
Hul. 1.68 1.09 3.01 0.70 1.22
Sol. 1.23 0.83 0.86 – –
21.8 14.1
42.2 28.4
39.1 9.1 15.8
29.4 – –
Jorg.
Hul.
1.23 0.83 0.17 0.49 0.19
1.23 0.83 0.43 1.09 0.19
42.2 28.4 70.6% 6.0 16.7 6.7 29.4%
42.2 28.4 14.8 37.5 6.7
accounting procedure. Although the figures change if we use Jorgenson’s model, the ranking among sources is fairly robust with respect to the approach, although in this case the contribution of ISTC to output growth is larger in South Korea (0.71 percentage points) than in the other three countries (around 0.50 percentage points). The differences in results depending on the approach is likely explained by Korean lower ratio of capital compensation and higher level of investment. In the literature we find several studies who focus on the measurement of the sources of output growth for Asian countries using traditional growth accounting methods. A number of studies (e.g., Young, 1995; Lee and Hong, 2012) concludes that the main source driving output growth in developing Asian countries in the last decades is capital accumulation, where TFP contribution to growth had been relatively limited. Nevertheless, other studies reach the opposite conclusion, where technological change is the main driver of growth in the Asian economies. For instance, Park (2012) studies TFP growth in 12 Asian economies, claiming that catch-up effect is the major source of TFP growth in past decades and suggesting that the growth accounting paradigm has shifted from input accumulation driven growth toward productivitybased growth for these economies. Additionally, he finds that human capital contribution to TFP growth has raised in Hong Kong, Korea, Singapore and Taiwan during the past decades. Our results for Korea are consistent with those obtained by the later. Our fifth and last result is about the contribution of Human Capital accumulation to growth. The contribution of human capital is around 1.2 percentage points for the Korean economy, around 0.4 percentage points in both Australia and Japan, and about 0.2 percentage points in the U.S. Again, we find a difference between South Korea and the other three countries which is consistent with the view of Korea as an economy in transition. Not surprisingly, such a pattern is consistent with the results of PISA survey on secondary education (see Hanushek and Woessmann, 2010, 2011a,b), and also consistent with empirical analysis of Serrano and Timmer (2002) on the supply of skills in South Korea. Given its primary role on growth, we provide a further investigation on the sources of ISTC. Using the asset categories in EU KLEMS classification, we disaggregate the contribution to output growth from implicit technical change in each single category of capital. Table 4 displays the results using both Jorgenson’s and Hulten’s approaches. Under the Hulten approach, ISTC contribution to output growth is 0.70 percentage points for South Korea, 0.79 for Japan, 0.96 for Australia, arriving to 1.09 percentage points for the U.S. economy. ISTC contribution from ICT is about a half, with the other half corresponding to non-ICT
B. Molinari et al. / Japan and the World Economy 28 (2013) 60–71
67
Table 4 ISTC contribution to output growth, 1980–2006. Total contribution (a + b)
Australia
ICT capital (a = i + ii + iii) Hardware (i) Software (ii) Communications (iii) Non-ICT capital (b = iv + v + vi) Transports (iv) Machinery (v) Other equipment (vi) Decomposition (%) ICT capital (i + ii + iii) Hardware (i) Software (ii) Communications (iii) Non-ICT capital (iv + v + vi) Transports (iv) Machinery (v) Other equipment (vi)
Japan
Korea
U.S.
Jor.
Hul.
Jor.
Hul.
Jor.
Hul.
Jor.
Hul.
0.58
0.96
0.46
0.79
0.71
0.70
0.49
1.09
0.30 0.20 0.04 0.06 0.28 0.12 0.13 0.02
0.50 0.34 0.06 0.11 0.45 0.12 0.22 0.05
0.26 0.17 0.02 0.06 0.20 0.06 0.10 0.03
0.43 0.28 0.04 0.11 0.36 0.09 0.21 0.06
0.28 0.12 0.05 0.11 0.43 0.15 0.29 0.00
0.31 0.18 0.03 0.09 0.39 0.12 0.27 0.00
0.30 0.14 0.05 0.10 0.19 0.06 0.12 0.01
0.62 0.33 0.10 0.20 0.47 0.14 0.32 0.02
52.2% 34.3 6.9 10.9 47.8% 20.4 23.4 4.1
52.6% 35.4 6.0 11.3 47.4% 19.1 23.5 4.8
56.7% 37.7 5.1 13.9 43.3% 13.0 23.0 7.3
54.6% 36.2 4.6 13.9 45.4% 11.0 26.5 7.9
39.2% 17.3 6.5 15.5 60.8% 20.8 40.0 0.0
44.4% 26.3 4.8 13.3 55.6% 17.0 38.6 0.0
60.6% 28.6 10.6 21.4 39.4% 12.6 25.1 1.7
56.9% 30.0 8.7 18.2 43.1% 12.6 28.8 1.7
equipment. With respect to ICT equipment, the larger contribution corresponds to hardware, representing this particular capital asset about 30% of total ISTC contribution. In the case of the non-ICT equipment, the larger contribution corresponds to machinery. We find important differences across countries in the case of the ICT equipment. In fact, ISTC corresponding to ICT in the U.S. doubles that in South Korea. By contrast, under Jorgenson’s approach the results changes substantially. For South Korea, ISTC contribution remains similar, as there is little difference between investment ratios and capital income shares for this country. However, for the other three countries ISTC contribution to output growth is lower, as capital income shares are higher than investment shares. In fact, now is South Korea the country with a higher contribution from ISTC to output growth (0.71 percentage points). For Australia this contribution is 0.58 percentage points, 0.49 for the U.S. and 0.46 for Japan. The same statistical growth accounting approaches can be applied to the decomposition of labor productivity growth. Results are shown in Table 5. The relative importance of technology on labor productivity growth is larger for the Korean economy (around 80% of productivity growth), but also important for the Australian economy, explaining about 63% of productivity growth, in spite that average productivity growth is very different between these two economy (1.84% in Australia versus 6.33% in South Korea). This comparison implies that technological progress is much more deeper in the Korean economy than in the Australian
economy, that is explained by the higher progress in both TFP and human capital in the former. For the U.S., technological change explains about half of productivity growth. Finally, in the case of Japan, technological progress only explain around 36% of productivity growth, where the rest 64% is explained by input accumulation. In summary, the statistical growth accounting approach reveals that the main factor explaining differences in output and productivity growth across the four selected countries is TFP or neutral technological change. Whereas in Australia, Japan, and the U.S., the contribution of TFP to output and productivity growth is negligible when we control by ISTC and human capital, in South Korea contribution from TFP change is very large (an average of around 3 percentage points by year). The growth decomposition exercise has been done at an industry level. We find a similar pattern by industry than at in the aggregate level for all the countries. This result confirms that the large contribution of TFP to growth in the Korean economy is a general phenomenon affecting most of the sectors of this economy and that gains in multifactor productivity may be due to changes in the intensity in the capital/ labor use. The alternative approach is the so-called equilibrium growth accounting decomposition. This growth decomposition method proposed by Greenwood et al. (1997) based on the estimation of the balance growth path for productivity using a general equilibrium model. Under this view, technological progress is the only long-run source of productivity growth. As pointed out by
Table 5 Contribution to labor productivity growth, 1980–2006. Labor productivity (gY/L)
Australia
Japan 2.55
1.84 Sol. Contribution Capital/hours TFP ISTC HCI Decomposition (%) Capital/hours TFP (i) ISTC (ii) HCI (iii) Technology (i + ii + iii)
0.67 1.17 – – 36.6 63.4 – –
Korea
Jorg. 0.67 0.17 0.58 0.42 36.6 9.4 31.4 22.6 63.4%
Hul. 0.67 0.20 0.96 0.42 36.6 11.1 51.9 22.6
Sol. 1.64 0.91 – – 64.3 35.7 – –
U.S.
6.33 Jorg. 1.64 0.04 0.46 0.41 64.3 1.8 17.8 16.1 35.7%
Hul. 1.64 0.29 0.79 0.41 64.3 11.3 30.9 16.1
Sol. 1.40 4.93 – – 22.1 77.9 – –
1.64 Jorg. 1.40 3.00 0.71 1.22 22.1 47.4 11.3 19.2 77.9%
Hul. 1.40 3.01 0.70 1.22 22.1 47.6 11.1 19.2
Sol. 0.78 0.86 – 47.8 52.2 – -
Jorg. 0.78 0.17 0.49 0.19 47.8 10.7 29.7 11.9 52.2%
Hul. 0.78 0.43 1.09 0.19 47.8 26.4 66.7 11.9
B. Molinari et al. / Japan and the World Economy 28 (2013) 60–71
68
Table 6 Contribution to labor productivity growth, 1980–2006. Labor productivity (gY/L)
Contribution (i + ii + iii) TFP (i) ISTC (ii) ICT Non-ICT HCI (iii) Decomposition (%) TFP (i) ISTC (ii) ICT Non-ICT HCI (iii)
Australia
Japan
Korea
U.S.
1.84
2.55
6.33
1.64
General equilibrium approach 0.19 0.60 3.96 1.39 1.26 0.84 0.72 0.67 0.38 0.67 0.59 0.46 0.64 0.69 1.53 10.3 75.5 39.1 36.4 34.8
23.4 49.5 26.3 23.2 27.1
62.6 13.3 6.0 7.3 24.2
0.26 1.60 0.90 0.70 0.30 15.9 97.6 54.9 42.7 18.3
Greenwood et al. (1997) this alternative approach fits better the determinants of productivity growth in the long-run, given that part of the observed growth in capital is the result of technological change. Rodrı´guez-Lo´pez and Torres (2012) use the same approach to study productivity growth in Germany, Japan and the U.S. but without considering the role of human capital and a consequence implicit technical change embedded in the labor input is included in the neutral technological change as the latter is calculated as a residual. The general equilibrium approach shows that the main factor explaining labor productivity growth in the long-run is ISTC, in the case of Australia, Japan and the U.S. For Korea the main contribution comes from TFP, followed by human capital. Additionally, we find that TFP contribution to productivity is positive for Korea and Japan, but negative for the U.S. and Australia. Similar results for the case of Japan are obtained by Braun and Shioji (2007). The larger contribution from ISTC to productivity growth corresponds to the U.S. (1.6 percentage points). Nevertheless, results for Australia and Japan are fairly similar (1.39 and 1.26 percentage points, respectively). A lower contribution from ISTC is obtained for the case of Korea (about 0.84 percentage points). We also find important differences for the contribution of human capital. In this case the lower contribution corresponds to the U.S. (only 0.3 percentage points), whereas the contribution in the case of the Korea economy is 1.53 percentage points. The main differences is found in the contribution from neutral technology change. We find negative contributions for the case of Australia and the U.S., whereas is positive for Japan and Korea. It is remarkable the case of Korea, where the contribution of the neutral change to productivity growth is close to 4 percentage points. In summary, we find that the relative importance of the different technological factors in explaining long-run productivity growth is similar across Australia, Japan and the U.S., but of different nature in the case of the Korean economy. Overall, we find that differences in labor productivity growth among selected countries are explained mainly by differences in the neutral technological change. This technological factor is the main factor explaining why average labor productivity growth during the period is larger in South Korea with respect to the other countries. Indeed, we observe that the average labor productivity for the Japanese economy (2.55%) is also larger than the ones of Australia and the U.S. (1.84% and 1.64%, respectively), which also can be attributed to the neutral technical change (Table 6). 5. Concluding remarks During the last twenty years, the identification of technological progress driving productivity growth has received a renewed
attention, in particular after the economic boom caused by the expansion of Information and Communication Technologies, i.e., the so-called ‘‘new economy’’. This paper identifies three of those sources – technological progress generated by enhanced human capital, neutral technological progress or TFP growth, and by innovations in productive capital – and assesses their relative importance in explaining output and labor productivity growth. This analysis is performed on a group of Pacific countries observed during the period 1980–2006, using two different approaches to quantify the contribution of each source to growth: a traditional growth accounting approach and a general equilibrium model consistent growth decomposition. Whereas the first approach is a good explanation of the importance of technological progress together with factors accumulation, the second approach better identifies the determinants of productivity growth in the long-run where only technological factors matter. The main conclusion of the paper is that the most important sources of growth in South Korea are different from the that of Australia, Japan and the U.S., which instead have similar composition of growth. Factors accumulation explains about 60–70% of output growth in Australia, Japan and the U.S., but only 36% in South Korea. Conversely, technological progress explains no more than 30–40% of growth in the first three countries, whereas it accounts for 64% of Korean growth. We show that the difference is determined by the importance of the neutral technological change. Whereas in the three first countries, TFP contribution to output growth is relatively low or even negative, in the case of South Korea neutral technological change plays a crucial role in explaining output growth. Similar results are obtained in terms of labor productivity growth. This high contribution to growth from TFP in the case of Korea can be explained by changes in the intensity in the capital/labor use. Our results also stress the importance of taking into account embedded technological progress to factor inputs, both labor and capital. Technical progress embedded within equipment, mainly ICT, has increases its role as a productivity contributor in the last decades which, together with human capital contribution can explain a large proportion of output and productivity growth. Acknowledgements We are grateful to A. Bongers, J. Ohnemus, M. Saam, J.M. O’Kean, J.J. Pe´rez, G. Ferna´ndez-de-Co´rdoba, participants in the VI Conference on Information Technologies, ZEW, July 2008, and one anonymous referee for their helpful comments and suggestions. The three authors acknowledge financial support from Research Grant Junta de Andalucı´a P07-SEJ-02479.
Appendix A. Data transformation For all countries, assets are classed into three categories: (i) Information and Communication Technologies (ICT) assets, (ii) non-ICT assets, and (iii) Structures. The ICT category includes three assets, namely, (i.1) Hardware, Office equipment and Peripherals, (i.2) Communication Equipment and (i.3) Software licenses. The non-ICT category include (ii.1) Transport equipment, (ii.2) Machinery and (ii.3) Other equipment. These, together with structures, sum up to seven assets and match the classification provided by the EU KLEMS database for the four countries under consideration, Australia, Japan, Korea, and the U.S. Given the information provided by the EU KLEMS database the sample period is 1980–2006.
B. Molinari et al. / Japan and the World Economy 28 (2013) 60–71
We define the investment specific technical change (henceforth, ISTC) of asset j, Q nj;t , as the ratio between the constant quality consumption, PC nt , and the relative price of quality adjusted investment qnj;t
Q nj;t ¼
PC nt ; qnj;t
(A.1)
where n is a country code. We will assume that all assets except structures have an ISTC.11 The EU KLEMS database only quality adjusts those assets related to the ICT by adapting the BEA-NIPA prices from the U.S. economy. For the rest of assets, EU KLEMS does not quality adjust the deflators. In this appendix, and following Schreyer (2002), we propose a method that quality adjust all assets except structures and non-residential assets. We first describe the properties of the quality adjusted series for equipment and software developed by Gordon (1990) and later extended by Cummins and Violante (2002) for the U.S. economy. Using these series, we build six series of quality adjusted deflators, weighted by the nominal investment shares given in the BEA. In the following sub-section, we describe how these series are adapted for the other selected countries, namely Australia, Japan, and Korea Schreyer (2002). In this case, nominal investment series from the EU KLEMS database serve to weight the harmonized quality adjusted deflators earlier obtained. We believe that, since this is a sufficiently disaggregated portfolio of capital assets, measurement distortions from the different composition in the portfolios are minimized. In a final sub-section of this appendix, we calibrate the depreciation rates.
A.1. Quality adjusted investment prices for the U.S.A. Unless otherwise note, in this section we will suppress the country code. For the U.S.A., series of investment have been quality adjusted using the 1947–2000 annual series provided by Cummins and Violante (2002), say qCV j;t , where the subindex refers to asset j. These series do not include quality adjust for structures and are disaggregated into 26 assets, that can be grouped into the previous six equipment categories of the EU KLEMS database. The following is a detailed schedule with the aggregation we have undertaken: 1. Hardware equipment include three sub categories, namely (a) computers and peripheral equipment, (b) instruments and photocopies and (c) office and accounting equipment. 2. Communication equipment. 3. Software licenses include three assets: (a) prepackaged software, (b) custom software and (c) own software. These series are available from 1960 to 2000. 4. Transport equipment include (a) trucks, buses, and truck trailers, (b) autos, (c) aircraft, (d) ships and boats, and (d) Railroad equipment. 5. Machinery and equipment include (a) fabricated metal products, (b) engines and turbines, (c) metalworking machinery, (d) special industry machinery, n.e.c., (e) general industrial, including materials handling, equipment, and (f) electrical transmission, distribution. 6. Other equipment include (a) furniture and fixtures, (b) tractors, (c) agricultural machinery, except tractors, (d) construction machinery, except tractors, (e) mining and oilfield machinery,
11 Gort et al. (1999) show that the growth rate of ISTC in structures is about 1% per year, accounting for a 15% of total output growth.
69
(f) service industry machinery, (g) electrical equipment, n.e.c. and (h) and Other equipment. In what follows, we will refer to items i = 1 . . .6 as categories i’s. The first step consists in aggregating the 26 assets into the 6 categories. For that purpose we use a To¨rnqvist price aggregate that weights growth rates of the price index of investment in asset j, qi,j,t, belonging to category i by their nominal shares X g ðqi;t Þ ¼ 0:5ðsi; j;t þ si; j;t1 Þg ðqi; j;t Þ; (A.2) j
! qi; j;t ; qi; j;t1
g ðqi; j;t Þ ¼ ln
where si,j,t is the nominal investment share of asset j in year t (table 5.3.5. Private Fixed Investment by Type and detailed Investment in Private Nonresidential Fixed Assets, from BEA). Note that P jsi,j,t = 1. The level of quality-adjusted price index for total investment is recovered recursively, qi;t ¼ qi;t1 exp½g ðqi;t Þ;
(A.3)
with i ¼ 1; 2; . . . 6: Following the EU KLEMS methodology, for each category i we select 1995 as the base year, qi,1995 = 1. A Price Index for Consumption, PCt, is constructed using a To¨rnqvist price index aggregate that weights growth rates of price indexes for non-durable consumption (food, clothing and shoes, and other goods) and services (household operations, transportation, medical care, recreation, and other services) by their nominal shares. Let PCi,t be the price index for non-durable consumption/ service good i in year t. Let sci;t be the corresponding nominal share of good i in period t. Thus, the growth rate of the price index for consumption is
g ðPC t Þ ¼
X
0:5ðsci;t þ sci;t1 Þg ðPC i;t Þ;
(A.4)
i
g ðPC i;t Þ ¼ ln
PC i;t : PC i;t1
The level of quality-adjusted price index for total investment is recovered recursively, PC t ¼ PC t1 exp½g ðPC t Þ;
(A.5)
where PC1995 = 1. From (A.3) and (A.5) we can measure the technical change specific to asset j = 1 . . .6, according to expression (A.1). Using the EU KLEMS nominal investment series for categories 1 through 6, we have aggregated the quality adjusted price series into ICT and non-ICT assets
g ðqict;t Þ ¼
3 X 0:5ðsict; j;t þ sict; j;t1 Þg ðqict; j;t Þ;
(A.6)
j¼1
g ðqict; j;t Þ ¼ ln
! qict; j;t ; qict; j;t1
with j = 1 (Hardware), 2 (Communication) and 3 (Software), where sict,j,t is the nominal investment share of asset j within the ICT
B. Molinari et al. / Japan and the World Economy 28 (2013) 60–71
70
chapter in year t. Note that we have
P3
j¼1 sict; j;t
¼ 1. For the non-ICT assets
Variable
6 X g ðqnict;t Þ ¼ 0:5ðsnict; j;t þ snict; j;t1 Þg ðqnict; j;t Þ;
(A.7)
j¼4
g ðqnict; j;t Þ ¼ ln
Table A OLS estimates of quality adjusted prices. Hardware
Communication
Software
lnðqCV hard;t Þ
lnðqCV com;t Þ
lnðqCV soft;t Þ
ICT assets Constant lnðqNIPA i;t Þ
! qnict; j;t ; qnict; j;t1
with j = 4 (Transport eq.), 5 (Machinery eq.) and 6 (Others). Analogously, snict,j,t is the nominal investment share of asset j P within the non-ICT chapter in year t, and 6j¼4 snict; j;t ¼ 1. Both in (A.6) and (A.7) the shares of nominal investment are now borrowed from the EU KLEMS database, but the basic source of these EU KLEMS series is the BEA. Therefore, for the U.S.A. economy we end up with seven series of assets, three ICT assets, three non-ICT assets plus structures. We use these quality adjusted prices to deflate the series of nominal investment. Nominal investment in structures are deflated using the Consumption price index for non durables and services. Finally, for those years uncovered by the Cummins–Violante database, from 2000 to 2006, we extend the series of prices according to the long run specification proposed by Cummins and Violante (2002) NIPA NIPA lnðqCV j;t Þ ¼ b j;0 þ b j;1 lnðq j;t Þ þ b j;2 lnðq j;t1 Þ þ b j;2 t
þ b j;3 Dlnðyt1 Þ þ e j;t ;
(A.8)
where qCV j;t is the Cummins–Violante quality adjusted price of asset j, qNIPA is the NIPA price of asset j, and D ln(yt1) is the lagged j;t growth rate of the U.S. GDP (table 1.1.3. of BEA Real Gross Domestic Product Quantity Indexes). We estimate the long run relation in (A.8) with OLS. These estimates are shown in Table A. All coefficients are statistically significant, excepts those associated to the growth rate of output, bj,3. For the non-ICT assets, the lagged values of the NIPA prices are not much significant. Using these estimates and the NIPA prices of the six assets, we extend the data base from 2001 to 2006. Table B reports the price based measure of the ISTC for the six categories of assets, according to (A.1). These results do not differ from those reported by Cummins and Violante (2002) (see their Table 2). The highest ISTC is viewed for the Hardware and Communication equipment. Transport equipment and machinery have also a non-negligible ISTC, mainly after the eighties. This is an important issue, as noted by Cummins and Violante (2002), since the use of NIPA price for growth accounting decomposition may be misleading when concluding that the upsurge in the US productivity growth after the mid nineties was almost solely due to the ICT assets. On the contrary, the use of quality adjusted prices evince that the ISTC has been embedded in every type of assets.
2.23 0.90
(4.47) (4.43)
2.48 5.63
lnðqNIPA i;t1 Þ
0.56
(3.20)
4.02
(4.43)
t 100 D ln(yt1) 2 R¯
6.34 1.13 0.99
(4.16) (1.52)
7.13 1.54 0.94
(11.87) (1.59)
Variable
R¯
0.56
(5.01)
1.22 0.12 0.98
(38.76) (0.79)
Machinery
Others
lnðqCV trans;t Þ
lnðqCV mach;t Þ
lnðqCV other;t Þ
lnðqNIPA i;t1 Þ
2
(34.20) (6.18)
0.44 0.68
Transport
Non-ICT assets Constant lnðqNIPA i;t Þ t 100 D ln(yt1)
(12.53) (6.84)
1.68 0.94
(19.30) (4.65)
0.37
(1.81)
3.63 0.13
(19.29) (0.45)
0.96
1.48 0.99
(14.59) (6.02)
0.19
(1.14)
0.11
(1.02)
(14.58) (0.95)
1.92 0.15
(16.39) (0.90)
3.19 0.25 0.98
0.88 1.22
(16.48) (10.81)
0.99
A.2. Quality adjusted investment prices for Australia, Japan, and Korea Next we have to deal with the question on how to construct a reliable measure of ISTC for Australia, Japan, and Korea. To this end, we follow the methodology proposed by Schreyer (2002) to adapt the ICT prices from the US economy to another set of countries for which National Accounts have not been quality adjusted. If both sets of equipment are viewed as perfectly tradable goods (not alike structures and residential constructions), the following assumption can be made: the relative price change of the asset under consideration should be the same across countries. This implies that the price of asset j in country n is given by qnj;t ¼
Ptn USA q ; PtUSA j;t
(A.9)
for j = 1, . . .6, where fPtn ; PtUSA g are the GDP deflators of country n and the U.S.A., respectively, and qUSA is the aggregate quality j;t adjusted price of category j (= 1, . . .6) given in expression (A.3) for the U.S. economy. Expression (A.9) is a version of the law of one price, where the relative deflators is proxying the exchange rate. Thus Schreyer’s method to construct harmonized deflators for both ICT assets and non-ICT assets. Yet this adjustment is the one used by the EU KLEMS database to adapt the ICT-NIPA prices to other countries (see O’Mahony et al., 2009). For an application of this methodology to the U.K. case, see Basu et al. (2003). Analogously to the U.S. case, we use the EU KLEMS nominal investment series for categories 1 through 6 to aggregate the quality adjusted price series into ICT assets and non-ICT assets.
Table B Investment specific technical change by asset, U.S.A.
Hardware Communication Software Transport eq. Machinery Others
1948–2006
1948–1960
1961–1970
1971–1980
1981–1990
1991–2000
2001–2006
18.3 9.3 4.1 3.7 2.5 1.8
– – – 3.7 1.5 1.7
15.9 5.4 3.8 4.6 2.8 1.8
22.8 7.9 4.4 2.1 1.5 0.4
15.6 9.0 4.9 3.3 2.2 2.0
22.1 13.8 4.1 4.6 3.6 2.5
14.3 13.2 2.6 4.1 4.4 3.0
B. Molinari et al. / Japan and the World Economy 28 (2013) 60–71
Therefore, this aggregation takes explicit account of the portfolio composition in each country. Finally, the ISTC of asset j in country n, Q nj;t , is calculated according to (A.1). Using this previous based-price measure of ISTC, we compute the following gross rate
hnj ¼
006 Q nj;t 1 2X ; 28 t¼1978 Q nj;t1
(A.10)
n ¼ Australia; Japan; Korea; U:S:A:
A.3. Physical versus economic depreciation A final item concerns the depreciation rate of the aggregated assets. As long as we measure capital in efficiency units, several arguments recommend the use of physical depreciation rather than economic depreciation rates (see Cummins and Violante, 2002 and references therein). The following adjustment for time varying depreciation rates is recommended d j;t ¼ 1 ð1 d j;t Þ
Q j;t1 ; Q j;t
(A.11)
for asset j, where dj,t is the economic rate and dj,t is the physical one. Note that when there is no quality improvement, Qj,t/Qj,t1 = 1, the physical rate coincides with the economic one. When Qj,t/Qj,t1 > 1, the economic rate exceeds the physical rate, the difference being due to obsolescence. For our calibration exercise, we take this average rate of ISTC, hj, and construct a constant physical depreciation rate in the following manner
d j ¼ 1 h j ð1 d j Þ:
(A.12)
This is the steady state version of expression (A.11). References Basu, S., Fernald, J.G., Oulton, N., Srinivasan, S., 2003. The case of the missing productivity growth, or does information technology explain why productivity accelerated in the United States but not in the United Kingdom? NBER Macroecon. Annu. 18, 9–63. Braun, A., Shioji, E., 2007. Investment specific technological change in Japan. Seoul J. Econ. 20 (1) 165–199. Colecchia, A., Schreyer, P., 2002. ICT investment and economic growth in the 1990’s: is the United States a unique case? A comparative study of nine OECD countries. Rev. Econ. Dyn. 5 (2) 408–442.
71
Cummins, J.G., Violante, G.L., 2002. Investment-specific technical change in the US (1947–2000): measurement and macroeconomic consequences. Rev. Econ. Dyn. 5, 243–284. DeJong, D.N., Ingram, B.F., 2001. The cyclical behavior of skill acquisition. Rev. Econ. Dyn. 4, 536–561. Denison, E., 1962. The Sources of Economic Growth in the United States and the Alternative Before Us, vol. 13. Committee for Economic Development, New York. Gordon, R.J., 1990. The Measurement of Durable Goods Prices. University of Chicago Press, Chicago, IL. Gort, M., Greenwood, J., Rupert, P., 1999. Measuring the rate of technological progress in structures. Rev. Econ. Dyn. 2, 207–230. Greenwood, J., Krusell, P., 2007. Growth accounting with investment-specific technological progress: a discussion of two approaches. J. Monet. Econ. 54, 1300–1310. Greenwood, J., Hercowitz, Z., Krusell, P., 1997. Long-run implication of investmentspecific technological change. Am. Econ. Rev. 87, 342–362. Hayashi, F., Prescott, E.C., 2002. The 1990 in Japan: a lost decade. Rev. Econ. Dyn. 5, 206–235. Hanushek, E.A., Woessmann, L., 2010. The High Cost of Low Educational Performance. The Long-run Economic Impact of Improving PISA outcomes. OECD, Paris. Hanushek, E.A., Woessmann, L., 2011a. The Economics of International Differences in Educational Achievement. Handbook of the Economics of Education. Elsevier, Amsterdam. Hanushek, E.A., Woessmann, L., 2011b. How much do educational outcomes matter in OECD countries? Econ. Policy 26 (67) 427–491. Hulten, C., 1992. Growth accounting when technical change is embodied in capital. Am. Econ. Rev. 82, 964–980. Jorgenson, D.W., 1966. The embodiment hypothesis. J. Polit. Econ. 74 (1) 1–17. Jorgenson, D.W., Stiroh, K.J., 2000. Raising the speed limit: U.S. economic growth in the information age. Brooking Pap. Econ. Act. 1, 125–211. Jorgenson, D.W., Gollop, F.M., Fraumeni, B.M., 1987. Productivity and US economic growth. Harvard University Press, Cambridge. Lee, J.W., Hong, K., 2012. Economic growth in Asia: determinants and prospects. Jpn. World Econ. 24, 101–113. OECD, 2001. Measuring capital. A manual on the measurement of capital stocks, consumption of fixed capital and capital services. OECD, Paris. O’Mahony, Mary, Marcel, P., Timmer, 2009. Output, input and productivity measures at the industry level: the EU KLEMS database. Econ. J. 119 (538) F374–F403. Oulton, N., 2007. Investment-specific technological change and growth accounting. J. Monet. Econ. 54, 1290–1299. Park, J., 2012. Total factor productivity growth for 12 Asian economies: the past and the future. Jpn. World Econ. 24, 114–127. Rodrı´guez-Lo´pez, J., Torres, J.L., 2012. Technological sources of productivity growth in Japan, the U.S. and Germany. Macroecon. Dyn. 16, 133–150. Schreyer, P., 2002. Computer price indices and international growth and productivity comparison. Rev. Income Wealth 48 (1) 15–31. Schultz, T., 1961. Investment in human capital. Am. Econ. Rev. 51 (1) 1–17. Serrano, L., Timmer, M., 2002. Is technical change directed by the supply of skills? The case of South Korea. Econ. Lett. 76, 289–293. Solow, R., 1956. A contribution to the theory of economic growth. Quart. J. Econ. 70, 65–94. Stiroh, K., 2002. Information technology and U.S. productivity revival: What do the industry data say? Am. Econ. Rev. 92 (5) 1559–1576. Timmer, M.P., Ypma, G., van Ark, B., 2003. IT in the European Union: Driving Productivity Divergence? GGDC Research Memorandum GD-67 (October 2003). University of Groningen. Timmer, M., van Ark, B., 2005. Does information and communication technology drive EU-US productivity growth differentials? Oxf. Econ. Pap. 57 (4) 693–716. Young, A., 1995. The tyranny of numbers: confronting the statistical realities of the East Asian growth experience. Quart. J. Econ. 110, 641–680.