The impact of high-tech capital on productivity: evidence from Australia.

Connolly, Ellis ; Fox, Kevin J.

I. INTRODUCTION

The bulk of the empirical research on the relationship between high-tech capital investment and productivity has been based on data for the United States. Given the relatively large size of the high-tech capital sector in the U.S. economy, it is likely that the United States is a special case. Most economies resemble Australia's in being net importers of high-tech capital. Not producing high-tech capital was often cited as a weakness in these economies relative to the United States. However, since the end of the late 1990s tech boom, there has been a reappraisal of whether the production or the use of high-tech capital is most beneficial. Hence, to shed light on this debate, the main purpose of this paper is to empirically examine the relationship between high-tech capital use and the multifactor productivity (MFP) of 10 industries in the Australian market sector.

Using annual data for 1966-2002, alternative specifications of production functions are considered and estimated. Alternative measures of high-tech capital are included, with combinations of computers, software, and electronics capital. In the first production function, high-tech capital is allowed to be more (or less) productive than other capital to determine whether there is a relationship between high-tech capital and MFP. It is also possible to test whether there are excess returns to high-tech capital. For robustness testing purposes, a second production function is specified with high-tech capital as a separate input into production. Other potential sources of MFP growth, such as public capital, research and development stocks, and microeconomic reform, are also included in the production functions.

We find evidence of a positive relationship between high-tech capital and MFP in the market sector. There is less evidence, however, of excess returns to high-tech capital, which suggests that current investment levels are not inadequate. The results by industry suggest that the benefits of investing in high-tech capital are not distributed evenly across the economy, with only some industries experiencing a positive relationship between high-tech capital and MFP.

The structure of this article is as follows. In section II, the theoretical contribution of high-tech capital to MFP and the evidence to date is reviewed. In section III, the methodology is outlined. In section IV, data sources and trends are described. In section V the results are presented and interpreted, and section VI concludes.

II. THE IMPACT OF HIGH-TECH CAPITAL ON OUTPUT AND PRODUCTIVITY: THE THEORY AND EVIDENCE

The impact of investment in high-tech capital has been the subject of debate between new economy sceptics and optimists for several years. It is widely accepted that MFP has increased substantially in U.S. high-tech producing industries, in which the rate of technological change has been phenomenal over the past 40 years. Several publications have found a quantifiable contribution by computer-producing industries to U.S. output growth. (1) However, Australia is not generally regarded as a producer of computers. Data from the Organisation for Economic Co-operation and Development (OECD 1998) indicate that the high-tech share of Australian manufacturing value added was 4.5% in 1995, compared to 15.8% in the United States. Therefore, the substantial U.S. economy-wide MFP improvements specifically associated with its computer-producing industry are unlikely to be replicated in the Australian economy.

Over the past 30 years, there has been rapid accumulation of high-tech capital in many countries. Australia was the third most intensive user of high-tech capital in the OECD in 1995 behind the United States and New Zealand. Simon and Wardrop (2002) find that this rapid accumulation of high-tech capital may have substantially contributed to Australia's economic growth during the 1990s. This accumulation has been driven by technological change in high-tech producing industries lowering the price of high-tech capital. Stiroh (1998) and Gordon (2000) argue that there is no evidence of nontraditional effects from this capital accumulation that would show up in MFP and therefore conclude that the impact of computerization on productivity in the computer-using sectors of the economy should be small.

"New economy" optimists such as Brynjolfsson and Hitt (1998) and Oliner and Sichel (2000) emphasise the ways in which the use of high-tech capital could improve aggregate MFP. First, investment in high-tech capital is complementary with new productivity-enhancing strategies and business processes such as the shift from mass production to mass customization. Second, high-tech capital can be used to provide new services such as automatic teller machines, electronic funds transfer point of sale, Internet banking and shopping, and e-procurement, which may provide more consumer satisfaction than the traditional labor intensive alternatives. Third, high-tech capital improves access to information that is essential for the efficient operation of markets and allocation of resources. Finally, the use of high-tech capital can increase the speed of innovation by making it possible to process the data required to develop new products faster and more cheaply.

Econometric studies in the United States have had mixed success in finding a relationship between high-tech capital and MFP. Berndt and Morrison (1995) conducted a pioneering "exploratory analysis" of the impact of high-tech capital on the productivity of two-digit classification U.S. manufacturing industries, and found little evidence of a relationship. Amato and Amato (2000) used data on U.S. manufacturing industries disaggregated to the four-digit level, and also produced inconclusive results, suggesting that the "computer productivity paradox" is still alive and well. However, Lehr and Lichtenberg (1999) and Brynjolfsson and Hitt (2003) used firm-level data to find a significant positive impact on MFP from high-tech capital use.

Some recent studies comparing the European experience to that of the United States have revealed considerable differences (e.g. Oulton 2002, Salvatore 2003). In a concise summary of the evidence, Daveri (2002) notes that it "looks as though the celebrated 'Solow paradox' on the lack of correlation between Information and Communication Technologies (ICT) investment and productivity growth has fled the USA and come to Europe."

Few studies have directly examined the relationship between high-tech capital use and Australian MFP. (2) Madden and Savage (1998) found that "investment in telecommunications and information technology," proxied by main telephone lines per capita, is a significant short-run source of labour productivity growth for the Australian economy from 1950 to 1994. However this is not necessarily evidence of high-tech capital spillovers, because unlike MFP, labor productivity can increase due to capital accumulation. Colecchia and Schreyer (2002) use the methodology of Oliner and Sichel (2000) and Jorgenson and Stiroh (2000) applied to a data set of nine OECD countries, including Australia. They found that although Australia has a very small ICT producing sector, it has benefited markedly from ICT capital services. Supporting evidence was found by OECD (2003), National Office for the Information Economy (2004), and by Bassanini and Scarpetta (2002). Gretton et al. (2002) used firm-level data to find "positive and significant links between ICT use and productivity growth in manufacturing and a range of service industry sectors." However, these studies do not inform us of industry differences in the impacts of high-tech capital use, nor on the possible existence of excess returns to high-tech capital investments. That is, high-tech capital typically has a higher user cost than other types of capital, which means that it must have a higher marginal product to make it a worthwhile investment, not simply contribute positively to growth (Jorgenson and Stiroh 1999; Lehr and Lichtenberg 1999; Stiroh, 2002a). None of the studies have examined this issue empirically for Australia.

III. METHODOLOGY

To measure the impact of high-tech capital on MFP, Cobb-Douglas production functions are estimated. In the first function, high-tech capital is specified as a productivity enhanced form of capital, so we can assess whether it has made a significant contribution to MFP. In the second function, high-tech capital is specified as a separate input, and the results between the different specifications are compared for robustness.

We begin with an aggregate production function specified as a function of technology and factor inputs:

(1) Y= A(t)f(K,L),

where Y is a measure of real output, K is the private capital stock, L is a measure of labour input and A(t) represents disembodied technological change. This production function is specified to have a Cobb-Douglas functional form rather than a more flexible functional form, such as the CES or the translog, because previous studies in Australia and elsewhere have found that the CES or translog functional forms produce results virtually identical to those using a Cobb-Douglas functional form.

To explore whether investment in high-tech inputs is influencing MFP, a specification similar to Lehr and Lichtenberg (1999) and Schreyer (2000) is used, where the capital stock (K) is decomposed into high-tech capital ([K.sub.H]) and nonhigh-tech capital ([K.sub.N]). The output elasticity of capital ([alpha]) is specified with respect to the "effective" capital stock [[K.sub.N] 4(1 + [theta])[K.sub.[theta]] where [theta] is a parameter measuring the extent to which a unit of [K.sub.H] is more (or less) productive than a unit of [K.sub.N]. Because high-tech capital typically has a higher user cost, it must have a higher marginal product to make it a worthwhile investment; see discussion to follow. The coefficient [beta] is the output elasticity of labor:

(2) Y = A(t) [[K.sub.N] + [(1 + [theta])[K.sub.H]].sup.[alpha] [L.sup.[beta]],

To reduce the possibility that high-tech capital is positively correlated with an unobserved input that is more directly responsible for increased MFP, other inputs to production which have been found to be significant in other studies are included as regressors. Romer (1986) and Lucas (1988) argue that education can improve the ability of the labor force to adapt to new technology, resulting in higher productivity. To account for this, labor is decomposed into skilled ([L.sub.H]) and unskilled labor ([L.sub.N]), with the productivity enhancing effect of human capital being measured by the coefficient [pi] in equations (3)-(5). Industry Commission (1995) indicates that the research and development (R&D) stock is a significant variable in the Australian production function, whereas Otto and Voss (1994) find that the stock of public capital has a significant and positive impact on private sector productivity.

Griliches (1997) argues that environment variables that are not considered as standard inputs can also be included in the production function to reduce misspecification and omitted variables bias. The environment variables used in this study have all been used previously in studies of Australian output or productivity, and where possible, they have been customized for the relevant industry. To control for the role of microeconomic reform in Australia's improved productivity performance in the late 1990s, international competitiveness and openness are included. International competitiveness is measured using the terms of trade, which Otto (1999) finds has strong "predictive content" for Australian MFP. Openness is measured as Australia's international trade, which according to Grossman and Helpman (1991), can also act as a carrier for international knowledge spillovers through foreign R&D. Other environment variables are energy prices, to capture the impact of the three oil price shocks on output; the weather as an influence on agricultural output; and a business cycle variable to account for the procyclical nature of productivity. A time trend variable is also included to control for the effect of financial deregulation on MFP in finance and insurance since 1985. Further data details are given in section IV and Appendix A.

Therefore, the production function is augmented to include the R&D stock of a particular industry (R), (3) public capital (G), and several environment variables, [Z.sub.j] where j = 1, 2, ..., n. Taking the natural logarithm, the parameters [alpha], [beta], [psi], [gamma], and [[omega].sub.j] are the output elasticities of the respective inputs:

(3) lnY = lnA + [alpha]1n(K + [theta][K.sub.H])

+ [beta]ln(L + [pi][L.sub.H]) + [gamma]lnR

+ [psi]lnG + [n.summation over j = 1] [[omega].sub.j]ln[Z.sub.j] + [kappa]t,

and by using the approximations ln(1 + [theta][K.sub.H]/K) [congruent to] [theta][K.sub.H]/K and ln(1 + [pi][L.sub.H]/L) [congruent to] [pi][L.sub.H]/L when [theta][K.sub.H]/K and [pi][L.sub.H]/L are small, (3) becomes an equation that can be estimated using ordinary least squares (OLS) if [alpha][theta] is treated as a single coefficient:

(4) ln Y [congruent to] lnA + [alpha]lnK + [alpha][theta][K.sub.H]/K + [beta]lnL + [beta][pi][L.sub.H]/L + [gamma]lnR + [pi]lnG + [n.summation over j = 1] [[omega].sub.j]ln[Z.sub.j] + [kappa]t.

Equation (4) can then be manipulated to produce an MFP specification. We assume constant returns to scale ([alpha] = 1 - [beta]), and subtract [alpha]lnK + [beta]lnL from both sides, consistent with Solow (1957). (4) MFP is then calculated using a standard growth accounting framework. In this framework, the factor share of income of labor is used as a proxy for the output elasticity of labour ([S.sub.L] = [beta]). Capital is then assumed to be the residual claimant on income ([S.sub.K] = 1 - [S.sub.L] = 1 - [beta] = [alpha]): (5)

(5) lnMFP [congruent to] lnA + [S.sub.K][theta][K.sub.H] / K + [S.sub.L] [pi][L.sub.H] / L + [gamma]lnR + [psi]1nG + [n.summation over j = 1] [[omega].sub.j]ln[Z.sub.j] + [kappa]t.

Therefore, we can see from equation (5) that in our productivity-enhanced inputs framework MFP is not only a function of disembodied technical change, but also R&D, government capital, environment variables, the high-tech share of capital, and the skilled share of labor.

An alternative functional form specifies high-tech capital and skilled labor as separate inputs. If increases in high-tech variables have

different returns-to-scale implications from the nonhigh-tech variables, and thus alter the shape of the production function, the production function could be specified as follows, taking the natural logarithm, where a, b, c, d, e, f, and g are parameters that measure the output elasticities of their respective explanatory variables:

(6) lnY = lnA + aln[K.sub.N] + bln[K.sub.H] + cln[L.sub.N] + dln[L.sub.H] + elnR + flnG + [n.summation over j = 1] [g.sub.j]ln[Z.sub.j] + [kappa]t.

This is similar to equation 14 of Stiroh (2002a, p. 53), except here we consider a longer list of explanatory variables, including a separation of labor into skilled and unskilled. This specification is more flexible than (5), which assumes an additive structure between [K.sub.N] and [K.sub.H]. However, the difficulty involved in obtaining precise estimates of the output elasticities of capital and labor in small sample regressions leads us to prefer the results to equation (5) over the results to equation (6). Nevertheless, estimating equation (6) is a useful robustness check, because if [S.sub.K][theta] is statistically significant in (5), b should also be significant in (6).

The Regression Equations

The models presented can be expressed as two linear regression equations which are estimated for industry i in period t, where [a.sub.x], [b.sub.x], and [c.sub.x] are unknown parameters where x = l, 2, ..., X: (6)

(7) lnMF[P.sub.it] = [b.sub.0] + [b.sub.1][K.sub.Hit]/[K.sub.it] + [b.sub.2][L.sub.Hit]/[L.sub.it] + [b.sub.3]1n[R.sub.it] + [b.sub.4]ln[G.sub.it] + [n.summation over.j = 1] [b.sub.5j]ln[Z.sub.itj] + [b.sub.6]t + [[epsilon].sub.1it],

which corresponds to equation (5), and

(8) ln[Y.sub.it] = [c.sub.0] + [c.sub.1]ln[K.sub.Nit] + [c.sub.2]ln[K.sub.Hit] + [c.sub.3]ln[L.sub.Hit] + (1 - [c.sub.1] - [c.sub.2] - [c.sub.3)ln[L.sub.Nit] + [c.sub.4]ln[R.sub.it] + [n.summation over j = 1][c.sub.5j]ln[Z.sub.itj] + [c.sub.6]t + [[epsilon].sub2it],

which corresponds to equation (6) with constant returns to scale imposed. To preserve degrees of freedom, we exclude environment variables which are individually insignificant at the 15% level.

The coefficient [b.sub.1] on [K.sub.H]/K indicates the sign of the relationship between MFP and the share of high-tech capital in the total capital stock. Because the relationship is log-linear, the slope and elasticity change for each value of [K.sub.H]/K and MFP, but they always have the same sign as the coefficient.

The coefficient [b.sub.1] can be used to derive an estimate of [theta], the extent to which high-tech capital is more productive than other capital, weighted by the output elasticity of K. From (7) it is possible to calculate [theta] using the formula [b.sub.1] [congruent to] [S.sub.K][theta] derived from equation (5), using capital's share of income as a proxy for the output elasticity of capital. Then if we assume capital's share of income is known with certainty, we can derive the standard error of [theta] from the standard error of [b.sub.1]. (7) We can then conduct hypothesis tests on whether there are excess returns to high-tech capital relative to other capital. Lehr and Lichtenberg (1999) did not derive such standard errors and so did not conduct the relevant hypothesis tests.

Regression equation (8), which is based on equation (6), estimates output with high-tech capital and human capital as separate inputs. From (8), we can obtain [c.sub.2], the output elasticity of [K.sub.H]. Although the magnitude of [c.sub.2] is not directly comparable with [b.sub.1], we would expect a positive coefficient on [b.sub.1] would coincide with a positive coefficient on [c.sub.2]. For consistency with equation (7), constant returns to scale are imposed in equation (8). (8) Public capital is excluded as a regressor in equation (8) because potential collinearity with private capital inhibits our ability to accurately estimate the output elasticity of capital.

Quantifying the Returns to High-Tech Capital

By differentiating (2) with respect to [K.sub.H], it is possible to calculate the marginal product of high-tech capital (MP[K.sub.H1]):

(9) MP[K.sub.1] = [alpha](1 + [theta])Y/[[K.sub.N] + (1 + [theta])[K.sub.H]],

which can then be compared to the marginal product of other capital (MP[K.sub.N1]. Lehr and Lichtenberg (1999) note that the ratio of the marginal products of high-tech capital and other capital (MP[K.sub.H1]/MP[K.sub.N1]) for the profit maximizing industry should equal the ratio of the user costs of high-tech capital and other capital ([R.sub.H]/[R.sub.N]):

(10) MP[K.sub.H1]/MP[K.sub.N1] = (1 + [theta]) = [R.sub.H]/[R.sub.N] = {[r + [d.sub.H] - E([p.sub.H])][P.sub.H]} /{[r + [d.sub.N] - E([p.sub.N])][P.sub.N]},

where r is the risk-adjusted discount rate, [d.sub.m] is the depreciation rate, [P.sub.m] is the purchase price of a unit of capital, and E([p.sub.m]) is the expected rate of price appreciation where m = H for high-tech capital and m = N for other capital. This relationship can be used to test whether the returns to high-tech capital in Australian industries equal the returns to other capital, as follows.

Using data from the Australian Bureau of Statistics (ABS) on computer capital (see section IV), the following values were calculated: [d.sub.H] = 0.20, [d.sub.N] = 0.05, which are the average depreciation rates from the ABS; E([p.sub.H]) = -0.15, E([p.sub.N]) = 0.6, where E([p.sub.H]) captures the trend in computer prices and is estimated by taking the long-run average price change; r = 0.04, as used by the ABS; and [P.sub.H]/[P.sub.K] is normalized to 1. The ratio of user costs in equation (10), [R.sub.H]/[R.sub.N], is then 13, implying [theta] = 12. Thus, a test of the null hypothesis [H.sub.0]: [theta] = 12 is a test of no excess returns to computers, as in Lehr and Lichtenberg (1999, p. 337). We consider different types of high-tech capital, so each may have a different ratio of user costs. For an aggregate of computers and software capital, the ratio turns out to be the same as for computers, so that a test of the null hypothesis [H.sub.0]: [theta] = 12 is also a test of no excess returns to computers and software. For an aggregate of electronics, computers, and software capital, [d.sub.H] = 0.15 and E([p.sub.H]) = -0.01, as electronics dominates this aggregate. In this case, a test of the null hypothesis [H.sub.0]: [theta] = 5 is a test of no excess returns to electronics, computers and software.

As noted by Lehr and Lichtenberg (1999, p. 357), a test for excess returns is much stronger than a test of whether high-tech capital is productive. If excess returns to capital are found, then this implies that profit-maximizing firms should be utilizing high-tech capital relatively more intensively.

There are two weaknesses inherent in this method of quantifying the returns to high-tech capital. First, [R.sub.H]/[R.sub.N] is based on the assumption of profit-maximizing behavior at the firm level. However, we are dealing with industries, which are not decision-making units. Therefore any economic theory that is applicable to the firm may not be applicable to the industry. Second, the additive framework for capital [K = [K.sub.N] + (1 + [theta])[K.sub.H]] implies that [R.sub.H]/[R.sub.N] (the ratio of marginal products) is constant over time if [theta] is constant, which is a strong assumption, especially considering the rapid technological progress embodied by high-tech capital. Nevertheless, this calculation can give us an approximate comparison of the returns to high-tech capital relative to other capital.

For equation (6), which specified high-tech capital as a separate input into production, we can also calculate the respective marginal products for high-tech and other capital, MP[K.sub.H2] and MP[K.sub.N2], using the formula:

(11) MP[K.sub.H2] = [differential]Y/[differential][K.sub.H] = bY/[K.sub.H].

It is then possible to compare the estimated values of these marginal products to those from using (9).

IV. DATA CONSTRUCTION, MEASUREMENT ISSUES, AND TRENDS

Data Construction

The ABS calculates annual capital (net capital stocks), labour (total hours worked) and output (value added) data for Australian industries under the Australian and New Zealand Standard Industrial Classification. (9) All yearly observations are from July 1 of the previous year until June 30 current year. Such data are available for agriculture, forestry, and fishing; mining; manufacturing; electricity gas and water supply; construction; wholesale and retail trade; and transport, storage, and communications from 1965/66-2001/02 and for accommodation, cafes, and restaurants; finance and insurance; and cultural and recreational services from 1974/75-2001/02. (10)

The ABS has recently calculated the net capital stocks of electronics, electrical machinery, and communications equipment (referred to as electronics), computer equipment and peripherals (referred to as computers), and software for each of the industries back to 1959/60. Because these three forms of capital could each be considered high-tech but are derived from different data sources by the ABS, three alternative measures of high-tech capital are used in this article to improve the robustness of the results: electronics, computer, and software capital stocks; computer and software capital stocks; and computer capital stocks. (11)

Other variables are constructed as follows. Similar to the approach of Otto and Voss (1994), the general government net capital stock is used as the measure for public capital, and is included in regressions for predominantly private sector industries. (12) R&D expenditure data is available from the ABS for 1976/ 77-2001/02 for mining, manufacturing, wholesale and retail trade, finance and insurance, and the domestic economy and is used to calculate R&D stocks for those industries and the domestic economy using the perpetual inventory method. (13) The dearth of industry-specific education data in Australia precludes the inclusion of human capital in regressions for each industry in the market sector, but at the aggregate level, a measure of the proportion of employed with further education from the ABS is used. The ANZ vacancies series is the measure of the business cycle (ANZ 2003; Foster 1996). Alternative measures of the terms of trade are used where most appropriate. (14) The quantity of imports and exports is the measure for openness and is used in the market sector regressions. The west Texas crude oil price is the measure of energy prices and is included in all regressions. Finally, the Southern Oscillation Index from the Bureau of Meteorology is included as a measure of the weather in the agriculture regressions to reduce weather-induced volatility.

Data Measurement Issues

Mismeasurement is often cited as a cause of the computer productivity paradox; see, for example, Griliches (1994), and Diewert and Fox (2001). Triplett (1999) and David (1999) both note the difficulty involved in measuring high-tech capital and the output of service industries, which tend to have the largest proportions of high-tech capital in their capital stocks. However, Triplett and David both conclude that mismeasurement biases are not sufficient to explain away the computer productivity paradox.

The problem of measuring the true quantity of computer capital when prices fall due to rapid technological change has encouraged Triplett and the U.S. Bureau of Economic Analysis (BEA) to construct hedonic price indices for computers. In hedonic regressions, the price of computers is regressed on the relevant quality characteristics. The prices are then adjusted to incorporate the effect of technology changes. The ABS uses the BEA's quality-adjusted computer price index in calculating the computer net capital stock for each industry. Unfortunately, hedonic indices are very expensive for statistical agencies to produce, and at this stage, the price indices for software and electronics are not constructed using hedonic methods. (15) Nordhaus (2001) even suggests that these hedonic price indexes do not capture all the price declines in computers, because they focus on inputs rather than output. Any measurement error resulting from insufficient quality adjustment of software and electronics capital could then appear as an MFP spillover.

The use of high-tech capital tends to be concentrated in industries where measurement problems are thought to be greatest. For example, construction, finance and insurance, and wholesale and retail trade could suffer from output mismeasurement. Triplett and Bosworth (2001) note that increased customer satisfaction due to the introduction of new products such as automated teller machines and Internet banking does not contribute to output in finance and insurance. In David (1999) it is argued that the economy is fundamentally changing from a mass production to a mass customisation orientation, involving more quality change and price differentiation than our statistical systems are designed to handle. If output is indeed becoming more difficult to measure over time, one would expect it to be harder to find a relationship between the recent rapid increase in high-tech capital and MFP.

Data Trends

The 10 industries in our sample, which sum to the market sector, accounted for around 55% of Australian gross domestic product in 2000/ 2001, the base year. The largest industry, manufacturing, represented 20% of the market sector, whereas the smallest, cultural and recreational services, was 3% of the market sector (Table 1). The average capital share of income for most industries ranged between 30% and 50%, whereas mining; electricity, gas and water; and agriculture were more capital intensive. The high-tech intensity (ratio of high-tech capital to total capital) of the market sector was 7% in 2000/2001, with finance and insurance and'cultural and recreational services the most high-tech intensive, and agriculture and mining were the least high-tech intensive. Some of these service industries, including finance and insurance and wholesale and retail trade, experienced an improvement in their MFP performance during the 1990s (Figure 1).

[FIGURE 1 OMITTED]

V. ECONOMETRIC ISSUES AND RESULTS

Several methods have been used to estimate equations (7) and (8), including OLS, seemingly unrelated regressions (SUR) and panel regressions. Panel data techniques are found to be inappropriate. (16) The OLS results are reported rather than the SUR results because the OLS results do not suffer from possible misspecification in one regression contaminating a whole system. However, the OLS and SUR results are similar.

Econometric Problems and Diagnostic Testing

Endogeneity is a potential problem in time-series regressions. If labor, measured by hours worked, is an endogenous regressor when output is the regressand, the results from equation (8) will be biased. A solution would be to use instrumental variables estimation with an instrument for labor. However there do not appear to be any good instruments available for labour over the required time period. (17) In any case, the results from estimating equation (7) will not suffer from this potential endogeneity, because labor has been incorporated into the regressand, MFP.

Equations (7) and (8) involve the use of time-series data in (log) levels, which could be nonstationary. With a larger sample size, it would be more feasible to check for unit roots and cointegration. However, given sample sizes that are no greater than 37, such testing is of little value. Therefore, the approach of Otto and Voss (1994) is followed. They augment their models with a linear time trend, which would be appropriate if the data are trend stationary and not detrimental if the data are in fact difference stationary. The Durbin-Watson test statistics provide informal evidence that suggests that the regressions are not spurious. Another alternative is to estimate equations (7) and (8) in differences, without a time trend. According to Box and Jenkins (1970), this approach has the advantage of being appropriate if the data are difference stationary. However this approach inhibits estimation of the long-run relationship between high-tech capital and productivity, because any common long-run stochastic trends in the data are removed by differencing. For the same reason, differencing may also accentuate problems with endogeneity. However, for completeness and comparison purposes, the results for equations (7) and (8) estimated in differences are presented in Appendix B.

Several other standard diagnostic tests are also reported: the p-values for White heteroskedasticity tests without cross-terms; the test statistic for the Durbin-Watson test for autocorrelation; and p-values for the Jarque-Bera test for normality. Generally, there is little evidence of heteroskedastic, autocorrelated, or nonnormal residuals.

Results for the Regression Equations

For each industry, the results using electronics, computers and software as the measure for high-tech capital are presented for equation (7) in Table 2. There is a positive coefficient significant at the 10% level on [K.sub.H]/K for the market sector, which suggests that in aggregate, high-tech capital is more productive than other capital in the measurable industries of the Australian economy. This finding is supported by the individual sectoral results, where there is a positive and significant relationship between the high-tech share of capital and MFP in wholesale and retail trade, construction, agriculture, finance and insurance and accommodation, cafes and restaurants at the 10% level. These results are robust to the use of the alternative measures of high-tech capital.

Significant coefficients on the high-tech share of capital in service industries are particularly noteworthy considering that output mismeasurement in these industries would be expected to reduce our ability to find a relationship with MFP. However, if the method of quality adjustment of high-tech capital used by the ABS underestimates quality change, part of the benefit of accumulating high-tech capital would be included in MFP. This could help explain the relationship. Alternatively, the coefficients may be capturing some of the productivity benefits of deregulation in these industries over the past 15 years, which has coincided with increased high-tech investment. This does not appear to be the case for finance and insurance--a trend dummy has been included from 1986/87 to capture the productivity-enhancing effect of financial deregulation, and the coefficient on high-tech capital is still significant. However, the results for wholesale and retail trade and accommodation, caf6s and restaurants may suffer from misspecification, as suggested by low Durbin-Watson statistics and evidence of heteroskedasticity. It is harder to develop variables to capture the effect of deregulation in these sectors, where the reforms were conducted at the state government level at different points in time.

There are, however, several industries that do not appear to be benefiting from the productivity enhancing effects of using high-tech capital. In manufacturing; transport, storage and communication; and cultural and recreational services, the coefficient on high-tech capital is insignificant. In electricity, gas and water and mining, there is actually evidence of a negative relationship between high-tech capital and productivity. However the result for mining is not robust to the use of the narrower measures of high-tech capital. Structural changes in these industries may be affecting our results. In particular, electricity, gas and water and transport, storage and communication underwent substantial restructuring in the 1990s, which may be inhibiting our efforts to identify the contribution of high-tech capital to MFP.

The robustness of these results can be tested by estimating equation (7) for each industry in differences rather than levels; see Appendix Table B1. The signs of the coefficients are consistent with Table 2 for most industries. However, the standard errors are much larger, which is unsurprising because these regressions are only estimating the short-run relationship between the high-tech share of capital and MFP.

Our alternative specification is equation (8), which specifies high-tech capital as a separate input into production. If there is a significant relationship between high-tech capital and MFP in equation (7), one would also expect high-tech capital to have a significant output elasticity, because MFP growth is one of the drivers of output growth. The results are presented in Table 3. The results for the market sector, finance and insurance, and agriculture are positive and significant at the 10% level, consistent with the results in Table 2. Also consistent are the results for manufacturing, cultural and recreational services, and transport, storage and communication, with small or insignificant contributions; and mining and electricity gas and water, with negative contributions. However, the results for wholesale and retail trade, construction, and accommodation, caf6s and restaurants are harder to reconcile. For these industries, we tend to prefer the results in equation (7), which impose more plausible output elasticities of capital.

Results for equation (8) in differences are presented in Appendix Table B2. Although the signs of the coefficients are generally consistent with those presented in Table 3, again the standard errors are much larger, and many of the output elasticities on other capital and labour are counterintuitive.

Overall, the results are broadly consistent with those from previous studies in the United States and suggest a stronger relationship between high-tech capital and MFP than has been previously estimated for Australian data. The positive significant coefficients on high-tech capital in Australian service industries are similar to the findings in overseas studies such as Brynjolfsson and Hitt (1998) and Lehr and Lichtenberg (1999). Previous studies into the determinants of Australian productivity, such as Kraemer and Dedrick (1990) and Madden and Savage (1998), did not have access to the new ABS capital stock data that have allowed us to obtain more robust estimates of the relationship between high-tech capital and productivity.

The results for nonservice industries such as mining and manufacturing are also consistent with the U.S. evidence, where Berndt and Morrison (1995), Amato and Amato (2000), and Stiroh (2002b) were unable to find a relationship between high-tech capital and manufacturing productivity in the United States.

The results for the other variables specified in the equations are broadly similar to those in previous studies. The coefficients on the human capital variables are insignificant in the market sector regressions, a result which is consistent with Gust and Marquez (2000) and Bassanini et al. (2000), suggesting that the failure to specify human capital in the industry-level regressions should not be a major source of bias. Also, public capital has a positive and significant productivity elasticity in the manufacturing and wholesale and retail trade industries in Table 2, consistent with the findings of Otto and Voss (1994).

The Marginal Product of High-Tech Capital and the Returns to High-Tech Capital

We also make an attempt to quantify the marginal product of high-tech capital and test whether there are excess returns to high-tech capital under the first-order conditions of profit maximisation. As noted in section III, estimates for 0, the extent to which high-tech capital is more productive than other capital, are calculated by dividing the coefficient on the high-tech share of capital, [b.sub.1], by [s.sub.K] the output elasticity of capital, proxied by capital's share of income. The calculations for each industry using the different measures of high-tech capital are reported in Table 4.

The dispersion of [theta]s is inversely related to the high-tech share of capital, suggesting that the estimates of [theta] are more accurate for industries with a higher share. Alternatively, there may be diminishing returns to investment in high-tech capital. This may help explain why industries such as agriculture and accommodation, cafes and restaurants, which have very small high-tech shares of capital, can have such large [theta] values.

The marginal product of high-tech capital in equation (7) is calculated at each point in time using the formula in (9) where [alpha] is proxied by [s.sub.K] and s0 is estimated by [b.sub.1]. We can also calculate the marginal product of high-tech capital using estimates of (8) and the formula in (11), where high-tech capital is specified as a separate input. MP[K.sub.H1] is plotted in Figure 2 using electronics, computers and software as the measure of high-tech capital for the market sector and the industries with the most robust results.

[FIGURE 2 OMITTED]

For the market sector, MPK[K.sub.H1] is downward sloping, falling from 132% in 1974/75 to 106% in 2001/2002. It is reasonable to expect MPK[K.sub.H1] to fall over time as [K.sub.H]/K rises, as diminishing returns to high-tech capital begin to set in. The marginal product of other capital, MPK[K.sub.N1], is around 15% and is much smaller than the corresponding MP[K.sub.H1] values. This is to be expected, because high-tech capital has a significantly shorter functioning life than other forms of capital. When we compare MP[K.sub.H1] and MP[K.sub.H1], we find broadly similar results, with MP[K.sub.H2] of 160 in 2001/ 2002. At the industry level, for those industries with a positive output elasticity of high-tech capital, the results display the same downward slope, but MP[K.sub.H2] tends to be somewhat higher than MP[K.sub.H1]. This is unsurprising, because the estimated output elasticities of capital for these industries are higher in equation (8) than those imposed in calculating MFP in equation (7).

The p-values for the hypothesis tests for each industry are presented in Table 3 (and the results from estimating equation (8) in differences are presented in Appendix Table B3). There is little evidence of excess returns to high-tech capital across our three measures of high-tech capital for the market sector and for most industries. Only in agriculture and accommodation, cafes and restaurants is there consistent evidence of excess returns across the three measures of high-tech capital. Both these industries have relatively low high-tech intensities, and the result suggests they could benefit from further investment in high-tech capital. However, there is also evidence of deficient returns to high-tech capital in manufacturing; electricity, gas and water; finance and insurance; and cultural and recreational services, suggesting that these industries may have overinvested. Consistent with this, some of these industries are relatively high-tech capital intensive. Overall, the results suggest that the benefits of further investment in high-tech capital are not spread evenly across the economy.

VI. CONCLUSION

This article has presented an analysis into the relationship between high-tech capital and Australian MFP. This is an important topic given the debate on whether the benefits of computerization lie in their production or their use. It is well accepted that the production of computers involves strong MFP growth. However, whether using computers increases MFP remains controversial. Because Australia is a relatively computerized society but, like most countries, a net importer of high-tech capital, it is particularly important to explore whether the use of high-tech capital is a source of increased MFP.

The contribution of high-tech capital use to the productivity of 10 market sector industries of the Australian economy was measured by estimating production functions with different specifications of capital embodied technical change. The robustness of the results was tested using different measures of high-tech capital and other inputs which may impact on MFP. The marginal product of high-tech capital was calculated and compared to the marginal product of other capital for the different specifications. Finally, it was possible to test whether there are excess or deficient returns to high-tech capital under the assumption of profit-maximising behaviour.

For the Australian market sector, there is some evidence of a relationship between high-tech capital use and MFP. At the industry level, the results indicate that the benefits of investment in high-tech capital are not spread evenly across the economy. The industries with evidence of a positive relationship between high-tech capital use and productivity are wholesale and retail trade; finance and insurance; accommodation, cafes and restaurants; and agriculture. However for electricity, gas and water, there is some evidence of a negative relationship. These results are somewhat surprising, with positive relationships being found in service industries where output measurement is generally thought to be most problematic.

There remains wide scope for the relationship between high-tech capital and productivity to be explored in future research. With additional observations and capital rental-price data, a more flexible approach could be adopted with, for example, translog cost or profit functions. The analysis has also been limited to the Australian market sector, which excludes key industries such as education, health, and government, where the use of computers has the potential to significantly influence productivity. Also, this article has not attempted to explicitly model the impact of the Internet and e-commerce because it is very difficult to explicitly measure its effect on productivity at this early stage. Nevertheless, much of the impact of the Internet should be captured by the high-tech variables used here, which include the hardware and software requirements of the Internet.

Because productivity is one of the main drivers of economic growth, these results may have implications for economic policy. If the strong productivity growth in Australia during the 1990s is partly due to contributions from investment in high-tech capital, then it is more likely that this improved productivity performance is structural rather than cyclical. However, whether further investment in high-tech capital should be encouraged across all sectors of the economy is an area for further research, with the returns to high-tech capital appearing to vary substantially between industries.

APPENDIX A: DATA SOURCES

Computers and computer peripherals net capital stocks: ABS cat. no. 5204.0.

Software net capital stocks: ABS cat. no. 5204.0.

Electronic and electrical machinery and communication equipment net capital stocks: ABS cat. mo. 5204.0.

Value added by industry: ABS cat., o. 5204.0 and 5211.0.

Labor quantity: total hours worked: ABS cat. no. 5204.0 for 1975 2002. 1966-74 from ABS cat no. 6204.0. Hours worked for Finance, Property and Business services spliced for Finance and Insurance 1975-83.

Net capital stocks by industry: ABS cat. no. 5204.0

The labor share of income = compensation of employees/total factor income: ABS cat. no. 5204.0 and Gretton and Fisher (1997) for Finance and Insurance, Cultural and Recreational and Accommodation labor shares.

R&D expenditure by industry and product field: ABS cat no. 8104.0 for 1977-2002. From 1976/77 to 1984/85, the ABS only ran a comprehensive Survey of Research and Experimental Development every second year, requiring the use of linear interpolation to fill in the gaps.

Employed (15 64) with post school qualifications: ABS cat. no. 6227.0 and 6504.0, with interpolation for 1975-76.

Public capital: ABS cat. no. 5204.0 general government net capital stock.

Terms of trade: ABS cat. no. 5302.0.

Energy price index: West Texas crude oil prices (Bloomberg).

ANZ job vacancies: Foster (1996) and ANZ (2001).

International trade: calculated as a Fisher index of exports plus imports. ABS Cat. No. 5302.

Weather: Southern Oscillation Index (Australian Bureau of Meteorology).

APPENDIX B: SUPPLEMENTARY TABLES

APPENDIX TABLE B1

Results for Equation (7), Estimated in Differences

 Transport,
 Storage and
 Manufacturing Mining Communication

(Electronics, computers -1.352 -29.937 -1.648
& software) / capital (0.153) (0.013) (0.319)
([K.sub.H]/K)
R&D
Business cycle
Energy prices
Terms of trade 0.272
 (0.030)
Deregulation
Weather
Constant 0.024 0.003 0.037
 (0.000) (0.778) (0.000)
Sample 1967-2002 1967-2002 1967-2002
[[bar.R].sup.2] -0.007 0.135 -0.016
Durbin-Watson 2.303 2.082 2.227
White heteroskedasticity (0.851) (0.416) (0.752)
Jarque-Bera normality (0.366) (0.083) (0.002)

 Electricity, Wholesale
 Gas and and Retail
 Water Trade Construction

(Electronics, computers -7.432 2.149 3.351
& software) / capital (0.000) (0.204) (0.238)
([K.sub.H]/K)
R&D
Business cycle 0.040 0.055
 (0.034) (0.084)
Energy prices
Terms of trade 0.187 0.284
 (0.030) (0.004)
Deregulation
Weather
Constant 0.037 0.005 -0.008
 (0.000) (0.444) (0.294)
Sample 1967-2002 1967-2002 1967-2002
[[bar.R].sup.2] 0.359 0.112 0.076
Durbin-Watson 1.874 1.400 2.345
White heteroskedasticity (0.883) (0.995) (0.001)
Jarque-Bera normality (0.570) (0.795) (0.001)

 Finance Accommodation,
 and Cafes and
 Agriculture Insurance Restaurants

(Electronics, computers -16.540 1.144 9.359
& software) / capital (0.501) (0.365) (0.015)
([K.sub.H]/K)
R&D -0.899
 (0.003)
Business cycle
Energy prices -0.071
 (0.007)
Terms of trade
Deregulation 0.036
 (0.001)
Weather 0.005
 (0.028)
Constant 0.026 -0.020 0.041
 (0.206) (0.000) (0.033)
Sample 1967-2002 1976-2002 1976-2002
[[bar.R].sup.2] 0.101 0.296 0.237
Durbin-Watson 3.004 1.967 2.144
White heteroskedasticity (0.091) (0.400) (0.789)
Jarque-Bera normality (0.392) (0.587) (0.774)

 Cultural and
 Recreational Market
 Services Sector

(Electronics, computers 1.104 1.355
& software) / capital (0.563) (0.401)
([K.sub.H]/K)
R&D -0.621
 (0.003)
Business cycle 0.048
 (0.018)
Energy prices
Terms of trade
Deregulation
Weather
Constant 0.022 0.010
 (0.085) (0.046)
Sample 1976-2002 1976-2002
[[bar.R].sup.2] 0.022 0.269
Durbin-Watson 2.770 2.209
White heteroskedasticity (0.742) (0.149)
Jarque-Bera normality (0.612) (0.540)

Notes: Numbers in brackets are p-values. The p-values on the
coefficients are calculated using heteroskedasticity and
autocorrelation robust Newey-West SE.

APPENDIX TABLE B2

Results for Equation (8), Estimated in Differences

 Transport,
 Storage and
 Manufacturing Mining Communication

Electronics, computers & -0.027 -0.388 0.114
software capital (0.43) (0.043) (0.219)
Other capital 0.393 1.190 0.791
 (0.009) (0.000) (0.000)
Labor 0.634 0.198 0.095
 (0.000) (0.080) (0.491)
Human capital
Business cycle
Energy prices
Terms of trade 0.239
 (0.057)
Weather
Openness
Deregulation
Constant 0.024 -0.004 0.019
 (0.000) (0.828) (0.030)
Sample 1967-2002 1967-2002 1967-2002
[[bar.R].sup.2] 0.626 0.070 0.061
Durbin-Watson 2.275 1.757 2.035
White heteroskedasticity (0.459) (0.701) (0.178)
Jarque-Berg normality (0.670) (0.265) (0.016)

 Electricity, Wholesale
 Gas and and Retail
 Water Trade Construction

Electronics, computers & -0.203 0.095 0.079
software capital (0.006) (0.483) (0.680)
Other capital 1.212 0.641 0.340
 (0.000) (0.000) (0.036)
Labor -0.010 0.264 0.581
 (0.906) (0.178) (0.000)
Human capital
Business cycle 0.037 0.046 0.053
 (0.001) (0.013) (0.081)
Energy prices 0.017
 (0.007)
Terms of trade 0.208
 (0.026)
Weather
Openness
Deregulation
Constant 0.023 0.000 -0.003
 (0.000) (0.989) (0.830)
Sample 1967-2002 1967-2002 1967-2002
[[bar.R].sup.2] 0.507 0.234 0.468
Durbin-Watson 1.342 1.355 2.329
White heteroskedasticity (0.942) (0.637) (0.000)
Jarque-Berg normality (0.852) (0.616) (0.004)

 Finance Accommodation,
 and Cafes and
 Agriculture Insurance Restaurants

Electronics, computers & -0.493 0.133 -0.072
software capital (0.468) (0.102) (0.648)
Other capital 1.920 0.068 0.635
 (0.022) (0.656) (0.047)
Labor -0.427 0.798 0.437
 (0.624) (0.000) (0.040)
Human capital
Business cycle 0.043
 (0.114)
Energy prices -0.063
 (0.009)
Terms of trade
Weather 0.004
 (0.131)
Openness
Deregulation 0.040
 (0.000)
Constant 0.021 -0.026 0.027
 (0.414) (0.015) (0.123)
Sample 1967-2002 1976-2002 1976-2002
[[bar.R].sup.2] 0.107 0.571 -0.167
Durbin-Watson 2.899 2.132 1.782
White heteroskedasticity (0.104) (0.727) (0.314)
Jarque-Berg normality (0.812) (0.735) (0.840)

 Cultural and
 Recreational Market
 Services Sector

Electronics, computers & 0.142 0.026
software capital (0.115) (0.817)
Other capital 0.632 0.399
 (0.003) (0.055)
Labor 0.226 0.219
 (0.203) (0.018)
Human capital 0.356
 (0.048)
Business cycle
Energy prices
Terms of trade
Weather
Openness 0.363
 (0.000)
Deregulation
Constant 0.018 -0.010
 (0.253) (0.332)
Sample 1976-2002 1976-2002
[[bar.R].sup.2] -1.209 0.633
Durbin-Watson 1.958 1.969
White heteroskedasticity (0.011) (0.185)
Jarque-Berg normality (0.504) (0.982)

Notes: Numbers in brackets are p-values. The p-values on the
coefficients are calculated using heteroskedasticity and
autocorrelation robust Newey-West SEs.

APPENDIX TABLE B3

[theta] from Equation (7), Estimated in Differences, and Testing for
Excess Returns

 Transport,
 Storage and
 Manufacturing Mining Communication

Electronics, computers & -3.864 -44.025 -3.833
software (0.002) (0.006) (0.026)
[H.sub.0]: [theta] = 5
Computers & software -2.272 -92.593 -10.720
[H.sub.0]: [theta] = 12 (0.066) (0.146) (0.011)
Computers -5.638 -200.01 -17.814
[H.sub.0]: [theta] = 12 (0.020) (0.010) (0.002)

 Electricity, Wholesale
 Gas and and Retail
 Water Trade Construction

Electronics, computers & -11.986 6.321 6.981
software (0.000) (0.788) (0.735)
[H.sub.0]: [theta] = 5
Computers & software -29.917 6.148 4.555
[H.sub.0]: [theta] = 12 (0.000) (0.820) (0.950)
Computers -33.566 10.661 23.198
[H.sub.0]: [theta] = 12 (0.000) (0.574) (0.119)

 Finance Accommodation,
 and Cafes and
 Agriculture Insurance Restaurants

Electronics, computers & -20.420 2.932 32.274
software (0.403) (0.521) (0.037)
[H.sub.0]: [theta] = 5
Computers & software 24.112 5.673 32.155
[H.sub.0]: [theta] = 12 (0.430) (0.814) (0.075)
Computers 44.396 16.011 43.035
[H.sub.0]: [theta] = 12 (0.309) (0.147) (0.040)

 Cultural and
 Recreational Market
 Services Sector

Electronics, computers & 2.629 3.011
software (0.602) (0.578)
[H.sub.0]: [theta] = 5
Computers & software 0.295 4.659
[H.sub.0]: [theta] = 12 (0.208) (0.940)
Computers -2.373 7.132
[H.sub.0]: [theta] = 12 (0.215) (0.791)

Notes: Numbers in brackets are p-values. The p-values on the
coefficients are calculated using heteroskedasticity and
autocorrelation robust Newey-West SEs.

REFERENCES

ANZ. "ANZ Job Advertisement Series." ANZ economic commentary Web site, 2003.

Amato, T., and C. Amato. "The Impact of High-Tech Production Techniques on Productivity and Profitability in Selected U.S. Manufacturing Industries." Review of Industrial Organisation, 16(4), 2000, 327-42.

Bassanini, A., and S. Scarpetta. "Growth, Technological Change, and ICT Diffusion: Recent Evidence from OECD Countries." Oxford Review of Economic Policy, 18(3), 2002, 324-44.

Bassanini, A., S. Scarpetta, and I. Visco. "Knowledge, Technology and Economic Growth: Recent Evidence from OECD countries." Paper presented at the 150th Anniversary Conference of the National bank of Belgium, How to Promote Economic Growth in the Euro Area, Brussels, May 11-12, 2000.

Bean, C. "The Australian Economic Miracle: A View from the North." Paper presented at the RBA conference The Australian Economy in the 1990s, HC Coombs Centre for Financial Studies Kirribilli, July 24-25, 2000.

Berndt, E. R., and C. J. Morrison. "High-Tech Capital Formation and Economic Performance in U.S. Manufacturing Industries, an Exploratory Analysis." Journal of Econometrics, 65, 1995, 9-43.

Box, G., and G. Jenkins. Time Series Analysis," Forecasting and Control. San Francisco: Holden Day, 1970.

Brynjolfsson, E., and L. M. Hitt. "Beyond the Productivity Paradox." Communications of the ACM, August 1998.

--. "Computing Productivity: Firm-Level Evidence." Review of Economics and Statistics, 85(4), 2003, 793-808.

Colecchia, A., and P. Schreyer. "ICT Investment and Economic Growth in the 1990s: Is the United States a Unique Case?" Review of Economic Dynamics and Control, 5, 2002, 408-42.

Connolly, E., and K. J Fox. "The Impact of High-Tech Capital on Productivity: Evidence from Australia." University of New South Wales School of Economics Discussion Paper 2003/4, 2003.

Daveri, F. "The New Economy in Europe, 1992 2001." Oxford Review of Economic Policy, 18(3), 2002, 345-62.

David, P. "Digital Technology and the Productivity Paradox: After Ten Years, What Has Been Learned?" Conference, Understanding the Digital Economy: Data, Tools and Research, U.S. Department of Commerce, Washington, DC, May 25-26, 1999.

Dewan, S., and K. Kraemer. "Information Technology and Productivity: Evidence from Country Level Data." Management Science, 46(4), 2000, 548-62.

Diewert, W. E., and K. J. Fox. "The Productivity Paradox and the Mismeasurement of Economic Activity," in Monetary Policy in a Worm of Knowledge-Based Growth, Quality Change, and Uncertain Measurement, edited by K. Okina. London: Macmillan, 2001.

Foster, R. A. Australian Economic Statistics, 1949-50 to 1994-95. Reserve Bank of Australia, Occasional Paper no. 8, June 1996.

Fox, K. J. "Problems with (Dis)Aggregating Productivity Growth and Another Productivity Paradox." School of Economics, University of New South Wales Discussion Paper 2003101, 2003.

Gordon, R. J. "Does the New Economy Measure up to the Great Inventions of the Past?" Journal of Economic Perspectives, 14, 2000, 49-74.

Gretton, P. K., and B. Fisher. "Productivity Growth and Australian Manufacturing Industry." Industry Commission Staff Research Paper, AGPS, Canberra, 1997.

Gretton, P. K., J. Gali, and D. Parham. "Uptake and Impacts of ICTs in the Australian Economy: Evidence from Aggregate, Sectoral and Firm Levels." Paper prepared for the Workshop on ICT and Business Performance, OECD, Paris, December 9, 2002, Productivity Commission, Canberra, 2002.

Griliches, Z. "Productivity, R&D, and the Data Constraint." American Economic" Review, 84, 1994, 1-23.

--. "The Simon Kuznets Memorial Lectures." Draft, October 30, 1997.

Grossman G. M., and E. Helpman. Innovation and Growth in the Global Economy. Cambridge, MA: MIT Press, 1991.

Gust, C., and J. Marquez. "Productivity Developments Abroad. Federal Reserve Bulletin, October 2000, 665-81.

Industry Commission. Research and Development, Report no. 44, vol. 3, Canberra, AGPS, 1995.

Jorgenson, D. W., and K. J. Stiroh. "Information Technology and Growth." American Economic Review, Papers and Proceedings, 89(2), 1999, 109-15.

--. "Raising the Speed Limit: U.S. Economic Growth in the Information Age." Brookings Papers on Economic Activity, 1, 2000, 125-212.

Jorgenson, D. W., M. S. Ho, and K. J. Stiroh. "Lessons from the US Growth Resurgence." Journal of Policy Modelling, 25, 2003, 453-70.

Kraemer, K. L., and J. Dedrick. "Payoffs from Investment in Information Technology: Lessons from the Asia-Pacific Region." Worm Development, 22(12), 1990, 1921-29.

Lehr, B., and F. Lichtenberg. "Information Technology and its Impact on Firm-Level Productivity: Evidence from Government and Private Data Sources, 1977-1993." Canadian Journal of Economics, 32(2), 1999, 335-62.

Lucas, R. "On the Mechanics of Development Planning." Journal of Monetary Economics, 22, 1988, 342.

Madden, G., and S. J. Savage. "Sources of Australian Labour Productivity Change 1950 1994." Economic Record, 74(227), 1998, 362-72.

National Office for the Information Economy. "Productivity Growth in Australian Manufacturing." Occasional Economic Paper. National Office for the Information Economy, Commonwealth of Australia, Canberra, 2004.

Nordhaus, W. D. "The Progress of Computing." Yale University and the NBER, 2001.

OECD. Science, Technology and Industry Outlook 1998. Paris: Organisation for Economic Co-operation and Development, 1998.

--. ICT and Economic Growth: Evidence from OECD Countries, Industries and Firms. Paris: Organisation for Economic Co-operation and Development, 2003.

Oliner, S.D., and D.E. Sichel. "The Resurgence of Growth in the Late 1990s: Is Information Technology the Story?" Journal of Economic Perspectives, 14, 2000, 3-22.

--. "Information Technology and Productivity: Where Are We Now and Where Are We Going?" Journal of Policy Modeling, 25, 2003, 477-503.

Otto, G. "The Solow Residual for Australia: Technology Shocks or Factor Utilisation?" Economic Inquiry, 37(1), 1999, 136-53.

Otto, G., and G. M. Voss. "Public Capital and Private Sector Productivity." Economic Record, 70(209), 1994, 121-32.

Oulton, N. "ICT and Productivity Growth in United Kingdom." Oxford Review of Economic Policy, 18(3), 2002, 363-79.

Romer, P. "Increasing Returns and Long-Run Growth." Journal of Political Economy, 94(5), 1986, 1002-37.

Salvatore, D. "The New Economy and Growth in the G-7 Countries." Journal of Policy Modeling, 25, 2003, 531-40.

Schreyer, P. "The Contribution of Information and Communication Technology to output growth: a study of the G7 countries." OECD STI working Paper 2000/2, 2000.

Simon, J., and S. Wardrop. "Australian Use of Information Technology and its Contribution to Growth." Reserve Bank of Australia Research Discussion Paper 2002-02, 2002.

Stiroh, K. J. "Computers, Productivity and Input Substitution." Economic Inquiry, 36, 1998, 175-91.

--. "Are ICT Spillovers Driving the New Economy?" Review of Income and Wealth, 48(1), 2002a, 33-57.

--. "Information Technology and the U.S. Productivity Revival: What Does the Industry Data Say?" American Economic Review, 92(5), 2002b, 1559-76.

Triplett, J. E. "The Solow Productivity Paradox: What Do Computers Do to Productivity?" Canadian Journal of Economics, 32(2), 1999, 309-34.

(1.) See, for example, Oliner and Sichel (2000, 2003), Gordon (2000), and Jorgenson et al. (2003).

(2.) Kraemer and Dedrick (1990), Dewan and Kraemer (2000), and Bean (2000) are cross-country studies into the impact of high-tech capital use on productivity. They include Australia but do not report country-specific results.

(3.) If industry-specific R&D data are unavailable, we use the domestic R&D stock.

(4.) An alternative to using capital stock data to calculate MFP is to use capital services data. However, because we only have high-tech capital stock data available, we use capital stock data rather than capital services data for consistency. In any case, the signs and significance of the results in Tables 1 and 2 are generally robust to the use of capital services instead of capital stocks in calculating MFP.

(5.) The labor and capital shares of income have been allowed to vary across industries and across time to obtain a more accurate measure of MFP. Although this is desirable for estimation in (log) levels, expressing equation (5) in terms of first differences results in an MFP index that is a combination of productivity growth and share changes, and thus is not a pure measure of (the log of) productivity growth (Fox 2003).

(6.) In Connolly and Fox (2003), the working paper version of this article, results were also presented for a regression corresponding to equation (4). However, the estimated output elasticities of capital and labor were often counterintuitive. Therefore the results for the MFP specification in equation (7) are preferred.

(7.) In Connolly and Fox (2003), [theta] was calculated using the estimated output elasticity of capital, and the standard error of [theta] was calculated using a Taylor series expansion. The standard error of [theta] was much wider due to our imprecise estimates of the output elasticity of capital.

(8.) When constant returns to scale are not imposed, some of the estimated output elasticities are implausible.

(9.) Details on data sources and definitions are provided in Appendix A.

(10.) Data are also available for transportation and storage, communications, wholesale trade, and retail trade from 1974/75-2001102. However, to lengthen the time series back to 1965/66, these industries have been aggregated into transport, storage and communications and wholesale and retail trade.

(11.) These series are constructed using T6rnqvist quantity indices.

(12.) Predominantly private sector industries are: agriculture, mining, manufacturing, and wholesale and retail trade.

(13.) The results presented in section V use R&D stocks data with a depreciation rate of 5%. The use of alternative depreciation rates were not found to affect the results.

(14.) Terms of trade in goods is used for agriculture, mining, and manufacturing; services is used for other industries; and goods and services is used for the market sector.

(15.) The BEA now has official quality-adjusted deflators for prepackaged software, but the ABS is still working on producing a price index for computer software.

(16.) F tests on the restriction of equal slope coefficients on [K.sub.H]/K across industries are rejected at the 1% significance level for all three [K.sub.H] proxies. Therefore the OLS results should be preferred to the panel results because they do not impose equality restrictions on the [K.sub.H]/K coefficient across industries.

(17.) Hausman tests suggest that there are not significant endogeneity problems in our results when the lag of labor and the other variables in the regressions are used as instruments for labor in equation (8) without imposing constant returns to scale. Instrumenting for labor also does not appear to affect the significance of the coefficients on high-tech capital.

ABBREVIATIONS

ABS: Australian Bureau of Statistics

BEA: Bureau of Economic Analysis

ICT: Information and Communication Technologies

MFP: Multifactor Productivity

OECD: Organisation for Economic Co-operation and Development

OLS: Ordinary Least Squares

R&D: Research and Development

SUR: Seemingly Unrelated Regressions

ELLIS CONNOLLY and KEVIN J. FOX *

* We thank two anonymous referees for their very helpful and constructive comments. The article has benefited from comments by Erwin Diewert, Glenn Otto, participants at the Conference of Economists, Perth, and the Economic Measurement Group Workshop 2003. Any errors are the responsibility of the authors. The views expressed in this article are those of the authors and do not necessarily reflect those of the Reserve Bank of Australia.

Connolly: Senior Economist, Reserve Bank of Australia, GPO Box 3947, Sydney 2001, Australia. Phone 02-9551 8827, Fax 02-9551 8833, E-mail connollye@rba.gov.au

Fox. Associate Professor, School of Economics, University of New South Wales, Sydney 2052 Australia. Phone 61-2-9385-3320, Fax 61-2-9313-6337, E-mail k.fox@ unsw.edu.au

TABLE 1

Value Added and High-Tech Capital in the Australian Market Sector

 Transport,
 Storage and
 Manufacturing Mining Communication

Value added in 2001 73.4 34.0 51.6
(A$billion)
Average capital share of 35.0 67.9 42.9
income (%)
High-tech intensity in 2001 6.1 1.0 6.8
Electronics 2.1 0.7 5.0
Computers 2.0 0.1 1.0
Software 2.0 0.2 0.9

 Electricity, Wholesale
 Gas and and Retail
 Water Trade Construction

Value added in 2001 15.4 66.4 34.9
(A$billion)
Average capital share of 61.7 34.3 47.7
income (%)
High-tech intensity in 2001 7.0 10.7 8.1
Electronics 6.0 4.6 3.7
Computers 0.7 2.4 1.7
Software 0.3 3.7 2.8

 Accommoda-
 tion Cafes
 Finance and and
 Agriculture Insurance Restaurants

Value added in 2001 21.6 44.9 14.7
(A$billion)
Average capital share of 80.5 38.8 29.2
income (%)
High-tech intensity in 2001 3.6 12.4 5.5
Electronics 2.9 1.7 3.5
Computers 0.3 3.1 1.0
Software 0.4 7.6 1.0

 Cultural and
 Recreational Market
 Services Sector

Value added in 2001 11.8 368.6
(A$billion)
Average capital share of 42.4 45.1
income (%)
High-tech intensity in 2001 13.8 6.7
Electronics 10.0 3.8
Computers 2.1 1.2
Software 1.6 1.7

TABLE 2

Results for Equation (7)

 Transport,
 Storage and
 Manufacturing Mining Communication

(Electronics, computers & -0.233 -29.253 0.918
software) / capital
([K.sub.H]/K) (0.558) (0.054) (0.304)
Public capital 0.150
 (0.025)
R&D
Business cycle -0.079
 (0.112)
Energy prices 0.017 0.031
 -0.056 (0.005)
Terms of trade 0.568
 (0.000)
Openness
Deregulation
Weather
Time 0.016 0.006 0.032
 (0.000) (0.080) (0.000)
Constant 2.498 2.455 4.005
 (0.002) (0.000) (0.000)
Sample 1966-2002 1966-2002 1966-2002
[R.sup.-2] 0.996 0.673 0.996
Durbin-Watson 1.661 0.553 1.436
White heteroskedasticity (0.116) (0.007) (0.62)
Jarque-Berg normality (0.813) (0.925) (0.830)

 Electricity, Wholesale
 Gas and and Retail
 Water Trade Construction

(Electronics, computers & -9.423 8.362 7.219
software) / capital
([K.sub.H]/K) (0.000) 0.000 (0.001)
Public capital 0.710
 0.000
R&D
Business cycle 0.076 0.080
 (0.004) (0.029)
Energy prices -0.029 0.041 0.066
 (0.036) -0.027 (0.003)
Terms of trade 0.487 -0.213
 (0.000) (0.081)
Openness
Deregulation
Weather
Time 0.040 -0.032 -0.022
 (0.000) (0.000) (0.000)
Constant 1.855 -3.729 5.465
 (0.004) -0.068 (0.000)
Sample 1966-2002 1966-2002 1966-2002
[R.sup.-2] 0.991 0.822 0.601
Durbin-Watson 0.820 0.654 1.005
White heteroskedasticity (0.075) (0.014) (0.344)
Jarque-Berg normality (0.581) (0.989) (0.881)

 Accommodation,
 Finance and Cafes and
 Agriculture Insurance Restaurants

(Electronics, computers & 30.573 0.876 14.235
software) / capital
([K.sub.H]/K) (0.002) (0.002) (0.001)
Public capital
R&D -0.772
 (0.004)
Business cycle -0.138 0.053
 (0.054) (0.001)
Energy prices -0.055
 (0.083)
Terms of trade -0.446
 (0.045)
Openness
Deregulation 0.038
 (0.000)
Weather 0.006
 (0.013)
Time 0.007 -0.017 0.017
 (0.242) (0.000) (0.192)
Constant 4.314 4.694 13.133
 (0.000) (0.000) (0.000)
Sample 1966-2002 1975-2002 1975-2002
[R.sup.-2] 0.774 0.977 0.887
Durbin-Watson 1.947 2.019 1.061
White heteroskedasticity (0.336) (0.127) (0.059)
Jarque-Berg normality (0.069) (0.734) (0.797)

 Cultural and
 Recreational Market
 Services Sector

(Electronics, computers & 0.734 3.046
software) / capital
([K.sub.H]/K) -0.305 (0.000)
Public capital
R&D -0.485
 0.000
Business cycle -0.045
 -0.038
Energy prices 0.020
 (0.13)
Terms of trade -0.147
 (0.038)
Openness 0.397
 (0.000)
Deregulation
Weather
Time 0.013 -0.018
 (0.007) (0.000)
Constant 8.806 1.064
 (0.000) (0.112)
Sample 1975-2002 1975-2002
[R.sup.-2] 0.973 0.982
Durbin-Watson 2.177 1.791
White heteroskedasticity (0.079) (0.167)
Jarque-Berg normality (0.631) (0.872)

Notes: Numbers in brackets are p-values. The p-values on the
coefficients are calculated using heteroskedasticity and
autocorrelation robust Newey-West SEs.

TABLE 3

Results for Equation (8)

 Transport,
 Storage and
 Manufacturing Mining Communication

Electronics, computers & 0.033 -0.346 0.194
software capital (0.079) (0.073) (0.001)
Other capital 0.222 1.046 0.414
 (0.008) (0.000) (0.000)
Labor 0.745 0.300 0.392
 (0.000) (0.016) (0.000)
Human capital
R&D
Business cycle
Energy prices 0.027 0.023
 (0.000) (0.022)
Terms of trade 0.508
 (0.006)
Weather
Openness
Deregulation
Time 0.021 0.007 0.019
 (0.000) (0.433) (0.000)
Constant 3.974 -3.419 1.238
 (0.000) (0.016) (0.023)
Sample 1966-2002 1966-2002 1966-2002
[[bar.R].sup.2] 0.992 0.979 0.998
Durbin-Watson 1.287 0.545 1.489
White heteroskedasticity (0.352) (0.056) (0.746)
Jarque-Bera normality (0.928) (0.645) (0.744)

 Electricity, Wholesale
 Gas and and Retail
 Water Trade Construction

Electronics, computers & -0.281 -0.043 -0.188
software capital (0.000) (0.764) (0.034)
Other capital 1.248 -0.001 0.309
 (0.000) (0.996) (0.000)
Labor 0.033 1.044 0.879
 (0.535) (0.000) (0.000)
Human capital
R&D
Business cycle 0.061
 (0.013)
Energy prices 0.058
 (0.092)
Terms of trade 0.453
 (0.000)
Weather
Openness
Deregulation
Time 0.026 0.014 0.017
 (0.000) (0.209) (0.013)
Constant -5.825 5.866 4.107
 (0.000) (0.000) (0.000)
Sample 1966-2002 1966-2002 1966-2002
[[bar.R].sup.2] 0.997 0.956 0.975
Durbin-Watson 0.982 0.309 1.277
White heteroskedasticity (0.026) (0.062) (0.227)
Jarque-Bera normality (0.456) (0.583) (0.964)

 Accommodation,
 Finance and Cafes and
 Agriculture Insurance Restaurants

Electronics, computers & 0.453 0.113 -0.388
software capital (0.086) (0.000) (0.001)
Other capital -0.226 0.187 0.476
 (0.381) (0.011) (0.039)
Labor 0.773 0.700 0.912
 (0.000) (0.000) (0.000)
Human capital
R&D
Business cycle -0.115 0.034 0.060
 (0.091) (0.075) (0.016)
Energy prices -0.086
 (0.011)
Terms of trade
Weather 0.006
 (0.009)
Openness
Deregulation 0.045
 (0.000)
Time 0.017 -0.026 0.025
 (0.006) (0.000) (0.005)
Constant 5.005 4.680 2.421
 (0.001) (0.000) (0.054)
Sample 1966-2002 1975-2002 1975-2002
[[bar.R].sup.2] 0.860 0.998 0.983
Durbin-Watson 2.249 2.145 1.597
White heteroskedasticity (0.261) (0.164) (0.508)
Jarque-Bera normality (0.105) (0.514) (0.486)

 Cultural and
 Recreational Market
 Services Sector

Electronics, computers & 0.074 0.249
software capital (0.101) (0.000)
Other capital 0.415 0.423
 (0.000) (0.001)
Labor 0.511 0.328
 (0.001) (0.000)
Human capital -0.041
 (0.713)
R&D -0.340
 (0.001)
Business cycle -0.051
 (0.009)
Energy prices
Terms of trade
Weather
Openness 0.293
 (0.004)
Deregulation
Time 0.006 -0.016
 (0.185) (0.064)
Constant 5.925 -8.149
 (0.000) (0.000)
Sample 1975-2002 1975-2002
[[bar.R].sup.2] 0.988 0.996
Durbin-Watson 1.963 1.695
White heteroskedasticity (0.088) (0.382)
Jarque-Bera normality (0.671) (0.631)

Notes: Numbers in brackets are p-values. The p-values on the
coefficients are calculated using heteroskedasticity and
autocorrelation robust Newey-West SEs.

TABLE 4

[theta] from Equation (7) and Testing for Excess Returns

 Transport,
 Storage and
 Manufacturing Mining Communication

Electronics, computers & -0.665 -43.019 2.134
software (0.000) (0.033) (0.170)
[H.sub.0]: [theta] = 5
Computers & software -0.322 -11.080 1.690
[H.sub.0]: [theta] = 12 (0.000) (0.500) (0.017)
Computers -1.022 -65.162 0.755
[H.sub.0]: [theta] = 12 (0.000) (0.266) (0.046)

 Electricity, Wholesale
 Gas and and Retail
 Water Trade Construction

Electronics, computers & -15.198 24.595 15.040
software (0.000) (0.000) (0.021)
[H.sub.0]: [theta] = 5
Computers & software -14.854 17.861 12.470
[H.sub.0]: [theta] = 12 (0.002) (0.117) (0.887)
Computers -17.700 33.653 28.545
[H.sub.0]: [theta] = 12 (0.002) (0.003) (0.012)

 Finance Accommodation,
 and Cafes and
 Agriculture Insurance Restaurants

Electronics, computers & 37.745 2.246 49.087
software (0.006) (0.000) (0.001)
[H.sub.0]: [theta] = 5
Computers & software 51.192 1.961 39.230
[H.sub.0]: [theta] = 12 (0.002) (0.000) (0.006)
Computers 88.449 6.114 61.438
[H.sub.0]: [theta] = 12 (0.001) (0.004) (0.002)

 Cultural and
 Recreational Market
 Services Sector

Electronics, computers & 1.747 6.768
software (0.063) (0.152)
[H.sub.0]: [theta] = 5
Computers & software -0.188 6.951
[H.sub.0]: [theta] = 12 (0.000) (0.000)
Computers -0.838 12.457
[H.sub.0]: [theta] = 12 (0.000) (0.824)

Notes: Numbers in brackets are p-values. The p-values on the
coefficients are calculated using heteroskedasticity and
autocorrelation robust Newey-West SEs.