Journal summary

On the contribution of information and communication technology to productivity growth in Australia

Md Shahiduzzaman1,2 • Allan Layton3 •

Khorshed Alam2,3

Received: 18 January 2015 / Accepted: 28 July 2015 / Published online: 11 August 2015

� Springer Science+Business Media New York 2015

Abstract This paper revisits the so-called ‘ICT-productivity paradox’ from a long-run perspective by using annual Australian data for 1965–2013. It provides

estimates of long-run and short-run elasticities of labour productivity with respect to

ICT capital deepening, and explores the nature of long-run causality among pro-

ductivity growth and ICT and non-ICT capital deepening. The estimates of long-run

elasticities are derived by employing both time-series and panel data econometric

techniques. The empirical results provide strong confirmatory evidence of the long-

run impact of ICT capital deepening on labour productivity in Australia.

Keywords Information and communication technology � Capital deepening � Productivity

JEL Classification O1 � O4 � O5

& Md Shahiduzzaman [email protected]

Allan Layton

[email protected]

Khorshed Alam

[email protected]

1 Australian Digital Futures Institute, University of Southern Queensland, Toowoomba, QLD,

Australia

2 Australian Centre for Sustainable Business and Development, University of Southern

Queensland, Toowoomba, QLD, Australia

3 University of Southern Queensland, Toowoomba, QLD, Australia

123

Econ Change Restruct (2015) 48:281–304

DOI 10.1007/s10644-015-9171-9

1 Introduction

The impact of information and communication technology (ICT) on productivity

has long been an ongoing debate in economic research. Economics Nobel Laureate,

Robert Solow, once quipped ‘you can see the computer age everywhere but in the

productivity statistics’ (Solow 1987). Solow’s aphorism, which became known as

the ‘productivity paradox’, was in response to apparently negligible measurable

effects of ICT investment on productivity in the United States (US) and elsewhere

(Draca et al. 2006; Triplett 1999), and led to intense research interest in this domain

in later years.

In many cases research results supported a strong contribution of ICT investment

to productivity and economic growth in the 1990s (e.g., Jorgenson and Stiroh 1999,

2000; Oliner and Sichel 2000; Parham et al. 2001). Oliner and Sichel (1994),

however, argued that, due to the very low share of computing equipment in total

capital stock, seeking to find a significant contribution of ICT to productivity was an

unrealistic expectation. Gordon (2000) noted that much of the productivity

resurgence in the 1990s in the US was aided by the favourable economic conditions

rather than the contribution of ICT. In a relatively recent study, Jorgenson et al.

(2008) found the role of ICT on economic performance to be less important in the

post-2000 period in the US suggesting that the long-run contribution of ICT capital

investment to productivity remains an ongoing research issue. This is especially so

in the context of the so-called Information Age, and in relation to even more recent

allusions to so-called ‘knowledge-based economies’.

There are various transmission channels through which greater ICT investment

can enhance economic growth and multifactor productivity (MFP). Whilst early

studies placed more emphasis on the innovation and productivity improvement in

the ICT-producing industries (Jorgenson and Stiroh 2000), the general consensus

has emerged over time on the widespread contribution of ICT across a wide range of

sectors of the economy (Stiroh 2002b). The relentlessly falling prices of ICT

equipment relative to other capital—as well as in relation to the price of labour—

provided strong incentives for the substitution of ICT equipment for other forms of

capital and labour services (Jorgenson 2001), thereby potentially contributing to

economic growth and MFP through the process of capital deepening and spill-over

effects in the economy (Jorgenson 2001; Oliner and Sichel 2002; Venturini 2007;

Vourvachaki 2009).

Thus far, the ICT-productivity paradox has been mainly investigated by applying

the growth accounting framework and predominantly using data for the 1980s and

1990s (See, Draca et al. 2006, for a review). However, the results from growth

accounting or decomposition methods ‘are suggestive, rather than conclusive, on the

nature and extent of the links between ICT and productivity growth’ (Gretton et al.

2004, p. 106). The growth accounting framework also fails to capture the dynamic

relationship among the variables, especially in the context of addressing the mutual

causation among the variables of interest. The important attribute of ICT as a

general purpose technology (GPT) means that its productive impacts are only fully

materialized in the long-run (Bresnahan and Trajtenberg 1995; Venturini 2009). In

282 Econ Change Restruct (2015) 48:281–304

123

addition, productivity gains of ICT could be further delayed due to several cycles of

complementary innovation and investment. A reasonably longer period of data

covering various cycles of economic activity may therefore be a prerequisite to

uncover the true relationship between ICT and productivity. 1

This study seeks to fill this gap in the literature by first using aggregate Australian

data for five decades (1965–2013) and by utilizing the time series econometric

techniques of cointegration and causality testing to determine whether there is

evidence of a long run causal relationship between ICT and productivity. 2

The

second part of the paper uses Australian industry level panel data for the period of

1986–2013 to broaden the analytical framework, reduce the possibility of

aggregation bias, and gauge the sensitivity of estimates and conclusions to the

two alternative estimation techniques as well as to the alternative use of aggregate

time series data versus industry level panel data. It uses a production function

framework where ICT capital is considered to be a separate factor of production

along with other inputs.

The case of Australia is particularly interesting for several reasons. Australia is

one of the few advanced countries that has apparently generated significant

productivity benefits from the use of ICT in the 1990s (e.g., Banks 2001; Colecchia

and Schreyer 2002; Parham 2005; Parham et al. 2001). From the mid-2000s,

however, the productivity performance in Australia has seemingly slowed down

significantly despite the substantial increases in both private and public ICT

investment. Presently, the Australian Government is pursuing the biggest telecom-

munications reform in the country’s history with a view to ultimately providing

high-speed broadband access to all Australian homes and businesses. The major

stated motivation for the initiative is to drive innovation and productivity in the

Australian economy. An adequate understanding of ICT-productivity interactions

from the broad historical perspective would therefore be value-adding to the process

of making better-informed and more cost-effective decisions about such large scale

and very costly ICT investment.

The organization of the paper is as follows. Section 2 provides a historical

overview of economic performance and ICT use in Australia. Section 3 gives a brief

review of the theory and evidence. Section 4 outlines the framework, data and

1 Accurately identifying productivity cycles is a daunting task. Estimation results have been found to be

quite sensitive even to the minor variations in the period selected (Parham et al. 2001). Moreover, time

series econometrics is highly data intensive and may not produce very useful results for any one particular

cycle. A longer sample is therefore preferable. Of course, the longer the sample, the greater the chance of

significant temporal structural breaks in parameter values and model specificity. 2

From an econometric point-of-view, the use of a larger sample is always preferable than using a shorter

sample as estimation is more efficient, provided any structural changes are properly allowed for in the

estimated model. For that reason, using data back to the 1960s allows one to cover various cycles of

economic activity. As reported in Parham (2004b), Australia experienced a relatively better performance

in output growth, labour and MFP—contributed to by strong growth in capital services and hours

worked—in the 1960s as compared to the 1970s and 1980s. Subsequently, labour productivity and MFP

boosted again in the 1990s, followed once again by an apparent slowdown in the 2000s. Modelling labour

productivity from the 1960s, therefore, allows a counterbalancing of these ups and downs and provides a

better understanding of the role of the various contributing factors to productivity from a longer term

perspective.

Econ Change Restruct (2015) 48:281–304 283

123

econometric methods, and Sect. 5 presents the empirical results. Conclusions,

possible limitations and some policy implications follow in the last section.

2 Australia’s measured productivity performance in the last five decades

Apart from a few short-lived downturns in the 1980s, measured MFP in the

Australian market sector 3

has grown at a positive rate during the last four decades

(Fig. 1). Productivity performance showed a major improvement in the second half

of 1990s following the 1990–1992 recession, which itself followed a deterioration in

productivity performance observed in the 1980s, catalysed by the business cycle

downturn early in that decade. Overall, productivity improvements remained

consistently positive from 1988 to 2004. This very long period of continuous

increase in productivity (17 years) was no doubt facilitated by the significant

microeconomic reforms implemented during the 1980s, 1990s and early 2000s

(Parham 2004b). These reforms included the dismantling of tariff protection,

floating of the Australian dollar (A$), deregulation of financial markets, the

introduction of a range of productivity enhancing competition and related policies,

significant changes to the wage setting process along with other productivity

enhancing labour market reforms, and the introduction of the Goods and Services

Tax. It would be expected that these would all have contributed to the more

productive combination of labour and capital (both ICT and non-ICT) inputs in

production processes. From 2005, however, productivity growth has seemingly

deteriorated with a substantial reduction in its mean (2005–12 average is -0.37 %)

and variance. Output growth, however, remained relatively robust during the same

period (2005–12 average is 2.86 %).

In terms of an international perspective, the apparently disappointing picture of

Australia’s productivity performance in the 2000s is not shared by other comparable

advanced countries. For example, Table 1 compares the growth rates of MFP and

GDP per capita (a widely used measure of economic well-being) for Australia and

the US. Whilst the average growth of productivity was 1.0 % in both countries

during 1965–2011, Australia, on average, showed a relatively better productivity

performance than that of the US during 1973–2000. However, the trend has reversed

in the 2000s. Notwithstanding this, Australia has outperformed the US in the growth

of GDP per capita for over two decades (Table 1). The deterioration of measured

productivity performance as well as an apparent delinking between output and

productivity growth in recent periods has therefore emerged as a key concern for the

Australian economy and its policy makers.

Before proceeding, an important caveat is worth noting at this point. In the above,

the emphasis is necessarily on ‘‘measured’’ productivity and words like ‘‘seem-

ingly’’ and ‘‘apparent’’ were used quite deliberately. Actual productivity is what is

3 Market sector in Australia consists of 16 out of 19 industrial sectors. It includes all industries except for

Public Administration and Safety; Education and Training; Health Care and Social Assistance; and

Ownership of Dwellings.

284 Econ Change Restruct (2015) 48:281–304

123

important for enhancing our standard of living and sometimes measured produc-

tivity can give misleading signals of what is really happening as far as actual

productivity is concerned. For instance, to illustrate the point with an extremely

important example (for Australia), Gruen (2012) recently commented that measured

productivity in the mining industry was negatively impacted in the second half of

the 2000s. However, he notes that this was not because of any real decline in

productive efficiency in that industry, but simply because of the massive capital

spending by mining companies during that time in response to the huge upswing in

Australia’s terms of trade (resulting in more mining projects becoming econom-

ically viable due to higher world minerals’ prices), along with the necessarily

considerable lags between the capital spending and subsequent actual increases in

Fig. 1 Changes in measured MFP and output of the market sector, 1965–2012. Source: ABS (2012)

Table 1 Growth of measured MFP and GDP per capita: Australia and the US

Growth of MFP Growth of GDP per capita

Australia US Australia US

1965–1973 1.1 1.7 3.1 2.8

1973–1980 1.8 0.2 1.8 2.2

1980–1990 0.8 0.8 1.9 2.1

1990–1995 1.1 0.4 1.2 1.1

1995–2000 2.1 1.5 3.1 2.8

2000–2007 0.7 1.4 2.2 1.7

2007–2011 -0.7 0.4 1.5 -0.2

1965–2011 1.0 1.0 2.1 1.9

GDP per capita in 2011 US$ (converted to 2011 price levels with updated 2005 EKS PPPs). MFP

figures are for the Australian market sector and the US private business sector (excluding government

enterprises), respectively

Sources: the conference board total economy database, September 2011, http://www.conference-board.

org/data/economydatabase/; Bureau of Labour Statistics http://www.bls.gov/mfp/

Econ Change Restruct (2015) 48:281–304 285

123

mining production. Whilst the current analysis will necessarily be conducted using

derived measured productivity data, this is nonetheless an important issue to always

be kept in mind.

The rapid deterioration of measured MFP also led to a similar decline in

measured labour productivity growth in the 2000s, but to a lesser extent due to some

offsetting effects from capital deepening. Figure 2 shows the growth rate of labour

productivity, MFP and capital deepening over productivity cycles in Australia. As

shown in the figure the growth of labour productivity has declined significantly

(about 37 %) despite an increase (about 65 %) in the growth of capital deepening

during the 2004–2011 cycle from the preceding cycle (1999–2004). If the growth of

MFP in the 2004–2011 cycle had been the same as in the 1999–2004 cycle (1.1 %),

labour productivity would have doubled in the 2004–2011 cycle from the

1999–2004 cycle. The decline in measured labour productivity in the recent

productivity cycle is therefore attributed to the sharp deterioration of measured

MFP.

Figure 3 shows the average ICT to total capital ratio by industry in Australia for

the period of 1986–2013. This figure reflects and demonstrates the considerable

heterogeneity in ICT capital exposure across industries.

In terms of sectoral composition, Australia’s productivity performance over the

past two decades has been associated with an increase in the share of ICT

investment by ICT-using sectors relative to the ‘‘ICT-producing’’ sector (Gretton

et al. 2002a; Parham 2002). The share of the Information, Media and Telecom-

munication 4

sector’s ICT net capital stock (chain value measure) to market sector

ICT net capital stock in Australia fell from around 17 % in 1990 to 16 % in the

2000s (ABS 2014b). However, during this time, ICT net capital stock has

experienced quite a rapid growth in its share of total net capital stock—from 1.5 %

in the 1990s up to 3.5 % in the 2000s, for the market sector—implying a more than

doubling in the share over this period. Furthermore, the reduced share of ICT net

investment by the ‘‘ICT-producing’’ sector over this period implies this observed

rapid growth in the overall share of ICT investment in the market sector can be

disproportionately attributed to the ICT-using sectors, reflecting the potential for

significant spill-over productivity effects of ICT investment throughout the

economy.

3 ICT investment and productivity: Theory and empirics

What factors cause productivity growth to accelerate or regress? Technological

innovation has long been noted as a key factor. Early growth models (e.g., Solow

1956) posited technological progress as exogenous. On the other hand, endogenous

growth theorists postulate that changes in technological progress are a consequence

as well as a cause of changes in economic growth (e.g., Aghion and Howitt 1992;

4 The Information, Media and Telecommunication sector includes units engaged in (1) creating,

enhancing and storing information products in media, (2) transmitting information products using

analogue and digital signals, and (3) providing transmission services and/or operating the infrastructure to

enable the transmission and storage of information and information products.

286 Econ Change Restruct (2015) 48:281–304

123

Romer 1990). Either way, ICT as a GPT, could certainly be regarded, on an ‘a

priori’ basis, as a putative factor in leading to accelerated aggregate productivity.

This may happen through three channels. First, innovations in ICT raise MFP in

sectors (and countries) involved in ICT capital production (Colecchia and Schreyer

2002; Jorgenson 2004). Second, increasing capital deepening from investment in

complementary intangible capital in the ICT-using sectors and, third, disembodied

MFP gains from network spillovers and associated positive externalities from the

use of ICT (Acharya and Basu 2010). 5

Innovations in the ICT-producing sector and

the resulting fall in the costs of capital and intermediate goods provide incentives in

this process. In this mechanism, sectoral (ICT-producing, ICT-using, ICT-not using)

Fig. 2 Growth in measured labour productivity, MFP and capital deepening in the Australian market sector. Source: ABS (2012) and authors’ own estimation

Fig. 3 ICT to total capital ratio in Australia: 1986–2013 average

5 These may happen through other contextual factors. As for example, Christopoulos and McAdam

(2013) found that expenditures on ICT bolsters the positive impact openness on technical efficiency

among the OECD manufacturing sector.

Econ Change Restruct (2015) 48:281–304 287

123

output growth rates differ, but the economy as a whole continues on its steady-state

growth path (e.g., Venturini 2007; Vourvachaki 2009).

The proposition that ICT is a GPT poses several implications. ICT has the

potential for pervasive use in a wide range of industries and to show technological

dynamism through technology spillovers to downstream sectors (Draca et al. 2006).

ICT as a GPT may also lead to innovation by those adopting firms by creating new

applications opportunities; e.g., combining it with their own technology to improve

the quality of final products. The innovation activities and spillover effects of GPT

in turn make other sectors more productive (Bresnahan and Trajtenberg 1995).

However, there are instances where the spillover effects of such a GPT are not

detected as significant when subjected to empirical testing. Gordon (2000) found

that the productivity resurgence in the US was driven by the ICT-production and

durable manufactures sectors rather than by the other sectors of the economy.

Daveri and Silva (2004) also found no evidence of technology spillover from the

ICT sector to the rest of the economy. Similar results were reported by Edquist

(2005) for the Swedish economy.

Empirically, the ICT-productivity paradox has been examined extensively, but

predominantly using US data and by making use of short spans of data (see, Draca

et al. 2006; Kretschmer 2012, for a review). A number of influential studies have

made use of the growth accounting method, based on the work by Solow (1957) and

Jorgenson and Griliches (1967), to quantify the contribution of ICT to economic

growth and labour productivity (Colecchia and Schreyer 2002; Jorgenson and Stiroh

2000; Oliner and Sichel 1994, 2000; Van Ark and Inklaar 2005). These studies have

shown substantial variation in ICT elasticities across countries (Kretschmer 2012).

A major criticism of the growth accounting method is that it does not consider the

spillover effects and MFP growth outside the ICT production sector (Kretschmer

2012).

Other studies have used econometric methods, typically using industry level data

and adopting first difference models, fixed effects or instrumental variable

techniques (Stiroh 2002a). Using industry data for the US and United Kingdom

and employing a dynamic panel model, O’Mahony and Vecchi (2005) found

evidence of a positive and significant effect of ICT on output growth. Kretschmer

(2012) reports a clustering of estimated output elasticities of ICT around the value

of 0.05–0.06, with some notable highly positive or negative outliers. Stiroh (2002a)

summarized the results from 20 econometric studies and found the median estimate

to be 0.046, with a considerable variation with estimates ranging from -0.06 to

0.24.

Importantly, there is a limited number of studies in the Australian context and

these are predominantly based on the growth accounting method. 6

Parham et al.

(2001) analysed the productivity gains from ICT by using growth accounting

methodology. This study supported the existence of a strong contribution of ICT to

labour productivity growth in the 1990s (1.1 % points, which approximately

6 For brevity, we focus on Australian literature on ICT and productivity. Please see Shahiduzzaman and

Alam (2014a) for a review of ICT-economic growth nexus for Australia as well as OECD countries.

Draca et al. (2006), Kretschmer (2012) provides a comprehensive review of international literature on

ICT-productivity paradox.

288 Econ Change Restruct (2015) 48:281–304

123

accounted for one-third of labour productivity growth). However, to an extent, ICT

substituted for other forms of capital, resulting in a modest overall contribution of

total capital (Gretton et al. 2002b; Parham 2004a; Parham et al. 2001).

On the other hand, Quiggin (2001, 2006) has questioned the 1990s ‘productivity

surge’ in Australia, suggesting it as a ‘statistical illusion’. Furthermore, Dolman

(2009) has suggested that, even accepting measured estimates of the 1990s

productivity surge, ICT played only a moderate role, being more of the nature of an

enabler rather than the engine of growth. Using a simple cross-country regression

for the OECD, Bean (2000) found a statistically significant, but very small,

contribution of Australian ICT investment to productivity growth over the period

1992–97 (0.12 % points of the excess Australian MFP growth).

Gretton et al. (2004) carried out an analysis using firm-level data for Australia.

They found evidence of significant positive links between ICT use and productivity

growth in manufacturing and a range of service industries. However, productivity

effects associated with selected ICT investments were found to taper off over time

after the initial boost. The authors therefore concluded that the effect of ICT on

Australia’s productivity growth might have been a transitory one-off boost rather

than permanent in nature (Gretton et al. 2004).

In a recent paper Shahiduzzaman and Alam (2014a) found significant impact of

ICT capital deepening on labour productivity and technical progress in the 1990s,

but the impact seemed to reduce in the 2000s. Utilizing data for 1975–2011, and

employing the Toda and Yamamoto (1995; TY) approach, the study found evidence

of unidirectional long-run causality from ICT capital deepening to labour and MFP.

The study, however, did not estimate the short-run dynamics and elasticity

parameters for a longer time horizon.

The present study differs from previous studies in several ways. First, the study

uses annual time series data for 1965–2013 for the aggregate level analysis, which is

considerably longer than in earlier studies. As noted earlier, using data back to the

1960s allows one to cover a number of significant economic cycles, thus potentially

allowing an econometric teasing out and disentangling of possible causes of

observed variations in productivity from a longer term perspective. Second, apart

from examining the long-run causality for such a long-period using the TY

approach, the long-run elasticities are estimated by employing alternative methods,

i.e., the Autoregressive Distributed Lag (ARDL) and Johansen testing approaches.

Finally, industry level data for the period of 1986–2013 is used to compare results

between the aggregate time series and the industry panel analyses to ascertain the

degree of consistency thereto.

4 Methodology

4.1 Framework

To investigate the effects of ICT, non-ICT capital deepening and technical progress

(i.e., MFP) on productivity, the Cobb–Douglas production function framework is

utilized (Cobb and Douglas 1928). The Cobb–Douglas production function is a

Econ Change Restruct (2015) 48:281–304 289

123

particular and quite useful functional form that describes the relationship between

production input and the possible maximum output under a given technology. It is

based on the micro-foundation that the production processes are well described by a

linear homogenous function of the first degree with diminishing marginal returns to

each factor of production.

The production function begins with the following form:

Y ¼ FðK; L; tÞ ð1Þ

In (1) Y represents output and K and L represent capital and labour inputs (hours

worked) respectively. The variable t represents time factor. The conventional

growth accounting framework describes output growth in an economy as being

generated through a change in inputs and through the change of technology, pop-

ularly known as ‘technical change’. A change in inputs, such as capital and labour,

represents a movement along a given production function, while technical change—

which may involve actual new production technologies as well as improved

organisational and management practices—leads to shifts in the production function

(Jorgenson and Stiroh 1999; Solow 1957; Swan 1956). It is, therefore, important to

incorporate the shift parameter when estimating a production function using time

series data.

Assuming the production function (1) satisfies a certain economic neutrality

condition, i.e., the marginal rate of substitution between capital and labour remains

constant, (1) can be written as:

Y ¼ AðtÞf ðK; LÞ ð1:1Þ

In (1.1) A(t) describes the cumulative effects of technical change, i.e., the shifts of

the production function over time. Note that the K in above formulations includes

both ICT and non-ICT capital stock. If we consider that ICT and non-ICT capital are

separate inputs of production, the production function could be specified as follows,

taking natural logarithm (ln):

ln Y ¼ f þ u ln A þ a ln Ko þ b ln Kit þ c ln L þ et ð1:2Þ

where, Ko and Kit refer to non-ICT and ICT capital, respectively. The parameters a, b and c are the output elasticities of the relevant factor inputs, and importantly are constrained to sum to one and f and et are the constant and error terms, respectively. The shift factor A is popularly known as the Solow Residual and the mathematical

formulation to compute it at each point of time is described in Solow (1957).

Because A is computed as residual values, it captures any possible externalities

related to the factor inputs, including the ICT capital.

Given that aggregate production function satisfies the properties of constant

returns to scale (CRS), Eq. (1.2) can be expressed in terms of output per unit of

labour (y = Y/L):

ln y ¼ f þ u ln A þ a ln ko þ b ln kit þ et ð1:3Þ

In Eq. (1.3) ko = Ko/L and kit = Kit/L, where the variables ko and kit represent

capital deepening in terms of non-ICT and ICT capital, respectively.

290 Econ Change Restruct (2015) 48:281–304

123

Equation (1.3) can be used to estimate the role of ICT, non-ICT capital

deepening and MFP on labour productivity. 7

4.2 Data

Data used in the study are annual data. The variable Y represents gross value-added

production using chain volume measures (in AUD$ million), and L represents

labour hours. Ko and Kit denote the chain volume measure of non-ICT and ICT net

capital stocks, respectively and A represents MFP. The data for the variables are

collected from Australian Bureau of Statistics (ABS) cat. no. 5204.0 and 5206.0.

ICT capital stock consists of three categories of capital stocks—computers and

peripherals, computer software and electrical and electronic equipment. This

categorization is based on ABS’s definition of information technology net capital

stock (Table 69, Cat no 5204). The sample covers 1965–2013 for the aggregate

analysis and represents data for the Australian market sector. 8

The industry level

panel data analysis covers 12 out of 16 market sector industries for 1986–2013. 9

Sample periods and the industrial sectors considered in the analysis are determined

by the availability of data.

Table 2 reports the descriptive statistics of the sample employed in the aggregate

analysis. As shown in the table, labour productivity grew at an average rate of 2.2 %

as compared to the average growth rate of ICT capital deepening of 9.1 % and non-

ICT capital deepening of 4.6 % during 1965–2013.

4.3 Econometric methods

4.3.1 Time series

This study uses Augmented Dickey-Fuller (ADF; Dickey and Fuller 1979, 1981),

Phillips-Perron (PP; Phillips and Perron 1988) and Kwiatkowski–Phillips–Schmidt–

Shin (KPSS; Kwiatkowski et al. 1992) to check the stationarity properties of the

data. The null hypothesis of the ADF and PP tests is non-stationarity, whilst, in the

case of the KPSS test, the null is stationarity. In addition, the Zivot and Andrews

(2002) unit root test, which consider the possibility of a structural break in data, is

implemented to examine the consistency of test results. The null hypothesis of the

Zivot and Andrews test is the same as ADF and PP tests, however, the alternative

7 One of the reviewers pointed to the possible role of complementary factors in the relationship between

ICT and productivity. However, these complementary variables work through the channel of MFP (Mc

Morrow et al. 2010), a shift parameter in the models. In addition, in our model, non-ICT capital consists

of R&D and other forms of capital. Nonetheless, it is possible to disaggregate different forms of capital

and show their impact on MFP. Given the length and scope of the paper, we intend to cover this very

important issue in our future research. We present results for quality-adjusted hours worked of labour for

a shorter sample in the panel data analysis. 8

The market sector accounted for about 83 % of Gross Value Added at basic prices (in chain value

measure) for all industries in 2012–13 (ABS 2014a). 9

While the ABS has expanded its market sector to cover the 16 industries listed earlier, insufficient

disaggregated data are available for the extra four industries to be included in the industry level analysis.

Econ Change Restruct (2015) 48:281–304 291

123

hypothesis postulates a series is stationary with a break occurring at an unknown

point of time.

If most of the variables in the model are found to be integrated of order one

(I(1)), testing for cointegration is necessary. In the presence of cointegration, a

vector error correction model (VECM) can be formulated for further modelling and

inference (Engle and Granger 1987). The econometric issues related to the analysis

of VECM are discussed, for example, in Johansen (1995) and Pesaran and Pesaran

(1997).

Cointegration amongst the variables is examined by using the Gregory and

Hansen (1996) cointegration technique, ARDL-based bound approaches (Pesaran

and Pesaran 1997; Pesaran et al. 2001) and Johansen (1991, 1995). The Gregory

and Hansen test is a residual-based test for cointegration, which allows for the

possibility of structural breaks and regime shifts (Esso 2012). In the bound

testing approach, the null hypothesis of no cointegration among the variables can

be tested by applying general F-statistics and comparing them with critical

values in Narayan (2005). 10

The bound testing approach implies rejection of the

null if the computed F statistics are higher than the upper bound of the critical

values in case of I(1) variables. The ARDL approach allows the application of

general-to-specific modelling techniques to estimate consistent estimates of

parameters of the model. In addition, the ARDL-based estimators of the long-run

coefficients are biased in small sample sizes (Kumar et al. 2015; Pesaran 1999).

In the Johansen procedure, the maximum likelihood method is applied to test for

the cointegration rank, r, which can be found using the Trace (ktrace) and Max- eigenvalue (kmax) test statistics (Johansen 1988; Stock and Watson 1988). Both ARDL and Johansen approaches correct for possible endogeneity of the

variables, effectively by allowing simultaneous estimation of the long-run and

short-run components within the VECM framework (Shahiduzzaman and Alam

2014; Squalli 2007).

In this study, the long-run causality among the employed variables is also

examined by implementing the TY approach. The key advantage of the TY

Table 2 Descriptive statistics of the variables

y kit ko A

Index (2006–07 = 100)

Initial value (1965) 39.4 2.68 16.47 61.5

Ending value (2013) 110.7 169.4 139.3 81.9

Avg. growth rate (1965–2013) 2.2 9.1 4.6 1.0

Mean 70.7 35.4 60.1 81.9

Median 66.3 14.0 56.5 80.8

Observations 49 49 49 49

10 The critical F values provided by Narayan (2005) are preferred over the values provided by Pesaran

et al. (2001) as the former is suited for a sample size of 30–80 as compared to 500–1000 observations for

the latter.

292 Econ Change Restruct (2015) 48:281–304

123

approach is that it addresses the possible simultaneity of the variables by

implementing simultaneous equations modelling through seemingly unrelated

regression (SUR) equation modelling (Rambaldi and Doran 1996; Shahiduzzaman

and Alam 2012, 2014b). The TY approach involves the estimation of an augmented

VAR (k ? dmax) model, where k is the optimal lag length in the original VAR

system and dmax is the maximal order of the integration of the variables. The

procedure uses a modified Wald (mWald) test for cases with zero restrictions on the

parameters of the first k lags. This statistic follows an asymptotic Chi squared

distribution with k degrees of freedom in the limit when a VAR (k ? dmax) is

estimated.

4.3.2 Panel data

Similar to time series analysis, both stationary and cointegration properties are

examined by alternative methods. The tests of panel unit roots are carried out by

employing Im et al. (2003, henceforth IPS), ADF and PP Fisher tests (Choi 2001;

Fisher 1932; Maddala and Wu 1999) as well as Pesaran (2007). A major advantage

of Pesaran (2007) panel unit root test is that it controls for cross-section dependence.

The test for cointegration is performed by employing the residual based

cointegration test developed by Pedroni (1999) and the error correction based

cointegration test developed by Westerlund (2007). Both tests are appropriate for

heterogeneous panels, however, Westerlund (2007) relaxes the common factor

restriction. Third, the long-run relationship is estimated using the residual-based

panel Fully Modified OLS (FMOLS) and Dynamic OLS (DOLS) estimators (Kao

and Chiang 2001; Pedroni 2001; Phillips and Moon 1999). It has been proposed in

the literature that both methods produce asymptotically unbiased, normally

distributed coefficient estimates.

5 Estimation results

The results of the ADF, PP and KPSS unit root tests suggest that each of the

variables is non-stationary in levels and stationary in the first differences (Table 3).

Therefore, the variables are each I(1). These test results are supported by Zivot and

Table 3 ADF, PP and KPSS tests for unit roots

Variables ADF PP KPSS

Level First diff. Level First diff. Level First diff.

y –2.24 –8.87 a

–2.14 –9.07 a

0.09 0.06

kit –1.67 –4.02 b

–1.46 –4.02 b

0.23 a

0.06

ko –2.23 –4.07 b

–2.07 –3.93 b

0.19 b

0.19 b

A -1.70 -8.97 a

-1.66 -8.97 a

-0.09 -0.09

Superscript a, b, c denotes the significance level at 1, 5 and 10 %, respectively

Econ Change Restruct (2015) 48:281–304 293

123

Andrews (2002), which control for a structural break in data (Table 4). The next

step is to test for cointegration.

The results for Gregory and Hansen (1996) cointegration test, which allows for

any regime shifts in the relationship among the variables in Eq. 1.3 is presented in

Table 5. The ADF* and Zt *

test statistics reject the null hypothesis of no-

cointegration at around 10 % level after allowing for a structural break. The

breakpoint is determined as 1991, which might have some implications in parameter

estimation and this is explored further below.

5.1 ARDL approach

Before proceeding to the estimation of the Eq. (1.3), it is important to test whether

the production function (Eq. 1.2) satisfies the CSR property. Estimation of the

production function (Eq. 1.2) using ARDL approach results in the sum of the three

coefficients being 1.01. The Wald test of restriction cannot reject the null of

constant returns to scale (v2 (1) = 0.764; p value .382). Given this, the estimation of Eq. (1.3) is presented below.

In further investigating the cointegration property, we proceed with the ARDL

bound testing approach by including a dummy variable for a possible structural

break in 1991. The dummy variable takes a value of 1 from 1991 onwards and zero

otherwise. The computed F statistics for the bound testing of cointegration for

Eq. (1.3) is found as Fyðy j kit; ko; AÞ ¼ 4:85, which is higher than the upper bound critical value of 4.002 at 5 % level. Therefore, the null hypothesis of non-

cointegration is rejected.

Table 6 presents the long- and short-run elasticities using the ARDL approach.

The ARDL model (1, 2, 3, 1) based on the Schwarz Bayesian Criterion (SBC) is

Table 4 Zivot and Andrews (2002) unit root test

Series Trend Trend and intercept

Level Break First diff. Break Level Break First diff. Break

y –2.86 1972 –8.96 a

1983 –4.349 1981 –9.36 a

1996

kit –2.784 1981 –4.375 c

2005 –2.68 1977 –5.01 b

1982

ko –2.39 1973 –5.17 a

1996 –2.57 1993 –5.47 b

1985

A –3.32 2005 –9.45 a

2003 –3.40 2002 –9.84 a

1996

Superscript a, b and c denote the significance level at 1, 5 and 10 %, respectively. Tests consider a 10 %

trim for Trend and intercept assumption

Table 5 Results of the Gregory-Hansen Test for

cointegration with regime shifts

Statistics Value Break year Asymptotic critical values

1 % 5 % 10 %

ADF *

-5.78 1991 -6.51 -6.00 -5.75

Zt *

-5.74 1991 -6.51 -6.00 -5.75

Za *

-40.26 1991 -80.15 -68.94 -63.42

294 Econ Change Restruct (2015) 48:281–304

123

preferred since the model selected based on the AIC criterion does not pass the

residual serial correlation test. 11

Overall, however, both provide similar estimates of

long-run parameters. The estimated model for the sample period is found to be

adequate in terms of the diagnostic tests as shown in the lower part of the panel. The

long-run labour productivity elasticity of ICT capital deepening is 0.06 which is

significant at the 1 % level. ICT capital is therefore found to be a long-run forcing

variable to improve labour productivity. The long-run labour productivity elasticity

of non-ICT capital is 0.23 which is also found to be highly significant.

The estimated short-run elasticities in the table indicate negative effects of ICT

capital and positive effects of Other Capital. This warrants a little further discussion.

The table shows a negative coefficient for ICT capital at lag 1. This could be a

reflection of some type of adjustment cost that temporarily constrains the

effectiveness of ICT capital investment (Stiroh 2002c). The short-run elasticity

coefficients related to Other Capital are positive and significant both at the zero lag

and lag 2. In addition, very importantly, the estimated error correction (ecm) terms

are found to be significant along with expected negative signs confirming

cointegration among the variables. Figures 4, 5 in the ‘‘Appendix’’ presents the

plots of Cumulative Sum of Recursive Residuals (CUSUM) and Cumulative Sum of

Squares of Recursive Residuals (Square of CUSUM) with 5 % level of significance.

The test results confirm the stability of the estimated parameters.

Table 6 Estimated long-run and short-run elasticity

A: Lagrange multiplier test of

residual serial correlation

B: Ramsey’s RESET test using

the square of the fitted values

C: Based on a test of skewness

and kurtosis of residuals

D: Based on the regression of

squared residuals on squared

fitted values

* Related to error correction

model. Superscripts a and b

denote the significance at 1 and

5 %, respectively. Figures in the

parenthesis () show the standard

errors

Regressors 1965–2013

Long-run elasticities (dependent variable y)

kit 0.06 (0.008) a

ko 0.23 (0.017) a

A 1.03 (0.053) a

Constant -1.46 (0.200) b

Short-run elasticities (dependent variable Dy)

Dkit 0.007 (0.057)

Dkit-1 -0.054 (0.020) a

Dko 0.35 (0.023) a

Dko-1 0.030 (0.023)

Dko-2 0.041 (0.018) b

DA 1.04 (0.016)a

ecm-1 -0.18 (0.046) a

Diagnostics tests* SIC (1, 2, 3, 1)

A. Serial correlation v2 (1) = 2.50 [0.114]

B. Functional form v2 (1) = 0.254 [0.614]

C. Normality v2 (1) = 2.591 [.274]

D. Heteroscedasticity v2 (1) = 0.169 [0.681]

11 Detailed diagnostic test results can be obtained upon request.

Econ Change Restruct (2015) 48:281–304 295

123

5.2 Johansen approach

Johansen’s approach requires the selection of an optimal lag length of the VAR at

the first stage. As economic theory is not explicit about lag lengths, purely statistical

techniques are applied. We begin with a maximum lag of four and use different

information criteria to select the optimal lag for the VAR. The unrestricted VAR

models are estimated with intercepts. Both sequential modified log likelihood ratio

(LR) and Akaike Information Criterion (AIC) criterion suggests a lag of three to be

optimal for the VAR model in levels. 12

The cointegration model should therefore be

estimated by using a lag of two. The cointegration rank of one is accepted based on

Trace statistics at the 5 % significance level.

Table 7 presents the estimation results for the long-run relationships from VEC

modelling using the Johansen approach. The estimated model is found to be

adequate in terms of residual properties (not reported but available upon request).

The normalized cointegration vectors related to ICT capital deepening is estimated

to be .07, which is only slightly higher than the estimated coefficient in the ARDL

method as reported in Table 6. The estimated cointegration vectors related to non-

ICT capital deepening and technical progress are found to be slightly lower than that

reported in Table 6. Overall, the Johansen results are broadly in agreement with

those in ARDL approach in relation to the apparent long-run effect of ICT capital

deepening on labour productivity.

5.3 Long-run causality (non-causality)

Analysis of long-run causality (non-causality) by the implementation of the TY

technique requires determination of k and dmax at the first step. The optimal lag

length for the TY procedure is set to 3 based on AIC and LR criteria as found before

and dmax is set to 1 as the variables in the model are found to be I(1). The TY long-

run causality results based on VAR (4) model are presented in Table 8. The TY

Table 7 Estimated long-run coefficients using Johansen approach (normalized to y)

Cointegration with restricted intercepts and no trend in the VAR Sample 1965–2013. No. of cointegration

vector r = 1

Vector

y 1.00

kit -0.07 (0.01) a

ko -0.22 (0.02) a

A -0.99 (0.08) a

Intercept 1.24

Superscripts ‘‘a’’ denotes the significance at 1 % level. Figures in the parenthesis () show the standard

errors

12 The results on the selection of optimal lags and Johansen cointegration are not reported here, but will

be made available from the corresponding author upon request.

296 Econ Change Restruct (2015) 48:281–304

123

approach supports the proposition that ICT capital per labour input ‘Granger causes’

labour productivity. 13

The long-run Granger non-causality from kit to y is rejected at the 1 %

significance level. ICT capital per labour input is also found to ‘Granger cause’

MFP (Table 8). The TY results presented in Table 8 indicate the presence of long-

run causality from ICT capital per labour input to other capital but not in the other

direction. The most important implication from the results is that they support the

proposition that both ICT and non-ICT capital deepening are forcing variables in

improving both labour productivity and MFP in the long-run.

5.4 Panel data analysis

In this section, Eq. (1.3) is re-estimated by using industry level panel data for the

period of 1986–2013. Table 9 shows the results from four alternative panel data unit

root tests. Similar to the time series tests, results from all four panel data unit root

tests indicate the variables are each I(1). Therefore, the usual approach is then to test

for cointegration. Table 10 reports the Pedroni (1999, 2004) panel cointegration test

and Table 11 reports the results from Westerlund (2007) error correction based

panel cointegration test. The test results indicate the existence of cointegration

among the variables. The third step is to estimate the long-run coefficients using

panel cointegration FMOLS and DOLS estimators. The results are reported in

Table 12.

The results from both FMOLS and DOLS estimators show a significant and

positive long-run effect of ICT capital on labour productivity in Australia

(Table 12). The estimated long-run elasticity for ICT capital for the industry panel

is 0.05–0.06, which is consistent with the time series analysis results. The estimated

long-run elasticity for non-ICT capital is found to be considerably higher than that

Table 8 TY causality tests (1965–2013)

Results are obtained through

seemingly unrelated regression

(SUR) estimation

Null hypothesis Modified Wald statistics p value

kit does not Granger-cause y 24.01 0.00

y does not Granger-cause kit 3.71 0.29

kit does not Granger-cause A 20.57 0.00

A does not Granger-cause kit 2.29 0.51

ko does not Granger-cause y 14.90 0.00

y does not Granger-cause ko 7.73 0.05

ko does not Granger-cause A 16.62 0.00

A does not Granger-cause ko 8.73 0.03

kit does not Granger-cause ko 13.43 0.01

ko does not Granger-cause kit 1.28 0.73

A does not Granger-cause y 1.22 0.75

y does not Granger-cause A 4.76 0.19

13 It should be noted that this result of unidirectional causality is quite sensitive to the inclusion of data

for 2013. In fact, bi-directionality is supported when the model is estimated for the period of 1965–2012.

Econ Change Restruct (2015) 48:281–304 297

123

found in the time series analysis. The coefficient related to technical progress is

found to be higher in the panel data analysis. Of course, this may reflect the

Table 9 Results from Panel Unit Root Tests

Variable IPS Test Fisher ADF Fisher PP Pesaran (2007)

y 1.65 (0.95) 21.84 (0.59) 20.92 (0.94) -2.48 (0.27)

Dy -14.54 (0.00) 201.81 (0.00) 212.04 (0.00) -4.07 (0.00)

kit 4.14 (1.00) 8.78 (0.99) 8.55 (0.99) -1.81 (0.97)

Dkit -7.59 (0.00) 102.23 (0.00) 109.46 (0.00) -2.91 (0.01)

ko 3.06 (0.99) 18.39 (0.78) 18.40 (0.78) -1.71 (0.99)

D ko -10.57 (0.00) 145.04 (0.00) 168.05 (0.00) -3.11 (0.001)

A 0.43 (0.67) 29.46 (0.20) 19.25 (0.74) -2.58 (0.169)

DA -11.63 (0.00) 166.61 (0.00) 201.87 (0.00) -4.15 (0.00)

The IPS, Fisher ADF, Fisher PP tests assume individual unit root process (Null). Probabilities for Fisher

tests are computed using an asymptotic Chi-square distribution. All other tests assume asymptotic nor-

mality. Figures in the parentheses represent the p values

Table 10 Results from Pedroni (1999, 2004) Panel

Cointegration Tests

Statistics Probabilities

Panel ADF -3.12 0.0009

Panel PP -3.29 0.0005

Group ADF -3.25 0.0006

Group PP -2.56 0.0005

Table 11 Results from Westerlund (2007) error

correction based Panel

Cointegration Tests

Statistics Value Z value p value

Gt -2.782 -2.017 0.022

Ga -11.868 -0.444 0.329

Pt -9.486 -2.664 0.004

Pa -9.931 -1.306 0.096

Table 12 FMOLS and DOLS estimates: industry panel

Dependent variable y FMOLS DOLS

Sample 1986–2013. Cross-sections included 12

kit 0.06 (0.00) a

0.05 (0.00) a

ko 0.38 (0.00) a

0.38 (0.00) a

A 1.05 (0.00) a

1.03 (0.00) a

Superscripts ‘‘a’’ denotes the significance at 1 % level. Figures in the parenthesis () show the standard

errors. FMOLS is estimated with pooled weight capturing for heterogeneous first-stage long-run coeffi-

cients. DOLS is estimated with pooled weight with 2 lags and 1 lead

298 Econ Change Restruct (2015) 48:281–304

123

difference in sample period and the size of sample. All coefficients in the panel data

analysis have the expected positive signs and are highly significant.

The actual size of the long-run elasticity coefficient related to ICT capital

deepening is quite significant given that the share the ICT capital to total capital is still

quite low in Australia (an average of 1.7 % over 1965–2013 and 3.5 % in 2000s, for

the market sector). This may reflect the possibility of increasing returns from the

accumulation of this particular form of capital. The elasticity coefficient related to

other capital is found to be between 0.34 and 0.37. Table 13 in the ‘‘Appendix’’

presents the results of using Quality-adjusted hours worked for the labour variable for

the period 1990–2013. The results indicate a considerably lower contribution of Other

capital, but a higher contribution of ICT capital for the sample period.

6 Conclusions, possible limitations and policy implications

This paper revisits the so-called ‘ICT-productivity paradox’ from a long-run

perspective by using Australian data for about five decades. The ‘ICT productivity

inducing effects’ issue has been examined in quite a large number of studies, however,

these have focussed predominantly upon data for the 1980s and 1990s. Moreover, the

majority of the existing macro level evidence is based upon a ‘growth accounting’

exercise and the application of time series econometrics is found to be quite limited.

This study seeks to fill this gap in the literature by employing an array of appropriate

time series and panel data econometrics techniques. It provides estimates of long-run

and short-run elasticities of labour productivity with respect to ICT capital deepening

and explores the direction of long-run causality among the variables.

Using alternative modelling approaches, and based on the quite long sample

period used, the results support long run labour productivity elasticity with respect

to ICT capital deepening of around 0.06–0.07. The corresponding long-run elasticity

coefficient estimated in the panel data analysis is found to be 0.05, suggesting a

reasonable degree of consistency in the results derived from the time-series and

panel-data model approaches in their application to the data used in this analysis.

The contribution of ICT capital is significant in the context that the average share of

ICT capital to total capital over the sample period was below 2 %. This suggests an

apparently very high marginal labour productivity gain from ICT capital deepening.

The elasticity estimate related to ICT capital is found to be consistent with the

cluster of elasticities around the value of 0.05–0.06 as reported in the survey by

Cardona et al. (2013).

The long run elasticity coefficient related to Other Capital was found to be

0.22–0.23 in the alternative time series models, whilst the industry panel estimates

yielded an elasticity coefficient of around 0.34–0.37. It may be inferred that the

aggregate time-series analysis estimate as reported in this paper might have

understated the contribution of Other Capital.

The results also show evidence of unidirectional causality from ICT capital

deepening to labour productivity. The result is consistent with recent literature on

Australia (e.g., Shahiduzzaman and Alam 2014a). Evidence of unidirectional

causality was also found from ICT capital to MFP. This suggests that increases

Econ Change Restruct (2015) 48:281–304 299

123

(decreases) in ICT capital deepening could accelerate (decelerate) both labour

productivity and MFP, if the past trend continues into the future. ICT capital

deepening is also found to ‘Granger cause’ non-ICT capital. The important

implication from the results is that both ICT and non-ICT capital investment appear

to be long-run forcing variables to improve labour productivity and MFP in

Australia.

Taken all together, the results from the study provide quite strong evidence of

long-run effects of ICT capital deepening on productivity growth in Australia along

with the possibility of increasing returns from the accumulation of ICT capital.

While the results are robust to alternative econometric methods, some limitations

remain. There is a longstanding and widespread criticism of accounting for

technical progress, or MFP, using a growth accounting framework as it considers

productivity as a residual value. Moreover, the methodological framework used in

this paper ignores the interaction between ICT investment and human and

organization capital as emphasized by the recent developments of growth literature

(see, for example, Aghion and Howitt 1998; Romer 1990).

For instance, over the period analysed, not only was there greater quantities of

additional ICT investment being made, but there was also very rapid advances in

ICT technologies going on. This virtuous circle of more advanced ICT technologies

driving even more advanced technologies, and the impact this process may have had

on disembodied technical change, human capital development and enterprise

organisation was presumably evident in the aggregated data used in this study.

Given that the pace of ICT technology change seemingly continues unabated, a

similar effect may also be expected into the future. If these factors could somehow

be properly accounted for, an even more accurate picture of the impact of ICT

capital investment on productivity growth could be determined. Unfortunately, this

kind of disaggregated data is not available at the macro level for a long enough time

period to allow the use of the sorts of techniques as used in this study.

Another example of this type of issue in the current context relates to the

microeconomic reforms of the 1980s, 1990s and early 2000s mentioned earlier.

These were also quite likely to have been significant drivers of the enhanced

productivity (both labour and MFP) in Australia as observed in the 1988–2004

period. Whilst the quite long length of the data sample used in the study

(1965–2013) can be expected to mitigate any potential confounding effects from

this source of productivity growth, it nevertheless would be interesting if these

potential sources of productivity gains could be disentangled and isolated from the

separate impacts of ICT and non-ICT capital deepening. Unfortunately, again, it is

not at all clear how this could be sensibly done in the context of the sorts of

econometric techniques employed in this paper.

Acknowledgments This project is supported through the Australian Government’s Collaborative Research Networks (CRN) program. The authors thank George Hondroyiannis, the Editor in Chief of the

journal, and the two anonymous reviewers for their very useful comments.

300 Econ Change Restruct (2015) 48:281–304

123

Appendix

See Figs. 4, 5 and Table 13.

The straight lines represent critical bounds at 5% significance level

-20

-10

0

10

20

1968 1980 1992 2004 2013

Fig. 4 Plot of cumulative sum of recursive residuals

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1968 1980 1992 2004 2013

The straight lines represent critical bounds at 5% significance level

Fig. 5 Plot of cumulative sum of squares of recursive residuals

Table 13 FMOLS and DOLS estimates of industry panel: quality adjusted hours worked basis

Dependent variable y FMOLS DOLS

Sample 1990–2013. Cross-sections included 12

kit 0.08 (0.001) a

0.07 (0.011) a

ko 0.20 (0.004) a

0.15 (0.03) b

A 1.10 (0.01) a

1.04 (0.045) b

Superscripts ‘‘a’’ and ‘‘b’’ denote the significance at 1 and 5 % levels. Figures in the parenthesis () show

the standard errors. FMOLS is estimated with pooled weight capturing for heterogeneous first-stage long-

run coefficients. DOLS is estimated with pooled weight with 2 lags and 1 lead

Econ Change Restruct (2015) 48:281–304 301

123

References

ABS (2012) Cat no. 5204, Australian System of National Accounts, various tables. Australian Bureau of

Statistics (ABS), Canberra, August 17, 2012. http://www.abs.gov.au/AUSSTATS

ABS (2014a) Cat no. 5204, Australian system of national accounts, Table 5, Australian Bureau of

Statistics (ABS), Canberra, August 17, 2012. http://www.abs.gov.au/AUSSTATS

ABS (2014b) Cat no. 5204, Table 69. Information technology net capital stock, selected items by

industry, Australian Bureau of Statistics (ABS), Canberra, April 06, 2014. http://www.abs.gov.au/

AUSSTATS

Acharya RC, Basu S (2010) ICT and total factor productivity growth: intangible capital or productive

external ties?. Micro-Economic Policy and Analysis Industry, Canada

Aghion P, Howitt P (1992) A model of growth through creative destruction. Econometrica 60(2):323–351

Aghion P, Howitt P (1998) Endogenous growth theory. MIT press, Cambridge

Banks G (2001) The role of ICT in Australia’s economic performance. In: Commonwealth research

forum proceedings of the Commonwealth research forum. Rydges Lakeside, Canberra

Bean C (2000) The Australian economic ‘‘miracle’’: a view from the north. In: Gruen D, Shrestha S (eds)

The Australian economy in the 1990s. Reserve Bank of Australia, Sydney, pp 73–114

Bresnahan TF, Trajtenberg M (1995) General purpose technologies ‘Engines of growth’? J Econ

65(1):83–108

Cardona M, Kretschmer T, Strobel T (2013) ICT and productivity: conclusions from the empirical

literature. Inf Econ Policy 25(3):109–125

Choi I (2001) Unit root tests for panel data. J Int Money Finan 20(2):249–272

Christopoulos D, McAdam P (2013) ‘Openness, efficiency and technology: an industry assessment. Scott

J Polit Econ 60(1):56–70

Cobb CW, Douglas PH (1928) A theory of production. Am Econ Rev 18(1):139–165

Colecchia A, Schreyer P (2002) ICT investment and economic growth in the 1990s: is the United States a

unique case? A comparative study of nine OECD countries. Rev Econ Dynam 5(2):408–442

Daveri F, Silva O (2004) Not only Nokia: what Finland tells us about new economy growth. Econ Policy

19(38):117–163

Dickey D, Fuller W (1979) Distribution of the estimators for autoregressive time series with a unit root.

J Am Stat Assoc 74(366):427–431

Dickey D, Fuller W (1981) Likelihood ratio statistics for autoregressive time series with a unit root. Econ

J Econ Soc 49(4):1057–1072

Dolman B (2009) What happened to Australia’s productivity surge? Austr Econ Rev 42(3):243–263

Draca M, Sadun R, Van Reenen J (2006) Productivity and ICT: a review of the evidence. Centre for

Economic Performance, The London School of Economics and Political Science, London

Edquist H (2005) The Swedish ICT miracle—myth or reality? Inf Econ Policy 17(3):275–301

Engle R, Granger C (1987) Co-integration and error correction: representation, estimation, and testing.

Econometrica 55(2):251–276

Esso LJ (2012) Re-examining the saving-investment nexus: threshold cointegration and causality

evidence from the ECOWAS. Econ Change Restruct 45(3):193–220

Fisher RA (1932) Statistical methods for research workers. Oliver and Boyd, Edinburgh

Gordon RJ (2000) Does the ‘‘New Economy’’ measure up to the great inventions of the past? J Econ

Perspect 14(4):49–74

Gregory AW, Hansen BE (1996) Residual-based tests for cointegration in models with regime shifts.

J Econ 70(1):99–126

Gretton P, Gali J, Parham D (2002a) Uptake and impacts of ICT in the Australian economy: evidence

from aggregate, sectoral and firm levels. In: OECD workshop on ICT and business performance,

productivity commission, Canberra, December, proceedings of the OECD workshop on ICT and

business performance, productivity commission, Canberra, December

Gretton P, Gali J, Parham D (2002b) Uptake and impacts of ICT in the Australian economy: evidence

from aggregate, sectoral and firm levels. In: Workshop on ICT and business performance.

Proceedings of the workshop on ICT and business performance. OECD, Paris

Gretton P, Gali J, Parham D (2004) The effects of ICTs and complementary innovations on Australian

productivity growth. In: OECD (ed) The economic impact of ICT. OECD, Paris

302 Econ Change Restruct (2015) 48:281–304

123

Gruen D (2012) The importance of productivity. In: Productivity perspective conference: productivity

commission—Australian Bureau of Statistics: proceedings of the productivity perspective confer-

ence: productivity commission—Australian Bureau of Statistics Canberra, November

Im KS, Pesaran MH, Shin Y (2003) Testing for unit roots in heterogeneous panels. J Econ 115(1):53–74

Johansen S (1988) Statistical analysis of cointegration vectors. J Econ Dynam Control 12(2–3):231–254

Johansen S (1991) Estimation and hypothesis testing of cointegration vectors in Gaussian vector

autoregressive models. Econ J Econ Soc 59:1551–1580

Johansen S (1995) Likelihood-based inference in cointegrated vector autoregressive models. Oxford

University Press, New York

Jorgenson DW (2001) Information technology and the US economy. Am Econ Rev 91(1):1–32

Jorgenson DW (2004) Information technology and the G7 economies. MIT Press, Cambridge

Jorgenson DW, Griliches Z (1967) The explanation of productivity change. Rev Econ Stud

34(3):249–283

Jorgenson DW, Stiroh KJ (1999) Productivity growth: current recovery and longer-term trends. Am Econ

Rev 89(2):109–115

Jorgenson DW, Stiroh KJ (2000) Raising the speed limit: U.S. economic growth in the information age.

Brook Papers Econ Act 2000(1):125–235

Jorgenson DW, Ho MS, Stiroh KJ (2008) A retrospective look at the U.S. productivity growth resurgence.

J Econ Perspect 22(1):3–24

Kao C, Chiang M-H (2001) On the estimation and inference of a cointegrated regression in panel data.

Advances in econometrics 15:179–222

Kretschmer T (2012) Information and communication technologies and productivity growth: a survey of

the literature. OECD Publishing, Paris

Kumar RR, Stauvermann PJ, Patel A (2015) Nexus between electricity consumption and economic

growth: a study of Gibraltar. Econ Change Restruct 48(2):119–135

Kwiatkowski D, Phillips PCB, Schmidt P, Shin Y (1992) Testing the null hypothesis of stationarity

against the alternative of a unit root. J Econ 54:159–178

Maddala GS, Wu S (1999) A comparative study of unit root tests with panel data and a new simple test.

Oxford Bull Econ Stat 61(S1):631–652

Mc Morrow K, Röger W, Turrini A (2010) Determinants of TFP growth: a close look at industries driving

the EU–US TFP gap. Struct Change Econ Dynam 21(3):165–180

Narayan P (2005) The saving and investment nexus for China: evidence from cointegration tests. Appl

Econ 37(17):1979–1990

Oliner SD, Sichel DE (1994) Computers and output growth revisited: how big is the puzzle? Brook

Papers Econ Activ 1994(2):273–334

Oliner SD, Sichel DE (2000) The resurgence of growth in the late 1990s: is information technology the

story? J Econ Perspect 14(4):3–22

Oliner SD, Sichel DE (2002) Information technology and productivity: where are we now and where are

we going?. Divisions of Research & Statistics and Monetary Affairs, Federal Reserve Board, USA

O’Mahony M, Vecchi M (2005) Quantifying the impact of ICT capital on output growth: a heterogeneous

dynamic panel approach. Economica 72(288):615–633

Parham D (2002) Productivity gains: importance of ICTs. Agenda 9(3):195–210

Parham D (2004a) Australia’s 1990s productivity surge and its determinants. University of Chicago Press,

Chicago, pp 41–70

Parham D (2004b) Sources of Australia’s productivity revival. Econ Record 80(249):239–257

Parham D (2005) Australia’s 1990s productivity surge: a response to Keith Hancock’s challenge. Austr

Bull Labour 31(3):295

Parham D, Roberts P, Sun H (2001) Information technology and Australia’s productivity surge. Staff

Research Paper, Productivity Commission, Canberra

Pedroni P (1999) Critical values for cointegration tests in heterogeneous panels with multiple regressors.

Oxford Bull Econ Stat 61(S1):653–670

Pedroni P (2001) Fully modified OLS for heterogeneous cointegrated panels. Adv Econ 15:93–130

Pedroni P (2004) Panel cointegration: asymptotic and finite sample properties of pooled time series tests

with an application to the PPP hypothesis. Econ Theory 20(03):597–625

Pesaran MH (1999) An autoregressive distributed lag modelling approach to cointegration analysis. In:

Proceedings of the Cambridge University Citeseer. Cambridge University

Pesaran MH (2007) A simple panel unit root test in the presence of cross-section dependence. J Appl

Econ 22(2):265–312

Econ Change Restruct (2015) 48:281–304 303

123

Pesaran M, Pesaran B (1997) Working with Microfit 4.0: interactive econometric analysis. Oxford

University Press, Oxford

Pesaran M, Shin Y, Smith R (2001) Bounds testing approaches to the analysis of level relationships.

J Appl Econ 16(3):289–326

Phillips PC, Moon HR (1999) Linear regression limit theory for nonstationary panel data. Econometrica

67(5):1057–1111

Phillips PCB, Perron P (1988) Testing for a unit root in time series regression. Biometrika 75(2):335–346

Quiggin J (2001) The Australian productivity miracle: a sceptical view. Agenda 8(4):333–348

Quiggin J (2006) Stories about productivity. Austr Bull Labour 32(1):18–26

Rambaldi AN, Doran HE (1996) Testing for Granger non-causality in cointegrated systems made easy,

vol. In: Working papers in econometrics and applied and applied statistics 88. Department of

Econometrics, The University of New England

Romer P (1990) Endogenous technological change. J Polit Econ 98(5):71–101

Shahiduzzaman M, Alam K (2012) Cointegration and causal relationships between energy consumption

and output: assessing the evidence from Australia. Energy Econ 34(6):2182–2188

Shahiduzzaman M, Alam K (2014) A reassessment of energy and GDP relationship: the case of Australia.

Environ Dev Sustain 16(2):323–344

Shahiduzzaman M, Alam K (2014a) Information technology and its changing roles to economic growth

and productivity in Australia. Telecommun Policy 38:125–135

Shahiduzzaman M, Alam K (2014b) The long-run impact of Information and communication technology

on economic output: the case of Australia. Telecommun Policy 38(7):623–633

Solow RM (1956) A contribution to the theory of economic growth. Q J Econ 70(1):65–94

Solow RM (1957) Technical change and the aggregate production function. Rev Econ Statist

39(3):312–320

Solow RM (1987) We’d better watch out. The New York Times Book Review, July 12

Squalli J (2007) Electricity consumption and economic growth: bounds and causality analyses of OPEC

members. Energy Econ 29(6):1192–1205

Stiroh K (2002a) Reassessing the impact of IT in the production function: a meta-analysis. Federal

Reserve Bank of New York, mimeo

Stiroh KJ (2002b) Information technology and the US productivity revival: what do the industry data say?

Am Econ Rev 92(5):1559–1576

Stiroh KJ (2002c) Are ICT spillovers driving the new economy? Rev Income Wealth 48(1):33–57

Stock JH, Watson MW (1988) Testing for common trends. J Am Stat Assoc 83(404):1097–1107

Swan TW (1956) Economic growth and capital accumulation. Econ Rec 32(2):334–361

Toda HY, Yamamoto T (1995) Statistical inference in vector autoregressions with possibly integrated

processes. J Econ 66(1–2):225–250

Triplett JE (1999) The Solow productivity paradox: what do computers do to productivity? Can J Econ

Revue canadienne d’Economique 32(2):309–334

Van Ark B, Inklaar R (2005) Catching up or getting stuck? Europe’s troubles to exploit ICT’s

productivity potential. Groningen Growth and Development Center, Groningen

Venturini F (2007) ICT and productivity resurgence: a growth model for the information age. BE J

Macroecon 7(1):1–26

Venturini F (2009) The long-run impact of ICT. Empir Econ 37(3):497–515

Vourvachaki E (2009) Information and communication technologies in a multi-sector endogenous growth

model. CERGE-EI Working Paper, no. 386

Westerlund J (2007) Testing for error correction in panel data*. Oxford Bull Econ Stat 69(6):709–748

Zivot E, Andrews DWK (2002) Further evidence on the great crash, the oil-price shock, and the unit-root

hypothesis. J Bus Econ Stat 20(1):25–44

304 Econ Change Restruct (2015) 48:281–304

123

Copyright of Economic Change & Restructuring is the property of Springer Science & Business Media B.V. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.