Time series analysis in the assessment of ICT impact at the aggregate level – lessons and implications for the new economy

Information & Management 42 (2005) 1009–1022 www.elsevier.com/locate/dsw

Time series analysis in the assessment of ICT impact at the aggregate level – lessons and implications for the new economy Sang-Yong Tom Leea,*, Roghieh Gholamib, Tan Yit Tongb a

College of Information and Communications, Hanyang University, 17 Haengdang-dong, Seongdong-gu, Seoul 133-791, Republic of Korea b Department of Information Systems, National University of Singapore, 3 Science Dr. 2, Singapore 117543, Singapore Received 3 April 2004; received in revised form 13 November 2004; accepted 29 November 2004 Available online 12 January 2005

Abstract The major role of information and communication technology (ICT) in the new economy is well documented: countries worldwide are pouring resources into their ICT infrastructure despite the widely acknowledged ‘‘productivity paradox’’. Evaluating the contribution of ICT investments has become an elusive but important goal of IS researchers and economists. But this area of research is fraught with complexity and we have used Solow’s Residual together with time-series analysis tools to overcome some methodological inadequacies of previous studies. Using this approach, we conduct a study of 20 countries to determine if there was empirical evidence to support claims that ICT investments are worthwhile. The results show that ICT contributes to economic growth in many developed countries and newly industrialized economies (NIEs), but not in developing countries. We finally suggest ICT-complementary factors, in an attempt to rectify possible flaws in ICT policies as a contribution towards improvement in global productivity. # 2004 Elsevier B.V.. All rights reserved. Keywords: Information and communication technology (ICT); Economic growth; Productivity paradox; Time series; Solow’s Residual

1. Introduction The Nobel laureate economist Robert Solow once cited the infamous ‘‘productivity paradox’’ of the US economy, where productivity stagnated despite increasing computing power. His quip that ‘‘the computer age is everywhere but in the productivity statistics’’ [53] might apply to other advanced * Corresponding author. Tel.: +82 2 2290 1814; fax: +82 2 2290 1886. E-mail address: [email protected] (S.-Y.T. Lee).

economies as well. Much of the early research of the 1970s and 1980s also indicated the negative effects of computers on productivity [2,51,55,49,46,50]. A number of authors attempted to provide justifications for the post-1970s ‘‘clash of expectations and statistics’’. Their review of the paradox [6,12] produced a gamut of explanations, including mismeasurement of outputs and inputs and lags due to learning and adjustment. The most widely recognized explanation for the post-1973 productivity slowdown was flaws in methodological frameworks and measurement errors. Another explanation was that the

0378-7206/$ – see front matter # 2004 Elsevier B.V.. All rights reserved. doi:10.1016/j.im.2004.11.005

1010

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022

main benefits from using computers – improved quality, timeliness, and customization [7] – were not properly measured in official productivity statistics. In the 1990s, a number of researchers sought to find a positive contribution of information and communication technology (ICT) to economic growth. Brynjolfsson and Yang [9] cited contemporary studies that had associated ICT with productivity growth. While refuting the productivity paradox, authors of recent studies have attributed this change to improved data quality and a new econometric framework that produced more satisfying empirical results. At the firm and industry level, several authors have noted positive evidence of returns from ICT investments [36,19,23]. Their results have been confirmed by a number of recent studies and initiated a large stream of research. However, there are limitations to the implications of the results. In particular, as the ICT productivity paradox was originally defined at the economy level, one natural concern was that most of these recent IS studies have addressed the productivity question at the micro level [10,45]. In contrast to investigations at the firm and industry levels, studies at the aggregate level have not been conclusive. They have also suffered from limitations in their analysis. For instance, in their country-level research in the Asia Pacific area, Kraemer and Dedrick [30] found that a positive correlation existed between ICT and economic growth. However, they acknowledged that they could not provide conclusive evidence of a causal relationship, given the relatively small portion of the economy allocated to ICT in the overall capital and the broad array of factors affecting economic growth. Similarly, Jorgenson and Stiroh [27] discovered that computer capital contributed to growth more than ordinary capital, suggesting a positive payoff from ICT. However, the extrapolation of total factor productivity (TFP) growth for long-term projections is questionable. Even the researchers themselves admitted: ‘‘Only as the statistical agencies continue their slow progress towards improved data and implementation of state-of-the-art methodology will this murky picture become more transparent.’’ Pohjola [48] indicated that disappointment in ICT is still chronicled in many macroeconomic studies, because the impact on productivity and economic growth has been much harder to detect. Therefore,

better measurement methods and definitions are required for more precise appraisal, especially in the Internet and e-commerce era. In response, we proposed the use of Solow’s Residual together with time-series analysis tools to overcome the methodology inadequacies of previous studies. Our goal was to establish empirical evidences to assess previous productivity strategies. Our hope was that our findings would help guide future ICT investment decisions in both developing and developed countries. The use of Solow’s Residual offers the ability of better measuring the productivity attributable to technology. The majority of earlier research used tangible outputs, such as gross domestic product (GDP), national wealth, and revenue; these output measures might not capture the full contribution of ICT to an economy’s productivity, because the impact of ICT usage is generally considered to be wideranging but intangible. Solow’s Residual, which provides more information about changes in technology than other productivity measures, should therefore better appraise the effectiveness of the use of ICT. For each of the countries sampled here, we first investigated the causal relationship between ICT and GDP by looking directly at the production function. Then we derived Solow’s Residual for the country in order to analyze the impact of ICT on its economic growth. In both analyses, we incorporated time series statistical tools because all the variables – GDP, capital, labor and ICT – were generally generated for a particular instance in time. Econometricians, in their investigations, have often imposed theories on data even when its temporal structure does not conform to their theories; this inadequacy is common in studies on the relationship between ICT and productivity. In our study, we implemented time series analysis tools to eliminate this spurious regression problem. Because time series tools allow the researcher to test data stationarity before making any further analyses, corrective measures can be incorporated in the statistical tests; the researcher would thus be spared the potential problems of ordinary regression. Thus, more consistent empirical findings can be expected from our methodology. Traditional regression methods are susceptible to the limitation of reliable forecasting; similarly, the predicted values of the variable would also have to be

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022

near the range of the sample values [39]. In contrast, time series tools allow the contribution of ICT capital to be more accurately projected into the future. This is absolutely important, because the value of ICT does not present itself at any particular point in time but rather unfolds as ICT applications and infrastructure are put into effective use. Moreover, finding a strong association between ICT investment and growth does not necessarily imply a causal relationship. If non-stationary time series variables are not cointegrated, then a high degree of correlation between two variables does not mean a causal relationship between them. The time series methodology thus allows us to recognize and avoid spurious results. Obviously, time does not go backwards. Using a time series methodology, we can use lags to identify causal relationships. This is not possible in cross-sectional studies. The time series methodology also allows us to answer some specific questions. For example, we can find out whether the relationship between variables is long- or short-run. The Granger causality test enables us to determine the direction of causality and find out whether ICT growth causes GDP growth, whether GDP growth causes ICT growth, or whether there is a feedback effect between ICT and GDP. Longitudinal or panel data analyses, although rather resource intensive, allow researchers to obtain a deeper understanding of the impact of technologies along the continuum of ICT investment [37]. As Kohli and Devaraj [29] suggested, researchers should gather larger samples including longitudinal or panel data to assess the effects of ICT payoff. Such data can often improve the accuracy of the results, because they can control for country (industry or firm) specific effects. Application of this data also allows the researcher to examine the lag effects of technological impact [13,47]. This is an advantage, as neglect of lag effects has been cited as a factor contributing to the productivity paradox [8,35]. Economic analyses of growth operate on the belief that it is in some way related to qualitative change. The assumption is that growth is not just the extension of an existing activity but involves doing new things with new processes, which entail a change in the character of the activity. For instance, Adam Smith [54] perceived that growth was associated with a more complex division of labor: the components of existing

1011

activities would spin off as separate activities and be subject to productivity growth as people specialized in them and became more skilful at carrying them out. Smith also predicted that knowledge creation would become a separate activity and that this would further drive productivity growth [5]. Cross-country studies on the productivity impact of ICT are still relatively scarce, primarily because comparable data sources are quite new. Despite the fact that the productivity paradox of ICT has been considered to be an international phenomenon, Dewan and Kraemer [14] noted, most of the existing studies involve firmlevel analysis [28,56,34,52,44] conducted mainly in the US. Mahmood and Mann [40] asserted that it was important for researchers to include an international dimension to ICT investment-performance relationships and extend their research focus beyond the US to encompass the experience of other countries. Hence, we sought to make a significant contribution to the value of ICT investment by using new findings on international experiences with ICT investment. The answers to our research questions should have important theoretical and practical implications. By making comparisons among countries to appraise the cause of growth disparities, we attempted to identify the characteristics of national innovation systems that were linked to strong innovative performance. More importantly, we intended to uncover evidence to support the view that the contribution of ICT can be a long-term, sustainable phenomenon. For practitioners, the empirical findings should then serve to shed light on ICT policy making.

2. Methodology 2.1. The production function To investigate the long-run relationship between variables, we first deployed the Cobb-Douglas production function. It can be viewed as a linear approximation to the underlying production function and has proven to be a good approximation in the ICT and productivity contexts [16]; it is also pervasive in productivity research literature [22]. The CobbDouglas functional form is an augmented neoclassical model of economic growth. The basic Solow model is thus extended to include investment in human

1012

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022

capital and ICT, in addition to investment in physical capital. Moreover, the production function approach has been widely used in previous studies of ICT impact on company performance [1,38]. The typical production function requires three resources, i.e., labor (L), capital (K), and ICT. Hence, we have the following production function: Y ¼ AICTb1 K b2 Lb3

(1)

where Y is output (GDP), A is a constant representing other factors of production, and b1, b2, and b3 are the elasticities of the production resources. This function can be converted into its log-linear form for analytical convenience: lnðYÞ ¼ a þ b1 lnðICTÞ þ b2 lnðKÞ þ b3 lnðLÞ

(2)

2.2. Solow’s Residual Assuming that the aggregate production function has a simple Cobb-Douglas functional form Y ¼ AK a L1
a

(3)

where Y is GDP or output, K is aggregate capital stocks, L is the labor force, A > 0 is the constant representing other factors of production, which measures mainly the productivity of the available technology, and 0 < a < 1 is the share parameter representing the elasticities of the production resources. Total factor productivity is popularly known as Solow’s Residual, and is what economic historians use to explain which part of a country’s economic productivity remains unexplained and is thus attributed mainly to technology. Using the production function from Eq. (3), Solow’s Residual, A, can be derived as: Y (4) K a L1
a Many growth accounting studies have estimated a either econometrically or using national account data for both developed and developing countries. There is also evidence that, for the typical developed (Organization for Economic Cooperation and Development or OECD) country, a roughly equals 0.33, and for the typical developing country, a roughly equals 0.4. In general, capital’s share of GDP is 1/3, regardless of the levels of K or L [41].1 Thus, following Ben-David and A¼

1

Estimated a’s from the OLS also support this idea.

Rahman [3], the value of this share parameter is set to both a = 0.3 and a = 0.4 in the derivation of Solow’s Residual; then the empirical results obtained from both values are compared. Solow’s Residual better measures the actual productivity attributed to technology. Most earlier research used tangible outputs, such as GDP, national wealth and revenue, and these might not capture the entire contribution of ICT in an economy’s productivity. Since it is widely acknowledged that ICT usage provides a wide range of intangible impacts, Solow’s Residual could better appraise the value of ICT. Once we had obtained Solow’s Residual as a measure of productivity, we needed to investigate the causal relationship between GDP (Y in the equations) and ICT, and between Solow’s Residual (A in the equations) and ICT. To study the long-run relationship between variables, we use Johansen’s [25] cointegration test. In time series analysis, two variables are said to have a long-run relationship if they are cointegrated. For the variables in Eq. (2) to be cointegrated, the order of integration of the left-hand-side variable (Y) should be equal to or greater than the highest order of integration of the right-hand-side variables (K, L and ICT). Otherwise, even without cointegration tests, they are obviously not cointegrated. Before the cointegration tests, we implemented the Augmented Dickey-Fuller (ADF) test to establish the order of integration of the variables [17,18]. This test was also useful for obtaining stationary variables for the Granger causality test. Depending on whether they were cointegrated, different tests would be necessary to determine the causal relationship between the two variables: if the variables were cointegrated, we deployed the vector error correction model (VECM); the Granger causality test was conducted for noncointegrated variables. 2.3. The unit root test We used the ADF test to check for the unit roots of the time series variables. Consider the following expression: m X DYt ¼ aYt
1 þ bi DYt
i þ d þ gt þ et (5) i¼1

where DY is the first difference of the series, d is a constant, m is the number of lags and t is the time

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022

trend. If the null hypothesis that a = 0 can be rejected, then the time series variable is stationary and the series is ‘‘integrated of order zero’’ or ‘‘I(0)’’. If the null hypothesis cannot be rejected, then the variable is nonstationary. In this case, we differenced the series of the variable and repeated the ADF test to determine the order of integration. For example, if a variable became stationary after the first difference, then it was ‘‘integrated of order one’’ or ‘‘I(1)’’. 2.4. Johansen’s cointegration test Johansen’s model, extended by Johansen and Juselius [26], was chosen for our cointegration test. This method applies the maximum likelihood procedure that is appropriate in a multivariate framework analysis. By the Granger representation theorem, the general form of VECM is given by:

The coefficients of the normalized cointegrating vector can be used as indicators of the actual long-run relationship between A (or Y) and ICT. The residual of the cointegrating vector becomes the error correction term that is later used in the error correction model. 2.5. Granger causality test In the absence of a long-run relationship (i.e., the variables are not cointegrated), two variables may still be causally related in the short run. If the series are found to be non-stationary in the unit root tests, then differencing has to be performed to make each series stationary [21]. Corresponding to the stationary time series data of A (or Y) and ICT, the causality test involves estimating the following vector autoregression (VAR) models:

Dyt ¼ G0 þ G1 Dyt
1 þ . . . þ Gp
1 Dyt
pþ1 þ Pyt
p þ et

(6)

ltrace ðrÞ ¼
T

n X

Zt ¼

n X

ai Zt
i þ

i¼1

D is a difference operator, G0 is an (n  1) intercept vector, G1, G2, . . . Gp
1, P are (n  n) matrices; yt is an (n  1) vector. By construction, P has rank r and can be decomposed as P = ab. The elements of a are known as speed of adjustment parameters. b is a (p  r) matrix of cointegrating vectors, the long-run coefficients in VECM. The Johansen procedure focuses on the rank of matrix P, which determines the number of distinct cointegrating vectors. The procedure involves two-test statistics for cointegration. The first, known as the trace test, is based on the trace of the stochastic matrix, and tests the hypothesis that there are at most r cointegrating vectors: lnð1
li Þ

(7)

i¼rþ1

where T is the number of observations, r is the number of cointegration vectors,
li is the eigenvalue obtained from the estimated P matrix, and i = r + 1, . . ., n. The second test is called the maximal eigenvalue test, which tests the hypothesis that there are r + 1 cointegrating vectors: n X lmax ðr; r þ 1Þ ¼
T lnð1
liþ1 Þ (8) i¼rþ1

1013

ICTt ¼

p X i¼1

l X

bi ICTt
i þ

i¼1

di ICTt
i þ

m X

ci Xt
i þ ut

(9)

i¼1 q X

ei Zt
i þ

i¼1

r X

fi Xt
i þ vt

i¼1

(10) Z is either Y or A. For the analysis of the relationship between Y and ICT, we needed additional variables, such as labor or capital. X stands for these variables. For the relationship between A and ICT, we did not need them. We selected the lag structure of the model based on Akaike Information Criteria (AIC), as reported by Eviews 4.1. It followed that, if the computed F-value exceeds the critical F-value at the chosen level of significance, the null hypothesis was rejected. i.e., ICT does indeed affect Y (or A). For the Granger causality test of Y ! ICT (or A ! ICT), the same procedure applied. 2.6. The error correction model When two variables are cointegrated, VECM has to be applied before the Granger causality test can be implemented. Applying VECM, which is a VAR model in first differences with the addition of a vector of cointegrating residuals, results in the pairwise VEC Granger causality test, which contrasts with the

1014

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022

standard Granger causality test of Eq. (9): Zt ¼ g 1 ECTt
1 þ

n X

ai Zt
i þ

l X

i¼1

þ

m X

bi ICTt
i

i¼1

ci Xt
i þ ut

(11)

i¼1

ICTt ¼ g 2 ECTt
1 þ

p q X X di ICTt
i þ ei Zt
i i¼1

þ

r X

fi Xt
i þ vt

i¼1

(12)

i¼1

where ECT is the error-correction term, which is the residual of the cointegration equation that explains the short-run disequilibrium among the variables. 2.7. Impulse response function When A and ICT are cointegrated, by generating the impulse response function of A and ICT, the response of Solow’s Residual can be traced back to the shock in ICT inputs. This determines whether the impact is causing a positive or negative temporary jump or having a long-run effect, and also predicts the responses from ICT investments to Solow’s Residual in the future.

3. The data We collected data on 20 countries; it consisted of annual time series data for four main variables: GDP (Y), capital (K), labor (L), and ICT investment. The data was from both developed and developing countries. The developing countries were China, India, Indonesia, Malaysia, and the Philippines. The developed countries/newly-industrialized economies (NIEs) were Australia, Austria, Canada, Denmark, Finland, France, Ireland, Italy, Japan, (South) Korea, Singapore, Spain, Sweden, United Kingdom, and the United States. The time series data ranged from 1980 to 2000. But for several countries, capital data was not available for certain years, mostly for 1980 and 1981. Nevertheless, we made sure that each country had at least 16 data points. GDP (in constant local currency) and labor

data were directly obtained from World Development Indicators (WDI). However, WDI only provided values for gross fixed capital formation (in constant local currency), which is actually gross domestic fixed investment and not the capital stocks that should be used to calculate Solow’s Residual. Thus, an adjustment to the capital variable data was crucial. In our data, non-high-tech capital stocks were, on average, depreciated with a 10% rate [capital income taxes and economic performance], and the individual annual values were derived as {K(t) = K(t
1) + I(t)
depreciation(t
1)}. Data series for the above-mentioned variables were then used to generate Solow’s Residual, A. Also a was set to 0.3 and 0.4 for the comparison of empirical results, denoted by A0.3 and A0.4, respectively. There were no sources that provided ICT investment data of more than 10 years for countries in our sample. Therefore, we use International Telecommunication Union’s (ITU’s) annual telecommunications investments as a proxy for the ICT variable for the period 1980–2000. Summary statistics of the variables are presented in Table 1.

4. Empirical results and discussion The unit root test results of Table 2, show that the majority of the variables were non-stationary, and that 12 of the countries were not cointegrated. This validated the concern that simple ordinary least square (OLS) analysis might produce spurious regression results, and reinforced the need for time series analysis tools in this area of research. On the other hand, we discovered that incorporating a = 0.3 and a = 0.4 in the derivation of Solow’s Residual yielded very similar empirical results in all the statistical tests. There were slight differences in the results of cointegration and the Granger causality tests for Y and ICT relative to that of A and ICT, but there were common results from these two tests for most countries in the sample. In general, the results showed that developing countries did not derive productivity improvements from their ICT investments while developed countries and NIEs experienced growth with ICT. Logical explanations can be furnished for this finding, except for Denmark, Finland, France, India, Sweden, and the

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022

1015

Table 1 Descriptive statistics (mean of each variable) Country

GDP (billion dollars)

Capital (billion dollars)

Labor (million)

ICT (million dollars)

Australia Austria Canada China Denmark Finland France India Indonesia Ireland Italy Japan Korea Malaysia The Philippines Singapore Spain Sweden UK US

328 211 524 493 167 126 1430 283 145 57 1000 4640 354 64 67 60 529 225 1020 6530

75.4 47.5 96.9 161.0 31.3 26.8 276.0 61.7 36.2 11.1 197.0 1300.0 111.0 20.6 14.5 21.1 114.0 38.0 172.0 1150.0

8.4 3.6 659 659 2.9 2.5 25 367 79 1.4 24.3 63.7 19.8 7.2 24.9 1.6 15.8 4.6 28.4 126

2120 1100 2990 5820 530 557 5110 1510 644 311 5230 19000 3340 782 535 297 2710 987 6110 21900

198.1

106.2

Average

912.7

4079.2

Note: Monetary terms are in 1995 constant US dollars. Table 2 Results of unit root and cointegration tests Degree of integration Country Australia Austria Canada China Denmark Finland France India Indonesia Ireland Italy Japan Korea Malaysia The Philippines Singapore Spain Sweden UK US

ln A0.3 I(1) I(1) I(1) I(0) I(0) I(2) I(0) I(0) I(1) I(2) I(0) I(2) I(1) I(1) I(1) I(1) I(1) I(1) I(2) I(1)

ln A0.4 I(1) I(1) I(1) I(0) I(0) I(2) I(0) I(0) I(1) I(2) I(0) I(2) I(1) I(1) I(1) I(1) I(1) I(1) I(2) I(1)

Cointegration ln ICT I(1) I(1) I(1) I(1) I(1) I(1) I(0) I(1) I(1) I(2) I(0) I(2) I(1) I(2) I(0) I(1) I(2) I(2) I(2) I(1)

Note: Vacant entries are attributed to data unavailability. * Significance of test statistics at 10%error rate. ** Significance of test statistics at 5% error rate. *** Significance of test statistics at 1% error rate.

ln Y I(2) I(1) I(0) I(1) I(1) I(2) I(1) I(1) I(2) I(2) I(1) I(2) I(1) I(2) – I(2) – I(1) I(1) I(0)

ln K I(2) I(1) I(2) I(2) I(2) I(0) I(0) I(1) I(2) I(2) I(1) I(1) I(1) I(1) – I(0) – I(2) I(2) I(1)

ln L I(2) I(2) I(2) I(2) I(0) I(2) I(2) I(2) I(2) I(2) I(2) I(2) I(2) I(0) – I(2) – I(2) I(1) I(2)

ln A0.3 ln ICT **

Yes Yes** Yes** No No No No No Yes* Yes* No No Yes* No No Yes** No No No Yes***

ln A0.4 ln ICT **

Yes Yes** Yes** No No No No No Yes** Yes** No No Yes** No No Yes** No No No Yes***

ln Y ln ICT Yes** No No No No Yes** No No Yes** Yes** No No No Yes** Yes** – No No No

1016

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022

Table 3 Results of Granger causality test Country Developed Group1: Group1: Group1: Group1: Group1: Group2: Group2: Group3: Group3: Group3: Group4: Group4: Group4: Group4: Group4:

country/NIE Australia Austria Canada Singapore US Ireland Korea Italy Japan Spain Denmark Finland France India UK

Developing country China Indonesia Malaysia The Philippines

Long-run equilibrium A and ICT

Short-run relationship A and ICT

Long-run equilibrium Y and ICT

Short-run relationship Y and ICT

Yes Yes Yes Yes Yes Yes Yes No No No No No No No No

ICT ! Aa ICT ! Aa ICT ! Aa ICT ! Aa ICT ! Aa No causalitya No causalitya ICT ! A ICT ! A ICT ! A No No No No No

Yes No No Yes No Yes No No No – No Yes No No No

ICT $ GDPa ICT ! GDP – No GDP ! ICT No Causalitya ICT ! GDP No No – GDP ! ICT No causalitya No No No

No Yes No No

A ! ICT A ! ICTa A ! ICT A ! ICT

No Yes Yes –

No No GDP ! ICTa –

Note: Vacant entries are attributed to data unavailability. a VEC Granger causality tests are conducted for these cointegrated cases. Standard Granger causality tests are applicable for the other cases since the series are NOT cointegrated.

United Kingdom. A summary of the principal findings is given in Table 3. 4.1. Developed countries and newly industrialized economies 4.1.1. Group 1: long-run and short-run interrelation of economic growth with ICT – Australia, Austria, Canada, Singapore, United States These five economies presented an optimistic view of the value of ICT investment; the results may be attributed to the results from the cointegration and causality tests (they point from ICT to A). Moreover, many of the normalized cointegrating coefficients implied a positive relationship between ICT investments and productivity. Australia’s impulse response functions in Fig. 1 showed that it experienced a positive improvement due to ICT innovation and that it did not diminish with time – a trend similar to Singapore’s. The Singapore government is known to have deployed ICT as an

economic tool to increase competitiveness in the international market, implementing an ICT-led development strategy [31]. The notable difference was that the initial negative impact of ICT on productivity suggested that high-tech investments might have been counter-productive at the start. This initial short period of negative responses turned out to be worthwhile: the redesign of job practices, the training that workers received, and the realignment of work scope and structures served to bring in a sustained period of high returns. Tradenet (currently IES) was an example that may have been useful in explaining the initial negative impact. In contrast, Austria’s impulse response function showed that its ICT investments had an initial positive impact on its economy but that it died off eventually. Canada presented yet another picture: ICT innovation did not affect its productivity in the beginning but the function then took off and fluctuated from its equilibrium position after some time. Similar to Singapore, Canada apparently underwent vigorous

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022

1017

Fig. 1. Impulse response function.

changes and infrastructure-building before the payoffs from ICT were seen. Numerous early studies cited the negative contributions of ICT in the US, but our statistics indicated differently. Cointegration test results pointed to a

long-term equilibrium between ICT and productivity in the US; the statistics also suggested a causality flow from ICT to A. Of more interest, perhaps, was its response function, which fluctuated irregularly with the passing of time. The United States appeared to

1018

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022

have experienced more haphazard movement due to ICT innovation. The variations could even dip below equilibrium, giving the perception that negative impact could have occurred during some adaptability periods of new ICT investment or even due to inappropriate management of the investment. 4.2. Group 2: long-run equilibrium of national productivity with ICT – Ireland and (South) Korea Referring to the diagrams for Ireland, it can be seen that there was no vigorous response of A to ICT innovation in any short-term period. There was nevertheless a gently sloping positive gradient, suggesting a long-run equilibrium of the two variables. This implies that, in the Irish economy, ICT investments may not have provided any visible jump in the country’s productivity but delivered a significant payoff over a longer time period when complementary investments were built up and pooled. This finding is in line with Ireland’s policy of ‘‘Industrialization by Invitation’’ [33], by which the Irish government sought to attract foreign investors with financial and tax-based incentives, excellent infrastructure and a computerliterate labor force. A longer time frame was needed for the ICT-led development policy to yield returns in the form of significant productivity gains and improvements as it took time for ICT usage to be efficiently assimilated into the work force and production lines. Also, the significant normalized cointegrating coefficients displayed an optimistic outlook – ICT did indeed contribute positively to productivity. Similarly, the Korean economy (whose past growth was based on traditional manufacturing industries) had shifted from cost- to efficiency-oriented policies in the past two decades. The Korean government made consistent efforts to provide the necessary infrastructure and an ICT-friendly business environment (for example, Cyber21, which was initiated to create an information-based society). As a result, the government established an extremely successful telecommunication infrastructure and managed to stimulate the country’s software development and e-commerce industries. Thus, the response functions for Ireland and Korea exhibited similar trends, implying that the effects of ICT on such countries had to unfold over a longer time period, and that positive effects were not captured and

conceptualized in a short time frame; thus, there was a cointegrated relationship but no apparent causality. 4.3. Group 3: short-run association of ICT with economic growth – Italy, Japan, Spain Italy and Japan are well known for their aggressive ICT policies and high-tech exports. The statistical results for the three economies, however, showed no apparent long-term association between the countries’ productivity and ICT investment, but there was a short-term relationship. This agrees with previous studies. Van Ark et al. [57], for example, indicated that there was a large variation among various countries in terms of their contribution of ICT capital to productivity growth between 1995 and 2000: the US, Australia, and Finland received the largest boost; Japan and Italy a much smaller one, and Spain and Portugal the least. The absence of a cointegrated relationship between ICT and A suggested that Japan, one of the world’s largest exporter of technological products, failed to reap long-term benefits from its high-tech investment. But maybe Japan’s NTT-DoCoMo, one of the most comprehensive telecommunication systems in the world, and the Information Technology Promotion Agency only delivered productivity growth for a short time. Could similar failures also be dogging Italy? Telecom Italia, with its fully digitized circuit-switched network serving approximately 26 million lines, and Italy’s status as the fifth largest world market in both television and telecommunications services do not appear to be furnishing long-term growth benefits for the Italian economy. If the empirical results of this paper are correct, there must be some areas of inadequacy in the Italian and Japanese ICT strategies to ensure a longer and sustained period of growth. A Japan Economic Update report pointed out that ICT investments could, on balance, have a negative impact on economic growth in the short run [24]. Also a recent OECD publication reported ICT investment accounted for between 0.3 and 0.8% point of growth in GDP per capita for the period 1995–2001 [43]. The US, Canada, Netherlands, and Australia had the largest boost; Japan and the UK a more modest ones, and Italy a much smaller one. Some OECD countries substantially increased labor productivity growth from the 1980s to the 1990s,

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022

including Denmark, Ireland, and Sweden. However, some of the larger OECD economies, such as France, Germany, and Italy, experienced a substantial decline in labor productivity growth in the same period. According to conventional wisdom, some European countries (including Italy) and Japan have been slow to invest in ICT, partly due to rigidities in the labor and product markets that reduce returns on such investments. During the second half of the 1990s, most OECD countries, except Italy, and Japan enjoyed faster productivity growth. Japan’s GDP growth was slower, because of a fall in non-ICT investment and weak employment due to deficient demand. In a similar vein, the European Information Technology Observatory indicated in 2000 that Italy’s ICT expenditure (as a percentage of its GDP) reached 5.5%, while in 1997 it was 3.9% against the European average of 6.3% [20]. Despite the importance of ICT, noticeable differences in the diffusion of ICT continued between OECD countries. New OECD data showed that the US, Canada, New Zealand, Australia, the Nordic countries, and the Netherlands typically had the highest rates of ICT diffusion. Many other OECD countries lagged in the diffusion of ICT, leaving scope for improvement. In their study on the impact of ICT on multifactor productivity (MFP) growth in Italy from 1996 to 1999 [42], Milana and Zeli found that it is positively affected by increase in intensity of ICT use. Despite the positive impact, the overall performance of Italy in the period was characterized by negative MFP growth, which they attributed to the limited scale of investment in ICT and the costs of adjusting to the new technology. It may also be assumed that ICT has not been a main contributor to growth in the Spanish economy, because Spain has been lagging in technological advancements, with low Internet penetration and a telecommunication system that was behind other European countries [58].

5. Developing countries 5.1. Causality from national productivity to ICT: China, Indonesia, Malaysia, the Philippines This group consisted of developing countries, for whom, in the past two decades, the opening of political

1019

doors led to tremendous growth and international competition. China, Indonesia, Malaysia, and the Philippines may indeed have undergone increased productivity, but it was not possible ascertained what specific factors had contributed to the improvement. A developing economy may be investing heavily in hightech capital because of its (mistaken) perception that jumping onto the ICT bandwagon is the quickest way to economic prosperity. Recent success and increased productivity may reinforce this idea, discouraging the country from carefully examining whether it is ICT investments that were contributing to prosperity. It is possible that developing nations have poured financial resources into ICT though higher productivity should have been attributed to other factors, such as the maturation of the work force, a rise in education level of the population, and the influx of foreign skilled workers. Do the elevated level of economic wealth and higher wages in a developing country lead to large ICT investments that cannot be justified? Ideally, growth convergence should occur, with ICT boosting the country’s productivity and high-tech capital serving as tools to allow it to catch up with the economic pace of developed nations. But this convergence cannot take place if the causality in developed countries is from ICT-to-growth while the causality in developing countries is from growth-toICT. Thus ICT should not be widening the gap between developed and developing nations and exacerbating the differences between the wealthy and the poor. Consequently, it is necessary for developing countries to re-examine their ICT policies, to affirm whether the country’s growth has been wrongly ascribed to inappropriate ICT investments. 5.2. Lesson learnt: ICT-complementary factors ICT-complementary factors are probably the reason for disparity in payoffs from ICT capital for developing and developed countries: the experience curves which must first be constructed by countries before ICT investments become productive. According to previous studies, these crucial components consist of specialized information infrastructure, human resources, research and development (R&D), low tariffs on computer imports, telecommunication liberalization, and adaptive business models, and reorganization of manual processes.

1020

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022

These fundamental factors must, to a large extent, be established by government policies and development strategies. From our statistical findings, it seems that Ireland and Singapore experienced long-term equilibrium of national productivity because of ICT investments. This is an acknowledgement that implementing policies that remove restrictions on imports and technology transfer attract multinational computer companies to subcontract with local firms, and promote both ICT use and its production; e.g., consider the example of the Infocomm Local Industry Upgrading Program, an effort by the Singapore Infocomm Development Authority to encourage the exchange of ideas and industry knowledge between multinational corporations and local small and medium enterprises [11]. Policies promoting the favorable conditions for ICT investments can be seen in other countries that established a long-term relationship between ICT and national growth. Notably, none of these countries has used protectionism in its ICT policies. On the other hand, Japan, which used to require foreign companies to license their technology to Japanese companies in return for access to the market (thus limiting ICT usage to promote production), was only found to have a shortterm causality from ICT to economic growth. It can be concluded that countries that have invested in ICT over a long period of time and accumulated a substantial installed base and complementary investments in telecommunications and human resources are better able to achieve positive and significant returns to ICT. This does not mean that developing countries should ignore ICT investments. On the contrary, it is possible that some threshold of ICT capital must be reached before their effect becomes measurable. The implication of our results is that developing countries can benefit by promoting ICT-use and creating the environmental conditions needed to support the effective use of ICT.

6. Limitations and conclusion ICT and productivity have been considered by many scholars. However, our study is among the first to approach the topic employing Solow’s Residual instead of tangible outputs and time series analysis tools at the country level. Consistent with previous

findings [15], we discovered that ICT investments have been contributing to improvement in the national productivity of several developed countries and newly industrialized economies (NIEs), but not of developing nations. To allow for the limitations in the data sets, the material was obtained from the ‘‘best’’ available – even if only one set of ICT data series provided positive evidence of the relationship, the results from the other ICT data series were discarded. Even though there is no agreement on the minimum number of periods required in time series statistics tests, scholars have tacitly agreed that cointegration testing on small samples still leaves something to be desired [4]. Thus the data period analyzed here may be insufficient to capture long-term effects. In addition, inferences from the Granger causality tests may be unfair for the relatively small sample set of annual values for A and ICT, which could ignore short-run effects. The values of the ICT variable are problematic also: there are no sources that provide these values over a period of more than 10 years. Consequently, telecommunications investments had to be used as a proxy, even though they could prove inadequate in reflecting the full effect of ICT in some countries’ productivity. Nonetheless, we believe that the study has contributed to development literature and provided useful groundwork for time series analyses in ICT and productivity research. In recent years, studies have recommended ICT production-close-to-use, which promotes the interaction between ICT producers and users, and facilitates the development of software and information services. According to Kraemer and Dedrick [32], policies could include: the promotion of small business ICT use, provision of financial support to ICT companies, and encouragement of partnerships between local firms and multinationals. Indeed, countrywide diffusion of technology could deliver increasing returns to investments and competitiveness, given today’s global markets and pervasiveness of the Internet and ecommerce. From the consistency between our findings and previous studies, countries may need to put in place ICT-complementary factors in tandem with their ICT investments. The proven existence of a long-term relationship between ICT and growth in several countries would encourage belief that the phenomenon is sustainable, and that nations can rely on ICT

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022

investments to attain economic expansion, rather than resort to an increase in the traditional production inputs of human labor and capital. References [1] P. Alpar, M. Kim, A microeconomic approach to the measurement of information technology value, Journal of Management Information Systems 7(2), 1991, pp. 55–69. [2] M.N. Baily, What has happened to productivity growth? Science (234) 1986. [3] D. Ben-David, A. Rahman, Technological convergence and International Trade, No. 1359 in CEPR Discussion Papers, 1996. [4] A. Blangievicz, W.W. Charmeza, Cointegration in small samples: empirical percentiles, drifting moments, and customized testing, Oxford Bulletin of Economics and Statistics (52) 1990, pp. 303–315. [5] K. Bruland, Technological revolutions, innovation systems and convergence from a historical perspective, ESST Report to the CONVERGE project, 2001. [6] E. Brynjolfsson, The Productivity Paradox of Information Technology: Review and Assessment, Communications of the ACM, December 1993. [7] E. Brynjolfsson, Technology’s true payoff, Informationweek 1994, pp. 34–36. [8] E. Brynjolfsson, L. Hitt, Paradox lost? Firm-level evidence on the returns to information systems spending Management Science 42(4), 1996, pp. 541–558. [9] E. Brynjolfsson, S. Yang, Information technology and productivity: a review of the literature, Advances in Computers (43) 1996, pp. 179–214. [10] Y.E. Chan, IT value: the great divide between qualitative and quantitative and individual and organizational measures, Journal of Management Information Systems 16(4), 2000, pp. 225– 261. [11] Computer Times, SMEs Get A Helping Hand, Computer Times 18, January 1, 2003. [12] J. Dedrick, V. Gurbaxani, K.L. Kraemer, Information technology and economic performance: a critical review of the empirical evidence, ACM Computing Surveys 35(1), 2003, pp. 1–28, March. [13] S. Devaraj, R. Kohli, Information technology payoff in the healthcare industry: a longitudinal study, Journal of Management Information Systems 16(4), 2000, pp. 41–67. [14] S. Dewan, K.L. Kraemer, International dimensions of the productivity paradox, Communications of the ACM 41(8), 1998, pp. 56–62. [15] S. Dewan, K.L. Kraemer, Information Technology and Productivity: Evidence from Country-Level Data, Center for Research on Information Technology, University of California, Irvine, 2000, December. [16] S. Dewan, C. Min, The substitution of information technology for other factors of production: a firm level analysis, Management Science 43(12), 1997, pp. 1660–1675.

1021

[17] D.A. Dickey, W.A. Fuller, Distribution of the estimations for autoregressive time series with a unit root, Journal of the American Statistical Association (74) 1979, pp. 427– 431. [18] D.A. Dicky, W.A. Fuller, Likelihood ratio statistics for autoregressive time series with a unit root, Econometrica (49) 1981, pp. 1057–1072. [19] W.E. Diewert, A.M. Smith, Productivity measurement for a distribution firm, National Bureau of Economic Research, Working Paper No. 4812, July 1994. [20] EITO, European Information Technology Observatory, Frankfort, 2001. [21] C.W.J. Granger, Investigation causal relations by econometric models and cross-spectral method, Econometrica (37) 1969, pp. 424–438. [22] Z. Griliches, R&D and Productivity: The Econometric Evidence, The University of Chicago Press, Chicago, IL, 1998. [23] V. Gurbaxani, N. Melville, K.L. Kraemer, Disaggregating the return of investment to IT Capital, in: Proceedings of the International Conference on Information Systems, Helsinki, Finland, 1998. [24] Japan Economic Update, July 8/11, 2000. [25] S. Johansen, Statistical analysis of co-integrating vectors, Journal of Economic Dynamics and Control (12) 1988, pp. 231–254. [26] S. Johansen, K. Juselius, Maximum likelihood estimation and inference on co-integration with applications for the demand for money, Oxford Bulletin of Economics and Statistics (52) 1990, pp. 169–210. [27] D.W. Jorgenson, K.J. Stiroh, Raising the Speed Limit: US Economic Growth in the Information Age, Brookings Papers on Economic Activity, 2000, 125–211. [28] H. Kivijarvi, T. Saarinen, Investment in information systems and the financial performance of the firm, Information and Management (28) 1995, pp. 143–163. [29] R. Kohli, S. Devaraj, Measuring information technology payoff: a meta-analysis of structural variables in firm level empirical research, Information Systems Research 14(2), 2003, pp. 127–145. [30] K.L. Kraemer, J. Dedrick, Payoffs from investment in information technology: lessons from the ASIA-PACIFIC region, World Development 22(12), 1994, pp. 1921–1931. [31] K.L. Kraemer, J. Dedrick, IT-led development in Singapore: from Winchester Island to Intelligent Island, Working Paper No. ITR-123, Center for Research on Information Technology and Organizations, Graduate School of Management, University of California, 1997. [32] K.L. Kraemer, J. Dedrick, Information technology and productivity: results and policy implications of cross-country studies, Working paper No. PAC-144, Center for Research on Information Technology and Organizations, Graduate school of Management, University of California, February 1999. [33] K.L. Kraemer, P. Tallon, The Impact of Technology on Ireland’s Economic Growth and Development: Lessons for Developing Countries, Center for Research on Information

1022

[34] [35]

[36]

[37]

[38]

[39] [40]

[41]

[42]

[43]

[44]

[45] [46]

[47]

[48]

[49] [50]

S.-Y.T. Lee et al. / Information & Management 42 (2005) 1009–1022 Technology and Organizations, Graduate School of Management, University of California, 1999. C. Lee, Modeling the business value of information technology, Information and Management (39) 2001, pp. 191–210. B. Lee, A. Barua, An integrated assessment of productivity and efficiency impacts of information technology investments: old data, new analysis and evidence, Journal of Productivity Analysis 12(1), 1999, pp. 21–43. F. Lichtenberg, The output contributions of computer equipment and personnel: a firm level analysis, Economics of Innovation and new Technology (3) 1995, pp. 201–207. H.C. Lucas, The business value of information technology: a historical perspective and thoughts for future research, in: R. Banker, R. Kauffman, M.A. Mahmood (Eds.), Strategic Information Technology Management, Idea Group, Harrisburg, PA, 1993, pp. 359–374. G.W. Loveman, An assessment of the productivity impact of information technology, Working paper No. 88-054, Management in the 1990s Project, MIT Sloan School, Cambridge, MA, 1988. G.S. Maddala, Introduction to Econometrics, 2nd ed., Prentice-Hall, 1992. M.A. Mahmood, G.J. Mann, Special issue: impacts of information technology investment on organizational performance, Journal of Management Information Systems 16(4), 2000, pp. 3–10. N.G. Mankiw, D. Romer, D.N. Weil, A contribution to the empirics of economic growth, Quarterly Journal of Economics (107) 1992 May, pp. 407–437. C. Milana, A. Zeli, The Contribution of ICT to Production Efficiency in Italy: Firm-level Evidence using Data Envelopment Analysis and Econometric Estimations in the Economic Impact of ICT; Measurement, Evidence and Implications, OECD, 2004. OECD, Information and Communications Technologies and Economic Growth; Evidence from OECD countries, industries and firms, 2003. K.M. Osei-Bryson, M. Ko, Exploring the relationship between information technology investments and firm performance using regression splines analysis, Information and Management 42(1), 2004, pp. 1–13. J. Park, H. Shin, S.K. Shin, Information Technology Investment and National Productivity Growth, ICIS, 2003. D.J. Parsons, C.C. Gotlieb, M. Denny, Productivity and computers in Canadian banking, Working Paper No. 9012, University of Toronto Department of Economics, June 1990. K. Peffers, B.L. Dos Sontos, Performance effects of innovative IT applications over time, IEEE Transactions on Engineering Management 43(4), 1996, pp. 381–392. M. Pohjola, Information Technology and economic growth: a cross-country analysis, Working Paper No. 173, The United Nations University, January 2000. S. Roach, American’s Technology Dilemma: A Profile of the Information Economy, Morgan Stanley, New York, 1987. S. Roach, White-Collar Productivity: A Glimmer of Hope? Morgan Stanley, New York, 1988.

[51] K. Schneider, Services hurt by technology: productivity declining, New York Times, 1987. [52] B.M. Shao, W.T. Lin, Technical efficiency analysis of information technology investments: a two-stage empirical investigation, Information and Management 39(5), 2002, pp. 391– 401. [53] R.M. Solow, We’d better watch out, New York Times Book Review, July 12, 1987, 36. [54] A. Smith, Wealth of Nations: An Inquiry into the Nature and Causes, 1776 [55] P.A. Strassmann, Information Payoff: The Transformation of Work in the Electronic Age, Free Press, New York, 1985. [56] T. Stratopoulos, B. Dehning, Does successful investment in information technology solve the productivity paradox? Information and Management 38(2), 2000, pp. 103–117. [57] B. Van Ark, J. Melka, N. Mulder, M. Timmer, G. Ypma, ICT investment and growth accounts for the European Union, 1980–2000, Paper for DG ECFIN, Brussels, June 2002. [58] J. Ward, J. Peppard, Strategic Planning for Information Systems, 3rd ed., Wiley Series in Information Systems, 2002 June. Sang-Yong Tom Lee is currently an assistant professor in College of Information and Communications, Hanyang University in Seoul, Korea. He received a PhD degree from Texas A&M University in 1999 and taught at Department of Information Systems, National University of Singapore. His research interests are economics of information systems, online information privacy, IT and growth, and value of IT. He is trying to build a bridge between theory and applied work in these areas. Roghieh Gholami is a PhD candidate in department of Information Systems, National University of Singapore. Her doctoral dissertation focuses on information and communication technology and economic growth. Her current research interests are information technology and productivity, information technology adoption, broadband technology, the deployment, and IT in developing countries. Tan Yit Tong obtained his BSc (Hons) from Department of Information Systems, National University of Singapore in June 2003, and was engaged as a research assistant for Dr Sang-Yong Tom Lee from Jan-June 2003. Over the same period, he was also appointed as the teaching assistant for economics of ebusiness. He is currently with Singapore Airlines, Singapore Sales, as Corporate Accounts Manager.