Industry Determinants of

(1)

Industry Determinants of

ICT

Investment Levels in Western Europe

By: Arnold Masselink1

ABSTRACT

Many papers about the impact of Information and Communication Technology (ICT) adoption on growth have been provided on the United States. As to Europe, insight in reasons for ICT investment is limited, partially due to scarcity of disaggregated data. I utilise recently developed databases by the

EUKLEMS consortium. These data sources are analysed with panel techniques (random-effects and Arrelano-Bond) and offer explanations about driving forces behind ICT investment decisions for three industries across The Netherlands, Germany and the UK. ICT investment in the rubber and plastics industry is fairly constant, just like for wholesale trade and commissions trade the consumption of ICT

capital can closely approximate new investment in ICT. Dynamic panel analysis on post and telecommunications sector characteristics shows that company age, market concentration, labour intensity and low-skilled labour as significant factors. This implies that mostly the oldest companies in this sector use ICT investment as a way of competing: by decreasing low-skilled labour need as a result of automation, labour intensity is lowered to reduce costs. ICT investment across industries differs from simply a fairly constant development over time to a complex mix of various determinants including the interaction between labour and ICT, the degree of competition and the company age for the more ICT dependant sector included in the study (post and telecommunications). Thus a one-variable proxy as R&D investment is too generic as an indicator of ICT investment levels.

JEL classifications: E01, E22, F21, M15, O11, O33, O47

Keywords: ICT investment, determinants, dynamic panel analysis, industry level

The University of Groningen, 27-06-2010

Written under the supervision of: dr. W. Westerman

1

Arnold Masselink (s1737996), a_masselink@hotmail.com

The helpful comments of Mathijs Bouman, Bert Scholtens, Marcel Timmer and Gelijn Werner are gratefully acknowledged. This paper was written while the author was finishing the International Business &

(2)

TABLE OF CONTENTS 1.INTRODUCTION... 3 2.CONCEPTUAL FRAMEWORK... 6 2.1 FRAMEWORK FOUNDATION... 7 2.2 HYPOTHESES... 8 3.DATA DESCRIPTION... 10 3.1 OPERATIONAL VARIABLES... 11 3.2 DESCRIPTIVE STATISTICS... 17 3.3 CHOICE OF COUNTRIES... 19 3.4 CHOICE OF INDUSTRIES... 20 3.5 TIME PERIOD... 21 4.METHODOLOGY... 21 5.RESULTS... 24 5.1 RANDOM EFFECTS... 24 5.2 ARELLANO-BOND... 28 6.ANALYSIS... 30 7.CONCLUSIONS... 33 8.DISCUSSION... 36 REFERENCES... 38

APPENDIX A–DATA THE NETHERLANDS... 41

APPENDIX B–DATA GERMANY... 43

(3)

1. INTRODUCTION

Not many research studies about the determinants of ICT investment have been conducted. This research tries to fill the gap by testing various macroeconomic variables and industry level factors on their correlations with ICT investment. Especially, about the how the different the potential determinants of ICT investment are cross industries (both service and production segments), there are not many papers available.

In this exploratory study the author is interested in the influence of the burst of the dot-com bubble on ICT investment for instance. The more realistic view on potential success of newcoming

ICT companies (in a broad sense: hardware, software and telecommunications as their main products and services) after the burst led to decreased prices of ICT companies’ shares of stock, but did ICT

products also become less attractive to invest in? The question is how this major stock market crash for the ICT segment of the stock exchanges was perceived across the chosen industries, which have a contrasting core-business and ICT dependency. Some industries suffer more from tight funds and some industries particularly do not feel the need to invest in ICT when the cash flows are strained and under pressure. This paper tries to determine the characteristics for the industries wherein the companies are interested in improving their ICT infrastructure. The study considers macro-economic trends and sector/industry factors that are potentially relevant when deciding about ICT investments.

The focus of this study is on the determinants within the industries and to find out if ICT

investment occurs for fundamentally different reasons across industries. Thus, it is assumed one-variable proxies for ICT investment are not the most precise indicators about the ICT investment trend throughout the years. Managerial, firm-specific, industry-specific, country-specific and time-period specific characteristics could all play their part. To achieve study feasibility and to adhere to the focus on long-term development explanation, a fairly stable long-term period is chosen (without the big financial crisis we are still in the aftermath of). Only West-European countries are studied, narrowed down to mostly factors at the industry level.

The study entails an extensive dataset collected from various sources, with the well-appreciated but not commonly used EUKLEMS database as the main source.

Purpose

(4)

Research question

“How do macro-economic and especially industry- level variables correlate with ICT investment within the rubber and plastics industry, the wholesale trade and commissions trade industry, as well as the post and telecommunications sector in Germany, the Netherlands and the United Kingdom? What are the relative differences between industries in investment level and exposure to these determinants?”

The main research question can be divided into several sub-questions, which logically come together to answer the main research question. First of all, there is the question of cyclical vs. anti-cyclical investment on the macro-economic level. There is an ongoing debate about whether ICT really contributes positively to company efficiency and/or growth. Assuming it does, ICT investment is not a big priority for many companies as the contribution of ICT to success is still not very clear to the managers of those companies. From this it follows that these managers prefer to invest in ICT when the company is doing well and there is excess cash available, as ICT investment is something they are sceptical about and they do not see as a fruitful or serendipitous move during rough times.

However, if the belief in the importance of a solid and modern ICT infrastructure is stronger, then ICT investment could be the key to improve efficiency and thereby performance during periods of a difficult business climate. To get out of a slump, ICT investment could reduce costs and streamline the business processes. Of course, there should be funds available in the first instance, but fixed and variable costs could be reduced for many years to come after the ICT invested in is effectively utilised.

The most important sub-question is related to industry data and its correlation with ICT

investment levels. Why do companies in a particular sector invest in ICT? Is it because of the sector’s characteristics as ICT asset and inventory levels, internal rates of return, a necessity to be competitive or is it just required when a high average turnover causes a lot of data and information to process?

Finally, the concluding sub-question is about the combination of factors to function as a proxy for ICT investment for the particular industries with the highest explanation power. Research and development expenditure (often published in annual reports) is one of the best proxies found so far for ICT investment (Arcangelis, Jona-Lasinio and Manzocchi, 2006). My research attempts to find a more specific proxy in other countries than Italy and the US. That would fill up the research gap somewhat, as there is only proof for the usability/validity of the R&D-proxy for non-Anglo-Saxon countries within Europe.

This study does not have the objective to find support for R&D percentage to net sales as a decent proxy; it rather attempts to find a better proxy on the industry level. For example, a combination of the independent variables as chosen may be a better proxy than just R&D levels, as

(5)

random-effects dynamic panel data analysis. The one-step Arellano-Bond method provides even less biased results than a random-effects regression does.

Summarizing the answers to the sub-questions would lead to being able to accept or reject the hypotheses about the variables tested for the chosen industries. The variables of significance and their explanation power will be compared across industries. For each industry or sector one can then pinpoint the most important factors correlating with ICT investment. As a conclusion, statements can be made about industry-specific determinants of the level of ICT investment. The main issue refers to the factors that correlate with the different ICT investment level across various non-similar industries and its development.

After this chapter, the conceptual framework is clarified in chapter 2 along with all the proposed hypotheses (for a particular industry as well as for multiple industries) well-founded by other literature. Thereby every concept (potential influence on ICT investment in basic understandable terms) is propelled for inclusion in the conceptual model. Various sources and private assumptions lead to this conceptual framework.

The third chapter deals with describing the unique dataset. In the fourth chapter it is shown how the dataset is methodologically dealt with: the formulas for panel analysis are presented and motivated. Another part of this chapter deals with the operationalisation of the variables by turning concepts into measurable variables. Furthermore, in this chapter the foundations for the timeframe of analysis and the choice of industries as well as countries are laid out more elaborately.

After the methods of statistical analysis are clarified, the findings are presented in chapter five. The findings supporting acceptation or rejection of the hypotheses stemming from the results of both models for every industry are combined in an analytical manner and are drawn into a wider context in chapter six. Concluding in chapter seven, the conceptual framework is presented along with the relationships or correlations supported by the foregoing analysis. In chapter seven, the conclusions are discussed in the light of the literature the research is based on. In the final discussion chapter it is explained which gaps for further research are still unfilled and how the results can be applied to the practise of business managers who want to invest in ICT or who want to sell ICT

(6)

2. CONCEPTUAL FRAMEWORK

In this chapter the most important building brick of the study is presented. The conceptual framework is modelled into figure 1 below and shows the scale and scope of the research, understandable even for those not that knowledgeable about the specific field of ICT or really into measurements of macro-economic activity. Furthermore, the expectations on outcomes of the research are raised.

Figure 1: initial conceptual model with expected relationships and direction of influence

The framework graphically represents the main contributions in each domain identified in the literature and serves as a 'sensitising framework' to guide empirical investigation. Specifically, the conceptual framework indicates how sector/industry characteristics potentially interact with other factors such as labour influences, previous ICT asset measurements and various macro-economic concepts to shape ICT investment. Given that causality cannot be attributed in a linear fashion within such a complex system, the focus of operationalising the framework was not on prediction but rather

ICT

investment

Country

IRR on capital

National dimension Gross ouput growth %

control

control Average turnover

Consumption of ICT assets

-

+

+/-

+

-

Constant Interest rates

+

Industry dimension

Burst of the Internet bubble

ICT capital compensation

Market concentration Average company age

age Labour intensity Low-skilled labour

(7)

on understanding what is contextually unique and why it is so (Brewster, 1999; Truss, 2002). As one can see in figure 1, one can try to predict causality based on the literature. However, eventually only correlations can be concluded about.

2.1 FRAMEWORK FOUNDATION

Even though this study has a different approach than most literature written on the topic of ICT

investment and the effects of it, in this chapter the link with prior research is identified. Most of the papers available are considering ICT investment levels as a given factor and then to try to prove an influence on growth or efficiency. However, effects may not appear until long after the initial ICT

investment, if at all visible in the numbers. For example, Matti Pohjola doesnot find any significant correlation between ICT investmentand economic growth in the period 1985–99 for a sampleof 42 countries for which ICT spending data are available (Pohjola, 2002).

About the ICT investment contribution to growth in the European Union not a great deal is known if it would not be for Bart van Ark, Johanna Melka, Nanno Mulder, Marcel Timmer and Gerard Ypma working for the Groningen Growth and Development Centre (GGDC). They have developed a unique database for the research behind their working paper “ICT Investment and Growth Accounts for the European Union, 1980-2000”. They raise the question of more industry-specific research and cross-country comparison of macro-economic influences affecting ICT implementation. This study takes a step back to focus on the potential determinants for ICT investment, partly also found within the GGDC databases.

One of the few papers that has a similar approach and is focused on Europe is as well, is written by De Arcangelis, Jona-Lasinio and Manzocchi (2006). They are looking into many sectors, but only for their country of residence. They conclude that the determinants and dynamics of ICT

investment can be approximated by taking the level of R&D investment as a proxy. Although the research finds a strong correlation between ICT investment and R&D expenditures, applying just one-instrument proxy is quite simplistic. It makes sense that innovation by researching and developing new ideas is often paired with ICT investment (especially in the ICT services industry itself, which is part of the sample in this Italian research), as during the years under investigation ICT was really hot due to the Internet upheaval. Therefore, ICT was quite popular for any firm in every industry to invest in (perhaps excluding the bakery at the corner of the block).

(8)

In earlier times there were measurement and definition problems, which is why the influence of substantial ICT investment was not clear. Nowadays, ICT is defined more clearly and research has shown ICT investment is facilitating company growth. A paper by Roeger (2001) supports this idea on a macro-economic level. Company growth can be measured as growth of the company’s sales, although newcomers threatening the incumbents and trying to capture market share often do not succeed in the short run. Investments in ICT need to be made when entering a market to streamline processes and improve efficiency, so there can be a narrower focus on growth of output. Potentially, sunk costs in ICT investment can be a barrier for entry.

2.2 HYPOTHESES

All the talk about ICT investment contributing to gross output growth does not consider the opposite influence. If gross output within a sector grows substantially, there must be a need to support this with additional ICT investment. This is easily explored by taking gross output growth as an independent variable instead of a dependent one. Formulated as hypothesis 1: “The ICT sector is depending on gross output growth within the respective industries whereas gross output growth fosters ICT investment”.

Several more independent variables are tested for their influence on ICT investment, based on several publications. Interest rates, for instance, are a decent predictor according to the guiding Italian study by De Arcangelis, Jona-Lasinio and Manzocchi (2006). To be able to confirm or reject their claim, hypothesis 2 is developed: “With relatively lower interest rates for a particular industry in a given year, ICT investment levels will be higher than during years of more expensive loans”.

There are many factors influencing economic growth possibilities and growth levels. A fraction of this can be attributed to the use of ICT. However, ICT is not the whole story. Thoughts about ICT became more realistic from the burst of the Internet bubble onwards (Oliner, 2000), after the ‘information age’ in the US (Jorgenson & Stiroh, 2000). For this reason the year when everyone could observe the dot-com stocks crash, was chosen as the switch-point for the dummy variable depicting the burst of the dot-com bubble. By the regressing the dummy variable hypothesis 3 can be tested: “The level of ICT investment significantly lowered after the dot-com bubble bursted suddenly”.

The other variables are chosen from the EUKLEMS databases. Research on the topic of ICT

investment has shown that effectively utilising ICT increases efficiency during the working hours (Van der Wiel, 2005). Therefore, it is hypothesised that when contribution of hours worked to value added growth is low, this is a pull-factor for ICT investment. Stated as hypothesis 4: “A low contribution of hours worked to value added growth can partially explain ICT investment levels in the year later, because ICT investments are made intentionally in an attempt to increase efficiency”.

(9)

their novelty. It is interesting to include data derived from company accounts, because it maps the companies on an extensive list of sectors and industries. Plus that for all the companies data is gathered and combined to arrive at averages for the sectors and industries.

For example, for each and every company the turnover is added up and the total is divided by the number of companies. This average turnover is expected to have an influence on ICT investment: “Net turnover divided by the number of companies active in the respective industries in the former year, positively influences ICT investment in the current year” (hypothesis 5). The inclusion of this hypothesis in the research is motivated by the Revest and Sapio (2009) claim that technology-based small firms mostly rely on internal funds for new investments. As companies with a negative profit margin would be driven out of the market, a higher average turnover should imply more internal funds to be available for investment.

Also, the age of the companies within the sectors is averaged and included in the analysis. Especially for post and telecommunications a relationship is expected. Formulated as hypothesis 6: “A rising average age of the companies in the post and telecommunications sector, influences the ICT

investment levels in a positive manner”. This is expected because the sector is ICT intensive (telecommunications is a part of Information and Communication Technology) and the older companies usually have a broader network of clients who increasingly desire ICT-supported services.

The third variable from the company accounts database verifies another Italian research by Paganetto, Becchetti, Bedoya, Andres (2000). Is ICT investment really necessary because of competitive forces or is ICT investment more interesting when competition is not that fierce and there are slack resources? More specifically, hypothesis 7 is formulated as follows: “The higher the number of competitors for the post and telecommunications industry, the more investment there is in

ICT”. Leading by technology is expected to have a positive effect on how well a company can compete.

This is in line with former research findings: “Based on studies for the US, one of the most efficient policy options seems to be creating a more competitive environment, which will cause firms to increase efficiency and to explore new ways of doing business” (Van der Wiel, 2005). This statement is related to ICT optimisation its visible effects in the Netherlands, confirming previous findings about ICT contribution to growth for the early adopters of ICT in the US (Fernald, 2004).

Another expected correlation seems rather logical when keeping the definition of ICT

investment in mind. ICT investment is about automating or digitalizing tasks that were formerly mostly labour intensive, but a part of it is only about keeping the value of ICT on the same level after depreciation charges. This is why hypothesis 8 is expected to hold: “The capital consumption of the former year affects the ICT investment levels of the current year in a positive manner”.

If current ICT capital is earning a good compensation, a lagged effect is expected on ICT

(10)

Automated processes are increasingly able to replace manual labour like data entry (Venkatraman, 1994). Then it could very well be possible that “Hours worked by low-skilled workers engaged is significantly determining ICT investment levels the next year; the more low-skilled (often manual) labour there is currently to replace, the higher the ICT investment level the next year will be” (hypothesis 10).

Hypothesis 11 is stated as follows: “The rubber and plastics industry is a distinct branch of business where the reliability on ICT assets is not that high. For this reason ICT investment levels are likely to be fairly constant over time, which is why it is expected that the constant in the equation would be significant”. Backing this statement up, Wade (2002) claims a new form of dependency on

ICT suppliers, but no direct need to innovate ICT-wise for certain sectors, indicating that ICT

implementation has reached a certain base level within some industries.

In this paper, the computer paradox (Triplett, 1999) is assumed to be solved. Thus, a positive influence of ICT utilisation on worker productivity is recognised. As ICT-supported business processes generally facilitate various tasks to be more easily accomplished, it is accepted that ICT

implementation improves growth opportunities, as more can be done in the same time with the help of new hardware and software. Thus, with the same number of employee FTEs, more products can be sold or especially more service can be delivered. This recognition of growth opportunities warrants the need to invest in more ICT as growth in the number of relationships of a company, for example, necessitates the use of bigger databases and more modern tools to keep in touch with everyone (Stiroh, 2000).

This study adds value to these articles by investigating whether potential efficiency gains by

ICT utilisation really go hand-in-hand with contribution of hours worked to value added growth (hypothesis 4) and/or at the cost of (low-skilled) employees (hypothesis 10). All the other hypotheses behind all the items and arrows in the conceptual framework are meant to test other study claims as well.

3. DATA DESCRIPTION

Concepts are nothing without a proper operationalisation in measurable variables. This chapter is an important stepping stone to justify the validity of the dataset for the quantitative research methods chosen. Considering the statistical nature of the research, the availability of data is quite essential. A good dataset to start with would up the value of the findings tremendously.

(11)

countries. The March 2008 and November 2009 release of data for various industries and sectors have been combined for specific countries in Western-Europe.

For Germany, The Netherlands and the United Kingdom there are some valuable Excel database-files available: in the basic file one can find figures about how value is added to the inputs during the whole process of creating output, data about prices (output, intermediate inputs, etc.), volumes, growth accounting variables (specific to labour, ICT or other input factors) and additional variables.

Most interesting is the capital input data, also found on the EUKLEMS website (http://www.euklems.net). This Excel-file filled with data about the capital within the various industries (capital formation, capital stock, capital compensation and capital consumption all included as real and nominal values with an index year or not) is the main source for this research. The database is under-utilised considering the effort it has taken to create it in all the countries that are members of the consortium and the database is very suitable for regression analysis. The latter is because the database has sufficient annual observations for a panel analysis, even though the data of the panels seems to be unbalanced. In other words, certain variables have more observations than other ones.

Special to this research is the combination of national factors, industry characteristics and industrial aggregates. The industrial aggregates are the most uncommonly used and therefore the most interesting potential determinants. Industry level indicators of market structure (concentration, turnover divided by number of firms and average age of firms) collected by the EUKLEMS

consortium, were derived by aggregating information from company accounts using the Amadeus database (aggregated micro-level data). This database contains information on around 120,000 companies in the EU-25. The number of firms varies considerably by country, with the highest coverage in the UK followed by Germany, France, Italy and Spain. The data covers the period from 1997 to 2006. Potential indicators like those described above allow for analyses investigating to what extent the differences in investments in ICT products and services across countries as evident from the core EUKLEMS data can be explained by differences in market structure and average market actor characteristics.

3.1 OPERATIONAL VARIABLES

(12)

Industry dimension

Real gross fixed capital formation

A critical success factor for this research is operationalising the concept of ICT investment as the dependent variable of interest. For this study, a line depicting the border between ICT and non-ICT

had to be drawn. In a very broad sense ICT can entail every invoice for the phone connection a company pays, but with the real gross fixed capital formation a narrower definition has been found. It includes software and hardware investment, but most ICT service costs are excluded by the GGDC. Of course, ICT is partly quite intangible. Not all software in utilisation every day within certain companies is legal and paid for, for example. This research is therefore limited to explain nominal values as reported yearly. Or rather, differences between nominal ICT capital figures year after year. ) Characterizing real gross capital formation from the EUKLEMS Capital Input Data database, the following can be quoted: “Capital formation may be viewed in several ways. First, since part of the capital stock wears out or loses some of its economic potential each year (depreciation), provision must be made for its replacement. A certain amount of capital formation must be allocated to depreciation in order for the economy's capital stock to remain constant. Second, after depreciation charges, additions to the capital stock are known as net investment or net capital formation” (http://www.thecanadianencyclopedia.com). The independent variable consists of investment in technology for other reasons than replacement investment and to compensate for depreciation, thus strictly considering ICT assets (hardware, software and telecommunication assets).

Industry

Industry differences are controlled for, as described above in the writing about the industry choices for this study. It could be allowed to have a big differentiation in the chosen industries of the national economies, so contrasts would be clear. It is expected that each part of the economy has a certain level of dependency on ICT probably developing (or more like fluctuating?) over time, which can be translated into (the trend in) ICT investment levels. Some industries and the developments within require investment in ICT because of competitive pressures and other industries are still not quite 'computerised'. These dissimilarities might become obvious by comparing essentially different industries.

Low-skilled labour compensation (share in total labour compensation)

(13)

Hours worked by low-skilled persons engaged (share in total hours)

As it is expected that labour characteristics within the chosen industries are essential push factors for

ICT investment, just including a monetary value for low-skilled labour is not sufficient. It is believed that the hours worked by low-skilled persons engaged captures more specifically the low-skill intensity of the industries. In addition, the variable has a broader definition than just low-skilled employees. Every person engaged by the businesses (also flexible, temporary and subcontracted personnel) within the industries is contributing hours to the grand total values used.

It is assumed low-skilled labour is more easily replaced by non-human work than high-skilled labour is. Complex selling, designing and managerial activities are not easily or not replaceable by automated business processes at all. Thus, a big share of hours worked by low-skilled persons engaged in total hours worked within the industry (no matter the skill level involved) could foster management decisions in favour of ICT investment.

Consumption of fixed ICT capital

Something else the EUKLEMS consortium has done beautifully is splitting up fixed capital figures in different categories of capital, for various European Union countries separated in industries and sectors. One of those categories within the EUKLEMS database input files is ICT capital. The consumption of ICT capital, with 1995 as the year of indexing, is related to the depreciation of ICT

assets (the gradual decrease in value).

Consumption is used in preference to depreciation to emphasize that fixed capital becomes less useful in the process of generating new output (especially technologically savvy products may be outdated sooner rather than later). Consumption may include other costs incurred when using fixed assets beyond actual depreciation charges. Unlike depreciation in business accounting, depreciation in national accounts is, in principle, not a method of allocating the costs of past expenditures on fixed assets over subsequent accounting periods. Rather, fixed assets at a given moment in time are valued according to the remaining benefits to be derived from their use. Consumption of fixed ICT capital represents the decrease in remaining benefits as a consequence of ICT assets becoming more outdated, less valuable and less efficient.

(14)

Contribution of hours worked to value added growth

The contribution of the subtotal of hours worked to value added growth within an industry can be interpreted in various ways depending on the trend one can see. Reaching a plateau could be described as a good standard for manual human efficiency from where onwards only automation solutions could improve business efficiency. Following this line of reasoning, a higher contribution of hours worked to value added growth would positively correlate with ICT investment. Hours worked are defined per year and this is variable not limited to low-skilled labour. Value added is basically the difference between input values and turnover (Timmer, 2008).

Capital compensation (ICT assets)

Capital compensation for ICT assets is potentially an interesting indicator of ICT investment, because of the intrinsic marginal value of one Euro extra ICT investment explained by it. Keeping in mind that

ICT capital might not have very rewarding immediate pay-offs right after the initial investments, a high ICT capital compensation is attractive and could drive ICT investment decisions. 1995 is the index year for the time series of this independent variable, gathered from capital input Excel-files per country created by the EUKLEMS consortium and most recently updated in 2008.

Gross output volume growth

The comprehensive mix of variables tests many possible influences, but most explanatory models include some sort of representation of a base level. Whether it be a constant or the basic growth in output, a fairly stable dependent variable that is developing over time can not be determined by only fluctuating variables. Gross output volume is a share of the gross domestic products for the specific industries and sector of choice. The growth thereof is measured, meaning that the absolute values are converted to year-to-year percentage differences. Sources are the EUKLEMS output databases cross-checked with the OECD publicly available statistical figures.

Internal rate of return (IRR) on capital

(15)

Turnover divided by number of companies

Turnover is expected to correlate with the level of ICT investment, especially a lagged effect is deemed to exist. By utilising the unique company accounts database constructed by the EUKLEMS

consortium, variables like this one are as close as one can get to firm-level research without compromising the strictly quantitative nature of this research.

When industry turnover rises steadily, this could be merely the consequence of new companies with differentiated products entering the industry. This is why industry turnover is divided by the number of companies. If the average turnover in the industry is rising, expenses (variable costs) would be rising just as well of course. Assuming we are dealing with healthy industries where companies have a positive profit margin, a higher turnover causally would mean a higher net profit. This would result in more slack financial resources to use as internal funding for investment in ICT.

Adjusted Herfindahl-index

The Herfindahl-index measures how fierce the competition is in the particular industries. It depicts the concentration level (or market share division) in a particular industry or sector. As such, it can be used see if we should talk about monopoly dominated market, an oligopoly or a well diversified range of suppliers. The definition of the degree of concentration refers to the number of firms offering products or services in a specific industry.

The regular Herfindahl-index is widely used, but also widely criticised. The main point of criticism is the sensitivity for the relevant market definition, in terms of geographical boundaries as well as product homogeneity. Therefore, the more robust version: the adjusted Herfindahl-index accounts for close substitutes, was chosen. That should generate better results than the traditional indicator (Lijesen, 2003). The variable stems from the EUKLEMS company accounts database of 2008 as well.

Average age of companies (weighted by turnover)

The average age of companies (weighted by turnover) is very applicable to investigate industry maturity. An industry, in which physical non-ICT products are being sold and that is at a low maturity or market saturation, is not a good market for ICT products and services providers. Competing by technology, process efficiency, information system added value and innovation often is more relevant in the later phases of an industry’s development. These assumptions are the reasons for using the average age of companies in the regression. This variable, for which the construction of company accounts’ data is required once more, is a tool to monitor the entry of new incumbents and the exit of (in)experienced players in the industry its markets. It is tested whether new incumbents face a barrier of ICT investment (besides other barriers to entry) or if ICT investment is rather a manner to compete against competitors. National dimension

(16)

National dimension

Country is the second control variable. As has been described, three economies in Western-Europe are the scope of the research and only these mature economies are accounted for by every variable.

Burst of the Internet bubble

Many new start-up ICT firms issued a lot of equity to raise capital for innovations and other R&D. Many initial public offerings were successful and the dot-com stocks kept on rising in value, providing excellent returns to investors and venture capitalists. The Internet hype was huge, especially in the US. The change in attitude from overly optimistic to more realistic was a sudden twist. That is why it is believed the end of the hype (when many dot-com stocks plummeted in value, leading to bankruptcies), could have affected the investment in ICT products and services in non- ICT

industries as well.

From the burst of the Internet bubble onwards, the period after 2001 has been chosen. From 1998 to 2001 the dummy variable is 1 and before and before/after this date the independent variable is always 0. This is in line with the existing literature on the subject and a glance on a graph of the

NASDAQ (an indicator of the performance of stocks of all technology companies), learns that there is an obvious build-up to 2001. After this year, stock prices more than halved, marking the switch (hence the dummy variable) to a less optimistic investor attitude. The bad years can be depicted by one side of the coin and all the other years are supposedly the other more normal side. Considering this assumption, the central question for this variable can be more advanced: are stock investor attitudes related to management attitudes about investing in ICT for their company?

Long-term interest rates

Interest rates are a common factor in investment decisions. Especially, ICT investment is often seen as something additional. It can not be afforded by free cash flows (internally generated funds), unless there is a big pressure to advance technologically. This is when interest rates come into play. ICT

projects are often funded (partly) by borrowed money. Expensive borrowed money is deemed to hamper investment of any kind. Lowering the interest rates is often a measurement taken by governments to foster investments. Interest rates might not explain between-country differences, but could correlate with the absolute level of ICT investment. Simply stated, the lower interest rates are, the lower the weighted average cost of capital (WACC) of an ICT project would be. A low WACC

would lead to a calculated positive net present value rather often, even though return on investment for ICT projects is considerably difficult to predict.

(17)

countries, the nations of analysis. This interest rate dictates the interbank interest rate and the interest rate for the actual loan-providing institutions.

Each country has a maximum of what they can borrow from the ECB and if they need more they have to bid a small percentage of extra interest at the ‘money auction’, to pay back on top of the yearly reimbursements. To battle the credit crunch the ECB has installed a policy of ‘full allotment’. This implies that there are no limits on how many billions a country borrow. However, during the time period studied credit allotment was limited to what a country wants to pay in the future.

Relevant for this research is the fact that different countries have diverging interest rates. The interest rates of Germany and The Netherlands follow the same trend, but there is deviation from the

ECB interest rates and not always by the same number of percentage points. 0,3 percent differences between the neighbouring countries are not special and the UK has its own monetary policy. The UK

has the Pound Sterling as its currency, which has a different currency risk than the Euro (Hamberg, 2004). Sometimes the British interest rates are three percent higher than the same kind of interest rates for the same year in a country that has the Euro. Intra Euro-country variation could be explained due to the credit risk differences for the same industries across national borders.

Different kinds of interest rates can be downloaded from the national accounts of the three countries, for instance through the OECD website. All the data on the macro-economic level is mostly downloaded from the OECD website (http://stats.oecd.org). More specific, parts of the sector indicators database are extracted. It has been chosen to use long-term interest rates in this case, because often long-term borrowing is necessary, as ICT projects are commonly too expensive and too large in scale to have only short-term implications. Implementation often takes a long and underestimated time. Pay-off occurs in the long run. After the initial investment, the project costs need to be earned back by efficiency gains or less paper waste for example.

The burst of the internet bubble dummy together with the interbank interest rates and GDP

growth variables together are conceptualised and operationalised in the national dimension. Furthermore, the national dimension is controlled: three specific yet similar countries are chosen.

3.2 DESCRIPTIVE STATISTICS

Combining the observations for these countries together in panels, the OECD and the EUKLEMS

(18)

Table 1: descriptive statistics for the rubber and plastics industry

Variable Description Obs Mean Std. Dev. Min Max

rnp_rgfcf Real gross fixed capital formation 87 119,84 147,13 1,23 622,02 rnp_tdn1 Turnover / number of companies (-1) 30 88413,02 27931,89 43949,25 132606,50 rnp_ahi Adjusted Herfindahl-index 30 0,12 0,05 0,02 0,18 rnp_aac Average age of companies 30 48,09 9,19 28,98 60,95 rnp_irr Internal rate of return 87 0,15 0,08 -0,01 0,32 rnp_cfitc1 Consumption of fixed ICT capital (1995 prices) 87 99,02 122,96 0,88 488,42 rnp_itcc1 Capital compensation (ICT assets) 87 123,69 144,73 0,94 468,50 rnp_govlg1 Gross output volume growth % 97 0,09 27,83 -266,36 20,55 rnp_lc1 Labour compensation (-1) 91 4240,06 4893,34 200,42 15343,18 rnp_chwvag Contribution of hours worked to value added growth 75 -0,09 2,93 -10,38 5,11 rnp_hwlspe1 Hours worked by low-skilled persons engaged (-1) 75 -1,08 1,40 -7,63 3,62 rnp_ltir Long-term interest rates 117 7,57 2,65 3,35 14,88 rnp_bib Bursting of the internet bubble 117 0,10 0,30 0 1

Table 2: descriptive statistics for wholesale trade and commissions trade sector

wnctr_rgfcf Real gross fixed capital formation 87 1423,87 1681,19 23,72 6841,98 wnctr_tdn1 Turnover / number of companies (-1) 30 184628,00 84142,16 82981,44 379719,70 wnctr_ahi Adjusted Herfindahl-index 30 0,10 0,05 0,03 0,19 wnctr_aac Average age of companies 30 29,11 3,37 24,93 35,27 wnctr_irr Internal rate of return 87 0,26 0,14 0,08 0,76 wnctr_cfitc1 Consumption of fixed ICT capital, 1995 prices (-1) 87 1121,32 1362,71 15,29 5423,49 wnctr_itcc1 Capital compensation (ICT assets) 87 1706,36 1626,28 44,71 5489,78 wnctr_govlg1 Gross output volume growth % 97 3,29 7,09 -9,20 55,18 wnctr_lc1 Labour compensation (-1) 91 21783,81 19873,34 1463,66 67059,12 wnctr_chwvag Contribution of hours worked to value added growth 75 0,14 2,13 -10,25 6,90 wnctr_hwlspe1 Hours worked by low-skilled persons engaged (-1) 75 -0,83 1,57 -8,09 5,03 wnctr_ltir Long-term interest rates 117 7,57 2,65 3,35 14,88 wnctr_bib Bursting of the internet bubble 117 0,10 0,30 0 1

Table 3: descriptive statistics for post and telecommunications sector

pnt_rgfcf Real gross fixed capital formation 87 2870,00 3914,27 165,97 17100,61 pnt_tdn1 Turnover / number of companies (-1) 30 825289,00 560343,60 232895,90 1762588,00 pnt_ahi Adjusted Herfindahl-index 30 0,20 0,09 0,07 0,35 pnt_aac Average age of companies 30 11,81 4,21 5,95 19,88 pnt_irr Internal rate of return 87 0,10 0,06 -0,04 0,24 pnt_cfitc1 Consumption of fixed ICT capital (1995 prices) 87 2098,47 2801,13 147,54 12172,34 pnt_itcc1 Capital compensation (ICT assets) 87 2597,56 2753,77 61,31 8784,17 pnt_govlg1 Gross output volume growth % 97 3,59 20,20 -185,39 24,28 pnt_lc1 Labour compensation (-1) 91 7845,60 7052,50 566,64 21443,79 pnt_chwvag Contribution of hours worked to value added growth 75 -0,26 2,48 -9,03 5,20 pnt_hwlspe1 Hours worked by low-skilled persons engaged (-1) 75 -0,66 1,56 -8,03 4,24 pnt_ltir Long-term interest rates 117 7,57 2,65 3,35 14,88 pnt_bib Bursting of the internet bubble 117 0,10 0,30 0 1

(19)

Figure 2: Real gross fixed capital formation development over time for the three industries

In the graph (figure 2) the development of ICT investment is shown. Levels are increasing, especially for post and telecommunications, with peaks right after the change of millennium for this sector of special interest. Clearly, the development of ICT investment for the post and telecommunications sector is not gradual at all unlike for the other two industries.

3.2 CHOICE OF COUNTRIES

In respect of the research bias towards the US, it has been chosen to focus the study over more than just my mother country by using panel analysis to acquire wider evidence for Westernised economies. Whilst including The Netherlands data, panels have been developed together with the UK

and Germany data to have a broader range of values for a more accurate and generic explanation of the development of ICT investment levels. All of the countries are deliberately chosen to be developed West-European economies (according to the OECD definitions), because this study has the focus on explaining differences between industries within countries in a similar stage of economic development. Emerging economies, for example, still are still growing by brute force of the population and less by ICT investment. Thus, lesser developed economies are not that interesting to investigate. For that reason, similar developed European economies of different scale with similar characteristics of their industries and sectors are researched.

The US, on the other hand, was the earliest adopter of ICT products and services and this fostered their growth before the dot-com bubble and the credit crunch. The effects of the early ICT adoption by the

(20)

adopted this early as well, but chose not to do so. The reasons are unclear as the researched countries are quite technologically advanced just as well. Therefore, a different approach has been chosen: contribution of ICT to growth is assumed in line with several studies’ outcomes and ICT investment is attempted to be explained. Asking our self why Western Europe missed the boat and started investing more in ICT relatively late compared to the US, is that because of certain aspects on the national level or rather on the industry/sector level?

What made the internet hype in the US so big around the turn of century is well researched (although still disagreements about it exist) and that is why this research is purely focused on Western Europe.

3.3 CHOICE OF INDUSTRIES

Moving on to the industry dimension in general, it was quite difficult to acquire a mix of interesting sources with sufficient observations, but eventually a unique combination of resources is acquired in cooperation with the GRDC.

As a source for industry level data the EUKLEMS database (http://www.euklems.net/) is utilised. This is an extensive and unique database. In a short time period of public existence, there are many requests for the data and added value to the database already. The database is highly interesting as it is seen as a great building block for further analysis on growth already. With the separation of ICT

assets from other more tangible assets, specific research on ICT-related bits of the databases can be fruitful. ICT investment levels, ICT consumption, labour compensation, capital formation, labour hours etc. are all suitable for analysis on the industry level. These variables are taken from the capital input sector/industry accounts of the 2008 output databases.

Only one industry of interest would give no possibility to find out whether industry variables are of significant differentiating importance in explaining ICT investment. That is why three very dissimilar industries have been chosen to compare various parts of countries’ total GDP. To be precise: rubber and plastic products (25), wholesale trade and commissions trade (51, without motor vehicles and motorcycle business) and post and telecommunications (64, in fact a whole sector) are researched. The numbers are given to the industries by the OECD and are recognised in international research contexts.

The industries selected are not specialised in providing advanced services for which a great deal of supporting ICT assets are needed, as in the financial services sector. The post and telecommunications sector is chosen, because it has been developing into a more ICT-driven way of doing business. The services became less machinery and manpower driven over time. The other two industries have increasingly been supporting the back-office tasks by ICT assets, but the core business therein is not that ICT dependant.

(21)

This was also a practical consideration during data selection, as for many small industries data about

ICT possession is too scarce.

3.4 TIME PERIOD

1970 up until 2007 is the chosen time period for analysis of yearly data, considering the limitations of the sources. For the years before 1970 too many observations are missing, because of the novelty of

ICT and the administration of the values of it. Starting the timeframe after 1970 would have the consequence of needing to drop interesting variables, which is a pity because the 1970s can be seen as a time of financial crisis. Comparing to the recent crisis, low growth, high inflation and soaring energy prices are commonalities. However, during the recent crisis the housing crisis and spiking commodities prices have been thrown into the mix. It looks like a prolonged recession, thus extending the timeframe up until now would be a worse option. That would include an abnormal period of recession, wherein all the companies in the sector and industries chosen are extremely limited by the credit crunch.

For certain independent variables considering the industry factors for a particular country, the available values would stretch over a period of a limited number of years (see appendix A, B and C). However, there are always at least thirteen observations and this can be multiplied by three because the panel nature of the research takes into account three countries for every variable.

4. METHODOLOGY

The conceptual framework seems quite extensive. However, the multiple dimensions are very well suited for dynamic panel analysis. This is because there is only one dependent variable and the development of the independent variables can be captured over time for the various industries for three similar nations all together in pools (dynamic panels). Panel data, also called longitudinal data or cross-sectional time series data, are data where multiple cases (people, firms, countries, etc.) were observed at two or more time periods (Stock and Watson, 2003). In my research the cases are multiple countries and by using a few countries the number of observations can be lifted up to very acceptable counts.

(22)

for the total number of observations. Consequently proper variances and F-tests are computed. Hence, the unbalanced panels are easy to accommodate very well suited for time-series analysis. The independent variables are separated on two separate dimensions and the most important dimension is the industry level dimension. On this lowest level of aggregation firms can be grouped in different sectors as well and on a higher level there is the macro-economic dimension. So from microeconomics to macroeconomics, many potential ICT investment impacting factors are included in the regressions.

Obviously, the model would be too complex and data collection would be too cumbersome if there would only be the separation of the independent and dependent variables. That is why both dimensions are controlled to limit the scope of the research, thus making it a feasible research and decreasing outcome bias by potentially unfortunate overrepresentation of a certain industry or sector in aggregates.

The reason for taking countries together and not industries is that the industries chosen have very different needs for their production or service processes in relation to the customers. ICT can vary broadly in degree of importance or essentiality within a country and across industries. For this reason, the results might become biased and insignificant if a random sample of companies investing in ICT will be used as the dataset.

Summarizing, industry dimension variables and the national level variables will be tested for their influence on ICT investment in a particular industry across countries. The results for one industry can be compared to other industries, as there will not be severe overlap of company entities between investigated sectors. Yet, it is possible that certain rubber and plastics producers have some wholesale trade business next to their core-business, for example.

Random-effects panel analysis and Arellano-Bond dynamic panel data analysis (one-step) are utilised. The Houseman-specification test is the classical test of whether the fixed or random effects model should be used. This test questions whether there is significant correlation between the unobserved person-specific random effects and the regressors. If there is no such correlation, the random effects model may be more powerful and parsimonious. If there is such a correlation, the random effects model would be inconsistently estimated and the fixed effects model would be the model of choice.

Random effects came out to be the preferred method after the Hausman-test indicated fixed effects regressions would be biased. P-values and Prob>chi2 are larger than 0,05 for all the industries (actually approaching one), so random effects applies better for my study than fixed effects does. For the random-effects dynamic panel analysis the following formula is regressed:

(23)

In this formula ind stands for the industry or sector being analysed. The possible abbreviations of the sector or industry consist of rnp for rubber and plastics, wnctr for wholesale trade and commissions trade and pnt for post and telecommunications. For every single variable the textual explanation is given in section 3.3 and the descriptive statistics of every variable, which clarify the whole formula, are provided in section 3.2. dbib is the dummy variable representing when the dot-com bubble bursted

and εt is a standard error term that captures the noise, which can not be explained by the model. All

the betas (β) are the coefficients of the variables (the multiplier of their influence on the dependent variable). It is attempted to gather various variables of the same order, but coefficients can still vary within quite a broad spectrum of values. This is not detrimental to the research as the focus is on discovery of significant correlations.

The research is reproducible by using the data gathered as can be seen in appendix A, B, C with all the collected data for respectively The Netherlands, Germany and the UK. Moving on to the second research method applied, because potential endogeneity in the random-effects model can be further mitigated by the Arellano-Bond regression and the robustness needs to be verified:

(2) ind_rgfcft= β0 + β1∆ind_rgfcft-1 β2∆ind_tdnt-1 + β3∆ind_ahit + β4∆ind_aact + β5∆ind_irrt +

β6∆ind_cfitct-1 + β7∆ind_itcct-1 + β8∆ind_govlgt-1 + β9∆ind_lct-1 + β10∆ind_chwvagt +

β11∆ind_hwlspet-1 + β12∆ind_ltirt + ∆dbib + εt

This second formula might seem to have a high level of complexity, but it has the same variables included as the first formula. Added to the base model are a few standard terms belonging to the Arellano-Bond panel specification: the lagged variation of the dependent variable is included as an independent variable. This is to test if the simplistic assumption of previous values of the level of ICT

investment predicting the ICT investment in the current year holds and especially to remove endogeneity.

Arellano and Bond (1991) introduced lagged dependent variables into their model to account for dynamic effects. The lagged dependent variables can be introduced to either fixed or random effects models. Their inclusion assumes that the number of temporal observations is greater than the number of regressors in the model. This assumption holds for the model used, thus lagged dependent variables can be added to the random-effects model in a confident manner.

For dynamic panels with lagged dependent variables, Arellano, Bond, and Bover have used general methods of moments, which are asymptotically normal (Wooldridge, 2002). With greater numbers of moment conditions, they are able to handle some missing data and they can attain gains in efficiency as long as there are three or four periods of data (Greene, 2002). Based on these sources, dynamic and unbalanced country panels with at least 10 years of observation for each and every variable are especially valid and suitable for this research.

(24)

econometric solution to the potential problem of endogeneity, one can use so-called dynamic panel estimators, which rely on instrumental variables estimation in the Generalised Method of Moments framework (GMM). For example, the Arellano and Bond (1991) estimator takes first differences of the panel data (thereby wiping out the individual effects) and resolves the endogeneity issue by using lagged levels and differences of the dependent variable as instruments in a GMM-framework. This estimator is for a large number of individuals with few time observations asymptotically unbiased (Elsas and Floryiak, 2008). By transforming the regressors by first differencing the fixed country-specific effect is removed as well, because it does not vary with time.

5. RESULTS

In this chapter the main findings for the random-effects and the Arellano-Bond panel regressions are presented in tables and described by words. First of all the random-effects GLS regression outcomes are presented for the industries and sector under investigation. After these basic results, the Arellano-Bond results will clarify whether the random-effects model does or does not suffer from endogeneity too much. The findings acquired should still hold for a similar model with a few added variables, so that after the analysis conclusions can be drawn from compliance-checked and robust models.

5.1 RANDOM EFFECTS

Rubber and plastics

Table 4: results random-effects GLS regression for the rubber and plastics industry

Variable Coefficient Significance

Average turnover (-1) -0,0006 0,302

Adjusted Herfindahl-index 179,6944 0,411

Average age of companies -4,3909 0,040**

Internal rate of return -242,0176 0,422

Consumption of fixed ICT capital 0,6107 0,020** Capital compensation (ICT assets) 0,4868 0,062*

Gross output volume growth % 3,3956 0,354

Labour compensation (-1) 0,0003 0,962

Contribution of hours worked to value added growth -4,6163 0,328 Hours worked by low-skilled persons engaged (-1) 5,5027 0,505

Long-term interest rates 30,8231 0,385

Bursting of the Internet bubble -38,5589 0,156

Constant 174,2653 0,097*

R-squared 0,98842

Note: *, **, *** represents significance at respectively 10%, 5%, and 1% levels.

2

(25)

Table 4 presents various estimates of Eq. (1):

(1) ind_rgfcfn,t = β1ind_tdnn,t-1 + β2ind_ahin,t + β3ind_aacn,t + β4ind_irrn,t + β5ind_cfitcn,t-1 +

β6ind_itccn,t-1 + β7ind_govlgn,t-1 + β8ind_lcn,t-1 + β9ind_chwvagn,t + β10ind_hwlspen,t-1 + β11ind_ltirnn,t + dbib + εn,t

All these variables their values (as can be found in the appendixes) are reshaped (converted from the wide to the long form) and then pooled together for the three countries. The results of rearranging (not altering) the data can be found in section 3.2 devoted to data description. In table 4 you see the outcomes of the random-effects generalised least squares regression run for the reorganised data about the rubber and plastics industry.

Reported are the coefficients and their significance level, as other results are not nearly as interesting when keeping the original research question in mind. The fit of the model is excellent (0.9884 R2), but not many variables are significant. Nevertheless, the interpretation of the results can be focused on the factors that are correlating with ICT investment levels.

The most significant factor is the consumption of fixed ICT capital in the former year. With significance at a 5% level (0,020) much certainty exists about the replacement cause for ICT

investment. The positive relationship does not have a high coefficient, but is still quite strong as consumption is of the same order of monetary value as ICT investment (the reported coefficient is 0,61). Next, the average age of the companies is notably interesting, as it is also significant at a 5% level (0,040). The correlation of company age is not as expected though: younger companies tend to invest more. This means that new companies trying to compete with the existing firm base tend to invest quite some money in ICT and younger firms have more development going on, for which ICT

investment is necessary. The older the average age of the companies in rubber and plastics, the lower the ICT investment is. Maybe there is a technological gap to close for new incumbents to enable them to compete with the settled businesses in the industry.

(26)

Wholesale trade and commissions trade

Table 5: results random-effects GLS regression for the wholesale trade and commissions trade industry

Average turnover (-1) 0,0047 0,446

Adjusted Herfindahl-index -12369,2000 0,215

Average age of companies -78,2416 0,470

Internal rate of return 2274,8140 0,335

Consumption of fixed ICT capital 0,8611 0,004*** Capital compensation (ICT assets) 0,3399 0,391

Gross output volume growth % 1,3533 0,975

Labour compensation (-1) -0,0027 0,945

Contribution of hours worked to value added growth 57,7322 0,587 Hours worked by low-skilled persons engaged (-1) 271,1279 0,019**

Bursting of the Internet bubble 18,6403 0,968

Constant 1439,3190 0,675

R-squared 0,97093

Note: *, **, *** Significance at respectively 10%, 5%, and 1% levels.

This second section of basic results depicts the outcome of formula (1) again with the same variables as for the rubber and plastics industry, utilising the random-effects model again. The totals of ICT

investment level for the industry are correlating with the independent variables as follows.

For this trade industry significance on a 1% level is reported for the consumption of fixed ICT

capital in the former year (0,004). However, the coefficient value is not high at all. Despite the low coefficient value, for this industry, investment to offset consumption of the same kind of assets (ICT) is necessary to keep businesses up and running as usual. The coefficient value lower than one shows that there is more to ICT investment than just regaining monetary values of ICT assets by replacement and renewal investment. The coefficient of 0,86 for the consumption of fixed ICT capital implies that there is a fraction of ICT investment related to the other factors as well. The only significant one among those other factors is hours worked by low-skilled persons engaged of the former year, with significance at a 5% level. The positive correlation brings support delivers support for the statement that increasing hours worked by low-skilled persons fosters ICT investment the next year. Either low-skilled labour generally entails working with ICT assets or increases in low-skilled labour hours tend to push for enhancing the role of ICT in various tasks in a manner that manual labour can be sacrificed.

3

(27)

Post and telecommunications

Table 6: results random-effects GLS regression for the post and telecommunications sector

Adjusted Herfindahl-index -34844,4500 0,040**

Average age of companies 349,9461 0,034**

Consumption of fixed ICT capital 0,1931 0,667 Capital compensation (ICT assets) 0,9921 0,142

Gross output volume growth % -18,6312 0,860

Contribution of hours worked to value added growth 388,3624 0,003*** Hours worked by low-skilled persons engaged (-1) -519,7528 0,072*

Constant -4936,3720 0,380

R-squared 0,98234

When interpreting table 6, one can notice that the most interesting findings arise for the post and telecommunications sector. Firstly, because of the relatively high significance levels of the factors correlating with ICT investment. Secondly, additional variables pop-up as significant compared to the other industries’ regressions. The most significant finding is obviously the contribution of hours worked to value added growth. Clearly, worker productivity is positively related to ICT investment, as there is significance at a 1% level (0,003). Then, further down the list of significance, one can see two variables stemming from the company accounts variables arising as significant factors in results. First of all, the concept of market concentration represented by the adjusted Herfindahl-index appears to play an important role with significance at a 5% level. The significance is slightly higher (6/10th of a percent) than the significance of the average age of the companies. The adjusted Herfindahl-index has a bigger and negative coefficient. Bigger does not imply anything as the scales of the variables are not very comparable. The negative sign is implying that when the competition is fiercer, this drives down ICT investment as resources are to be applied to be competitive in the short run. ICT

investment does not fit in this picture.

The positive sign for the average age of the companies in this big sector is basically saying that older companies tend to invest more in ICT than the later market entrants do. Finally, there is another sign of interrelatedness between the hours worked by low-skilled persons and ICT investment. Although this link is only significant at a 10% level (0,072), for this sector there is reason to believe

4

(28)

ICT investment is conducted to replace manual labour for which not very specialised skills are needed.

5.2 ARELLANO-BOND

In this paragraph several findings of the random-effects model are confirmed by the Arellano-Bond dynamic panel analysis. Some new findings are offered by this more extensive model with more measures against endogeneity. The Arrelano-Bond regression is mostly used to verify whether regression of a model rather similar to the random-effects model, adding a few basic derivatives of variables already included in the random-effects, acquires the same results. Conducting the Arellano-Bond dynamic panel-data estimation (one-step version), should confirm that the main random-effects model was a good choice, clarifies whether the dataset is robust to changes in the method of statistical analysis and concludes how sensitive the results are to adding variables and lagging the dependant variable as another independent variable.

Rubber and plastics

Table 7: Arellano-Bond dynamic panel-data estimation: one-step result rubber and plastics

Real gross fixed capital formation (-1) -1,1794 0,020**

Adjusted Herfindahl-index 559,1667 0,124

Average age of companies -3,4181 0,269

Consumption of fixed ICT capital 2,2082 0,001*** Capital compensation (ICT assets) 0,3412 0,329

Gross output volume growth % -2,6699 0,513

Contribution of hours worked to value added growth -11,6470 0,047** Hours worked by low-skilled persons engaged (-1) -2,3117 0,789

Constant 353,4987 0,049**