The Economics of Technological Change and Employment in the Digital Machine Age

Contents

1 Introduction

2 The Economic History of Technological Change
2.1 From Industrial Revolution to Automation
2.2 From Automation to Digitization

3 The Neoclassical Theory of Technological Change and Employment
3.1 Types and Drivers of Innovation
3.2 Compensation Theory of Labor-Saving Technological Change
3.3 Labor-Saving Technological Change
3.4 A Neoclassical Model of Technological Change and Employment

4 Estimating the Impact of Technological Change on Employment Measures
4.1 Description of the Model
4.2 Building the Model
4.3 Estimation Results

5 Conclusion

Bibliography

Appendix

List of Figures

1 German Manufacturing and Service Industry as a Percentage of GDP

2 Share of Capital Stock younger than five years in German Manufacturing

3 Change of Employment in Three German Sectors

4 Labor Productivity in Germany

5 Wage Share of GDP vs. Corporate Profit Share of GDP

6 Development of Education Level in Germany

7 Development of Gross Incomes in Germany

8 Product Life Cycle

9 Invention Possibility Frontier

10 Localized Technical Progress

11 Types of Technical Change According to Hicks

12 Equilibrium Analysis - Introduction of a Labor-Saving Technology

13 Harrod and Hicks-neutral technological change

14 The Aggregate Production Function

15 The Labor Market

16 The Goods Market

17 Technical Change in a Neoclassical Framework

18 Monopoly as a Consequence of Innovation

19 Consequences of a Decrease in Wages

20 Employees in German Manufacturing as Percentage of Total Employment

21 The Vintage Approach

22 Trend of Scrappings and Investments in German Manufacturing

23 Trend of Productivity Measures in German Manufacturing

24 Trend of Hours Worked in German Manufacturing

25 Trend of Payroll in German Manufacturing

26 Development of Capital Stock in German Manufacturing

27 US Manufacturing and Service Industry as a Percentage of GDP

List of Tables

1 Dependent Variables

2 Explanatory Variables

3 Correlation of Untransformed Independent Variables

4 Composition of Estimated Models

5 The Effect of Capital, Productivity and Utilization Rate on Hours Worked

6 The Effect of Capital, Productivity and Utilization Rate on the Payroll

7 The Effect of Capital, Productivity and Utilization Rate on the Number of Employees

1 Introduction

Technology as a Matter of Economic Interest A timeliness object of discussion since the first noteworthy appearance of economic literature has always been the construct of technological unemployment. The fear of people being put out of work by some machines really took off during the industrial revolution in the late 18 th century owed to the invention of the steam machine, which eventually lead to the automation of the weaving craft.

The economic interest in the relationship between new technology and its social impact is mostly based on the idea of increased productivity. If we consider a production func- tion containing an arbitrary number of input factors, then given a certain amount of output, profits will decrease, if any of those inputs is becoming more expensive. In the course of time before human and property rights were installed, this have been most often the factors land and labor. Labor insofar, as it depended massively on the ability of keeping people healthy and well-nourished and also on the expected life span which was certainly lower, centuries ago. The United Kingdom was affected first and severest by the invention of the steam-engine. Ulrike Herrmann (2013) explains the necessity of this invention on the scarcity of labor due to the wars and famines plaguing the Europe of that era. This disruptive general purpose technology is the first encounter of em- ployees with massive technological unemployment and the almost obliteration of certain professions, i. e. weavers.

No matter if we talk about a Japanese high-tech hotel that employs human-like robots (The Henn-na Hotel), an intelligent oven that takes over the cooking process for us (June Life Inc., 2015) or about Apple Inc.’s huge investment plan of over $ 10 billion in robots and machinery to automate production (appleinsider.com, 2013), the pervasion of autonomous robots in every ones life are indispensable. It is not the new technology itself that generates an immense interest in this development, it is the pace with which the environment is adapting to digital and automating technologies per se. Compared to the first industrial revolution from the introduction of the power loom propelled by steam-power to the first riots of the Luddite movement it took more than 30 years. The steam machine was only replaced bit by bit by electric motors which enabled the further division of labor in decentralized production sites. Referring to Hemingway, bankruptcy comes gradually, then suddenly. Same is to be considered to happen in the digital ma- chine age (see Brynjolfsson and McAfee (2014)). Major multinational corporations that do not see the change of that digitization brings along, are running the risk of losing out on the competition that small and agile start ups bring to markets by turning business models upside down due to scale effects enabled by digitization.

In recent years manufacturing sectors in high-wage countries find themselves in pronounced turnover, which in Germany is called Industry 4.0. This evoked fourth indus trial revolution, cyber physical systems or the so called Internet of Things all adhere the same idea: the pervasion of the production process by information technology. A fully integrated network of intermediate and final products that are able to automate the complete production process by communication between each other over the Internet or other networks without human interference, at best.

This development will change not only the portfolio of products and services, as it is proposed, but probably the work environment of employees in the sector affected. Higher skills will be needed to engage in complex work environments. Employees will have to acquire additional human capital in order to match the labor demand driven by tech- nological improvements in the production process. The smart factory, as it is called, is not only object of research in federal funded science projects (Centrum Industrial IT, 2015) but also the innovative idea of implementing technologies enabling integrated and linked manufacturing. Implementation will need a high degree of investments into new equipment and software which should be observable via data collected by established organizations.

The share of gross domestic product (from now on GDP) of the manufacturing sector declined over the past decades in high-wage countries, as it was commonly assumed that former prospering times of the manufacturing industry were past due to intensive labor costs and losing competitiveness to emerging markets (Sendler, 2013). The financial in- dustry with its services was claimed to lay foundations for future economic development. This wishful thinking abruptly came to halt in 2009 when the world economy got hit by the financial crises.

Figure 1 shows the differences between the German manufacturing and services sectors. Remarkably, the share of GDP of German manufacturing is residing at around 30 % of GDP since the early 1990s. In the United States, a much lower share of total gross domestic product is observable at around 12 - 13%¹ in recent years. This might be referable to higher degrees of off-shoring of US production to low-wage countries and potential markets. In Germany, the share of GDP contributed by the service industry steadily increased over the past decades, but now being dead level with manufacturing related to growth rates. At least for Germany, the industrial sector still is one of the major employment and value adding factors in the economy. This can be related to decisive promotion of automation instead of off-shoring production facilities (Sendler, 2013). Thanks to delivering high quality products and appropriate reliability, Germany is still one of the leading economies when it comes to exporting high tech manufactured goods.

Figure 1: German Manufacturing and Service Industry as a Percentage of GDP

illustration not visible in this excerpt

Source: German Federal Statistical Office - 2015;

Besides production that is affected by the ongoing digitization of manufacturing, new business models are created by the possibility of integrating clients into the manufac- turing process. Mass customization, as a buzzword, enable whole new areas of con- ducting business via the Internet instead of face to face business, meeting demand of consumers more accurately. Producers will optimize their production by being aware of each of the consumer’s needs. One of the first companies implementing such strategies is sportswear manufacturer Nike. By using this technique, the manufacturer generates massive amounts of data (big data), which is essential for product innovation and market analyses and could bring competitive advantages over competitors.

This thesis will provide an overview on the concept of technology in economics. The main focus lies on the impact technological change has on matters of employment based on neoclassical views. A special focus lies on the idea of embodied technical change in new machinery, which will be analyzed in section 4 with the help of a vintage-capital approach. The relation between technological change and employment effects are of economic interest regarding the automation and digitization of tasks and, thus, labor. All the above discussed aspects of the fourth industrial revolution of course must come at some societal cost. Every innovative technical disruption concerning the production of goods pushes some workers out of work as their tasks become obsolete, but also cre- ates new job opportunities. As already mentioned, tasks might become more complex and therefore new skills have to be acquired by employees. Is the wedge between low- skilled and high-skilled becoming more severe, as some researchers claim (Brynjolfsson and McAfee, 2014, Rifkin, 2014, Acemoglu and Autor, 2011)? This question will not be answered concludingly, but section 2 will indicate that the findings of Frey and Osborne (2013), as well as those cited by Brynjolfsson and McAfee (2014), may not be conferable to German manufacturing for the last two decades.

The IAB (Institut für Arbeitsmarkt- und Berufsforschung) analyzed the effect of the on- going digitization on employment matters until 2030. The conducted scenario analysis estimated an overall prosperity due to the structural change of industry 4.0 but also a loss of 60.000 employees due to automation, performing mostly unskilled jobs (Wolter et al., 2015).

In section 2, an overview of the economic history on the topic of technological change is presented. This is followed by selected theories and their contribution to the subject, demonstrated via theoretical models, in section 3. Section 4 will introduce the methodology of the empirical analysis. The work closes with concluding remarks on the results of the empirical analysis and suggests some ideas for further research to be conducted in the field of technological change and its impact on employment.

2 The Economic History of Technological Change

2.1 From Industrial Revolution to Automation

At the end of the 18 th century, the introduction of water and steam power initiated the change from an agricultural to an industrial economy. The mechanization of work enabled production to become more productive. Transportation and logistics have been facilitated by the new technologies due to powered carriers and cranes able to lift and transport heavy duty equipment that was formerly not possible. Wage labor arose due to the possibility of dividing complex tasks into less complex ones enabling a higher number of less skilled workers to do the tasks formerly only skilled artisans could have done. It was this development that lead economists as Smith, Say and Marx, to think about social justice and inequality related to this area of concern and also the overall economic picture that emerged from the implementation of labor-augmenting and -saving innovations.

Classical Economics and Technological Change

This section will give a selected summary of important contributions to the economics of technological change, beginning with Adam Smith and James Steuart. In his work An Inquiry into the Principles of Political Economy (Steuart, 1767, p. 120), he writes:

”A room cannot be swept without raising dust, one cannot walk abroad without dirtying one’s shoes; neither can a machine, which abridges the labour of men, be introduced all at once into an extensive manufacture, without throwing many people into idleness.” which reveals the ever recurring fear of unemployment caused by job automation. Steuart’s theory comes close to the classical economic theory, i. e. he also assumes compensation of short-term unemployment, caused by more efficient factors of production (capital), via the demand for labor in the sector that produces the newly introduced machinery and the decreasing prices which lead to increasing demand of the product produced by the novel equipment.

Adam Smith (1776), however, was more concerned about the division of labor that is entailed by newly introduced machinery. In the Wealth of Nations, Smith stresses the importance of the connection between the progressive division of labor and the ongoing increase of labor productivity, as one worker is more and more able to do the work of many, owing to machines that increase the productivity of labor. All in all, these two pioneers in the area of economics of technological change were concerned about the socio-economic effects of an ample use of technologies in the production process that would push people out of work - at least temporarily until adjustment processes bring the economy towards equilibrium.

In 1815, Thomas R. Malthus (1986) remarks differences in productivities between the agricultural and the manufacturing sector. Industrialized production is more and more mechanized and labor- and capital-saving technologies can be implemented up to a point, where the price of an industrially produced good is equal to the price of production from the most efficient equipment. Whereas, productivity growth in the agricultural sector would lead to population growth, but not to per capita income growth in the long-run. This supposition originates from the argumentation, that increasing per capita income above an equilibrium level of consumption will result in population growth. In his way of thought, Malthus concludes, that every increase in population attenuates per capita ca- pabilities, recycling consumption to its initial equilibrium level. Also, given the scarcity of fertile lands, Malthus concludes, that while prices for goods in the manufacturing sector decrease, prices for agricultural produce increases up to the point, where labor costs of land-workers equal the cost of machinery doing the same job. Along with West (1815), Malthus and Ricardo established the idea of diminishing returns in production.

In his essay On Machinery (Ricardo, 1817, ch. 31), David Ricardo conceptualized the idea of progressive mechanization: capitalists were to invest in fixed capital, that is ma- chinery, in order to increase the productivity of their workers, induced by an increase in goods demand, only if the cost of a good produced by a machine is less expensive compared to one produced manually by an artisan worker. In other words, if an increase in wages is observable, capitalists will substitute capital for labor, if profitable. If the increase in labor productivity outruns the rise in goods demand, so that the capitalist would produce an excess supply of goods, he would adapt his labor force and demand less workers in the production process. Within this essay he revises his former view, that if profits increase, wages of the working class would rise similarly. Hayek (1942) named the described substitution effect the Ricardo Effect. Ricardo states that the two factors of production, i.e. capital and labor, are permanently in competition and capital is not employed until the price for labor is relatively higher (Ricardo, 1817, p. 479).

Nevertheless, Ricardo also notes that if technical progress is not promoted, given an open economy the domestic one will suffer from technical advancement of the foreign. Samuelson (1988) advanced Ricardo’s idea of a diminishing population induced by the Ricardo Effect in the long-run. He consequently extends and formalizes Ricardo’s model and shows, considering progressing technological change in the form of robots replacing manual labor, that the human labor force will not survive and be replaced completely by robots.

Say (1803) also concludes in his 1803 work A Treatise on Political Economy that tech- nical change, at least in the short run, has negative effects on labor as new machines are introduced into the production system and no alternative capital goods ensure the employment of the disposed workforce. In the long run, Say’s law states that total ex- penditure equals the national product in an economy (given a closed economy without a public sector), so that there is no excess production or demand. Disposal of labor might only occur in Say’s framework, if the rate of investment cannot keep up with the velocity of changing capital-productivity, which, nevertheless, is rather unlikely because imple- mentation of technical change tends to need time and entrepreneurs are often skeptical due to uncertainty of return on new investments. Interestingly, Say already mentions that new means of production mostly change the essence of a good which eventually means that refusing new technology leads to refusing new products.

Neoclassical and Keynesian Economics

The neoclassical school of thought, beginning in the late 19 th century, developed the an- alytical methodology on the foundation of decision making based on margins. Walras, Jevons and Menger are the most commonly known pioneers of the neoclassics. Still, the underlying utility maximization is commonly used in contemporary economic analyses. The three underlying neoclassical assumptions of rational expectations, utility or profit maximization and full information can be found in the later (section 3) described theory of compensation mechanisms.

Keynes popularized the term Technological Unemployment and thought about the eco- nomic possibilities of our grandchildren (Keynes, 1931), in which he concludes that in the distant future people will not have to work all day long to earn for their living due to productivity increases that enable more free time to be consumed. What actually is observable in German Manufacturing, is a work week that has diminishing weekly hours of work, if one considers the development of the last century. Technological unemploy- ment is considered a rather short term phenomenon by optimists, evoked by disruptive innovations making a certain skilled human labor obsolete. Pessimists see appreciable harmful effects on employment induced by innovation. These concerns were expressed excessively by manifestations of the Luddite movement, a formation of British workers fearing social impoverishment by the incipient industrialization at the beginning of the 19 th century. This gratuitous fear is nowadays labeled the Luddite Fallacy as none of the later described compensation mechanisms are considered. Essentially, the fallacy states that short term unemployment is possible but due to more efficient production and decreasing prices, demand will increase across all other industries not immediately affected which in turn should create a higher demand for labor.

In the digital machine age, in which we are in, an important question is now, what if automation and digitization occurs in each industry and there are no jobs that are labor intensive enough to absorb the disposed workforce? Partly, Sachs et al. (2015) show how the transition into a fully automated world could look like. They state, that autonomous robots increase output due to higher productivity but also may peril employment, put downward pressure on wages and decrease investment complementing labor. In an over- lapping generations model, they show, depending on specific parameters, how increasing robotic productivity is decreasing welfare of future generations and young workers but simultaneously benefiting retired workers disproportionately by increasing the rates of return on capital (machinery), a source of income for retired workers. Their conclusion to encounter such a development depends strongly on redistribution policies to provide benefits for all generations affected by increasing productivity of robots.

Skills and Tasks in the Reticle of Automation

Another scope of technical change in economics relates to skill-biased technological change and resulting wage inequalities. Tinbergen (1974) emphasizes the problem of substitution between different types of labor. Acemoglu and Autor (2011) discuss the increasing income inequalities between different types of skills, especially between col- lege graduates and high school graduates. Technological change increases the demand for skilled labor, as tasks are becoming more sophisticated. They stress the differentia- tion of the terms ‘task’ and ‘skill’, as skills are related to education and tasks are related to production. The increased demand for college graduates relative to lower educated workers has been studied by Autor et al. (1998). These authors relate this development to the ascent of the use of computers in work environments. An interrelation between the increased high-skill demand and wage inequalities is carved out and associated with new production techniques, organizational changes, and capital deepening. Interesting findings come from Card and DiNardo (2002), who analyzed wage inequal- ities since the 1980s and found a stabilization in the 1990s. They remark that skill biased technological change is not able to explain the different conditions of several types of wage gaps. The focus on the demand for skills evoked by technological change and its effect on wages is one aspect of the ever proceeding technical advance, but if we consider a marginal cost society, as Rifkin (2014) labels it, the focus can also lie on the income distribution and its relation to the prevailing market structure. Markets for information goods show these characteristics and generate monopoly structures in general, as is shown by Jones and Mendelson (2011). Real world observations are for example markets where Facebook, Google, Amazon, WhatsApp and Microsoft engage. Monopoly structures are known to create lower employment levels compared to compet- itive market structures therefore leading to lower aggregate employment levels. Also, in non-competitive market structures the distribution of profits becomes narrower.

2.2 From Automation to Digitization

What is new about technological change nowadays compared to the big waves of tech- nical progress during the past centuries? If we consider the transition from the steam engine to electrification, a nameable scale effect has taken place. Power was no longer locally bound where it was produced, but could be transfered over certain distances and therefore enabled completely new kinds of business models. Production was able to take place decentralized and machines could be designed to fit to the work-place, not vice versa.

Today, however, the economy faces a new revolutionary change. The transition from analogous, mechanical manual labor to automated labor carried out by robots that think for themselves, making human monitoring almost obsolete due to the Internet of Things where communication between products and machines is the normal state. The scale effect can be enormous and might make the majority of the work force currently employed redundant. This is at least what can be found in current scientific surveys from Frey and Osborne (2013), who analyzed the US job market, and the identified change in the structure of the workforce by Goos et al. (2009), who conducted empirical analyses for European job markets. Researchers even go further and theorize about de- serted manufacturing or even an entire robot economy (Sachs et al., 2015, Benzell et al., 2014). Hemous and Olsen (2014) also construct an endogenous growth model of directed technical change with automation and horizontal innovation and show that an increase in low-skill wages increases incentives to automate the tasks conducted by low-skilled workers crowding out low-skilled employment in the long-run. Would this be a thread to the minimum wage policy of governments?

One of the major threats to the demand for human labor is considered to be artifi- cial intelligence. Elon Musk, co-founder of Tesla Motors Inc. and visionary innovator along with scientists Nick Bolstrom, contributor to the notion of existential risks and the former president of the Royal Society Lord Martin Rees, founder of the Centre for the Study of Existential Risk, share the same reservations towards the continuously improving level of the substitute for human intelligence. The ability of algorithms in understanding languages, recognizing images and the like (The Economist, 2015) has drastically increased over the last two decades of heavy Internet usage. Millions of users deliver the so called big data which enables machines to learn via algorithms to increase accuracy in certain fields only humans were able to execute before. Basically, machine learning via algorithms can also be seen as an indicator for job automation in distinct ar- eas of employment, as machines gather more information and can use this via brute force.

Figure 2: Share of Capital Stock younger than five years in German Manufacturing

illustration not visible in this excerpt

Source: Own calculation based on German Federal Statistical Office - 2015

Assuming that new technologies are embodied in newly acquired capital it is worthwhile looking at the age structure of the capital stock². In figure 2 the capital stock younger than five years in German manufacturing is mapped. The calculations, explained in detail in section 4, provide a picture that shows a fluctuating age of the younger capital stock, first declining, also due to calculation biases, but rising towards 2013. Noticeably, capital stock fluctuates between 25 and 32% during the last 20 years. The change rates of employment mentioned in figure 3 show a slack during the 1990s with decreasing employment in all three sectors. In the transition to the 21 st century, a steep increase of employment in the information technology sector can be observed. This volatile increase fits to the development of the dotcom bubble that burst in 2000. From then on, information technology, transportation and warehousing show cyclical changes. Manufacturing, however, needs longer until 2005 to create new demand for employment, interrupted by the financial crisis in 2009. Overall, figure 3 depicts the ongoing structural change from a manufacturing towards a service oriented economy.

Figure 3: Change of Employment in Three German Sectors

illustration not visible in this excerpt

Source: German Federal Statistical Office - 2015

Increasing Productivity - Increasing Inequality?

The majority of US researchers in the field of labor market inequality relate technological change to this phenomenon. Autor et al. (2006) find a polarization of the US labor market with employment being driven either towards the upper or the lower end of the income distribution. The traditional middle-wage jobs are being more and more substituted by computerization. Katz and Autor (1999) discuss the wage structure and

Robert Solow (Jorgenson, 1966, Hercowitz, 1998), it is not yet fully solved whether it is not reasonable to imply embodied technical change. If it holds, up-to-date capital should be more effective and increase labor productivity. overall US and OECD earnings inequalities. They analyze the effects of several variables influencing the labor markets, i.e. skill-biased technological change, globalization forces, changes in demographics and more.

Figure 4: Labor Productivity in Germany

illustration not visible in this excerpt

Source: German Federal Statistical Office - 2015

Taking labor productivity as an indicator for technological change, figure 4 depicts a steady increase in this measurement since 1970. The time series is split due to measurement revisions made necessary by the reunification of Germany in 1990. Noticeably, the financial crisis in 2009 has a considerable impact on the productivity measurement, as aggregate output and employment data were diminished during that time. Nevertheless, we can see that an efficiency increase per worker is continuously proceeding as time moves on. To see whether this productivity increase affects the well being of workers negatively, we next focus on the labor’s share of income.

Karabarbounis and Neiman (2014), Azmat et al. (2012) and Brynjolfsson and McAfee (2014) find that a decrease in the labor’s share of income is observable globally due to computerization and the elimination of middle-class jobs. For Germany, the picture is shown in figure 5 - the share of income dedicated to labor is decreasing since the late 1970s. The trend is, again, disturbed by difficulties in measuring consistently the aggre- gation of east and west Germany during the reunification period in 1990. The solid black line indicates the share of income related to wages and the dashed line relates income to corporate profits. While observing a negative trend in wages’ share of income since reunification, corporate profits obtain a dramatically increasing share of income since 2003. This shows that corporate profits increased disproportionate compared to wages. Brynjolfsson and McAfee (2014) interpret the decline of the labor’s share of income in the United States on the basis of two trends. First, they find less employment prevailing

Figure 5: Wage Share of GDP vs. Corporate Profit Share of GDP

illustration not visible in this excerpt

Source: German Federal Statistical Office - 2015

and those who are employed earn less than before. Considering data from the German federal statistical office, this trend is not observable for the German economy as a whole for the period of 1970 to 2014, but for the manufacturing sector, which is analyzed in section 4 further below. Concerning the second trend, observed by the American re- searchers, average gross income did not change significantly over the period of ten years, see figure 7.

Figure 6: Development of Education Level in Germany

illustration not visible in this excerpt

Source: German Federal Statistical Office (2001, 2006, 2010)

Regarding the distribution of the German labor force in figure 6, it is observable that the share of vocationally trained labor declined from 2001 to 2010, from about 67% to about 53%. During that time span, the number of people that obtained a higher educational degree (college and university) doubled during that time span. The amount of labor having no education at all, except schooling, has remained identical but the share of unknown education, which relates to employees where no information about their graduation status is available, has increased from almost 9 % in 2001 to about 15% in 2010. The share of low skilled labor is not growing.

Figure 7: Development of Gross Incomes in Germany

illustration not visible in this excerpt

Source: German Federal Statistical Office - 2001, 2006, 2010

Gross incomes for the time from 2001 to 2010 notably decreased for all education levels except university degrees. By trend, this confirms the observed claim of increased demand for high skilled labor, but which might also be caused by the financial crisis immediately before.

[...]

---

¹ See figure 27 in the appendix for illustration.

² Although the embodiment hypothesis was discussed controversially between Dale Jorgenson and