2014 Theses Doctoral
General Purpose Technologies: engines of change?
This dissertation examines the relevance of technology in explaining structural and cyclical changes in the labor and product markets. The first chapter focuses on computers and the labor market, the second chapter on electricity and the labor market, and the third chapter focuses on computers and the goods market. The three chapters rely on a General Purpose Technology and on the distinction between routine and nonroutine jobs. A General Purpose Technology has three characteristics: it has pervasive use in all industries, it improves over time, and it is able to foster other innovations. This dissertation considers the technology of computers in the second half of the 20th century and electricity in the first half. It defines "pervasive use" as the technology's ability to substitute for some types of jobs--called routine jobs--more than others. Routine jobs consist of repetitive tasks, follow an explicit set of rules, and can easily be automated by the technology. Nonroutine jobs are the remaining jobs, which cannot be easily replaced by the technology. For computers, examples of routine jobs are clerks and secretaries, since their work can be automated with information processing software, whereas examples of nonroutine jobs are managers and health aides, since their work requires creativity or personal interactions. For electricity, examples of routine jobs are laborers on the factory floor, since their work can be automated by the conveyor belt, whereas examples of nonroutine jobs are foremen and engineers, since their work requires attention or detailed calculations. The distinction between routine and nonroutine jobs depends on the technology: accountants can be nonroutine relative to electricity and routine relative to computers. The first chapter examines computers as a theoretical explanation for changes in the US labor market in recent decades. When computers become cheap and competitive compared to workers, they diffuse more rapidly and become more important in the conventional mechanism of capital-labor substitution. The model can account for recent structural changes with this trend of automation: employment has shifted away from routine occupations and the labor share of income has declined. The model also predicts that recessions accelerate the decline in routine occupations--firms prefer to destroy routine jobs during a downturn, when the opportunity cost of restructuring is low. This acceleration can account for recent cyclical changes of the labor market: routine job losses are concentrated in recessions and the ensuing recoveries are jobless.It asks whether computer adoption can account for several changes in the US labor market since the 1980s, such as the shift away from routine occupations, the decrease in the labor share of income, and jobless recoveries. As computers become better and cheaper, firms substitute away from routine occupations and into computers. This substitution becomes quantitatively more important when computers are cheaper--if computers are becoming cheaper but are still too expensive, firms hire workers in routine occupations instead and adjust their demand for computers by a small amount. This level effect is the reason why the price of computers has been decreasing since 1950 but the effects on the labor market start in the 1980s. The trend of automation when computers become sufficiently cheap can account for the structural changes in the labor market: employment shifts away from routine occupations, the labor share of income decreases, and the capital share of income increases. Furthermore, a friction in the labor market such as hiring costs implies that recessions accelerate the decline in routine occupations, even as nonroutine occupations are unaffected. Firms know that nonroutine jobs complement the technology and are increasing: if firms destroy nonroutine jobs in the recession, they will need to create them again in the recovery and pay a hiring cost. So firms prefer to hoard workers in nonroutine occupations during the recession and the burden of adjustment falls on routine occupations, which do not entail a hiring cost since their trend is declining. In the recovery, firms do not hire workers in nonroutine occupations, since they did not fire them during the recession. Workers in routine occupations may be hired back but not back to the peak level because their trend is declining. Total employment is stagnant even as output recovers, which is the definition of a jobless recovery. The second chapter examines the labor market and electricity in the first half of the 20th century. The 1920s and 1930s witnessed large changes in the US labor market, with a shift away from dexterity-intensive occupations, a productivity speedup, and low job creation. The second chapter asks whether the model of the first chapter, which explained labor market changes since the 1980s with the adoption of computers, can also explain labor market changes in the 1930s with the adoption of electricity. It supports the model's main assumption by empirically testing the model's prediction for the labor share of income. The identification strategy uses a state's initial loading on the technology to generate electricity--hydroelectric power or coal power--as an instrument for changes in the price of electricity. It also uses a newly digitized dataset for the concrete industry from 1929 to 1935 to provide plant-level measures of the labor share of income. Technical progress in electric utilities caused a decrease in the labor share of income of the downstream industry of concrete. This result supports the mechanism in the model, which can in turn explain other features of the 1920s and 1930s: structural changes in employment, a productivity speedup, and a weak recovery of employment after the Great Depression.Unlike computers, electricity prices vary across states depending on the geography of the power source--hydroelectric power or coal power--and have different trends. Hydroelectric power was highly efficient from the start and provided few opportunities for cost-saving innovations, so states like California have cheap electricity but the price of electricity is constant. Coal power was relatively inefficient and had many opportunities for cost-saving innovations, so states like New Jersey have initially expensive electricity but the price of electricity decreases over time. A state's initial loading on the coal technology is an instrument for the supply-side changes in the price of electricity. This first-stage of the regression allows identification of the causal effect on the labor market of downstream industries. The chapter focuses on one prediction of the model and asks whether states where the price of electricity decreased more also decreased their labor share of income. It uses labor market variables from the concrete industry from 1929 to 1935, digitized for the first time for this dissertation. Concrete plants sell a non-traded product and locate near their customers, not near cheap electricity, so their location decisions are orthogonal to the geography of power sources and support the validity of the instrument. The chapter finds that, consistent with the mechanism in the model, concrete plants decreased the labor share of income in response to the decrease in the price of electricity. The third chapter examines the behavior of consumption in the second half of the 20th century. The recoveries from the last three recessions in the United States were not only jobless, they were also slow. The growth rate of output and consumption after the trough of the business cycle is twice as small for the last three recessions compared to previous ones. This chapter asks whether the structural decline in employment of routine occupations can also account for recent slow recoveries in consumption. It assumes that workers in nonroutine occupations are optimizing agents who can smooth consumption by saving, whereas workers in routine occupations are hand-to-mouth agents who consume all of their income. Before the 1980s, workers in routine occupations can easily find another routine job right after the recession, so consumption decreases in the recession and "bounces back" in the recovery. After the 1980s, workers in routine occupations need to go through a period of retraining in order to find a new job, so the recovery of consumption is delayed until they finish retraining. In a simulation of the model, the recovery of consumption is twice smaller after the 1980s than before, which suggests that this mechanism may be quantitatively important in explaining recent slow recoveries.
- Morin_columbia_0054D_12051.pdf binary/octet-stream 2.77 MB Download File
- Academic Units
- Thesis Advisors
- Reis, Ricardo
- Ph.D., Columbia University