Has Monetary Policy Accelerated Job Polarization Part 1

Has Monetary Policy Accelerated Job Polarization? Abstract Since the late 1960s, the share of U.S. employment in occupations involving primarily routine tasks has declined by about 30 percent, a trend that has largely affected workers with a high-school degree but no college. This paper argues that contractionary monetary policy has accelerated this change. In part by disproportionately affecting industries with high shares of routine occupations, contractionary monetary policy shocks lead to large and very persistent shifts away from routine employment. Expansionary shocks, on the other hand, have little effect on these industries. Indeed, monetary policy’s effect on overall employment is concentrated in routine jobs. These results highlight monetary policy’s role in generating fluctuations not only in the level of employment, but also the composition of employment across occupations and industries. Keywords: Job polarization; monetary policy; asymmetries; occupational employment; industry employment. 1 1 Introduction Job polarization refers to the “hollowing out” of middle-skill, routine occupations that has occurred in the United States since 1980.1 Over the last 40 years, jobs that involve routine tasks (e.g., assembly line workers, data entry technicians, office administrators) have been replaced with those involving nonroutine tasks (e.g., management and personal services), as illustrated in Figure 1. Autor et al. (2003) and others have argued that this trend is due to routine-biased technological change (RBTC). Routine tasks are amenable to automation, and, in recent decades, employers have substituted away from labor and toward capital to accomplish these tasks. As Jaimovich and Siu (2014) show, and as is evident from Figure 1, the majority of job losses in these occupations since the mid-1980s occurred during, just before, or after the last three NBER recessions. They link this phenomenon to the so-called “jobless recoveries” that accompanied these recessions. They also highlight that the decline was not driven primarily by changes in industry composition. The long-term trend decline in routine employment is even more evident in Figure 2, which depicts how the relative decline in routine-task employment has translated into the share of total employment in routine jobs. In February 1967, routine-task occupations made up 62 percent of employment. By December 2019, that number had dropped more than 20 percentage points to less than 42 percent. In this paper, I argue that monetary policy—specifically, contractionary monetary policy— accelerates this process. In particular, a contractionary monetary policy shock that increases the Fed Funds Rate by 100 basis points (b.p.) produces a persistent decline in the share of employment in routine occupations that peaks at 1 percentage point. An expansionary policy shock of the same size, on the other hand, does not produce a statistically significant change. Moreover, the decline in employment in routine jobs drives almost all of the decline in total employment after a contractionary monetary policy shock. I find that monetary policy shocks account for up to 40 percent of the changes in the share of employment in routine occupations over a two- to three-year horizon. After establishing that monetary policy shocks have large and asymmetric effects on employment in routine jobs (and essentially no effect on nonroutine employment), I explore three possible channels for this pattern. One possibility is that monetary policy shocks affect the relative price of investment in a way that induces substitution toward capital and away 1 This phenomenon is described in detail in Acemoglu (1999), Autor, Levy and Murnane (2003), Jaimovich and Siu (2014), Foote and Ryan (2015), and others. 2 Figure 1: Per capita employment in routine occupations, 1967-2019. Shaded dates are NBER recessions. Figure 2: Share of total employment in routine occupations, 1967-2019. Shaded dates are NBER recessions. 3 from labor, an important mechanism that the job polarization literature has highlighted as an explanation for the trends evident in Figures 1 and 2. I find little evidence that monetary policy works through this channel. In fact the contractionary monetary policy shocks that drive the declines in routine employment increase the price of investment goods. Although there is some evidence of substitution toward existing capital, it is qualitatively different from the type of substitution toward new technologies emphasized in the job polarization literature. Another possible way for monetary policy to have an outsize role on routine-task employment is that its effects differ by industry, and industries differ in the share of employment made up by routine-task jobs. In contrast to Jaimovich and Siu’s (2014) analysis of routine job loss during recessions, I find this industry composition mechanism to be an important driver of the effects of monetary policy shocks on routine employment.2 Monetary policy shocks have larger effects on total employment in construction and durable goods manufacturing, and employment in each of these industries is concentrated in routine-task occupations.3 By having stronger effects overall on those industries that utilize a greater share of routine-task employment—and even stronger effects when policy is contractionary— monetary policy has persistent effects on the mix of occupations in economy and potentially disproportionate welfare consequences for the “middle-skill” workers affected. Finally, I present some evidence of the drivers of these differing effects by industry. Contractionary shocks lead to large and persistent appreciation of the dollar and declines in exports volume (especially of durable goods), as well as increases in mortgage rates and declines in housing starts and residential investment. The former is related to the role of trade emphasized in the job polarization literature (cite papers), and the latter points to the important role of the housing market in the transmission of monetary policy to the real economy (cite papers). Nevertheless, this industry effect does not explain the asymmetry per se; monetary policy has asymmetric effects on employment in virtually all industries. The source of this asymmetry is likely the persistent endogenous effect of monetary policy. 2 It is worth emphasizing that these results are not contradictory. I find only that industry composition is important in accounting for the effects of monetary policy, while they show that long-run changes in the industry makeup of the U.S. economy cannot explain the majority of job polarization. Foote and Ryan (2015), on the other hand, do argue that the decline of U.S. manufacturing employment played an important role in job polarization. 3 In 1972, the earliest year for which annual occupation-by-industry data are available from the CPS, more than 80 percent of employment in these industries was in routine employment, and these industries made up 17.2 percent of all nonroutine employment. 4 In Section 2, I review the recent literature on job polarization and discuss Autor et al.’s (2003) occupation classification system used throughout the paper. In Section 3, I describe the data on employment and monetary policy I use in subsequent sections. In Section 4, I discuss the estimation of linear and asymmetric impulse responses for occupational employment data and present results. In Section 5, I discuss the historical contribution of monetary policy shocks to job polarization. In Section 6, I discuss possible reasons for the results from the previous section. Section 7 concludes. 2 Literature review and occupation classifications The literature on job polarization grew out of an earlier literature on earnings inequality, skill-biased technical change (SBTC), and international trade.4 Research on these issues typically focused on two groups—low- and high-skill workers—and studied the differential effects of SBTC, off-shoring, and immigration on these types of workers. Autor et al. (2003) introduced a more nuanced categorization of jobs with the “taskbased” framework. Their system highlights the tasks involved in performing a job, rather than the characteristics of the person holding that job. They delineate occupations along two dimensions: manual vs. cognitive, and routine vs. nonroutine. The first category describes whether the job’s tasks are primarily physical or mental; the second, whether the job consists of “carrying out a limited and well-defined set of ... activities, those that can be accomplished by following explicit rules” (Autor et al. (2003)). They document a compositional shift away from routine occupations beginning in the 1970s, coincidental with the start of rapid computerization. Indeed, the change occurred most rapidly in those industries that more quickly adopted computing technology. Computers substituted for labor in performing routine tasks, while they complemented labor in nonroutine—escpecially nonroutine cognitive—occupations. Subsequent research noted that routine occupations, both cognitive and manual, were held by workers in the middle of the wage and educational distribution. Cognitive nonroutine jobs (such as journalists, professors, or CEOs) were typically held by high-wage, high-education workers, while low-wage, low-education workers were in manual nonroutine jobs (such as janitors, barbers, groundskeepers). The growth of low-skill and high-skill jobs and simultaneous decline of middle-skill jobs, led to Goos and Manning’s (2007) introduction of the term “job polarization.” The trend of job polarization has since been documented in 4 See Autor and Katz (1999) for an overview. 5 many European countries and has been confirmed in the U.S. across numerous datasets and methodologies.5 More recently, researchers have begun to examine the cyclicality of job polarization. Jaimovich and Siu (2014) document that the decline in middle-skill jobs in the U.S. occurred almost entirely during recessions and never recovered during subsequent expansions. They show that this phenomenon can partially explain the so-called “jobless recoveries” after the three most recent recessions. Foote and Ryan (2015) also examine middle-skill employment over the business cycle and argue that its cyclicality is driven by middle-skill jobs’ concentration in highly cyclical industries like construction and manufacturing. They also argue job polarization can partially explain recent declines in labor force participation, especially for men. This paper links the asymmetry documented by Jaimovich and Siu (2014)—that routine employment falls during economic downturns but does not recover—with the composition channel that Foote and Ryan (2015) study—that routine jobs are concentrated in cyclical industries—and examines them in the context of monetary policy. The approach taken here is closely related to the literature examining asymmetric and state-dependent effects of shocks. While much emphasis has been given to the state-dependence of fiscal and tax shocks,6 a more recent literature has considered the state- and sign-dependence of monetary policy shocks. Tenreyro and Thwaites (2016) find that monetary policy shocks have smaller effects on the real economy during recessions than in booms, and that contractionary shocks have larger effects than expansionary ones. Angrist, Jordà and Kuersteiner (2016) and Barnichon and Matthes (2017), using very different methods, also find that contractionary monetary policy shocks have larger effects. This paper confirms these findings with respect to employment and highlights that the asymmetry exists only in employment in routine occupations. The literature on the real effects of monetary policy is too long to review in detail here7 ; in the next section, however, I describe in detail the two empirical approaches I take. 5 Autor (2015) provides an excellent overview of this research. See for example Auerbach and Gorodnichenko (2012a) and (2012b), Owyang, Ramey and Zubairy (2013), and Ramey and Zubairy (2017). 7 See Christiano, Eichenbaum and Evans (1999) for an overview of the early literature on monetary policy shocks, and Ramey (2016) for an overview of more recent developments. 6 6