|
Article Excerpt
THE ROLE OF TECHNOLOGY SHOCKS in business cycle fluctuations has recently received considerable attention. A myriad of papers has emerged on this topic addressing the controversial conclusion reached by Gall (1999) that technology shocks cannot be the main driving force behind cyclical movements in macroeconomic data. This conclusion challenges the core of the long-standing real business cycle (hereafter, RBC) theory; thus, it comes as no surprise that so many recent papers have been written either in defense of or to challenge Gali's findings (see Gali and Rabanal 2004 for a review of the literature, Basu, Fernald, and Kimball 2006 for an alternative approach that yields similar findings, and Fisher 2006 who proposes an alternate technology shock that produces results consistent with the traditional RBC paradigm).
Standard RBC theory teaches that all factor inputs should rise when there is a positive technological innovation. However, recent empirical tests of the theory find that labor input falls in response to a positive shock to technology, a finding that has sparked a debate for the last decade with little resolution. The crux of the debate has to do with the data-generating process assumed for per capita labor input in empirical models. If one were to rely on econometrics, which fails to reject the presence of a unit root in per capita labor, one would be led to enter labor input in first differences when estimating a (structural) vector autoregression (VAR). Entered in differences the results of a typical VAR predict a fall in labor input in response to a positive shock to technology, opposite of that predicted by the standard RBC model. However, common sense tells us that per capita labor being a bounded series cannot have a unit root. For this reason, several papers have assumed per capita labor is stationary and, thus, should enter the VARs in levels. When entered in levels the standard result emerges that labor input rises when there is a positive innovation to technology.
In this paper, we show that there are significant low-frequency demographic and sectoral movements, over the postwar period, that are features of the commonly used measure of hours worked per capita as well as the growth rate of labor productivity. Our premise is that these low-frequency movements have nothing to do with the kinds of technology shocks typically modeled in RBC theory. These low-frequency movements in the standard measures distort unit root tests (which have low power to begin with), make the time series for per capita labor inconsistent over time and with RBC theory, and are the source of conflicting results in the levels versus first difference debate. Previous researchers, such as Flaim (1990), Shimer (1998), and Aaronson et al. (2006), have highlighted the effects of demographics on trends in aggregate series. None, however, has shown that demographic influences can seriously affect the results of structural VARs.
We begin by showing that extraction of these low-frequency movements from hours alone using a conservative Hodrick-Prescott (HP) filter produces results similar to those obtained using first-differenced hours. We then show that an important source of the low-frequency movements in both hours and productivity growth is the movement of the baby boom generation through the labor market. We also show that sectoral changes involving government and nonprofit employment induce trends in the standard measure of hours per capita, which counts only hours worked in private business. We demonstrate these points both empirically and theoretically. Ngai and Pissarides (2008) offer a model that generates a U-shape in market hours based on total factor productivity (TFP) growth across sectors and home production. We show most of the U-shape in the data is simply due to demographics and incomplete counting of market hours.
Once we account for demographics and sectoral shifts, we show consistent implications for the role of technology shocks in business cycle fluctuations. Positive technology shocks, identified with long-run restrictions, lead to a short-run decrease in hours worked regardless of the stationary assumption made for per capita labor. On the other hand, the results are mixed for investment-specific technology shocks. In some specifications, positive investment-specific shocks raise hours while they lower productivity. In the specification that controls for demographics in a less structured way, positive investment-specific shocks raise productivity and lower hours. All specifications, however, have the common feature that hours move in the opposite direction of productivity.
Our results also have broader implications beyond the technology-hours debate. For example, our analysis of institutional and demographic changes explains the findings of Kahn and Rich (2007) and Fernandez-Villaverde and Rubio-Ramirez (2007). Kahn and Rich use a regime-switching dynamic factor model to detect breaks in trend productivity growth. They show that there are two important sources of low-frequency movements in per capita macroeconomic variables: technology and slow movements in labor supply that look like preference shifts. Similarly, Fernandez-Villaverde and Rubio-Ramirez estimate a dynamic stochastic general equilibrium (DSGE) model and find that low-frequency movements in the preference parameter are an important part of fluctuations. Our results suggest that the sources of these movements are demographic changes in the "representative household" and institutional changes in the sectors of employment.
1. THE PROBLEM OF LOW-FREQUENCY MOVEMENTS IN HOURS PER CAPITA
According to RBC models with standard preference specifications, the hours worked per capita variable should be stationary in the absence of permanent shifts in government spending, labor income taxes, and preference shifts. Yet the most widely used measure of private hours per capita shows significant low-frequency movements. Figure 1 shows the behavior of private hours in business divided by the civilian non-institutional population ages 16 and over during the post-WWII period. Hours show a U-shape, with a downward trend until the mid 1970s, which partially reverses by 1997, but then falls again afterward. The peak in 1979 is 14% below the peak in 1948. The low-frequency movements are so pronounced that the series does not return to its mean for decades at a time.
While these low-frequency features are not an issue for analyses that HP filter data before analyzing it, they are very problematic for structural VARs (SVARs) in which assumptions about stationarity are key parts of the identification. In particular, these low-frequency movements in hours per capita have important implications for empirical structural VAR models that identify technology shocks using long-run restrictions. Based on the results of standard unit root tests, Francis and Ramey (2005) assume that hours per capita have a unit root, and thus enter hours in first differences in the model. They find that a positive technology shock leads to a decline in hours worked. In contrast, Christiano, Eichenbaum, and Vigfusson (CEV, 2003) argue that hours per capita cannot logically have a unit root, and offer alternative empirical tests against a unit root. They enter hours in levels and find that a positive technology shock leads to a rise in hours worked, t
[FIGURE 1 OMITTED]
To illustrate, we reestimate the structural VAR used by Gall (1999) and Ramey (2005), and CEV using the...
|
