|
COPYRIGHT 2001 MIT Press Journals
I. Introduction
BEGINNING with Friedman (1968), the natural rate of unemployment (hereafter, used synonymously with the nonaccelerating inflation rate of unemployment, or NAIRU) has often been defined as the unique level of the unemployment rate such that, after controlling for supply shocks, inflation remains unchanged when actual unemployment equals that rate. The large and sustained changes in unemployment rates that have occurred in many industrialized economies since the early 1970s have focused attention on econometric models that allow for variation over time in the NAIRU. From the perspective of economic policy, a particularly important question is how precisely the NAIRU can be measured. This article presents estimates of the NAIRU for seven economies based on several versions of an unobserved-components model, and investigates the consequences of different model specifications for the precision of the resulting NAIRU estimates.
Numerous econometric models have been used to obtain estimates of the NAIRU that allow for time variation. The NAIRU is alternatively modeled as a deterministic function of time (Staiger, Stock, & Watson, 1997a, 1997b; Cross, Darby, & Ireland, 1997), as an unobserved stochastic process (King, Stock, & Watson, 1995; Staiger et al., 1997b; Gordon, 1997), or as a function of demographics or labor market variables (Weiner, 1993; Staiger et al., 1997b). This article explores the ability of the second approach, in which the NAIRU is treated as an unobserved stochastic process, to deliver precise estimates of the NAIRU for the G7 excluding Japan, and Australia, over the period from 1971 to 1998. The approach is based on the presumption that the determinants of the NAIRU are unknown but persistent. The presumption that the determinants are unknown (hence, modeled as random) seems appropriate in a cross-country study, in which different factors may cause changes in the NAIRU in different countries. In addition, there seems to be no consensus in the literature on which factors are responsible for changes in the NAIRU.(1) Assuming that among these factors are demographics, technological changes, and so forth, motivates modeling them as persistent.
The model as used in previous work consists of the Phillips relation linking unexpected inflation to the deviation of the unemployment rate from the NAIRU (the "unemployment gap"), and an equation specifying the NAIRU as a random walk without drift. This article extends the basic model in two steps. First, specifying the NAIRU as a random walk without drift seems a plausible assumption for U.S. unemployment data over the samples considered in the aforementioned studies. A look at the unemployment rate series used in this study, presented as solid lines in figure 1, suggests that, for most of the countries, excluding a drift from the NAIRU specifications may not be justified. Accordingly, results are presented for specifications in which the NAIRU is modeled as a random walk with drift. Because no pattern of mean reversion in the drift is apparent from the series in figure 1, the drift itself is assumed to be a random walk.
[ILLUSTRATION OMITTED]
The second dimension in which the basic model is extended is to impose some structure on the unemployment gap. Without this additional structure, the value of the unemployment gap--and hence the position of the NAIRU--is inferred from (unexpected) inflation only. By assuming some stationary process for the unemployment gap, changes in the unemployment rate itself yield information about the NAIRU. The assumption that the unemployment gap has a tendency to revert to its mean (zero) over time can be viewed as a consequence of Friedman's 1968 formulation of the natural-rate hypothesis, that the unemployment rate can be kept away from its natural rate only by ever accelerating inflation or deflation. The inflation data displayed as dotted lines in figure 1 do not show any signs of explosive behavior, and one would therefore conclude that the unemployment gap must have averaged close to zero.
In general, the results presented in this article confirm those of Staiger et al. (1997b) for the United States, and of Cross et al. (1997) for the G7 economies, that the measured uncertainty around the NAIRU estimates is large. However, there is substantial variation in the precision of the estimates, both across countries and across specifications. Across countries, NAIRU estimates for the United States are by far the most precise, whereas European data are uninformative about the existence, let alone the level, of the NAIRU. Across specifications, allowing for a drift in the NAIRU leads to a dramatic decline in the precision of the estimates when no additional structure is imposed on the unemployment gap, and the opposite is true once the unemployment gap is assumed to follow an AR process. The finding that information from unemployment data greatly improves the precision of the NAIRU estimates raises the question whether these estimates are in fact linked to inflation, or just smoothed unemployment series. The evidence presented in the article suggests that the latter interpretation is not warranted for the United States and possibly Canada, but that it cannot be rejected for most of the European economies.
The remainder of the article is organized as follows. The specifications of the unobserved components model are introduced in the next section, and estimation issues are discussed. Section III presents the estimation results, and section IV offers some conclusions. Details of the data used in this article are presented in appendix A.
II. The NAIRU as Unobserved Stochastic Process
The purpose of this article is to report results from estimating several related econometric specifications of the NAIRU. The specifications are discussed in the first subsection. The second subsection addresses issues pertaining to the estimation of the various specifications.
A. Specifications of the NAIRU Model
In this article, as in most articles cited in the introduction, the NAIRU is inferred from a Phillips curve-type regression of the following form:
(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
where [[Pi].sub.t] and [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] denote realized and expected inflation, [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] denotes the NAIRU at time t, [X.sub.t] is a vector of variables capturing supply shocks (in this article changes in the nominal exchange rate and in commodity prices), and the disturbance [[Epsilon]].sub.t] is assumed to be i.i.d, normal with mean zero and variance [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. Two measures of inflation are considered, based on the all-items CPI and the GDP deflator, respectively.
One feature of specification (1) is that all right-side variables enter as lagged, not contemporaneous values. Because the model is estimated using single-equation methods, it is important to avoid simultaneity issues. In this respect the specification (1) is closest to those in King et al. (1995), and Staiger et al. (1997b), although the latter allows supply shock variables to enter contemporaneously. Gordon (1997), by contrast, allows the unemployment rate to enter contemporaneously, which requires that there be no contemporaneous feedback from inflation to unemployment.(2) The view taken in this article is to interpret equation (1) as a reduced-form, not as a structural equation.
A related issue is that any specification of inflation expectations in equation (1) is not model endogenous, and therefore somewhat arbitrary. Because for all the countries and measures of inflation examined in this article, the hypothesis of a unit root is not rejected at the 5% level by augmented Dickey-Fuller tests, this article follows the simplest approach by setting [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], which implies that...
Read the full article for free courtesy of your local library.
|
Bookmark this article
Print this article
Link to this article
Email this article
Digg It!
Add to del.icio.us
RSS
