Printer Friendly

The unemployment paradigms revisited: a comparative analysis of U.S. state and European unemployment.

I. INTRODUCTION

The persistent rise in unemployment levels recorded in the developed world, particularly in continental Europe, ever since the first great oil shock has brought about a renewed interest among macroeconomists and labor economists in understanding the dynamic behavior of unemployment rates. It is often argued that major postwar macroeconomic shocks such as the slowdown in productivity growth (Hoon and Phelps, 1997), the steep rises in oil prices in the 1970s (Carruth, Hooker, and Oswald, 1998) and the changes in world interest rates (Phelps, 1994) could be responsible for the persistent rise in unemployment. However, this provides an incomplete picture since the U.S. unemployment rate--which has also been subject to these shocks--has reverted back to preshock levels rather than steadily rising as in Europe. Thus, authors have turned their attention to a line of argument related to the degree of labor market flexibility. Blanchard and Wolfers (2000) and Phelps and Zoega (1998), among others, have claimed that more stringent job-security legislation, higher impediments to hiring and firing, higher levels and duration of unemployment benefits, and a higher degree of wage indexation in the European Union (European Union) relative to the United States may have contributed to the marked differences in unemployment levels recorded in both economic areas.

At the theoretical level, three main hypotheses have been entertained in accordance with the evolution of unemployment levels over the postwar era. The traditional natural rate theory pioneered by the work of Friedman (1968) and Phelps (1967, 1968) holds the view that the unemployment rate tends to fluctuate around some equilibrium level associated with a fully equilibrated labor market where all adjustments have worked themselves out. This natural rate is set depending on fundamentals of the economy, which are taken to be exogenous. Shocks to unemployment are assumed to be temporary, which implies that unemployment is path independent and reverts to its preshock level. However, the difficulties in interpreting European unemployment movements since the first oil shock in terms of the natural rate theory gave rise to two alternative views: the structuralist and hysteresis hypotheses.

The structuralist hypothesis--see Phelps (1994) and Phelps and Zoega (1998)--tries to explain the rise in European unemployment through the adjustment to an underlying equilibrium rate of unemployment which has increased in response to structural factors of the labor market and the economy in general. Some of the factors trying to explain the "movement" of the natural rate of unemployment are labor productivity, technological change, world real interest rates, real exchange rates, stock prices, and energy prices. Other factors specific to the functioning and adjustment of labor markets are the level and duration of unemployment benefits and employment protection legislation. According to the structuralist view, most shocks to unemployment are temporary, but infrequently large variations in the aforementioned structural factors occur, which lead to shifts in the now endogenous natural rate of unemployment.

In contrast to the natural rate and structuralist approaches, the hysteresis hypothesis--see Blanchard and Summers (1986, 1987) that focused on insider-outsider dynamics in wage formation--implies that the unemployment rate is path dependent as its current level shows high dependence on past levels. As a result, temporary shocks can affect unemployment permanently. (1) Hysteresis' defenders claim that the key to the rise in European unemployment should not be sought on the occurrence of adverse supply shocks--such as the productivity slowdown or the oil crises--or demand shocks--such as those caused by tight monetary policies leading to sharp rises in interest rates. Instead, the answer is in the way countries adjust to shocks. For these authors, unemployment rates can remain at a new higher level indefinitely even if the recession causing the unemployment rise has ended. So every policy measure geared to increase labor market flexibility, thus making the labor market more resilient in the aftermath of adverse shocks, should be welcome. (2)

In order to discriminate among these competing paradigms that try to characterize the distinct behavior of unemployment rates in European countries and the United States, unit root and stationarity tests have been widely used in applied work. In essence, the hysteresis hypothesis is formulated as a unit root process and its rejection gives support to the natural rate hypothesis--if no breaks are included in the specification which implies mean stationarity in unemployment--or the structuralist hypothesis--if mean shifts are included which implies regime-wise stationarity in unemployment.

The objective of this article was to investigate which theoretical paradigm most closely represents the behavior of unemployment rates in the states of the United States and the countries forming the European Union-15 over the past three decades. By establishing the time series properties of the unemployment rates in both economic areas, a recommendation can be made as to which can be considered as an appropriate set of to Ordinary Least Squares to fight unemployment. As a matter of fact, stabilization policy can affect the unemployment rate permanently in the event of hysteresis, while temporarily if the unemployment rate is stationary. In addition, policy measures aimed at providing greater labor market flexibility will help more in the case of hysteresis.

Much of the empirical literature on the analysis of the time series properties of unemployment rates has used univariate unit root tests of the augmented Dickey-Fuller type (Augmented Dickey-Fuller Type), which normally lack power to discriminate between a nonstationary and a near unit root process. A recent avenue to increase power has been the use of panel unit root tests that exploit the cross-sectional variability of unemployment rate series. Applying the test by Levin, Lin, and Chu (2002, LLC) to the unemployment rates of a sample of 48 U.S. states and 15 countries of the Organisation for Economic Co-operation and Development (Organisation for Economic Co-operation and Development) over the postwar era, Song and Wu (1997, 1998) are able to reject the null of nonstationarity in both cases. Using the test by Im, Pesaran, and Shin (2003, IPS), Leon-Ledesma (2002) confronts 50 U.S. states plus the District of Columbia with 12 European Union countries and finds that the most plausible hypotheses are hysteresis for European Union countries and the natural rate for U.S. states. However, these studies are based on panel unit root tests that do not explicitly allow for cross-sectional dependence across individuals, which is likely to lead to dramatic size distortions (Strauss and Yigit, 2003). In addition, none of these studies have considered the issue of structural breaks in the trend function of the unemployment rate, which according to Perron (1989) could bias the results toward the nonrejection of the unit root hypothesis. The only exception to this is Romero-Avila and Usabiaga (2007) that applied the panel Lagrange Multiplier (Lagrange Multiplier) unit root test allowing for up to two breaks in the mean unemployment rate of U.S. states. Their results point toward the rejection of the nonstationary null in favor of stationarity around a broken mean. However, that analysis presents two main caveats: it cross-sectionally demeans the data, which constitutes a very restrictive form of cross-correlation and only allows for a maximum of two breaks. Thus, by employing a panel stationarity test that allows for both multiple breaks and cross-sectional dependence, we may be able to draw definitive conclusions regarding the comparison between the dynamic behavior of U.S. state unemployment and the European Union unemployment. (3) Overall, our analysis renders clear-cut evidence in favor of regime-wise stationarity in U.S. state unemployment, while hysteresis in European Union unemployment.

This article extends the empirical literature in this field by exploiting the cross-sectional variability of the data through the recently developed panel stationarity test of Carrion-i-Silvestre, del Barrio, and Lopez-Bazo (2005, CBL hereafter), which allows for a highly flexible trend function by incorporating an unknown number of level shifts. In addition, unlike previous studies, we will carry out a formal analysis of the prevalence of cross-sectional dependence in unemployment innovations by applying the test for cross-dependence (Cross-Dependence) recently developed by Pesaran (2004). As a further departure from most studies in this field, we will allow for more general forms of cross-sectional dependence by simulating the bootstrap distribution of the panel stationarity test with multiple breaks, thereby correcting for finite sample bias.

The plan of the rest of the article is as follows. Section II briefly describes the data and the methodology of the tests employed in this article. Section III presents the results of the analysis of the time series properties of unemployment rates for U.S. states and European Union countries. Section Instrumental Variable pays special attention to explaining the breaks and puts forward some policy implications and Section V concludes.

II. DATA AND METHODOLOGY

A. Data Description

We employ seasonally adjusted monthly data on unemployment rates for 50 U.S. states plus the District of Columbia over the period 1976-2004, which were obtained from the Bureau of Labor Statistics of the U.S. Department of Labor (2005). For the unemployment rate of the European Union countries, we use seasonally adjusted quarterly data over the period 1970-2004 obtained from the Organisation for Economic Co-operation and Development Economic Outlook. We remove Greece from the sample since the unemployment rate series for the Greek economy was largely incomplete. (4) The main differences between the coverage of our data set and that analyzed by Leon-Ledesma (2002) are the following: (1) we extend the span of the data set to cover the period 1976-2004 for U.S. states and 1970-2004 for European Union countries rather than 1985-1999 and (2) we add to the analysis Austria and Luxembourg.

B. Methodology

Pesaran's (2004) Cross-Dependence test is based on the average of pair-wise correlation coefficients ([^.[rho].sub.ij]) of ordinary least squares (Ordinary Least Squares) residuals obtained from standard Augmented Dickey-Fuller Type regressions for each individual. The test is given by:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (1)

where T and N refer to time dimension and state/country dimension, respectively. The Cross-Dependence statistic tests the null of cross-independence, is distributed as a two-tailed standard normal distribution, and exhibits good finite sample properties.

The CBL test allows for an unknown number of structural breaks under the null hypothesis of stationarity for all panel members. This test is a generalization of the panel stationarity test of Hadri (2000) for the case of multiple breaks. Let the unemployment rate ([U.sub.i,t]) be the stochastic process, which under the null hypothesis is characterized by:

(2) [U.sub.[i,t]] = [[alpha].sub.[i,t]] + [[epsilon].sub.[i,t]] and [[alpha].sub.[i,t]] = [[m.sub.i].summation over ([k = 1])][[gamma].sub.[i,k]][DU.sub.[i,k,t]] + [[alpha].sub.[i,[t-1]]] + [[upsilon].sub.[i,t]],

where [[upsilon].sub.[i,t]] [right arrow] i.i.d. (0, [[sigma].sub.[upsilon],t.sup.2]) and [[alpha].sub.i,0] = [[alpha].sub.i] represent a constant. {[epsilon].sub.[i,t]]} and {[[upsilon].sub.[i,t]]} are assumed mutually independent. The dummy variable for the change in level is given by [DU.sub.[i,k,t]] = 1 for t>[T.sub.[b,k]].sup.i] and 0 otherwise, with [T.sub.[[b,k].sup.i] denoting the kth break location for the ith individual, for k = 1, ..., [m.sub.i], [m.sub.i] [greater than or equal to] 1. Using back substitution, the set of stochastic processes given by Equation (2) can be rewritten under the null hypothesis of stationarity as follows:

(3) [U.sub.[i,t]] = [[alpha].sub.[i,0]] + [[m.sub.i].summation over ([k = 1])] [[theta].sub.[i,k]][DU.sub.[i k,t]] + [e.sub.[i,t],

where [e.sub.[i,t]] = [[sigma].sub.t.sup.t] = 1 [[upsilon].sub.[i,t]] + [[epsilon].sub.[i,t]]. Thus, if [[sigma].sub.[[upsilon],i]].sup.2] = 0, [e.sub.[i,t]] reduces to [[epsilon].sub.[i,t] which is stationary. Therefore, the panel stationarity test with multiple breaks considers the null hypothesis of panel regime-wise stationarity implied by [[sigma].sub.[[upsilon],i].sup.2] = 0 for all i versus the alternative that [[sigma].sub.[[upsilon],i].sup.2] > 0 for some i. The panel stationarity test is computed as the average of univariate Kwiatkowski, Phillips, Schmidt, and Shin (1992, KPSS hereafter) tests:

(4) [eta][^.[lambda]] = [N.sup.-1] [N.summation over (i = 1)] ([[eta].sub.i]([[^.[lambda]].sub.i])),

where [[eta].sub.i]([[^.[lambda]].sub.i]) = [[^.[omega]].sub.i.sup.-2] [T.sup.-2] [[SIGMA].sub.[t=1].sup.T] = [[^.S].sub.[i,t].sup.2] is the univariate KPSS test for individual i and [[^.S].sub.[i,t]] = [[SIGMA].sub.[j=1].sup.t] [[^.e].sub.[i,j]] stands for the partial sum of the estimated Ordinary Least Squares residuals from Equation (3). [[^.[omega]].sub.i.sup.2] represents an autocorrelation and heteroskedasticity consistent estimate of the long-run variance of the residuals, which is computed nonparametrically following the method proposed by Kurozumi (2002) using the quadratic spectral kernel with fixed bandwidth.

Since the test is dependent on the location of the breaks ([[lambda].sub.i]), which is unknown, we estimate it for each unit using the procedure of Bai and Perron (1998), which is based upon the global minimization of the sum of squared residuals. Once the dates for all possible [m.sub.i] [less than or equal to] [m.sup.max] for each i are estimated, where [m.sup.max] is the maximum number of breaks, we select the appropriate number of structural breaks using the sequential procedure of Bai and Perron (1998). (5) We then compute the normalized test statistic as follows: (6)

(5) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [bar.[zeta]], and [bar.[zeta]].sup.2] are computed as averages of individual means and variances of [[eta].sub.i]([[lambda].sub.i]). Since the assumptions of cross-independence and asymptotic normality are unlikely to hold, we compute the bootstrap distribution of the panel statistic following Maddala and Wu (1999). (7)

III. EMPIRICAL RESULTS

Traditional panel unit root tests are derived under the assumption of cross-sectional independence of innovations, which is unlikely to hold in practice given the high degree of inter-dependencies across units. To shed some light on this issue, we apply the Cross-Dependence statistic of Pesaran (2004) to innovations in unemployment rates for the panel of U.S. states and European Union countries. For each unit i, we compute Ordinary Least Squares residuals from Augmented Dickey-Fuller Type regressions whose optimal lag order is determined using a step-down approach with a maximum lag order of 8. As shown in Table 1, the null hypothesis that unemployment innovations are cross-sectionally independent is strongly rejected for both the panel of U.S. states and the panel of European Union countries. This result is plausible for U.S. states and also for the sample of industrialized countries considered here, which show a high degree of dependence due to trade links, a high degree of capital mobility and common shocks such as the oil embargos of the 1970s.
TABLE 1
Cross-Sectional Dependence Test

 US States European Union Countries

Cross-Dependence test 30.881 (a) 3.535 (a)
p Value 0.000 0.000

Notes: The Cross-Dependence statistic tests for the null of
cross-sectional
independence and is distributed as a two-tailed standard normal
distribution.
(a) implies rejection of the null hypothesis at 1% level.


Before presenting the results from the panel stationarity test with multiple breaks, we first report the results from the application of the panel stationarity test without breaks. As shown in Table 2, the evidence points to the presence of a unit root in the unemployment rate for both the European Union countries and the U.S. states. This occurs irrespective of the assumption regarding cross-sectional independence. Thus far, our results are supportive of the hysteresis hypothesis.
TABLE 2
Panel KPSS Stationarity Test without Structural Breaks (Hadri, 2000)

 Bootstrap Critical Values

 Test p Value 10% 5% 2.5% 1%

Part A: U.S. state unemployment rates
 Panel KPSS test 34.530 (a) 0.000 5.990 9.490 13.042 18.308

Part B: EU countries' unemployment rates
 Panel KPSS test 13.716 (a) 0.000 3.165 4.833 6.756 9.454

(a) implies rejection of the null hypothesis at the 1% significance
level.


However, as stressed by Perron (1989), failure to allow for structural breaks may have favored the unit root hypothesis. To deal with this issue, we now proceed to present the results from the panel stationarity test of CBL, which allows for multiple level shifts as well as for cross-sectional dependence through bootstrap methods. Table 3 reports the results for the panel of U.S. states. Panel A displays the individual KPSS tests allowing for a maximum of five structural breaks upon which the panel KPSS statistic with multiple breaks is based. In order to control for finite sample bias that could arise from assuming asymptotic critical values even for moderate sample sizes, we will simulate finite sample critical values for the univariate KPSS tests with multiple breaks by means of Monte Carlo simulations using 20,000 replications.
TABLE 3 Panel KPSS Stationarity Test with Structural Breaks for
Unemployment Rate in U.S. States, 1976(1) to 2004(12)

State KPSS Test [m.sub.i] Mean [[^.T].sub.b,1.sup.i]
 before the
 Break

Panel A: State-Specific Test

Alabama 0.063 4 6.56 10.13, 1980 (6)

Alaska 0.058 3 9.21 9.93, 1982 (3)

Arizona 0.047 3 7.74 6.10, 1983 (8)

Arkansas 0.071 3 6.37 8.67, 1980 (4)

California 0.056 4 7.52 8.68, 1981 (7)

Colorado 0.041 4 5.65 6.91, 1982 (1)

Connecticut 0.082 (b) 3 6.49 3.89, 1983 (10)

Delaware 0.107 (b) 3 7.54 3.64, 1985 (2)

District of 0.063 (c) 4 7.85 9.71, 1981 (5)
Columbia

Florida 0.050 3 7.53 5.85, 1984 (4)

Georgia 0.056 3 6.73 5.71, 1981 (6)

Hawaii 0.07l (b) 4 7.13 5.39, 1980 (4)

Idaho 0.036 4 5.67 7.91, 1980 (4)

Illinois 0.037 4 6.15 9.72, 1980 (4)

Indiana 0.075 (b) 4 6.16 9.19, 1980 (4)

Iowa 0.082 (a) 4 4.21 7.18, 1980 (7)

Kansas 0.046 (c) 5 3.62 5.31, 1981 (12)

Kentucky 0.049 4 5.47 9.53, 1980 (4)

Louisiana 0.056 3 6.95 10.97, 1981 (11)

Maine 0.067 3 7.55 4.76, 1984 (3)

Maryland 0.099 (b) 3 6.70 4.36, 1983 (12)

Massachusetts 0.066 (b) 4 7.07 4.01, 1984 (1)

Michigan 0.063 (c) 4 8.37 12.91, 1980 (4)

Minnesota 0.048 4 4.93 6.41, 1981 (10)

Mississippi 0.049 4 7.08 11.27, 1981 (10)

Missouri 0.059 4 5.46 8.43, 1980 (4)

Montana 0.063 (b) 4 5.66 7.63, 1981 (7)

Nebraska 0.059 (c) 4 3.34 4.98, 1981 (9)

Nevada 0.049 (b) 5 6.44 8.24, 1981 (8)

New Hampshire 0.055 4 5.08 2.96, 1980 (11)

New Jersey 0.074 (c) 3 8.17 4.83, 1984 (3)

New Mexico 0.042 3 7.39 8.84, 1982 (2)

New York 0.076 3 8.30 5.57, 1984 (10)

North Carolina 0.047 (c) 5 5.44 7.60, 1980 (5)

North Dakota 0.044 3 4.15 5.48, 1980 (4)

Ohio 0.044 4 6.41 9.90, 1980 (4)

Oklahoma 0.051 4 4.35 7.76, 1982 (6)

Oregon 0.051 4 7.38 9.76, 1980 (4)

Pennsylvania 0.054 (b) 5 7.54 9.72, 1981 (8)

Rhode Island 0.057 3 7.64 4.09, 1984 (4)

South Carolina 0.053 (b) 5 6.21 8.20, 1980 (6)

South Dakota 0.053 (b) 5 3.46 4.99, 1980 (4)

Tennessee 0.049 4 6.28 9.43, 1980 (9)

Texas 0.042 4 5.17 7.61, 1982 (5)

Utah 0.049 4 4.90 6.93, 1980 (4)

Vermont 0.062 3 6.48 4.01, 1984 (4)

Virginia 0.065 3 5.75 4.53, 1986 (4)

Washington 0.044 5 7.75 9.57, 1981 (2)

West Virginia 0.063 (c) 4 6.91 13.53, 1981 (12)

Wisconsin 0.064 (c) 4 5.03 7.98, 1980 (4)

Wyoming 0.040 4 3.53 7.33, 1982 (4)

Panel B: Panel KPSS Test with Multiple Breaks Assuming Cross-Section
Independence

 Tests p Value

Z([^.lambda]) 7.886 (a) 0.000

Panel C: Bootstrap Distribution (%)

 1 2.5 5 10

z([^.lambda]) 1.624 2.104 2.530 3.089

State [[^.T].sub.b,2.sup.i] [[^.T].sub.b,3.sup.i]

Panel A: State-Specific Test

Alabama 6.69, 1984 (10) 4.53, 1989 (2)

Alaska 7.6, 1988 (5) 6.72, 1997 (4)

Arizona 4.77, 1995 (1) 5.22, 2000 (8)

Arkansas 7.13, 1988 (5) 5.14, 1993 (6)

California 6.03, 1985 (11) 8.50, 1991 (1)

Colorado 5.61, 1988 (5) 3.73, 1994 (4)

Connecticut 5.82, 1990 (3) 3.75, 1997 (6)

Delaware 4.90, 1990 (8) 3.66, 1994 (2)

District of Columbia 6.00, 1985 (10) 8.26, 1991 (2)

Florida 7.39, 1990 (9) 4.83, 1995 (1)

Georgia 4.61, 1985 (10) 4.24, 1994 (7)

Hawaii 3.03, 1986 (9) 5.24, 1992 (4)

Idaho 5.82, 1987(11) 5.22, 1994 (2)

Illinois 7.08, 1986 (5) 4.97, 1994 (4)

Indiana 5.41, 1986 (3) 3.61, 1993 (6)

Iowa 4.45, 1987 (3) 3.22, 1994 (2)

Kansas 4.45, 1987 (3) 4.70, 1991 (12)

Kentucky 6.97, 1984 (8) 5.57, 1988 (12)

Louisiana 6.94, 1989 (2) 5.50, 1996 (10)

Maine 6.26, 1990 (6) 4.24, 1996 (4)

Maryland 5.51, 1990 (6) 4.08, 1996 (4)

Massachusetts 7.16, 1990 (4) 3.62, 1994 (12)

Michigan 8.33, 1984 (11) 4.98, 1994 (3)

Minnesota 4.77, 1986 (2) 3.34, 1994 (5)

Mississippi 8.40, 1987(10) 6.62, 1992 (10)

Missouri 6.05, 1984 (8) 4.15, 1994 (4)

Montana 6.25, 1987 (10) 5.44, 1993 (9)

Nebraska 2.78, 1988 (2) 2.62, 1993 (8)

Nevada 5.36, 1985 (12) 6.29, 1991 (2)

New Hampshire 6.31, 1985 (3) 3.16, 1989 (12)

New Jersey 6.92, 1990 (11) 4.81, 1996 (11)

New Mexico 7.04, 1988 (3) 5.61, 1998 (9)

New York 7.10, 1990 (12) 5.52, 1995 (4)

North Carolina 4.33, 1985 (3) 5.18, 1990 (8)

North Dakota 4.42, 1988 (2) 3.22, 1994 (3)

Ohio 6.33, 1986 (3) 4,59, 1994 (7)

Oklahoma 5.99, 1988 (1) 4.08, 1994 (12)

Oregon 6.23, 1986 (6) 5.40, 1994 (3)

Pennsylvania 5.37, 1985 (12) 6.72, 1990 (12)

Rhode Island 7.44, 1990 (4) 4.89, 1995 (6)

South Carolina 4.99, 1984 (10) 6.18, 1990 (11)

South Dakota 4.34, 1986 (6) 3.58, 1990 (10)

Tennessee 5.99, 1985 (1) 5.22, 1959 (5)

Texas 6.96, 1988 (8) 5.32, 1996 (2)

Utah 4.67, 1987 (8) 3.54, 1995 (6)

Vermont 5.19, 1990 (2) 3.47, 1994 (6)

Virginia 5.15, 1990 (9) 3.35, 1996 (10)

Washington 6.01, 1985 (6) 6.71, 1989 (10)

West Virginia 9.97, 1986 (4) 7.19, 1994 (6)

Wisconsin 4.60, 1987 (2) 3.43, 1994 (10)

Wyoming 5.41, 1988 (3) 4.92, 1994 (3)

Panel B: Panel KPSS Test with Multiple Breaks Assuming Cross-Section
Independence

Z([^.lambda])

Panel C: Bootstrap Distribution (%)

 90 95

z([^.lambda]) 8.279 9.326

State [[^.T].sub.b,4.sup.i] [[^.T].sub.b,5.sup.i]

Panel A: State-Specific Test

Alabama 5.36, 1994 (2)

Alaska

Arizona

Arkansas

California 5.96, 1996 (6)

Colorado 5.18, 2000 (8)

Connecticut

Delaware

District of Columbia 6.75, 1998 (10)

Florida

Georgia

Hawaii 3.92, 1999 (1)

Idaho 4.98, 1998 (9)

Illinois 6.10, 2000 (8)

Indiana 4.80, 2000 (8)

Iowa 4.02, 2000 (8)

Kansas 3.82, 1996 (4) 5.04, 2000 (8)

Kentucky 5.11, 1993 (10)

Louisiana

Maine

Maryland

Massachusetts 4.79, 2000 (8)

Michigan 6.22, 2000 (8)

Minnesota 4.42, 2000 (8)

Mississippi 5.91, 1997 (2)

Missouri 5.14, 2000 (8)

Montana 4.49, 2000 (1)

Nebraska 1.60, 2000 (8)

Nevada 4.51, 1995 (6) 5.06, 2000 (8)

New Hampshire 3.94, 1994 (4)

New Jersey

New Mexico

New York

North Carolina 3.66, 1994 (12) 5.92, 2000 (8)

North Dakota

Ohio 5.50, 2000 (8)

Oklahoma 4.58, 2000 (8)

Oregon 7.22, 2000 (8)

Pennsylvania 4.73, 1996 (1) 5.30, 2000 (8)

Rhode Island

South Carolina 4.22, 1995 (3) 6.01, 2000 (8)

South Dakota 3.04, 1996 (4) 3.32, 2000 (8)

Tennessee 4.75, 1993 (11)

Texas 5.90, 2000 (8)

Utah 5.12, 2000 (8)

Vermont

Virginia

Washington 5.00, 1996 (4) 6.66, 2000 (8)

West Virginia 5.68, 1998 (10)

Wisconsin 4.95, 2000 (8)

Wyoming 4.04, 1999 (11)

Panel B: Panel KPSS Test with Multiple Breaks Assuming Cross-Section
Independence

 Tests p Value

Z([^.lambda])

Panel C: Bootstrap Distribution (%)

 97.5 99

Z([^.lambda]) 10.269 11.489

 Finite Sample Critical Values

State 10% 5% 1%

Panel A: State-Specific Test

Alabama 0.067 0.082 0.125

Alaska 0.069 0.081 0.110

Arizona 0.081 0.099 0.141

Arkansas 0.083 0.101 0.144

California 0.056 0.066 0.089

Colorado 0.053 0.061 0.078

Connecticut 0.068 0.079 0.104

Delaware 0.078 0.093 0.128

District of Columbia 0.054 0.063 0.083

Florida 0.076 0.090 0.123

Georgia 0.079 0.095 0.131

Hawaii 0.054 0.062 0.082

Idaho 0.055 0.064 0.084

Illinois 0.056 0.065 0.086

Indiana 0.058 0.069 0.092

Iowa 0.054 0.063 0.081

Kansas 0.043 0.049 0.061

Kentucky 0.070 0.087 0.129

Louisiana 0.068 0.079 0.104

Maine 0.071 0.084 0.112

Maryland 0.070 0.082 0.112

Massachusetts 0.056 0.065 0.088

Michigan 0.060 0.072 0.098

Minnesota 0.056 0.065 0.088

Mississippi 0.054 0.063 0.085

Missouri 0.062 0.075 0.106

Montana 0.052 0.060 0.077

Nebraska 0.053 0.061 0.080

Nevada 0.043 0.048 0.060

New Hampshire 0.066 0.080 0.119

New Jersey 0.069 0.080 0.106

New Mexico 0.074 0.088 0.128

New York 0.076 0.091 0.125

North Carolina 0.042 0.048 0.061

North Dakota 0.078 0.094 0.133

Ohio 0.057 0.066 0.089

Oklahoma 0.053 0.061 0.080

Oregon 0.055 0.065 0.086

Pennsylvania 0.043 0.048 0.060

Rhode Island 0.072 0.086 0.117

South Carolina 0.043 0.049 0.063

South Dakota 0.043 0.049 0.064

Tennessee 0.069 0.085 0.128

Texas 0.055 0.064 0.084

Utah 0.056 0.065 0.084

Vermont 0.078 0.094 0.128

Virginia 0.076 0.091 0.125

Washington 0.044 0.050 0.065

West Virginia 0.056 0.065 0.087

Wisconsin 0.056 0.065 0.085

Wyoming 0.052 0.059 0.075

Panel B: Panel KPSS Test with Multiple Breaks Assuming Cross-Section
Independence

 Tests p Value

Z([^.lambda])

Panel C: Bootstrap Distribution (%)

z([^.lambda])

Notes: Z([lambda]) denotes the panel KPSS test with multiple breaks
developed by CBL. The finite sample critical values for univariate KPSS
tests with multiple breaks are obtained through Monte Carlo simulations
with 20.000 replications. The bootstrap distribution for Z ([lambda])
is obtained through bootstrap methods with 20.000 replications
following Maddala and Wu (1999). The entries in columns (4)-(9) are the
local means in each regime, with the break dates reported in italics.
(a), (b), and (c) imply rejection of the null hypothesis at 1%, 5%, and
10% levels, respectively.


It is noticeable that all U.S. states experience at least three changes in their equilibrium rate of unemployment over the period 1971976-2004. Only for Iowa are we able to reject the null of regime-wise stationarity at the 1% level. In addition, we reject the null at the 5% level for 11 states--Connecticut, Delaware, Hawaii, Indiana, Maryland, Massachusetts, Montana, Nevada, Pennsylvania, South Carolina, and South Dakota--and at the 10% level for another 8 states--District of Columbia, Kansas, Michigan, Nebraska, New Jersey, North Carolina, West Virginia, and Wisconsin. With the panel KPSS test with multiple breaks assuming asymptotic normality and cross-sectional independence--shown in Panel B--we are able to reject the null of joint regime-wise stationarity at the 1% level for the panel as a whole. However, these conclusions are completely overturned when we compare the value of the panel KPSS statistic with the bootstrap critical values--shown in Panel C--which are capable of accommodating general forms of cross-sectional dependence. This backs up the well-known finding that highlights the existence of large-size distortions in panel unit root and stationarity tests that fail to control for cross-sectional dependence. Therefore, there appears to be strong evidence supportive of regime-wise stationarity in the unemployment rate of the states of the United States over the past three decades, which is congruent with the structuralist hypothesis.

Table 4 reports the results from the KPSS tests with multiple breaks for the panel of 14 European Union countries. As with the states of the United States, we find strong evidence of at least three shifts in mean unemployment rates for each of the 14 European Union countries over the period 1970--2004. As shown in Panel A, we are able to reject the null of regime-wise stationarity for Belgium and Ireland at the 1% level; for France at the 5% level; and for Italy, Portugal, and the United Kingdom at the 10% level. With the panel KPSS test with multiple breaks assuming cross-sectional independence, we can reject the null of regime-wise stationarity at the 1% level. This result holds true even when we compare the panel KPSS statistic with the bootstrap critical values. Therefore, given that our test takes stationarity as the null--which will tend to be rejected when there is strong evidence against it--we can conclude for the presence of hysteresis in unemployment for the panel of European Union countries.
TABLE 4
Panel KPSS Stationarity Test with Structural Breaks for Unemployment
Rate in European Union Countries, 1970(1) to 2004(4)

Panel A: Country-Specific Test

Country KPSS Test [m.sub.i] Mean before the Break

Austria 0.076 3 1.27
Belgium 0.074 (a) 5 2.36
Denmark 0.057 4 1.32
Finland 0.085 3 2.22
France 0.090 (b) 4 2.78
Germany 0.057 4 0.77
Ireland 0.084 (a) 4 6.18
Italy 0.063 (c) 4 4.42
Luxembourg 0.069 3 0.04
Netherlands 0.044 4 1.76
Portugal 0.063 (c) 4 2.95
Spain 0.059 4 2.31
Sweden 0.047 4 2.05
United Kingdom 0.058 (c) 4 2.33

Panel B: Panel KPSS Test with Multiple Breaks
Assuming Cross-Section Independence

 Tests p Value

Z([^.lambda]) 5.404 (a) 0.000

Panel C: Bootstrap Distribution (%)

 1 2.5 5

Z([^.lambda]) 0.488 0.757 0.987

Panel A: Country-Specific Test

Country [[^.T].sub.[b,1].sup.i] [[^.T].sub.[b,2].sup.i]

Austria 3.27, 1981 (1) 4.20, 1986 (4)
Belgium 6.98, 1975 (2) 10.25, 1981 (1)
Denmark 4.83, 1975 (1) 6.73, 1980 (2)
Finland 5.06, 1976 (1) 14.45, 1991 (4)
France 5.35, 1975 (1) 9.35, 1981 (1)
Germany 2.33, 1975 (1) 6.24, 1982 (2)
Ireland 8.64, 1975 (1) 14.65, 1982 (4)
Italy 5.96, 1977 (2) 9.26, 1983 (3)
Luxembourg 0.60, 1976 (4) 1.50, 1982 (1)
Netherlands 3.70, 1975 (1) 8.29, 1981 (3)
Portugal 8.00, 1976 (1) 4.92, 1987 (1)
Spain 4.99, 1975 (3) 13.61, 1980 (4)
Sweden 3.01, 1981 (2) 2.01, 1986 (3)
United Kingdom 4.54, 1975 (2) 10.41, 1980 (4)

Panel B: Panel KPSS Test with Multiple Breaks
Assuming Cross-Section Independence

 Tests p Value

Z([^.lambda])

Panel C: Bootstrap Distribution (%)

 10 90

Z([^.lambda]) 1.280 3.783

Panel A: Country-Specific Test

Country [[^.T].sub.[b,3].sup.i] [[^.T].sub.[b,4].sup.i]

Austria 5.37, 1992 (3)
Belgium 7.24, 1988 (1) 9.40, 1993 (2)
Denmark 7.76, 1990 (3) 4.96, 1996 (2)
Finland 9.76, 1997 (2)
France 11.53, 1992 (3) 9.35, 1999(3)
Germany 5.46, 1987 (3) 8.25, 1992 (4)
Ireland 12.45, 1989 (2) 4.82, 1997 (4)
Italy 11.54, 1993 (3) 9.34, 1999 (3)
Luxembourg 3.12, 1993 (3)
Netherlands 6.16, 1988(3) 3.69, 1997 (4)
Portugal 6.66, 1993 (1) 5.02, 1998 (2)
Spain 16.76, 1993(1) 11.13, 1998 (4)
Sweden 7.45, 1992 (1) 4.83, 1998 (3)
United Kingdom 8.20, 1987 (3) 5.58, 1997 (3)

Panel B: Panel KPSS Test with Multiple Breaks
Assuming Cross-Section Independence

 Tests p Value

Z([^.lambda])

Panel C: Bootstrap Distribution (%)

 95 97.5

Z([^.lambda]) 4.234 4.636

Panel A: Country-Specific Test

 Finite Sample Critical Values

Country [[^.T].sub.[b,5].sup.i] 10% 5% 1%

Austria 0.082 0.098 0.137
Belgium 7.34, 1999 (3) 0.043 0.048 0.059
Denmark 0.059 0.069 0.092
Finland 0.090 0.114 0.170
France 0.061 0.073 0.104
Germany 0.063 0.076 0.108
Ireland 0.054 0.062 0.080
Italy 0.057 0.066 0.088
Luxembourg 0.078 0.092 0.125
Netherlands 0.055 0.063 0.082
Portugal 0.061 0.073 0.103
Spain 0.064 0.078 0.112
Sweden 0.061 0.074 0.102
United Kingdom 0.056 0.066 0.088

Panel B: Panel KPSS Test with Multiple Breaks
Assuming Cross-Section Independence

 Tests p Value

Z([^.lambda])

Panel C: Bootstrap Distribution (%)

 99

Z([^.lambda]) 5.132

Note: See Table 3.


Our results for the panel of U.S. states appear in line with those obtained by Song and Wu (1997) who reject the null of joint non-stationarity using the LLC test for a panel of 48 U.S. states with annual data over the period 1962-1993. By contrast, our results for the panel of European Union countries clearly differ from those in Song and Wu (1998) who apply the LLC test to a sample of 11 European Union countries with quarterly data over the period 1960-1992 and strongly reject the null of nonstationarity. However, as stressed by Leon-Ledesma (2002), Song and Wu (1998) findings may be contaminated by large-size distortions characterizing panel unit root tests that fail to properly account for cross-sectional correlation. Our results appear more in line with those in Leon-Ledesma (2002) who, using quarterly data over the period 1985-1999, provides broad support for the natural rate hypothesis for the states of the United States, while hysteresis for a sample of 12 European Union countries for the case with cross-sectionally demeaned data. However, the IPS t-statistic fails to reject the null of nonstationarity for both U.S. states and European Union countries when raw data are used. Since cross-sectional demeaning implies a very restrictive form of cross-correlation, it is not clear ex-ante which of his results is preferable. In addition, Leon-Ledesma (2002) does not control for structural change in unemployment rates, which seems quite plausible if one visually inspects the unemployment rate series for both the U.S. states and the European Union countries. More recently, in line with our results, Romero-Avila and Usabiaga (2007) find evidence supportive of the structuralist paradigm in the monthly unemployment rates of 51 U.S. states over the period 1976-2004 applying the panel Lagrange Multiplier unit root test allowing for up to two breaks and a restrictive form of cross-correlation. Finally, we would like to highlight that the results of the present study should be taken very seriously as we try to properly account for cross-sectional dependence, multiple mean shifts, and the finite sample bias that results from assuming asymptotic normality in relatively short samples. (8)

IV. DISCUSSION OF THE BREAKS

Having established that U.S. state unemployment is best described as regime-wise stationary, while European unemployment is characterized by a unit root, we shift the focus to discuss in detail the structural breaks identified and the patterns followed by the equilibrium rate of unemployment. Following the work by Staiger, Stock, and Watson (1997), we provide an estimate of the equilibrium rate of unemployment associated with each regime, which is obtained as a local mean of the series using Equation (3). Therefore, the equilibrium rate of unemployment for each regime is calculated as [[bar].U.sub.k] = [alpha] + [[gamma].sub.k] for [T.sub.b,k] - l [less than or equal to] t < [T.sub.b,k], k = 1, ..., m.

Beginning with the states of the United States, we find that 17 states exhibit three breaks, 27 states present four breaks, and seven states experience five mean shifts. This provides strong evidence of the need for controlling for structural change due to the occurrence of infrequently permanent shocks to U.S. state unemployment over the past three decades, which led to shifts in the equilibrium rate of unemployment as held by the structuralist paradigm. Furthermore, we can observe some clustering of the timing of breaks and a clear pattern in the fluctuations of the equilibrium rate of unemployment across regimes that roughly coincide with the phases of the business cycles witnessed over the past decades in the United States. (9)

Among the 194 breaks detected, 39 breaks occurred during the period 1980-1982; 56 breaks took place during the period 1983-1989 of which 48 are concentrated in the period 1984-1988; 19 breaks are detected over the period 1990-1991; 54 breaks during the period 1993-1999 of which 45 are concentrated in the period 1993-1996; and, finally, 24 breaks occurred in 2000. From the variations in the equilibrium rate of unemployment for each regime--shown in columns (5)-(9) of Panel A in Table 3--we can infer the sign of the mean shifts. Among the breaks identified during the period 1980-1982, 92% are positive. This coincides with the rise in unemployment rates during the recessionary period that followed the second oil shock of the late-1970s and the Volcker disinflation of 1979, which led to a steep rise in interest rates. As opposed to the first cluster of breaks, the period 1983-1989 shows a prevailing tendency for unemployment rates to fall as indicated by 96% of the breaks being negative. This roughly coincides with the recovery period that set in from around the mid-1980s to the late-1980s. The third cluster of breaks is related to the period 1990-1991 that is characterized by the widespread rise in unemployment rates as confirmed by the fact that 95% of the breaks occurring during this period are positive. This was probably the result of the demand crisis caused by the loss of consumer confidence originating in Iraq's invasion of Kuwait. The U.S. authorities reacted to this situation mainly with markedly expansionary monetary policy, which led to a fall in interest rates. In the period 1993-1999, 96% of the breaks are negative, indicating the prevailing tendency for unemployment rates to fall after the economic downturn of the early 1990s.10 Finally, in the year 2000, unemployment rates rose again as indicated by the fact that 96% of the breaks are positive. The U.S. authorities reacted to the investment crisis with expansionary monetary and fiscal policies. It is also interesting to note that the equilibrium rate of unemployment estimated for the final regime is usually lower than that estimated for the first regime. The only exceptions are Kansas, Nebraska, North Carolina, Oklahoma, Texas, Utah, and Wyoming, but with a difference in equilibrium rates between both regimes always below 1.5 percentage points.

Turning now to the analysis of the breaks for the panel of 14 European Union countries, we can observe a somewhat similar pattern to the U.S. state unemployment rate since European Union unemployment also appears to move countercyclically, thus rising in economic downturns and dropping in upturns. (11) Looking at the number of breaks, 3 countries--Austria, Finland, and Luxembourg--exhibit three breaks, 10 countries--Denmark, France, Germany, Ireland, Italy, Netherlands, Portugal, Spain, Sweden, and the United Kingdom--present four breaks, and Belgium has five breaks. As with the states of the United States, we also detect five clusters of breaks showing similar timing. Among the 54 breaks identified, 12 breaks occurred during each of the periods 1975--1977 and 1980-1983 of which 11 correspond to each of the periods 1975-1976 and 1980-1982; eight breaks took place during the period 1986-1989; 11 occurred between 1990 and 1993 of which 9 are concentrated in the years 1992-1993 and 11 breaks came during 1996-1999. (12)

It is also worth noting that all the breaks detected during the periods 1975-1977 and 1980-1983 are positive, which is explained by the widespread rise in European unemployment during the recessionary periods that set in after the first and second oil shocks of the 1970s. Other factors that may have contributed to the rise in unemployment are the productivity slowdown recorded since the mid-1970s and the tighter monetary and fiscal policies implemented in the first half of the 1980s with the aim of reducing inflation. In addition, it is noticeable that between the mid-1960s and mid-1970s, there was an expansion in the level and range of welfare entitlements such as unemployment insurance benefits, which could have aggravated quitting and shirking. Unlike the clusters related to the mid-1970s and early 1980s, most of the breaks detected during the period 1986-1989 are negative (87.5%), coinciding with the economic upturn that spread across Europe in the second half of the 1980s. The period 1990-1993 is characterized by a sharp increase in unemployment rates in many European Union economies, as confirmed by all the breaks during this period being positive. This may be the result of tighter monetary policies targeting inflation that caused steep rises in real interest rates and tighter fiscal policies aimed at achieving nominal convergence by fulfilling Maastricht criteria. (13) And again, after the deep recession of the early 1990s, European Union economies began recovering and then expanding during the mid-1990s and late-1990s due in part to the decline in world interest rates, thus helping unemployment rates to steadily decrease. This is confirmed by the cluster of breaks related to 1996-1999 that are all negative.

Interestingly, there is some evidence of business cycle asymmetries in regard to the regime shifts in the mean unemployment rate: there are marked increases in unemployment during recessions--mainly those that followed the two oil crises of the 1970s--which sharply contrast with smaller declines during expansions. This is confirmed by the fact that the mean unemployment rate estimated for the final regime is usually much higher than that associated with the first regime before the occurrence of any mean shift, thus providing evidence that the drops in unemployment during cyclical upturns do not fully compensate for the rises in unemployment during cyclical downturns. (14) The only exception is Ireland that witnessed a large decline in unemployment rates over the 1990s, reaching an equilibrium rate of 4.82 in the fourth quarter of 1997, which is below the 6.18 associated with the first regime. This is partly the result of fast economic growth fueled by large inflows of foreign direct investment into sectors related to high technology.

To conclude, the prevailing tendency for unemployment to rise in European Union countries over the past 35 years--in contrast to U.S. states--along with the evidence of a unit root in the series can be seen as evidence that shocks have impacted more severely European Union countries than the U.S. economy. (15) This fact may be attributed to differences in the institutional arrangements governing the functioning of labor markets in the European Union and the United States, which lead to marked differences in the way the economies adjust to macroeconomic shocks. (16)

From an economic policy perspective, our results call for further policy measures in the European Union aimed at improving labor market flexibility conditions, which speed up the adjustment process in response to adverse shocks, thereby preventing upward shifts in unemployment from becoming permanent. For instance, any impediment to the adjustment mechanism in the form of a wage reduction in response to an increase in unemployment should be removed, thus lowering the persistence of unemployment. Some of the main policy measures proposed in the literature to improve flexibility in labor market conditions are the following: (17) (1) the reduction in legislated firing costs, which would lower the degree to which firm's current employment decisions depend on past employment levels, thus becoming more responsive to changes in economic activity; (2) the reduction in the level and duration of unemployment benefits, which would help combat long-term unemployment that causes the loss of skills as well as the fall in the willingness to work; (3) the reduction in the extent to which current wages depend on past wages; (4) the reduction in the legislated minimum wages, which would make it easier for the youngest and least educated workers to find a job; (5) the dismantlement of any regulations that inhibit the creation of new start-ups and/or impose high administrative costs on employers; (6) the increase in the degree of coordination in the wage bargaining process; (7) the introduction of policy measures aimed at increasing the degree of competition in labor, capital, and product markets; and (8) the implementation of measures to enhance geographical and sectoral mobility of the labor force.

V. CONCLUSIONS

Panel unit root and stationarity tests are used in unit root testing in order to increase statistical power. However, as with univariate tests, panel tests need to control for structural breaks so that stationarity with structural change is not misinterpreted as a unit root. In addition, under a panel framework, there exists the limitation that most tests used to analyze the time series properties of unemployment are derived under cross-sectional independence. This latter assumption is unlikely to hold in practice as demonstrated by the Cross-Dependence statistic of Pesaran (2004) that points to the existence of a strong degree of cross-sectional dependence in the error structure of our panels of unemployment rates for U.S. states and European Union countries over the past three decades. To account for these facts, we have employed the panel stationarity test of CBL, which explicitly allows for multiple structural breaks, thereby controlling for cross-sectional dependence and finite sample bias through bootstrap methods.

Our findings, supportive of the existence of regime-wise stationarity in U.S. state unemployment, are in line with the structuralist paradigm, while a unit root in European unemployment is in accordance with the hysteresis paradigm. Interestingly, the timing of the breaks broadly coincides with major macroeconomic shocks mainly associated with the oil crises of the 1970s and the marked changes in interest rates in the 1980s and 1990s. We have also observed that upward shifts in unemployment in response to adverse shocks have tended to become permanent in the European Union as opposed to the U.S. experience where state unemployment appears to revert back to preshock levels. Arguably, labor market institutions may be responsible for such marked differences in unemployment behavior between both economic areas. Finally, an important policy implication arises from our analysis: stabilization policy may affect permanently European unemployment, while only temporarily U.S. state unemployment. This calls for appropriate stabilization policy management and, more importantly, greater labor market flexibility in the European Union economies so as to prevent upward shifts in unemployment from becoming permanent.

APPENDIX

DESCRIPTION OF THE BOOTSTRAP PROCEDURE

The purpose of this appendix is to provide a brief description of the residual-based bootstrap procedure used to generate the empirical distribution of the panel stationarity test with breaks.

Step 1: We run a regression like Equation (3) and then save the resulting residuals ([e.sub.i,t) as well as the corresponding fitted values [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. It is important to calculate the residuals with the null hypothesis of joint stationarity imposed, so that the subsequent bootstrap test is consistent.

Step 2: We generate bootstrap residuals ([e*.sub.i,t]) following the sampling strategy suggested by Maddala and Wu (1999). This consists in permuting the time dimension of the panel in order to create bootstrap residuals, which preserve the cross-correlation structure of the error term. We draw with replacement samples of t + 30 values (and then discard the first 30 values) from the residual matrix obtained in Step 1 by only permuting entire rows, that is, keeping the cross-section index fixed. Therefore, instead of resampling [e.sub.[i,t]] we resample [e.sub.t] = [[e.sub.1,t], [e.sub.2,t], ..., [e.sub.N,t]]' to obtain [e*.sub.t].

Step 3: We calculate the bootstrap sample of observations [U*.sub.i,t] as [U*.sub.i,t] = [^.[U.sub.i,t]] + [e*.sub.i,t] where [[^.U].sub.i,t] is given in Step 1.

Step 4: We construct the Z([^.[lambda]) statistic based on Equation (5).

Step 5: We repeat Steps 2-4 twenty thousand times and the collection of realized Z([^.[lambda]]) statistics provides us with the distribution of Z([^.[lambda]]) under the null hypothesis of regime-wise stationarity.

REFERENCES

Bai, J., and P. Perron. "Estimating and Testing Linear Models with Multiple Structural Changes." Econometrica, 66, 1998, 47-78.

Balakrishnan, R., and C. Michelacci. "Unemployment Dynamics Across Organisation for Economic Co-operation and Development Countries." European Economic Review, 45, 2001, 135-65.

Blanchard, O. J. "European Unemployment: The Evolution of Facts and Ideas." Economic Policy, 21, 2006, 5-59.

Blanchard, O. J., and L. Summers. "Hysteresis and the European Unemployment Problem." NBER Macroeconomics Annual, 1, 1986, 15-78.

--. "Hysteresis in Unemployment." European Economic Review, 31, 1987, 288-95.

Blanchard, O. J., and J. Wolfers. "The Role of Shocks and Institutions in the Rise of European Unemployment: The Aggregate Evidence." Economic Journal, 110, 2000, C1-C33.

Brainard, S. L., and D. M. Cutler. "Sectoral Shifts and Cyclical Unemployment Reconsidered." Quarterly Journal of Economics, 108, 1993, 219-43.

Camarero, M., J. L. Carrion-i-Silvestre, and C. Tamarit. "Testing for Hysteresis in Unemployment in Organisation for Economic Co-operation and Development Countries. New Evidence using Stationarity Panel Tests with Breaks." Oxford Bulletin of Economics and Statistics, 68, 2006, 167-82.

Carrion-i-Silvestre, J. L., T. del Barrio, and E. Lopez-Bazo. "Breaking the Panels: An Application to the GDP Per Capita." Econometrics Journal, 8, 2005, 159-75.

Carruth, A. A., M. A. Hooker, and A. J. Oswald. "Unemployment Equilibria and Input Prices: Theory and Evidence from the United States." Review of Economics and Statistics, 4, 1998, 621-28.

Coakley, J., A. Fuertes, and G. Zoega. "Evaluating the Persistence and Structuralist Theories of Unemployment from a Nonlinear Perspective." Studies in Nonlinear Dynamics and Econometrics, 5, 2001, 179-202.

Friedman, M. "The Role of Monetary Policy." American Economic Review, 58, 1968, 1-17.

Hadri, K. "Testing for Stationarity in Heterogeneous Panel Data." Econometrics Journal, 3, 2000, 148-61.

Hoon, H. T., and E. S. Phelps. "Growth, Wealth and the Natural Rate: Is the European Job Crisis a Growth Crisis?" European Economic Review, 41, 1997, 549-57.

Im, K. S., M. H. Pesaran, and Y. Shin. "Testing for Unit Roots in Heterogeneous Panels." Journal of Econometrics, 115, 2003, 53-74.

Karanassou, M., and D. J. Snower. "How Labour Market Flexibility Affects Unemployment: Long-Term Implications of the Chain Reaction Theory." Economic Journal, 108, 1998, 832-49.

Kurozumi, E. "Testing for Stationarity with a Break." Journal of Econometrics, 108, 2002, 63-99.

Kwiatkowski, D., P. C. B. Phillips, P. Schmidt, and Y. Shin. "Testing the Null Hypothesis of Stationarity Against the Alternative of a Unit Root: How Sure Are We that Economic Time Series have a Unit Root?" Journal of Econometrics, 54, 1992, 159-78.

Layard, R., S. Nickell, and R. Jackman. Unemployment, Macroeconomic Performance and the Labour Market. Oxford: Oxford University Press, 1991.

Leon-Ledesma, M. A. "Unemployment Hysteresis in the US States and the European Union: A Panel Approach." Bulletin of Economic Research, 54, 2002, 95-103.

Levin, A., C. Lin, and C. Chu. "Unit Root Tests in Panel Data: Asymptotic and Finite-Sample Properties." Journal of Econometrics, 108, 2002, 1-24.

Lilien, M. L. "Sectoral Shifts and Cyclical Unemployment." Journal of Political Economy, 90, 1982, 777-93.

Maddala, G. S., and S. Wu. "A Comparative Study of Unit Root Tests with Panel Data and a New Simple Test." Oxford Bulletin of Economics and Statistics, 61, 1999, 631-52.

Nickell, S."Unemployment and Labour Market Rigidities: Europe versus North America." Journal of Economic Perspectives, 11, 1997, 55-74.

--. "Unemployment: Questions and Some Answers." Economic Journal, 108, 1998, 802-16.

Perron, P. "The Great Crash, the Oil Price Shock, and the Unit Root Hypothesis." Econometrica, 57, 1989, 1361-401.

Pesaran, M. H. "General Diagnostic Tests for Cross-Section Dependence in Panels." IZA Discussion Paper Series, DP No 1240, 2004.

Phelps, E. S. "Philips Curves, Expectations of Inflation and Optimal Unemployment." Economica, 34, 1967, 254-81.

--. "Money-Wage Dynamics and Labour-Market Equilibrium." Journal of Political Economy, 76, 1968,678-711.

--. Structural Slumps: The Modern Equilibrium Theory of Unemployment, Interest and Assets. Cambridge, MA: Harvard University Press, 1994.

Phelps, E. S., and G. Zoega. "Natural-Rate Theory and Organisation for Economic Co-operation and Development Unemployment." Economic Journal, 108, 1998, 782-801.

Roed, K. "Hysteresis in Unemployment." Journal of Economic Surveys, 11, 1997, 389-418.

Romero-Avila, D., and C. Usabiaga. "Unit Root Tests, Persistence, and the Unemployment Rate of US States." Southern Economic Journal, 73, 2007, 698-716.

--. "On the Persistence of Spanish Unemployment Rates." Empirical Economics, 35, 2008, 77-99.

Runner, D. "Changes in the State Unemployment Insurance Legislation in 1995." Monthly Labour Review, 1, 1996, 73-8.

Song, F. M., and Y. Wu. "Hysteresis in Unemployment: Evidence from 48 US States." Economic Inquiry, 35, 1997, 235-43.

--. "Hysteresis in Unemployment: Evidence from Organisation for Economic Co-operation and Development Countries." Quarterly Review of Economics and Finance, 38, 1998, 181-92.

Staiger, D., J. H. Stock, and M. W. Watson. "How Precise Are Estimates of the Natural Rate of Unemployment?" in Reducing Inflation: Motivation and Strategy, edited by C. D. Romer and D. H. Romer. Chicago: University of Chicago Press, 1997, 195-246.

Strauss, J., and T. Yigit. "Shortfalls of Panel Unit Root Testing." Economics Letters, 81, 2003, 309-13.

U.S. Department of Labor. Employment and Earnings. Bureau of Labor Statistics Washington, DC: US Government Printing Office, 2005.

DIEGO ROMERO-AVILA and CARLOS USABIAGA*

* The authors thank seminar participants at Centro de Estudios Andaluces and Pablo de Olavide University and participants at the Unit Roots and Cointegration Testing Conference (Faro, 2005), 5th Annual Conference of the European Economics and Finance Society (Crete, 2006), and 63rd International Atlantic Economic Conference (Madrid, 2007) for helpful comments and suggestions. We are also indebted to the referees and the Editor of this Journal for valuable comments and suggestions that led to substantial improvements of the article. We acknowledge financial support from Junta de Andalucia (CICE Excellence Project 01252, PAISEJ-246) and Ministerio de Education y Ciencia (SEJ-2006-04803).

Romero-Avila: Associate Professor, Department of Economics, Pablo de Olavide University, Carreterade Utrera, Km. 1, 41013 Seville, Spain. Phone +34 954348381, Fax +34 954349339, E-mail: dromtor@upo.es

Usabiaga: Professor, Department of Economics, Pablo de Olavide University, Carretera de Utrera, Km. 1, 41013 Seville, Spain. Phone +34 954348553, Fax +34 954349339, E-mail: cusaiba@upo.es

ABBREVIATIONS

ADF: Augmented Dickey-Fuller Type

CD: Cross-Dependence

EU: European Union

OECD: Organisation for Economic Co-operation and Development

OLS: Ordinary Least Squares

LM: Lagrange Multiplier

(1.) The chain reaction theory stresses the importance of labor market flexibility in reducing the degree of unemployment persistence (see Karanassou and Snower, 1998, and references therein).

(2.) Persistence should not be confused with hysteresis since the former implies that (1) a slow adjustment process of mean reversion and (2) temporary shocks to unemployment would have long lasting but not permanent effects. Hysteresis can be modeled as a unit root process and persistence as a near unit root. There are three main approaches explaining the sources of hysteresis: insider-outsider theory, long-term unemployment, and capital scrapping. See Roed (1997) for a survey on the theories of hysteresis.

(3.) Camarero, Carrion-i-Silvestre, and Tamarit (2006) and Romero-Avila and Usabiaga (2008) have applied this methodology to the unemployment rate of the Organisation for Economic Co-operation and Development countries and Spanish regions, respectively. The former study provides evidence of regime-wise stationarity, while the latter supports the hysteresis paradigm.

(4.) We refrained from using monthly data for the European Union countries since they were not available for many countries.

(5.) It is important to note that the Bai and Perron (1998) procedure used to identify the location of breaks is valid only when the series investigated is stationary in variance. In the current scenario, we can safely apply the Bai and Perron (1998) procedure since breaks are only allowed under the null hypothesis of stationarity for all cross-sectional units.

(6.) The value of the Z([^.[lambda]]) statistic must be compared with the critical values from the upper tail of the standard normal distribution.

(7.) See the Appendix for an outline of the bootstrap methodology employed in our analysis.

(8.) Our results differ from those of Camarero, Carrion-i-Silvestre, and Tamarit (2006) who apply the panel KPSS test with multiple breaks for a sample of 19 Organisation for Economic Co-operation and Development countries with annual data over the period 1956-2001. Their failure to reject the null of regime-wise stationarity in unemployment may be the result of mixing countries with different time series properties, with the stationary component dominating the nonstationary one. This is confirmed by our rejection of the null of regime-wise stationarity in favor of hysteresis for the panel of 14 European Union countries for which persistence in unemployment is generally much higher than for those Organisation for Economic Co-operation and Development countries that are not part of the European Union such as Australia. Canada, Japan, and the United States.

(9.) Lilien (1982) finds evidence that part of the shifts in the equilibrium rate of unemployment over the 1970s and 1980s is caused by the slow adjustment of labor to shifts in employment demand between sectors of the U.S economy. He argues that sectoral shifts played a major role in shaping the fluctuations of the U.S. economy rather than the other way around. Brainard and Cutler (1993) also provide evidence that intersectoral reallocation shocks account for a large share of fluctuations in long-duration unemployment in the United States over the postwar period.

(10.) Another factor that may have contributed to the drop in state unemployment rates may be the welfare reform carried out at the state level in the mid-1990s, which led to a reduction in unemployment insurance benefits and other welfare entitlements. This gave the right incentives for former unemployed to actively seek for jobs (Runner, 1996).

(11.) Despite European Union unemployment displaying the cyclical pattern of unemployment--which is more typical of the U.S. experience--each upward shift in unemployment in response to adverse shocks has tended to become permanent in the European Union as opposed to U.S. state unemployment that appears to revert back to preshock levels.

(12.) The elimination of endpoints due to a 15% trimming prevents us from detecting structural breaks associated with some institutional reforms aimed at making European labor markets more flexible which took place after 1999.

(13.) Blanchard and Wolfers (2000) stress the importance of another factor: the occurrence of adverse shifts in labor demand as indicated by the decrease in the labor share since the early 1980s, which could be caused by technological bias away from labor or the decrease in the wage relative to marginal labor productivity. This factor may have contributed to the fact that unemployment levels remained high in the 1980s and 1990s and never went back to the low levels attained before the energy crises of the 1970s.

(14.) Using nonlinear autoregressive threshold models, Coakley, Fuertes, and Zoega (2001) also document some evidence of asymmetric adjustment of unemployment rates for Germany and the United Kingdom.

(15.) Balakrishnan and Michelacci (2001) find strong evidence that labor markets in Europe are much more sluggish in adjusting to macroeconomic shocks than in the United States.

(16.) See Layard, Nickell, and Jackman (1991), Nickell (1997, 1998), Blanchard and Wolfers (2000), and Blanchard (2006) for detailed explanations of how labor market institutions may explain the higher persistence of unemployment in response to adverse shocks in the European Union relative to the United States.

(17.) Nickeil (1998) analyses the impact of labor market institutions on unemployment across Organisation for Economic Co-operation and Development countries over the postwar era. He provides evidence that the level and duration of unemployment benefits, employment protection, union density, and total tax burden are associated with higher long-term unemployment, while coordination between unions and firms and active labor market policies reduce it. Along similar lines, Blanchard and Wolfers (2000) find that the unemployment benefit replacement rate, the duration of unemployment benefits, employment protection, and high taxes on labor income appear to slowdown adjustment, thereby increasing long-term unemployment.
COPYRIGHT 2009 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2009 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Romero-Avila, Diego; Usabiaga, Carlos
Publication:Contemporary Economic Policy
Article Type:Report
Geographic Code:4E
Date:Jul 1, 2009
Words:10251
Previous Article:Benefits and costs of intensive foster care services: the Casey family programs compared to state services.
Next Article:The effects of U.S.-China trade on employment and wages in the U.S.-Mexico border region.
Topics:

Terms of use | Privacy policy | Copyright © 2022 Farlex, Inc. | Feedback | For webmasters |