REVIEW OF ECONOMIC PERSPECTIVES – NÁRODOHOSPODÁŘSKÝ OBZOR,
VOL. 14, ISSUE 4, 2014, pp. 411–426, DOI: 10.1515/revecp-2015-0007
Is the Labour Force Participation Rate NonStationary
in Romania?
Aviral Kumar Tiwari1
, Mihai Mutascu2
Abstract: The purpose of this paper is to test hysteresis of the Romanian labour force
participation rate, by using time series data, with quarterly frequency, covering the period
1999Q1-2013Q4.The main results reveal that the Romanian labour force participation
rate is a nonlinear process and has a partial unit root (i.e. it is stationary in the first
regime and non-stationary in the second one), the main breaking point being registered
around year 2005. In this context, the value of using unemployment rate as an indicator
for capturing joblessness in this country is debatable. Starting from 2005, the participation
rate has not followed long-term changes in unemployment rate, the disturbances
having permanent effects on labour force participation rate.
Key words: Labour, Participation, Hysteresis, Process, Nonlinearity, Policy Implica-
tions
JEL Classification: J01, J21, C12
Introduction
In the past years, many researchers have focused their analyses on labour market characteristics
or functions, especially by using the unemployment rate as the main measure
of labour market state. There are two main hypotheses on employment behaviour.
The first one refers to the natural rate of unemployment and represents the “structuralist”
theory. According to Camarero et al. (2008), this hypothesis characterizes unemployment
dynamics as a mean reverting process, which means that in spite of its cyclical
movements, unemployment rate tends to revert to its equilibrium in the long-run. The
concept was developed by Friedman (1968), Phelps (1968) and Roed (1997). Lee et al.
(2009) show that, in this case, unemployment will fluctuate around a certain natural rate,
so that shocks to the series dissipating over time will only have a temporary effect.
The second hypothesis is the so called “hysteresis” hypothesis that was formulated by
Blanchard and Summers (1986). The authors state that due to labour market rigidities,
1
Faculty of Management, IBS Hyderabad, IFHE University, Room No.: Block C- 204, Dontanpalli
(Village), Shankerpalli Road, Hyderabad, A.P., Pin: 501203, India,
Emails: aviral.eco@gmail.com, aviral.kr.tiwari@gmail.com
2
Corresponding author, West University of Timisoara, Faculty of Economics and Business Administration,
16 J.H. Pestalozzi Street, 300115, Timisoara, Romania, and LEO (Laboratoire d'Economie
d'Orléans) UMR7322, Faculté de Droit d'Economie et de Gestion, University of Orléans,
Rue de Blois - B.P. 6739, 45067, Orléans, France, Email: mihai.mutascu@gmail.com
REVIEW OF ECONOMIC PERSPECTIVES
412
shocks have permanent effects on the level of unemployment. In other words, cyclical
fluctuations have permanent effects on the level of unemployment, due to labour market
restrictions. In this case, the level of unemployment can be characterized as a nonstationary
process, or unit root process.
Unfortunately, the use of unemployment rate in the hysteresis hypothesis generates
several issues (i.e. the individuals frequently move in and out of the workforce) which
are related to the concepts of ‘discouraged worker’ and ‘added worker’ effects (Benati,
2001; Stephens, 2002). The effect of ‘discouraged worker’ shows that the labour force
moves in and out of the labor market with the business cycle, while the ‘added worker’
effect appears as labor supply feed-back of wives to their husbands’ job loss.
In this context, Murphy and Topel (1997) conclude while investigating the trends in
joblessness among American males: “The unemployment rate has become progressively
less informative about the state of the labour market. We find that long-term changes in
labour demands have reduced the returns to work, most notably among the least skilled.
Declining labour market opportunities have led to reduced employment rates among
these men, reflected in both higher unemployment, and withdrawal from the labour
force. Since unemployment data exclude persons who have withdrawn from the labour
force for market driven reasons…these data may miss a key part of the story” (p. 295).
Having the above explanation, some of the current investigations started to explore the
hysteresis in unemploymentby using the labour force participation rate. Madsen et al.
(2008) offers two arguments for this choice: “First, the participation rate is preferred to
the unemployment rate, because unemployment is unlikely to reflect the number of
individuals wanting to work as discussed above. Second, theories of hysteresis are derived
from models of employment and not models of unemployment”. (p. 167).
The relationship between the structuralist and hysteresis hypothesis is grounded in the
property of participation rate series which is related to the “stationarity” process. More
precisely, a series is stationary or reveals a non-unit root process when the shocks do not
have permanent effects. On the contrary, if the shocks are permanent, the series is nonstationary
or represents a unit root process. Two research alternatives outline the analysis
of participation rate property.
In the first case, according to the structuralist theory, the participation rate is stationary,
being mean reverting. As Liu (2011) notes, “Structuralist theories imply that the most
shocks result in temporary fluctuations of employment around the natural rate, but some
shocks cause a structural change in the natural rate itself. …As results, the unemployment
would be stationary around its equilibrium path and is subject to structural
breaks.”(p. 390).
In the second case, when participation rate is non-stationary, the value of using unemployment
rate as an indicator for capturing joblessness is debatable. More precisely,
when the hysteresis in unemployment increases the participation rate, the unemployment
rate rises. Conversely, the unemployment rate remains unaffected, as long as the
unemployment rate equalises the natural rate through the adjustments of wage rate. It
can be concluded thataccording to Ozdemir et al. (2013) “if the labor force participation
rate series is non-stationary, that is the unemployment is characterized by hysteresis,
Volume 14, Issue 4, 2014
413
changes in unemployment rates do not translate into changes in employment rates” (p.
s142).
Thus, the new explorations inliterature have shifted their attention from unemployment
rates to labour force participation rate. It is this researching “platform” that the paper
tests the participation rate stationarity in Romania’s case on, in the presence of nonlinearity
by using the TAR model of Caner and Hansen (2001). The reason for using a nonlinear
threshold model stems from the evidence suggesting that labour responds differently
when employment prospects weaken and when they improve. These asymmetric
responses, in turn, contribute to the non-linear behaviour observed in the labour force
participation rate (and unemployment rate) over time. Several studies found a tendency
for the proportion of labour force participants who drop out of the labour market to
increase quickly during periods of economic weakness and fall slowly during times of
economic recovery. These asymmetric effects differ according to age, education and
gender. Such a model was already used in labour market studies (e.g. Ghosh and Dutt,
2008; Madsen et al., 2008; and Tiwari, 2014). We focus on Romania because this country
registered strong fluctuation in labour force participation rate, and was an emerging
economy during the period of analysis (i.e. January 1994 - December 2013).In the case
of the Romanian labour market, the main results provide a clear evidence of nonlinearity
in the full sample data, and non-stationary of participation rate for major part
of the investigated period, mainly starting from year 2005.
The paper follows a new labour market topic “wave”, which considers the property of
participation rate to be the main research point. The work extends the literature in the
field offering the first results regarding the hysteresis of participation rate in the case of
Romania. We note that the existent literature about this country regarding the hysteresis
as registered on labour market focuses on unemployment rate exclusively. Another
novelty of this investigation is related to the linear and nonlinear tests performed in
order to check the main assumed hypotheses.
The rest of the paper is organized as follows: Section 2 contains the literature review.
Section 3 presents methodology, description of variables and data. Section 4 shows the
estimations and empirical results, while Section 6 concludes.
Literature Review
Over the years, many scientists have focused their analyses on the labour market, especially
testing the hysteresis hypothesis in unemployment. Some of them claim the existence
of the unit root in unemployment rate series, while others state the opposite.
A new tendency arose in the last years and has the hysteresis implications of the participation
rate as the main target. In this context, the literature that explores the participation
rate is not so prolific.
One of the pioneer papers recognizing the importance of participation rate in the study
of labour market was written by Lazear (1987). The author states that, on the one hand,
decisions related to the labour force participation are designed in an irreversible way,
and on the other hand, any adverse shocks generate a decline of temporary participation
into persistent reductions. In the same note, but looking for the evidence of hysteresis in
REVIEW OF ECONOMIC PERSPECTIVES
414
the unemployment rate in the case of U.S., Roberts and Morin (1999) emphasize that
participation rate can be another direction of investigation in the labor market topic.
However, the authors argue that this topic has a secondary importance in discussing
unemployment as labor force participation involves consideration of demographic de-
velopments.
Madsen et al. (2008) offer the first consistent contribution which considers participation
rate in empirical labor explorations. They follow a nonlinear approach and use a unit
root test with threshold in order to analyze the hysteresis of participation rate in the case
of G7 countries, for a period of 130 years. The authors find mix outputs and conclude
that the labor force participation rate is trend reverting for Canada, Italy, Japan and the
U.S., while for Italy and the US, the rate is mean reverting and trend reverting. Gustavsson
and Österholm (2010) show strong evidence against mean reversion in disaggregated
participation rates of subpopulations of the U.S. labour force. The major
implication is that resorting to unemployment rates for subpopulations does not overcome
the informational problems of a non-stationary aggregate participation rate. The
authors suggest that unemployment ratesbe combined with other labour market statistics
before conclusions can be drawn about labour market conditions.
Duval et al. (2010) select 30 industrial countries, over the period of 1960-2008, in order
to investigate the effects of downturns on labour force participation. The main outputs
reclaim a partial rather than full participation hysteresis, as result of cohort effects. In
the same year, Ozdemir et al. (2013) investigate the unit root in the presence of endogenously
determined multiple structural breaks. The main targeted series are total, female
and male labour force participation rates for Australia, Canada and the U.S.A. Using an
extension of Gil-Alana (2008) procedure, they find that under endogenously determined
structural breaks, the total, female and male series are stationary, or mean reverting.
In their recent paper, Queneau and Sen (2013) focus their attention on a group of 12
OECD countries. The authors explore the characteristics of participation rates across
gender, obtaining different results. They reveal that the female labour participation rates
are relatively more persistent in Australia, Canada, Germany, Japan, the Netherlands,
Portugal, and Spain, while in the case of male participation rate, the persistence is registered
only for Finland, Norway, Sweden, and the U.S. Finally, the last contribution
belongs to Liu (2014) who, using the Fourier stationarity test proposed by Enders and
Lee (2012), extends the paper of Gustavsson and Österholm (2012) in the case of the
U.S. The paper concludes by demonstrating the stationary property of labor force rate.
The literature that uses the participation rate in the case of Romania is practically absent;
several papers focus only on unemployment rate hysteresis. The results are contradictory.
Some of them put in evidence the unemployment hysteresis (e.g. Dinu et al., 2011;
Gozgor, 2013; and Furuoka, 2014), while others reveal the mean reverting status of
unemployment rate (e.g. León-Ledesma and McAdam, 2004).
The main literature findings that used the participation rate are heterogeneous, especially
as a result of different econometric tools used to explore the main properties of analyzed
data set. On this framework, following a “battery” of econometric tests, our paper
investigates the hysteresis property of participation rate in Romania, offering the first
such output for this country.
Volume 14, Issue 4, 2014
415
Empirical Methodology: The Caner and Hansen (2001) Threshold Autoregressive
(TAR) Unit Root Test
We already stated that the relationship between the structuralist and hysteresis hypothesis
of participation rate is connected to the concept of “stationarity” process. The structuralist
theory implies the idea of stationary series (i.e. that the shocks do not have permanent
effects), while the hysteresis hypothesis is related to non-stationary series (i.e.
the shocks have permanent effects, the series is a unit root process). The econometrics
field offers many tools to investigate the stationary property of a series. These tools are
often called unit root tests. Thus, if we statistically test the unit root property of participation
rate, we can obtain important information regarding the persistence of shocks.
In the case of the Romanian labour market, three sets of unit root tests are performed in
order to investigate the properties of participation rate. The first one uses traditional
tests, the second one focuses on the presence of a deterministic trend in data coupled
with one or more structural breaks, while the last one tests both non-linearity and unit
root in the data series.
The nonlinear unit root approach includes nonlinear unit root tests which consider in
estimations non-linear functions (i.e.graphs do not have shape of a straight line), such as
polynomial function, exponential function, logarithmic function etc. The classical unit
root tests follow linear approach and use linear functions in estimations (i.e. the graphs
represent straight lines).The participation rate is taken from Eurostat and expresses the
percentage of people of working age (from 15 to 64 years) in the total population who
are actually employed. The sample has a quarterly frequency andcovers the period of
1999Q1-2013Q4. The seasonal component has been removed by using the Tramo/Seats
method. For robustness, we also use the participation level as number of workers expressed
in thousand persons, with monthly frequency, from 1994M1-2010M11. This
series is taken from the Monthly Bulletin of the Romanian National Bank.
In the first step of the analysis we follow the standard unit root tests of Dickey and
Fuller (1979) - ADF, Phillips and Perron (1988) - PP, and Kwiatkowski et al. (1992) KPSS.
One might argue that the lack of support for stationarity of these traditional tests
might be caused by the presence of a deterministic trend in the data coupled with one or
more structural breaks over a long period. Ng and Perron (2001) - NP test is employed
to investigate the trend stationarity of the series. In this case, four main test components
are offered: MZa, MZt, MSB, and MPT. Actually, MZa and MZt represent modified
versions of the Phillips’ (1987) and Phillips-Perron’s (1988) Za and Zt tests, while the
MSB is a test related to Perron and Ng’s (1996) approach. MPT test also is important
because its limiting distribution is the same with that of the feasible point optimal test
performed by Elliott, Rothenberg and Stock (1996).
However, the tests such as NP are found to give misleading results (i.e. biased towards
the non-rejection of null hypothesis when structural breaks are present in the data series
(Perron, 1989). Therefore, we have adopted in the present study two different tests of
unit root to check the stationary property of the data in the presence of structural breaks.
The first test is that by Zivot and Andrews (1992), while the second one by Lee and
Strazicich (2003, 2004). The Zivot and Andrews (1992) approach tests the null of nonstationary
against a single-break stationary. However, the Zivot and Andrews (1992),
REVIEW OF ECONOMIC PERSPECTIVES
416
and Lumsdaine and Papell (1997) ADF-type endogenous break unit root tests both have
the limitation that the critical values are derived, while assuming no break(s) under the
null hypothesis.
Nunes et al. (1997) show that this assumption leads to size distortions in the presence of
a unit root with structural breaks. As a result, when using ADF-type endogenous break
unit root tests, one might conclude that a time series is trend stationary, when in fact it is
non-stationary with break(s), which means that spurious rejections might occur. In contrast
to the ADF-type tests, the LM unit root test is unaffected by breaks under the null
hypothesis (Lee and Strazicich, 2001). Therefore, the Lee and Strazicich (2003, 2004)
test is adopted as the appropriate approach. This test uses the Lagrange Multiplier (LM)
test statistics and incorporates one and two structural breaks in the null and alternative
hypotheses.
None of the aforementioned tests takes the non-linearity implications of series into
account. In this context, we also follow – in our last methodological step –the approach
of Caner and Hansen (2001). They made the first attempt to develop a rigorous asymptotic
theory which would test for non-linearity and unit root in the data series simultaneously,
without assuming stationarity to be a priori. Caner and Hansen (2001) proposed a
TAR model based approach which jointly tests for non-stationarity and non-linearity.
Following Caner and Hansen (2001), a TAR model with a two-regime and an autoregressive
unit root can be described by the following data generating process:
Δ ′ ′ , (1)
where 1, … , , , ′ , Δ , … , Δ ′, is the labour force participation
rate for 1,2, … , , • is the indicator function, is a white noise process,
! " is the threshold variable, # is the delay parameter such that
$ # $ %, is a vector of exogenous variables that includes intercept and liner time
trend, & the threshold value. The & is unknown and & ∈ Λ )& , & * where & and &
are selected according to + $ & 0.10 and + $ & 0.90, based on the
tabulation offered by Andrew (1993). The component of and can be partitioned
as follows:
/
0
1
2
3 , /
0
1
2
3, (2)
where 0 and 0 are scalar terms, , 1 and 1 have the same dimensions, and 2 and
2 are k-vectors. Thus, the slope cofficients on are 0 , 0 , the slope on determinstec
componetns are 1 , 1 , and the slope coefficients on into two regimes are
2 , 2 .
In the equation (1), the null hypothesis of 45: , tests for the presence of the
threshold effect by using the Wald statistic of the following form:
W W 8&9: ;<= ∈>W & ?
@A5
@A &
! 1B, (3)
Volume 14, Issue 4, 2014
417
where @A5 and @A respectively, are the residual variances from least squares estimation
of the liner and TAR model.
Stationarity of the process can be established in two ways. The first case is related to
the evidence when we have a unit root in both regimes. This is called a complete unit
root process. In the first case, thus, the null hypothesis is of the form 45: 0 0 0,
against the unrestricted alternative 0 0	DE	0 0 is tested by using the Wald statistic.
The 0 and 0 that from equation (1) will control the regime dependent unit root
process of the labour force participation rate. The labour force participation rate has a
unit root, if 0 0 0 holds and the test statistics is:
R G , (4)
From the ordinary least squire estimation for 0̂ and 0̂ , and is obtained. However,
Caner and Hansen (2001) proposed the use of one-sided Wald statistic, arguing that the
two-sided Wald statistic may have less power over the one-side version of the test. The
one sided Wald test may be defined as follows:
R G IJ 5 IK 5 , (5)
Caner and Hansen (2001) suggest, using individual t-statistic, that it is and what
distinguish between the stationary (4 ) and the partial unit root case (4 ). The partial
unit root supposes a series for a particular regime (i.e. for a sample of time period over
the entire period under consideration) which may have unit root (based on one-sided
Wald statistic), but not for the whole period under consideration. Of course, this is useful
not only when we find that the unit root hypothesis is rejected (one-sided Wald statistic)
for the whole period under consideration, but also when it is not rejected. In this
case, the whole period under consideration may be influenced by the dominant nature of
the particular regime.
The evidence of the partial unit root, 4 , is consistent, if and onlyone of ! and ! is
statistically significant. This implies that significant evidence of partial unit root in the
labour force participation rate will be found when it is a non-stationary process in one
regime, and a stationary process in the other. We have generated critical values using
bootstrap simulations with 10,000 replications in order to obtain maximum power from
these tests, as suggested by Caner and Hansen (2001).
Even so, their approach has several disadvantages, as Smith (2006) notes: “the test of
threshold effect is only nearly optimal, ... , the procedure is analytically complex and
computationally expensive, while the sequential structure of the approach subjects it is
to a pretesting bias. Finally, the Caner and Hansen’s testing procedure is not easily
generalizable to alternative nonlinear specifications, such as Markov switching processes”
(p. 8).
Empirical Results
In the case of Romania, the evolution of the participation rate for the period of 1999Q1-
2013Q4 is illustrated in Figure 1 in the Appendix. The participation rate reveals two
REVIEW OF ECONOMIC PERSPECTIVES
418
main trend parts: The first part accentuates descending tendency (until 2005), while the
second one shows an ascending segment (starting from 2005).The participation rate has
a sinusoidal dynamics for the entire analysed period.
The empirical section follows, considering the three sets of unit root tests described
above. The results of the first set of unit root tests are presented in Table 1 which includes
the standard ADF, PP and KPSS tests, with the constant included in the model.
Table 1 Traditional ADF, PP and KPSS Unit Root Tests With The Constant Included in the
Model
Test Level 1st
difference
ADF -2.133 -7.847***
PP -2.091 -7.881***
KPSS 0.407** 0.303***
Note: ***, **, and * reflect significance at 1%, 5% and 10 % level of significance, respectively.
The outputs clearly indicate that onthat level the participation rate in Romania has a unit
root for all conventional levels of significance, except for the KPSS test (i.e. the null of
stationarity is rejected only for 5% level of significance).
In the second step of the empirical part we further more analysed unit root property of
the data series with the NP tests of Ng and Perron (2001), with a constant and trend.
This was carried out in order to test whether labour force participation rate is trend stationary.
The results of the NP tests are presented in Table 2.
Table 2 Univariate Unit Root Tests: The Constant and Trend Included in the Model
Romanian labour force
participation rate
Test statistics in level form
MZa MZt MSB MPT
-4.14 -1.301 0.314*** 20.58***
Note:
(1) *, ** and *** denote statistical significance at the 10%, 5% and 1% levels respectively.
(2) Lag length selection is based on Schwarz information criterion (SIC) and it is one here.
These results indicate that the null of unit root is not rejected, the labour force participation
rate revealing a non-stationary process3
for two components of the NP approach. At
the same time, the outputs of the unit root analysis with incorporation of endogenously
determined structural breaks are presented in Table 3.
3
Tiwari (2010, 2012) mentioned that MZa and MZt are the more powerful test statistics of Ng
and Perron (2001) tests.
Volume 14, Issue 4, 2014
419
Table 3 Univariate Unit Root Tests: Constant and Trend Included in the Model with Structural
Breaks
Lee-Strazicich’s LM unit Root Test
TB1 TB2 k
LM Test statis-
tics
Bt1 Bt2 Dt1 Dt2
Results for univariate LM unit root test with one and two structural break in intercept/constant
only
2001:04 0 (-2.79) (-4.28)***
2001:04 2009:02 0 (-3.07) (-4.32)*** (1.17)
Results for univariate LM unit root test with one and two structural break in intercept/constant
and trend both
2002:01 0 (-4.64)** (-0.11) (-2.99)***
2001:02 2002:03 0 (-5.35)* (2.12)** (0.86) (-3.77)*** (2.54)***
Zivot-Andrews unit root test: Allowing for Break in both Intercept and Trend
2002:01 0 -6.94***
Note:
(1) Critical Values for Zivot-Andrews Unit Root Test are -5.57 and -5.08 at 1% and 5% level of
significance respectively.
(2) TB1 and TB2 are the dates of the structural breaks; k is the lag length; Bt1 and Bt2 are the
dummy variables for the structural breaks in the intercept. Dt1 and Dt2 are the dummy variables
for the structural breaks in the slope. Figures in parentheses are t-values. Critical values of LM
test statistics of both test (that is when breaks occur intercept and intercept and trend jointly are
reported in Lee and Strazicich (2003, 2004) two-break and one-break cases respectively. * (**)
*** denote statistical significance at the 10%, 5% and 1% levels respectively.
Table 3 clearly shows that Zivot-Andrews unit root test of one structural break rejects
the null hypothesis of unit root in labour force participation rate at the 1% level of significance.
The LM unit root test of Lee-Strazicich also corroborates with the findings of
Zivot-Andrews unit root test particularly when one or two breaks occur both in constant/intercept
and trend. Hence, this provides evidence to support our argument that
labour force participation rate is mean and trend reverting.
Table 4 Wald Test for Threshold Effect Using Fixed Delay Parameter
m Wald statistics
Bootstrap critical values
Bootstrap p-value
10 % 5% 1%
1 60.33 26.50 32.25 46.34 0.004
2 24.86 26.21 31.61 44.81 0.124
3 16.56 26.32 31.68 44.79 0.339
4 41.99 26.05 31.59 45.19 0.016
Note: “m” denotes optimal delay Parameter. We set a maximum lag of 4 and base all our bootstrap
tests on 10,000 replications.
REVIEW OF ECONOMIC PERSPECTIVES
420
As a final empirical step, Caner and Hansen (2001) proposed a nonlinear TAR unit root
test which is used to jointly check for the presence of nonlinearities and unit root hypothesis
in the Romanian labour force participation rate; the results obtained are presented
in Tables 4–7. We begin with testing for the presence of non-linearity in the
Romanian labour force participation rate through the Wald test. The related results are
provided in the Table 4.
The bootstrap critical values and bootstrap p-values are obtained with 10,000 bootstrap
replications for the labour force participation rate of the form 	 !	 "
for the delay parameters # 1, … ,4. Following Caner and Hansen (2001), the m is
endogenously determined by selecting the least squares estimate of m that minimizes
the residual variance. In other words, such m is selected at which the value of M statistic
is maximized, and in the presented case the M statistic is maximized for labour
force participation rate when # 1. Based on the bootstrapped p-value we find strong
evidence to reject the null hypothesis of linearity at the 1% level of significance, which
implies that simple linear models are inappropriate for testing the unit root property of
labour force participation rate, and thus we got motivation to use the TAR model. Finally,
based on the N G and N G statistics, and for the each delay parameter # 1, … ,4, we
explored the threshold unit root properties of labour force participation rate with emphasising
our preferred model. The results related to the N G and N G statistics, together
with the bootstrap critical values at the 1%, 5% and 10% levels of significance and the
bootstrap p-value, are reported in Table 5.
Table 5 One- and Two-sided Unit Root Tests
m
R1T R2T
Wald
statis-
tics
Bootstrap critical
values
Boot-
strap
p-value
Wald
statis-
tics
Bootstrap critical
values
Boot-
strap
p-value10 % 5% 1% 10 % 5% 1%
1 16.07 12.74 16.42 26.23 0.054 16.07 13.99 17.40 27.56 0.066
2 6.11 13.35 17.25 26.63 0.365 6.11 14.30 18.32 27.52 0.438
3 10.58 13.42 17.38 27.84 0.167 10.58 14.51 18.41 28.51 0.202
4 34.54 13.97 18.03 28.34 0.004 34.54 14.83 18.68 29.28 0.004
Note: “m” denotes optimal delay Parameter. We set a maximum lag of 12 and base all our bootstrap
tests on 10,000 replications.
For # 1, the null hypothesis of unit root for the Romanian labour force participation
rate is rejected at the 10% level of significance in Table 5 which provides an evidence
of stationarity property of labour force participation rate. Given that the N G and N G are
unable to discriminate the complete and partial unit root in labour force participation
rate, we examine the evidence on the unit root hypothesis through the individual tstatistics,
and and report related results in Table 6.
Table 6 shows that, for # 1, the test statistics is higher than the critical value at the
5% level of significance. This leads us to conclude that Romanian participation rate is
characterized by partial unit roots. Specifically, we find that the Romanian participation
rate is stationary in the first regime and non-stationary in the second one. Finally, for
Volume 14, Issue 4, 2014
421
# 1, the least square estimates for the threshold model, as well as the point estimate
of the threshold, &9 , are presented in Table 7.
Table 6 Partial Unit Root Results
m
t1
2
Statistic t2
2
Statistic
t-
Statis-
tic
Bootstrap critical
values
Boot-
strap
p-value
t-
Statis-
tic
Bootstrap critical
values
Boot-
strap
p-value10 % 5% 1% 10 % 5% 1%
1 3.76 2.80 3.35 4.56 0.031 1.38 2.82 3.37 4.41 0.416
2 2.25 2.85 3.37 4.55 0.190 1.01 2.90 3.49 4.55 0.527
3 3.22 2.88 3.51 4.66 0.069 0.40 2.89 3.41 4.53 0.687
4 5.84 2.97 3.62 4.76 0.001 0.61 2.94 3.43 4.50 0.651
Note: “m” denotes optimal delay Parameter. We set a maximum lag of 4 and base all our bootstrap
tests on 10,000 replications.
Table 7 Partial Unit Root Results
Regressors
Estimates, # 1, &9 !0.511139
Test for equality of individual
coefficients
Q &9 R &9 Wald statis-
tics
Bootstrap pvalueEstimates
St Error Estimates St Error
Constant 23.83 6.72 4.22 3.02 7.06 0.36
Y(t-1) -0.38 0.10 -0.06 0.04 8.09 0.22
DY(t-1) -0.43 0.24 -0.05 0.17 1.63 0.58
DY(t-02) 0.24 0.20 -0.17 0.11 3.11 0.45
DY(t-03) -2.01 0.49 0.12 0.09 17.97 0.03
DY(t-04) 0.98 0.30 -0.08 0.10 10.82 0.11
Note: “m” denotes optimal delay Parameter. We set a maximum lag of 4 and base all our bootstrap
tests on 10,000 replications.
For # 1, the value of threshold point estimate 8&9: is !0.511139 and indicates that
the TAR model breaches the regression equation into two regimes contingent upon
whether a quarter change in the Romanian labour force participation rate lies below, or
above !0.511139.4
The first regime transpires for Q !0.511139, which occurs
when the labour force participation rate has fallen over more than 51.1 per cent over a
quarter period. When R !0.511139, the second regime appears which constitutes
all the remaining observations. The first regime shows that Romanian labour force participation
rate has 9 numbers of observations and the second regime, which shows nonstationary
behaviour of the Romanian labour force participation rate, has 46 observa-
tions.
4
The value of &9 is calculated by minimizing @ & . See also Figure 2 in the Appendix for better
understanding. The figure shows the partition of the sample data in both regimes and presents
about which observation point belongs to upper or lower thresholds.
REVIEW OF ECONOMIC PERSPECTIVES
422
As far as Romanian labour market is concerned, there is clear evidence of non-linearity
and non-stationary of the participation rate for the major part of the investigated period,
mainly starting from 2005. This last characteristic is also reinforced by both figures in
Appendix, the main breaking point being registered around year 2005.
Finally, when verifying for robustness, we find that similar results can be replicated by
using the participation level as number of workers, with monthly frequency, for the
period of 1994M1-2010M125
.
Conclusions
The basic objective of the study was to provide empirical validity of hysteresis of labour
force participation rate in Romania. We used quarterly data for the period of 1999Q1-
2013Q4. For the analysis, we adopted three main approaches. The first one follows
traditional unit root tests, the second one focuses on the presence of a deterministic
trend in data, coupled with one or more structural breaks, while the last one investigates
both nonlinearity and unit root in the data series, by using the TAR model of Caner and
Hansen (2001).
All of the traditional tests allow us to consider the participation rate in Romania as a
unit root process. Finally, the non-linear framework approach of Caner and Hansen
(2001) reveals that the Romanian labour force participation rate is a non-linear process
and has a partial unit root (i.e. it is stationary in the first regime, and non-stationary in
the second one).
The value of using unemployment rates as an indicator for capturing the joblessness in
this country is debatable, especially after 2005. Starting from this year, the participation
rate does not follow long-term changes in unemployment rate. In this context, on the
second part of the analysed period, the cyclical shocks have permanent effects on the
labour force participation rate, the Romanian labour market revealing a high level of
rigidity. The levels of unemployment rate are not translated into long-term changes in
employment rates.
Regarding the policy implications, as the participation rate in Romania is a nonlinear
process, it is required for Romanian policy-makers to follow forecasting nonlinear models
in any further estimations of the participation rate. On the other hand, regulations of
Romanian government should be made in order to obtain “laxity” on the labour market.
References
ANDREWS, D. (1993). Tests for Parameter Instability and Structural Change with
Unknown Change Point.Econometrica. 61, pp. 821–856.
BENATI, L. (2001). Some Empirical Evidence on the “Discouraged Worker” Effect.
Economics Letters. 70, pp. 387-395.
5
Access to these results can be requested from the authors.
Volume 14, Issue 4, 2014
423
BLANCHARD, O., SUMMERS, L. (1986). Hysteresis and the European Unemployment
Problem. in S. Fischer ed., NBER Macroeconomics Annual, Cambridge, MA:
MIT Press.
CAMARERO, M., CARRION-I-SILVESTRE, J.L., TAMARIT, C.
(2008).Unemployment Hysteresis in Transition Countries: Evidence using Stationarity
Panel Tests with Breaks.Review of Development Economics.123, pp. 620-635.
CANER, M., HANSEN,B.E. (2001). Threshold Autoregression with a Unit
Root.Econometrica. 69, pp. 1555-1596.
DICKEY, D., FULLER, W.A. (1979). Distribution of the Estimators for Autoregressive
Time Series with a Unit Root.Journal of the American Statistical Association.744, pp.
27-431.
DINU, M., SOCOL, C., MARINS, M-C., SOCOL, A-G. (2011). Testing of the Okun’s
Law in Romania, Economic Computation & Economic Cybernetics Studies & Research.
45, pp. 1-15.
DUVAL,R., ERIS, M., FURCERI, D. (2010).Labour Force Participation Hysteresis in
Industrial Countries: Evidence and Causes, Working Paper, OECD Economics Depart-
ment.
ELLIOT, G., ROTHENBERG, T.J., STOCK, J.H. (1996). Efficient tests for an autoregressiveunit
root. Econometrica. 64, pp. 813–836.
ENDERS, W., LEE, J. (2012). A unit root test using a Fourier series to approximate
smooth breaks. Oxford Bulletin of Economics and Statistics. 74, 574–599.
FRIEDMAN, M. (1968).The Role of Monetary Policy. American Economic Review.58,
pp.1-17.
FURUOKA, F. (2014). Does Hysteresis Exist in Unemployment? New Findings from
Fourteen Regions of the Czech Republic. Czech Journal of Economics and Finance. 64,
pp. 59-78.
GIL-ALANA, L.A. (2008). Fractional Integration and Structural Breaks at Unknown
Periods of Time. Journal of Time Series Analysis. 29, pp. 163-185.
GHOSH, D., DUTT, S. (2008). Nonstationarity and nonlinearity in the us unemployment
rate: a re-examination. Journal of Economic Educators. 8, pp. 43–53.
GOZGOR, G. (2013). Testing Unemployment Persistence in Central and Eastern European
Countries. International Journal of Economics and Financial Issues. 3, pp. 694-700.
GUSTAVSSON, M., ÖSTERHOLM, P.(2010). Labor-Force Participation Rates and the
Informational Value of Unemployment Rates: Evidence from Disaggregated U.S. The
National Institute of Economic Research NIER, Working Paper No. 120.
GUSTAVSSON, M., ÖSTERHOLM, P. (2012). Labor-force participation rates and the
informational value of unemployment rates: Evidence from disaggregated US data.
Economics Letters. 116, 408–410.
REVIEW OF ECONOMIC PERSPECTIVES
424
KWIATKOWSKI, D., PHILLIPS, P.C.B., SCHMIDT, P., YONGCHEOL, S. (1992).
Testing the Null Hypothesis of Stationarity against the Alternative of a Unit
Root.Journal of Econometrics. 54, pp. 159-178.
LAZEAR, E. (1987). Retirement from the Labor Force, in O. Ashenfelter, O. and R.
Layard, (ed.), Handbook of Labor Economics, Vol. 1.
LEE, J.-D., LEE, C.-C., CHANG, C.-P. (2009).Hysteresis In Unemployment Revisited:
Evidence From Panel Lm Unit Root Tests With Heterogeneous Structural
Breaks.Bulletin of Economic Research.614, pp. 325-334.
LEE, J., STRAZICICH, M. (2001). Break Point Estimation and Spurious Rejections
with Endogenous Unit Root Tests.Oxford Bulletin of Economics and Statistics. 635, pp.
535–558.
LEE, J., STRAZICICH, M. (2003). Minimum Lagrange Multiplier Unit Root Test with
Two Structural Breaks.Review of Economics and Statistics. 854, pp. 1082–1089.
LEE, J., STRAZICICH, M. (2004). Minimum LM Unit Root Test with One Structural
Break, Department of Economics, Appalachian State University.
LEÓN-LEDESMA, M., MCADAM, P. (2004). Unemployment, Hysteresis and Transition.
Scottish Journal of Political Economy, 51, pp. 377-401.
LIU, D-H. (2012). Hysteresis Hypothesis in Job Creation and Destruction: Evidence
from the U.S., Annals of Economics and Finance. 12, pp. 447-455.
LIU, D-H. (2014).Labor-Force Participation Rates and the Informational Value of Unemployment
Rates in US: Evidence from Regional Data, Social Indicators Research.
116, pp. 447–455.
LUMSDAINE, R., PAPELL, D. (1997). Multiple Trend Breaks and the Unit-Root Hypothesis.
Review of Economics and Statistics. 792, pp. 212–218.
MADSEN, J.B., MISHRA, V., SMYTH. R. (2008). Are labour force participation rates
nonstationary? Evidence from 130 years for G7 countries.Australian EconomicPapers.47,pp.
166-189.
MURPHY, K., TOPEL, R. (1997).Unemployment and Nonemployment.American Economic
Review.872, pp. 295-300.
NG, S., PERRON, P. (2001). Lag Length Selection and the Construction of Unit Root
Tests with Good Size and Power.Econometrica. 69, pp. 1519-1554.
NUNES, L., NEWBOLD, P., KUAN,C.-M. (1997). Testing for Unit Roots with Breaks:
Evidence on the Great Crash and the Unit Root Hypothesis Reconsidered.Oxford Bulletin
of Economics and Statistics. 594, pp. 435-448.
OZDEMIR, Z.A., BALCILAR, M., TANSEL, A. (2013).International Labour Force
Participation Rates by Gender: Unit Root or Structural Breaks?Bulletin of Economic
Research. 65, Issue Supplement s1, pp. s142–s164.
PERRON, P. (1989). The Great Crash, the Oil Price Shock and the Unit Root Hypothesis.Econometrica.
57, pp.1361–401.
Volume 14, Issue 4, 2014
425
PHELPS, E. (1968). Money Wage Dynamics and Labor Market Equilibrium.Journal of
Political Economy.76, pp.678–711.
PHILLIPS, P.C.B. (1987). Time Series Regression with Unit Roots. Econometrica. 55,
pp. 277-302.
PHILLIPS, P., PERRON, P. (1988). Testing for a Unit Root in Time Series Regression.Biometrika.752,
pp. 335-346.
ROBERTS, J., MORIN, N. (1999). Is Hysteresis Important for U.S. Unemployment?
Finance and Economics Discussion Series, Board of Governors of the Federal Reserve
System, No. 1999-56.
ROED, K. (1997). Hysteresis in unemployment.Journal of Economic Surveys. 11, pp.
38-418.
STEPHENS, M. (2002). Worker Displacement and the Added Worker Effect. Journal of
Labor Economics. 20, pp. 504–537.
TIWARI, A.K. (2010). Globalization and Wage Inequality: A Revisit of Empirical
Evidences with New Approach.Journal of Asian Business Management.21, pp. 173-187.
TIWARI, A.K. (2012). On the dynamics of Energy Consumption, CO2 Emission and
Economic Growth: Evidence from India.Indian economic Review. 47, pp.57-87.
TIWARI, A.K. (2014). Unemployment hysteresis in Australia: evidence using nonlinear
and stationarity tests with breaks, Quality & Quantity: International Journal of Methodology.
48, pp. 681-695.
QUENEAU, H., SEN, A. (2013). On the Persistence of Labor Force Participation Rates
by Gender: Evidence from OECD Countries, preliminary version.
ZIVOT, E., ANDREWS, D.W.K. (1992). Further Evidence on the Great Crash, the OilPrice
Shock, and the Unit-Root Hypothesis.Journal of Business and Economic Statistics.
103, pp. 251–270.
*** (2000-2014). Monthly Bulletin of Romanian National Bank.
*** 2014. Online data-base provided by Eurostat.
REVIEW OF ECONOMIC PERSPECTIVES
426
Appendix
Figure 1 The participation rate in Romania, in the period 1991Q1-2013Q4 (%)
Figure 2 Thresh hold regime plotof participation rate in Romania (%)
60
61
62
63
64
65
66
67
68
69
99 00 01 02 03 04 05 06 07 08 09 10 11 12 13
0 10 20 30 40 50 60
60
61
62
63
64
65
66
67
68
69
Observations
Regime 1
Regime 2
obs=corresponding time period
10=2001Q2
20=2003Q4
30=2006Q2
40=2008Q4
50=2011Q2
60=2013Q4