-
Institute for Empirical Research in Economics University of
Zurich
Working Paper Series
ISSN 1424-0459
Working Paper No. 54 (Revised version)
Economic Growth and Business Cycles:
A Critical Comment on Detrending Time Series
Klaus Reiner Schenk-Hopp
May 2001
-
Economic Growth and Business Cycles:
A Critical Comment on Detrending Time Series
Klaus Reiner SchenkHoppe
Institute for Empirical Research in EconomicsUniversity of
Zurich, Switzerland
[email protected]
May 1, 2001
Abstract
In this paper we pursue an approach based on economic theory
toillustrate possible shortcomings of widely-used detrending
methods. Weanalyze a simple model of economic growth and business
cycles in whichinvestment and technical progress are stochastic.
The Hodrick-Prescottand the Baxter-King lter are shown to detect
spurious business cycleswhich are not related to actual cycles in
the model. Our results castdoubts on the validity of
commonly-accepted stylized business cycle facts.We also discuss the
relation of business-cycle dating based on indicatorsof economic
activity, as e.g. applied by the NBER, and the
detrendingresults.
1 Introduction
Business cycles can be dened as deviations of macroeconomic data
from anunderlying trend which, however, is not observable in
general. Since the decom-position of a time series into a trend and
a remaining cyclic part is in principlearbitrary, any attempt to
identify, or approximate, business cycles has to bebased on
economic theory. This way one can break open the
above-mentionedcircularity of specifying one unobservable variable
with the other.
There is a large strand of literature, in particular due to
research associatedto the real business cycle school (Stadler
[22]), which assumes that the trend is
Institute for Empirical Research in Economics, University of
Zurich, Blumlisalpstrasse10, CH-8006 Zurich, Switzerland. Phone +41
(0)1 6343714, Fax +41 (0)1 6344907, [email protected]
I beneted from comments of the participants of the 8th Annual
Meeting of the Society forNonlinear Dynamics and Econometrics. I am
grateful to Hermann Garbers, Mordecai Kurz,and Rafael Lalive
DEpinay for helpful discussions. The referees comments have helped
todistinctly improve the rst draft of this paper.
1
-
smooth and all uctuations are driven by small transient
productivity shocks.Starting from this assumption, aggregate data
such as output and employmentare detrended using the
Hodrick-Prescott lter and, more recently, the band-pass lter by
Baxter and King [3]. The well-known stylized facts of the
thusspecied business cycles provide a benchmark for any business
cycle model,see e.g. King and Rebelo [14], Danthine and Donaldson
[9], and Stock andWatson [23].
Both the assumptions imposed and the detrending methodology
appliedhave undergone thorough inspections and recently face severe
criticisms on twogrounds.
On the one hand, detrending of time series with the
Hodrick-Prescott lter isshown to produce business cycle dynamics
even if non are present in the original(articial) time series, see
e.g. Cogley and Nason [8], Harvey and Jaeger [12],and Jaeger [13],
who carry out spectral analyses of structural time-series
models.Using actual US macroeconomic data, Canova [6] and Gregory
and Smith [11]nd that most stylized facts are sensitive to the
particular lter applied. Seealso the seminal paper by Nelson and
Kang [16].
On the other hand, recent studies provide evidence that the time
series ofU.S. GDP is not dominated by a smooth trend. Nelson and
Plosser [17] andMurray and Nelson [15] claim that permanent shocks
dominate. Fatas [10]questions the hypothesis of a smooth trend on
the grounds of the empirical factthat long-term growth rates and
persistence of output uctuations have a strongpositive
correlation.
Blanchard and Fischer [4, p. 6] remarked that Macroeconomists
are, andshould be, schizophrenic about the use of time series
methods. Recent discus-sions, as manifested e.g. in Burnsides [5]
comment on Canova [6] and Canovas [7]reply, cast doubts on the
validity of the claim that the economics profession isfully aware
of the potential pitfalls in using detrending methods.
This paper pursues a new direction in the study of the
shortcomings ofwidely-used detrending methods. While in the
above-mentioned literature ei-ther empirical macroeconomic data or
structural time-series models have beenemployed, we propose and
pursue a theory-based approach. The advantage ofour approach is
two-fold. First, the statistical properties of the time series
gener-ated by the underlying model are completely known while there
is disagreementamong econometricians about the actual statistical
properties of macroeconomictime series. Moreover, the trend and the
cyclic component of the time seriesare explicitly given by the
model. Second, the structure of the decompositionof the time series
into trend and cycle is not merely assumed (as in
structuraltime-series models) but based on economic theory.
Moreover, we can comparethe qualitative dynamical behavior of the
cyclic part of original stochastic eco-nomic model with that of the
detrended time series of the model and thus gobeyond a mere study
of the statistical properties.
The model considered here is a neoclassical growth model with
stochastictechnical progress and stochastic investment. It takes
the form of a stochasticdierence equation. The rst process is the
main source of long-run growthwhile the latter is the main source
of the short-run uctuations. Any correla-
2
-
tion of both sources of uctuations can be allowed for. Using
results due toSchenkHoppe and Schmalfuss [20], we can completely
determine the dynami-cal behavior and the statistical properties of
the time series of this model. Inparticular, we can characterize
the dynamics of model, which is stochastic andnon-linear, without
any approximations such as log-linearization. In particular,for
each realization of the exogenous stochastic processes, the sample
paths ofall initial capital intensities are identical in the
long-run. This property ensurese.g. that the numerical simulation
of the long-run dynamics of the model yieldsreliable results.
Assuming that technical progress is driven by a stationary
process of innova-tions, it turns out that the trend of the
stochastic capital intensity is a dierencestationary process. We
study dierent scenarios with respect to the statisticalproperties
of the two sources of randomness, and thus can precisely quantify
theerratic results of both Hodrick-Prescott and Baxter-King lter.
Some qualita-tive properties of the detrending methods are also
analyzed and compared withthe business-cycle-dating methodology
based on indicators of economic activity,as is applied by the
National Bureau of Economic Research.
We nd that even if technical progress is smooth and investment
shocks aresmall and independent of the trend, these two detrending
methods generatespurious business cycles. The higher the
persistence of shocks, due to smallerrates of depreciation, the
more pronounced is the misspecication of the businesscycles.
The remainder of the paper is organized as follows. Section 2
presents thestochastic model of economic growth and business cycles
and provides an anal-ysis of its dynamics; in particular existence
and uniqueness of a globally stablerandom xed point of the capital
intensity is proved. A numerical study of themodel is carried out
in Section 3. There, the actual business cycles of the modeland the
result of detrending methods are compared. Section 4 concludes.
2 A Stochastic Economy
The starting point of our study of detrending methods is a
theoretical eco-nomic model of growth and business cycles with
stochastic technical progressand stochastic uctuations of
investment. The model is strongly inuenced bythe seminal work of
Solow [21] and Swan [24]. The technology is described bya
neoclassical production function, technical progress is
labor-augmenting, andthe investment-consumption decision of
households is not explicitly modelledbut assumed. We enrich this
basic model by allowing for stochastic technicalprogress and a
stochastic saving rate.
The analysis of the model applies random dynamical systems
theory, cf.Arnold [1]. The main result on the long-run dynamics
relies of previous workdue to SchenkHoppe and Schmalfuss [20], and
SchenkHoppe [18]. Under theassumption that the stochastic
uctuations are ergodic and that the productionfunction satises an
Inada-type condition, we can show that all sample paths ofcapital
intensities asymptotically follow the same trajectory. This result
enables
3
-
us to derive the true decomposition of the time series of the
model into a trendand a cyclic part as well as the statistical
properties of the two components.
We consider an economy in which a single homogeneous good is
produced atany period in time. The good can be either consumed or
used as capital input.Two factors, capital and labor, are needed in
the production process, describedby the linear homogeneous
production function
Yt = F (Kt, at Lt)
where Kt 0 is the capital stock at the beginning of period t, at
Lt 0 is theecient labor supply, i.e. at is a measure of technical
progress at time t and Ltis aggregate labor supply. Technical
progress is labor-augmenting. We assumethat (K,L) F (K, at L) is
neoclassical, exhibits constant returns to scale,and satises the
Inada conditions for each possible realization of the
exogenousvariable at, cf. Barro and Sala-i-Martin [2, Sec. 1.2.1].
Households do not havedisutility from work and inelastically supply
their total endowment of labor. Wefurther assume a closed economy,
i.e. the endowment of capital at the beginningof period t+ 1 is
equal to the resources not consumed in the preceding period.Thus
the law of motion of the capital stock is given by,
Kt+1 = F (Kt, at Lt) + (1 t)Kt Ct (1)where Ct denotes aggregate
consumption in period t and t is the rate of depre-ciation.
Analogously to the standard Solow-Swan model we assume that each
house-hold consumes a fraction 1 st of the total output in every
period in time, i.e.Ct = (1 st)F (Kt, at Lt).
We make the following specic assumption on the process of
technical inno-vations.
Assumption 2.1 The evolution of the ecient labor supply, atLt,
is given byat+1Lt+1 = (1 + nt) atLt; and the exogenous variable
(nt, t, st) is an ergodicprocess.
Appropriate assumptions on the range of values for these
processes are im-posed below. For the moment it is sucient to
assume at Lt > 0 for all t.
Dene the capital per ecient unit of labor kt = Kt/(at Lt),
henceforthcalled capital intensity. Under assumption 2.1, (1)
yields the following stochasticlaw for the capital intensity,
kt+1 =Kt+1
at+1 Lt+1=
(1 t)Kt + st F (Kt, at Lt)(1 + nt) at Lt
=(1 t) kt + st f(kt)
1 + nt
where f(k) := F (k, 1) is the intensity form of F (also a
neoclassical productionfunction).
We model the ergodic process (nt, t, st) by an ergodic dynamical
system(,F ,P, ). That is, the probability space (,F ,P) is the
sample path space,and is the shift map. In this notation, the
stochastic law becomes,
kt+1 = h(t, k) :=(1 (t)) kt + s(t) f(kt)
1 + n(t)(2)
4
-
For a given initial state k0 of the capital intensity and any
sample path ofthe exogenous stochastic process, the random dierence
equation (2) denes asample path of the capital intensity. (2) is
henceforth called the stochastic Solowmodel.
Equation (2) generates a random dynamical system on the state
space R+in the following sense. Dene,
(t, , k) =
h(t1) . . . h()k for t 1k for t = 0h(t)1 . . . h(1)1k for t
1
(3)
where h() := h(, ) : R+ R+. (t, , k) is the state of the
stochasticsystem (2) at time t which has been started at k0 = k
under the perturbationdetermined by .
The family of maps (t, , k) is called a random dynamical system.
Thatis, : Z R+ R+, (t, , k) (t, , k) is a measurable mapping
suchthat (0, ) = idR+ and (s+ t, ) = (t,
s) (s, ) for all s, t Z and all . Note that these properties
replace the ow property of a deterministicdynamical system which is
generated by the iteration of a map. Obviously,(t, ) inherits the
regularities (such as continuity or smoothness) of h for t 0and of
h1 for t 0.
We dene the concept of a xed point which is central to our
subsequentanalysis of the model.
Denition 2.1 A random xed point of the random dynamical system
gen-erated by the stochastic Solow model is a random variable k :
R+ suchthat almost surely
k() = (1, , k()) := h(, k()). (4)
We are now in a position to state the main auxiliary result of
this section.A proof can be found in SchenkHoppe and Schmalfuss
[20].
Theorem 2.1 Assume that () [min, max] [0, 1], n() [nmin, nmax] ]
1,[, and s() [smin, 1] ]0, 1]. Assume further that f is
non-negative,increasing, strictly concave, and continuously
dierentiable.
Suppose that
(i) max + nmax > 0;
(ii) 0 limk
f (k) 0, |(t, , k)k(t)| 0 as ta.s.
The result ensures that the long-run behavior of all sample
paths is uniquelydetermined by the random xed point k. For each
initial capital intensity, thesample path asymptotically moves
jointly with t k(t). The dynamics isthus governed by the ergodic
process k(t).
Recall that the evolution of the aggregate capital stock is
described by the(non-stationary) function Kt = atLtkt =
t1u=0(1+n(
u)) a0L0 kt for each ini-tial value of the ecient labor supply
a0L0. The rst part is dierence stationaryafter taking logarithms
because log(at+1Lt+1) = log(1+n(t))+ log(atLt), bydenition, and
n(t) is ergodic and thus stationary.
We therefore obtain the following result.
Corollary 2.1 Fix any initial state of ecient labor supply a0L0
> 0. Thenthe sample path of the capital stock Kt = atLtkt is
governed asymptotically bythe sample path t atLtk(t) for each
initial state K0 > 0 and for almostall .
The asymptotic motion of the logarithm of the capital stock can
therefore bedecomposed into a dierence stationary part and an
ergodic part with expectedvalue zero:
logKt = [log(atLt) + E log k] +[log k(t) E log k] (5)
The two bracketed terms in (5) are referred to as trend and
cycle, respectively.
For each realization of the exogenous process, a variation of
the initial stateof ecient labor supply a0L0 results in a parallel
translation of the samplepath of the capital stock. We therefore
can and do assume a0L0 = 1. Thedecomposition (5) can be written
as
logKt =
[t1u=0
log(1 + n(u)) + E log k]+[log k(t) E log k]
We discuss the properties of the two stochastic processes dened
in thedecomposition (5) trend and cycle in turn. The trend is a
non-stationary yetdierence stationary process. Its systematic
contribution to the growth of thecapital stock is given by E logKt
= tE log(1 + n) + E log k. The uctuationsof the trend are
completely attributable to stochastic variations of the ecientlabor
supply, i.e. to log(atLt).
The cycle is an ergodic process with mean zero. No systematic
tendencyof the growth of logKt is caused by this part of the
decomposition. The uc-tuations of the cycle are stationary and stem
from the variation of the capitalintensity k. The cyclic part
therefore depends on the process describing the
6
-
stochastic investment as well as on the process of innovation.
The latter causesan indirect dependence of the cycle on the state
of technical progress.
Put dierently, the ergodic investment process causes uctuations
only inthe cycle part whereas the ergodic innovation process
triggers uctuations ofboth trend and cycle.
The decomposition of the capital stock process dened in
Corollary 2.1 alsoyields a decomposition of the total output. We
have
log Yt = [log(atLt) + E log f(k)] +[log f(k(t)) E log f(k)]
(6)
It is left to the reader as an easy exercise to derive the
decomposition of theinvestment, the interest and wage rate (which
are both ergodic), and the capitaland labor share (which have
similar decompositions as given above). We willnot need these
processes in the further study.
In summary, we have set up and completely analyzed the model in
thissection. The main result is the decomposition of the motion of
the capital stockinto trend and cycle. Both components are derived
by the principles of economictheory and have clear-cut economic
interpretations. The statistical propertiesof both components have
also been described in detail.
3 Numerical Analysis
In this section we apply the Hodrick-Prescott and Baxter-King
lter to time se-ries generated by our stochastic economic model.
These widely-used detrendingmethods yield a decomposition of the
stochastic aggregates into a trend and acyclic part. We examine the
relation between this decomposition and the trendand the cycle
which have been dened in the previous section using economictheory.
In other words, we study numerically whether the Hodrick-Prescott
orthe Baxter-King lter detects the actual business cycles in the
time series gen-erated by the model. The main emphasis in this
study is on qualitative ratherthan statistical properties of the
cycle.
The software used in the simulations is a collection of MatlabR
scripts.It is available on the web, SchenkHoppe [19]. Using the
software the readerreproduce and check our results as well as
analyze other interesting cases whichlack of space does not permit
to present here.
Before presenting our numerical study, some remarks on the
dierent usageof the terms trend and cycle in econometrics and
economic theory are inorder. In econometrics, these two notions
refer to dierent spectral properties,i.e. they are distinguished
with respect to frequencies. The cycle is associatedto frequencies
between 4 and 32 quarters whereas the trend is related to thelower
frequencies in a time series.
In economic theory the trend refers to that part of a time
series which isrelated to technical progress whereas the cycle
corresponds to the business cycle.The trend is commonly believed to
increase steadily and to show only sluggishvariations over time
whereas the business cycle is associated to economic policiesthat
have a short- or medium-run eect on the economy. This point of view
is
7
-
manifested in the fact that the two components are studied in
dierent elds,growth theory and business cycle theory, respectively.
Only recently strongerties between these two elds started to
develop due to empirical and theoreticalprogress. However, the
above-mentioned distinction, or decomposition, does notrule out any
possible dependencies between both factors and their impact onthe
growth of an economy. Even in our simple model the innovation
processexhibits an eect on the short-run uctuations.
We need to make specic assumptions on the stochastic processes
governingthe evolution of innovation and investment as well as on
the fundamentals.
Assumption 3.1 (i) Labor supply is xed and normalized to one,
i.e. Lt 1. The rate of depreciation is constant.
(ii) The process of innovation n() is dened as follows. n() =
0.0075+(),where is the ergodic process generated by t+1 = At + t
with i.i.d.process t being uniformly distributed on [, ].
(iii) The process of investment s() is dened as follows. s() =
0.25 (1 +z()), where z is the ergodic process generated by zt+1 = B
zt + t withi.i.d. process t being uniformly distributed on [,
].
(iv) The technology is described by the CobbDouglas production
function
f(k) = k 0 < < 1 (7)
We further assume that t and t are independent.
The expected value of the saving rate is 0.25 and expected value
of technicalprogress is about 3% per year, where we interpret each
period in time as onequarter of the year in the simulation.
We study three cases. The parameter settings are chosen as
follows. We xA = 0.95, B = 0.95, = 5 104, and = 5 103 throughout
the analysis.The other parameters are set to,
Case 1: = 0.75, = 0.9.
Case 2: = 0.25, = 0.9.
Case 3: = 0.25, = 0.1.
We rst need to ensure that the random xed point theorem 2.1
applies forthe above parameter-settings.
First note that () [/(1 A),+/(1 A)] and z() [/(1 B),+/(1 B)] for
all . Second, elementary calculations yield the steadystate of the
associated deterministic model,
k(max, nmax, smin) = (smin/(max + nmax))1/(1)
8
-
Using this expression, the contraction condition (iii) of
Theorem 2.1 is satisedif and only if,
E log(1 + s() + nmax
smin
)< E log(1 + n()) (8)
since is deterministic. Validity of (8) in the cases 1-3 can be
checked numeri-cally, see SchenkHoppe [19].
3.1 Case 1
We start with a case in which production is relatively capital
intensive, = 0.75,and the rate of depreciation is high, = 0.9. The
setting of the productionparameter is roughly in line with
empirical studies employing the deterministicSolow model.
The simulation of the stochastic economy described in Section 2
is carriedout as follows. First the initial states of all processes
are set to their expectedvalue. Second the model is iterated 600
periods to ensure that the sample path ofthe capital intensity is
close to the path of the random xed point t k(t).Numerical studies
show that this is indeed the case for the parameter
settingsconsidered here. Third the model is simulated for 200
periods, where a period isunderstood as representing a quarter of a
year. Thus the data generated in thelast 200 periods represent the
time series of output and capital of our model-economy over a
time-horizon of 50 years. Fourth we calculate the actual trendand
cycle for the time series of logarithms of aggregate output and
capital stock,log(Yt) and log(Kt), according to the denition given
in (5) and (6). Finally,we apply the Hodrick-Prescott lter HP(1600)
with parameter w = 1600 andthe band-pass lter BK(6,32,12), K = 12,
introduced by Baxter and King [3],to the time series log(Yt) and
log(Kt).
Figures 1 and 2 depict the results for case 1. We rst note that
the Hodrick-Prescott and the Baxter-King lter show a close
correspondence. This can beobserved in all simulations and is in
agreement with the ndings of Baxterand King [3]. We will therefore
mention only the Hodrick-Prescott lter in thesubsequent
discussions. The BK(6,32,12) lter produces the smoother line dueto
the fact that it is a band-pass lter and also removes the higher
frequencycomponents from a time series which is not true for the
HP(1600) lter.
In gure 1 the actual cycle of the logarithm of the total output
exhibitsvalues of roughly between 4%. It is straightforward to
check that this impliesa maximal deviation of total output Yt from
the true trend of about 9.5%. Oneobserves a clear pattern of
recurrent periods in which the logarithm of the totaloutput is
above resp. below its trend, i.e. the cycle is positive resp.
negative.The length of these periods as well as the magnitude of
the cycle vary over time.However, the longer a period the larger
the deviation from the trend. It is alsonoteworthy that the actual
cycle is quite smooth. In summary, we can clearlydistinguish the
dierent features of the actual business cycle in total output inour
stochastic model.
9
-
0 5 10 15 20 25 30 35 40 45 50
0.04
0.03
0.02
0.01
0
0.01
0.02
0.03
0.04
0.05
Figure 1: [Case 1] Business cycles of GDP (deviation of log(Yt)
from dierenttrends): HP(1600) (magenta), BK(6,32,12) (green),
actual cycle (blue).
0 5 10 15 20 25 30 35 40 45 500.06
0.04
0.02
0
0.02
0.04
0.06
Figure 2: [Case 1] Business cycles of Capital Stock (deviation
of log(Kt) fromdierent trends): HP(1600) (magenta), BK(6,32,12)
(green), actual cycle (blue).
The Hodrick-Prescott lter predicts a business cycle which
exhibits about thesame number of booms and recessions as existent
in the actual cycle. However,the average magnitude of the HP cycle
is smaller than that of the actual cycle.The dating of the cycles
due to the Hodrick-Prescott lter is mainly erratic.Between year 10
and 30 the HP cycle shows a similar behavior as the actualcycle but
predates the booms and recessions by 2 to 3 years.
Figure 2 depicts the results for the logarithm of the capital
stock. The fea-tures of the actual cycle and the HP cycle are very
similar to those discussedabove. This is due to the fact that the
rate of depreciation is high. The varia-tions caused by the
innovation and investment process aect output and capitalalmost in
the same magnitude.
3.2 Case 2
We next consider a case in which production is relatively
intensive in humancapital, = 0.25. We keep the same rate of
depreciation as in case 1. Thus
10
-
only the production parameter is changed compared to the
previous case. Thesimulations have been carried out exactly as
explained above.
Figures 3 and 4 depict the results for case 2.
0 5 10 15 20 25 30 35 40 45 50
0.025
0.02
0.015
0.01
0.005
0
0.005
0.01
0.015
0.02
0.025
Figure 3: [Case 2] Business cycles of GDP (deviation of log(Yt)
from dierenttrends): HP(1600) (magenta), BK(6,32,12) (green),
actual cycle (blue).
0 5 10 15 20 25 30 35 40 45 50
0.025
0.02
0.015
0.01
0.005
0
0.005
0.01
0.015
0.02
0.025
Figure 4: [Case 2] Business cycles of Capital Stock (deviation
of log(Kt) fromdierent trends): HP(1600) (magenta), BK(6,32,12)
(green), actual cycle (blue).
Due to the fact that physical capital is used less intensive in
the productionprocess than human capital, the logarithm of the
total output of the stochasticeconomy exhibits smaller deviations
from the trend than in case 1. Aggregateoutput deviates from the
actual trend by less than 1%. There are no pronouncedbooms and
recessions in the stochastic economy. The Hodrick-Prescott
lter,however, predicts business cycle of comparatively large
magnitude. There is norelation between the HP and the actual cycle.
The spurious business cycles asdated by the HP lter last roughly
about 4 years.
The situation is somewhat dierent in the time series of the
capital stock.There actual and HP cycle are of the same magnitude.
The HP cycle exhibits acorrelation with the actual cycle to a
certain degree. The business-cycle datingis not completely erratic
though the agreement between the HP and the actual
11
-
data is quite bad. Again we observe a predating of the cycle by
1 to 2 years.
3.3 Case 3
We now study our stochastic economy with a comparatively low
rate of depreci-ation, = 0.1. All other parameters are set as in
case 2, in particular productionis relatively intensive in human
capital, = 0.25. Again the simulations followthe same procedure as
discussed above.
Figures 5 and 6 depict the results for case 3.
0 5 10 15 20 25 30 35 40 45 500.025
0.02
0.015
0.01
0.005
0
0.005
0.01
0.015
0.02
Figure 5: [Case 3] Business cycles of GDP (deviation of log(Yt)
from dierenttrends): HP(1600) (magenta), BK(6,32,12) (green),
actual cycle (blue).
0 5 10 15 20 25 30 35 40 45 500.06
0.04
0.02
0
0.02
0.04
0.06
0.08
Figure 6: [Case 3] Business cycles of Capital Stock (deviation
of log(Kt) fromdierent trends): HP(1600) (magenta), BK(6,32,12)
(green), actual cycle (blue).
Due to the assumption that capital depreciates slower than in
the previouscases, the time series of the logarithm of the capital
stock, gure 6, exhibits largedeviations from its trend and is
relatively smooth. The same pattern can beobserved for the actual
cycle of the aggregate output, gure 5. Since productionis
relatively intensive in human capital, the deviations from the
trend are smallerthan for the capital stock.
12
-
In gure 5 the Hodrick-Prescott lter predicts business cycle of
approxi-mately the true magnitude for total output. The dating as
well as the lengthof the HP business cycle is erratic. The spurious
business cycles have a lengthof about 3 years. There is almost no
relation between the HP and the actualcycle.
The result of the Hodrick-Prescott lter is even worse for the
time series ofthe capital stock, see gure 6. The magnitude of the
HP cycle is about eighttimes too small. The dating of the cycle is
also unrelated to the actual behavior.
3.4 Summary of the numerical results
We have analyzed numerically three dierent cases of our articial
economy.In neither case have the Hodrick-Prescott or Baxter-King
lter tracked downthe actual cycle. In fact the approximation of the
true cycle is very poor. Inall cases both lters produced spurious
cycles of an average length of about 3to 5 years. In most cases the
magnitude of the deviation of the time seriesfrom the true trend is
either over- or underestimated. In case 2 (resp. 3) inwhich
production is labor intensive the uctuation of total output (resp.
capitalstock) is predicted to be 5 times larger (resp. smaller)
than it actually is for high(resp. low) depreciation. Any
business-cycle dating based on these lters leadsto incorrect
statements.
Of course it is mandatory to realize the causes of the observed
shortcom-ings when applying these two widely-used lters. The
qualitative results insection 2 ensured that the time series under
study are integrated of order one(i.e. dierence-stationary).
Together with the econometric results on the appli-cation of the
Hodrick-Prescott lter to integrated time series to which we
havealready pointed the reader in the introduction we have a good
understandingof those causes.
3.5 NBER business-cycle dating revisited
In this section we focus on those causes of the
detrending-problems that can bedetected by making use of the
additional knowledge we have about the inno-vation and investment
processes in our model. It will be shown that there is astrong link
between the predictions of the Hodrick-Prescott (and
Baxter-King)lter and the changes in economic activity in our model.
Since the NationalBureau of Economic Research (NBER) business-cycle
dating is based on indi-cators of economic activity, this
observation highlights a possible explanationwhy the HP and NBER
cycles exhibit a very close correspondence in empiricalstudies.
Figure 7 depicts the time series of the innovation process.
Comparison of thisgure with the time series of total output and
capital stock in case 3, gures 5and 6, yields the following
observation. The sample path of either time seriesis above (resp.
below) the actual trend if the sample path of the innovationprocess
is below (resp. above) its expected value. This behavior is due to
thedenition of the trend in our model: the trend is that part of
the time series
13
-
0 5 10 15 20 25 30 35 40 45 50
0.99
0.995
1
1.005
1.01
Figure 7: [Case 3] Time series of technical progress 1 +
n(t).
which is due to technical progress. If technical progress is
slowing down thenthe slope of the trend decreases. If investment
does not fall accordingly, thecapital per ecient unit of labor
increases faster than the trend. Therefore thecycle is positive
during these periods and we thus observe a boom. Clearly
thisdenition of trend and cycle is not related to economic activity
in a strict sense.
Interpreting economic activity in our model as (major) changes
in the invest-ment process we can state the following observation.
The Hodrick-Prescott (andBaxter-King) lter gives a clear-cut
prediction on the changes in investment, cf.gures 5 and 7. The HP
cycle is positive throughout year 5 to 10. In this periodinnovation
is slowing down whereas investment is above its expected value
andexperiences a temporary high in year 8. The actual cycle is
negative. Similarpatterns can be observed also in the period from
year 28 to 35. Both processesare below their expected value
throughout this period. In year 30 investmenthas a local minimum
whereas innovation experiences a local maximum. The HPcycle is
negative but has a local minimum at year 30. The actual cycle
positivethroughout this period. Summarizing we may state that the
HP cycle is closelyrelated to changes in the investment
process.
4 Conclusions
This paper illustrates the dangers of detrending non-stationary
macroeconomictime series by lters. It provides a critical
assessment of a common practice inempirical research, where
econometric methods are employed without a soundtheoretical
foundation. To this end we presented a model of stochastic
economicgrowth in which the actual business cycles are not detected
by the most com-monly applied lters in real business cycle theory.
Both the Hodrick-Prescottand the Baxter-King lter generate spurious
business cycles when applied tothe data generated by the model.
Our reasoning is based on economic theory, dening the trend as
that part ofa time series that is caused by technical progress. Our
criticisms gives further
14
-
0 5 10 15 20 25 30 35 40 45 50
0.2493
0.2494
0.2495
0.2496
0.2497
0.2498
0.2499
0.25
0.2501
0.2502
0.2503
Figure 8: [Case 3] Time series of investment s(t).
support to the purely econometric approaches due to Canova [6,
7], Cogleyand Nason [8], Harvey and Jaeger [12], Jaeger [13], and
Nelson and Kang [16],among others. We suggest that the ndings of
this paper and related work areunderstood as a motivation to strive
for a new methodology (or the resuscitationof classical approaches)
for determining the trend in economic growth paths.
References
[1] L. Arnold. Random Dynamical Systems. Springer-Verlag, New
York, 1998.
[2] R. J. Barro and X. Sala-i-Martin. Economic Growth.
McGraw-Hill, NewYork, 1995.
[3] M. Baxter and R. G. King. Measuring business cycles:
Approximate band-pass lters for economic time series. Review of
Economics and Statistics,81:575593, 1999.
[4] O. J. Blanchard and S. Fischer. Lectures on Macroeconomics.
The MITPress, Cambridge (Mass.), 1989.
[5] C. Burnside. Detrending and business cycle facts: A comment.
Journal ofMonetary Economics, 41:513532, 1998.
[6] F. Canova. Detrending and business cycle facts. Journal of
MonetaryEconomics, 41:475512, 1998.
[7] F. Canova. Detrending and business cycle facts: A users
guide. Journalof Monetary Economics, 41:533540, 1998.
[8] T. Cogley and J. M. Nason. Eects of the Hodrick-Prescott
lter on trendand dierence stationary time series: Implications for
business cycle re-search. Journal of Economic Dynamics and Control,
19:253278, 1995.
15
-
[9] J. P. Danthine and J. B. Donaldson. Methodological and
empirical issuesin real business cycles theory. European Economic
Review, 37:135, 1993.
[10] A. Fatas. Endogenous growth and stochastic trends. Journal
of MonetaryEconomics, 45:107128, 2000.
[11] A. W. Gregory and G. W. Smith. Measuring business cycles
with business-cycle models. Journal of Economic Dynamics and
Control, 20:10071025,1996.
[12] A. C. Harvey and A. Jaeger. Detrending, stylized facts and
the businesscycle. Journal of Applied Econometrics, 8:231247,
1993.
[13] A. Jaeger. Mechanical detrending by Hodrick-Prescott
ltering: A note.Empirical Economics, 19:493500, 1994.
[14] R. G. King and S. T. Rebelo. Resuscitating real business
cycles. In J. B.Taylor and M. Woodford, editors, Handbook of
Macroeconomics Vol. 1B,chapter 14. Elsevier, Amsterdam, 1999.
[15] C. J. Murray and C. R. Nelson. The uncertain trend in U.S.
GDP. Journalof Monetary Economics, 46:7995, 2000.
[16] C. R. Nelson and H. Kang. Spurious periodicity in
inappropriately de-trended time series. Econometrica, 49:741751,
1981.
[17] C. R. Nelson and C. Plosser. Trends and random walks in
macroeconomictime series. Journal of Monetary Economics, 10:139162,
1982.
[18] K. R. Schenk-Hoppe. Random dynamical systems in economics.
Stochasticsand Dynamics, 1:6383, 2001.
[19] K. R. Schenk-Hoppe. Matlab scripts for business cycle
simulationin the stochastic Solow model. Software available at the
web pagehttp://www.iew.unizh.ch/home/klaus/numerics/, Institute for
EmpiricalResearch in Economics, University of Zurich, May 2001.
[20] K. R. Schenk-Hoppe and B. Schmalfuss. Random xed points in
a stochas-tic Solow growth model. Journal of Mathematical
Economics. Forthcoming.
[21] R. Solow. A contribution to the theory of economic growth.
QuarterlyJournal of Economics, 70:6594, 1956.
[22] G. W. Stadler. Real business cycles. Journal of Economics
Literature,XXXII:17501783, 1994.
[23] J. H. Stock and M. W. Watson. Business cycle uctuations in
US macroe-conomic time series. In J. B. Taylor and M. Woodford,
editors, Handbookof Macroeconomics Vol. 1A, chapter 1. Elsevier,
Amsterdam, 1999.
[24] T. W. Swan. Economic growth and capital accumulation.
Economic Record,32:334361, 1956.
16
-
The Working Papers of the Institute for Empirical Research in
Economics can be downloaded in PDF-format
fromhttp://www.unizh.ch/iew/wp/
Institute for Empirical Research in Economics, Blmlisalpstr. 10,
8006 Zurich, Switzerland
Phone: 0041 1 634 37 05 Fax: 0041 1 634 49 07 E-mail:
[email protected]
Working Papers of the Institute for Empirical Research in
Economics
No.1. Rudolf Winter-Ebmer and Josef Zweimller: Firm Size Wage
Differentials in Switzerland: Evidence from Job
Changers, February 19992. Bruno S. Frey and Marcel Kucher:
History as Reflected in Capital Markets: The Case of World War II,
February
19993. Josef Falkinger, Ernst Fehr, Simon Gchter and Rudolf
Winter-Ebmer: A Simple Mechanism for the Efficient
Provision of Public Goods Experimental Evidence, February 19994.
Ernst Fehr and Klaus M. Schmidt: A Theory of Fairness, Competition
and Cooperation, April 19995. Markus Knell: Social Comparisons,
Inequality, and Growth, April 19996. Armin Falk and Urs
Fischbacher: A Theory of Reciprocity, July 20007. Bruno S. Frey and
Lorenz Goette: Does Pay Motivate Volunteers?, May 19998. Rudolf
Winter-Ebmer and Josef Zweimller: Intra-firm Wage Dispersion and
Firm Performance, May 19999. Josef Zweimller: Schumpeterian
Entrepreneurs Meet Engels Law: The Impact of Inequality on
Innovation-
Driven Growth, May 199910. Ernst Fehr and Simon Gchter:
Cooperation and Punishment in Public Goods Experiments, June
199911. Rudolf Winter-Ebmer and Josef Zweimller: Do Immigrants
Displace Young Native Workers: The Austrian
Experience, June 199912. Ernst Fehr and Jean-Robert Tyran: Does
Money Illusion Matter?, June 199913. Stefan Felder and Reto
Schleiniger: Environmental Tax Reform: Efficiency and Political
Feasibility, July 199914. Bruno S. Frey: Art Fakes What Fakes?, An
Economic View, July 199915. Bruno S. Frey and Alois Stutzer:
Happiness, Economy and Institutions, July 199916. Urs Fischbacher,
Simon Gchter and Ernst Fehr: Are People Conditionally Cooperative?
Evidence from a Public
Goods Experiment, July 200017. Armin Falk, Ernst Fehr and Urs
Fischbacher: On the Nature of Fair Behavior, August 199918. Vital
Anderhub, Simon Gchter and Manfred Knigstein: Efficient Contracting
and Fair Play in a Simple
Principal-Agent Experiment, September 200019. Simon Gchter and
Armin Falk: Reputation or Reciprocity?, September 199920. Ernst
Fehr and Klaus M. Schmidt: Fairness, Incentives, and Contractual
Choices, September 199921. Urs Fischbacher: z-Tree - Experimenters
Manual, September 199922. Bruno S. Frey and Alois Stutzer:
Maximising Happiness?, October 199923. Alois Stutzer:
Demokratieindizes fr die Kantone der Schweiz, October 199924. Bruno
S. Frey: Was bewirkt die Volkswirtschaftslehre?, October 199925.
Bruno S. Frey, Marcel Kucher and Alois Stutzer: Outcome, Process
& Power in Direct Democracy, November
199926. Bruno S. Frey and Reto Jegen: Motivation Crowding
Theory: A Survey of Empirical Evidence, November 199927. Margit
Osterloh and Bruno S. Frey: Motivation, Knowledge Transfer, and
Organizational Forms, November 199928. Bruno S. Frey and Marcel
Kucher: Managerial Power and Compensation, December 199929. Reto
Schleiniger: Ecological Tax Reform with Exemptions for the Export
Sector in a two Sector two Factor Model,
December 199930. Jens-Ulrich Peter and Klaus Reiner Schenk-Hopp:
Business Cycle Phenomena in Overlapping Generations
Economies with Stochastic Production, December 199931. Josef
Zweimller: Inequality, Redistribution, and Economic Growth, January
200032. Marc Oliver Bettzge and Thorsten Hens: Financial
Innovation, Communication and the Theory of the Firm,
January 200033. Klaus Reiner Schenk-Hopp: Is there a Golden Rule
for the Stochastic Solow Growth Model? January 200034. Ernst Fehr
and Simon Gchter: Do Incentive Contracts Crowd out Voluntary
Cooperation? February 200035. Marc Oliver Bettzge and Thorsten
Hens: An Evolutionary Approach to Financial Innovation, July
200036. Bruno S. Frey: Does Economics Have an Effect? Towards an
Economics of Economics, February 200037. Josef Zweimller and Rudolf
Winter-Ebmer: Firm-Specific Training: Consequences for
Job-Mobility, March 2000
-
Working Papers of the Institute for Empirical Research in
Economics
No.
The Working Papers of the Institute for Empirical Research in
Economics can be downloaded in PDF-format
fromhttp://www.unizh.ch/iew/wp/
Institute for Empirical Research in Economics, Blmlisalpstr. 10,
8006 Zrich, Switzerland
Phone: 0041 1 634 37 05 Fax: 0041 1 634 49 07 E-mail:
[email protected]
38. Martin Brown, Armin Falk and Ernst Fehr: Contract
Inforcement and the Evolution of Longrun Relations, March2000
39. Thorsten Hens, Jrg Laitenberger and Andreas Lffler: On
Uniqueness of Equilibria in the CAPM, July 200040. Ernst Fehr and
Simon Gchter: Fairness and Retaliation: The Economics of
Reciprocity, March 200041. Rafael Lalive, Jan C. van Ours and Josef
Zweimller: The Impact of Active Labor Market Programs and
Benefit
Entitlement Rules on the Duration of Unemployment, March 200042.
Reto Schleiniger: Consumption Taxes and International
Competitiveness in a Keynesian World, April 200043. Ernst Fehr and
Peter K. Zych: Intertemporal Choice under Habit Formation, May
200044. Ernst Fehr and Lorenz Goette: Robustness and Real
Consequences of Nominal Wage Rigidity, May 200045. Ernst Fehr and
Jean-Robert Tyran: Does Money Illusion Matter? REVISED VERSION, May
200046. Klaus Reiner Schenk-Hopp: Sample-Path Stability of
Non-Stationary Dynamic Economic Systems, Juni 200047. Bruno S.
Frey: A Utopia? Government without Territorial Monopoly, June
200048. Bruno S. Frey: The Rise and Fall of Festivals, June 200049.
Bruno S. Frey and Reto Jegen: Motivation Crowding Theory: A Survey
of Empirical Evidence, REVISED
VERSION, June 200050. Albrecht Ritschl and Ulrich Woitek: Did
Monetary Forces Cause the Great Depression? A Bayesian VAR
Analysis
for the U.S. Economy, July 200051. Alois Stutzer and Rafael
Lalive: The Role of Social Work Norms in Job Searching and
Subjective Well-Being, July
200052. Iris Bohnet, Bruno S. Frey and Steffen Huck: More Order
with Less Law: On Contract Enforcement, Trust, and
Crowding, July 200053. Armin Falk and Markus Knell: Choosing the
Joneses: On the Endogeneity of Reference Groups, July 200054. Klaus
Reiner Schenk-Hopp: Economic Growth and Business Cycles: A Critical
Comment on Detrending Time
Series, May 2001 Revised Version55. Armin Falk, Ernst Fehr and
Urs Fischbacher: Appropriating the Commons A Theoretical
Explanation, September
200056. Bruno S. Frey and Reiner Eichenberger: A Proposal for a
Flexible Europe, August 200057. Reiner Eichenberger and Bruno S.
Frey: Europes Eminent Economists: A Quantitative Analysis,
September 200058. Bruno S. Frey: Why Economists Disregard Economic
Methodology, September 200059. Armin Falk, Ernst Fehr, Urs
Fischbacher: Informal Sanctions, September 200060. Rafael Lalive:
Did we Overestimate the Value of Health?, October 200061. Matthias
Benz, Marcel Kucher and Alois Stutzer: Stock Options: the Managers
Blessing. Institutional Restrictions
and Executive Compensation, October 200062. Simon Gchter and
Armin Falk: Work motivation, institutions, and performance, October
200063. Armin Falk, Ernst Fehr and Urs Fischbacher: Testing
Theories of Fairness Intentions Matter, September 200064. Ernst
Fehr and Klaus Schmidt: Endogenous Incomplete Contracts, November
200065. Klaus Reiner Schenk-Hopp and Bjrn Schmalfuss: Random fixed
points in a stochastic Solow growth model,
November 200066. Leonard J. Mirman and Klaus Reiner Schenk-Hopp:
Financial Markets and Stochastic Growth, November 200067. Klaus
Reiner Schenk-Hopp: Random Dynamical Systems in Economics, December
200068. Albrecht Ritschl: Deficit Spending in the Nazi Recovery,
1933-1938: A Critical Reassessment, December 200069. Bruno S. Frey
and Stephan Meier: Political Economists are Neither Selfish nor
Indoctrinated, December 200070. Thorsten Hens and Beat Pilgrim: The
Transfer Paradox and Sunspot Equilibria, January 200171. Thorsten
Hens: An Extension of Mantel (1976) to Incomplete Markets, January
200172. Ernst Fehr, Alexander Klein and Klaus M. Schmidt: Fairness,
Incentives and Contractual Incompleteness,
February 200173. Reto Schleiniger: Energy Tax Reform with
Excemptions for the Energy-Intensive Export Sector, February
2001
-
Working Papers of the Institute for Empirical Research in
Economics
No.
The Working Papers of the Institute for Empirical Research in
Economics can be downloaded in PDF-format
fromhttp://www.unizh.ch/iew/wp/
Institute for Empirical Research in Economics, Blmlisalpstr. 10,
8006 Zrich, Switzerland
Phone: 0041 1 634 37 05 Fax: 0041 1 634 49 07 E-mail:
[email protected]
74. Sandra Gth, Thorsten Hens and Klaus Schenk-Hopp: Evolution
of Trading Strategies in Incomplete Markets.February 2001
75. Ernst Fehr and Klaus Schmidt: Theories of Fairness and
Reciprocity Evidence and Economic Applications,February 2001
76. Bruno S. Frey and Alois Stutzer: Beyond Bentham Measuring
Procedural Utility, April 200177. Reto Schleiniger: Global
CO2-Trade and Local Externalities, April 2001