JARI KUULUVAINEN
ESTIMATING SUPPLY AND
DEMAND FOR ROUNDWOOD:
How to incorporate the data
and theory?
METSÄNTUTKIMUSLAITOKSEN
TIEDONANTOJA 397
Metsäekonomian tutkimusosasto
Finnish Forest Research Institute
Department of Forest Economics
Osoite: Unioninkatu 40 A 00170 Helsinki
Puhelin:
(90)
857051 Telefax:(90)
625 308 Telex: 121286 metla sfYlijohtaja: Eljas
Pohtila Director:THE FINNISH FOREST RESEARCH INSTITUTE
Address: Unioninkatu 40 A SF-00170 Helsinki Finland
Phone: +358 0 857 051 Telefax: +358 0 625308 Telex: 121286 metla sf
Tutkimusjohtajat:
EeroPaavilainen,
Jari Parviainen, RistoSeppälä
Research directors:Metsäntutkimuslaitoson maa-
ja
metsätalousministeriönalainenvuonna 1917perustettu
valtiontutkimuslaitos. Senpäätehtävänä
on Suomen metsätaloutta sekämetsävarojen ja
metsientarkoituksenmukaistakäyttöä
edistävätutkimus.Metsäntutkimustyötä
tehdään lähes 800hengen
voimin kolmellatutkimus osastollaja kymmenellä
tutkimus-ja
koeasemalla. Tutkimus-ja
koetoimintaa varten laitoksella on hallinnassaan valtionmetsiäyhteensä
noin 150 000hehtaaria, jotka
onjaettu
17 tutkimusalueeseenja joihin sisältyy
kaksi kansallis-ja
viisiluonnonpuistoa.
Kenttäkokeita onkäynnissä
maankaikissa osissa.TheFinnishForest Research
Institute,
establishedin1917,
isa stateresearch institution subordinatedto theMinistry
ofAgriculture
andForestry.
Its main taskis to carry outresearch work tosupport
thedevelopment
offorestry
and theexpedient
use offorestresources andforests. The workis carriedoutby
meansof800persons in threeresearch
departments
andtenresearch and field stations. The institute administers state-owned forests of over 150 000 hectares for researchpurposes,including
two nationalparks
andfive strict naturereserves. Fieldexperiments
are inprogress inallparts
ofthecountry.
FinnishForest ResearchInstitute,
Department
ofForestEconomics, Working Paper
397ESTIMATING SUPPLY AND DEMAND FOR ROUNDWOOD: HOW TO
INCORPORATE THE DATA AND THEORY ?
Lauri Hetemäki and Jari Kuuluvainen
Helsinki 1991
ABSTRACT
This paper examines the aggregate
pulpwood
market in Finlandusing
the econometricapproach
advocatedby Hendry
et al.(1988)
andSpanos (1990)
who propose a
statistically
consistent way to estimatesimultaneous-equations
models. The effects ofcapital
marketimperfections
onprivate
nonindustrial timbersupply
are allowedand athree-input
demandfunction is used(capital,
labourandwood).By comparing
theresults with anearliermodelspecification
of theFinnishpulpwood market,
itis concludedthatthenewapproach provides statistically
more robust and informativeresults than the earlierspecification.
The results indicate that short-term
supply
reactspositively
to an increase in stumpageprice,
whilethelong-run (total) elasticity
isnegative
andrathersmall in absolute terms. The short-termelasticity
of demand with respect to the stumpageprice
ispositive
and thelong-run elasticity
isnegative. Capital
is acomplement
while labor is a substitute for roundwoodinput.
Thedynamic adjustment
process, substitutionandcross-price
effectsand thecapital
marketimperfections implied by
thepresentstudy
differfrom the results obtainedinprevious
studies.Key
words: roundwood markets, simultaneousequations model,
statisticalvalidity
Authors' address: The Finnish Forest Research
Institute,
Box37,
SF-00381 Helsinki, FinlandISBN951 - 40- 1187-2 ISSN 0358 - 4283
HAKAPAINO OY, HELSINKI 1992
CONTENTS
Page
PREFACE 4
1. INTRODUCTION 5
2. THEORETICAL MODEL 6
2.1
Supply
ofPulpwood
62.2Demandfor
Pulpwood
73. RESEARCH RESULTS 8
3.1
Properties
oftheTimeSeries 83.2
Specification
and EstimationoftheStatistical and Econometric Model 11
3.3 Statistical Model 12
3.4 Econometric Model 13
3.5Results forthe
Supply Equation
163.6Results fortheDemand
Equation
223.7
Encompassing
274. CONCLUSIONS 28
FOOTNOTES 31
REFERENCES 32
APPENDICES 35
PREFACE
The present
study
was conducted at theDepartment
of Forest Economicsof theFinnishForest Research Institute. The
study originates
from themeetings
of an informal"econometric
study group"
attheDepartment
ofForestEconomics.The workofthis groupled to the construction and estimation of an econometric model for the Finnish roundwood
market
(Kuuluvainen
et ai.1988).
In the presentstudy,
thepulpwood
marketpart of themodel
reported
inKuuluvainenetai. (1988)isfurtherdeveloped using
therecentadvancesin timeseries econometrics. An
abridged
versionof this reportispresented
in HetemäkiandKuuluvainen
(forthcoming).
The authors
gratefully acknowledge helpful
comments from DariusAdams,
DavidNewman,and Ville Ovaskainen. Wealso want to express our
gratitude
to the late JormaSalo for the valuable contributions to Finnish roundwood market studies.
Finally,
wewouldliketo thankall theresearchers whotookpart in the workoftheeconometric
study
group for the enthusiastic and
inspiring, although occasionally long
andexhaustive,
discussions on the
subject.
The first authoracknowledges
financial support from theAcademy
ofFinland.Lauri Hetemäki and Jari Kuuluvainen
1. INTRODUCTION
Econometric
analyses
ofroundwoodand forestproduct
marketsbased on timeseries datahave a
relatively long
tradition.Sincethepioneering study by McKillop (1969),
a numberof studies have
appeared
intheUSA(e.g.,
Robinson 1974, Adams &Haynes
1980andNewman
1987)
and in Scandinavia(e.g.,
Brännlund, Johansson andLöfgren 1985,
Kuuluvainen 1986, Kuuluvainen et ai. 1988 and Hultkranz and Aronsson
1989).
Thesestudies have
greatly
increased ourunderstanding
of the basicrelationships affecting
roundwood markets. However, recent
developments
in time series econometrics andsimultaneous
equations
estimation(Engle
andGranger 1987, Hendry
etal.1988, Phillips
and Durlauf
1986, Spanos 1990)
canprovide
newinsights
into these markets. Forexample, assumptions
made inearlierstudiesconcerning
short-termdynamics,
substitutionand
cross-price
effects, and the effects ofcapital
marketimperfections
on nonindustrialprivate
woodsupply
can now beexaminedina moresystematic
way.The purposeofthis
paper is to
provide
newevidence on thefunctioning
ofthepulpwood
marketinFinland.
However,
thestudy
shouldalso beofinterestforempirical applications
of small simultaneous
equations
systems ingeneral.
Weapply
recent advances in timeseries econometricsandsimultaneous
equations
systemestimation,
asproposed by Hendry
et al.
(1988)
andSpanos (1990) (hereafter
theHSapproach),
to estimatethe demandforand
supply
ofpulpwood using
annualdata from 1960to 1988.The results arecompared
with an earlier
specification
ofthepulpwood
market(Kuuluvainen
etal.1988)
which were-estimate and examine
using
a numberof tests and evaluation criteria notpreviously
used in roundwood market studies.
The results reveal
misspecification problems
in Kuuluvainen et al., while the newspecification
is found to bestatistically
consistent and more informative. As aresult,
therobustness ofshort-termforecasts forthe
pulpwood
marketareimproved.
Further, thenew model
implies
revisions of some of thehypotheses concerning
the behavioralfunctioning
ofthepulpwood
market.The remainder of the paper
provides
adescription
of the theoreticalframework,
theempirical
results and discussionoftheirimplications.
Thevariablesanddata are describedinthe
Appendix.
2. THEORETICAL MODEL
2.1
Supply of Pulpwood
Nonindustrial
private
forest(NIPF)
owners are themainsuppliers
ofraw materialto theforest-based industriesin Finland. The institutional framework of the Finnish
pulpwood
markethasbeen describedin Kuuluvainenetal. andwe donot repeatthediscussion here.
The
empirical
evidence from the Finnish dataindicates that marketimperfections
affectthe timber
supply
behavior of NIPF owners(Kuuluvainen
&Salo).
Ifuncertainty (cf.
Koskela,
1989a,b), imperfect capital
markets(cf.
Kuuluvainen,1990),
and/or nontimbervalues
(cf. Binkley)
are assumed,optimal
timbersupply
andconsumption
mustbe decidedsimultaneously.
Therefore, the variableswhich affect theoptimal consumption
decisionsalso affect the
optimal harvesting
decisions.Using
a twoperiod
Fisherianconsumption-savings
model fordetermining
theoptimal harvesting
decisions underselective credit
rationing,
thebehavioralsupply equation
becomes(Kuuluvainen, 1990),
where
P,
is thepulpwood price,
net ofharvesting
costs(i
=t,t+l),
Rt+l is the marketinterest rate, V
t
is the stock of
growing
merchantabletimber, I,
is the exogenous(1) q[
=Q
S
(P
t,P t+l,R
l+l, V
t)
I„
Il+l,B
b
5),
+/--/+- + - + - +
nonforestry
income(i=t,t+l),
B is theexogenouscredit limitfacedby
an individualforestowner, <5is the
subjective
rate oftimepreference.
Under selective credit
rationing,
the effects ofprices
and the interest rate cannot bedetermineda
priori.
Theprice
effectofthecredit-rationedforestownercontains boththepositive
substitutioneffect andthenegative
incomeeffect. Thesign
of theinterestrate isundeterminedas some forestowners face restrictions in the
capital
marketandothers donot. Forthecredit-rationed
suppliers
the interestrate hasonly
anegative (income)
effect,while for the nonrationed ones the effect is
unambiguously positive.
Note that theimplications
ofothertypesofmarketimperfections
fortheempirical analysis
ofpulpwood
supply
arefairly
similarto those ofselectivecreditrationing.
Theessentialpoint
isthat,
inthe presence ofmarket
imperfections,
the income level of forestowners also affects thesupply.
Whenaggregated
data are used it is reasonable to assume that creditrationing
faced
by
individualforestowners as wellas thesubjective
rate of timepreference
donothavea
systematic
patternover time.Furthermore, it is notgenerally possible
to measurethese variablesin
aggregated
data.Consequently, they
are not included in theempirical
analysis. Finally,
dueto dataproblems,
allowabledrain series is usedtoapproximate
thestockof
growing
timber(seeAppendix).
2.2Demand
for Pulpwood
Thepaper
industry's
outputcan bedescribedusing
theproduction
functionwhere Kis
capital input,
L is labourinput
andQ
is thewoodraw-materialinput
neededtoproduce
an amount Ytofthefinal
product.
Ineq.(2)
it isimplicitly
assumedthattheK,Land
Q inputs
areweakly separable
as a group from the residualinputs (materials
and(2)
Yt
=
F(K,L,Q)
energy).
Firms in thepulp
and paperindustry
are assumed to sell their finalproducts
oncompetitive
export markets atgiven prices,
PXt.Ignoring
decisions on theholding
ofraw-material inventoriesand the
uncertainty
containedinshort-termproduction
decisions,the
profit-maximizing problem
of therepresentative
firm can be used to derive theshort-termdemandfunction for
pulpwood (cf.
Brännlundetai.1985):
where
Q
td is thedemandfor
pulpwood,
PXtis the
export
price
offinalproducts,
Ptis the
stumpage
price,
Wtis theunit labourcost, and C
t
is the
price
ofcapital.
The effects ofwages and
capital
are uncertainbecause itcannot be deducedapriori
whetherroundwoodis a technical
complement
ora substitutefor theseinputs.
3. RESEARCH RESULTS
3.1
Properties of
theTimeSeriesDepending
on theparticular properties
of the time series used forestimation,
differentmodelling strategies
must bechosen.Forexample,
the "errorcorrection"or"cointegration"
modelshave become
popular
foranalyzing nonstationary
timeseries data(see,e.g.,Engle
and
Granger 1987, Hendry
and Ericsson1991).
Further, for thevalidity
of statisticalinference it is
important
toknow whethertheseries arestationary
ornot (see,e.g.,Phillips
1986and
Phillips
andDurlauf1986).
Inordertoexaminetheproperties
ofthedataseries,theautocorrelationfunctionsand
autoregressive
processes oftheseries werecomputed
andnormality
andstationarity
testsperformed.
Alltheseries are
logarithmic
transformationsoftheoriginal
series andthemonetaryseriesare deflated
(see
theAppendix
fordatadescription).
Theresults in Table 1 show thatthe(3) Q
td=
(3
d(PX„
Pt,W
t
,
Ct),
+- ? ?
firstautocorrelation coefficientis rather
high
anddecreasesslowly
with increases inlags
for theusercost
(CO,
wages(W
t),
exogenousincome(I
t)
and allowabledrain(V
t) series,
indicating
thatthelevelsoftheseseries are notstationary.
However,their first differencesappear to be
stationary,
andconsequently
these series can beregarded
as1(1)
seriesaccording
to theautocorrelation functions.All theother series used in themodelsappearto be
1(0)
series. The results(Table 1)
oftheregression
ofeach ofthe levels series ontheir
respective
fivelags
indicatedthatall the series follow a first-orderautoregressive
process
(i.e. AR(1) process).
Table1. TimeSeries
Properties
oftheData.Notes:InTable 1,
*
indicatessignificantt-valuesat the5 % significancelevel fortheautoregressive coefficients and Adenotes thedifferenceoperator.All monetaryunitsand the interestrate aredeflated.
Series Autocorrelation function Autoregressive process
lag 1 2 3 4 5 1 2 3 4 5
Pulpwood quantity Q
t .48 .27 .03 -.06 -.07 .54* .13 -.24 .06 .00
aq,
-.30 -.03 -.21 -.01 -.09 Stumpage pricePt .64 .35 .17 .15 .18 .82* -.06 -.35 .31 -.01
AP,
-.05 -.16 -.33 .01 .17 Export pricePXt .76 .53 .45 .44 .33 .08* -.27 .15 .16 -.11
APX
t
-.01 -.34 -.12 .17 .16 User cost
c
t .90 .81 .82. .80 .72 .74* -.43 .46 .01 .02
AC
t -.05 -.47 .08 .22 -.06 Wages
W[ .96 .94 .92 .93 .95 .74* -.16 -.08 .15 .24 Aw
t -.30 -.05 -.22 -.04 .13 Interest rate
Rt .62 .26 .06 .03 -.01 .76* -.20 .01 .03 -.03
AR,
-.05 -.27 -.15 -.01 -.15 Disposable incomeIt .99 .99 .98 .98 .97 .71* -.46 .79 -.54 .41
AI,
-.15 -.16 .16 -.01 -.22Allowable drain v
t .97 .94 .89 .85 .80 .97* .12 -.13 -.06 .11
AV
t -.06 .04 -.09 -.09 .03
In Table2 the results ofthe
Jarque-Bera normality
test(£
2N
~test) and the
Dickey-Fuller
(DF)
and Durbin-Watson(CRDW)
unitroot tests arepresented (for
adescription
of thetest statistics, see
Hendry,
1989, andEngle
&Granger 1987).
Since thegraphs
of theseries (and theirdistribution
functions)
indicatedthat therehavebeenone-timechanges
inthe structure ofthe series (as a resultofthedevaluationin 1967andthe severe recession
aftertheoil crisis in
1973),
wealsouse the modifiedDickey-Fuller
test, denotedby DF*,
suggested by
Perron(1989).
The CRDW andDFtests arevalid,
since theresults indicatedthatall theseries followthe
AR(1)
process asrequired.
However, it should be borne inmindthatsmall
sample
biasmaybe present.Theresults in Table2 indicatethat alltheseries, except
Q
(,
are
normally
distributed.Thenon-normality
oftheQ
(
series is dueto the
significant
outliersin thesample
in 1976and1978.
According
to the DF and CRDW test, theQ
t
series followsa
1(0)
process, thePseries
being just
below the 5 percent criticalvalue, whileall theother series areclearly
nonstationary.
However, theDF* test shows that iftheone-timechanges
are filtered outanda timetrend
included,
all theseries,except theWtandV
t
series,
are1(0).
1The above results indicate that the
nonstationarity,
apparent in manyexpanding
roundwoodmarkets
(cf.
Brännlundetai. 1985 andNewman1987),
is not aproblem
in thepulpwood
market in Finland. Therefore, for the data used in the presentstudy,
cointegration
models are not relevant formodelling
thesupply
of and demand forpulpwood (Engle
andGranger
1987).2Further, tests on Wt
and C
t series, on the one hand, andon Vtand It
, on theother, indicatedthat
although
the series themselvesare 1(1), their linearcombinationsarestationary (i.e., they
arecointegrated).
3 Therefore, theordinary
*
least squares
(OLS)
estimates of these variables are(super)
consistent, but theirdistributions are not normal.
Table2.
Normality
Test(x
2n
)
andUnitRootTests(DF, DF, CRDW)
Notes:Critical value for DF testat 5% significance level and25 observations is3.00 (Fuller 1976);
forDF* testcritical values vary between -3.68and -3.80(Perron 1989). Critical valuefor CRDW -test
at5% significancelevel and 30 observations is0.79 (Bhargava1986).Critical value for
"/}
-test at5%level and 2 degrees offreedom is5.99. Itdoes nothave significant outliers andDF* testwasnot
computed forit.
3.2
Specification
andEstimationof
theStatisticalandEconometricModelFollowing Hendry
et al.(1988)
andSpanos (1990),
we draw a distinctionbetween thestatisticalmodel
(the
system orreducedform)
andthe econometricmodelofthesystem.Thestatisticalmodelis defined
by
theset ofvariablesofinterest(suggested by
economictheory),
their status (classification into modelledand nonmodelledvariables)
and thelag
polynomials
involved. The statistical model summarizes thesample
informationandensures that the statistical
assumptions underlying
the modelare validforthedata used.IfSeries *2
N
DF DF* CRDW
Pulpwood quantity Qt
7.64 -3.00 -6.20 1.00
Stumpage price Pt 0.35 -2.70 -9.10 0.67
Export price PXt 0.75 -1.79 -5.50 0.47
User costC
t 2.45 -1.66 -5.40 0.22
Wages Wt 2.20 -1.49 -3.30 0.10
Bank lending rate R(
4.16 -2.37 -5.00 0.73
Disposable income It 1.62 -2.31 0.04
Allowable drain V, 1.53 -0.18 -2.20 0.07
thesystem is not
statistically
valid, thereis littlepoint
inimposing
furtherrestrictions onit, and tests thereofwillbe
against
aninvalidbaseline. Oncethesystem hasbeen foundtobe
statistically adequate,
the structural econometric model is derivedby imposing
zerorestrictions,
implied by
the economictheory,
on the reduced form. Thevalidity
of thestructuraleconometric modelis
judged
onthe basis oftheoveridentifying
restrictionsandusing diagnostic
testsformisspecification.
The HS
approach emphasizes
statisticaladequacy
rather than the modeller'ssubjective
decisions in the
process of
specifying
a model.Consequently,
theproblem
of"multiple
hypotheses", i.e.,
too many structural modelssupporting
the data, is tackled in asystematic
way. Thus,in order to makestructuralmodelsdirectly comparable,
a commonstatistical model
(reduced form)
isfirst
estimatedand its statisticalvalidity
is checked.Second,
theoveridentifying
restrictionsimplied by
differenttheoreticalhypotheses
andtheirstatistical
validity
canbe tested.Finally,
for themodels which survive the first twostages, theselection ismadeintermsofparameterconstancy,robustness and
parsimony.
Below, we first formulate the statistical model, check its
validity,
thensimplify
it viaparameterrestrictions in orderto
specify
thestructural econometricequations
and test theresulting
econometricmodel.3.3 Statistical Model
Theoretical
supply
anddemandequations (eqs. (1)
and (3)) determinethe basic variablestobe included in thestatistical model, but the
dynamics
is dictatedby
thedata. Even inthe case of a
"two-period"
theoreticalpulpwood supply equation,
it is notpossible
toderive,
apriori,
theexplicit adjustment
process. Thetheory
behindroundwoodsupply
anddemand does not yet
specify
howquickly
agents react tochanges
in
prices
and thechoice ofexpectations
structure(static, rational, adaptive, etc.),
apriori,
is ad hoc. Wemake noa
priori assumptions
aboutthedynamic
behaviorandconsequently
therole of
expectations
in the modelis not addressedexplicitly.
Short-rundynamics
isdetermined
solely by
thedata.Totake intoaccount short-term
dynamics, lagged
variables are included.Because ofthesmall
sample
size and thelarge
numberofvariablesinthesystem,only
a limitednumberof
lags
can be introducedsimultaneously.
Based on theexperiments
with differentlag
structures, a system
consisting
ofeq.(4)
andeq.(5)
was found to be (onthe basis ofthetest criteria
reported
in Table3) statistically
the mostadequate
summaryof thesample
information. The estimation results are shown in Table 3. Because the reduced form is a
statisticalsummaryof
sample
information,parsimony
is notrequired
atthisstage and the
modelis
deliberately overparameterized,
When estimated
using Ordinary
LeastSquares (OLS), eq.(4)
and eq.(5) passed
all thetests
concerning
the classicalassumptions
of a linearregression
model(see
Table3).
However, due to the small
sample
size, theprecise
estimation of parameters is notpossible.
3.4 Econometric Model
The structural
supply equation
was derivedby imposing
zero restrictionsimplied by
eq.(1) on eq. (4).
Similarly,
to obtain a demandequation,
weimposed
zero restrictions(4)
Pt=ao+
aiQ,_i+ a2Qt-2+
«3pt-i+«4PXt+ a6R
t„!+ a
7
It+agVt+
09W t
+
«loQ+Mt
(5) Q
t=
j3
0+j3
1Q
t_ 1+/3
2Qt
_2+ftP
t_,+/3
4PX l+j3
5Rt+j3
6Rt_l+frit
+&V
t+
ftW t
+
ftoC
t+£t-suggested by
eq.(3)
oneq.(5).
4 It turnedout that we were unableto find astatistically
valid demand
equation
when the exportprice (PX
t)
was includedexplicitly
in thestructural model.5 The
high collinearity
of the real exportprice
and thereal stumpageprice
indicatesthat fluctuationsin exportprices
are transmitted to stumpageprices (c.f.,
Forsman & Heinonen
1989).
Becausethedemandequation
ishomogeneous
ofdegree
zeroin
product
and factorprices,
one oftheprices
can befactored out. We usedtheproduction
price
index in thedemandequation
as a proxy for theprice
ofthe "otherinputs" (i.e.,
energy andresidual
materials)
andtheexportprice,
andconsequently,
as adeflator.6Sincethe
production price
index turnedout tobe almost identical to the nominal exportprice
series,
theinformationof theexportprice
is includedin themodel,although
we are notable toobtainanestimateforthe
elasticity
ofdemandwithrespect totheexportprice.
On the basisof different
diagnostic
tests(shown
in Table 4and5),
themostsatisfactory
representation
ofthedataproved
to bethestructuralmodelshown ineq.(6)
andeq.(7)
where aQ and
/3q
denote the constant terms. It may be notedthat both thesupply
anddemand
equations
includevariablesthatare 1(1) series. However,since Itand V
t
series are
cointegrated
ineq.(6),
as are the Wt
and C
t
series in
eq.(7),
the parameterestimates ofthese series are super consistent in
large samples, although
the distribution of theirt-valuesis nonstandard
(Fuller 1976, Engle
andGranger 1987).
Eqs. (6)
and(7)
were estimatedusing OLS,
recursive leastsquares(RLS),
two-stage leastsquares
(2SLS)
andthree-stage
least squares(3SLS)
estimation methods. However, sincethe 3SLS estimation results did not differ
significantly
from the 2SLS results, weonly
report thelatter.
(6) Q
ts =Oq +ociP
t+ a 2Pt_i +
oi3l
t+ a 4Vt+ aSAR
t + +et
(7) Q
td
=A)
+APi
+&Pl-1
+ +/3
4Ct+ftQt-i
+Mt»Table3.TheEstimatedOLSResults for theReducedForm
Equations (Statistical Model),
1960-1988.
Notes: df.denotes
degrees
offreedom.Symbols
ofteststatistics areexplained
in theAppendix,
t-statistics areinparentheses.
Independent variable
Pulpwood price, P[
Quantity traded, Qt
Constant
Pt-1
Qt-i
Qt-2
Ct
w t
It
R t
Rt-i
Vt
PX t
-8.14
(1-02) 0.42 (1.96) 0.09
(0.51) 0.14
(0.78) 0.25 (0.33) -0.07 (0.18) -0.27 (0.36) 0.01 (0.96) -0.01
(1.34) 0.66 (0.28) 1.35 (2.12)
-9.12
(1.15) -0.82 (3.80) 0.29 (1.70) -0.11
(0.59) 0.45
(0.59) -0.33 (0.88) -0.05 (0.06) 0.01
(0.33) -0.01
(0.45) 3.06 (1.32) 0.34
(0.53)
Model RSS R2 DW F
RAC
y2
F Fc f
arch
Pt df.
0.396 0.74 1.99 2.00 3,13
0.08 2
4.66 10,16
0.20
1,15
0.63 3,10
Qt df.
0.397 0.75 2.07 2.16 3,13
1.38 2
4.74 10,16
2.87 1,15
0.28 3,10
3.5
Resultsfor
theSupply Equation
Theestimatedresults forthe
supply equation (6)
aregiven
in Table4(referred
to asSIAand SIB). The re-estimationresults ofthe
pulpwood supply equation
of Kuuluvainenetal., where
are also
reported (referred
to as S2A and S2B) inTable4. In eq. (8)adaptive pulpwood
price expectations
are assumed, the cross effect from thesawtimberstumpageprice
(SPt)is included and the effect of
disposable
income is taken into account as the firstdifference.7 The estimations of the Kuuluvainen et al.
(1988)
model wereoriginally
computed
for theperiod
1965-1985using
2SLS and a correctionforautocorrelation.Herewe have re-estimated it for the
period
1960-1988 without the correction forautocorrelation.8
Foreq.
(6),
theestimatedparametersofthe1(0)
variablesarestatistically significant
in theOLS estimations. Thetest resultfor the overidentificationrestrictions
Ct
2 01~
test)
in the2SLS estimations indicatedthat therestrictions
imposed
on the reduced form are valid.This
implies
that the structural modelparsimoniously
encompasses the unrestrictedreduced form (see
Hendry 1988).
The parameter estimates are not very sensitive to theestimationmethod
(OLS,
2SLS or3SLS),
except forthe stumpageprice
variable. Alltheothercoefficients are of similar
magnitude
and the t-values show that the2SLS methodsdo not increase the
efficiency
of the parameter estimates, OLS estimationproducing
markedly higher
t-valuesfor some of theparameters(Pt
,
Pt_i,ADt,
AR
t). The
endogeneity
of stumpageprice
isrejected by
theGranger causality
test, which may be an indicationthatthedifferencesin OLSestimationand2SLS estimationsare dueto small
sample
biasin 2SLS.9
(8) Q
t=
(j)o+ <j)lP
t+<j>2Pl-l
+<J>3SPi+ 4>
4Qt-l
+Table4.EstimatedResults for
Supply
ofPulpwood,
1960-1988.Notes:df. denotes degrees offreedom. Symbols ofteststatistics are explained inthe Appendix, t-statisticsare inparentheses. Due tothedifferent estimation methodA and Bmodelshave different teststatistics.
Independent S1A SIB S2A S2B
variable OLS 2SLS OLS 2SLS
Constant -7.07 -7.42 3.26 2.00
(4.16) (3.75) (2.72) (1.85)
Pulpwood price, Pt 0.65 0.81 0.53 0.07
(4.95) (1.86) (2.57) (0.17)
Lagged pulpwood -1.00 -1.07 -0.82 -0.61
price, Pt_j (8.68) (4.99)) (5.20) (2.70)
Disp. income,It -0.59 -0.68 (2.85) (2.15)
ADisp. income,Alt -0.40
(-0.19) (0.49)
Allowable drain, 4.41 4.66
v
t (5.24) (4.30))
Lagged endogenous 0.44 0.49
variable, Qt_j (3.34) (3.27)
ALaggedendog. 0.20 0.21
variable,AQt_i (2.46) (2.38)
AInterest rate, -0.01 -0.02
ARt (2.37) (1.63)
Sawtimber price, -0.12 0.20
sp, (0.46) (0.56)
Model RSS R2 DW F Y2
F F F F_
RAC AN c ARCH F
S1A 0.205 0.85 1.94 0.33 1.20 22.32 1.33 0.23 0.66
df. 3,17 2 6,20 1,19 3,14 3,17
S2A 0.539 0.66 1.29 4.12 0.33 8.49 7.54 0.17 0.96
df. 3,19 2 5,22 1,20 3,16 3,19
Model RSS *2
I
DW y2 y2
K RAC K N
Ayi a=0
y2 x01
f
arch *2 f
SIB 0.221 0.44 1.80 2188 1.19
df. (2)12 (3)/3 2 (7)/7 5 3,14 (3)/3
S2B 0.663 5.29 1.60 968 32.46 1.84
df. C2)12 (3)/3 2 (6)/6 4 3,16 (3)/3
Comparing
the OLS estimationresults of the two differentmodelling approaches (Table
4),
one can see that thenewspecification (SIA)
has a betterfitthan theoldspecification
(S2A).
Furthermore, modelSIAis astatistically
validrepresentation
of the data, whilemodel S2A is not. ModelS2A does not pass theF -test for
serially
correlatederrors, as also indicated
by
thelow DW statistics.Kuuluvainen etal. usedacorrection forserial correlation, but the correct
procedure
would have been tore-specify
the model.Furthermore, model S2Adoes not pass the test for correct functionalform
specification
(F
c
-testfor
linearity)
andisoveridentified(x
2qj-test).
The better
performance
ofmodel SIA becomes more obvious when one looks atFigures
1-4.From
Figure
1, whichgives
theactualandfittedvalues, one can see thatmodel S2Asystematically underpredicts
the observedquantity
tradedduring
the 1980s(even
aftercorrection for
autocorrelation,
as can be seen from Kuuluvainen et al.1988)
andis,
without
dummy variables,
unabletoproduce
theturning points
of therecession after themid-19705.
Figure
2 shows theRLS estimates forthe stumpageprice coefficient,
Pt,for
both models.The
plots
make itpossible
to trace the evolutionofPt asmore andmore ofthe
sample
dataare used in the estimation.Clearly,
the Pt coefficient for SIA is morestable than for S2A.
Figures
3 and 4 indicate that the standard error lines of the RLSestimates for the
1-step
residuals for model S2A are morespread
out (+0.3)
than formodel SIA
(+ 0.2). Further,
model S2Aoverpredicts
theresidual pattern for theperiod
1977-1988 more than does model SIA. The lower
portion
of theplots
shows theprobability
values for thosesample points
where thehypothesis
ofparameter constancywouldbe
rejected
atthe5,
10or 15percent levels.Figure
1.Actualand fittedvalues forS 1AandS2aFigure
2. Recursiveestimates forstumpageprice
coefficientFigure
3. Recursiveresiduals for S1AFigure 4. Recursive residuals forS2A
ForthemodelSIA, theshort-term
supply
reactspositively
to anincrease inthestumpage
price (the
elasticities are 0.64 for OLS and 0.81 for2SLS).
However, thelong-term
ortotal
supply elasticity
isnegative
andrather small in absolute terms(-0.36
for OLS and-0.27 for
2SLS).
10 The F-test for omitted variables(Hendry, 1989)
suggests that thelong-term
effectsofdisposable
incomeand annualallowabledrain mustbe included.Theelasticity
ofsupply
with respect todisposable
income is-0.58,
which islarger
than theelasticities obtainedfrommicro data
(cf.
Kuuluvainen& Salo1991).
Thelarge elasticity
for theallowablecut should not be
interpreted
as theelasticity
ofsupply
with respecttochanges
in thegrowing
timber stock as the variable measures theaggregated
annualincrement of the stock. The coefficientsof the difference of thereal interestrate and the
lagged endogenous
variable capture short-term structural shocks. These variables donotseem to have
long-term
effects on timbersupply
as theirlevels(present
orlagged)
wererejected by
the F-test for omitted variables when the difference terms were included.Although,
theimpact
effect oftheinterestrateappearsreasonable,as amajor
partofloanshave not had fixedinterestrates, itis difficult to
judge
whetherthevariable isrelatedtoinflationary expectations,
tochanges
in monetarypolicy,
ortosomeotherfactor.Theelasticitiesof
stumpage
prices
inmodels SIA an S2A are notmarkedly different,
butthe
dynamics implied by
the twospecifications
arecompletely
different. This is animportant
resultif,
forexample,
short-termforecasting
is of interest.Furthermore,
according
to model SIA, disposable
income seems to have along-term
effect onpulpwood supply, compared
to the short term effect indicatedby
the oldspecification
(note
thattheelasticity
ofthedifferenceterm inKuuluvainenetai.(1988)
was -23.7 whencorrectionfor serial correlationwasused,
hardly
arealisticorderofmagnitude).
3.6Results
for
theDemandEquation
As inthe case of the
supply equation,
the estimatedresults for thedemandequation (7)
are
compared
to those obtainedfromre-estimating
Kuuluvainenet ai.'s(1988)
demandequation,
where D75/76 and D7B/79 are
dummy
variablestaking
account of structuralchanges
connectedwith export market
developments
after theenergy crisis. Theresults from theOLS and2SLSestimationsofeqs.
(7)
and(9)
are shownin Table5(referred
to asDIA, Band
D2A,
B,respectively).
Theresults in Table5 show thatthe newdemand
specification (DIA, B)
has a betterfitthan the Kuuluvainen et ai.
(1988) specification (D2A, B).
Model DIA passes all thediagnostic
tests, while model D2Afails the test for correct functionalform (F -test forlinearity)
and the low t-statistics indicatesproblems
with thespecification.
Infact,
theFarch
"test shows thatheteroscedasticity
is aproblem
for model D2A. When theheteroscedastic consistent standard errors (not
reported here)
are used to compute thet-values,
only
theestimate for theD77/78dummy
variable had a t-value above2. Thus,model D2A is not a
satisfactory representation
ofthe datageneration
process.Although
the
diagnostic
test results for 2SLS estimations of model 2 (i.e., D2B) indicate nosignificant problems
with thespecification,
the test statistics are poorer than for modelDIB.In
particular,
theresidual sum ofsquares(RSS)
ishigh, indicating high
varianceinthemodel.Furthermore,modelD2B
clearly
failedtopass theoveridentificationtest.(9) Q
t=
<SQ
+<siP
t +SzPX,
+SjQ,-,
+ 54D75/76+65D78/79+ tj,
,