CROSS-COMMODITY PRICE TRANSMISSION AND INTEGRATION … · Horisontaalinen hintasiirtymä...
Transcript of CROSS-COMMODITY PRICE TRANSMISSION AND INTEGRATION … · Horisontaalinen hintasiirtymä...
PTT työpapereita 170 PTT Working Papers 170
CROSS-COMMODITY PRICE TRANSMISSION AND INTEGRATION
OF THE EU LIVESTOCK MARKET OF
PORK AND BEEF: PANEL TIME-SERIES APPROACH
Hanna Karikallio
Helsinki 2015
2
PTT työpapereita 170 PTT Working Papers 170 ISBN 978-952-224-170-2 (pdf) ISSN 1796-4784 (pdf) Pellervon taloustutkimus PTT Pellervo Economic Research PTT Helsinki 2015
3
Karikallio, H. 2015. CROSS-COMMODITY PRICE TRANSMISSION AND INTEGRATION OF THE EU LIVESTOCK MARKET OF PORK AND BEEF: PANEL TIME-SERIES APPROACH. Working Papers 170. 102 p. ISBN 978-952-224-170-2 (pdf), ISSN 1796-4784 (pdf).
Abstract: At this study we analyze horizontal cross-commodity price transmission and
integration of the EU livestock market of pork and beef. The study seeks to investigate
whether or not there are short-term and long-term relationships between pork and beef
prices in the EU. Our focus is on cross-commodity price transmission which provides
valuable insights into market integration and efficiency. We utilize recently developed
panel time-series techniques. Our data consists of monthly data on pork and beef prices
in the EU member states during the period from February 1995 to June 2014. The
estimation results reveal that there exists bi-directional relationship between pork and
beef prices in the EU in long run. Cross-commodity price transmission between pork
and beef has increased remarkably during the past ten years in the EU15 member states.
Also the convergence to the equilibrium has sped up. In short run, we found evidence
only for price transmission from pork prices to beef prices, not vice versa. Overall,
short-run dynamics is significantly different in the EU livestock market compared to
long run dynamics.
Key words: Price transmission, market cointegration, panel time-series, pork and beef
prices
Karikallio, H. 2015. HINTASIIRTYMÄT JA INTEGRAATIO EU:N SIANLIHA-
JA NAUDANLIHAMARKKINOIDEN VÄLILLÄ: PANEELI-AIKASARJOJEN
ANALYYSI. PTT työpapereita 170. 102 s. ISBN 978-952-224-170-2 (pdf), ISSN
1796-4784 (pdf).
Tiivistelmä: Tutkimuksessa tarkastellaan EU:n lihamarkkinoiden integraatiota
analysoimalla tuottajahintamuutosten välittymistä sianliha- ja naudanlihamarkkinoiden
välillä. Tutkimuksessa selvitetään, onko sianlihan ja naudanlihan tuottajahintojen välillä
lyhyen tai pitkän aikavälin yhteyttä ja tasapainoa. Mielenkiinnon kohteena on siis
hintasiirtymät hyödykemarkkinoiden välillä, mikä tarjoaa arvokasta tietoa markkinoiden
integroitumisesta ja tehokkuudesta. Empiirinen analyysi perustuu viime vuosina
kehitettyihin paneeli-aikasarjamenetelmiin. Tutkimuksessa hyödynnettävä aineisto
koostuu EU-maiden sianlihan ja naudanlihan kuukausittaisista tuottajahinnoista 2/1995-
6/2014. Tulokset osoittavat, että sianlihan ja naudanlihan tuottajahinnan välillä on
EU:ssa kaksisuuntainen pitkän aikavälin tasapaino. Hintasiirtymät ovat voimistuneet ja
nopeutuneet huomattavasti viimeisen kymmenen vuoden aikana. Lyhyen aikavälin
tarkastelu osoittaa, että hintasiirtymiä tapahtuu vain sianlihan tuottajahinnoista
naudanlihantuottajahintaan, ei toisinpäin. Kaiken kaikkiaan EU:n lihamarkkinoiden
lyhyen aikavälin hintadynamiikka on hyvin erilainen verrattuna pitkän aikavälin
dynamiikkaan.
Asiasanat: Hintasiirtymä, markkinaintegraatio, paneeli-aikasarja, lihan tuottajahinnat.
4
5
KESKEISET TULOKSET
Tulokset osoittavat, että sianlihan ja naudanlihan tuottajahinnan välillä on EU:ssa
kaksisuuntainen pitkän aikavälin tasapaino. Hintasiirtymät sianlihan ja naudanlihan
välillä ovat merkittävimpiä vanhoissa jäsenmaissa (EU15), mutta hintamuutosten
välittyminen lihamarkkinoilla on tilastollisesti merkitsevää myös uusien jäsenmaiden
sisältyessä analyysiin. Sianlihan tuottajahinnan vaihtelut siirtyvät naudanlihan
tuottajahintaan voimakkaammin kuin naudanlihan tuottajahinnan vaihtelut sianlihan
tuottajahintaan. EU:ssa sianliha hallitseekin lihamarkkinoita. Hintasiirtymät sianlihan ja
naudanlihan markkinoiden välillä ovat voimistuneet ja nopeutuneet huomattavasti
viimeisen kymmenen vuoden aikana. Hintojen konvergoituminen pitkän aikavälin
tasapainoon on kuitenkin edelleen melko hidas prosessi, mikä kertoo EU:n lihamarkki-
noilla vielä olevista markkinoiden integroitumisen vaikeuksista. Lyhyellä aikavälillä
hintasiirtymiä tapahtuu vain sianlihan tuottajahinnoista naudanlihantuottajahintaan, ei
toisinpäin. Kaiken kaikkiaan EU:n lihamarkkinoiden lyhyen aikavälin hintadynamiikka
on hyvin erilainen verrattuna pitkän aikavälin dynamiikkaan. Sianlihamarkkinoita
EU:ssa hallitsevat Tanska, Saksa ja Hollanti. Kyseisten maiden sianlihamarkkinoilla
tapahtuvat hintavaihtelut vaikuttavat voimakkaimmin naudanlihamarkkinoille
välittyvien hintasiirtymien suuruuteen ja nopeuteen. Naudanlihamarkkinoita EU:ssa
hallitsevat vastaavasti Ranska, Irlanti ja Iso-Britannia.
Tulosten perusteella voidaan todeta, että EU:ssa on vuorovaikutteiset ja yhtenäistyvät
lihamarkkinat. Lihamarkkinoiden integraatio on edennyt varsin nopeasti. Maatalous-
tuotteiden kaupan esteiden vähentäminen on keskittänyt toimialaa voimakkaasti.
Unionin sääntelyn on jatkossakin oltava oikeudenmukaista jäsenmaihin nähden:
kaupanesteitä pitää torjua, kilpailuedellytyksiä yhtenäistää sekä huolehtia, että yhteisiä
normeja noudatetaan tasapuolisesti. Kilpailukykyä heikentäviä yksipuolisia velvoitteita
ei tule ottaa käyttöön. Lihamarkkinoiden integroituminen voimistui samaan aikaan, kun
markkinaheilahtelut maataloustuotemarkkinoilla lisääntyivät. Integraatio on näin ollen
edistänyt myös hintaheilahteluiden leviämistä markkinoiden välillä. Markkina- ja
hintariskien kasvaessa niiden hallintaan on panostettava enemmän. Markkinaintegraati-
on näkökulmasta lihamarkkinat EU:ssa ovat toimivat ja tehokkaat. Politiikkatoimenpi-
teiden vaikutukset menevät yhä nopeammin lihamarkkinoiden läpi. Hintaintegraatiosta
johtuen kohdennetuilla politiikkatoimenpiteillä on vaikutusta koko lihamarkkinoiden
toimintaan. EU:n lihamarkkinoita koskevat politiikkapäätökset onkin tehtävä koko
lihamarkkinoiden ja niiden toimivuuden näkökulmasta.
6
7
YHTEENVETO
Markkinoiden integroitumisella on keskeinen merkitys EU:n kasvu- ja kilpailustrategi-
assa. Yhtenäinen sisämarkkina-alue tarjoaa tuotteille laajat markkinat, mutta samalla
lisääntyvä kilpailu pakottaa yhdenmukaiseen hinnoitteluun koko EU:n alueella.
Taustalla on yhden hinnan lain (Law of One Price, LOP) toteutuminen, joka on eräs
kansainvälisen talouden perusoppeja. Sen mukaan samalla tuotteella tulee olla sama
hinta eri maissa, kun kansalliset hinnat ilmaistaan samassa valuutassa.
Tietämys markkinoiden yhdentymisestä ja markkinoiden välisestä vuorovaikutuksesta
on tärkeää maan kilpailukyvyn näkökulmasta. Hintojen välttyminen markkinoilta
toisille heijastaa markkinoiden integraation syvyyttä ja markkinoiden tehokkuutta.
Hintasuhteiden analysointi onkin keskeinen työkalu markkinoiden integraation
analysoinnissa.
Maatalousmarkkinoiden yhdentyminen on ollut aina EU:n keskiössä. Maatalousmarkki-
noiden toimintaedellytyksiin vaikutetaan Euroopan unionin yhteisellä maatalouspolitii-
kalla (Common Agriculture Policy, CAP), jonka tavoitteena on muun muassa
maatalouden tuottavuuden parantaminen, elintarvikemarkkinoiden vakauttaminen ja
kohtuullisten elintarvikehintojen varmistaminen kuluttajille EU:n jäsenmaissa.
Yhteisellä maatalouspolitiikalla integroidaan kansalliset maataloustuotemarkkinat
jäsenvaltioiden sisällä ja välillä, jolloin tuotantokustannuksiin perustuvat kansalliset
hintatiedot välittyvät tehokkaasti muiden EU-maiden maataloustuotemarkkinoille.
Monissa tutkimuksissa on tarkasteltu maataloustuotteiden hintojen välittymistä
kansallisten maatalousmarkkinoiden välillä.1 Useimmissa kyseisistä tutkimuksista myös
päädytään siihen, että maiden väliset hintaerot ovat EU:ssa pienentyneet. Aikasarja-
analyyseihin perustuvien tutkimuksia vaikeuttaa kuitenkin maatalousmarkkinoita
häirinneet toistuvat eksogeeniset shokit, joiden huomioiminen mallinnuksessa on
haastavaa.
1 Esimerkiksi Sanjuan et al.(2001), Serra et al. (2006), Fousekis (2007), Ihle et al. (2012), Liu (2011) ja Meyer (2012).
8
Viime vuosina empiirinen tutkimus hintojen välittymisestä elintarvike- ja maatalous-
markkinoilla on saanut paljon huomiota. Hintasiirtymät voidaan karkeasti jakaa kahteen
pääryhmään: vertikaalisiin ja horisontaalisiin hintasiirtymiin. Vertikaalisilla hintasiirty-
millä (vertical price transmission) tarkoitetaan hintojen välittymistä elintarvikeketjun
sisällä. Horisontaalinen hintasiirtymä (horizontal price transmission) taas viittaa hinnan
välittymiseen elintarvikeketjussa samassa asemassa olevien markkinoiden välillä.
Määrittelyyn liittyy markkinoiden rajaaminen jollakin kriteerillä. Yleensä horisontaali-
sella hintasiirtymällä tarkoitetaan hinnan välittymistä tietyin alueen markkinoilta toisen
alueen vastaaville markkinoille (spatiaalinen hintasiirtymä). Se voi kuitenkin viitata
myös hinnan välittymiseen hyödykemarkkinoiden välillä. Kahden tuotteen voidaan
katsoa olevan samoilla markkinoilla ja näin ollen läheisiä substituutteja, jos niiden
hintojen suhde pysyy ajan kuluessa vakaana. Tämä perustuu tuotteiden keskinäiseen
korvattavuuteen, mikä puolestaan riippuu tuotteiden kysyntäfunktioista ja kuluttajien
preferensseistä. Jos tuotteet ovat substituutteja, toisen tuotteen markkinoita häiritsevän
shokin vaikutukset siirtyvät myös toisen tuotteen markkinoille, jos kyseisten tuotteiden
markkinat ovat integroituneet. Hinnan nousu saa tässä tapauksessa kuluttajat
vaihtamaan ostokohteekseen korvaavan tuotteen. Näin ollen läheiset substituutit tulisi
luokitella kuuluviksi samoille markkinoille. Hyödykemarkkinoiden välisten vuorovai-
kutusten ja hintasiirtymien ymmärtäminen muodostuu erityisen tärkeäksi, jos sääntelyllä
halutaan vaikuttaa hyödykemarkkinoiden toimintaan.
Tutkimuksen tavoitteet
Tässä tutkimuksessa tarkastellaan EU:n lihamarkkinoiden integraatiota analysoimalla
tuottajahintamuutosten välittymistä sianliha- ja naudanlihamarkkinoiden välillä.
Mielenkiinnon kohteena on siis hintasiirtymät hyödykemarkkinoiden välillä, mikä
tarjoaa arvokasta tietoa lihamarkkinoiden integroitumisesta ja markkinoiden
tehokkuudesta.
EU:ssa naudan ja sian ruhot luokitellaan standardoidulla laatuluokituksella2. Yhtenäinen
luokittelu on kansallisten lihamarkkinoiden integroitumisen perusta, mutta edelleen
maiden välisiä hintavertailuja vaikeuttaa eri luokkiin kuuluvien eläinten määrien
maakohtaiset vaihtelut. Sen sijaan siipikarjan ruhojen osalta laatuluokittelu ei vielä ole
EU-maissa yhtenäistä, minkä takia tuottajahinnatkaan eivät ole vertailukelpoisia. Tästä
syystä tutkimuksessa onkin rajoituttu tuottajahinnan muutosten välittymiseen
2 Naudan ruhot luokitellaan lihakkuuden mukaan kirjaimin E, U, R, O ja P. Lihasiat luokitetaan Hennessy GP4 -mittarilla.
9
sianlihamarkkinoiden ja naudanlihamarkkinoiden välillä ja sipikarjanlihamarkkinat
jätetään analyysin ulkopuolelle3. Tutkimuksessa selvitetään, onko sianlihan ja
naudanlihan tuottajahintojen välillä EU:ssa lyhyen aikavälin tai pitkän aikavälin
yhteyttä ja tasapainoa.
Tutkimuksen merkitys suhteessa olemassa olevaan kirjallisuuteen maataloustuotemark-
kinoiden integroitumisesta ja hintasiirtymistä on kahtalainen. Ensinnäkin tutkimuksen
kohdentuminen hyödykemarkkinoiden välisiin hintasiirtymiin ei ole kirjallisuudessa
yhtä yleistä kuin vertikaalisten ja spatiaalisten hintasiirtymien tarkastelut. Tutkimus
tarjoaakin mielenkiintoista tietoa substituuttihyödykkeiden hintojen välisestä yhteydestä
ja markkinoiden integraatiosta.
Tutkimuksen toinen kontribuutio aihepiirin aikaisempaan tutkimukseen nähden liittyy
tilastolliseen analysointiin ja tutkimuksessa hyödynnettäviin ekonometrisiin menetel-
miin. Tutkimuksen empiirinen analyysi perustuu viime vuosina kehitettyihin paneeli-
aikasarjamenetelmiin. Toisin sanoen hintamuutosten välittymistä lihamarkkinoilla
testataan menetelmillä, jotka hyödyntävät sekä aineiston ajallista vaihtelua että EU-
maiden välistä vaihtelua. Tutkimuksessa hyödynnettävä kuukausitason aineisto koostuu
EU-maiden sianlihan ja naudanlihan tuottajahinnoista ajanjaksolta 2/1995-6/20144.
Tutkimuksen tulokset
Tulokset osoittavat, että sianlihan ja naudanlihan tuottajahinnan välillä on EU:ssa
kaksisuuntainen pitkän aikavälin tasapaino. Tulosten mukaan hintasiirtymät sianlihan ja
naudanlihan välillä ovat merkittävimpiä vanhoissa EU:n jäsenmaissa (EU15), mutta
hintamuutosten välittyminen lihamarkkinoilla on tilastollisesti merkitsevää myös uusien
jäsenmaiden sisältyessä analyysiin. Sianlihan tuottajahinnan vaihtelut siirtyvät
naudanlihan tuottajahintaan voimakkaammin kuin naudanlihan tuottajahinnan vaihtelut
sianlihan tuottajahintaan. EU:ssa sianliha hallitseekin lihamarkkinoita. Tarkastelupe-
riodista riippuen sianlihan tuottajahinnan prosentin nousu nostaa naudanlihan
tuottajahintaa 0.15–0.33 prosentilla. Naudanlihan tuottajahinnan vaihteluiden
siirtyminen sianlihan tuottajahintaan voidaan havaita vain tuloksissa, jotka koskevat
tarkasteluperiodia 1/2007–6/2014. Kyseisen tarkasteluperiodin tulokset osoittavat, että
3 Tutkimuksessa käytetyt lihan laatuluokat: nauta R3-sonni, sika luokka E. 4 Tutkimuksessa hyödynnetään kolmea eri aineistoa EU15 2/1995-6/2014, EU25 1/2005-6/2014 ja EU27 1/2007-6/2014.
10
prosentin nousu naudanlihan tuottajahinnassa aiheuttaa 0.1–0.2 prosentin nousun
sianlihan tuottajahinnassa. Tarkastelut pidemmältä periodilta paljastavat, että vain
sianlihan tuottajahinta on välittynyt merkitsevästi naudanlihan tuottajahintaan.
Hintasiirtymät sianlihan ja naudanlihan markkinoiden välillä ovat voimistuneet ja
nopeutuneet huomattavasti viimeisen kymmenen vuoden aikana EU15-maissa. Ennen
vuotta 2005 osaotoksiin perustuvissa estimointituloksissa ei havaittu yhteyttä sianlihan
ja naudanlihan tuottajahintojen välillä. Kyseisen vuoden jälkeen EU:n lihamarkkinoiden
integroituminen on ollut nopeaa ja hintasiirtymät tuottajahintojen välillä ovat
tilastollisesti merkitseviä. Lihamarkkinoiden integroitumisen voimistuminen on
nähtävissä myös EU25- ja EU27-maita tarkasteltaessa.
Sianlihan ja naudanlihan tuottajahintojen ajautuessa pois pitkän aikavälin tasapainosta,
kestää tasapainon uudelleen saavuttaminen EU15-maissa noin vuoden. Ennen vuotta
2005 tasapaino saavutettiin uudestaan noin 1,5 vuoden hintojen sopeutumisella. Pitkän
aikavälin tasapainon savuttaminen onkin nopeutunut selvästi viime vuosina. Hintojen
konvergoitumisen voidaan kuitenkin katsoa olevan edelleen melko hidas prosessi, mikä
kertoo EU:n lihamarkkinoilla vielä olevista markkinoiden integroitumiseen vaikuttavis-
ta hidasteista ja jäykkyyksistä.
Lyhyellä aikavälillä tulokset osoittavat, että hintasiirtymiä tapahtuu vain sianlihan
tuottajahinnoista naudanlihantuottajahintaan, ei toisinpäin. Lisäksi lyhyen aikavälin
hintasiirtymiä ei voitu havaita EU:n lihamarkkinoilla ennen vuotta 2007. Tulokset myös
osoittavat, että havaittu lyhyen aikavälin hintasiirtymä sianlihan tuottajahinnasta
naudanlihan tuottajahintaan on negatiivinen. Tämä kertoo ennen muuta markkinoiden
reagointikyvyssä olevista viiveistä, jota ovat naudanlihamarkkinoilla suuremmat kuin
sianlihamarkkinoilla. Kaiken kaikkiaan EU:n lihamarkkinoiden lyhyen aikavälin
hintadynamiikka on hyvin erilainen verrattuna pitkän aikavälin dynamiikkaan.
Tulosten mukaan sianlihamarkkinoita EU:ssa hallitsevat Tanska, Saksa ja Hollanti.
Kyseisten maiden sianlihamarkkinoilla tapahtuvat hintavaihtelut vaikuttavat
voimakkaimmin EU:n naudanlihamarkkinoille välittyvien hintasiirtymien suuruuteen ja
nopeuteen. Tanska, Saksa ja Hollanti ovatkin selvästi lisänneet EU:n lihamarkkinoiden
integraatiota; EU-maiden naudanlihamarkkinat ovat selvimmin integroituneet kyseisten
maiden sianlihamarkkinoihin. Naudanlihamarkkinoita EU:ssa hallitsevat tulosten
mukaan vastaavasti Ranska, Irlanti ja Iso-Britannia. Sen sijaan ainakin Kreikan,
Kyproksen, Slovenian ja Slovakian sianlihan ja naudanlihan tuottajahintojen vaihtelut
vaikuttavat negatiivisesti tuloksiin EU:n lihamarkkinoiden integraatiosta. Kyseisten
11
maiden lihamarkkinoilta lähtevät hintavaihtelut välittyvätkin vain heikosti EU:n
lihamarkkinoille.
Johtopäätökset
Johtopäätöksinä tutkimuksen empiirisistä tuloksista voidaan ensinnäkin todeta, että
EU:ssa on vuorovaikutteiset ja yhtenäistyvät lihamarkkinat. Lihamarkkinoiden
integraatio on edennyt varsin nopeasti. Maataloustuotteiden kaupan esteiden
vähentäminen on keskittänyt ja keskittää edelleen toimialaa voimakkaasti. Unionin
sääntelyn on jatkossakin oltava oikeudenmukaista jäsenmaihin nähden: jäsenmaiden
välisiä kaupanesteitä pitää torjua, kilpailuedellytyksiä yhtenäistää sekä huolehtia, että
yhteisiä normeja noudatetaan tasapuolisesti. Kilpailukykyä heikentäviä yksipuolisia
velvoitteita ei tule ottaa käyttöön.
Maataloustuotemarkkinoiden toimintaympäristöä kuvaa nykyään nopeat ja merkittävät
heilahtelut. Suuret hintavaihtelut ovat tulleet osaksi markkinoiden toimintaa.
Lihamarkkinoiden integroituminen voimistui samaan aikaan, kun markkinaheilahtelut
lisääntyivät ja niiden merkitys toimialalla kasvoi. Integraatio on näin ollen edistänyt
myös hintaheilahteluiden leviämistä markkinoiden välillä. Markkina- ja hintariskien
kasvaessa niiden hallintaan on panostettava integroituvassa maataloudessa enemmän.
Markkinaintegraation näkökulmasta lihamarkkinat EU:ssa ovat toimivat ja tehokkaat.
Politiikkatoimenpiteiden vaikutukset menevät yhä nopeammin läpi koko lihamarkkinoi-
den. Hintaintegraatiosta johtuen kohdennetuilla politiikkatoimenpiteillä on vaikutusta
koko lihamarkkinoiden toimintaan. EU:n lihamarkkinoita koskevat politiikkapäätökset
onkin tehtävä koko lihamarkkinoiden ja niiden toimivuuden näkökulmasta.
12
13
TABLE OF CONTENTS
Yhteenveto ......................................................................................................................... 5
1 Introduction .............................................................................................................. 15
2 Background............................................................................................................... 18
2.1 Market integration and horizontal price transmission ....................................... 18
2.2 Meat market in the EU ...................................................................................... 20
2.3 Modelling price transmission in the panel framework ...................................... 22
3 Econometric methodology........................................................................................ 25
3.1 Panel unit root tests ........................................................................................... 25
3.1.1 First generation panel unit root tests (cross-country independence) ... 26
3.1.2 Second generation panel unit root tests (cross-country dependence) ... 30
3.1.3 Panel unit root tests allowing structural breaks .................................... 32
3.2 Cointegration tests ............................................................................................. 35
3.3 Estimation methods ........................................................................................... 43
4 Data description ........................................................................................................ 51
5 Empirical results ....................................................................................................... 57
5.1 Panel unit root results ........................................................................................ 57
5.2 Cointegration test results ................................................................................... 65
5.3 Estimation results .............................................................................................. 73
5.4 Robustness check against estimation method ................................................... 81
5.5 Robustness check against definitions of the data sample .................................. 83
6 Concluding remarks.................................................................................................. 86
References ....................................................................................................................... 89
Appendix 1. LM unit root test results for individual EU member states assuming
one structural breakpoint. .................................................................................. 96
Appendix 2. Pooled Mean Group ECM estimates, EU15 over different sub-periods97
Appendix 3. Cross-sectional stabilities of the long-run elasticity and the speed of
adjustment ......................................................................................................... 98
14
15
1 INTRODUCTION
In recent years, the empirical research on agricultural price transmission has gathered
considerable attention. It is frequently utilized concept in market integration analysis.
Price transmission can be separated into two types: horizontal and vertical. Vertical
price transmission refers to price linkages along a supply chain, whereas horizontal
price transmission means the price linkage occurring among different markets at the
same position in the supply chain.
The literature analyzing vertical price linkages has concentrated on evaluations of the
links between farm, wholesale and retail prices. The asymmetric vertical price
transmission, i.e., increasing and decreasing prices at one level of supply chain transmit
at different rates to another level, has aroused considerable discussion in agricultural
economics (Vavra and Goodwin, 2005). The price relationships along the supply chain
provide insights into marketing efficiency and consumers’ and farmers’ welfare (Aguiar
and Santana, 2002). Meyer and von Cramon-Taubadel (2004) and Frey and Manera
(2005) provide reviews of the literature on asymmetry price transmission.
The notion of horizontal price transmission usually refers to price linkages across
market places (spatial price transmission). However, it can also concern the transmis-
sion across different agricultural commodities (cross-commodity price transmission)5,
from non-agricultural to agricultural commodities (for example from energy prices to
agricultural prices; Serra et al., 2010 and Hassouneh et al., 2012), and across different
purchase contracts for the same commodity (for example from futures to spot markets
and vice versa; Baldi et al., 2013).
The key underlying theoretical explanation of spatial price transmission is the spatial
arbitrage and the consequent Law of One Price (LOP)6. On the contrary, for cross-
commodity price transmission, the co-movement of prices is mostly driven by the
substitutability and complementarity relations among the products (Saadi, 2011), which,
in turn, depends on the respective demand functions and on the underlying preferences.
Substitutability and complementarity essentially implies that the prices of two
commodities have a long-run relationship and shocks to one of the markets will get
transmitted to another if markets are integrated. Understanding of inter-commodity
price relationships and shock transmissions becomes important when price volatility and
5 Detailed discussion: Esposti and Listorti (2012, 2013). 6 The Law of One Price (LOP) is an economic concept which posits that "a good must sell for the same price in all locations". The intuition behind LOP is based on the assumption that differences between prices are eliminated by market participants taking advantage of arbitrage opportunities.
16
market failures formed the basis for public interventions. Even though, the background
theory differs between spatial and cross-commodity price transmission, the empirical
framework and the econometric implications of these different cases of horizontal price
transmission are the same.
At this study we analyze horizontal cross-commodity price transmission and integration
of the EU livestock market of pork and beef. The study seeks to investigate whether or
not there are short-term and long-term relationships between pork and beef prices in
EU. The contribution of this study to the existing literature on the price transmission in
agricultural market is twofold. The first contribution is the focus on cross-commodity
price transmission. Studies on this field are not as frequent as studies on vertical or
spatial price transmission although they provide valuable insights into market
integration.
The second contribution of our study refers to econometric methods: we utilize recently
developed panel time-series techniques. In other words, to study the causality
relationship between pork and beef prices, we catch both time-series and cross-country
variation in our data set. Our data consists of monthly data on pork and beef prices in
the EU member countries during the period from February 1995 to June 2014.
Panel models have many advantages compare to time series or cross-section models:
panel models make more information available, hence more degrees of freedom and
more efficiency. They allow controlling for individual heterogeneity. They also allow
identifying effects that cannot be detected in simple time series or cross-section data.
One benefit is better properties of the testing procedures when compared to more
standard time series methods. In the panel framework, we can analyze long-run
relationship across the panel while allowing the associated short-run dynamics and fixed
effects to be heterogeneous across different members of the panel (Banerjee, 1999;
Maddala and Wu, 1999).
To examine the linkages between pork and beef prices, we follow the standard three-
step approach consisting of (i) assessing the stationarity of the time series, (ii) in case
the variables are not stationary, checking whether they are characterized by a
cointegration relationship, (iii) in case cointegration holds, testing for long-run and
short-run relationships between pork and beef prices by estimating error correction
mechanism (ECM), which permits to analyse the long-run relationship between the
variables jointly with the short-term adjustment towards the long-run equilibrium.
In analyses of the price linkages in the meat sector, panel causality approach has been
utilized mainly in vertical price transmission studies, for example, Boetel and Liu
17
(2008), Goodwin and Holt (1999), Goodwin and Harper (2000) and Carraro and
Stefani (2010).
In our study, testing for unit root is performed using the panel unit root test of Levin,
Lin and Chu (2002), Im, Pesaran and Shin (2003), Breitung (2000), Maddala and Wu
(1999), Choi (2001), Pesaran (2003) and Im, Lee and Tislau (2005, 2010). The panel
cointegration test based on Pedroni (1999, 2004), Kao (1999), Westerlund (2005, 2007),
Westerlund and Edgerton (2008) are considered. In addition, we carry out Johansen’s
Fisher panel cointegration test suggested by Maddala and Wu (1999). Moreover, Mean-
Group estimator (MG) by Pesaran and Smith (1995), Pooled Mean Group estimator
(PMG) by Pesaran, Shin and Smith (1999), group-mean Dynamic OLS (DOLS) and
group-mean Fully Modified OLS (FMOLS) by Pedroni (2000) are implemented to test
for Granger-causality in panel models. We use the software Stata for empirical analyses.
The main objective of the study is to analyze the dynamics of meat price transmission
and market integration between pork and beef in the EU. The specific objectives are:
- To examine the trend in prices of pork and beef in the EU,
- To analyze the price integration between pork and beef markets in the EU,
- To identify the long and short run price transmission of pork and beef in the EU.
The rest of the paper is organized as follows: In the next section we briefly discuss on
the concept of market integration and price transmission and how it can be modelled in
the panel framework. We also take a quick snapshot on the EU meat market. In section
3 we describe the theory of the econometric methods we have utilized. Section 4 defines
the data and its limitations. Section 5 gives the results and discussion and the last
section concludes.
18
2 BACKGROUND
2.1 Market integration and horizontal price transmission
Market integration is an indicator that explains how much different markets are related
to each other. Price transmission reflects the extent of market integration and the extent
of market efficiency. Consequently, analysis of relationships between prices is a
common tool in market integration analysis. Even though theory in general argues that
other variables (product attributes) are equally important in describing and explaining
market integration, this interest in prices can be justified because data on prices are
easier to obtain, and often the only data available.
Price transmission can be understood from two aspects: vertical and horizontal. The
literature analyzing vertical price linkages has concentrated on evaluations of the links
between farm, wholesale and retail prices. In general, the primary attention in studies
that analyze vertical price transmission is the magnitude, speed and nature of the price
adjustments through the supply chain and especially the extent to which adjustments are
asymmetric.7
Although studies into vertical price transmission are the most frequent, horizontal price
transmission is the focus of this study. The literature analyzing horizontal price linkages
dates back more than one-hundred years and is typically concerned with links between
prices at different locations (spatial price arbitrage). Spatial price arbitrage is an
equilibrium concept, where in a well-functioning market, transactions between spatially
dispersed agents will ensure that price shocks occurring in one market evoke responses
in other market places (Serra et al., 2006). In theory, prices of homogeneous goods in
two separate locations will differ by, at most, the cost of moving the commodity form
the cheapest market place to the most expensive one. This arbitrage condition is
equivalent to the weak version of the Law of One Price (LOP) (Fackler and Goodwin,
2001). Numerous concepts such as market integration and market efficiency have been
used to describe spatial price arbitrage (Serra et al., 2006 and Fousekis, 2007). The
majority of the studies on spatial price relationships have been focused on food and raw
material markets.
7 Studies on vertical price transmission: Peltzman (2000), Meyer and von Cramon-Taubadel (2004), McCorriston and Sheldon (1996), Vavra and Goodwin (2005).
19
Markets are not related only between regions but also between commodities. Horizontal
price transmission also concerns the transmission across different agricultural
commodities (cross-commodity price transmission) (Esposti and Listorti, 2013). In this
case of horizontal price transmission the co-movement of prices is mostly driven by the
substitutability and complementarity relations among the commodities (Saadi 2011),
which, in turn, depends on the respective demand functions and on the underlying
preferences. However, the question is not whether the two commodities are comple-
ments or substitutes, but what degree they are first and what degree they are second.
When goods and information flow freely, shocks occurring in one market will evoke
responses in other markets. In cross-commodity price transmission studies are often
examined the question of how shocks in one of the commodity markets will get
transmitted to the rest if markets are integrated. As the price of one commodity
increases following an international price shock, demand may shift towards another
commodity resulting in higher prices on these as well. Thus, stabilization of prices of
essential agricultural commodities continues to remain an area of major concern for
policy makers. An understanding of inter-commodity price relationships and shock
transmissions forms the basis for public interventions (Alderman, 1993; Rashid, 2011;
Sharma and Kumur, 2001). Price instability affects both producers and consumers and
has macroeconomic implications as well. World widely, this was an important aspect
during the global food crisis from the early 2007 to the middle 2008, during which the
prices of a number of food commodities increased sharply. It was followed by a period
of collapsing prices in the second half of 2008.
There are only few studies on cross-commodity prices transmission. For example,
Asche et al. (2005) examined cross-commodity market integration between wild and
farmed salmon on the Japanese market and found that the species were close substitutes
on the market, and that the expansion of farmed salmon had resulted in price decreases
for all salmon species. Nielsen et al. (2007) found that markets for farmed trout are
related toothed fish markets in Germany, and that markets for these trout are more
closely linked to markets for captured fish than to farmed salmon. Timmer (2009)
addresses the long-run relationships among the prices of the three basic cereal staples,
rice, wheat and corn (maize) in Asian market. He found that wheat and corn are more
closely connected with each other than with rice markets. Rashid (2011) examined inter-
commodity price relationships to assess the relative importance of each of the three
major cereals (maize, wheat, teff) in Ethiopian cereal markets. He concluded that maize
is the most significant in exacerbating price variability with respect to the persistence of
shocks to itself and the two other cereals. He noticed that focusing on maize helps to
stabilize prices and also to reduce costs of stabilization. Sharma and Kumur (2001)
found complex long run relationships among rice, wheat, groundnut seed, mustard seed,
cottonseed, groundnut oil, vanaspati oil, mustard oil and other edible oils in Indian
20
market. They concluded that simply analyzing the price behavior of one commodity
while ignoring the behavior of the prices of substitutes will not be meaningful for price
stabilization purposes. The earlier researches includes for example Alderman (1993)
who investigated how information is transmitted across commodities (maize, millet,
sorghum) in Ghana. He noted imperfections in the way markets process information: the
lagged price of maize conveys information that is not contained in the past price of
sorghum or millet.
Although the studies specifically related to cross-commodity price movements are only
a few, after the biofuels issue has being raised by the recent food crisis, a considerable
number of papers have been dealing with the impact of crude oil on major agricultural
commodities.8
2.2 Meat market in the EU
Europe constitutes the largest single market in the world and at the same time qualifies
as one of the largest agricultural exporters in the world. Due to the growing liberaliza-
tion of world markets and the continuing European integration, agriculture in the EU is
undergone and is still undergoing constant restructuring in order to meet the demands of
global competition. The meat industry is very competed branch of industry and is
currently in its mature stage of development. Large multinational firms are prominent in
the meat market. The success of the largest firms is linked to their ability to achieve
economies of size and scope.
For decades, the meat sector has been one of the most important of the European
Union’s (EU) agriculture. Half of all the EU farms have livestock. The EU also provides
large support to livestock sector.
With approximately 150 million pigs and a yearly production of about 20 million
carcass weight the EU is the world’s second biggest producer of pig meat after China
and also the biggest exporter.9 The EU's main producer countries are Germany, Spain
and France. They represent together already half of the EU's total slaughter. Although
the pig herd in EU has been decreasing since 2006, the EU has a self-sufficiency of
about 110% and exports about 12% of its total production. Especially the Danish and
8 For example Arshad and Hameed (2009), Saghaian (2010), Baffes (2007), Muhammad and Kebede (2009), Bakhat and Wurzburg (2013), Ciaian and Kancs (2011) and Kristoufek, Janda and Zilberman (2012). 9 These rounded numbers has been valid during the last 5 years (2008-2013).
21
Dutch pork industries are export driven. Denmark, for example, is one of the world’s
largest pork exporters with over 75% of its output going to some 100 countries. Main
export destinations are Russia and East Asia, in particular China.
The EU is the world’s third largest producer of beef after the USA and Brazil. Also the
beef sector is important to the EU. With a herd of around 85 million heads and a total
yearly production of about 8 million tons of beef, the EU had a self-sufficiency of
approximately 100%. France, Germany, Italy and UK are the main producing member
states. These four member states represent together slightly more than 56% of the EU
total production. However, similarly to pork production, beef production in Europe has
declined, mainly due to disease crises and the consequent reduction in demand. In
Europe beef production is to a large extent a by-product of milk production with around
two-thirds of beef production coming from the dairy herd. Thus, the EU’s milk
production quota system affects also directly to beef production.
The EU limits meat imports with high tariffs and a complex set of quotas. In addition,
the EU has introduced sanitary barriers unrelated to the spread of disease among meat
animals. Strict regulations on slaughter and processing plants and a decision to ban
imports of meat from animals that received hormones in their feed have placed strong
restrictions on trade. The net effect of the meat barriers is to limit imports to special,
country-specific quotas, with small imports outside the quotas (Dyck and Nelson,
2003).
The integration of agricultural markets has always been very important for the member
states of the EU. Therefore, they merged the organization of the sector in the (Common
Agricultural Policy, CAP). Under the EU’s Common Agricultural Policy, the
agricultural markets are required to become spatially integrated within and between all
member states. In an integrated market, price information related to the production costs
should be efficiently transmitted between the member states. Specially, the pork market
is said to be homogenous because of its concentration on some dominating countries.
Several authors found a stable long-run equilibrium using cointegration tests on small
samples of countries, for instance Sanjuan and Gil (2001) and Serra et al. (2006).
Nevertheless, investigating a larger sample of 14 EU countries, Fousekis (2007) found
that the prices are not homogenous. The bovine market of the EU is more heterogeneous
due to the disturbances caused by policy measures (CAP) and a large variety of
production strategies (e.g. suckler cow husbandry, bull-mast or as a side product of milk
production). Because of a large number of exogenous shocks, such as policy changes,
the standard cointegration measures are problematic. For that reason Ihle et al. (2012)
used a methodology that is robust for break points and found cointegration between the
calf prices of four EU countries.
22
The developmental steps towards market efficiency in the EU pork and beef markets are
analyzed by Liu (2011). She evaluated the extent to which the Finnish domestic meat
market responds to changes in the European price (German and Danish prices).
According to Liu, both pork and beef prices in Finland are found to have slowly
cointegrated with German prices, but the cointegration relationship of the two counties
is only found to be symmetric for pork prices, while it is asymmetric for beef prices.
Producer price for pork in Finland is symmetrically cointegrated with the Danish price,
but the Finnish and Danish beef prices show a random walk. This implies that the price
transmission to the Finnish pork producer market from the EU market is smoother and
more efficient than for the beef market. However, the speed of transmission is still slow
compared to that between the Danish and German markets.
Meyer (2012) applied panel convergence tests with individual adjustment processes in
order to analyze the convergence of the EU livestock markets of pork and beef. The
estimation results reveal that the overall price heterogeneity is reducing within the EU
for both beef and pork. Meyer confirms that exogenous changes, such as the
enlargement and policy measures (European Food Monitoring Tool and Mid-Term
Review) improved the functioning of the internal markets. According to the study, larger
heterogeneity of the beef prices is the result of the still remaining differences of policy
measures within the member states of the EU. Focusing on the analysis of the EU
enlargement in 2004, the study found a stronger convergence of the prices of the new
member states compared to the old member states, which confirms that the accession
countries are catching-up. Nevertheless, the prices in the more segmented beef markets
of the accession countries were less strongly progressing towards homogeneity. A
detailed study of the EMU and non-EMU countries indicated that the dropping of the
currency risk has indeed had an influence on agricultural markets. The countries within
the Eurozone converge faster than the other countries, which significantly reduced the
welfare losses of consumers and producers. The study also reveals that at the beginning
of the ongoing euro crisis the prices within the EU only temporarily dispersed, but then
converged again.
2.3 Modelling price transmission in the panel framework
Let us consider agricultural prices observed over three different dimensions: space,
commodity and time. The generic price is pi,k,t where: i=1,…,j,…,N is the market place
(spatial dimension); k=1,…,h,…,K is the commodity; t=1,…,s,…,T is the period of
observation (time dimension). By more conventionally distinguishing between a cross-
sectional dimension, given by the combination of the dimensions ik, and a time
23
dimension t, we can identify any generic price observation as pik,t (scalar) and any
generic price series (vector) as pik . Notice that here we consider the logarithms of
prices. This monotonic transformation facilitates the economic interpretation of results,
in particular considering that regression coefficients may be interpreted as elasticities.
Consequently, henceforth pik identifies the time series of the price logarithm of the kth
commodity in the ith market place.
The behavior of pi,k,t over its three dimensions can be represented within suitable
structural models as the combination of market fundamentals such as supply, demand
and stock formation. However, such models are complex and hardly tractable in the
empirical analysis, whereas the investigation of price evolution and linkage is more
frequently afforded within reduced-form models. When all the three dimensions are
explicitly considered, reduced-form models are actually and by far more feasible and of
immediate use to generate price predictions, i.e. the estimation of E(pi,k,t pj,k,t-s, pi,h,t-s,
pi,k,t-s), given the available observations.
By distinguishing a cross-sectional dimension, ik, and a time dimension, t, a generic
reduced-form model of price formation and transmission over these two dimensions is
the following:
���,� = ��� + ∑ �����,����������� + ∑ ∑ � ��,��
���� ��
�������� ���,��� + ���,� (1.1)
where S is the maximum time lag and ���,� ~ �(0, ���,�� ). In a more compact matrix
form equation can be written as:
� = � + ∑ ���������� �� + �� (1.2)
where P, Ps and t are (T×(N×K)) matrices, s expresses the time lag, α is a (T×(N×K))
matrix of time invariant parameters, that is ���,� = ���,��� = ���, ∀�, �, � (any column
of α contains T elements with constant value αik) and ��~ �(�, ��). Ws is a ((N×K) ×
(N×K)) matrix of unknown parameters incorporating the correlation across prices
within both the time and cross-sectional (space-commodity) dimensions. The diagonal
elements, � ��,��� , indicate the auto-correlation over time, with the exclusion of the matrix
W0, where diagonal elements are evidently � ��,��� = 0 ∀ ik. The off-diagonal elements,
� ��,��� , represent the cross-sectional dependence of prices; in other words, they express
the interdependence among the different prices and, therefore, the degree and the
direction of transmission of the price shocks.
In particular:
24
- if h=k but i≠j, we are considering the price transmission for the same commodity
across space, that is, different market places. In this case, under perfect spatial
arbitrage, the validity of the Law of One Price (LOP) implies that � ��,�� = 1;
- if i=j but h≠k, we are considering the price transmission between two different
commodities in the same market. In this case, elements � ��,��indicate the degree
of substitutability between the different goods. � ��,�� will be close to 1 (-1) un-
der perfect substitutability (complementarity) between h and k, while it will be
close to 0 under low substitutability (complementarity).
As pik indicates the logarithms of prices, the elements of Ws express the price
transmission elasticities. Within the logarithmic form, the implicit assumption is that all
factors possibly contributing to price differentials but not explicitly taken into account
in the model (for example, transportation and transaction costs) are a constant
proportion of prices.
These constant multiplicative terms (that can be naturally intended as percentages)
apply to price pik,t-s, to obtain pjh,t and are captured by the elements of α.
If the matrix of unknown parameters, Ws, contains all the information about price
linkages over the three dimensions, we can expect that the transmission equations (1.1-
1.2) get rid of possible autocorrelation and heteroskedasticity across both the time and
cross-sectional dimensions; we can assume that spherical error terms are restored:
�� ~ �(�, ��) and E(t, t-s) = 0. The proper specification of (1.1-1.2) aims indeed at
restoring such conditions.10
10 More detailed discussion: Esposti and Listorti (2013)
25
3 ECONOMETRIC METHODOLOGY
In this section we represent econometric methods we have utilized: panel unit root tests,
panel co-integration tests and the panel data techniques to estimate the long-run and
short-run relationships between price series.
3.1 Panel unit root tests
The unit root test is an important tool in econometric analysis to test whether a time
series is stationary or not. It is known that univariate unit root tests are of low power
when the sample size is medium or small. To overcome this problem, testing the unit
root in the panel setting was developed and has been intensively studied in the last two
decades.
A variety of procedures for the analysis of unit roots in a panel context have been
developed. The logic behind the use of a panel unit root test is to combine the
information from time series with the information from cross-sectional units. Since the
power of unit root tests depend on the total variation in the data used (both in the
number of observations and their variation), panel unit root tests are more powerful than
standard time-series unit root tests. The variation across cross-section units improves
estimation efficiency, leading to smaller standard errors and, consequently, to higher t-
ratios and potentially more precise parameter estimates. With the increasing availability
of quite rich panel data sets in a number of contexts, these kinds of tests would seem
very attractive. One of the advantages of panel unit root tests is that their asymptotic
distribution is standard normal. This is in contrast to individual time series unit roots
which have non-standard asymptotic distribution.
However, a variety of issues arise when panel data are employed in testing for unit
roots. Early panel unit root tests generally ignore cross-sectional dependence that is
common for most of the macroeconomic series. Despite the shortcoming of early unit
root tests, they have helped researchers to develop new tests to deal with dependence
across cross-section units. These recent tests try to find out the same question of non-
stationarity with early ones but have different approaches to resolution and different
implications for empirical research. Therefore, in order to compare results predicated by
26
these tests and to discover what the impact of cross-sectional dependence assumption is
on unit root testing, we use more than one unit root tests. The first-generation tests we
conduct are Levin, Lin and Chu (2002), Im, Pesaran and Shin (2003), Breitung (2000),
Maddala and Wu (1999) and Choi (2001). After the first-generation unit root tests, we
conduct Pesaran’s (2003) second-generation unit root test.
We also employ unit root test based on the Lagrangian multiplier (LM) principle
developed by Im, Lee and Tislau (2005, 2010) which applies when a structural break
occurs at different time period in each time series as well as when the structural break
occurs in only some of the time series. The proposed test is not only robust to the
presence of structural breaks, but is also powerful in the basic case where no structural
breaks are involved.
3.1.1 First generation panel unit root tests (cross-country independence)
Levin, Lin and Chu (2002)
Levin, Lin and Chu (2002) (LLC ) suggest a more powerful panel unit root test than
performing individual unit root tests for each cross section. The null hypothesis of LLC
test is that each individual time series contains a unit root against the alternative that
each time series is stationary. The structure to be tested has the following form, similar
to an Augmented Dickey-Fuller (ADF) test but into a panel framework:
Δyit = ρiyi,t-1 + ∑ ������ iLΔyi,t-L + αmidmt + εit, m = 1,2,3 (3.1)
where y is the variable to be tested for unit root, Δ is the first difference operator and pi
is the lag order, which is allowed to vary across cross sections and is determined into the
test procedure. These terms are included to take into account heterogeneous serial
correlation across cross-sectional units; dmt can take three values depending on the
model specification: d1t ={empty set}, d2t ={1} including an individual constant and d3t
={1, t} including an individual constant and an individual linear trend. ρi, �iL and αmi are
parameters to be estimated. it is an error term that is assumed to be independently
distributed across i and t, i = 1,..., N, t = 1,..., T. Levin Lin and Chu tested the null
hypothesis: ρi = ρ = 0 for all i against the alternative hypothesis: ρi = ρ < 0 for all i. The
test based on the test statistic )ρ̂/se(ρ̂t*ρ (where ρ̂ is the OLS estimate of ρ in
equation (3.1), and )ρ̂se( is its standard error). Levin, Lin and Chu (2002) showed that
the estimator ��∗
is asymptotically normal distributed as N(0,1) .
27
Im, Pesaran and Shin (2003)
As stated before, LLC test is restrictive in the sense that it requires ρ to be homogeneous
across individuals. Im, Pesaran and Shin (2003) (IPS) permit a heterogeneous
coefficient on yi,t-1, proposing an alternative testing procedure that averages the
individual unit root test statistics. Im, Pesaran and Shin begin by specifying a separate
ADF regression for each cross section with individual effect and no time trend. The
estimated model is also the one given in equation (3.1). However, the null hypothesis is
that each series in the panel has unit root ρi = ρ = 0, and the alternative hypothesis states
that some individual series have unit roots while some are stationary. It can be
expressed as ρi < 0 for i = 1, 2,…, N and for i = N + 1,…,N. Thus, the null hypothesis of
this test is that all series are non-stationary process under the alternative that fraction of
the series in the panel are assumed to be stationary. Therefore, the null is rejected if
there is a subset (N1) of stationary individuals.
The first test (IPSLM) that they propose is the standardised group-mean Lagrange
Multiplier (LM) bar test statistic which allows different serial correlation patterns across
cross-section units:
,
11
11
iNi
iNi
LMLMVarN
LMENLMN
(3.2)
with iNi LMNLM
1
1 , where LMi denotes the individual LM test for testing i = 0
in the equation (3.1). E(LMi) and Var(LMi) are obtained by Monte Carlo simulation. The
second test (IPSt) is a standardised group-mean t bar test statistic, with an expression
similar to the equation (3.2) but replacing LM and LMi by t and it , respectively. The
IPS statistic is defined as the average of all the N individual ADF statistics:
�̅ =�
�� ����
��
��� (3.3)
where ��� denotes the individual pseudo t-ratio for testing i = 0 in (3.1), and E(ti) and
Var(ti) are obtained using Monte Carlo simulation. Im et al. show that when the lag
order is non zero for some cross-sections, and as N ∞, T ∞ and kTN / , the
limiting distribution of both test statistics is normal N(0,1).
Im et al. show that if we select a large enough lag order for the ADF regressions, the
small sample properties of IPS test outperform those from LLC test. However, they
found that both LLC and IPS tests present important size distortions when either N is
small or N is relatively large with respect to T.
28
Besides the popular LLC and IPS tests, we performed three more sophisticated first
generation panel unit root tests which try to correct some flaws that the former tests may
present. They are the Breitung (2000) test which shows a higher power than LLC or
IPS tests when they are compared in Monte Carlo experiments and the Maddala and
Wu (1999) and Choi (2001) Fisher type tests, which can be applied using ADF or
Phillips-Perron (PP) versions of the unit root tests for each cross section, and is also
found to be superior to the IPS test.
Breitung (2000)
Breitung (2000) consider the same basic ADF specification as Levin at al. (2002).
However, Breitung finds that both the LLC and IPS tests suffer from a substantial loss
of power if individual-specific trends are included. This is due to the bias correction that
also removes the mean under the sequence of local alternatives. He proposes a test
statistic whose power is substantially higher than that of LLC or IPS. Breitung considers
a panel unit root test which employs unbiased t-statistics. Breitung’s test assumes the
unit root process is common to all cross sections. By allowing for heterogeneous
deterministic trends and short-run dynamics across countries without the need of bias
adjustment, the Breitung test has more power to reject a false null and is not sensitive to
the degree of augmentation of the ADF specifications
The panel data yit is generated by the simple components model:
yit = µi + βit xit, (3.4)
where the unobserved error term xit follows
xit = i xit-1 + it. (3.5)
The null hypothesis is to test the presence of a unit root in all cross-sectional units, i =
1 for all i. For this, it is assumed that it ~iid(0,σ2 ) with E(εit4), the initial observations xi0
are independent and identically distributed (i.i.d.) across i with E(ε0t4 ) < ∞ and
independent of it for all t ≥ 1 and i.
This testing problem is invariant to the following linear transformation:
yit*=y
it+ μ
it * + β
i*t. To construct a test that is invariant to the transformation, Breitung
suggested the use of the transformed data:
(Δyit)* = st[Δyit – (1/T-t)(Δyi,t+1 + … + ΔyiT)], (3.6)
29
for t = 1,..,T – 1, where ��� = (T-t)/(T-t+1) and �����
∗ = yi,t-1 – yi0 – ((t-1)/T(yiT – yi0) for t =
2,…,T. The panel unit root test for the null hypothesis proposed by Breitung follows
statistic:
BnT = (
�
√��) ∑ ∑ �∆��,�� ��
∗�� ����
���� �∗
�,�� �
� ���
� �� ∑ ∑ ��∗�,�� ��
��� ����
����
(3.7)
or
BnT = (B2nT)-1/2B1nT (3.8)
where BnT is Breitung t-statistic and ��2 is a consistent estimator of σ2.
Maddala and Wu (1999) and Choi (2001)
An alternative approach to panel unit root tests uses Fisher’s (1932) results to derive
tests that combine the p-values from individual unit root tests. The ADF and PP Fisher
panel unit root tests proposed by Maddala and Wu (1999) and Choi (2001) combine the
p-values of the test statistics for a unit root in each cross-sectional unit. The p-values are
computed from the ADF test (Maddala and Wu, 1999) and the PP test (Choi, 2001). The
simplicity of this test, its robustness to the choice of lag length and sample size and the
advantage of allowing for as much heterogeneity across units as possible make its use
attractive.
Choi (2001) considers the following model using the properties of Fisher test:
yit = dit + xit (3.9)
The observed data (yit) comprises of two components, namely a non-stochastic
component (dit) and a stochastic component (xit):
dit = -2∑ ������ ��� (3.10)
and
xit = ixi,t-1 + µit (3.11)
where µit is integrated of order zero, I(0), and may be heteroskedastic. Thus, the null
hypothesis is i = 1 for all i which implies unit root nonstationarity. The alternative
hypothesis for finite N is i< 1 for at least one i for finite N.
30
GiTi is a one-sided unit root test statistic (ADF or PP) for ith unit in the model following
assumptions hold:
under the null, as Ti ⟶ ∞, GiTi ⟶ Gi, (Gi being a non-degenerate random
variable)
µit is independent of µjs for all t and s when i≠j
��
� ⟶k as N ⟶ ∞, (k being a fixed constant).
Let pi be the p-value of a unit root test for a cross-sectional unit i, such that pi = �(����)
where F(.) is the distribution function of Gi. The proposed inverse chi-squared Fisher-
type panel unit root test statistic has the form:
P = -2∑ (������)���� . (3.12)
It combines the p-values from unit root tests for each cross-sectional unit i to test for
unit root in the panel. Under the null hypothesis of unit root, P is distributed as χ2(2N)
as Ti ⟶ ∞ for all N.
In this study we employ Fisher-type tests based on augmented Dickey-Fuller (Fisher-
ADF) and Phillips-Perron (Fisher-PP) Chi-square unit root tests. Using Fisher-type tests
are advantageous since they do not require balanced panels. Other advantages are the
availability for completely heterogeneous specifications and possibility to use different
lag lengths in the individual ADF or PP regressions. Additionally, Maddala and Wu
(1999) compare Fisher-type, IPS (2003) and LLC (2002) panel unit root tests and
present Monte-Carlo simulations as evidence in favor of Fisher-type tests in case of
cross-sectional correlation among variables. They also point out that when a mixture of
stationary and non-stationary series in the group is included in the alternative
hypothesis, the Fisher-types are the best among others because they are more powerful
in distinguishing the null and the alternative hypotheses.
3.1.2 Second generation panel unit root tests (cross-country dependence)
These so-called first-generation panel unit root tests are examples where the cross-
sectional dimension is used to construct tests that have higher power than individual
unit root tests. However, all the first-generation tests rely on independence along the
cross-sectional dimension. The first-generation panel unit root tests focus on panels
where the idiosyncratic errors were cross-sectionally uncorrelated. It was soon realized
that cross-sectional independence is a highly unrealistic assumption for most settings
encountered in practice, and it has been shown that the first-generation tests exhibit
31
large size distortions and low power in the presence of cross-sectional dependence (e.g.
O’Connell, 1998; Banerjee et al., 2004; Strauss and Yigit, 2003). Therefore, so called
second-generation panel unit root tests have been constructed to take the cross-sectional
dependence into account. These second-generation tests assume specific forms of the
cross-sectional dependence as their application depends on modelling the structure of
the dependence. As argued by Quah (1994), the modelling of cross-sectional
dependencies is a difficult task since no natural ordering exists in unit observations.
This is why various tests have been proposed including the works of Bai and Ng (2002),
Phillips and Sul (2003), Moon and Perron (2004), Choi (2002), Ploberger and Phillips
(2002), Moon, Perron and Phillips (2005), Chang (2002) and Pesaran (2003). Two main
approaches can be distinguished. The first one relies on the factor structure approach
and includes the contributions of Bai and Ng (2001), Phillips and Sul (2003), Moon and
Perron (2004), Choi (2002) and Pesaran (2003). The second approach consists in
imposing few or none restrictions on the residuals covariance matrix. This approach has
been adopted by Chang (2002) among others, who proposed the use of instrumental
variables in order to solve the nuisance parameter problem due to cross-sectional
dependency.
Pesaran (2003)
Pesaran (2003) considered a one-factor model with heterogeneous loading factors for
residuals. However, instead of basing the unit root tests on deviations from the
estimated common factors, he augmented the standard Dickey-Fuller or Augmented
Dickey-Fuller regressions with the cross-section average of lagged levels and first-
differences of the individual series. If residuals are not serially correlated, the regression
used for the ith cross-sectional unit is defined as:
∆yi,t = αi + iyi,t-1 + ci����� + di∆��t + vi,t (3.13)
where ��t-1 = 1/N � ���,�����
��� and ∆��� = 1/N � �∆��,��
�
���. The Pesaran’s test is
based on these individual cross-sectionally augmented ADF statistics, denoted CADF. A
truncated version, denoted CADF*, is also considered to avoid undue influence of
extreme outcomes that may arise in small time T dimensions. In both cases, the idea is
to build a modified version of IPS t-bar test based on the average of individual CADF or
CADF* statistics (respectively denoted CIPS and CIPS*, for cross-sectionally
augmented IPS).
CIPS = �
�∑ ��
���� (�, �), (3.14)
CIPS* = �
�∑ ��
∗���� (�, �) (3.15)
32
where ti(N,T) denotes the t-statistic of the OLS estimate of i. The truncated CADF
statistic is defined as:
K1 if ti(N,T) ≤ K1
��∗(�, �) = ti(N,T) if K1 < ti(N,T) < K2 (3.16)
K2 if ti(N,T) ≥ K2
The constants K1 and K2 are fixed such that the probability that ti(N,T) belongs to
[K1,K2] is near to one. All the individual CADF (or CADF*) statistics have similar
asymptotic null distributions which don’t depend on the factor loadings. However, they
are correlated due to the dependence on the common factor. It is possible to build an
average of individual CADF statistics, although standard central limit theorems don’t
apply to these CIPS or CIPS* statistics. Pesaran showed that, the null asymptotic
distribution of the truncated version of the CIPS statistic exists and is free of nuisance
parameter. He proposed simulated critical values of CIPS and CIPS* for various sample
sizes. The critical values for CIPS for various deterministic terms are tabulated by
Pesaran (2007).
3.1.3 Panel unit root tests allowing structural breaks
The major weakness of the traditional unit root test is the failure to reject the unit root
hypothesis if the series has a structural break. This implies that series that are found to
be I(1) may in fact be stationary around the structural break; that is, I(0) but mistakenly
classified as I(1). Perron (1989, 1990) shows that failure to allow for break leads to a
bias that reduces the ability to reject a false unit root hypothesis. To overcome this
problem, Perron proposed allowing for a known or exogenous structural break in the
Augmented Dickey-Fuller (ADF) tests. Based on the short coming of this approach,
Zivot and Andrews (1992) and Perron (1997) propose determining the break point
endogenously. Lumsdaine and Papell (1997) extended the Zivot and Andrews (1992)
model to accommodate two structural breaks. However, Lee and Strazicich (2003)
criticized the endogenous break point tests for their treatment of breaks under the null
hypothesis. Lee and Strazicich (2003) propose a two break minimum Lagrange
Multiplier (LM) unit root test for structural breaks both under the null and the
alternative hypothesis that do not suffer from the spurious rejection of the null
hypothesis
Im, Lee and Tislau (2005, 2010)
33
The latest direction has been merged recently in the development of panel unit root tests
that allow for the presence of structural breaks and cross-sectional dependence. Im, Lee
and Tislau (2005) extended the univariate Lagrange Multiplier (LM) unit root test of
Lee and Strazicich’s (2003) to a panel LM test (ILT). Im et al. (2010) made use of a
simple transformation in order to obtain a Lagrange Multiplier (LM) panel unit test
statistic which is invariant to both the location and the size of breaks in the level or
trend of the series in the panel. This test depends only on the number of breaks in the
series and, therefore, has significantly greater power than all previous panel tests. In
addition, the test corrects for the presence of cross-correlations in the innovations of the
panel by applying the cross-sectionally augmented procedure of Pesaran (2007) that is
found to perform robustly under various specifications of cross-sectional dependence
(Baltagi et al., 2007).
The LM test of Lee and Strazicich (2004) which is obtained using the following
regression:
∆�� = ��∆�� + ������ + ∑ ��
���� ����� + ��, (3.17)
where Zt is the vector describing the breaks. Specifically, the case of one structural
change in the mean is formed as Zt = [1,t,Dt ]’, where Dt = 1 for t ≥ TB+1, and zero
otherwise. TB is the time period of the structural break. The LM t-test statistic is given
by the �̃, the t-statistic for the null hypothesis ϕ = 0. The location of the break is
determined endogenously by utilizing a grid search over all possible break points. Lee
and Strazicich (2004) provide simulated critical values of the minimum LM unit root
test.
Following Im et al. (2005), the test statistic (ILT) is based on the panel framework of the
equation (3.18) by implementing the testing regression on each cross-section unit:
Δ��� = ���Δ��� + �����,��� + ∑ ���
���� ���,��� + ��,�, (3.18)
where i is the cross-section unit, i = 1,2,…,N and t is the time period, t = 1,2,…,T. The
test statistic is based on the null hypothesis ϕi = 0 for all i, against the alternative
hypothesis ϕi < 0 for some i. The panel LM statistic can be constructed as the average of
univariate LM unit root t-test statistic estimated for each individual i:
��̅�� =
�
�∑ �̃��
���� (3.19)
The standardized panel LM unit root test statistic ILT is:
34
��� =
√�(�̅��� ��(����))
��(����), (3.20)
where �(�̃��) and �(�̃��) are the expected value and variance of the individual �̃��
statistic, respectively. Thus, as N and T → ∞ as long as �(�̃��) and �(�̃��) exist and
N/T → k, we have ���� → N(0,1) and the asymptotic distribution is not affected by the
presence of structural breaks.
The panel LM test of Im et al. (2005) will critically depend on the nuisance parameters
indicating the size and location of breaks when the series under investigation exhibits
breaks in both the intercept and the slope, and thus can be subject to serious size
distortions. To address this problem, Im, Lee and Tislau (2010) propose a new Lagrange
multiplier (ILT*) panel unit root test that is invariant to the nuisance parameters.
Following Lee and Strazicich (2009) the dependency of the test statistic on the nuisance
parameter can be removed by defining
���∗ = �
�
�����, for � < ��
�
�������, for �� < � < �
(3.21)
Using the transformed series, Im et al. (2010) formulate a test equation similarly to
equation (3.18) by replacing ���,��� with ���,���∗ . The transformed panel LM statistic can
be obtained as the standardized statistic of the following average test statistic:
��̅� = �
�∑ �̃��
∗���� (3.22)
Formally, the standardized panel LM unit root test statistic ILT* is obtained as:
����∗ = √�(�̅�����(�̅��))
���(�̅��), (3.23)
where ��(��̅�) and ��(��̅�) are the estimated values of the average of the means and
variances of �̅ as reported in Table 2 of Im et al. (2010). The standardized LM panel unit
root test follows a standard normal distribution.
The previous panel LM unit root tests assumed no correlations in the innovations across
the panel. To correct for the presence of cross-sectional dependence, Im et al. (2010)
apply the cross-sectionally augmented procedure of Pesaran (2007) that is found to also
be robust to the presence of other sources of cross-section dependence such as the
spatial form (Baltagi et al., 2007). Therefore, they formulate the transformed testing
35
regression augmented by the cross-section averages of lagged levels and first-
differences of the individual series:
��� = ������ + �����,���
∗ + ���̅��∗ + ℎΔ��̅
∗ + ∑ ���Δ��̅��∗�
��� + ∑ ������� Δ���,��� + ���,
(3.24)
with ��̅��∗ = ��� ∑ ��,���
∗���� and Δ��̅
∗ = ��� ∑ Δ���∗�
��� = ��̅∗ − ��̅��
∗ . Therefore, the t-
statistic on ϕi is used in order to construct the mean statistic �̅ as in equation (3.22),
which in turn can be used to construct the ILT*CA test statistic equivalently to equation
(3.23), which again follows a standard normal distribution.
3.2 Cointegration tests
Using unit root tests we can verify the stationarity of the series. However, empirical
questions often concern multivariate relationships; it becomes essential to find out if
variables are cointegrated. In the time series framework, cointegration refers to the idea
that if a set of variables is individually integrated of order one, it is possible that some
linear combinations of these variables are stationary. In this case, the vector of slope
coefficients is referred to as the cointegrating vector.
Compared to panel unit root tests, the analysis of cointegration in panels is still at an
early stage of development although during the recent years there have been seen quite
big steps forward. So far, the focus of the panel cointegration literature has been on
residual-based approaches, although there have been a number of attempts to develop
system approaches as well. The residual-based tests were developed to ward against the
spurious regression problem that can arise in panels when dealing with I(1) variables.
Kao (1999) and Pedroni (1999, 2004) were among the first to propose residual-based
tests for the null hypothesis of no cointegration in cross-sectionally independent panels.
However, it has since then become clear that these tests do not work in general, as
cross-section dependence is likely to be the rule rather than the exception. In fact, as
Gengenbach, Palm, and Urbain (2006) showed the presence of unattended cross-section
dependence in the form of non-stationary common factors can actually cause the test
statics to diverge as N and T grows. As a response to this, they propose to estimate
separately the common and idiosyncratic components of Xit and Yit using the principal
components method of Bai and Ng (2004), and then to test for cointegration in the
resulting component estimates. Banerjee and Carrion-i Silvestre (2006), Westerlund
(2005, 2007) and Westerlund and Edgerton (2007) propose a similar test but instead of
applying the principal components method to Xit and Yit directly, they apply it to the
36
residuals of a first-stage regression of Yit onto Xit. Cointegration requires that both the
common and idiosyncratic components of the residuals are stationary.
With confirmation on the integrated order of variables of interest, we look for a
relationship between pork and beef prices using firstly the panel cointegration technique
developed by Pedroni (1999, 2004). This technique is a significant improvement over
conventional cointegration tests applied on a single country series. As explained in
Pedroni (1999), conventional cointegration tests usually suffer from unacceptable low
power when applied on data series of restricted length. The panel cointegration
technique addresses this issue by allowing one to pool information regarding common
long-run relationships between a set of variables from individual members of a panel.
Further, with no requirement for exogeneity of the regressors, it allows the short-run
dynamics, the fixed effects, and the cointegrating vectors of the long-run relationship to
vary across the members of the panel. Furthermore, it provides appropriate critical
values even for more complex multivariate regressions. Pedroni (2004) considers a set
of residual-based test statistics for the null of no cointegration in the general case of
fully endogenous regressors, no pooled slope coefficients and varying dynamics. The
advantage of these tests is that they pool only the information regarding the possible
existence of the cointegrating relationship that comes from the statistical properties of
the estimated residuals.
Pedroni (2004)
The implementation of Pedroni’s cointegration test requires estimating first the
following long run relationship:
yit = αi + δit + β1ix1it + … + βMixMit + εit, i = 1,…,N, t = 1,…,T and m=1,…,M
(3.25)
where N is the number of cross-sectional members in the panel; T is the number of
observation over time and M refers to the number of exogenous variables. The variables
yit and xit are assumed to be I(1), for each member i of the panel, and under the null of
no cointegration the residual εit will also be I(1). αi and δi are scalars denoting fixed
effects and unit-specific linear trend parameters, respectively and βi are the cointegra-
tion slopes. Both the slope coefficients β1i,…,βMi, and the member specific intercept αi
can vary across each cross-section.
For the computation of the panel test statistics we take the first-difference of the original
series and estimate the residuals of the following regression:
∆yit = β1i∆x1it + … + βMi∆xMit + πit (3.26)
37
Using the residuals from the differenced regression with a Newey-West (1987)
estimator the long run variance of ���� is calculated. It is symbolized as ������ :
������ =
�
�∑ ���,�
����� +
�
�∑ �1 −
�
�����
����� ∑ ���,�
������ ���,���. (3.27)
Pedroni considers the use of seven residual-based panel cointegration statistics for this
test. They are panel v -statistic, panel PP t -statistic, panel PP -statistic, Panel ADF t-
statistic, group PP t -statistic, group PP -statistic and group ADF t -statistic. The first
four statistics known as panel cointegration statistics are based on the within approach.
The last three statistics are group panel cointegration statistics and are based on the
between approach. The four within–dimension statistics are based on pooling the
autoregressive coefficients across the different cross-sectional units for the unit root
tests on the estimated residuals, whereas the three between-dimension statistics are
based on estimators that simply average the individual estimated coefficients for each
unit. Between-dimension-based statistics are therefore just the group mean approach
extensions of the within-dimension-based ones. In the presence of cointegrating
relationship, the residuals are expected to be stationary. The panel v- test is a one sided
test with the null of no cointegration being rejected when the test has a large positive
value. The other statistics reject the null hypothesis of no cointegration when they have
large negative.
Another distinction between the two sets of test is based on the alternative hypothesis
specification. Both sets of test verify the null hypothesis of no cointegration: ρi = 1 for
all i, where ρi is the autoregressive coefficient of estimated residuals under the
alternative hypothesis (�̂it = �� �̂it-1 + uit).
Alternative hypothesis specification is, however, different. The panel cointegration
statistics impose a common coefficient under the alternative hypothesis: ρi = ρ <1 for
all i. The group mean cointegration statistics allow for heterogeneous coefficients under
the alternative hypothesis and it results: ρi <1 for all i.
It is evident that the tests based on the between dimension are more general allowing for
cross-section heterogeneity.
Defining �̂it the estimated residuals from (3.25), the seven Pedroni’s statistics are:
����� =�
∑ ∑ ������ � �̂�,�� �
�����
����
(3.28)
38
����� =∑ ∑ �����
� ����� ��̂�,�� �∆�̂��������
���
∑ ∑ ������ � �̂�,�� �
�����
����
(3.29)
���� =∑ ∑ �����
� ����� ��̂�,�� �∆�̂��������
���
������ ∑ ∑ �����
� � �̂�,�� ���
�������
(3.30)
����∗ =
∑ ∑ ������ ��
��� ��̂�,�� �∆�̂�������
��̃��∗� ∑ ∑ �����
� � �̂�,�� ���
�������
(3.31)
�̃������ = ∑∑ ��̂��� �∆�̂��������
���
�∑ �̂�,�� ���
��� �
���� (3.32)
�̃��� = ∑∑ ��̂��� �∆�̂��������
���
�����∑ �̂�,�� �
�����
���� (3.33)
����∗ = ∑
∑ �̂�,�� �∗ ∆�̂��
∗����
�∑ �̃�∗��̂�,�� �
∗�����
���� (3.34)
Kao (1999)
Kao (1999) was the first author to suggest the test for cointegration in homogeneous
panels. The Kao test statistics are calculated by pooling all the residuals of all cross-
sections in the panel; it is assumed that all the cointegrating vectors in every cross-
section are identical.
Kao considered the following system of cointegrated regressions in the homogeneous
panels:
xit = xi,t−1 + it (3.35)
yit = yi,t−1 + vit (3.36)
Let’s consider the regression:
yit = αi + βxit + uit, i = 1,…,N and t = 1,…,T (3.36)
where αi are individual constant terms, β is the slope parameter, it and vit are stationary
disturbance terms and so yit and xit are integrated processes of order 1 for all i.
The zero mean vector ξit = (vit, it)’ satisfies
�
√�∑ ���
|��|��� ⇒ ��(Ω) (3.37)
39
for all i as T ⟶ ∞, where Bi(Ω) is a vector of Brownian motion with asymptotic
covariance Ω. Kao derives two types of panel cointegration tests based on residuals
from panel least-squares dummy variable (LSDV) estimation.
The first is of a Dickey-Fuller (DF) type, which can be applied to the residuals using:
���� = ���,��� + eit (3.38)
The OLS estimate of is:
�� =∑ ∑ �������,�� �
����
����
∑ ∑ ���,�� ���
�������
(3.39)
The null hypothesis that = 1 is tested by:
√��(�� − 1) =
�
√��
�
∑�
�∑ �������,�� �
����
����
∑�
�� ∑ ���,�� ���
�������
(3.40)
Totally, Kao (1999) proposed four DF-type panel cointegration tests based on the OLS
residuals from the homogeneous panel regression. The first two DF statistics are based
on assuming strict exogeneity of the regressors with respect to the errors in the equation,
while the remaining two DF statistics allow for endogeneity of the regressors.
Kao also uses ADF-type panel cointegration test for cointegation in heterogeneous
panel. In this test version, the parameters are allowed to differ across the cross-sections.
Kao’s (ADF) type test can be calculated from:
���� = ����,��� + ∑ ��
���� ���,��� + ���� (3.41)
where p is chosen so that the residuals eitp are serially uncorrelated. The ADF test
statistic is the usual t-statistic with = 1 in the ADF equation (Kao, 1999). The null
hypotheses is = 1 and the alternative hypotheses < 1. The asymptotic distributions
of the test converge to a standard normal distribution N(0,1) as T → ∞ and N → ∞.
Residual based (“first generation”) panel cointegration tests are often based on the
assumption of independent panel units. Breitung and Pesaran (2005) note that time
series are contemporaneously correlated in many macroeconomic applications using
country or regional data. Cross-sectional dependence can arise due to a variety of
factors, such as omitted observed common factors, spatial spillover effects, unobserved
common factors, or general residual interdependence, all of which could remain even
40
when all observed and unobserved common effects have been taken into account. The
literature on how to model cross-sectional dependence in large panels is still
developing.
Residual based panel cointegration tests result in low power for small samples. In the
presence of cross section dependencies, the tests are subject to large size distortions.
The situation gets worse if the number of cross sections is increased. To overcome these
deficits, panel cointegration tests have been developed that control for the dependencies
via a common factor structure (Banerjee and Carrion-i-Silvestre, 2006). The existence
of panel cointegration can be taken into account by the second generation panel
cointegration tests among of which we employ Westerlund (2007) and Westerlund and
Edgerton (2007), which seems to be more powerful when compared with the residual-
based panel cointegration tests.
Westerlund (2007) and Westerlund and Edgerton (2007)
We performed panel cointegration tests by Westerlund (2007), which are based on
structural rather than residual dynamics. Tests are based on error correction model
(ECM). These structural kind of test does not impose any common factor restriction,
which is a main reason associated to loss of power for residual-based cointegration tests.
The tests are based on the estimation of the following error correction equation:
Δyit = δ’idt +α(yi,t-1 – �’xi,t-1) +∑ ������ ijΔyi,t-j +∑ ���
����� ∆xi,t-j + εit, i = 1,…,N and t =
1,…,T (3.42)
where yit is the dependent variable, xit is a vector of independent variables, dt =(1,t)’ is
the set of deterministic components and Δ is the first difference operator. Equation can
be rewritten as:
Δyit = δ’idt +αyi,t-1 + ���xi,t-1 +∑ �
����� ijΔyi,t-j +∑ ���
����� ∆xi,t-j + εit, (3.43)
where λi = -αi���. The parameter αi determines the speed at which the system Yi,t-1 –
���Xi,t-1 corrects back to the equilibrium after a shock.
Westerlund (2007) states that if αi < 0, then there is error correction, which implies that
yit and xit are cointegrated, whereas if αi = 0, there is no error correction and no
cointegration. It is indicated that the tests have limiting normal distributions and that
they are consistent.
41
Westerlund proposes 4 panel cointegration tests. Pα and Pτ are panel statistics which are
based on pooling the information regarding the error correction along the cross sectional
units. The null hypotheses for the panel tests is αi = 0 and alternative hypotheses is αi =
α < 0 for all i. Gα and Gτ are group statistics which do no exploit the information
regarding the error correction. The null hypotheses for the group tests is αi = 0 and
alternative hypotheses is αi < 0 for at least some i. The test proposed by Westerlund
(2007) does not only allow for various forms of heterogeneity, but also provides p-
values which are robust against cross-sectional dependencies via bootstrapping.
The test developed by Westerlund and Edgerton (2007) relies on the popular Lagrange
multiplier test of McCoskey and Kao (1998), and permits heteroskedastic and serially
correlated errors, unit specific time trends and cross-sectional dependence. It also allows
an unknown structural break in both the intercept and slope of the cointegrated
regression, which may be located different dates for different units. In addition, this
bootstrap test is based on the sieve-sampling scheme, and has the advantage of
significantly reducing the distortions of the asymptotic test. Hence, the test also works
well in the small sample analysis. Another appealing advantage is that here the null
hypothesis is now cointegration which implies, if not rejected, the existence of a long-
run relationship for all panel members. The alternative hypothesis is that there is no
cointegrating relationship for at least one country of the panel.
Johansen’s Fisher panel cointegration test (Maddala and Wu, 1999)
The last cointegration test we perform is Johansen-Fisher based panel cointegration test
suggested by Maddala and Wu (1999). Johansen’s methodology takes its starting point
in the vector autoregression (VAR) of order p given by
�� = � + ������ + ⋯ + ������ + ��, (3.44)
where �� is an n×1 vector of variables that are integrated of order one, I(1), and �� is an
n×1 vector of innovations. This VAR can be re-written as
∆�� = � + ����� + ∑ �������� Δ���� + �� (3.45)
where
� = � ��
�
���
− � and �� = − � ��
�
�����
If the coefficient matrix Π has reduced rank r<n, then there exist n×r matrices α and β
each rank r such that Π = αβ′ and β′yt is stationary. r is the number of cointegrating
42
relationships, the elements of α are known as the adjustment parameters in the vector
error correction model and each column of β is a cointegrating vector. It can be shown
that for a given r, the maximum likelihood estimator of β defines the combination of yt-1
that yields the r largest canonical correlations of �� with ���� after correcting for
lagged differences and deterministic variables when present. Johansen proposes two
different likelihood ratio tests of the significance of these canonical correlations and
thereby the reduced rank of the Π matrix: the trace test and maximum eigenvalue test,
shown in equations (3.46) and (3.47) respectively.
������ = −� ∑ ln (1 − ��������� ) (3.46)
���� = −�ln(1 − �����) (3.47)
Here T is the sample size and ��� is the ith largest canonical correlation. The trace test
tests the null hypothesis of r cointegrating vectors against the alternative hypothesis of n
cointegrating vectors. The maximum eigenvalue test, on the other hand, tests the null
hypothesis of r cointegrating vectors against the alternative hypothesis of r+1
cointegrating vectors.
Using Johansen’s (1988) test for cointegration, Maddala and Wu (1999) consider
Fisher’s (1932) suggestion to combine individuals tests, to propose an alternative to the
two previous tests, for testing for cointegration in the full panel by combining individual
cross-sections tests for cointegration.
If πi is the p-value from an individual cointegration test for cross-section i, then under
the null hypothesis for the whole panel,
−2 ∑ log�(��
���� ) (3.48)
is distributed as ���� .
43
3.3 Estimation methods
It is well know that the panel cointegration relationship provides evidence in favor of
the existence of causal relationships but it does not indicate the direction of the causal
linkages between variables. Recently, econometricians have begun to modify Granger-
causality11 tests to incorporate panel dynamics (see for example Arellano and Bond
(1991), Holtz-Eakin et al. (1988), Hurlin (2005) and Hurlin and Venet (2001)). Within
panel frameworks, Granger-causality tests generate meaningful results with significant-
ly shorter time, incorporate significantly more observations, and produce more efficient
results than Granger-causality tests in the conventional context (Hurlin and Venet,
2001).
The recent literature highlights four approaches that could be implemented to test for
Granger-causality in panel models. The first one, based on the random or fixed effects
or on the Generalized Method of Moments (GMM), is usually employed to estimate
homogenous panel model by eliminating the fixed effect. The main disadvantage of this
method is that it assumes that slope parameters are similar which may lead to
inconsistent and misleading long-term coefficients (Pesaran et al., 1999). Furthermore,
this method does not allow for heterogeneity and cross-sectional dependence among
individuals. The second approach to test Granger causality, initiated by Hurlin (2007)
takes into account the heterogeneity across units but pay no attention to the
possible cross-sectional ependence (i.e., existence of common shocks). The third
approach proposed by Konya (2006), deals with both heterogeneity and cross-sectional
dependence across units and allows for testing Granger-causality on each individual
panel member separately by taking into account simultaneously contemporaneous
correlation across units. The forth approach, launched by Pesaran and Smith (1995)
proposes the Mean Group estimator (MG) for testing Granger causality. Constructed
separately for each group, the model computes a simple arithmetic average of the
coefficients consenting to the intercepts, short-run coefficients and error variances to
differ across groups. According to Pesaran and Smith (1995), this estimator provides
consistent estimates of the parameters’ averages. Pirotte (1999) also shows that the
mean group estimator provides efficient long-run estimators for a large sample size. It
allows the parameters to be freely independent across groups and does not consider
potential homogeneity between groups. Finally, the forth approach developed by
Pesaran, Shin and Smith (1999), focuses as well on non-stationary dynamic panels with
11 The term of Granger-causality, proposed by Granger (1969), is not a true causality concept but a statistical tool which in principle concerns only the predictability between time-series variables. It could be understood as following: X is said to “Granger cause” Y if and only if Y is better predicted by using the past values of X than by not doing so. Although not a real causality identification, Granger-causality analysis is widely applied in business forecast and policy-modeling.
44
heterogeneous parameters. It offers an intermediate estimator, the Pooled Mean Group
estimator (PMG) which allows the intercept, short-run coefficients and error variances
to be different across the groups but constraints the long-run coefficients to be equal
across these groups.
We are interested to detect short-run and long-run causal relationships between pork and
beef prices among the EU member states by implementing the MG and the PMG
estimators.
Mean Group and Pooled Mean Group estimators
The original Pesaran, Shin and Smith (1999) paper starts with an autoregressive
distributed lag (ARDL) dynamic panel specification, with p being the number of lags
for the dependent variable, and q the number of lags for the explanatory variables. The
specification takes the form:
yit = ∑ �����,��� + ∑ ���
����� ��,��� + �� + ���
���� , I = 1,2,…,N and t = 1,2,…,T
(3.49)
where λij are coefficients of the lagged dependent variable, Xit is a set of regressors and
δij is a (k×1) vector of its coefficients. Group-specific effects are represented by μi while
it is the error term. This model specification requires that T is large enough, which
means that there is a reason to expect that some variables might not be stationary. In
case the variables are integrated of order one, and consequently cointegrated, then we
expect that the error term is stationary for all i. Typically, cointegrated variables react to
deviations from their long-run equilibrium, and adjust in the short-run. These sorts of
deviations and reactions are usually presented as error-correction models by reparame-
terizing (3.49):
Δyit = ��(��,��� + θ�
����) + ∑ ���∗ ∆��,��� + ∑ ���
∗����� ∆��,��� + �� + ���
���� (3.50)
where
�� = − �1 − � ���
�
���
� , θ� = ∑ ���
����
1 − ∑ ����, ���
∗ = − � ���,
�
�����
� = 1,2, … , � − 1 and ���∗ = − � ���
�
�����
45
The error coefficient, or the speed of adjustment term, is presented here as ϕi. In case ϕi
is statistically significant, there is evidence of a long-run relationship, while a negative
ϕi implies that the variables return to the equilibrium after a deviation in the short-run.
��� is the vector of long-run coefficients, while ���
∗ and ���∗ are short-run coefficients of
the lagged dependent and explanatory variables, respectively.
Pesaran et al. (1999) recommend using maximum likelihood (ML) estimation method
for estimating equation (3.50), as it is nonlinear in its parameters. However, prior to ML
estimation, we must make a few assumptions about this specification. Firstly, equation
(3.50) is rewritten by stacking the time-series observations for each group:
Δyi = ��(��,��� − ����) + ∑ ���∗ ∆��,��� + ∑ ���
∗������� ∆��,��� + ��� + ��
������ (3.51)
where � = (1,...,1)’ is a (T×1) vector of ones, and the disturbance term is i = (i1,…,iT)’.
The error terms are distributed independently across groups, time, and of the regressors.
The last assumption is necessary for a consistent estimate of short-run coefficients, as
we allow them to differ across groups. To control for the long-run relationship, and for
the adjustment to the long-run equilibrium, the speed of error-correcting adjustment
term, ϕi, is negative. This is ensured when the model given by equation (3.50) is stable in its roots that lie outside the unit circle, or that ∑ ������
��� = 1. Now there exists a
long-run relationship between the dependent variable and regressors, with �i the long-
run coefficients and it a stationary process:
��� = −����� + ���. (3.52)
Once we confirm there is a long-run relationship, we can rewrite its coefficients without
the group-specific term: �� = �.
We rearrange equation (3.51):
��� = ����(�) + ���� + ��, (3.53)
where ��(�) = ��,��� − ���, �� = �Δ��,���, … , Δ��,�����, Δ��, Δ��,���, … , Δ��,�����, ��
and �� = ���,�∗ , … , ��,���
∗ , ��,�∗�
, ��,�∗�
, … , ��,���∗�
, ����.
There are no restrictions on the short-run coefficients. The error variances are allowed
to differ across groups, var(it) = ���. The nonlinearity of the parameters � and ϕi
implies that a likelihood approach should be used in panel estimation. Just for these
purposes, it is assumed that the error term is normally distributed. It is possible to
express the likelihood of the panel as the product of each, separate, group-specific
likelihoods:
46
��(�) = �
2� ln(2���
�)
�
���
−1
2�
1
���
�
���
[Δ�� − ����(�)]���[Δ�� − ����(�)] (3.54)
where
�� = �� − ��(��
���)����
�, � = (��, ��, ��)�, � = (��, ��, … , ��)� and �
= (���, ��
�, … , ���)�
In order to get consistent and asymptotically normal estimators, it is necessary to add
some further assumptions that can be found in the original paper (Pesaran, 1999, p. 624-
625).
Maximizing equation (3.54) with respect to φ gives us estimates of the long-run
coefficients �� and of the error-correction coefficients ���:
�� = − �����
�
����
�
���
��������
��
× �����
����
�
���
������Δ�� − �����,�����,
��� = �����������
�����
����� (3.55)
where ���� = ������ = ��,��� − ����. Estimators in (3.55) are called Pooled Mean Group
(PMG) estimators because they reflect both pooling, inherent in ��, and averaging across
groups, inherent in ���. Solving the maximization problem provides the error variance
estimate as well:
���� = ����Δ�� − �������
����Δ�� − �������. (3.56)
The PMG estimators are computed using the algorithm of back-substitution or in other
words, by iterating the obtained estimates. Using the initial estimate ��, one can get
�� and ���� from equations (3.55) and (3.56), respectively. These estimates can then be
replaced into equation (3.54) to get a new estimate of θ, used to get new estimates of φ
and ���, repeated until convergence is achieved. It should be noted that the ML estimates
of long-run and short-run parameters are asymptotically distributed independently of
each other, implying that once we get the long-run parameters, we can use them to
consistently estimate the short-run and the error-correction coefficients.
Indeed, the PMG estimator allows to evaluate two different Granger-causality
relationships: a short-run causality, testing the significance of the coefficients related to
the lagged difference of variables (H0: δi = ��� = 0 for all i) and a long-run causality
47
related to the coefficient of ECT term (H0: ϕ = 0 for all i). The parameter ϕ represents
the speed of adjustment at which the values of variables come back to the long-run
equilibrium levels, once they deviate from the long-run equilibrium relationship. The
negative sign of the estimated speed of adjustment coefficients are in accordance with
the convergence toward the long-run equilibrium. The larger the value of ϕi, the stronger
is the response of the variable to the deviation from the long run equilibrium. On the
flip-side, in the case of low coefficients values, any deviation from long-run equilibrium
requires much longer time to force the variables back to the long-run equilibrium.
Although the parameters obtained by iteration are identical to those obtained from full-
information maximum likelihood, the covariance matrix is not. Nevertheless, we can
recover the covariance matrix for all estimated parameters, because we know the
distribution of the PMG parameters. The covariance matrix can be estimated by the
inverse of:
⎣⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎡�
������
���
����
�
���
−�����
� ���
���� … −
������ ���
���� −
������ ��
���� … −
������ ��
����
���� ���
���� … 0
���� ��
���� … 0
⋱ ⋮ ⋮ ⋱ ⋮���
� ���
���� 0 …
���� ��
����
�����
���� … 0
⋱ ⋮��
� ��
���� ⎦
⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎤
With the mean group (MG) estimator proposed by Pesaran and Smith (1995), the
intercepts, slope coefficients, and error variances are all allowed to differ across groups.
MG estimator runs ARDL equations separately and calculates the mean of the short and
long-run parameters across groups by the simple arithmetic average of the unit specific
coefficients. The MG estimator assumes that all of the parameters can differ across units
but does not allow for long-run homogeneity. Though MG estimator can provide
consistent estimates, it depends on quite strong assumptions and does not take into
account that certain parameters may be the same across groups. The MG parameters are
then simply unweighted means of individual coefficients, or:
����= N-1∑ ������� (3.57)
48
with the variance
���� ������ = �
�(���)∑ (���
���� − ��)� (3.58)
It is possible to test for the suitability of the PMG estimator vs. the MG estimator based
on the consistency and efficiency properties of the two estimators, using a likelihood
ration test or a Hausman test. If the restrictions imposed on the long-run coefficients are
valid then the PMG is more efficient than the MG estimator, while it becomes
inconsistent if the restrictions do not apply. Rejection of the test would suggest that the
sample is too heterogeneous to be pooled.
It is worth noting that obtaining consistent and efficient PMG estimator requires several
conditions to be satisfied. First, the time dimension has to be long enough for the
estimation of the model for each cross-sections separately. Second, the lag order must
be chosen to ensure that the residuals of the error correction model are serially
uncorrelated, but it should not cause loss of degrees of freedom as well. Third, PMG
assumes cross-sectional independence of the regression residuals εit. Arising from
omitted common effects (e.g. time-specific effects or common shocks affecting
countries), cross-sectional dependence influences ARDL process and causes misspecifi-
cation. Pesaran et al. (1999) offer either to use cross-sectional means of the existent
regressors as additional regressors or include all of the variables as deviations from their
respective cross-sectional means in each period. The fourth condition is the existence of
long-run relationship between the variables and it requires a negative and significant
error correction term (φi). Finally, PMG estimator is both consistent and efficient if and
only if the long-run parameters are homogenous across countries.
Dynamic OLS (DOLS) and Fully Modified Ordinary Least Squares (FMOLS)
To check the robustness of our results, we also apply another two heterogeneous
dynamic panel estimators, the Dynamic OLS (DOLS) and the Fully Modified Ordinary
Least Squares12 (FMOLS). The methods are asymptotically equivalent (Banerjee, 1999).
The PMG is based on the maximum likelihood estimation, while the DOLS and the
FMOLS, as their names suggest, are modified OLS estimators. The DOLS and the
FMOLS are popular in conventional time series econometrics, for they are believed to
eliminate endogeneity in the regressors and serial correlation in the errors.
12 The FMOLS is popular in conventional time series econometrics, for it is believed to eliminate endogeneity in the regressors and serial correlation in the errors.
49
The presence of cointegration and unit roots considerably affects the asymptotic
distributions in time series as well as in panel analysis. However, cointegration
equations have attractive properties: as the number of observations increase in T and N,
the OLS estimation of the cointegrated variables converges in the long-run equilibrium
to the true value. FMOLS and DOLS methodologies are proposed by Kao and Chiang
(2000) and Pedroni (2000) to estimate the long-run cointegration vector for non-
stationary panels. They found that while the OLS estimator is normal distributed with
non-zero mean, the fully modified FMOLS and dynamic DOLS estimators are
asymptotically normal with zero mean. The basic idea behind both the estimators is to
correct the standard pooled OLS for endogeneity bias and serial correlation that are
normally present in long-run relationship and thereby allow for standard normal
inference.
Let us consider the following fixed effect panel regression:
��� = �� + ���� + ���, � = 1, … , �; � = 1, … , �, (3.59)
where yit is a matrix (1,1), β is a vector of slopes (k,1) dimension, α is individual fixed
effect, uit are the stationary disturbance terms. It is assumed that xit (k,1) vector are
integrated processes of order one for all i, where:
��� = ��,��� + ��� (3.60)
A standard panel OLS estimator for the coefficient in equation is given by:
���,��� = �� �(��� − �̅�)�
�
���
�
���
�
��
� �(��� − �̅�)
�
���
�
���
(��� − ���) (3.61)
Where �̅� and ��� refer to the individual means for each i member of the cross section.
According to Pedroni (2000) this estimator is asymptotically biased and its distribution
is dependent on nuisance parameters associated with the dynamics underlying the
processes determining x and y. Only if x is strictly exogenous and the dynamics are
homogeneous across i members of the panel ���,��� is unbiased.
Pedroni (2000) suggested the group-mean FMOLS and DOLS estimators. According to
Pedroni group mean tests are preferred over the pooled tests since they allow greater
flexibility under alternative hypotheses. In our analysis we utilize group-mean versions
of FMOLS and DOLS by Perdoni (2000).
50
The test statistics derived from the group-mean estimators are constructed to test the
null hypothesis βi = β0 for all i against the alternative hypothesis βi ≠ β0, so that the
values for βi are not constrained to be the same under the alternative hypothesis.
Let’s define ��� = (���, ���) ~ �(1) and �� = (��� + ���) ~ �(0) with long run
covariance matrix Ωi = ����� (Li is a lower triangular decomposition of Ωi). In this case,
the variables are said to be cointegrated for each member of the panel, with cointegrat-
ing vector β. The terms αi allow the cointegrating relationship to include member
specific fixed effect. The covariance matrix can also be decomposed as
�� = ��
� + �� + ���, (3.62)
where ��� is the contemporaneous covariance and is a weighted sum of autocovariances.
The Panel FMOLS estimator which is parametric approach for the coefficient β is
defined as follows:
���,����� = ��� ∑ (∑ (��� − �̅�)��
��� )������ (∑ (��� − �̅�)
���� ���
∗ − ��̂�) (3.63)
where
���∗ = (��� − ���) −
�����
�����∆���, �̂� = Γ���� + Ω����
� − �����
������Γ���� − Ω����
� � (3.64)
and Li is a lower triangular decomposition of Ωi defined as follows:
� = ����� ���
�
���� ����� (3.65)
Pedroni (2001) also considers a group-mean DOLS estimator which is non-parametric
approach. This is done by modifying equation (3.64) and by including lead and lag
dynamics:
��� = �� + ����� + ∑ ����������
��,��� + ��� (3.66)
and the estimated coefficient β is given by:
���,���� = [��� ∑ (∑ ��������
��� )��(∑ ������� ���
∗ )���� ]� (3.67)
where ��� = ���� − �̅�, Δ��,���, … , Δ��,���� is the 2(K+1)×1 vector of regressors.
51
4 DATA DESCRIPTION
In order to analyze the price integration of pork and beef in the EU, we use the data on
monthly pork and beef prices in the EU member states, which are obtained from Tike
(Agricultural Statistics Finland)13. Tike is responsible for agricultural price monitoring
in Finland and price monitoring reports include data on both Finland and other EU
countries.
In our dataset, the EU member countries are categorized into three groups of EU1514,
EU2515 and EU2716. Data on EU15 covers the period from February 1995 to June 2014.
Correspondingly, the data on EU25 covers the period from January 2005 to June 2014
and the data on EU27 covers the period from January 2007 to June 2014. In other words
this means that the single price time series from EU15 includes 221 observations; the
single price time series from EU25 consists of 102 observations and the single price
time-series from EU27 includes 78 observations. In empirical analysis we carry out our
investigations by utilizing six different data specifications: we have one group of EU
member states (EU15 countries) when investigating period 2/1995-6/2014, we have two
groups of EU member states (EU15 and EU25) when investigating period 1/2005-
6/2014 and we have three groups of EU member states (EU15, EU25 and EU27) when
investigating period 1/2007-6/2014. By repeating the estimations with different group of
member states we can make conclusions regarding stability of the results over time and
also regarding cross-sectional stability.
We have constituted our monthly price data by taking averages from weekly data.
Weekly data includes many missing values which are problematic in our investigation.
We have estimated and filled some missing values with the help of data authors. As a
result, we have succeeded to create balanced monthly level panel data. However, one
should notice that Malta is missing in the panels because its time-series includes serious
13 I am grateful to Pia Outa and Lauri Juntti for sending me the excellent price data. 14 EU15 member states are Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden and United Kingdom. 15 EU25 member states are EU15 plus Cyprus, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Malta, Poland, Slovak Republic and Slovenia. 16 EU27 member states are EU25 plus Bulgaria and Romania.
52
shortages17. Regarding other countries, quality of data is good and time-series are
completed.
Carcass classification plays an important role in the European Union meat market. The
aim of the carcass classification scheme is to ensure a common classification standard
throughout the European Union which enables the EU to operate a standardized price
reporting system. The EU classification of the analyzed pork is labeled “E”, which
indicates that 55% or more of the carcass has to be lean meat. The quality of the beef is
correspondingly “R3”. According to the EU grading scheme, it is qualitatively good
meat, which means that the overall profiles are straight, the muscle development is good
and the content of fat is medium. These two are also the most typical carcass classes.
The descriptive statistics of pork and beef prices are reported for each member state in
Tables 4.1 and 4.2. Notice that in empirical analysis we use the natural logarithms of
these variables in order to obtain the elasticities.
17 Despite the fact that Malta is missing in our data sets, we use terms “EU25” and “EU27” in our empirical investigations.
53
Table 4.1. Descriptive statistics of pork prices in the EU member states
Mean Std. Min Max Mean Std. Min Max Mean Std. Min Max
Austria 148.5 21.8 86.2 227.1 151.6 15.7 122.7 195.3 152.6 16.9 122.7 195.3
Belgium 140.7 21.3 82.3 228.9 141.9 13.8 117.4 183.5 142.8 14.8 117.4 183.5
Bulgaria 179.0 14.8 153.9 230.7
Cyprus 173.7 24.1 126.9 234.0 172.2 24.9 126.9 234.0
Czech Republic 155.0 16.9 122.2 196.4 157.5 17.7 122.2 196.4
Denmark 131.8 19.4 88.2 195.9 134.4 16.5 107.0 176.0 136.9 17.4 107.0 176.0
Estonia 152.5 13.5 133.4 184.4 156.1 12.7 133.6 184.4
Finland 143.0 15.3 109.2 184.8 149.7 14.4 130.7 184.8 153.4 13.9 135.1 184.8
France 140.3 19.5 93.9 206.4 142.3 15.4 116.0 190.0 143.4 16.3 116.0 190.0
Germany 150.5 22.1 84.5 232.1 155.4 15.4 118.3 197.9 156.6 16.3 118.3 197.9
Greece 172.8 24.9 110.4 241.4 179.0 18.1 130.6 213.9 177.8 18.9 130.6 213.9
Hungary 153.6 15.9 123.3 193.1 156.1 16.6 126.3 193.1
Ireland 136.8 16.6 88.7 179.7 142.8 14.1 117.5 174.2 144.7 14.9 117.5 174.2
Italy 159.4 22.4 99.7 226.7 162.4 21.6 120.7 226.7 165.9 22.4 125.4 226.7
Latvia 160.9 17.3 129.4 203.4 163.8 18.1 129.4 203.4
Lithuania 156.7 16.9 126.4 206.9 160.1 17.0 126.5 206.9
Luxembourg 157.7 22.3 104.1 257.2 155.7 14.3 126.2 196.4 156.9 15.1 126.2 196.4
Netherlands 132.3 21.3 64.3 226.5 139.0 14.6 110.9 179.0 140.3 15.4 110.9 179.0
Poland 148.1 20.6 112.2 197.5 152.6 20.3 112.2 197.5
Portugal 154.6 22.9 85.3 223.5 160.0 16.4 129.0 201.0 161.6 16.4 129.0 201.0
Romania 164.7 17.1 136.7 205.6
Slovakia 157.5 17.7 124.4 201.8 160.7 18.1 124.4 201.8
Slovenia 150.4 14.8 118.6 191.4 151.0 16.2 118.6 191.4
Spain 150.0 24.8 74.2 217.1 157.4 20.9 118.1 217.1 159.7 21.7 118.1 217.1
Sweden 144.7 19.3 104.4 198.2 152.7 19.8 118.1 198.2 157.1 19.8 118.1 198.2
UK 155.1 19.6 89.9 201.5 163.2 16.8 127.8 201.5 167.1 16.7 127.8 201.5
PORK2/1995-6/2014 (EU15) 1/2005-6/2014 (EU25) 1/2007-6/2014 (EU27)
54
Table 4.2. Descriptive statistics of beef prices in the EU member states
The descriptive statistics reveal, among other things, that in the new member states pork
prices are higher than in the old member states. However, in the case of beef, the
situation is opposite. In figures 4.1, 4.2 and 4.3 are showed how the pork and beef prices
has evolved during the period 2/1995(EU15) or 1/2004(EU25) or 1/2007(EU27)-6/2014
in the EU member states. In order to make the figures more comparable, we have
indexed the data. From the user’s point of view, the price indexes are the most practical,
because index revisions do not interrupt the series. Additionally, visual inspection of the
pork and beef price time series, and in advance of the subsequent econometric analysis,
may help to assess sustainability issues in individual cases.
All in all, the figures confirm that there is some relationship between the pork and beef
prices. The relationship between the price series has strengthened since the beginning of
the 2000s in the EU15 member states. Before that, the developments in the pork and
beef price series were divergent. Notable, fluctuations in pork price series are much
larger than fluctuations in beef price series in the EU15. However, when examining the
price series of new member states the situation is opposite: beef prices have fluctuated
more than pork prices.
Mean Std. Min Max Mean Std. Min Max Mean Std. Min Max
Austria 303.4 41.4 215.0 410.0 335.8 34.9 243.3 410.0 343.6 35.1 243.3 410.0
Belgium 259.1 29.9 201.8 332.3 274.0 26.2 209.4 332.3 280.4 24.5 244.9 332.3
Bulgaria 242.2 44.8 163.6 360.5
Cyprus 286.5 22.9 235.5 345.0 293.3 20.5 251.7 345.0
Czech Republic 294.7 30.5 235.5 362.4 303.6 27.8 251.7 362.4
Denmark 301.2 49.3 219.2 413.8 340.7 39.6 259.3 413.8 350.7 37.5 277.4 413.8
Estonia 262.5 41.3 179.3 357.0 275.6 35.4 179.3 357.0
Finland 315.0 42.4 242.5 424.9 346.5 35.4 286.8 424.9 358.3 30.4 305.7 424.9
France 300.8 41.9 215.0 401.4 332.5 35.1 262.0 401.4 337.7 37.0 262.0 401.4
Germany 296.2 49.5 168.2 423.3 334.9 39.7 261.6 423.3 343.6 39.7 261.6 423.3
Greece 392.2 27.8 313.1 455.5 413.2 21.5 362.6 455.5 422.1 12.7 371.4 455.5
Hungary 242.9 19.0 185.1 296.0 239.4 19.5 185.1 296.0
Ireland 281.5 52.6 188.4 437.7 322.1 46.3 238.5 437.7 334.1 44.5 272.0 437.7
Italy 328.6 39.9 197.3 422.3 359.7 29.3 277.0 422.3 366.8 27.2 277.0 422.3
Latvia 207.1 39.4 127.9 321.0 217.9 36.8 147.0 321.0
Lithuania 250.8 46.1 164.3 335.2 261.3 45.5 178.5 335.2
Luxembourg 301.7 38.9 182.3 399.7 327.6 33.8 247.8 399.7 335.2 33.2 247.8 399.7
Netherlands 277.0 40.4 171.1 380.3 303.7 29.1 244.0 380.3 309.6 29.0 256.8 380.3
Poland 271.5 38.0 210.4 354.7 281.7 36.2 217.7 354.7
Portugal 324.0 31.4 247.2 390.7 347.0 21.5 283.9 390.7 351.6 17.4 308.4 390.7
Romania 249.4 40.4 159.9 346.1
Slovakia 291.5 41.4 182.4 390.7 302.5 39.5 216.5 390.7
Slovenia 314.4 32.8 254.1 392.0 323.2 31.0 258.7 392.0
Spain 305.4 40.6 208.1 400.7 336.7 32.4 279.0 400.7 343.7 31.7 283.9 400.7
Sweden 287.8 49.0 211.3 460.3 316.7 54.3 231.3 460.3 329.6 54.0 231.3 460.3
UK 292.4 55.9 195.3 450.9 330.5 56.6 238.7 450.9 345.2 54.4 276.8 450.9
BEEF2/1995-6/2014 (EU15) 1/2005-6/2014 (EU25) 1/2007-6/2014 (EU27)
55
The figures also reveal that there exist seasonal cycles particularly in pork prices and
mainly in the Southern European countries. In order to control for the potential seasonal
differences in our econometric investigations in the next chapter, we add dummy
variables for the months of January and March through December. For example, Osborn
(1993) argued that seasonal non-stationarity can often be adequately represented by
seasonal dummies.
Figure 4.1. The development of pork and beef prices in the EU15 member states
during the period 2/1995-6/2014, weekly data, index: week 5/1995=100.
50
100
150
200
50
100
150
200
50
100
150
200
50
100
150
200
50
100
150
200
1995 20142004 1995 20142004 1995 20142004
Austria Belgium Denmark
Finland France Germany
Greece Ireland Italy
Luxembourg Netherlands Portugal
Spain Sweden UK
Beef Pork
Year
Graphs by EU15 Member State
56
Figure 4.2. The development of pork and beef prices in the EU25 member states (excl.
EU15) during the period 1/2005-6/2014, weekly data, index: week 1/2005=100.
Figure 4.3. The development of pork and beef prices in the EU27 member states (excl.
EU25) during the period 1/2007-6/2014, weekly data, index: week 1/2007=100.
50
100
150
200
50
100
150
200
50
100
150
200
2005 20142010 2005 20142010 2005 20142010
Cyprus Czech Republic Estonia
Hungary Latvia Lithuania
Poland Slovenia Solvakia
Beef Pork
Year
Graphs by EU25 new Member States
50
100
150
200
2007 2010 2014 2007 2010 2014
Bulgaria Romania
Beef Pork
Year
Graphs by EU27 new Member States
57
5 EMPIRICAL RESULTS
5.1 Panel unit root results
Panel cointegration testing requires that variables are integrated in the same order. To
determine the level of integration of the variables, panel unit root tests have been carried
out. Thus, we start by presenting the results of five panel unit root tests discussed in
Section 3.1, namely, LLC, IPS, Breitung, ADF-Fisher and PP-Fisher tests. In this
application of these tests the dependence between the series has not been taken into
account. Table 5.1 show the statistics and the p-values from the panel unit root tests
concerning beef price series and Table 5.2 concerning pork price series. We use the
natural logarithms of the prices.
The first sections in Table 5.1 and 5.2 show the empirical result with price series that
include individual effects only, while the latter sections include individual effects and
individual trends. The trends amount to fixed effects in the first difference specification.
Probabilities for Fisher tests are computed using an asymptotic Chi-square distribution,
χ2 with 2*N degrees of freedom. IPS, LLC and Breitung tests assume asymptotic
normality. Selection of lags is based on the Schwarz criterion (SIC)18 and the Newey-
West bandwidth19 selection using the Bartlett-Kernel20 method.
The alternative hypothesis of a stationary process differs between panel unit root tests.
Namely, the alternative hypothesis of LLC and Breitung tests assume that all panels are
stationary. In other tests the alternative hypotheses assume that some or at least one
cross section is stationary. The null hypothesis is the same in all the tests: all panels
contain unit roots.
18 Estimating the lag length of autoregressive process for a time series is a crucial econometric. When fitting models, it is possible to increase the likelihood by adding parameters, but doing so may result in overfitting. Schwarz criterion (SIC) (or Bayesian information criterion (BIC)) resolves this problem by introducing a penalty term for the number of parameters in the model. Thus, SIC is a criterion for model selection and it is based on maximizing or minimizing the likelihood function. 19 Because the PP test differs from the ADF test in the treatment of serial correlation, the information-based method (such as SIC) does not apply to the PP test. In the PP test, an issue concerns the choice of a lag truncation parameter is a choice of the bandwidth (autocovariance lag). Newey and West (1994) proposed a method for estimating the optimal bandwidth from truncated sample autocovariances. 20 A weighting function used in non-parametric estimation techniques.
58
It is also worthwhile to member that both LLC and IPS tests present important size
distortions when either N is small or N is relatively large with respect to T. In our case N
is relatively small with respect to T.
According to the results, more or less, a unit root is detected for the level variables,
while the first differences appear to be stationary. For the beef and pork price series,
only the LLC and Breitung tests reject the null in some cases, while other three tests can
not reject the unit root hypothesis. After first-differencing the variables and repeating
the tests, nonstationarity is eliminated as all panel unit root tests reject the null of
nonstationarity for the first-differenced variables at the 1% level of significance. We can
thereby conclude that according to first generation panel unit root tests first-differenced
variables are stationary so that panel variables are integrated of order one, I(1). These
results imply that the long-run panel variables are not stationary, and that these variables
could be cointegrated.
59
EU1
5EU
15
EU2
5EU
15
EU2
5EU
27
EU15
EU1
5EU
25EU
15
EU25
EU
27
2/1
99
5-6
/20
14
1/2
00
5-6
/20
141
/20
05-6
/20
14
1/2
00
7-6
/20
141
/20
07-
6/2
01
41
/20
07
-6/2
014
2/1
99
5-6
/20
14
1/2
00
5-6
/20
14
1/2
00
5-6
/20
14
1/2
00
7-6
/20
14
1/2
00
7-6/
20
14
1/2
007
-6/2
01
4
Seri
es in
lev
els
Levi
n, L
in a
nd
Chu
-3.4
52-4
.08
3-7
.121
-4.9
14
-2.6
60
-2.4
97
-6.3
36
-4.4
85
-9.2
82
-5.6
42-2
.14
4-3
.951
(0.0
09
)***
(0.0
06)*
**(0
.00
5)*
**(0
.01
0)**
(0.0
15)*
*(0
.01
9)*
*(0
.00
1)*
**(0
.03
4)*
*(0
.000
)***
(0.0
04)
***
(0.0
57
)*(0
.01
8)*
*
Im, P
esa
ran
and
Shi
n0.
665
0.4
960.
551
0.7
280.
617
0.3
82
0.85
40
.63
60
.37
90
.93
80
.477
0.5
13
(0.5
51
)(0
.438
)(0
.47
6)
(0.5
81)
(0.5
27
)(0
.35
9)(0
.74
3)
(0.5
22)
(0.0
41)*
*(0
.86
2)(0
.02
4)**
(0.3
9)**
Bre
itu
ng
0.36
31
.193
-0.8
701
.851
-0.9
27
-1.2
25
-0.3
44
0.9
42
-0.3
23
0.2
52
-0.2
88
-0.9
82
(0.2
47
)(0
.646
)(0
.019
)**
(0.9
76)
(0.0
09
)***
(0.0
03)
***
(0.0
07
)***
(0.0
84)*
(0.0
38)*
*(0
.02
0)*
*(0
.03
4)**
(0.0
07
)***
Fish
er A
DF
-1.4
28-2
.04
6-1
.872
-2.7
81
-2.5
47
-1.8
92
-1.6
76
-2.5
46
-2.0
47
-2.9
24-2
.49
2-1
.973
(0.6
69
)(0
.772
)(0
.72
7)
(0.8
31)
(0.8
08
)(0
.76
4)(0
.72
1)
(0.7
99)
(0.7
23
)(0
.85
4)(0
.802
)(0
.79
3)
Fish
er P
P-2
.038
-1.5
14
-1.1
13-1
.87
1-1
.09
0-1
.55
5-1
.86
5-1
.23
2-0
.88
1-1
.660
-1.0
59
-1.4
36
(0.6
57
)(0
.726
)(0
.90
9)
(0.8
31)
(0.9
66
)(0
.83
8)(0
.74
3)
(0.9
24)
(0.9
75
)(0
.81
6)(0
.957
)(0
.87
5)
Seri
es in
fir
st d
iffe
ren
ces
Levi
n, L
in a
nd
Chu
-12
7.45
2-1
07.3
63
-76.
481
-169
.71
7-1
50
.626
-90
.990
(0.0
00
)***
(0.0
00)*
**(0
.00
0)*
**(0
.00
0)**
*(0
.00
0)*
**(0
.00
0)**
*
Im, P
esa
ran
and
Shi
n-6
0.3
84
-45
.279
-78.
919
-89
.258
-90.
316
-82
.623
(0.0
00
)***
(0.0
00)*
**(0
.00
0)*
**(0
.00
0)**
*(0
.00
0)*
**(0
.00
0)**
*
Bre
itu
ng
-30.
78
2-1
9.6
77-1
6.51
1-3
8.4
62-2
4.95
3-1
2.8
74
(0.0
00
)***
(0.0
00)*
**(0
.00
0)*
**(0
.00
0)**
*(0
.00
0)*
**(0
.00
0)**
*
Fish
er A
DF
-10.
95
0-1
5.3
23-1
2.12
2-1
8.6
451
6.5
48-1
3.5
78
(0.0
00
)***
(0.0
00)*
**(0
.00
0)*
**(0
.00
0)**
*(0
.00
0)*
**(0
.00
0)**
*
Fish
er P
P-3
9.5
70
-32
.451
-25.
811
-40
.624
-24.
975
-29
.557
(0.0
00
)***
(0.0
00)*
**(0
.00
0)*
**(0
.00
0)**
*(0
.00
0)*
**(0
.00
0)**
*
Indi
vid
ual e
ffec
ts +
indi
vid
ual t
rend
ln b
ee
fIn
divi
dua
l eff
ects
Tab
le 5
.1. R
esu
lts
of p
anel
un
it r
oot
test
s: B
eef
Not
e:
***,
**,
* d
enote
sig
nif
ican
ce a
t 1%
, 5%
an
d 10
%,
resp
ecti
vely
. p
-val
ues
are
rep
orte
d in
par
enth
eses
. P
rob
abil
itie
s fo
r F
ishe
r te
sts
are
com
put
ed u
sing
an
asym
ptoti
c C
hi-s
qu
are
dist
rib
uti
on.
All
oth
er t
ests
ass
ume
asym
ptot
ic n
orm
alit
y. S
elec
tio
n of
lag
s ba
sed
on
SIC
. N
ewey
-W
est
ban
dw
idth
sel
ecti
on u
sin
g B
artl
ett
kern
el. N
ull
hyp
othe
sis:
uni
t ro
ot.
60
EU1
5EU
15
EU2
5EU
15
EU2
5EU
27
EU1
5EU
15
EU2
5EU
15
EU2
5EU
27
2/1
99
5-6
/20
141/
20
05
-6/2
014
1/2
00
5-6
/20
141/
20
07
-6/2
01
41
/20
07-
6/2
01
41/
20
07
-6/2
01
42
/19
95-6
/20
14
1/2
00
5-6
/20
14
1/2
005
-6/2
01
41
/20
07-
6/2
014
1/2
007
-6/2
01
41
/20
07-
6/2
014
Se
rie
s in
leve
ls
Levi
n, L
in a
nd
Ch
u-4
.43
4-2
.895
-2.5
53
2.0
76
2.7
92
4.6
17
-6.2
71
-4.3
90
-1.7
23
2.4
72
4.76
67
.58
8
(0.0
04
)***
(0.0
05
)***
(0.0
18
)**
(0.2
57
)(0
.33
3)(0
.17
9)
(0.0
02)*
**(0
.00
6)*
**(0
.01
4)*
*(0
.35
1)(0
.46
8)
(0.7
63
)
Im, P
esar
an a
nd
Shin
-0.6
22
-0.7
52-0
.43
4-0
.44
5-0
.27
9-0
.39
0-0
.27
7-0
.08
96
-0.8
12
-0.5
43
-0.6
14
-1.1
54
(0.3
56
)(0
.29
2)
(0.4
58
)(0
.45
7)
(0.5
72)
(0.4
69
)(0
.01
4)*
*(0
.00
6)*
**(0
.299
)(0
.04
1)*
*(0
.34
3)
(0.0
97
)
Bre
itun
g-1
.46
6-2
.441
1.3
90
0.3
13
0.2
53
0.2
18
-3.8
97
-5.1
26
0.3
64-2
.35
20.
221
0.4
10
(0.0
19
)**
(0.0
05
)***
(0.0
51
)(0
.36
5)
(0.2
86)
(0.2
24
)(0
.001
)***
(0.0
01
)***
(0.4
18)
(0.0
89)
(0.2
63
)(0
.47
5)
Fish
er A
DF
-0.8
04
-0.7
36-0
.37
8-0
.64
9-0
.58
9-0
.42
5-0
.76
5-0
.70
4-0
.31
5-0
.68
6-0
.62
5-0
.55
2
(0.9
64
)(0
.95
7)
(0.9
22
)(0
.94
5)
(0.9
40)
(0.9
31
)(0
.95
0)(0
.94
1)
(0.9
25)
(0.9
46)
(0.9
48
)(0
.93
7)
Fish
er P
P-0
.63
2-0
.532
-0.3
11
-0.6
19
-0.6
28
-0.4
55
-0.6
61
-0.4
74
-0.3
93
-0.5
56
-0.5
12
-0.4
20
(0.9
43
)(1
.00
0)
(1.0
00
)(0
.96
7)
(0.9
60)
(1.0
00
)(0
.94
0)(1
.00
0)
(1.0
00)
(0.9
74)
(1.0
00
)(1
.00
0)
Se
rie
s in
fir
st d
iffe
ren
ces
Levi
n, L
in a
nd
Ch
u-1
48
.66
1-1
12.4
14
-96.
526
-224
.13
3-1
97
.06
7-1
90.4
71
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
Im, P
esar
an a
nd
Shin
-62.
335
-71
.45
6-7
8.98
0-8
5.4
38
-46
.17
7-5
0.5
12
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
Bre
itun
g-7
4.58
9-5
1.1
26
-43.
229
-80
.16
5-6
2.8
34
-37
.24
4
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
Fish
er A
DF
-24.
878
-20
.24
7-1
2.33
2-1
7.7
11
-15
.94
8-1
3.2
86
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
Fish
er P
P-4
7.25
4-5
2.8
15
-39.
729
-58
.61
0-3
2.8
82
-25
.56
9
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
Ind
ivid
ual e
ffec
tsIn
divi
dua
l eff
ects
+ in
divi
dua
l tre
nd
ln p
ork
Tab
le 5
.2. R
esu
lts
of p
anel
un
it r
oot
test
s: P
ork
Not
e:
***,
**,
* d
enot
e si
gni
fica
nce
at 1
%,
5%
and
10%
, re
spec
tive
ly.
p-v
alue
s ar
e re
por
ted
in
pare
nthe
ses.
Pro
bab
ilit
ies
for
Fis
her
test
s ar
e co
mp
uted
usi
ng
an a
sym
ptot
ic C
hi-s
qua
re d
istr
ibut
ion.
All
oth
er t
ests
ass
um
e as
ympt
otic
nor
mal
ity.
Sel
ecti
on o
f la
gs b
ased
on
SIC
. N
ewey
-Wes
t ba
ndw
idth
sel
ecti
on u
sin
g B
artl
ett
kern
el. N
ull
hyp
othe
sis:
uni
t ro
ot.
61
Unit root tests that assume cross-sectional independence can have low power if
estimated on data that have cross-sectional dependence. As a result, Pesaran's (2007)
CIPS test for unit roots is calculated. This is a second generation unit root test that
allows for cross-sectional dependence. Pesaran included cross-sectional averages of the
lagged levels as the common factor to filter out the cross-section dependence. The
statistic is constructed from the average of the cross-sectionally ADF (CADF) t-
statistics. The averaging of the group-specific results follows the procedure in the Im,
Pesaran and Shin (2003) test and brings out the CIPS t-statistic. Under the null of
nonstationarity the test statistic has a non-standard distribution. The critical values are
tabulated by the author for different combinations of N and T.
The panel unit root test results using CIPS test are reported in Table 5.3. These tests
were estimated with a constant term or with a constant term and trend. We choose to
include maximum 5 lags in each regression. We get qualitatively similar findings when
the number of lags in the regression is increased. Values of statistics for different
number of lags are reported for comparison purposes in order to assess the robustness of
the conclusion regarding non-stationarity of the data.
The Pesaran (2007) test is performed using the Stata -multipurt- command written by
Markus Eberhardt (2012).21
In this case the series were not stationary at the levels. So we made the regression
analysis with first differences of the variables. The results of the CIPS test indicate that
each series contains a unit root.
21 -multipurt- uses Scott Merryman's -xtfisher- and Piotr Lewandowski's -pescadf-.
62
EU1
5E
U1
5E
U2
5EU
15
EU
25
EU2
7E
U1
5EU
15
EU
25
EU
15
EU
25
EU
27
2/1
995
-6/2
014
1/2
005
-6/2
01
41
/20
05-
6/2
014
1/2
007
-6/2
014
1/2
00
7-6
/20
14
1/2
007
-6/2
014
2/1
995
-6/2
01
41
/20
05-6
/201
41
/200
5-6
/20
14
1/2
00
7-6
/20
141
/20
07-6
/201
41
/20
07
-6/2
01
4
Seri
es
in le
vels
LAG
1-1
.75
3-1
.55
6-1
.48
5-1
.52
1-1
.31
9-1
.21
0-2
.02
0-1
.81
3-1
.74
7-1
.77
8-1
.59
6-1
.50
2
LAG
2-1
.34
7-1
.59
8-1
.09
1-1
.31
9-1
.37
2-1
.06
3-1
.87
2-1
.83
3-1
.63
8-2
.55
6**
-1.6
17
-1.2
9
LAG
3-1
.42
4-1
.10
5-0
.89
1-1
.32
8-0
.95
3-0
.81
4-1
.95
1-1
.72
0-1
.42
0-1
.85
3-2
.18
1**
-1.3
62
LAG
4-1
.44
0-1
.21
9-1
.02
4-2
.04
5*
-1.4
25
-0.7
58
-2.7
52
**-1
.97
1-1
.94
4*
-1.5
78
-1.8
55
-1.4
96
LAG
5-1
.39
5-1
.34
6-0
.97
1-1
.75
7-1
.15
2-1
.54
3
Seri
es
in f
irst
dif
fere
nce
s
LAG
1-3
.25
6**
*-3
.07
1**
*-2
.82
7**
*-2
.79
0**
*-2
.54
8**
*-2
.50
1**
*
LAG
2-3
.21
*5**
-2.8
77
***
-2.7
16
***
-2.7
14
***
-2.6
84
***
-2.3
83*
**
LAG
3-3
.03
3**
*-2
.86
1**
*-2
.69
0**
*-2
.66
5**
*-2
.48
1**
*-2
.30
6**
LAG
4-4
.01
3**
*-3
.05
7**
*-2
.84
2**
*-2
.84
8**
*-2
.75
2**
*-2
.53
5**
*
LAG
5-3
.87
3**
*-2
.84
6**
*-2
.89
4**
*-2
.80
6**
*-2
.75
7**
*2
.66
4**
*
EU1
5E
U1
5E
U2
5EU
15
EU
25
EU2
7E
U1
5EU
15
EU
25
EU
15
EU
25
EU
27
2/1
995
-6/2
014
1/2
005
-6/2
01
41
/20
05-
6/2
014
1/2
007
-6/2
014
1/2
00
7-6
/20
14
1/2
007
-6/2
014
2/1
995
-6/2
01
41
/20
05-6
/201
41
/200
5-6
/20
14
1/2
00
7-6
/20
141
/20
07-6
/201
41
/20
07
-6/2
01
4
Seri
es
in le
vels
LAG
1-2
.45
5-2
.67
3-2
.69
0-2
.81
8-2
.88
1-2
.89
1-2
.87
4-2
.98
0**
-2.8
18
-3.0
42*
*-2
.95
5-3
.07
6*
LAG
2-2
.17
1-2
.25
3-2
.54
5-2
.73
6-2
.80
8-2
.78
2-2
.86
4-2
.88
1-2
.87
0-2
.93
8-2
.84
4-3
.21
5**
LAG
32
.06
7-2
.32
9-2
.38
1-2
.61
9-2
.79
2-2
.67
0-2
.80
62
.65
4-2
.48
2-2
.84
6-2
.69
1-2
.98
2*
LAG
4-2
.11
4-2
.47
2-2
.41
3-2
.55
7-2
.75
5-2
.70
6-2
.94
7**
-2.7
56
-2.7
15
-2.9
29
-2.7
56
-3.0
44
**
LAG
5-1
.98
1-2
.29
0-2
.33
3-2
.78
4-2
.69
2-2
.91
6-3
.05
5**
-2.8
45
-2.5
24
-2.9
92
*-2
.72
1-2
.89
0
Seri
es
in f
irst
dif
fere
nce
s
LAG
1-5
.01
7**
*-5
.63
7**
*-5
.09
6**
*-5
.45
4**
*-5
.67
3**
*-5
.56
5**
*
LAG
2-4
.86
8**
*-4
.93
7**
*-4
.67
3**
*-5
.02
4**
*-4
.88
1**
*-5
.29
0**
*
LAG
3-4
.60
8**
*-4
.87
1**
*-4
.92
3**
*-4
.89
5**
*-4
.86
6**
*-5
.01
4**
*
LAG
4-4
.87
5**
*-5
.33
6**
*-4
.94
4**
*-5
.23
1**
*-5
.69
0**
*-5
.28
0**
*
LAG
5-4
.71
7**
*-5
.47
6**
*-4
.83
3**
*-5
.08
4**
*-5
.32
2**
*-5
.41
7**
*
Ind
ivid
ual e
ffec
ts
Ind
ivid
ual e
ffec
tsIn
divi
dua
l eff
ects
+ in
divi
dual
tre
nd
Indi
vid
ual e
ffec
ts +
indi
vidu
al t
rend
ln b
ee
f
ln p
ork
Note
: *
**, *
*, *
den
ote
sig
nifi
can
ce a
t 1
%, 5
% a
nd 1
0%
, res
pec
tive
ly. N
ull
hyp
oth
esis
: un
it r
oot.
Tab
le 5
.3. R
esu
lts
of P
esa
ran
's C
IPS
tes
t
63
The first and second generation panel unit root tests may suffer from significant loss of
power if the data contains structural breaks. It is complicated to distinguish between unit
root and stationary processes with structural breaks. Today, the panel data unit root tests
which allow for structural breaks have received considerable attention among
econometricians. Therefore, we extend our analysis by performing Im et al. (2005)
panel LM test (ILT), that is an extension of the Lee and Strazicich’s (2004) minimum
LM test with one structural break, as well as the Im et al. (2010) test that allow for the
presence of heterogeneous structural breaks (ILT*) and cross-sectional dependence in
the data (ILTCA*). The objective of this analysis is to test unit root hypotheses in the
presence of structural change at the unknown time of the break.
In Table 5.4 we present the results of two versions of the panel LM test that is with and
without cross-sectional dependence, and perform the tests considering a heterogeneous
break only in level and both in level and trend. We employ the one-break version of the
panel LM tests of Im et al. (2005) (assuming cross-sectional independence) and Im et al.
(2010) (assuming cross-sectional dependence). The lag order selection are chosen using
the recursive t-statistic procedure with an upper bound of kmax=2. The null hypothesis of
the LM panel test is that all series contain unit roots, with the alternative that some of
the series in the panel are stationary.
The results show that in almost all cases, after taking into account the fact that the two
meat price series are subject to a structural break in mean and both in mean and the
slope of the series, the null hypothesis of a unit root is not rejected, indicating that these
series are not stationary with the presence of a structural break. The previous results
from unit root tests without breaks generally indicate that pork and beef price series
follow unit root process. Allowing for a structural break does not change the results
from non-stationary to stationary. However, it should be noted that if we allow more
than one structural break point, the results could be altered.
To make the analysis robust, the results of panel data unit root tests are compared with
those obtained with individual unit root tests. Appendix 1 includes LM unit root test
results for each individual country assuming one structural breakpoint. We do not
consider structural break-dates as being exogenously determined, but we test
endogenously for them, i.e. assuming the break-date to be unknown. According to the
results in 55 of the 67 pork series and in 49 of the 67 beef series, the unit root null
cannot be rejected at the 10% significance level. According to the break locations, there
seems to be many important events that have caused significant breaks in the time paths
of price series in the EU member states.
64
EU15
EU15
EU25
EU15
EU25
EU27
EU15
EU15
EU25
EU15
EU25
EU27
2/19
95-6
/201
41/
2005
-6/2
014
1/20
05-6
/201
41/
2007
-6/2
014
1/20
07-6
/201
41/
2007
-6/2
014
2/19
95-6
/201
41/
2005
-6/2
014
1/20
05-6
/201
41/
2007
-6/2
014
1/20
07-6
/201
41/
2007
-6/2
014
C-S
Ind
ep
en
de
nce
ILT
-0.6
21-0
.375
-0.0
98-0
.846
-1.3
10-1
.210
ILT*
1.5
742.
885
0.75
32.
267
0.55
51.
068
1.07
51.
852
0.71
22.
086
0.47
10.
656
C-S
De
pe
nd
en
ce
ILT C
A*
5.2
174.
267
2.98
26.
334
1.58
73.
014
-3.5
76**
-3.1
84**
-2.0
95-2
.894
-1.6
48-1
.286
EU15
EU15
EU25
EU15
EU25
EU27
EU15
EU15
EU25
EU15
EU25
EU27
2/19
95-6
/201
41/
2005
-6/2
014
1/20
05-6
/201
41/
2007
-6/2
014
1/20
07-6
/201
41/
2007
-6/2
014
2/19
95-6
/201
41/
2005
-6/2
014
1/20
05-6
/201
41/
2007
-6/2
014
1/20
07-6
/201
41/
2007
-6/2
014
C-S
Ind
ep
en
de
nce
ILT
-1.5
23-1
.222
-1.8
25-1
.085
-2.3
68-1
.454
ILT*
5.8
524.
561
6.11
34.
219
2.69
63.
224
2.00
52.
854
3.04
51.
328
1.78
91.
151
C-S
Ind
ep
en
de
nce
ILT C
A*
9.6
5210
.854
7.56
28.
881
5.75
46.
229
-2.3
43-1
.975
-2.8
58-1
.567
-2.6
41-2
.753
Leve
lLe
vel +
tre
nd
Leve
lLe
vel +
tre
nd
ln b
ee
f
ln p
ork
Not
e:
***,
**,
* d
enot
e si
gni
fica
nce
at 1
%,
5% a
nd
10%
, re
spec
tive
ly. N
ull
hyp
oth
esis
: un
it r
oot.
Tab
le 5
.4. R
esu
lts
of p
anel
LM
tes
ts w
ith
on
e st
ruct
ura
l br
eak
65
We employed a series of panel unit root tests that assume cross sectional independence
(the so called first generation panel data unit root tests) and cross-sectional dependence
(second generation panel data unit root tests). These tests do not allow for structural
breaks. For this reason we completed our analysis by employing the Lagrange
Multiplier (LM) panel unit root test with a break. In summary, the results of panel unit
root test we have employed indicate quite unanimously that both beef and pork series
contain a unit root and are integrated of order one, I(1). Next, we can carry out the
second part of the empirical analysis (panel cointegration tests) and check for the
existence of a long-run relationship between the price variables.
5.2 Cointegration test results
We are interested in the long-run equilibrium relationship between the beef and pork
prices in the EU. However, it cannot be consistently estimated if single series have a
unit root, unless they are cointegrated in the long run. Thus, the next step of our analysis
is to perform panel cointegration tests.
In panel settings, there are several ways of testing the null hypothesis of no cointegra-
tion. Namely, these test are grouped in two large families: the residual-based ones
(Pedroni, 1999, 2004; Kao, 1999), constructed on the basis of the Engle and Granger’s
(1987) two step test and likelihood-based ones (Maddala and Wu, 1999) which
represent the generalization of the widely-used Johansen (1991, 1996) test for vector
autoregressive models to panel data. It proceeds by applying the Johansen test to each
cross-section unit individually and then combining the marginal significance values for
each of these to arrive at a single p-value. Also, we use panel cointegration tests
developed by Westerlund (2007) and Westerlund and Edgerton (2007) that are based on
the structural dynamic, as opposed to the residual dynamic. The main idea is to test the
null hypothesis of no cointegration by inferring whether the error-correction term in the
conditional panel error-correction model is equal to zero.
In Table 5.5, we consider Pedroni test with deterministic intercept and trend.
Deterministic time trend has been included to account for cross sectional dependence.
Among the seven Pedroni's panel cointegration tests, four are based on the within
dimension (panel cointegration statistics) and the three others on the between dimension
(group mean panel cointegration statistics). All tests are based on the null hypothesis of
no cointegration for all countries. Under the alternative hypothesis, for the panel
statistics, there is cointegration for all countries i. However, the group statistics allow
for heterogeneity across countries under the alternative hypothesis.
66
The estimations are performed in both directions with the dependent and independent
variables interchanged to find evidence for the existence of a bi-directional relationship
between the pork and beef prices. The majority of Pedroni’s test statistics for
heterogeneous panels indicate the possibility of a bi-directional cointegrating (or long-
run equilibrium) relationship between pork and beef prices. Overall, we can claim that
only one among the seven statistics used by the Pedroni test (ie. Panel ν-statistics) does
not reject the null hypothesis of no cointegration. The cointegrating relationship seems
to be stronger when we use the beef price series as a dependent variable.
EU1
5EU
15
EU2
5EU
15
EU2
5EU
27
EU1
5EU
15
EU2
5EU
15
EU2
5EU
27
2/1
99
5-6
/20
14
1/2
00
5-6
/20
14
1/2
005
-6/2
014
1/2
007
-6/2
014
1/2
007
-6/2
014
1/2
007
-6/2
014
2/1
995
-6/2
014
1/2
005
-6/2
014
1/2
00
5-6
/20
14
1/2
00
7-6
/20
14
1/2
00
7-6
/20
14
1/2
00
7-6
/20
14
Pa
nel C
oin
teg
rati
on S
tati
stic
s (W
ith
in-D
imen
sio
n)
Pan
el ν
-sta
tist
ics
2.2
622
.698
1.6
383
.747
2.5
452
.290
1.9
54
2.4
78
1.0
25
0.8
85
1.4
251
.653
(0.0
31)*
*(0
.011
)**
(0.0
93)*
(0.0
00)*
**(0
.01
6)*
*(0
.02
9)**
(0.0
57)*
(0.0
12
)**
(0.1
16
)(0
.30
5)
(0.0
88
)*(0
.07
1)*
Pan
el P
P ρ
-sta
tist
ics
-5.8
23
-3.8
75
-3.1
12
-5.7
06
-2.4
35-2
.809
-2.4
35-2
.603
-2.0
58-2
.914
-2.3
11
-2.2
21
(0.0
00)*
**(0
.000
)***
(0.0
13)*
*(0
.000
)***
(0.0
21
)**
(0.0
15)
**(0
.01
2)*
*(0
.00
8)*
**(0
.05
5)*
(0.0
05
)***
(0.0
10)*
*(0
.021
)**
Pan
el P
P t
-sta
tist
ics
-5.0
64
-5.8
88
-4.4
46
-4.7
73
-3.2
83-2
.849
-2.8
38-2
.031
-3.0
47-1
.878
-3.0
02
-2.6
22
(0.0
00)*
**(0
.000
)***
(0.0
00)*
**(0
.000
)***
(0.0
02)*
**(0
.00
7)*
**(0
.01
5)*
*(0
.08
3)*
(0.0
23)*
*(0
.00
4)*
**(0
.020
)**
(0.0
17)*
*
Pan
el A
DF
t-st
atis
tics
-10.
577
-12.
544
-14.
886
-11.
42
9-9
.005
-7.0
29-6
.132
-7.1
17-5
.027
-8.1
65-5
.74
9-3
.81
9
(0.0
00)*
**(0
.000
)***
(0.0
00)*
**(0
.000
)***
(0.0
00)*
**(0
.00
0)*
**(0
.00
7)*
**(0
.00
0)*
**(0
.07
1)*
(0.0
00
)***
(0.0
54
)*(0
.12
5)
Gro
up
Mea
n P
anel
Co
inte
gra
tio
n St
ati
stic
s (B
etw
ee
n-D
imen
sion
)
Gro
up P
P ρ
-sta
tist
ics
-3.9
59
-5.7
11
-2.2
62
-3.3
44
-2.7
48-1
.855
-2.8
45-3
.559
-2.0
07-3
.985
-2.3
60
-1.7
62
(0.0
00)*
**(0
.000
)***
(0.0
25)*
*(0
.002
)***
(0.0
11
)**
(0.0
71)*
(0.0
00
)***
(0.0
00
)***
(0.0
28)*
*(0
.00
0)*
**(0
.025
)**
(0.1
31
)
Gro
up P
P t
-sta
tist
ics
-4.5
64
-5.6
33
-7.9
47
-6.3
85
-5.7
06-6
.91
-2.8
11-3
.384
-3.8
75-6
.135
-4.5
64
-5.8
28
(0.0
00)*
**(0
.000
)***
(0.0
00)*
**(0
.000
)***
(0.0
00)*
**(0
.00
0)*
**(0
.00
5)*
**(0
.00
0)*
**(0
.00
0)*
**(0
.00
0)*
**(0
.000
)***
(0.0
00)*
**
Gro
up A
DF
t-st
atis
tics
-12.
078
-15.
125
-14.
281
-13.
11
2-9
.179
-10
.73
3-7
.316
-8.3
04-6
.926
-9.0
01-8
.07
8-6
.40
4
(0.0
00)*
**(0
.000
)***
(0.0
00)*
**(0
.000
)***
(0.0
00)*
**(0
.00
0)*
**(0
.00
0)*
**(0
.00
0)*
**(0
.00
0)*
**(0
.00
0)*
**(0
.000
)***
(0.0
09)*
**
Num
ber
of
test
sta
tist
ics
that
rej
ect
H0
at 5
%
leve
l
7/7
7/7
6/7
7/7
7/7
6/7
6/7
6/7
4/7
6/7
5/7
4/7
ln P
ork
Dep
. var
. of
coin
t. r
eg.
Dep
. var
. o
f co
int.
reg
.
ln B
eef
Tab
le 5
.5.
Res
ult
s of
Ped
ron
i re
sidu
al
coin
tegr
atio
n t
ests
Not
e:
***,
**,
* d
enot
e si
gni
fica
nce
at 1
%,
5%
and
10
%,
resp
ecti
vely
. p
-val
ues
are
rep
orte
d i
n pa
rent
hese
s. S
elec
tio
n of
lag
s ba
sed
on
SIC
. New
ey-W
est
ban
dw
idth
sel
ecti
on u
sin
g B
artl
ett
kern
el. N
ull
hypo
thes
is:
no c
oint
egra
tio
n.
67
The Kao (1999) test follows the same basic approach as the Pedroni’s tests. Totally, Kao
proposed four DF-type panel cointegration tests based on the OLS residuals from the
homogeneous panel regression. Kao also uses ADF-type panel cointegration test for
cointegration in heterogeneous panel. In this test version, the parameters are allowed to
differ across the cross-sections. In our study, we only report Kao’s ADF- type test.
In the null hypothesis, the residuals are nonstationary (there is no cointegration) and in
the alternative hypothesis, the residuals are stationary (there is a cointegrating
relationship among the variables). Table 5.6 reports the results of Kao’s residual panel
cointegration tests, which reject the null of no cointegration at the 5% significance level
in most of the specifications. Findings of Kao’s residual cointegration test corroborates
the findings of Pedroni’s residual cointegration test in Table 5.5.
Both the Pedroni and the Kao tests use the Schwartz Information Criterion (SIC) to
automatically select the appropriate lag length. Further, spectral estimation is
undertaken by the Bartlett kernel with the bandwidth selected by the Newey-West
algorithm. Deterministic time trends are included in all specifications.
68
EU1
5EU
15
EU2
5EU
15
EU2
5EU
27
EU1
5EU
15
EU2
5EU
15
EU2
5EU
27
2/1
995
-6/2
01
41
/20
05
-6/2
014
1/2
00
5-6
/20
14
1/2
007
-6/2
01
41
/20
07
-6/2
014
1/2
007
-6/2
014
2/1
995
-6/2
014
1/2
00
5-6
/20
141
/20
05-6
/20
14
1/2
00
7-6
/20
141
/20
07
-6/2
01
41
/20
07-6
/20
14
AD
F-st
atis
tics
-1.6
42
-2.5
52
-0.9
50
-3.2
51
-1.2
38
-1.4
44
-1.3
22
-3.2
47
-0.7
89
-2.6
61
-1.5
21
-1.7
28
(0.0
02)
***
(0.0
00)
***
(0.0
41
)**
(0.0
00)
***
(0.0
71)
*(0
.02
5)*
*(0
.00
5)**
*(0
.00
0)*
**(0
.06
6)*
(0.0
00)
***
(0.0
35
)**
(0.0
02)
***
Dep
. var
. of
coin
t. r
eg.
Dep
. var
. o
f co
int.
reg
.
ln B
ee
fln
Po
rk
Not
e:
***
, **
, *
den
ote
sig
nif
ican
ce a
t 1
%,
5%
and
10%
, re
spec
tive
ly.
p-v
alue
s ar
e re
por
ted
in
pare
nthe
ses.
Sel
ecti
on o
f la
gs b
ased
on
SIC
. N
ewey
-Wes
t
ban
dw
idth
sel
ecti
on
usin
g B
artl
ett
kern
el. N
ull
hyp
oth
esis
: n
o co
inte
grat
ion.
Tab
le 5
.6. R
esu
lts
of K
ao r
esid
ual
coi
nte
grat
ion
tes
t
69
Since we might have cross-sectional dependence in our series, cross-sectional
dependence in cointegration vectors is likely. Therefore, we perform the Westerlund
(2007) cointegration test with bootstrap under the assumption of cross-sectional
dependence.
Westerlund (2007) developed four new panel cointegration tests that are based on
structural dynamics and do not impose any common-factor restriction. The idea is to test
whether the error-correction term in the panel error-correction model is equal to zero.
The four kinds of tests can be divided into two groups by the difference of the
alternative hypotheses. The two tests called group-mean tests are designed to test the
alternative hypothesis that at least one unit is cointegrated, while the panel tests are
designed to test the alternative hypothesis that the panel is cointegrated as a whole.
These two tests can be used in both cases of cross-sectional dependency and
independence cases. These tests allow also heterogeneity among the units forming the
panel.
We adopt the Westerlund (2007) panel cointegration tests selecting the lead and lag
orders based on the minimum AIC (Akaike’s Information Criterion) and the Bartlett
kernel window width. We perform cointegration tests with both a constant and a trend.
To take consideration of the cross-sectional dependence we introduce bootstrap into the
test to get the robust critical values for the test statistics. The number of replication for
the bootstrap is 2000.22
The Westerlund’s test takes no cointegration as the null hypothesis. Table 5.7 below
report the panel cointegration results. All the group mean panel cointegratin statistics
and most of the panel cointegration statistics reject the null of no cointegration at the
5% level of significance. Overall, the null hypothesis of no-cointegration is rejected in
both asymptotic standard distribution and in bootstrap method. The results suggest that
cointegration relationship exists between the series and they are expected to move
together in the long run.
22 We then used Stata command -xtwest- to test for cointegration.
70
EU1
5EU
15
EU2
5E
U1
5E
U2
5EU
27
EU1
5EU
15
EU2
5EU
15
EU
25
EU
27
2/1
99
5-6
/20
14
1/2
00
5-6
/20
14
1/2
00
5-6
/20
14
1/2
00
7-6
/20
14
1/2
007
-6/2
01
41
/20
07-6
/20
14
2/1
995
-6/2
014
1/2
00
5-6
/20
14
1/2
00
5-6
/20
14
1/2
007
-6/2
01
41
/20
07-6
/20
14
1/2
007
-6/2
01
4
Gro
up M
ean
Pan
el C
oin
teg
rati
on
Sta
tist
ics
Gτ
-9.2
11
-12
.755
-7.5
42-8
.221
-13
.63
5-1
1.7
9-1
0.4
61-9
.35
7-7
.15
9-8
.446
-10
.95
4-1
2.3
49
asym
pto
tic
p-va
lue
(0.0
00)*
**(0
.000
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00)*
**(0
.000
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
boo
tsta
p p
-va
lue
(0.0
02)*
**(0
.000
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00)*
**(0
.000
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
Gα
-8.9
28
-10
.212
-8.3
17-1
1.4
26
-15
.66
2-9
.34
3-9
.52
3-9
.87
6-1
0.2
25-1
2.3
13
-12
.84
5-1
0.6
12
asym
pto
tic
p-va
lue
(0.0
00)*
**(0
.000
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00)*
**(0
.000
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
boo
tsta
p p
-va
lue
(0.0
00)*
**(0
.000
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00)*
**(0
.000
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
Pane
l Co
inte
grat
ion
Stat
isti
cs
P τ-5
.71
5-8
.14
5-4
.021
-3.6
46-7
.722
-7.1
48
-5.2
13
-7.8
58
-4.1
18
-3.4
29-7
.058
-6.9
24
asym
pto
tic
p-va
lue
(0.0
01)*
**(0
.000
)***
(0.0
02
)***
(0.0
04
)***
(0.0
00
)***
(0.0
02
)***
(0.0
01
)***
(0.0
00)*
**(0
.002
)***
(0.0
05
)***
(0.0
00
)***
(0.0
03
)***
boo
tsta
p p
-va
lue
(0.0
07)*
**(0
.002
)***
(0.0
05
)***
(0.0
10
)**
(0.0
03
)***
(0.0
03
)***
(0.0
08
)***
(0.0
02)*
**(0
.006
)***
(0.0
10
)**
(0.0
03
)***
(0.0
04
)***
P α-5
.22
4-8
.45
7-3
.228
-6.1
13-3
.832
-4.5
57
-4.8
88
-8.3
71
-3.3
47
-6.0
89-3
.665
-4.4
26
asym
pto
tic
p-va
lue
(0.0
00)*
**(0
.000
)***
(0.0
00
)**
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)**
(0.0
00)*
**(0
.000
)***
(0.0
00
)***
(0.0
00
)***
(0.0
00
)***
boo
tsta
p p
-va
lue
(0.0
45
)**
(0.0
16
)**
(0.2
18
)(0
.03
6)*
*(0
.18
4)
(0.0
56
)*(0
.05
0)*
(0.0
20
)**
(0.1
92
)(0
.03
8)*
*(0
.18
5)
(0.0
58
)*
Dep
. va
r. o
f co
int.
reg
. ln
Po
rkD
ep.
var.
of
coin
t. r
eg.
ln B
ee
f
Tab
le 5
.7.
Res
ult
s of
Wes
terl
un
d’s
seco
nd-
gen
erat
ion
pan
el c
oin
tegr
atio
n
test
s
Not
e:
***
, **,
* d
enot
e si
gnif
ican
ce a
t 1
%,
5%
and
10%
, re
spec
tive
ly.
Tes
t re
gres
sion
is
fitt
ed w
ith
a co
nsta
nt a
nd o
ne l
ag a
nd
lead
wit
h th
e ke
rnel
ban
dw
idth
bei
ng
set
acco
rdin
g t
o t
he r
ule 4
(T/1
00)2/
9 . p
-val
ues
are
rep
orte
d in
par
enth
eses
. T
he p
-val
ues
are
for
a o
ne-s
ided
tes
t ba
sed
on 2
000
bo
ots
trap
rep
lica
tion
s. N
ull
hypo
thes
is:
no c
oin
tegr
atio
n.
71
In order to assess the robustness of our findings, we also implemented the panel
cointegration test proposed by Westerlund and Edgerton (2007). Unlike the other panel
data cointegration tests, here the null hypothesis is now cointegration. This test relies on
the popular Lagrange multiplier test of McCoskey and Kao (1998), and permits
correlation to be accommodated both within and between the individual cross-sectional
units. The test is also robust to unknown heterogeneous breaks in the intercept and/or
the slope of the cointegrating regression. In addition, the bootstrap suggested by
Westerlund and Edgerton (2007) is based on the sieve-sampling scheme23, and has the
appealing advantage of significantly reducing the distortions of the asymptotic test.
The results reported in Table 5.8 for a model including a constant and a trend clearly
indicate the cointegrating relationship between pork and beef prices in the EU since the
null hypothesis of cointegration is always accepted. This result is not modified if one
refers to the bootstrap critical values compared to asymptotic ones. Results also confirm
that the cointegrating relationship is robust to potential level and regime shifts.
The last panel cointegration test applied is the Johansen-Fisher type panel cointegration
test developed by Maddala and Wu (1999). They use Fisher’s result to propose an
alternative approach to test for cointegration in panel data by combining tests from
individual cross-sections to obtain a test statistic for the full panel. The Johansen-Fisher
panel cointegration test is a panel version of the individual Johansen cointegration test.
The Johansen-Fisher panel cointegration test is based on the aggregates of the p-values
of the individual Johansen maximum eigenvalues and trace statistic.
The results from Johansen-Fisher panel cointegration test are presented in Table 5.9. We
used the Akaike Information Criterion (AIC) and the Schwarz Information Criterion
(SIC) to determine the optimal lag length. We perform the cointegration test with both a
constant without trend and a constant with trend.
23 Sampling method proposed by Rietveld (1978), a simple and practical strategy for selecting line items with PPS (probability to proportional to line item size).
72
EU
15
EU1
5EU
25
EU
15
EU2
5E
U2
7EU
15
EU
15
EU2
5E
U1
5E
U2
5EU
27
2/1
99
5-6
/20
14
1/2
00
5-6
/20
14
1/2
00
5-6
/20
141
/20
07-
6/2
01
41
/20
07
-6/2
01
41
/20
07-
6/2
01
42
/19
95
-6/2
01
41
/20
05-
6/2
01
41
/20
05
-6/2
01
41
/20
07
-6/2
014
1/2
00
7-6
/20
14
1/2
00
7-6
/20
14
LM-s
tat
1.0
54
1.3
47
1.5
48
2.0
12
1.7
58
1.9
64
2.0
42
2.1
87
1.8
38
2.5
73
2.2
25
2.3
49
asym
pto
tic
p-va
lue
(0.5
15)
(0.3
86
)(0
.21
1)
(0.2
51
)(0
.18
7)
(0.1
98)
(0.3
35
)(0
.30
9)(0
.22
7)
(0.2
64
)(0
.34
9)
(0.3
76
)
boo
tsta
p p
-va
lue
(0.3
19)
(0.2
04
)(0
.25
1)
(0.1
28
)(0
.15
7)
(0.1
41)
(0.1
02
)(0
.08
2)(0
.12
6)
(0.0
66
)(0
.11
6)
(0.0
84
)
Dep
. va
r. o
f co
int.
reg
. ln
Be
ef
Dep
. var
. o
f co
int.
reg
. ln
Po
rk
Tab
le 5
.8.
Res
ult
s of
sec
ond-
gen
erat
ion
pa
nel
coi
nte
grat
ion
tes
t pr
opos
ed b
y W
este
rlu
nd
and
Edg
erto
n
Note
: *
**,
**,
* d
enot
e si
gnif
ican
ce a
t 1%
, 5%
and
10%
, re
spec
tive
ly.
Tes
t re
gres
sion
is
fitt
ed w
ith
a co
nsta
nt a
nd o
ne l
ag a
nd l
ead
wit
h th
e ke
rnel
ba
ndw
idth
bei
ng s
et a
ccord
ing
to
th
e ru
le
4(T
/100
)2/9 .
p-v
alue
s ar
e re
por
ted
in p
aren
thes
es.
The
p-v
alue
s ar
e fo
r a
one-
side
d te
st b
ased
on
2000
bo
ots
trap
rep
lica
tions
. The
Wes
terl
und
an
d E
dge
rton
(2
007)
tes
ts a
re p
erfo
rmed
usi
ng t
he S
tata
“xt
wes
t” c
om
man
d. N
ull
hyp
othe
sis:
co
inte
grat
ion.
Tab
le 5
.9. R
esu
lts
from
th
e Jo
ha
nse
n-F
ish
er p
anel
coi
nte
grat
ion
tes
t
Not
e:
***,
**,
* de
note
sig
nifi
can
ce a
t 1%
, 5
% a
nd 1
0%,
resp
ecti
vely
. p
-val
ues
are
rep
orte
d i
n pa
rent
hese
s. S
elec
tion
of
lags
bas
ed o
n A
IC a
nd
SIC
. N
ull
hyp
oth
esis
: no
coi
nteg
rati
on.
73
Based on the maximal eigenvalue statistics, we reject the null hypothesis of no
cointegrating relationship. Results from the trace statistics also support the finding of
maximal eigenvalue statistics where there is rejection towards null hypothesis of r=0
and r≤1, respectively. In a nutshell, based on the statistical results of the Johansen-
Fisher panel cointegration test, there is sufficient evidence to conclude the existence of
the long run relationship between the pork and beef prices.
In summary, the results of panel cointegration tests are generally highly significant
which gives strong evidence that the variables have a long run relationship.
5.3 Estimation results
The cointegration tests are only able to indicate whether or not the variables are
cointegrated and if a long-run relationship exists between them. Since they do not
indicate the direction of causality, the long-run equilibrium coefficients should be
estimated by using appropriate estimators.
When panels of data are available, there exist a number of alternative estimation
methods that vary on the extent to which they account for parameter heterogeneity. The
dynamics of traditional estimators is simply pooled and treated as homogeneous. Only
the intercepts are allowed to differ cross countries. Early and prominent examples
include fixed effects (FE), random effects (RE), and generalized methods of moments
(GMM). These methods are typically focused on solving the problem of fixed effect
heterogeneity in the case of large N and small T panels; whereas they are not designed
to correct for the endogeneity induced by the latent heterogeneity. Pesaran and Smith
(1995) show that the traditional procedures for the estimation of pooled models can
produce inconsistent and potentially misleading estimates of the lagged dependent
variable’s parameter in dynamic panel data models if latent heterogeneity is present.
In this study, we utilize panel-based vector error correction models (VECM) including
MG (Mean Group) and PMG (Pooled Mean Group) estimation methods. The MG
estimator allows for complete (short-run and long-run) parameter heterogeneity across
panel cross-sections. The PMG estimator is an intermediate case between the averaging
and pooling methods of estimations.24
24 OLS estimators are super-consistent in the case of co-integrated variables, but they are based on strong homogeneity assumptions among countries by imposing single slope coefficient in pooled estimation, which is inappropriate for this study regarding potential country heterogeneity. This is the reason for using MG and PMG estimators instead of traditional panel techniques.
74
The first technique, MG, introduced by Pesaran and Smith, (1995) calls for estimating
separate regressions for each country and calculating the coefficients as unweighted
means of the estimated coefficients for the individual countries. It allows for all
coefficients to vary and be heterogeneous in the long-run and short-run. Pesaran and
Smith showed that the MG method produces consistent estimates of the average of the
parameters when the time-series dimension of the data is sufficiently large. However,
for small N the average estimators (MG) in this approach are quite sensitive to outliers
and small model permutations (Favara, 2003).
The MG estimator does not take into account that some economic conditions tend to be
common across countries in the long run. The PMG estimator does because it combines
both pooling and averaging. The long run coefficients are constrained to be the same
across sections, while the intercepts, short run coefficients, and error variances are
allowed to differ. The PMG estimator is believed to offer the best compromise between
consistency and efficiency, because one would expect the long-term growth path to be
driven by a similar process across EU member states while the short-term dynamics
around the long-term equilibrium path differ because of idiosyncratic member state
features and shocks to fundamentals. The PMG estimator is appropriate when data have
complex country-specific short-term dynamics which cannot be captured imposing the
same lag structure on all countries. This estimator combines the properties of efficiency
of the pooled dynamic estimators while avoiding the inconsistency problem deriving
from slope heterogeneity. Moreover, since the PMG estimator does not impose any
restriction on short–term coefficients, it provides important information on country
specific values of the speed of convergence towards the long-run relationship linking
pork and beef price series.
The variables are I(1) and cointegrated, which means that the error term is I(0) process
for all i. A principal feature of cointegrated variables is their responsiveness to any
deviation from long-run equilibrium. This feature implies an error correction model in
which the short-run dynamics of the variables in the system are influenced by the
deviation from equilibrium. Since the model is nonlinear in the parameters, Pesaran,
Shin, and Smith (1999) develop a maximum likelihood method to estimate the
parameters.
The error-correction model to be estimated is given by the following equations: ��������� =
���������,��� − ����������� + ∑ ��������� ∆�������,��� + ∑ ���
������ ∆�������,��� +
�� + ��� (5.1)
75
��������� = �′��������,��� − �′���������� + ∑ �′�������� ∆�������,��� +
∑ �′�������� ∆�������,��� + �′� + �′�� (5.2)
where the number of groups i=1,2,…,N and the number of periods t=1,2,…,T. δij are the
coefficient vectors, �ij are scalars and μi is the group specific effect. θi is the vector
which contains the long-run relationships between the variables and the parameter φi is
the error-correcting speed of adjustment term. The long run coefficients are estimated
using the Maximum Likelihood (ML) estimation
The long-run relationship imposes under the null hypothesis the condition: φi = 0 for all
i. This hypothesis means that there is no long-run stable relationship between the
independent variable and the dependent variable in the model. The decision rule says
that when the error correction term (ECT) is negative and significant, the null
hypothesis of no causality would be rejected.
ECT is a measure of the extent by which the observed values in time t-1 deviate from
the long-run equilibrium relationship. Since the variables are cointegrated, any such
deviation at time t-1 should induce changes in the values of the variables in the next
time point, in an attempt to force the variables back to the long-run equilibrium
relationship.
In practice, the MG and PMG procedure involves first estimating autoregressive
distributed lag (ARDL) models separately for each country i. In a series of papers,
Pesaran and Smith (1995), Pesaran (1997), and Pesaran and Shin (1999) show that one
can use the ARDL approach to produce consistent and efficient estimates of the
parameters in a long-run relationship between both integrated and stationary variables,
and to conduct inference on these parameters using standard tests. The main require-
ments for the validity of this methodology are that, first, there exists a long-run
relationship among the variables of interest and, second, the dynamic specification of
the model is sufficiently augmented so that the regressors become weakly exogenous
and the resulting residual is serially uncorrelated25.
Tables 5.10 and 5.11 report MG and PMG estimates of the ECM. We have already
revealed that there is bi-directional relationship between pork and beef prices in the EU.
Thus, the dependent variable in the estimation results of Table 5.10 is lnbeef and
25 Augmenting the model with lags addresses the potential endogeneity of remittances. In this respect, Pesaran (1997) and Pesaran and Shin (1999) show that for inference on the long-run parameters, sufficient augmentation of the order of the autoregressive distributive lag model can simultaneously correct for the problem of residual serial correlation and endogenous regressors.
76
correspondingly the dependent variable is lnpork in Table 5.11 results. Lags are chosen
on the basis of AIC and are allowed to vary across countries. We report results both with
and without trend. A constant country-specific term is included. Natural logarithms are
used in order to obtain directly the elasticities.
We analyze the stability of our results over time by re-estimating the models for EU15
countries in two sub periods (1/2005-6/2014 and 1/2007-6/2014) and for EU25
countries in one subperiod (1/2007-6/2014). This is also a check for cross-sectional
stability: we can compare if new member state distort the estimation results.
77
tren
dn
o t
ren
dtr
end
no
tre
nd
tre
nd
no
tre
nd
tre
nd
no
tre
nd
tren
dn
o t
ren
dtr
end
no
tre
nd
tren
dn
o t
ren
dtr
end
no
tre
nd
tren
dn
o t
ren
dtr
end
no
tre
nd
tren
dn
o t
ren
dtr
end
no
tre
nd
Lon
g R
un
ln p
ork
0.25
10.
253
0.28
20.
288
0.21
30.
217
0.42
30.
426
0.4
090.
413
0.3
920.
394
0.1
930.
197
0.2
150.
219
0.14
70.
155
0.31
50.
331
0.28
10.
308
0.30
10.
310
(10.
385
)***
(12.
127)
***
(16.
291
)***
(18.
754)
***
(6.3
72)*
**(6
.479
)***
(29.
490)
***
(30.
572
)***
(27.
477)
***
(28.
582
)***
(25.
115)
***
(25.
626
)***
(16.
476)
***
(17.
554
)***
(19.
889)
***
(21.
216
)***
(10.
124
)***
(12.
397)
***
(33.
262
)***
(36.
661)
***
(32.
910
)***
(35.
090)
***
(29.
431
)***
(33.
995)
***
tim
e tr
en
d-0
.001
0.00
2-0
.002
0.00
1-0
.002
-0.0
01-0
.002
-0.0
01-0
.001
-0.0
02-0
.001
-0.0
03
(-2.
714)
**(2
.217
)**
(-2.
456)
**(1
.694
)(-
1.8
80)
(-1.
755)
(-2.
113)
*(-
1.68
6)(-
1.95
8)*
(-1.
224)
(-1
.517
)(-
2.43
7)**
Sho
rt R
un
ECT
(ad
j. s
pee
d)
-0.0
40-0
.042
-0.0
61-0
.065
-0.0
55-0
.057
-0.0
98-0
.099
-0.0
85-0
.085
-0.1
01-0
.105
-0.0
35-0
.038
-0.0
57-0
.06
-0.0
47-0
.054
-0.0
86-0
.093
-0.0
71-0
.073
-0.0
82-0
.088
(-1
1.11
8)**
*(-1
3.5
46)*
**(-
15.6
43)*
**(-
17.1
19)*
**(-
14.5
54)*
**(-
15.7
25)*
**(-
20.4
39)*
**(-
21.4
77)*
**(-
16.9
86)*
**(-
18.
442)
***(
-22.
971
)***
(-24
.811
)***
(-8
.530
)***
(-9.
158)
***(
-12.
227
)***
(-12
.972
)***
(-13
.114
)***
(-13
.785
)***
(-17
.218
)***
(-19
.076
)***
(-1
4.96
2)**
*(-1
5.9
21)*
**(-
22.1
24)*
**(-
21.8
50)*
**
Δ ln
po
rk-0
.003
-0.0
03-0
.010
-0.0
110.
006
0.0
070.
050
0.05
10.
061
0.06
20.
036
0.03
8-0
.002
-0.0
03-0
.017
-0.0
19-0
.005
-0.0
05-0
.037
-0.0
47-0
.072
-0.0
74-0
.055
-0.0
68
(-1.
828)
*(-
1.8
59)*
(-1.
231
)(-
1.26
1)(2
.290
)**
(2.4
29)*
*(3
.154
)**
(3.1
57)*
*(3
.793
)***
(3.8
46)*
**(2
.208
)**
(2.2
11)*
*(-
1.48
5)(-
1.5
56)
(-1.
237)
(-1.
281
)(-
2.02
8)*
(-2.
123
)*(-
2.17
3)*
(-2
.445
)**
(-3.
906
)***
(-3.
949)
***
(-2.
682)
**(-
3.0
19)*
**
Δ ln
po
rkt-
1-0
.001
-0.0
01-0
.002
-0.0
02-0
.001
-0.0
010.
015
0.01
70.
023
0.02
30.
009
0.00
9-0
.001
-0.0
01-0
.001
-0.0
020.
004
0.00
50.
008
0.0
100.
007
0.0
070.
003
0.0
05
(-0
.944
)(-
0.95
4)(-
1.1
07)
(-1.
136)
(-0.
753
)(-
0.76
1)(2
.685
)**
(2.6
82)*
*(2
.508
)**
(2.5
18)*
*(1
.957
)*(1
.974
)*(-
0.66
2)(-
0.7
33)
(-0.
940)
(-1.
019
)(1
.750
)(1
.828
)(1
.938
)*(2
.181
)*(-
1.8
74)
(1.9
55)*
(1.4
78)
(1.6
10)
Δ ln
po
rkt-
20.
001
0.0
010.
002
0.0
030.
001
0.0
010.
001
0.00
10.
002
0.00
30.
001
0.00
1-0
.001
-0.0
020.
002
0.00
32-0
.001
-0.0
01-0
.001
-0.0
01-0
.001
-0.0
01-0
.001
-0.0
01
(0.4
09)
(0.4
01)
(0.8
07)
(0.8
73)
(0.5
45)
(0.5
55)
(1.2
24)
(1.2
39)
(1.5
87)
(1.6
26)
(0.3
71)
(0.3
90)
(-0.
274)
(0.3
59)
(0.6
52)
(0.6
95)
(-0.
286)
(-0.
334)
(-0.
655)
(-0.
680
)(-
0.7
17)
(-0.
771)
(-0.
423
)(-
0.46
7)
Δ ln
be
ef t
-10.
617
0.6
210.
692
0.6
950.
448
0.4
510.
523
0.52
40.
438
0.43
80.
644
0.65
50.
549
0.55
50.
576
0.60
60.
367
0.37
20.
432
0.4
440.
608
0.6
150.
584
0.5
88
(3.0
56)*
*(3
.172
)**
(3.4
08)*
*(3
.422
)**
(2.6
37)*
*(2
.676
)**
(3.6
65)*
**(3
.712
)***
(3.3
55)*
**(3
.363
)***
(5.6
54)*
**(5
.690
)***
(4.2
63)*
**(4
.407
)***
(2.5
56)*
*(2
.752
)**
(2.4
37)*
*(2
.481
)**
(2.9
81)*
*(3
.117
)**
(4.2
58)*
**(4
.336
)***
(3.7
15)*
**(3
.743
)***
Δ ln
be
ef t
-20.
005
0.0
050.
006
0.0
070.
004
0.0
050.
002
0.00
20.
007
0.00
80.
005
0.00
50.
002
0.00
20.
004
0.00
50.
002
0.00
20.
001
0.0
030.
003
0.0
030.
002
0.0
04
(1.0
23)
(1.0
89)
(1.1
89)
(1.2
29)
(1.0
61)
(1.1
30)
(1.0
03)
(1.0
18)
(0.7
26)
(0.7
64)
(0.8
08)
(0.8
10)
(0.6
86)
(0.8
37)
(0.9
54)
(0.9
81)
(0.9
30)
(0.9
75)
(1.5
66)
(1.6
18)
(1.2
28)
(1.2
62)
(0.7
15)
(0.7
34)
Ob
serv
atio
ns
3315
3315
1530
1530
2550
2550
1170
1170
1950
1950
2106
2106
3315
3315
1530
1530
2550
2550
1170
011
7019
5019
5021
0621
06
Log
like
lih
oo
d-5
14.6
54-5
55.2
65-3
28.6
47-3
44.3
80-2
75.5
81-3
02.0
88-1
89.6
18-1
71.2
82-1
48.7
23-1
59.3
3519
6.73
6-2
19.4
81-5
89.2
44-5
25.4
67-3
69.4
13-3
91.5
82-3
37.9
11-3
15.3
90-1
73.9
30-1
78.5
10-1
56.9
77-1
57.6
5-2
39.2
92-2
46.6
29
Hau
sman
te
st0.
221
0.1
950.
089
0.1
020.
242
0.2
550.
075
0.07
10.
152
0.15
30.
179
0.18
5
(0.6
40)
(0.6
28)
(0.7
78)
(0.7
25)
(0.5
62)
(0.5
28)
(0.8
98)
(0.9
10)
(0.6
79)
(0.6
76)
(0.6
42)
(0.6
59)
EU25
EU27
1/20
07-6
/201
41/
2007
-6/2
014
EU25
EU27
1/20
07-6
/201
41/
2007
-6/2
014
EU25
1/20
05-6
/201
4
EU15
1/20
07-6
/201
4
MG
PM
G
EU15
1/20
05-6
/201
4
EU25
1/20
05-6
/201
4
EU15
1/20
07-6
/201
4
EU15
2/1
995-
6/2
014
EU15
2/19
95-6
/201
4
EU15
1/20
05-6
/201
4
Note
: *
**,
**,
* d
enot
e si
gni
fica
nce
at 1
%,
5% a
nd 1
0%,
resp
ecti
vely
. Aut
om
atic
sel
ecti
on
of
lags
bas
ed o
n A
kaik
e in
form
atio
n cr
iter
ion.
t-s
tati
stic
are
re
port
ed i
n p
aren
thes
es.
The
Hau
sman
tes
t is
a t
est
of p
ool
abil
ity
of t
he l
ong
-run
coe
ffic
ient
(i.
e. o
f th
e re
stri
ctio
n th
at a
ll c
ount
ries
hav
e th
e sa
me
long
-ru
n el
asti
city
). A
ll e
qua
tio
ns
incl
ude
a co
nsta
nt c
ount
ry-s
peci
fic
term
.
Tab
le 5
.10.
MG
an
d P
MG
reg
ress
ion
res
ult
s, d
epen
den
t va
riab
le:
lnbe
ef
78
tre
nd
no
tre
nd
tre
nd
no
tre
nd
tre
nd
no
tre
nd
tre
nd
no
tre
nd
tren
dn
o t
ren
dtr
end
no
tre
nd
tren
dn
o t
ren
dtr
en
dn
o t
ren
dtr
en
dn
o t
ren
dtr
en
dn
o t
ren
dtr
en
dn
o t
ren
dtr
end
no
tre
nd
Lon
g R
un
ln b
ee
f0.
091
0.09
20.
108
0.1
140.
067
0.0
710.
194
0.1
960.
136
0.1
420.
217
0.22
20.
066
0.06
70.
08
0.0
830.
042
0.0
460.
119
0.1
220.
099
0.1
030.
105
0.1
08
(1.7
90)
(1.8
48)
(1.7
36)
(1.8
12)
(1.3
04)
(1.5
16)
(3.7
75)*
**(3
.888
)***
(2.7
46)*
*(2
.851
)**
(3.8
82)*
**(4
.219
)***
(1.5
76)
(1.6
17)
(1.6
55)
(1.6
90)
(1.0
77)
(1.1
45)
(5.3
61)*
**(5
.552
)***
(3.1
83)*
*(3
.652
)***
(5.1
66)*
**(5
.648
)***
tim
e t
ren
d-0
.001
-0.0
02-0
.002
-0.0
01-0
.001
-0.0
03-0
.003
-0.0
03-0
.002
-0.0
02-0
.004
-0.0
06
(-1.
462)
(-1.
276
)(-
1.6
37)
(-1.
446
)(-
1.2
85)
(-2.
483
)**
(-1.
919
)*(-
2.0
27)*
(-1.
954
)*(-
1.2
28)
(-2.
734
)**
(-4.
203
)**
Sho
rt R
un
ECT
(ad
j. s
pee
d)
-0.0
24-0
.026
-0.0
45-0
.048
-0.0
33-0
.037
-0.0
53-0
.059
-0.0
72-0
.075
-0.0
81-0
.082
-0.0
2-0
.021
-0.0
38-0
.041
-0.0
27-0
.030
-0.0
51-0
.054
-0.0
66-0
.067
-0.0
74-0
.076
(-1.
480)
(-1.
507
)(-
2.72
6)**
(-2.
812
)**
(-1.
619
)(-
1.71
9)(-
4.1
20)*
**(-
4.27
5)**
*(-
6.7
33)*
**(-
6.84
6)**
*(-
8.1
91)*
**(-
8.2
19)*
**(-
1.22
3)(-
1.2
54)
(-1.
648)
(-1.
706
)(-
1.5
65)
(-1.
587)
(-2.
380)
**(-
2.5
82)*
*(-
2.8
28)*
*(-
2.90
1)**
(-5.
317
)***
(-5.
356
)***
Δ ln
be
ef
0.0
010.
001
0.0
020.
002
0.0
030.
004
0.02
30.
025
0.02
70.
027
0.0
280.
028
0.0
010.
001
0.0
030.
004
0.0
050.
006
0.00
50.
005
0.00
60.
007
0.0
070.
008
(0.6
92)
(0.7
44)
(1.0
12)
(1.0
78)
(1.5
99)
(1.6
23)
(2.8
80)*
*(2
.919
)**
(3.0
26)*
*(3
.047
)**
(3.2
19)*
*(3
.248
)**
(0.4
90)
(0.5
34)
(0.8
73)
(0.9
19)
(1.7
27)
(1.7
74)
(1.8
28)
(1.8
30)
(1.2
85)
(1.3
66)
(1.4
69)
(1.5
90)
Δ ln
be
ef t
-1-0
.001
-0.0
01-0
.001
-0.0
01-0
.001
-0.0
01-0
.008
-0.0
09-0
.01
-0.0
11-0
.004
-0.0
05-0
.001
-0.0
01-0
.001
-0.0
02-0
.002
-0.0
03-0
.005
-0.0
07-0
.008
-0.0
09-0
.007
-0.0
07
(-0.
755)
(-0.
785
)(-
0.5
67)
(-0.
582)
(-0.
753
)(-
0.76
0)(-
1.6
36)
(-1.
648)
(-2.
031)
*(-
2.1
14)*
(-1.
617)
(-1.
646
)(-
0.66
7)(-
0.7
38)
(-0.
944)
(-1.
010
)(-
1.7
59)
(-1.
828)
(-1.
953
)*(-
2.0
07)*
(-2.
186)
*(-
2.2
24)*
(-1.
955)
*(-
1.9
39)*
Δ ln
be
ef t
-20.
001
0.00
10.
001
0.0
010.
001
0.0
010.
001
0.0
010.
002
0.0
020.
002
0.00
20.
001
0.00
10.
001
0.0
010.
001
0.0
010.
001
0.0
010.
001
0.0
020.
003
0.0
03
(0.3
67)
(0.3
80)
(0.5
13)
(0.5
52)
(0.6
29)
(0.6
55)
(0.8
38)
(0.8
89)
(1.4
12)
(1.4
20)
(1.4
88)
(1.4
95)
(0.2
76)
(0.2
76)
(0.3
37)
(0.3
54)
(0.5
91)
(0.6
08)
(0.9
63)
(0.9
47)
(1.1
74)
(1.2
27)
(1.6
62)
(1.6
90)
Δ ln
po
rkt-
10.
559
0.56
70.
537
0.5
420.
396
0.4
060.
471
0.4
810.
522
0.5
290.
492
0.49
70.
518
0.52
00.
555
0.5
890.
351
0.3
540.
454
0.4
620.
479
0.4
830.
467
0.4
76
(5.5
39)*
**(5
.563
)***
(6.0
76)*
**(6
.137
)***
(4.1
64)*
*(4
.242
)**
(4.4
18)*
**(4
.555
)***
(5.4
18)*
**(5
.476
)***
(4.7
14)*
**(4
.661
)***
(6.0
30)*
**(6
.105
)***
(5.9
29)*
**(5
.959
)***
(4.8
18)*
**(4
.880
)***
(5.2
43)*
**(5
.316
)***
(5.6
97)*
**(5
.731
)***
(4.6
19)*
**(4
.641
***)
Δ ln
po
rkt-
20.
003
0.00
30.
004
0.0
030.
001
0.0
030.
002
0.0
020.
005
0.0
060.
006
0.00
60.
002
0.00
20.
003
0.0
030.
002
0.0
020.
001
0.0
020.
003
0.0
040.
003
0.0
04
(0.6
64)
(0.7
34)
(1.3
40)
(1.2
53)
(0.4
17)
(0.9
36)
(0.7
05)
(0.8
18)
(1.4
18)
(1.5
67)
(1.6
45)
(1.6
81)
(0.4
73)
(0.5
28)
(0.7
16)
(0.7
80)
(0.6
31)
(0.7
42)
(0.2
38)
(0.5
53)
(0.7
17)
(0.9
14)
(1.1
15)
(1.1
65)
Ob
serv
atio
ns
3315
3315
1530
1530
2550
2550
1170
1170
1950
1950
2106
2106
3315
3315
1530
1530
2550
2550
1170
011
7019
5019
5021
0621
06
Log
like
lih
oo
d-3
96.7
51-4
11.2
59-2
52.6
74-2
70.1
25-1
98.6
47-2
13.3
46-1
45.7
82-1
61.8
38-1
40.2
38-1
58.6
39-2
01.9
61-2
22.3
77-4
10.6
93-4
33.5
84-2
95.6
42-3
16.1
98-3
37.5
25-3
54.6
12-1
68.7
67-1
73.9
73-1
55.9
51-1
71.3
60-2
07.6
91-2
19.6
34
Hau
sman
te
st0.
071
0.07
50.
042
0.0
430.
035
0.0
380.
055
0.0
590.
092
0.0
950.
101
0.10
5
(0.9
10)
(0.9
02)
(0.9
58)
(0.9
55)
(0.9
72)
(0.9
68)
(0.9
29)
(0.9
22)
(0.7
95)
(0.7
83)
(0.7
30)
(0.7
22)
1/2
005-
6/2
014
1/2
007-
6/2
014
EU25
EU27
1/20
07-6
/201
41/
2007
-6/2
014
EU25
EU15
2/1
995-
6/20
14
EU25
1/2
005-
6/2
014
1/2
005-
6/2
014
1/20
07-6
/201
4
EU27
1/20
07-6
/201
41/
2007
-6/2
014
EU15
EU15
EU25
EU15
EU15
2/1
995-
6/20
141/
200
5-6/
2014
EU15
MG
PM
G
Tab
le 5
.11.
MG
an
d P
MG
reg
ress
ion
res
ult
s, d
epen
den
t va
riab
le:
lnpo
rk
Note
: *
**,
**,
* de
note
sig
nif
ican
ce a
t 1%
, 5%
and
10%
, re
spec
tive
ly. A
uto
mat
ic s
elec
tio
n o
f la
gs b
ased
on
Aka
ike
info
rmat
ion
crit
erio
n. t
-st
atis
tic
are
rep
orte
d i
n p
aren
thes
es.
The
Hau
sman
tes
t is
a t
est
of
poo
labi
lity
of
the
long
-run
coe
ffic
ient
(i.
e. o
f th
e re
stri
ctio
n t
hat
all
coun
trie
s ha
ve
the
sam
e lo
ng-r
un
elas
tici
ty).
All
equ
atio
ns i
ncl
ude
a co
nsta
nt c
oun
try-
spec
ific
ter
m.
79
Overall, our baseline results in Table 5.10 seem to be relatively robust: there is long run
equilibrium between pork and beef prices. Pork price is a significant factor affecting
beef price in the EU in long run. Since variables are defined in logarithmic terms, the
estimated coefficients are directly the price elasticities of beef price with respect to pork
price. All MG and PMG regressions results in Table 5.10 show a significant and positive
long run coefficient. On the contrary, the magnitude of the long-term coefficient is
affected by the estimation method. By choosing the MG rather than the PMG method
slightly higher values are obtained. We employ the Hausman test of long-run
homogeneity restriction to choose the appropriate estimator. Namely, homogeneity of
long run coefficients implied by PMG estimating procedure cannot be assumed a priori
but needs to be tested. When long-run homogeneity restriction can be accepted, PMG
estimates would be more efficient compared to MG. Respectively, when the slope
coefficients are heterogeneous, then PMG estimates would be inconsistent and the MG
estimator provides consistent estimates of the mean of long-run coefficients. According
to test results shown at the bottom of Table 5.10, we cannot reject the null of long-run
homogeneity restriction, implying that the PMG estimator is efficient under the null
hypothesis and is preferred over the MG estimator.
We can now concentrate on the estimation results of PMG estimation method. The PMG
estimates find strong evidence of positive relationship between the EU pork and beef
prices. According to the results, there exists cross-commodity price transmission from
pork prices to beef prices in the EU. All PMG regressions results in Table 5.11 show a
significant and positive coefficient of 0.15 to 0.33 in the long run, implying that a 1%
increase in pork prices, ceteris paribus, would raise beef prices 0.15-0.33%. According
to the results of PMG estimation the relationship between the two price series is the
strongest and, thus, the price transmission is the most significant in the old member
states, EU15 countries. Also, in the case of EU15, we notice that the price transmission
is more apparent and the coefficients are higher when we investigate estimation results
from subperiod 1/2007-6/2014 compared to estimation results from subperiod 1/2005-
6/2014 or from the full period 2/1995-6/2014. This indicates that integration of the EU
livestock market of pork and beef has increased during the investigation period. This is
also valid when analyzing price transmission and market integration between pork and
beef in the EU25 countries: the significance of long run coefficient is higher and the
long run price elasticity measured by the estimated long run coefficient is larger in
estimation results from subperiod 1/2007-6/2014 than in estimation results from the full
period 1/2005-6/2014.
The estimated error-correcting speed adjustment term ECT is, as expected, significantly
negative and less than 1 in absolute value, implying convergence towards long run
equilibrium relationship. The lowest value of the error correction coefficient in the PMG
estimation results is 0.035 and the highest is 0.093, implying a speed of adjustment of
80
about 1-2 years. The speed of adjustment has also improved during the investigation
period; In the case of EU15 countries, the adjustment term is 0.056 (trend)/0.060 (no
trend) in the estimation results from the period 1/2005-6/2014 and 0.086 (trend) 0.093
(no trend) in the results from the period 1/2007-6/2014. This means that after the year
2005 the convergence to the equilibrium has sped up from 1.5 years to 1 year. Price
transmission between the EU meat markets has fastened significantly but is still quite
slow. This indicates that there exist some rigidities that decelerates the price transmis-
sion process.
Furthermore, we can note that in short run the variation in pork prices has affected beef
prices only after 2007. This means that short run relationship between pork and beef
prices has only existed since 2007. Moreover, short term price transmission from pork
price to beef price in the EU is interestingly negative according to our PMG estimation
results. This means that increase in pork price lower the beef price in short term.
However, this short term effect is minor although statistically significant.
By repeating the estimation of the model for EU15 in period 1/2005-6/2014 and for
EU15 and EU25 in period 1/2007-6/2014 we also check the sensitivity of the results at
the type of countries included in the panel; i.e. we check what kind of influence the new
member states have had on the results. We notice that the panel estimates indicate cross-
sectional stability over results. The inclusion of the new member states to the data has
not changed the interpretation of the estimation results.
In table 5.11 we repeat the estimation but now we specify pork as a dependent variable
and beef as an independent variable. Consequently, we are interested in the price
elasticities of pork with respect to beef price. The long-run slope homogeneity
hypothesis of PMG is tested via the Hausman test. On the basis of the Hausman test it is
not possible to reject the hypothesis of poolability of the long-run elasticity of the beef
price and we can therefore focus on the results of PMG estimation method. All in all,
MG estimation method provides larger coefficients than PMG method. According to the
results of PMG estimation, there exists cross-commodity price transmission in long run
from beef prices to pork prices in the EU only in the estimation results from subperiod
1/2007-6/2014. This can be interpreted as overall increased integration and efficiency in
livestock market in the EU during the past ten years; not only the variation of pork price
will transmit to the prices of other meat types but also vice versa. There is bi-directional
relationship between pork and beef prices in the EU in long run although price
transmission coming from the fluctuations in pork price is larger and more significant.
Notably, this bi-directional relationship has not been valid until 2007 in our investiga-
tion. In table 5.11 the estimation results from subperiod 1/2007-6/2014 show a
significant and positive coefficient of 0.099 to 0.122 in the long run, implying that a 1%
increase in beef price ceteris paribus, would raise pork price 0.1-0.12% in the EU.
81
The error correction terms are negative and significant; hence the null hypothesis of no
long-run relationship is rejected. According to the results of PMG estimation, the speed
of adjustment from the deviation in the long run relationship between the beef and pork
prices is between -0.051 and -0.076. The model implies moderate adjustment inertia; it
converges to the equilibrium, with 5-7.5% percent of discrepancy corrected in each
period. No significant price transmission from beef price to pork price is found in short
run in the EU livestock market. Overall, this suggests that short-run dynamics is
significantly different in livestock market compared to long run dynamics.
5.4 Robustness check against estimation method
Next, we use alternative heterogeneous panel cointegration techniques that correct for
endogeneity, namely Fully Modified OLS (FMOLS) and Dynamic OLS (DOLS). We
use FMOLS and DOLS between-dimension estimators (Group Mean Estimator)
proposed by Pedroni (2001). Unlike the PMG, which uses a maximum likelihood
method, FMOLS and DOLS are based on a modified OLS26. FMOLS and DOLS
estimators allow us to relax the assumption of long-run homogeneity. Both methods
allow for regressors’ complete endogeneity, and treat all parameters, i.e. dynamics and
cointegrating vectors, as heterogeneous across panel members. FMOLS runs a static
OLS with fixed effects for each panel member individually and uses the estimated
residuals to (non-parametrically) build member-specific adjustment terms, which are
then used to correct for each member’s endogeneity. Differences in regressors are used
as internal instruments. Instead, DOLS individually corrects for endogeneity
parametrically by running OLS with fixed effects for each panel member including
leads and lags of differenced regressors.
It is worth to note that FMOLS and DOLS are static models and thus, we can only
estimate long term relationship between pork price and beef price in the EU member
states.
The DOLS estimates employ two lags and two leads. Overall, the results in Table 5.12
are robust with respect to the choice of the lag structure. A country-specific constant has
been incorporated while a time trend was not included.
26 It is worth noting that under endogeneity and no cointegration, using OLS produces first-order bias, and external instruments are needed. However, under endogeneity and cointegration, OLS produces superconsistent estimates and second-order endogeneity bias (i.e. inconsistent estimates of standard errors). Internal instruments are then used (such as in FMOLS and DOLS).
82
These results of DOLS and FMOLS estimates confirm the existence of a long-run
relationship between pork and beef prices in the EU countries. All the long run
coefficients obtained by applying FMOLS and DOLS have the same signs although
higher magnitudes as the previous PMG results, suggesting equilibrium derived from
the PMG method are roughly equal to those from FMOLS and DOLS. The results from
FMOLS and DOLS estimation also confirm the observation that price transmission and
market integration has increased between pork market and beef market in the EU during
FMO
LSD
OLS
FMO
LSD
OLS
FMO
LSD
OLS
FMO
LSD
OLS
FMO
LSD
OLS
FMO
LSD
OLS
Ind
epe
nd
en
t va
ria
ble
: ln
po
rk
ln b
eef
0.1
640
.203
0.1
49
0.1
65
0.1
32
0.1
720
.333
0.3
610
.311
0.3
570
.37
50
.39
0
(7.4
53
)***
(9.0
48)*
**(6
.761
)***
(8.8
19)*
**(6
.520
)**
*(7
.476
)**
*(1
3.0
62)*
**(
14
.281
)***
(12
.373
)***
(12.
98
2)*
**(1
5.2
07
)***
(15.
916
)***
Ob
serv
ati
ons
331
53
315
153
01
530
25
50
25
50
117
01
170
195
01
950
210
62
106
Ind
epe
nd
en
t va
ria
ble
: ln
be
ef
ln p
ork
0
.109
0.1
310
.12
70
.15
60
.11
90
.123
0.2
880
.312
0.1
890
.216
0.2
31
0.2
68
(2.4
59
)**
(3.1
09)*
**(3
.061
)***
(3.7
95)*
**(2
.227
)**
*(2
.258
)**
*(7
.136
)***
(7.7
88
)***
(5.1
61
)***
(5.9
07
)***
(6.3
44)*
**(6
.765
)***
Ob
serv
ati
ons
331
53
315
153
01
530
25
50
25
50
117
01
170
195
01
950
210
62
106
1/2
007-
6/2
014
EU1
5E
U1
5EU
25
EU1
5EU
25
EU2
7
2/1
995-
6/2
014
1/2
005-
6/2
014
1/2
005
-6/2
01
41
/200
7-6
/201
41
/200
7-6
/201
4
Not
e:
***,
**,
* d
eno
te s
igni
fica
nce
at 1
%, 5
% a
nd 1
0%, r
espe
ctiv
ely.
t-st
atis
tic
are
rep
orte
d i
n pa
rent
hese
s.
Tab
le 5
.12.
FM
OL
S a
nd
DO
LS
reg
ress
ion
res
ult
s
83
the observation period. In addition these estimation results confirm the cross-sectional
stability: results are not alert the type of countries included in the panel (EU15, EU25 or
EU27).
5.5 Robustness check against definitions of the data sample
In this section, the robustness of the relation between pork and beef price series is
checked against alternative definitions of the data sample. We address the following
questions. How is the relation changed over time? Are there countries with a
significantly different behavior?
We first check the stability of our results over time via PMG estimation. The entire
estimation period of 2/1995-6/2014 is divided into 4 subperiods (2/1995-12/1999,
1/2000-12/2004, 1/2005-12/2009, 1/2010-6/214) and estimation is repeated for every
subperiod including EU15 member states. These estimation results are presented in
Appendix 2.
The estimation results in Tables 5.10 and 5.11 already revealed that integration in the
EU livestock market of pork and beef has increased during the investigation period. The
purpose of this robustness check is to get a better picture of this development and also
examine the meat market integration and cross-commodity price transmission between
pork and beef during the early stage of the EU.
The estimation results show that the long-term elasticity and the speed adjustment term
have changed substantially between the subperiods considered. Overall, estimations
suggest a significantly different and higher relation between price series for the latter
subperiods. We can observe that EU meat market was diverged before year 2005. It
cannot be found statistically significant relation between pork and beef prices during the
periods 2/1995-12/1999 and 1/2000-12/2004. Since 2005, however, EU meat market
integration has increased sharply and price transmission between meat markets has
become significant. This is true in both cases: when we have lnpork as an independent
variable and when we have lnbeef as an independent variable. This is clear evidence that
meat market in Europe is well functioned and efficient and the direction of the EU
livestock market is towards stronger integration.
Next, we proceed by studying robustness of our results across countries. The findings
from our estimation might be affected also by the relative small number of countries in
the data sample. As a further robustness check we have re-estimated the model via PMG
estimation method excluding from the data one country at a time. This permits to
84
understand whether the results are strongly driven the behavior of a single country.
Figures in Appendix 3 plot the value of the long-run elasticity and speed adjustment
term on the country excluded from the sample.
According to the data availability, there are figures for EU15 member states, EU25
member states (excl. EU15) and EU27 member states (excl. EU25). In the first set of
figures independent variable is lnbeef. In the latter set lnpork is treated as an
independent variable.
In the first figures, EU15 countries that appear to influence significantly on the
estimation of the long-run elasticity are Denmark, Germany, Greece, Netherlands and
Portugal. When Denmark, Germany, and Netherlands are excluded from the sample, the
estimated elasticity is significantly lower: values of the long-run elasticity falling
outside the 95% confidence band estimated from the whole sample. This indicates that
the presence of these countries contributes to keep high the value of the long-run
elasticity estimated on the whole sample. Furthermore, these results can also be
interpreted that Denmark, Germany and Netherlands have the most significant influence
on the meat market integration and meat price transmission in the EU. Without these
countries EU meat market would not be as integrated and as efficient. They are also the
leaders in the EU pork market. According to the results, Germany is the leader when we
consider effectiveness of the meat market measured by the speed of price transmission
(responses are the fastest)27 but Denmark is the price leader which has the greatest
impact on the cross-commodity price transmission in the EU meat market (responses are
the largest)28.
Respectively, Greece and Portugal appears to decrease significantly the value estimated
across the whole panel of EU15 member states: the exclusion leads to an estimated
elasticity of about 0.22, while the estimated elasticity of the whole panel data sample is
0.197. This also means that Greece and Portugal have negative effect on meat market
integration in the EU and their inclusion reduces effectiveness of the EU meat market.
When we focus on new member states, inclusion of Cyprus, Slovakia, Slovenia and
Bulgaria has also statistically significant negative influence on transmission between
pork and beef prices and on the meat market integration in the EU.
When considering the speed of adjustment terms the values of variables come back to
the long-run equilibrium level Denmark and Germany have the most significant
influence on the meat market integration. When they are excluded from the data sample,
27 The estimated speed of adjustment term is the lowest when Germany is excluded from the data. 28 The estimated long-run elasticity is the lowest when Denmark is excluded from the data.
85
the estimated speed of adjustment term is statistically significantly lower. Greece has
opposite influence. In the figures on the speed of adjustment terms covering EU25
member states (excl. EU15), Latvia and Lithuania have statistically significant negative
effect on the meat market integration. When they are excluded from the sample, the
estimated speed of adjustment term is significantly higher.
In the latter set of figures, lnpork is treated as an independent variable. According to the
results, EU15 countries influencing statistically significantly on the estimation results of
the long-run elasticity are France, Germany, Greece, Ireland, and the UK. Presence of
France, Germany, Ireland and the UK contributes to keep high the value of the long-run
elasticity estimated on the whole sample. When they are one at time excluded from the
data, the estimated coefficients of long-run elasticity fall outside the 95% confidence
band. Without these countries price transmission from beef to pork in the EU would not
be as large and as efficient. They appear to be the leaders in the EU beef market.
Presence of Greece in the data sample affects another direction: Greece has negative
effect on meat market integration in the EU and its inclusion reduces effectiveness of
the EU meat market. When we focus on the new member states, inclusion of Slovakia
and Slovenia has statistically significant negative influence on price transmission from
beef to pork and on the meat market integration among the EU25 member states.
When lnpork is an independent variable and lnbeef is an explanatory variable, the
estimated speed of adjustment term is significantly higher when Greece, Cyprus,
Solavakia, Slovenia and Bulgaria are one by one excluded from the data set. France,
Germany, Ireland and the UK have instead opposite influence and they have statistically
significant positive effect on the meat market integration when we are considering price
transmission from beef prices to pork prices.
All in all, the impacts, though significant, appear to be quite moderate and they are not
strong enough to alter qualitative results. Moreover, they do not alter considerably
quantitative results. We may conclude that our results are robust against estimation
method and also against data sample definition.
86
6 CONCLUDING REMARKS
The main objective of the study is to analyze the dynamics of price transmission and
market integration in the EU livestock market of pork and beef. Our focus is on
horizontal cross-commodity price transmission which is not as common approach as
vertical or spatial price transmission. However, it provides us interesting information on
the connection between commodity prices and integration of commodity markets.
We utilize recently developed panel time-series techniques to analyze the price linkages
in the EU meat sector. Our data consists of monthly data on pork and beef prices in the
EU member countries during the period from February 1995 to June 2014.
The results indicate that there exists bi-directional relationship between pork and beef
prices in the EU in long run. According to the estimation results price transmission is
the most significant in the old member states, EU15 countries. Price transmission
coming from the fluctuations in pork price is larger and more significant than price
transmission coming from the fluctuations in beef price. Depending on investigation
period, 1% increase in pork prices, ceteris paribus, would raise beef prices 0.15-0.33%.
Long run price transmission from beef price to pork price can be observed only in the
results from subperiod 1/2007-6/2014. These results imply that a 1% increase in beef
price ceteris paribus, would raise pork price 0.1-0.12% in the EU.
Cross-commodity price transmission between pork and beef has increased remarkably
during the past ten years in the EU15 member states. It cannot be found statistically
significant relation between pork and beef prices before year 2005. However, ever since
EU meat market integration has increased sharply and price transmission between meat
markets has become significant. This is also valid when analyzing price transmission
and market integration between pork and beef in the EU25 and EU27 countries. This
indicates that integration of the EU livestock market of pork and beef has increased
during the investigation period.
Also the convergence to the equilibrium has sped up. In the case of EU15 countries after
the year 2005 the convergence to the equilibrium has sped up from roughly 1.5 year to 1
year. Although the equilibrium is achieved faster than before, the convergence process is
still quite slow. Price transmission from pork to beef in the EU livestock market is
87
strong in magnitude but there still exist some rigidities that decelerates the process on
the whole.
In short run, we found evidence only for price transmission from pork prices to beef
prices in the EU, not vice versa. Short run relationship between pork and beef prices has
only existed since 2007. Moreover, according to the estimation results, short term price
transmission from pork price to beef price is interestingly negative. This means that
increase in pork price lower the beef price in short term. Overall, short-run dynamics is
significantly different in the EU livestock market compared to long run dynamics. This
can partly be explained by different time lags in the adaptation of markets. The
inclusion of the new EU member states to the data has not changed the interpretation of
the estimation results.
We also checked whether the results are strongly driven the performance of a single
country. The results indicate that Danish, German and Dutch pork markets and Irish,
French, German and British beef markets have the most significant influence on the
meat market integration and meat price transmission in the EU. Without these countries
EU meat market would not be as integrated and as efficient. Denmark, Germany and
Netherlands are also the leaders in the EU pork market. Similarly, Ireland, France,
Germany and the UK appear to be the leaders in the EU beef market. Countries that
have negative effect on meat market integration in the EU are most clearly Greece,
Cyprus, Slovakia, Slovenia and Bulgaria. However, the impacts caused by the
significance of meat market of a single country, appear to be moderate – though
statistically significant – and they do not alter our main results.
Based on our empirical results we can conclude that the EU livestock market is
interactional and integrated. The meat market integration in the EU has proceeded
rapidly in recent years. Reduction of barriers to the trade of agricultural products has
made the integration possible. Furthermore, it is still important to ensure that regulation
measures of the Union treat all member state equally: prevent the barriers to trade
between the member countries, harmonize the conditions of competition and monitor
that common rules are widely abided by as they stand.
Today, operational environment of the EU meat market is characterized by rapid and
significant price fluctuations. The major price fluctuations are affecting increasingly
operation of the meat market. The integration of meat market has intensified at the same
time as the price fluctuations have generalized. Consequently, this means that the
market integration has promoted the price fluctuations to spread between product
markets. As the market and price risks grow, managing them is crucial in the EU
agriculture policy.
88
From the point of view of the market integration, the EU meat market is well-functioned
and effective. The impacts of the policy measures go faster and faster through the whole
meat market. Due to the price integration, targeted policy measures for the one branch
of the meat market has impact on the other meat market branches too. Policy decisions
concerning the EU meat market must be designed by the perspective of the performance
of the whole meat market.
89
REFERENCES
Aguiar, D. and Santana, J. A. (2002): Asymmetry in Farm to Retail Price Transmission: Evidence for Brazil. Agribusiness, 18, 37-48.
Alderman, H. (1993): Intercommodity Price Transmittal: Analysis of Food Markets in
Ghana. Oxford Bulletin of Economics and Statistics, 55, 1, 43-64. Arellano, M. and Bond, S. R. (1991): Some tests of specification for panel data: Monte
Carlo evidence and an application to employment equations. Review of Eco-nomic Studies, 58, 277-297.
Arshad, F. M. and Hameed, A. A. (2009): The Long Run Relationship between
Petroleum and Cereals Prices: Evidence From Cointegration Tests. Global Economy & Finance Journal, 2, 2, 91-100.
Asche, F., Guttormsen, A. G., Sebulonsen, T. and Sissener, E. H. (2005): Competition
between farmed and wild salmon: the Japanese salmon market. Agricultural Economics, 33, 333-340.
Baffes, J. (2007): Oil spills on other commodities. Resources Policy, 32, 126-134. Bai, J. and Ng, S. (2002): Determining the number of factors in approximate factor
models. Econometrica, 70, 191-221. Bakhat, M. and Wuerzburg, K. (2013): Price relationships of crude oil and food
commodities. Working Paper FA06/2013, Economics for Energy, Vigo. Baltagi, B. H., Song, S. H., Jung, B. C. and Koh, W. (2007): Testing for serial
correlation, spatial autocorrelation and random effects using panel data. Journal of Econometrics, 140, 5-51.
Banerjee, A. (1999): Panel Data Unit Roots and Cointegration: An Overview. Oxford
Bulletin of Economics and Statistics, 61, 1, 607-629. Banerjee, A. and Carrion-i-Silvestre, J. (2006): Cointegration in panel data with breaks
and cross-section dependence. ECB Working Paper 591. Benerjee, A., Marcellino, M. and Osbat, C. (2004): Some Cautions on the Use of Panel
Methods for Integrated Series of Macro-Economic Data. Econometrics Journal, 7, 2, 322-340.
Boetel, B. L. and Liu, D. J. (2008): Incorporating Structural Changes in Agricultural and
Food Price Analysis: An Application to the U.S. Beef and Pork Sectors. Work-ing Papers 44076, Food Industry Center, University of Minnesota.
90
Breitung, J. (2000): The Local Power of Some Unit Root Tests for Panel Data. In Baltagi, B. (ed.): Nonstationary Panels, Panel Cointegration, and Dynamic Panels. Advances in Econometrics, 15, 161-178.
Breitung, J. and Pesaran, M. H. (2005): Unit Roots and Cointegration in Panels.
Cambridge Working Papers, Econometrics 0535, University of Cambridge. Carraro, A. and Stefani, G. (2010): Vertical price transmission in Italian agri-food
chains: does the transmission elasticity change always or sometimes? DIPSA - University of Florence, XLVII convegno SIDEA.
Chang, Y. (2003): Nonlinear IV Panel Unit Root Tests. Working Papers 2003-06, Rice
University, Department of Economics. Choi, I. (2001): Unit Root Tests for Panel Data, Journal of International Money and
Finance, 20, 249-272. Ciaian, P. and Kancs, D. (2011): Interdependencies in the Energy-Bioenergy-Food Price
Systems: A Cointegration Analysis. Resource and Energy Economics, 33, 326-348.
Dyck, J. H. and Nelson, K. E. (2003): Structure of the global market for meat.
Agricultural Economic Report, No. 785, Economic Research Service, US De-partment of Agriculture, Washington D.C.
Eberhardt, M. (2012): Estimating panel time-series models with heterogeneous slopes.
Stata Journal, 12, 1, 61-71. Esposti, R. and Listorti, G. (2012): Horizontal Price Transmission in Agricultural
Markets: Fundamental Concepts and Open Empirical Issues. Bio-based and Applied Economics, 1, 1, 81-108.
Esposti, R. and Listorti, G. (2013): Agricultural price transmission across space and
commodities during price bubbles. Agricultural Economics, 44, 1, 125-139. Fackler, P. and Goodwin, B. K. (2001): Spatial Market Integration. In Rausser, G. and
Gardner, B. (eds): Handbook of Agricultural Economics. Elsevier Publishing, Amsterdam.
Favara, G. (2003): An Empirical Reassessment of the Relationship between Finance and
Growth. IMF Working Paper 03/123. Fisher, R. A. (1932): Statistical Methods for Research Workers. Oliver and Boyd, 4th
ed., Edinburgh. Fousekis, P. (2007): Multiple Markets in the EU? Empirical Evidence from Pork and
Poultry Prices in 14 EU Member Countries. Economics Bulletin, 3, 65, 1-12. Frey, G. and Manera, M. (2005): Econometric models of asymmetric price transmission.
Working Papers, 100. Fondazione Eni Enrico Mattei.
91
Gengenbach, C., Palm, F. C. and Urbain, J-P. (2010): Panel Unit Root Tests in the Presence of Cross-Sectional Dependencies: Comparison and Implications for Modelling. Econometric Reviews, Taylor & Francis Journals, 29, 2, 111-145.
Gengenbach, C., Palm, F. and Urbain, J. (2006): Cointegration Testing in Panels with
Common Factors. Oxford Bulletin of Economics and Statistics, 68, 683-719. Goodwin, B. K. and Harper, D. C. (2000): Price transmission, threshold behavior, and
Asymmetric Adjustment in the US pork sector. Journal of Agricultural and Ap-plied Economics, 32, 543-553.
Goodwin, B. K. and Holt, M. T. (1999): Price Transmission and Asymmetric
Adjustment in the U.S. Beef Sector. American Journal of Agricultural Econom-ics, 81, 630-637.
Granger, C. W. J. (1969): Investigating causal relation by econometric and cross-
sectional method. Econometrica, 37, 424-438. Gutierrez, L. (2005): Panel Unit Roots Tests for Cross-Sectionally Correlated Panels: A
Monte Carlo Comparison. Oxford Bulletin of Economics and Statistics, De-partment of Economics, University of Oxford, 68, 4, 519-540.
Hassouneh, I., Serra, T., Goodwin, B. K. and Gil, J. M. (2012): Non-Parametric and
Parametric Modeling of Biodiesel, Sunflower Oil, and Crude Oil Price Rela-tionships. Energy Economics, 34, 5, 1507-1513.
Holtz-Eakin, D., Newey, W. and Rosen, H. S. (1988): Estimating Vector Autoregres-
sions with Panel data. Econometrica, 56, 6, 1371-1395. Hurlin C. (2005): Un Test Simple de l.Hypothèse de Non Causalité dans un Modèle de
Panel Hétérogène, Revue Economique, 56, 3, 799-809. Hurlin, C. and Venet, B. (2001): Granger Causality Tests in Panel Data Models with
Fixed Coefficients. Miméo, University Paris IX. Hutlin, C. (2007): Testing Granger Non-Causality in Heterogeneous Panel Data Models
with Fixed Coefficients. Working Papers 1547, Orleans Economic Laboratorys, University of Orleans.
Ihle, R., Brümmer, B. and Thompson, S. R. (2012): Structural Change in European Calf
Markets: Decoupling and the Blue Tongue Disease. European Review of Agri-cultural Economics, 39, 1, 157-180.
Im, K. S., Pesaran, M. H. and Shin, Y. (2003): Testing for Unit Roots in Heterogeneous
Panels. Journal of Econometrics, 115, 53-74. Im, K.-S., Lee, J. and Tieslau, M. (2005): Panel LM Unit-Root Tests with Level Shifts.
Oxford Bulletin of Economics and Statistics, 67, 393-419.
92
Im, K.-S., Lee, J. and Tieslau, M. (2010): Stationarity of Inflation: Evidence from Panel Unit Root Tests with Trend Shifts. 20th Annual Meetings of the Midwest Econometrics Group, Oct 1-2, 2010.
Jang, M. J. and Shin, D. (2005): Comparison of panel unit root tests under cross-
sectional dependence. Economics Letters, 89, 12-17. Johansen S. (1991): Estimation and hypothesis testing of cointegration vectors in
Gaussian vector autoregressive models. Econometrica, 59, 1551-1580. Johansen, S. (1996): Likelihood-based inference in cointegrated vector auto-regressive
models. Oxford University Press, Oxford. Johansen, S. (1988): Statistical analysis of cointegration vectors. Journal of Economic
Dynamics and Control, 12, 231-254. Reprinted in Engle, R. F. and Granger, C. W. J. (eds.): Long-Run Economic Relationships. Oxford University Press, Ox-ford, 131-152.
Kao, C. (1999): Spurious Regression and Residual Based Tests for Cointegration in
Panel Data. Journal of Econometrics, 90, 1-44. Kao, C. and Chiang, M. H. (2000): On the estimation and inference of a cointegrated
regression in panel data. Advances in Econometrics, 15, 179-222. Kónya, L. (2006): Exports and growth: granger causality analysis on OECD countries
with a panel data approach. Economic Modelling, 23, 978-992. Kristoufek, L., Janda, K. and Zilberman, D. (2012): Correlations between biofuels and
related commodities before and during the food crisis: A taxonomy perspective. Energy Economics, 34, 1380-1391.
Lee, J. and Strazicich, M. C. (2009): LM Unit Root Tests with Trend Breaks at
Unknown Dates. Mimeo. Lee, J. and Strazicich, M. C. (2003): Minimum LM unit root test with two structural
breaks. Review of Economics and Statistics, 85, 4, 1082-1089. Lee, J. and Strazicich, M. C. (2004): Minimum LM Unit root test with one structural
break. Working Paper, Department of Economics, Appalachain State Universi-ty.
Levin, A., Lin, C. F. and Chu, C. J. (2002): Unit root tests in panel data: asymptotic and
finite-sample properties. Journal of Econometrics, 108, 1-24. Liu, X. (2011): Horizontal price transmission of the Finnish meat sector with major EU
players. MTT Discussion Papers 1/2011, MTT Economic Research, Finland. Lumsdaine, R. L and. Papell, D. H. (1997): Multiple trend breaks and the unit root
hypothesis. Review of Economics and Statistics, 79, 2, 212-218.
93
Maddala, G. S. and Wu, S. (1999): A comparative study of unit root tests with panel data and new simple test. Oxford Bullertin of Economics and Statistics, Special issue, 631-652.
McCorriston, S., and Sheldon, I. M. (1996): Trade Policy in Vertically-Related Markets.
Oxford Economic Papers, 48, 664-672. McCoskey, S. and Kao, C. (1998): A Residual-based Test of the Null of Cointegration in
Panel Data. Econometric Reviews, 17, 57-84. Meyer, J. and von Cramon-Taubadel, S. (2004): Asymmetric Price Transmission: A
survey. Journal of Agricultural Economy, 55, 3, 581-611. Meyer, S. (2012): Estimation of the Long-Run Food Price Equilibrium in Germany, the
U.S. and Europe. Dissertation, Georg-August-Universität Göttingen. Moon, H. R. and Perron, B. (2004): Testing for a unit root in panels with dynamic
factors. Journal of Econometrics, 122, 81-126. Moon, H. R., Perron, B. and Phillips, P. C. B. (2005): Incidental Trends and the Power
of Panel Unit Root Tests. Cowles Foundation Discussion Paper No. 1435; USC CLEO Research Paper No. C03-17; IEPR Working Paper No. 05-38.
Muhammad, A. and Kebede, E. (2009): The Emergence of an Agro-Energy Sector: Is
Agriculture Importing Instability from the Oil Sector? Choices, 24, 12-15. Newey, W. and West, K. (1987): A Simple, Positive Semi-Definite, Heteroscedastic and
Autocorrelation Consistent Covariance Matrix. Econometric, 55, 703-708. Nielsen, M., Setala, J., Laitinen, J., Saarni, K., Virtanen, J. and Honkanen, A. (2007):
Market integration of farmed trout in Germany. Marine Resource Economics, 22, 2, 195-213.
O`Connell, P. (1998): The Overvaluation of Purchasing Power Parity. Journal of
International Economics, 44, 1-19. Osborn, D. R. (1993): Discussion: Seasonal Cointegration. Journal of Econometrics, 55,
299-303. Pedroni, P. (1999): Critical Values for Cointegration Tests in Heterogeneous Panels with
Multiple Regressors. Oxford Bulletin of Economics and Statistics, 61, 653-670. Pedroni, P. (2000): Fully modified OLS for heterogeneous cointegrated panels. In
Baltagi, B. H. (ed.): Nonstationary panels, cointegration in panels and dynamic panels. Elsevier Publishing, Amsterdam.
Pedroni, P. (2004): Panel Cointegration. Asymptotic and Finite Sample Properties of
Pooled Time Series Tests with an Application to the PPP Hypothesis. Econo-metric Theory, 20, 597-625.
94
Peltzman, S. (2000): Prices rise faster than they fall. Journal of Political Economy, 108, 3, 466-502.
Peri, M., Baldi, L. and Vandone, D. (2013): Price discovery in commodity markets.
Applied Economics Letters, 20, 397-403. Perron, P. (1989): The great crash, the oil price shock, and the unit root hypothesis.
Econometrica, 57, 6, 1361-1401. Perron, P. (1990): Testing for a unit root in a time series with a changing mean. Journal
of Business & Economic Statistics, 8, 2, 153-62. Perron, P. (1997): Further evidence on breaking trend functions in macroeconomic
variables. Journal of Econometrics, 80, 2, 355-385. Pesaran, H. (1997): The Role of Econometric Theory in Modelling the Long Run.
Economic Journal, 107, 440, 178-191. Pesaran, H. and Shin, Y. (1999): An Autoregressive Distributed Lag Modelling
Approach to Cointegration. In Strøm (ed.): Econometrics and Economic Theo-ry in the 20th Century; the Ragnar Frisch Centennial Symposium. Chapter 4, 371-413, Cambridge University Press.
Pesaran, H. M. (2003): A Simple Panel Unit Root Test in the Presence of Cross Section
Dependence. Mimeo, University of Southern California. Pesaran, M. H. (2007): A simple panel unit root test in the presence of cross section
dependence. Journal of Applied Econometrics, 27, 265-312. Pesaran, M. H. and Smith, R. P. (1995): Estimating long-run relationships from dynamic
heterogeneous panels. Journal of Econometrics, 68, 1, 79-113. Pesaran, M. H., Shin, Y. and Smith, R. P. (1999): Pooled mean group estimation of
dynamic heterogeneous panels. Journal of the American Statistical Association, 94, 289-326.
Phillips, P. C. B. and Sul, D. (2003): Dynamic panel estimation and homogeneity testing
under cross section dependence. Econometrics Journal, 6, 217-259. Pirotte, A. (1999): Convergence of the static estimation toward the long run effects of
dynamic panel data models. Economics Letters, 63, 2, 151-158. Ploberger, W. and Phillips, P. C. B. (2002): Optimal Testing for Unit Roots in Panel
Data, Mimeo. Quah, D. (1994): Exploiting cross-section variation for unit root inference in dynamic
data. Economics Letters, 44, 9-19. Rashid, S. (2011): Inter commodity price transmission and food price policies: An
analysis of Ethiopian cereal markets. Discussion Papers 1079, International Food Policy Research Institute.
95
Rietveld, M. R (1978): A new method for estimating the regression coefficients in the formula relating solar radiation to sunshine. Agricultural Meteorology, 19, 243-252.
Saadi, H. (2011): Price Co-movements in International Markets and Their Impacts on
Price Dynamics. In Piot-Lepetit, I. and M’Barek, R. (eds.): Methods to Analyse Agricultural Commodity Price Volatility. Berlin, Springer Verlag, 149-164.
Saghaian, S. H. (2010): The impact of the oil sector on commodity prices: correlation or
causation? Journal of Agricultural and Applied Economics, 42, 477-485. Sanjuán, A. I, and Gil, J. M. (2001): Price transmission analysis: a flexible methodolog-
ical approach applied to European pork and lamb markets. Applied Economics, 33, 123-131.
Serra, T., Gil, J. M. and Goodwin, B. K. (2006): Local Polynomial Fitting and Spatial
Price Relationships: Price Transmission in EU Pork Markets. European Review of Agricultural Economics, 33, 415-436.
Serra, T., Zilberman, D., Gil, J. M. and Goodwin, B. K. (2010): Nonlinearities in the US
Ethanol-Crude Oil Price System. Agricultural Economics, 42, 1, 35-45. Sharma, A. K. and Kumar, S. (2011): Effect of Working Capital Management on Firm
Profitability: Empirical Evidence from India. Global Business Review, 12, 1, 159-173.
Strauss, J. and Yigit, T. M. (2003): Shortfalls of Panel Unit Root Testing. Economics
Letters, 81, 3, 309-313. Timmer, P. (2009): Did Speculation Affect World Rice Prices? ESA Working Paper
No.07. Vavra, P. and Goodwin, B. K. (2005): Analysis of Price Transmission along the Food
Chain. OECD Food, Agriculture and Fisheries Working Papers, No. 3. OECD Publishing.
Westerlund, J. (2005): New Simple Tests for Panel Cointegration. Econometric
Reviews, 24, 297-316. Westerlund, J. (2007): Testing for Error Correction in Panel Data. Oxford Bulletin of
Economics and Statistics, 69, 709-748. Westerlund, J. and Edgerton, D. (2007): Simple Tests for Cointegration in Dependent
Panels with Structural Breaks, Oxford Bulletin of Economics and Statistics, 70, 665-704.
Zivot, E. and Andrews, D. W. K. (1992): Further evidence on the great crash, the oil-
price shock, and the unit-root hypothesis. Journal of Business & Economic Statistics, 10, 3, 251-270.
96
Appendix 1. LM unit root test results for individual EU member states
assuming one structural breakpoint.
Note: ***, **, * denote significance at 1%, 5% and 10%, respectively.
Note: ***, **, * denote significance at 1%, 5% and 10%, respectively.
Univariate LM
Stat Optimal Lag Break Location
Univariate LM
Stat Optimal Lag Break Location
Univariate LM
Stat Optimal Lag Break Location
Austria -4.987** 6 10/2004 -6.451*** 4 3/2011 -6.623*** 3 3/2011
Belgium -6.212*** 5 12/2003 -7.397*** 3 9/2012 -8.055*** 3 9/2012
Denmark -8.758*** 3 2/2001 -10.113*** 3 10/2010 -12.438*** 3 10/2010
Finland -7.055*** 2 7/2001 -8.656*** 2 1/2011 -9.174*** 2 1/2011
France -6.826*** 8 4/2003 -7.904*** 6 7/2010 -7.843*** 5 7/2010
Germany -8.492*** 4 3/2001 -9.649*** 2 4/2011 -10.963*** 2 4/2011
Greece -3.367 7 5/2006 -4.007* 5 2/2013 -3.898* 4 2/2013
Ireland -5.813*** 5 8/2003 -4.937*** 5 5/2012 -5.674*** 5 5/2019
Italy -4.064* 5 6/2005 -4.711** 5 7/2012 -6.058*** 4 7/2012
Luxembourg -5.824*** 7 9/2003 -5.868*** 6 11/2012 -7.187*** 4 11/2012
Netherlands -9.142*** 3 6/2001 -10.674*** 4 3/2011 -10.008*** 4 3/2011
Portugal -3.548 6 5/1998 -4.216* 3 1/2013 -5.286*** 3 1/2013
Spain -5.371** 5 10/2000 -6.542*** 4 6/2012 -6.131*** 4 6/2012
Sweden -6.809*** 5 3/2002 -7.396*** 2 8/2011 -7.770*** 2 8/2011
United Kingdom -7.471*** 7 11/2001 -9.060*** 3 1/2012 -8.898*** 3 1/2012
Cyprus -5.112*** 6 12/2010 -5.973*** 5 12/2010
Czech Republic -6.087*** 7 12/2009 -6.244*** 5 12/2009
Estonia -6.596*** 5 4/2011 -6.111*** 4 4/2011
Hungary -3.859* 5 2/2011 -3.563 1 2/2011
Latvia -3.635* 4 9/2009 -4.502** 4 9/2009
Lithuania -4.738** 3 1/2010 -5.562*** 3 1/2010
Poland -6.953*** 4 10/2010 -7.324*** 4 10/2010
Slovakia -5.044** 6 7/2010 -5.129** 5 7/2010
Slovenia -5.729*** 1 3/2011 -6.376*** 3 3/2011
Bulgaria -6.275*** 2 3/2012
Romania -4.996*** 3 5/2013
PORK
EU15 (2/1995-6/2014) EU25 (1/2005-6/2014) EU27 (1/2007-6/2014)
Univariate LM
Stat Optimal Lag Break Location
Univariate LM
Stat Optimal Lag Break Location
Univariate LM
Stat Optimal Lag Break Location
Austria -5.893*** 4 2/2000 -7.045*** 3 10/2010 -7.545*** 4 10/2010
Belgium -7.114*** 3 3/2003 -8.205*** 3 5/2011 -8.803*** 4 5/2011
Denmark -6.252*** 3 9/2001 -6.989*** 3 3/2013 -11.806*** 2 3/2013
Finland -5.387*** 5 2/2002 -6.364*** 3 1/2012 -10.532*** 3 1/2012
France -3.071 4 9/2009 -3.552 4 8/2012 -3.683 1 8/2012
Germany -7.646*** 3 6/2001 -7.898*** 4 5/2012 -8.199*** 4 5/2012
Greece -2.640 5 5/2005 -2.877 4 1/2007 -2.898 4 1/2007
Ireland -5.257*** 5 2/2004 -4.763** 5 6/2011 -5.094*** 5 6/2011
Italy -3.165 3 9/2003 -3.332 5 11/2011 -3.541 4 11/2011
Luxembourg -5.764*** 6 11/2003 -6.286*** 3 1/2013 -6.687*** 4 1/2013
Netherlands -7.259*** 2 2/2002 -7.648*** 2 9/2011 -8.190*** 3 9/2011
Portugal -5.863*** 3 3/2004 -5.905*** 2 2/2008 -5.673*** 5 2/2008
Spain -3.617 6 8/2004 -3.070 4 4/2013 -3.486 5 4/2013
Sweden -5.836*** 3 7/2007 -6.450*** 1 6/2011 -7.117*** 2 6/2011
United Kingdom -8.451*** 5 1/2008 -7.937*** 4 2/2012 -7.493*** 4 2/2012
Cyprus -6.491*** 5 9/2007 -6.832*** 6 9/2007
Czech Republic -7.514*** 5 10/2009 -8.091*** 4 10/2009
Estonia -5.284*** 5 2/2006 -5.333*** 2 11/2012
Hungary -2.525 5 1/2011 -2.661 3 1/2011
Latvia -3.061 6 11/2011 -2.864 4 11/2011
Lithuania -3.357 4 9/2010 -3.070 6 9/2010
Poland -5.153*** 4 11/2010 -5.486*** 5 11/2010
Slovakia -4.266** 6 1/2011 -4.962*** 5 1/2011
Slovenia -4.189** 5 12/2009 -4.221** 1 12/2009
Bulgaria -7.920*** 4 1/2010
Romania -5.873*** 3 9/2007
BEEF
EU15 (2/1995-6/2014) EU25 (1/2005-6/2014) EU27 (1/2007-6/2014)
97
Appendix 2. Pooled Mean Group ECM estimates, EU15 over different
sub-periods
Independent variable: lnbeef
Note: ***, **, * denote significance at 1%, 5% and 10%, respectively. Automatic selection of lags based on Akaike information criterion. t-statistic are reported in parentheses. All equations include a constant country-specific term.
Independent variable: lnpork
Note: ***, **, * denote significance at 1%, 5% and 10%, respectively. Automatic selection of lags based on Akaike information criterion. t-statistic are reported in parentheses. All equations include a constant country-specific term.
trend no trend trend no trend trend no trend trend no trend
Long Run
ln pork 0.121 0.128 0.145 0.153 0.288 0.296 0.315 0.324
(1.173) (1.228) (1.692) (1.809) (24.582)*** (25.074)*** (38.616)*** (39.445)***
time trend 0.001 0.002 0.003 0.005
(1.556) (2.724)** (3.883)*** (4.690)**
Short Run
ECT (adj. speed) 0.042 0.044 -0.068 -0.069 -0.183 -0.190 -0.204 -0.209
(0.791) (0.836) (-1.567) (-1.615) (-20.770)***(-22.181)***(-21.338)***(-21.919)***
Δ ln pork 0.002 0.002 0.001 0.002 0.022 0.022 0.025 0.027
(0.660) (0.691) (1.072 (1.109) (3.136)** (3.123)** (3.395)*** (3.417)***
Δ ln porkt-1 -0.001 -0.001 -0.001 -0.001 0.009 -0.008 0.010 0.012
(-0.553) (-0.573) (-0.817) (-0.829) (2.850)** (2.866)** (3.125)** (3.157)**
Δ ln porkt-2 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
(0.517) (0.506) (0.584) (0.582) (0.749) (0.720) (1.019) (1.033)
Δ ln beeft-1 0.456 0.462 0.604 0.611 0.531 0.535 0.489 0.492
(4.125)*** (4.201)*** (5.228)*** (5.321)*** (4.919)*** (5.074)*** (5.461)*** (5.547)***
Δ ln beeft-2 0.009 0.009 0.010 0.009 0.004 0.004 0.008 0.006
(2.081)* (2.063)* (2.314)** (2.276)** (1.871) (1.850) (1.929)* (1.818)
Observations 885 900 900 900 900 900 810 810
Log likelihood -102.541 -89.578 -125.673 -100.284 -222.548 -208.649 -267.590 -238.181
2/1995-12/1999 1/2000-12/2004 1/2005-12/2009 1/2010-6/2014
trend no trend trend no trend trend no trend trend no trend
Long Run
ln beef 0.051 0.053 0.072 0.076 0.109 0.112 0.137 0.142
(0.890) (0.932) (1.257) (1.295) (1.945)* (2.163)** (5.368)*** (5.559)***
time trend -0.005 -0.005 -0.002 -0.002
(-3.118)** (-2.926)** (-1.647) (-1.779)
Short Run
ECT (adj. speed) 0.014 0.016 -0.004 -0.004 -0.073 -0.077 -0.110 -0.112
(0.135) (0.191) (0.813) (0.866) (-4.614)*** (-4.658)*** (-9.229)*** (-9.289)***
Δ ln beef 0.001 0.001 0.001 0.001 0.010 0.009 0.017 0.020
(0.410) (0.448) (0.382) (0.402) (2.375)** (2.327)** (3.456)** (3.713)***
Δ ln beeft-1 -0.001 -0.001 -0.001 -0.001 -0.003 -0.003 -0.006 -0.008
(-0.103) (-0.125) (-0.187) (-0.218) (-0.591) (-0.644) (-1.226) (-1.232)
Δ ln beeft-2 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
(0.151) (0.108) (0.219) (0.252) (0.723) (0.764) (1.513) (1.567)
Δ ln porkt-1 0.701 0.708 0.722 0.731 0.687 0.689 0.743 0.752
(9.147)*** (9.306)*** (10.148)***(10.135)*** (8.990)*** (8.981)*** (9.357)*** (9.456)***
Δ ln porkt-2 0.005 0.005 0.004 0.004 0.002 0.003 0.004 0.004
(0.747) (0.793) (1.156) (1.188) (0.917) (0.944) (1.362) (1.381)
Observations 885 900 900 900 900 900 810 810
Log likelihood -77.890 -90.261 -82.696 -98.987 -131.254 -140.383 -144.229 -153.627
2/1995-12/1999 1/2000-12/2004 1/2005-12/2009 1/2010-6/2014
98
Appendix 3. Cross-sectional stabilities of the long-run elasticity and
the speed of adjustment
EU15 (period: 2/1995-6/2014)
lnbeef as an independent variable
Long-run elasticity
Speed of adjustment
0,12
0,14
0,16
0,18
0,20
0,22
0,24
Aust
ria
Belg
ium
Denm
ark
Fin
land
Fra
nce
Germ
any
Gre
ece
Irela
nd
Italy
Luxe
mbourg
Neth
erl
ands
Port
ug
al
Spain
Sw
ede
n
Unite
d K
ingdom
-0,042
-0,040
-0,038
-0,036
-0,034
-0,032
Au
str
ia
Be
lgiu
m
De
nm
ark
Fin
lan
d
Fra
nce
Ge
rma
ny
Gre
ece
Ire
lan
d
Ita
ly
Lu
xe
mb
ou
rg
Ne
the
rla
nd
s
Po
rtu
ga
l
Sp
ain
Sw
ed
en
Un
ite
d K
ing
do
m
99
EU25 (excl. EU15; period: 1/2005-6/2014)
lnbeef as an independent variable
Long-run elasticity
Speed of adjustment
0,10
0,12
0,14
0,16
0,18
0,20
Cyp
rus
Cze
ch R
epublic
Est
onia
Hunga
ry
Latv
ia
Lith
uania
Pola
nd
Slo
vaki
a
Slo
ven
ia
-0,060
-0,058
-0,056
-0,054
-0,052
-0,050
-0,048
Cyp
rus
Cze
ch
Re
pu
blic
Esto
nia
Hu
ng
ary
La
tvia
Lith
ua
nia
Po
lan
d
Slo
va
kia
Slo
ve
nia
100
EU27 (excl. EU25; period: 1/2007-6/2014)
lnbeef as an independent variable
Long-run elasticity
Speed of adjustment
0,20
0,22
0,24
0,26
0,28
0,30
0,32
0,34
Bulgaria Romania
-0,096
-0,094
-0,092
-0,090
-0,088
-0,086
-0,084
-0,082
-0,080
Bulgaria Romania
101
EU15 (period: 2/1995-6/2014)
lnpork as an independent variable
Long-run elasticity
Speed of adjustment
0,050
0,055
0,060
0,065
0,070
0,075
Aust
ria
Belg
ium
Denm
ark
Fin
land
Fra
nce
Germ
any
Gre
ece
Irela
nd
Italy
Luxe
mbourg
Neth
erl
ands
Port
ug
al
Spain
Sw
ede
n
Unite
d K
ingdom
-0,026
-0,024
-0,022
-0,02
-0,018
-0,016
Au
str
ia
Be
lgiu
m
De
nm
ark
Fin
lan
d
Fra
nce
Ge
rma
ny
Gre
ece
Ire
lan
d
Ita
ly
Lu
xe
mb
ou
rg
Ne
the
rlan
ds
Po
rtu
ga
l
Sp
ain
Sw
ed
en
Un
ite
d K
ing
do
m
102
EU25 (excl. EU15; period: 1/2005-6/2014)
lnpork as an independent variable
Long-run elasticity
Speed of adjustment
0,035
0,040
0,045
0,050
0,055
Cyp
rus
Cze
ch
Re
pu
blic
Esto
nia
Hu
ng
ary
La
tvia
Lith
ua
nia
Po
lan
d
Slo
va
kia
Slo
ve
nia
-0,034
-0,032
-0,03
-0,028
-0,026
Cyp
rus
Cze
ch
Re
pu
blic
Esto
nia
Hu
ng
ary
La
tvia
Lith
ua
nia
Po
lan
d
Slo
va
kia
Slo
ve
nia
103
EU27 (excl. EU25; period: 1/2007-6/2014)
lnpork as an independent variable
Long-run elasticity
Speed of adjustment
0,095
0,100
0,105
0,110
0,115
Bulgaria Romania
-0,090
-0,085
-0,080
-0,075
-0,070
-0,065
Bulgaria Romania
104
PTT julkaisuja, PTT publikationer, PTT publications 22. Hanna Karikallio. 2010. Dynamic Dividend Behaviour of Finnish Firms and
Dividend Decision under Dual Income Taxation 21. Satu Nivalainen. 2010. Essays on family migration and geographical mobility in
Finland 20. Terhi Latvala. 2009. Information, risk and trust in the food chain: Ex-ante valuation
of consumer willingness to pay for beef quality information using the contingent valuation method.
19. Perttu Pyykkönen. 2006. Factors affecting farmland prices in Finland PTT raportteja, PTT rapporter, PTT reports 250. Noro, K ja Lahtinen, M. 2015. Pohjoismainen asuntomarkkinaselvitys. 249. Holm, P., Hietala J. ja Härmälä, V. 2015. Liikenneverkko ja kansantalous –
Suomi–Ruotsi vertailua 248. Alho, E. – Noro, K. – Pyykkönen, P. 2014. Ruokakorista sijoitussalkkuun –
Näkemyksiä kotimaisesta ruokaketjusta sijoituskohteena. 247. Hietala, J., Alhola, K., Horne, P., Karvosenoja, N., Kauppi, S., Kosenius, A-K.,
Paunu, V-V., Seppälä, J. 2014. Kaivostoiminnan taloudellisten hyötyjen ja ympä-ristöhaittojen rahamääräinen arvottaminen.
246. Holm, P. ja Kerkelä, L. 2014. Voisiko Suomi seurata Ruotsin ja Norjan esimerkkiä? Näkökohtia perintö- ja lahjaverosta sekä luovutusvoittoverosta.
245. Kerkelä, L., Lahtinen, M., Esala, L., Kosunen, A. ja Noro, K. 2014. Suomen pitkän aikavälin energia- ja ilmastopolitiikka ja teollisuuden kilpailukyky.
244. Kosenius, A-K., Haltia, E., Horne, P., Kniivilä, M. and Saastamoinen O. 2013. Value of ecosystem services? Examples and experiences on forests, peatlands, agricultural lands, and freshwaters in Finland.
PTT työpapereita, PTT diskussionsunderlag, PTT Working Papers 169. Holappa, V., Huovari, J., Karikallio, H. ja Lahtinen, M. 2015. Alueellisten
asuntomarkkinoiden kehitys vuoteen 2017 167. Huovari, J. 2015. Päästökaupan epäsuorien kustannusten kompensaatio. 166. Peltoniemi, A., Arovuori, K., Karikallio, H., Niemi, J. ja Pyykkönen, P. 2014.
Viljasektorin hintarakenteet. 165. Kosenius, A-K., Tulla, T., Horne, P., Vanha-Majamaa I. ja Kerkelä, L. 2014.
Metsäpalojen torjunnan talous ja ekosysteemipalvelut ‒ Kustannusanalyysi Poh-jois-Karjalasta.
164. Hietala, J., Kosenius, A-K., Rämö, A-K. ja Horne, P. 2014. Metsätalouden taloudellinen tulos eri kasvatustavoissa.
163. Rämö, A-K., Kerkelä, L. ja Horne, P. 2014. Marjojen, sienten ja yrttien kaupallinen hyödyntäminen Pohjois-Karjalassa ja Kainuussa.
162. Kämäräinen, S., Rinta-Kiikka, S. ja Yrjölä, T. 2014. Maatilojen välinen yhteistyö Suomessa.