Price transmission in three Italian Food Chains: a structural break
approach
Carraro, A., Stefani G.
Dep. of Plant Soil and Environmental Sciences
University of Florence (Italy).
Paper prepared for presentation at the EAAE 2011 Congress
Change and Uncertainty
Challenges for Agriculture,
Food and Natural Resources
August 30 to September 2, 2011
ETH Zurich, Zurich, Switzerland
Copyright 2011 by A. Carraro and G. Stefani. All rights reserved. Readers may make
verbatim copies of this document for non-commercial purposes by any means, provided
that this copyright notice appears on all such copies.
Price transmission in three Italian food chains: a structural break
approach
Abstract
Recently a wide instability of food prices has been observed in world and European
agricultural and food markets. Both media and policy makers have dealt with the
unsatisfactory patterns of marketing margins and price transmission along the food chain
which may bring about distributive issues and affect inflationary trends. Although price
transmission and margins dynamics have attracted so much interest at the policy level,
few Italian studies deal with this topic.
Our aim is to provide a first analysis of the price transmission mechanism in three
Italian agri-food chains (lamb, pork and pasta), within a structural change framework.
Results show that structural breaks in the price transmission mechanism are an issue
in the food chain of pasta and pork with the regime change arising in occasion of the
price bubble of 2007-2008.
Keywords: price transmission, cointegration, structural breaks
JEL codes: Q13, L11.
Price transmission in three Italian food chains: a structural break
approach
1. Introduction
Recently a wide instability of food prices has been observed in world and European
agricultural and food markets. Both media and policy makers have dealt with this issue
highlighting the unsatisfactory patterns of marketing margins and price transmission
along the food chain. At European level, studies about price transmission were
commissioned by Institutions such as the European Parliament (Agra CEAS, 2007) or the
UK Department for Environment Food and Rural Affaires (London Economics, 2004)
concerned about the possible impacts of the new Common Agricultural Policy on
consumer prices. According to a recent Commission Communication (COM 2009(591))
“a better functioning food supply chain is crucial for consumers and for ensuring a
sustainable distribution of value added along the chain, thus contributing towards raising
its overall competitiveness”.
A related issues concerns the degree of pass through from raw commodity prices to
consumer food prices and its impact on inflationary (or deflationary) trends as food
account for about 20% of Euro area consumption (Ferrucci et al. 2010; National Bank of
Belgium, 2008)).
Italian Institutions have commissioned studies on this topic too. The Ministry for
Economic Development recently published a research on the dynamic of prices along the
wheat chain (IPI, 2008). The Italian Antitrust Authority (AGCM, 2007; Giangiulio and
Mazzantini, 2010), concerned about the presence of market power in the food chain,
launched an inquiry on the food retail sector to check for anti-competitive practices along
the marketing chain.
A theoretical question related to the above issue is about the nature and causes of the
observed patterns of price transmission along the food chain. Surprisingly, although price
transmission and margins dynamics have attracted so much interest at the policy level,
few Italian studies deal with this topic. Frey and Manera (2007) in a recent literature
reviewed list only 4 papers on Italian markets out of 64 studies, and they were about
gasoline.
Most of the literature on price transmission is about estimating elasticities and possibly
detecting the presence of asymmetries in the transmission mechanism. This field of
econometrics has witnessed during the 80’s the so called unit root revolution when the
concept of cointegration was introduced and applied to empirical studies on price
transmission (Meyer and Von Cramon Taubadel, 2004).
Only recently, the potential confounding between non stationarity of time series and
structural changes was highlighted (Boetel and Liu, 2008). However, few studies have
applied this framework to food prices so far. Non stationarity or lack of cointegration for
a number of agricultural price was questioned by Wang and Tomek (2007): once
structural breaks were accounted for, most series previously considered integrated of
order one turned to be stationary. Boetel and Liu (2008), which provide a brief review of
earlier studies, found evidence of cointegrating relationships along the pork and beef
chains in the U.S only after accounting for structural breaks. Adachi and Liu (2009)
identified several regimes in the Japanese pork retail-farm price relationship.
The recent food price instability urges applied economists to take into account structural
breaks in analysing long term price relationships in the food sector. Our aim is to provide
a first analysis of the price transmission mechanism in three Italian agri-food chains
(lamb, pork and pasta), within a structural change framework.
The paper is organized as follows. Section 2 presents a brief discussion of the recent
structural change methodologies employed in applied research. We considered in turn the
Zivot and Andrews (1992) test for stationarity, tests for cointegration (Gregory and
Hansen, 1996; Carrion y Silvestre, 2006) and procedures to test for presence of single or
multiple structural breaks and to identify breaks dates (Bai and Perron 2003).
Section 3 contains an empirical application to three Italian food chains price data. We
checked for stationarity of price series and, whenever applicable, we tested for
cointegration between the couples of price series. Finally, we analyzed the long run price
relationship within each regime delimited by the break dates estimated with the Bai and
Perron procedure.
Section 4 concludes and provides a summary of major results and suggestions for future
research.
2. Structural change vs non stationarity in linear regression
Unit root tests
As clearly stated by Perron (2005) testing for a unit root against trend stationarity is
equivalent to addressing the following question: “do the data support the view that the
trend is changing every period or never?”. The problem with conventional unit root tests,
such as the widely used ADF - Augmented Dickey Fuller (1979), is precisely that the unit
root null is tested against the extreme alternative of a trend that always changes,
discarding the case of a trend that changes only “sometimes”. Once allowance is made
for one or more changes in the trend function the question addressed by the test is “do the
data favor a view that the trend is 'always' changing or is changing at most occasionally?”
(Perron, 2005, p. 48-49).
As a standard procedure to test the non stationarity of a series the ADF test is based on
the regression:
k
y t =μ+βt+αy t−1 ∑ c i Δy t−i +et
(1)
non stationarity is refused when the test suggests that α is different from 1. However,
when a structural break is present in the data generating process the conventional ADF
test is biased toward the acceptance of the null resulting in a dramatic loss of power.
Considering the case of a one time change Perron (1989, 1990) proposes a modified
version of the ADF test applicable to four main cases of structural change:
a) non trending series: change in level
b) trending series: change in level
c) trending series: change in slope
d) trending series: change in level or in slope
i= 1
Each of the four cases can be modelled as if the change occurs instantaneously (additive
outlier) or gradually (innovation outlier). The test proposed by Perron was based on know
break dates, that is the date when the shift occurs must be known by the researcher before
analysing the data. As this procedure may lead to some sort of data mining (for example
when the break date is inferred by looking at the graphs of the time series), Zivot and
Andrews (1992) suggested to determine the break date endogenously via a search
algorithm. They proposed to search for the break date that gives the minimal value of the
t statistics of the adjusted ADF:
t a [ inf ]=inf t
(2)
∈
where λ is the fraction (Tτ/T) of the structural break point with respect to the whole
sample, and ∆ is a closed subset of (0,1).
The adjusted ADF statistics of Zivot and Andrews (henceforth ZA) is based on variations
of the following regression which refers to case (d) above with gradual shifts (innovation
outlier):
k
y t =μ+θDU t +βt+γDT t +αy t−1∑ c i Δy t−i +e t
(3)
i=1
where DUt (λ)=1 and DT*t=T-λT for t>λT and zero otherwise. In the simple change in
level case (b above), DT is always zero, while in the change in slope case (c above) it is
DU that is equal to zero. One possible drawback of the ZA test is that it considers a null
of unit root process with no break against the alternative of a stationary process with one
break. When a break is actually present under the null, the test involves size distortion.
However such distortions are relevant only with implausible large shifts and they are
hardly relevant in practice (Perron, 2005). Methodological developments in the area of
unit root tests with structural breaks include generalizations to multiple breaks, to unit
root with breaking trend null hypothesis and extension to tests for the null of stationarity
such as the Kwiatkowski et al. (1992) or KPSS test (see Perron (2005) for a review and
Adachi and Liu (2009) for an application to food price series).
Cointegration tests
Let us consider a static regression between I(1) variables:
y t =μ+αx t +ut
(4)
where xt is a vector of independent variables. The system is cointegrated if the errors u t
are I(0). In that case the relation (4) may be interpreted as a long run equilibrium toward
which the process yt tends.
Within the conventional Engle and Granger (1987) test for cointegration the static
equation (4) is first estimated via OLS and then the stationarity of the residuals of this
relationship is tested via an ADF test using the critical values proposed by MacKinnon
(1991). If the residuals are found to be stationary, the two series are maintained to be
cointegrated. Gregory and Hansen (1996) extend the residual test to take into account a
possible break in the long-run relationship of unknown date. As in ZA, the test statistic is
the minimal value of the t statistics across all possible break dates. The authors consider
three modified version of equation (4) that includes dummies for the structural change :
C
y t =μ+θDU t +αx t +ut
(5a)
C/T
C/S
y t =μ+θDU t +βt+αx t +ut
y t =μ+θDU t +α 1 x t +α 2 DU t x t +ut
(5b)
(5c)
Model C entails a level shift in the equilibrium relationship, model C/T adds a trend to
the previous model whilst model C/S deals with regime shift by adding a change in the
slope coefficients. The authors provide asymptotically critical values for both the ADF
test and the Phillips et al. (1988) Za and Zt statistics.
Cointegration tests in the presence of structural breaks have been also framed maintaining
the reversed null of cointegration . Carrion y Silvestre and Sanso (2006) discuss six
modified versions of equation 4 adding to the models in (5) three further specifications1:
B
C
E
y t =μ+β 1 t+β 2 DT t +αx t +ut
y t =μ+θDU t +β 1 t+β 2 DT t +αx t +ut
y t =μ+θDU t +β 1 t+β 2 DT t +α1 x t +α 2 DU t x t +ut
(6a)
(6b)
(6c)
As pointed out by Perron (2005), the LM-type test statistics proposed by the authors is a
modification of the Gardner's (1969) Q statistics 2. The residuals for the Q statistics are
obtained from the OLS estimation of equations (5) or (6 ) scaled by an estimate of the
long run variance. In the general case, when x t is allowed to be endogenous, the dynamic
least squares (DOLS) estimator (Stock and Watson, 1993) is used. The break date is
estimated by minimizing the sum of squared residuals.
Recently, both the Gregory and Hansen and the Carrion and Sanso tests have been
extended to the two breaks case ( Abdulnasser, 2008 ; Carrion and Sanso, 2007).
Estimation of structural breaks and break dates
Bai and Perron (1998) provide a procedure to estimate structural changes in a linear
model with stationary variables. The procedure may be illustrated looking at a simpler
pure structural change model, that is a model were all coefficients are subject to m+1
regime changes:
y t =x' t δ i +u t for t = Ti-1+1, …., Ti and i = 1, …., m+1
(7)
where xt is a vector of regressors (among which possibly a constant and/ or a trend), the
corresponding vector of coefficients δ i is indexed over the m regimes and the indices T
are the break points.
For a given m partition of the sample (T1,....Tm), δ i can be estimated by OLS
minimizing the sum of squared residuals:
m+ 1
S T T 1 ,. .. ,T m = ∑
T
i
∑ [ y t −x' t δ i ]
(8)
i=1 t=T i−11
Then an estimate of the m break points is given by the set of break points that minimize
the estimated S T over all possible m-partitions of the sample ( provided that a minimum
size condition for each segment is fulfilled).
1
2
Model An, A and D by Carrion i Silvestre and Sanso correspond respectively to models C, C/T and C/S
by Gregory and Hansen.
A similar test is proposed by Arai and Kurozumi (2005)
Bai and Perron (2003) proposes a dynamic programming algorithm in order to reduce the
dimension of the search problem to manageable size. They also demonstrate that break
dates are asymptotically independent when all variables in the model are stationary 3
providing methods to calculate confidence intervals.
As far as the number of breaks is considered, a number of tests are discussed by Bai and
Perron (1998). Sup F test is given by the maximum value of a Wald test for the null
hypothesis of no structural change versus the alternative of k changes, calculated over all
possible k partitions with a common minimal length for each segment.
Double maximum tests are used to make inference without specifying a given number of
breaks. As the name suggests they are the (weighted or not) maximum of the previous sup
F test across all possible number of breaks up to a pre-specified maximum. Their use is
advised since the power of sup-F test may be low when the actual number of breaks is
greater than the one specified (Perron, 2005).
A sequential testing procedure can be based on the test of the null of l breaks against the
alternative of l+1 breaks. Each step requires to carry out l+1 sup F test for one break in
each of the l+1 segments obtained with the usual minimization of the sum of square
residuals in (8). The hypothesis of one additional break is retained if the overall minimum
value of S T across all l+1 break models is sufficiently smaller than the sum of squared
residuals from the l break model.
Finally, the number of breaks can be estimated with information criteria such as:
Schwarz’s Bayesian Information Criteria (BIC, Schwarz, 1978) and Akaike (1973)
Information Criteria (AIC). According to Bai and Perron (2003) the AIC performs
always badly. In absence of serial correlation and when the breaks are actually present the
BIC performs reasonably well. However when serial correlation is an issue or a lagged
dependent variable is included in equation (7) none of the criteria is adequate.
3. An application to three Italian food chains
We applied the methodologies illustrated in section 2 to three Italian food chains for
which price series are available at producer, wholesale (or industrially transformed) and
consumer stages: Pasta, Lamb and Pork. First we checked for stationarity of price series.
Second, we tested for cointegration between couples of price series. Finally, break dates
were estimated with the Bai and Perron procedure and the long run price relationships
were analysed within each regime.
Data
Datasets on farm, wholesale and retail monthly prices for pasta, lamb and pork were
sourced from Datima and from the household panel ISMEA-Nielsen provided by
ISMEA. The dataset spans from January 1994 through December 2008 for the farm and
wholesale prices and from February 2000 through June 2010 for the retail price series.
Datima is a collection of statistical databases including foreign trade and agricultural
markets data, whereas ISMEA-Nielsen is a household panel designed to analyse the
growth of domestic food consumption. Price data refers to aggregated product categories.
3
Methods to construct confidence intervals when the variables are I(1) but integrated are provided in
Kerjwal and Perron (2008). In this case distributions of break dates are not asymptotically independent
.
Monthly farm prices of lamb and hogs are in Euro per Kg and in Euro per Kg slaughter
weight respectively, as well as wholesale and retail prices about lamb and pork cuts.
Durum wheat, semolina and pasta prices are in Euro per kg.
The analysis has been carried out on the period from February 2000 to June 2010 for
which data are available at all market stages. All series were transformed in natural
logarithms and deseasonalised by regressing the transformed series on monthly dummies.
Testing for unit roots
Stationarity of the series were first checked with the conventional ADF test that does not
allow for any break in the data generation process. Results were compared with those
obtained with the Zivot and Andrews (1992) test, described in section 24.
In order to select the appropriate number of lags for our analysis we applied the Schwert
(1989) criterion:
Lag max=12∗ T /100
1/4
(9)
Schwert criterion provides a larger number of lags with respect to other criteria such as
AIC or BIC. However, a larger number of lags makes the actual size of the test closer to
its nominal value (Harris, 1999, p.36).
Results from the ADF tests for pasta, lamb and pork meat are illustrated in table 1, 2 and
3 respectively. In the no-break column of each table the ADF test with both trend and
drift and with drift alone is reported, whereas the one break column displays the ZA test
results. The ADF test suggests the presence of a unit root in all series since the null
hypothesis cannot be refused at the 5% level of significance everywhere.
On the contrary, the ZA test provides a different picture. Once a break in the
deterministic trend is allowed for, the null hypothesis of a unit root process is rejected in
three out of nine series. We run the test in the three versions illustrated in section 2, that is
including a deterministic trend and allowing a shift either in the intercept or in the slope
of the trend or in both.
In the pasta chain series a structural break was found only at the retail stage. The
estimated date is January 2006 with a model fitted with a drift and a change in the trend
slope. Also farm prices of lamb appear to be stationary with a break in March 2006 (a
trend is included in the model).
In the case of pork meat we reject the null hypothesis of presence of a unit root both in
retail and wholesale prices (ZA test with drift and with both change in trend slope and
drift, respectively). These series appear to be stationarity with a structural break, showing
a structural change in September and June 2007 respectively. Producer prices are found to
be I(1) confirming the ADF test.
4
ADF and ZA test were carried out in R, an open sources statistical software, employing respectively the
functions ur.df from package uroot and ur.za from package urca.
Table 1 - Unit root test results for pasta series
Durum Wheat
Semolina
Pasta
wheat prices
semolina
prices
pasta prices
No Break
Augmented Dickey & Fuller
(ADF)
One Break
Zivot & Andrews (ZA)
Trend & drift
drift
w/change
in drift
w/change
in Trend
w/ change
in Trend &
Drift
-4,70
Test value
-2,69
-2,99
-4,08
-4,71
Break date
n.a.
n.a.
n.a.
n.a.
n.a.
Test value
-2,25
-0,42
-4,31
-2,81
-2,86
Break date
n.a.
n.a.
n.a.
n.a.
n.a.
Test value
-2,69
-1,15
-5,11**
-2,93
-5,39**
Break date
n.a.
n.a.
gen-06
n.a.
** Null hypothesis of non stationarity rejected at 5% of significance. t indicates a trend included in the ADF model
d indicates a drift included in the ADF model
gen-06
Table 2 - Unit root test results for lamb series
No Break
Augmented Dickey & Fuller
(ADF)
Lamb
One Break
Zivot & Andrews (ZA)
Trend & drift
drift
w/change
in drift
w/change
in Trend
w/ change
in Trend &
Drift
-2,89
-4,92
-3,69
-5,55**
Farm prices
Test value
Break
date
-2,97
n.a.
n.a.
n.a.
n.a.
mar-06
Wholesale
prices
Test value
Break
date
-3,04
-0,81
-4,27
-3,48
-4,31
n.a.
n.a.
n.a.
n.a.
n.a.
Retail prices
Test value
Break
date
-3,18
-1,77
-4,54
-3,38
-4,88
n.a.
n.a.
n.a.
n.a.
n.a.
** Null hypothesis rejected at 5% of significance. t indicates a trend included in the model
d indicates a drift included in the model
Table 3 - Unit root test results for pork series
No Break
Augmented Dickey & Fuller
(ADF)
Pork
One Break
Zivot & Andrews (ZA)
Trend & drift
drift
w/change
in drift
w/change
in Trend
w/ change
in Trend &
Drift
-1,77
-3,84
-2,09
-4,44
Farm prices
Test value
Break
date
-1,87
n.a.
n.a.
n.a.
n.a.
n.a.
Wholesale
prices
Test value
Break
date
-2,23
-1,03
-4,29
-2,12
-5,36**
n.a.
n.a.
-1,94
-1,76
Test value
Retail prices
Break
date
n.a.
n.a.
** Null hypothesis rejected at 5% of significance. t indicates a trend included in the model
d indicates a drift included in the model
n.a.
n.a.
giu-07
-5,35**
-3,65
-4,58
set-07
n.a.
n.a.
Cointegration analysis
Although some of the series we checked for unit roots were found to be stationary with a
breaking trend, we conservatively run tests of cointegration for all possible couple of
series at the three marketing stages: producer-consumer, producer-wholesale and
wholesale-consumer5. We first run the conventional test of Engle and Granger (EG) and
then the two tests that account for a break in the cointegration relationship described in
section 2: Gregory and Hansen (GH) and Carrion y Silvestre and Sanso (CS) 6. In order to
maintain comparability across tests we always included in the equilibrium equation both
constant and trend7.
We set the maximum lag to 6 and we used the BIC to select appropriate lag lengths.
Specifications adopted in this procedure allow for a level shift in a trending equation or
level and regime shifts (see equation 5b and 5c above) in the case of GH test and for
level and regime shift only in the CS case (equation 5c displayed above). ADF statistics
for EG and GH as well as estimated break dates are shown in table 4.
We also report break dates and LM-type statistics 8 for the CS test. However, it must be
recalled that the break date estimated through the CS procedures is obtained by
minimizing the sum of squared residuals of the long run equation and is not consistent
with the one of the GH test that is selected through minimization of the t statistic of the
ADF.
The cointegrating relationships are confirmed either in EG, GH or CS tests for the wheatsemolina and lamb wholesale-retail couples of series. Interestingly, we cannot reject the
null of no-cointegration in the pork farm-wholesale case according to EG, whilst this
couple of series appears to be cointegrated once a break is allowed for.
With the GH test we found a stronger evidence of a cointegrating relationship (rejecting
the null hypothesis of absence of cointegration) when a regime change rather than a trend
is included in the model. Overall, the CS test confirms the GH but for the lamb farmretail and the pork farm-retail and wholesale-retail couples.
As far as break dates are concerned, GH test statistics agree with the CS tests for all the
series with the sole exception of the lamb wholesale-retail couple.
5
6
7
8
In the case of the pasta marketing chain the wholesale stage refers to the product of the first
transformation durum wheat semolina.
All estimates were obtained using the software R . To this purpose, we ported in R from Gauss both the
script made available by Hansen in his webpage (http://www.ssc.wisc.edu/~bhansen/) and the code
posted by Carrion y Silvestre (http://riscd2.eco.ub.es/~carrion/Welcome.html). R codes are available
from the authors upon request.
This implies running a simple ADF test on residuals not including neither a trend nor a constant when
looking for stationarity in residuals of the equilibrium equation with the EG test.
We used DOLS residuals following Carrion y Silvestre and Sanso (2006) suggestion that this version of
the test has better size and power properties irrespective of the endogeneity or exogeneity of regressors.
Table 4 – Cointegration Tests
Engle &
Granger (BIC)
Chains
Pasta
Wheat/Semolina
Semolina/Pasta
Farm/Retail
Lamb
Farm/Wholesale
Wholesale/Retail
Farm/Retail
Pork
Farm/Wholesale
Test value
Break
date
Carrion-i-Silvestre & Sansò
(min SSR)
C/T
C/S
D
-2.7
Trend &
Drift
-2.82
-4.19
-4.02
0.30**
n.a.
n.a.
Oct-08
Oct-08
Oct-07
-5.26**
5.96**
0.04
Drift
Wheat/Pasta
Gregory &
Hansen (BIC)
Test value
Break
date
-3.35
-4.15**
n.a.
n.a.
Aug-07
Jun-07
Jun-07
Test value
Break
date
-2.53
-2.65
-4.66
-4.54
0.27**
n.a.
n.a.
Oct-08
Oct-08
Dec-07
-6.91**
5.91**
0.34**
Test value
Break
date
-4.80**
-4.93**
n.a.
n.a.
Test value
Break
date
-3.56
-3.86
n.a.
n.a.
Apr-04
Dec-08
-4.61
0.19**
Mar-03
Nov-04
Sep-04
-6.79**
7.29**
0.04
Test value
Break
date
-4.67
n.a.
n.a.
Aug-01
Agu-01
Oct-00
Test value
Break
date
-2.66
-3.03
-2.94
-4.24
0.05
n.a.
n.a.
Jul-01
Jan-06
Jul-05
0.1
Test value
Break
date
-6.43**
Aug-01
-7.12**
-2.58
-3.32
-6.52**
6.87**
n.a.
n.a.
Feb-07
Feb-07
-1.82
-1.99
-3.87
-3.94
Test value
Wholesale/Retail
Break
date
n.a.
n.a.
Nov-01
Jul-05
** Reject the null hypothesis at the 5% of significance; Null for Carrion y Silvestree and Sanso test is cointegration.
Jan-07
0.04
Jan-06
Estimating structural breaks and break dates
To estimate break dates for the cointegration relationships above, we employed the Bai
and Perron (2003) dynamic programming algorithm as implemented by Zeileis et al
(2005)9. The algorithm provides confidence intervals for break dates that are valid only
for relationships among I(0) variables. However, point estimates remain consistent even
for I(1) cointegrated variables (Kejriwal and Perron, 2008).
For each possible breaking date, we carried out a sequence of F statistics based on the
null hypothesis of no shifts against the alternative of a single-shift. The corresponding
sup-FT statistics provides a test for structural change against a single break alternative of
9
All estimates were obtained using the R package strucchange.
unknown timing. We also carried out tests for the number of breakpoints based on information criteria, notably BIC and simple Residual Sum of Squares (RSS)10.
Once the break dates have been estimated, a pure structural change model for the price
transmission equilibrium (or long run) relationships is given by equation (7):
y t =x' t δ i +u t
(7)
where the subscript i refers to the m+1 regime delimited by the break dates. In our case x
is a bi-dimensional vector with a constant ( x 1t ) and the logarithm of the upstream
price series ( x 2t ).
Table 5 – Bai & Perron test – break dates, regimes and confidence intervals for pasta
n° of
breaks
1
Pasta - Bai & Perron test - (Farm - Wholesale)
breakdates
Partitions
Lower bound (2.5%) Upper bound (97.5%)
2007(6)
2000(2) - 2007(6)
2007(7) - 2010(4)
2007(5)
2007(8)
Table 6 – Bai & Perron test – break dates, regimes and confidence intervals for lamb
n° of
breaks
1
Lamb - Bai & Perron test - (Wholesale - Retail)
breakdates
Partitions
Lower bound (2.5%) Upper bound (97.5%)
2001(7)
2000(2) - 2001(7)
2000(8)
2001(12)
2001(8) - 2010(4)
Table 7 – Bai & Perron test – break dates, regimes and confidence intervals for pork
Pork - Bai & Perron test - (Farm - Wholesale)
n° of
breaks
breakdates
Partitions
Lower bound (2.5%)
Upper bound (97.5%)
1
2007(1)
2000(2) - 2007(1)
2006(12)
2007(2)
2007(2) - 2010(4)
10
We set the trimming rate for both tests to 0.15, whereas the maximal number of breaks is built by
default from the trimming parameter (generally five breaks maximum).
Figure 1 – BIC, RSS models for m break points and SupF test plots for wheat-semolina
Figure 2 – BIC, RSS models for m break points and SupF test plots for lamb wholesale retail
seriews
Figure 3 – BIC, RSS models for m break
points and SupF test plots for hog farm wholesale series
Results are reported with reference to the 3 cointegrated couple of series. Although BIC
or RSS criteria would have suggested for some series (notably lamb and pasta) a larger
number of breaks we retained a single break given the relatively small number of
observations. Tables 5, 6, and 7 show the estimates of optimal break dates and related
regime partitions. Coefficient estimates for each regime are provided in table 9 where the
i the coefficients 1,i , 2,i refer respectively to the constant and the upstream price in
the long run equation. Plots of the BIC and RSS values as well as SupF curves are
provided in figures 1 to 3 while figures 4 to 6 show the original series with regime
delimiters.
Table 9 - Bai & Perron test – Coefficients estimates
Pasta - Bai & Perron test - coefficients estimates
Coefficient
δ1,1
Farm/Wholesale
Estimstd error
Partitions
ate
-0.05
0.04
2000(2) - 2007(6)
δ1,2
0.25
0.02
2007(7) - 2010(4)
δ2,1
0.70
0.02
2000(2) - 2007(6)
δ2,2
0.83
0.01
2007(7) - 2010(4)
Lamb - Bai & Perron test - coefficients estimates
Coefficient
δ1,1
Wholesale/Retail
Estimstd error
Partitions
ate
1,03
0.20
2000(2) - 2001(7)
δ1,2
1,55
0,06
2001(8) - 2010(4)
δ2,1
0,59
0,12
2000(2) - 2001(7)
δ2,2
0.33
0.03
2001(8) - 2010(4)
Pork - Bai & Perron test - coefficients estimates
Coefficient
δ1,1
Farm/Wholesale
Estimstd error
Partitions
ate
1,03
0.01
2000(2) - 2007(1)
δ1,2
1,36
0.02
2007(2) - 2010(4)
δ2,1
0,81
0,04
2000(2) - 2007(1)
δ2,2
0.70
0,11
2007(2) - 2010(4)
Notably, we do not find a clear evidence of the correct number of breaks from the SupF
test plot which shows a major break in 2007 and two minor structural changes in 2002
and, possibly in 2005 .
The single break estimate appears to be related to the beginning of the commodity bubble
(june ‘07 break). Equilibrium coefficients estimates suggest that elasticity of transmission
as well intercept estimates increase in the commodity bubble period. This results in a
larger reactivity of semolina prices to changes in prices of durum wheat. The overall
effect on margins is ambiguous as log of prices are negative in this case.
For the lamb wholesale-retail couple a single breakpoint is also suggested by the BIC
criterion and by the sup-F test graphic that shows a single major spike at the beginning of
the series. Elasticity of transmission decrease after the break. However, the intercept
increases. In the second regime margins tend to decrease as the growth of wholesale
prices is only partially transmitted to the retail stage.
Finally, for the hog farm-wholesale transmission equation both the BIC and the graphical
pattern of the sup-F test (the latter with a clear single spike) suggest a single breakpoint.
The change occurs at the beginning of the commodity bubble (January 2007) that led to
dramatic increase of prices in the food sector (fig.6).
The second regime is characterized by a larger intercept and a smaller elasticity of
transmission. As the farm price shows no definite trend this results in wider margins.
Along the period considered the Italian pork marketing chain has witnessed a move
toward concentration of the slaughtering industry, this process may have impacted also
the price transmission mechanism in the context of the 2007-09 price crisis as we observe
a shrinking of the share of wholesale value accruing to farming.
Figure 4 - Nominal price series in natural logarithms – wheat-semolina
Note: farm price (in black) / wholesale price (in red)
Figure 5 - Nominal price series in natural logarithms – Wholesale / Retail - lamb
Note: Wholesale price (in black) / Retail price (in red)
Figure 6 - Nominal price series in natural logarithms – Farm / Wholesale - hog
Note: farm price (in black) / wholesale price (in red)
3. Conclusions
This article investigated long run transmission elasticities in three Italian food chains
(pasta, lamb and pork) accounting for structural breaks. We analysed three price
transmission relationships for each chain: farm-wholesale, wholesale-retail and farm
retail. Within the classical cointegration framework we found evidence of equilibrium
relationships only for the durum wheat-semolina and lamb wholesale-retail series.
However, once structural breaks are accounted for a long run relationship emerges also
for the hog farm-wholesale series. For each cointegrated couple of prices we examined
the presence of structural breaks.
Main changes in the long run relationship were found for the pasta and the pork chain.
Both equations show a break at the beginning of the price bubble that altered the
transmission mechanism across different stages of the food chain notably in the direction
of a larger elasticity of transmission in former case and a smaller elasticity in the latter .
Although, long run transmission elasticities are valuable findings per se, the analysis
could be easily extended to the study of possible asymmetries in the transmission
mechanism and to modelling of short run relationship via error correction models.
Further research is also needed to study food prices using a wider range of unit root and
cointegration tests within the multiple breaks framework that has been recently developed
and applied in the fields of financial and macroeconomic series.
4. References
Abdulnasser Hatemi-J (2008), Tests for cointegration with two unknown regime shifts with an
application to financial market integration, Empir Econ 35:497–505.
Adachi K., Liu D.J. (2009), Estimating Long-Run price relationship with structural change of
unknown timing: an application to the Japanese pork market, American Journal of Agricultural
Economics, 91: 1440-1447
AGCM (2007). Indagine conoscitiva sulla distribuzione agroalimentare, Indagini Conoscitive,
28, Autorità garante per la Concorrenza e il Mercato, Roma.
Agra CEAS (2007).The Gap between Producer Prices and the Price paid by the Consumer,
European Parliament, Brussels.
Akaike H (1973) Information theory and an extension of the maximum likelihood principle. In:
Petrov BN., Csake F. (eds) Second international symposium on information theory. Akademiai
Kiado, Budapest, Hungary: 267–281
Andrews D.W.K. (1993). Tests for parameter instability and structural change with unknown
change point, Econometrica, 61: 821–856.
Bai J., Perron P. (1998). Estimating and Testing Linear Models with Multiple Structural Changes,
Econometrica, 66: 47-78.
Bai J., Perron P. (2003). Computation and Analysis of Multiple Structural Change Models,
Journal of Applied Econometrics, 18: 1-22.
Boetel, Brenda L. & Liu, Donald J. (2008). Incorporating Structural Changes in Agricultural and
Food Price Analysis: An Application to the U.S. Beef and Pork Sectors, Working Papers 44076,
University of Minnesota, The Food Industry Center.
Carrion-i-Sylvestre, J.L., Sansó-i-Rosselló, A.S., (2006). Testing the null hypothesis of
cointegration with structural breaks, Oxford Bulletin of Economics and Statistics, 68: 623-646
Carrion-i-Sylvestre, J.L., Sansó-i-Rosselló, A.S., (2007). The KPSS test with two structural
breaks, Spanish Economic Review, 9(2):105-127
Commission of the European Communities (2009) - A better functioning food supply chain in
Europe – Brussels 28.10.09 - COM 2009(591). Available at
http://ec.europa.eu/economy_finance/publications/publication16061_en.pdf
Dickey, D.A., Fuller, W.A., (1979). Distribution of the estimators for autoregressive time series
with a unit Root, Journal of the American Statistical Association, 74: 427-431.
Engle R. F., e Granger C.W.J. (1987). Co-integration and error correction: representation,
estimation and testing, Econometrica, 55, 251–276.
Ferrucci G, Jiménez-Rodríguez R., Onorante L. (2010) - Food price pass-through in the euro
area, the role of asymmetries and non-linearities. European Central Bank Working Paper Series
n° 1168 - April 2010. Available at http://www.ecb.europa.eu
Frey G., Manera M. (2007). Econometric Models Of Asymmetric Price Transmission, Journal of
Economic Surveys 21(2): 349-415.
Gardner, L.A., 1969. On detecting changes in the mean of normal variates, The Annals of
Mathematical Statistics 40: 116-126.
Giangiulio D., Mazzantini G. (2010). Il trasferimento dei prezzi lungo le filiere agroalimentari:
una possibile metodologia di analisi per i profili di interesse Antitrust, XLVII Convegno SIDEA,
Campobasso
Gregory, A.W., and B.E. Hansen. (1996). Tests for cointegration in models with regime and trend
Shifts, Oxford Bulletin of Economics and Statistics 58:555–60.
Harris, R. (1999), Using Cointegration Analysis in Econometric Modelling, London, Prentice
Hall.
IPI (2008). Le dinamiche dei prezzi nella filiera del frumento, Istituto per la Promozione
Industriale.
Kejriwal, M., and Perron, P. (2008). The limit distribution of the estimates in cointegrated
regression models with multiple structural changes, Journal of Econometrics 146: 59–73.
Kwiatkowski, D., Phillips, P. C. B., Schmidt, P., and Shin, Y. (1992). Testing the null hypothesis
of stationarity against the alternative of a unit root, Journal of Econometrics 54, 159-178.
Lee, J., Strazicich, M.C. (2003). Minimum LM unit root test with two structural breaks, Review
of Economics and Statistics 63:1082-1089.
London Economics (2003), Investigation of the determinants of farm-retail price spreads. Report
prepared for the Department for Environment, Food and Rural Affairs. Available at
http://statistics.defra.gov.uk/esg/reports/pricespreads/default.asp.
MacKinnon J. (1991), Critical Values for Cointegration Tests. In Engle, R. and Granger, C. (eds),
Long-Run Economic Relationships, Oxford University Press, Oxford.
Meyer, J. and S. von Cramon-Taubdel. (2004). Asymmetric Price Transmission: A Survey.
Journal of Agricultural Economics 55:581-611.
National Bank of Belgium (2008). Processed food: inflation and price levels, NBB Economic
Review, Special Edition, April: 53-72.
Peltzman, S. (2000). Prices Rise Faster than they fall, Journal of Political Economy 108(3): 466502.
Perron, P. (1989). The great crash, the oil price shock, and the unit root hypothesis, Econometrica
57: 1361-1401
Perron, P. (1990). Testing for a Unit Root in a Time Series with a Changing Mean, Journal of
Business and Economic Statistics 8: 153-162.
Perron, P. (2005). Dealing with Structural Breaks, Mimeo forthcoming in the Vol. 1 Handbook of
Econometrics: Econometric Theory.
Phillips, P.C.B., Perron, P., (1988). Testing for a unit root in time series regression, Biometrika 75:
335-346.
Schmidt, P. and Phillips, P.C.B. (1992). LM Test for a unit root in the presence of deterministic
Trends, Oxford Bulletin of Economics and Statistics 54(3): 257–287.
Schwarz G (1978). Estimating the dimension of a model, Annals of Statistics 6(2): 461–464 .
Stock, J.H., Watson. M.W., (1993). A simple estimator of cointegrating vectors in higher order
integrated systems, Econometrica, 64: 783-820.
Wang, D., and W.G. Tomek. (2007). Commodity Prices and Unit Root Tests, American Journal of
Agricultural Economics, 89: 873-889.
Zeileis A., Kleiber C., Krämer W., Hornik K. (2003). Testing and Dating of Structural Changes in
Practice, Computational Statistics and Data Analysis 44:109-123.
Zeileis A, Leisch F, Kleiber C, Hornik K (2005). Monitoring Structural Change in Dynamic
Econometric Models, Journal of Applied Econometrics, 20(1): 99-121.
Zivot, E., Andrews, D.W.K., 1992. Further evidence on the great crash, the oil price shock and the
unit root hypothesis, Journal of Business and Economic Statistics, 10: 251-270.