ECONOMETRIC MODELS OF

ASYMMETRIC PRICE TRANSMISSION

Giliola Frey
Fondazione Eni Enrico Mattei

Matteo Manera
University of Milan-Bicocca and Fondazione Eni Enrico Mattei
Abstract. In this paper, we review the existing empirical literature on price
asymmetries in commodities, providing a way to classify and compare different
studies that are highly heterogeneous in terms of econometric models, type
of asymmetries and empirical findings. Relative to the previous literature, this
paper is novel in several respects. First, it presents a detailed and updated
survey of the existing empirical contributions on price asymmetries in the
transmission mechanism linking input prices to output prices. Second, this paper
presents an extension of the traditional distinction between long-run and short-
run asymmetries to new categories of asymmetries, such as: contemporaneous
impact, distributed lag effect, cumulated impact, reaction time, equilibrium
and momentum equilibrium adjustment path, regime effect, regime equilibrium
adjustment path. Each empirical study is then critically discussed in the light of
this new classification of asymmetries. Third, this paper evaluates the relative
merits of the most popular econometric models for price asymmetries, namely
autoregressive distributed lags, partial adjustments, error correction models,
regime switching and vector autoregressive models. Finally, we use the meta-
regression analysis to investigate whether the results of asymmetry tests are
not model-invariant and find which additional factors systematically influence
the rejection of the null hypothesis of symmetric price adjustment. The main
results of our survey can be summarized as follows: (i) each econometric
model is specialized to capture a subset of asymmetries; (ii) each asymmetry
is better investigated by a subset of econometric models; (iii) the general
significance of the F test for asymmetric price transmission depends mainly on
characteristics of the data, dynamic specification of the econometric model, and
market characteristics. Overall, our empirical findings confirm that asymmetry, in
all its forms, is very likely to occur in a wide range of markets and econometric
models.
Key words. Cointegration; Meta-regression analysis; Partial adjustment; Price
asymmetries; Threshold regime switching

0950-0804/07/02 0349–67 JOURNAL OF ECONOMIC SURVEYS Vol. 21, No. 2


C 2007 The Authors

Journal compilation 
C 2007 Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK

and 350 Main Street, Malden, MA 02148, USA.

 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
350 FREY AND MANERA

1. Introduction
Consumers are generally very concerned when retailers decide to increase the price
of their products as a consequence of increases of wholesale prices, but not to reduce
the price as a consequence of a fall in wholesale prices. This sharp attention to
product price variations is particularly addressed to those goods which significantly
contribute to the consumers’ daily expenditure.
One of the products whose price variations consumers are particularly sensible
to is gasoline. Given the importance of individual mobility in modern societies, it
is quite natural to think that a reduction in the price of gasoline makes consumers
very happy just as a price rise makes them very upset. What is less obvious is
whether this last statement could be considered an appropriate description of the
real consumers’ sentiment. Would it not be better to say that consumers are very
happy if the price of gasoline decreases, while they are very, very upset if the price
of gasoline increases?
As illustrated, among others, by Brown and Yucel (2000) for the US gasoline
market, many consumers complain of the existence of price asymmetries, which
they interpret as evidence of monopolistic behaviour in the markets for oil and
petroleum products.
The perception of price asymmetries in the mechanism of transmission linking
input prices to output prices is not confined to the gasoline market, but it is typical of
many agricultural products (e.g. vegetables, meat, dairy products, etc.) and financial
markets (e.g. interest rates, bank deposits, etc.). In any case, the crucial question
is whether output prices respond symmetrically to variations of input prices, or if
prices behave as the consumers’ common sense seems to suggest.
This question has received growing attention in the last decade, as demonstrated
by the existence of a very large and mixed empirical literature on price transmission
asymmetries. Studies generally differ in terms of analyzed goods, dependent and
explanatory variables (Table 1), countries under scrutiny (Table 2), time frequencies,
time periods, specifications of the models employed, and even type of journal

Table 1. List of Variables Used in the Surveyed Literature.

cr Crude oil price D Dummy variable

dp Domestic producer price er Exchange rate
fb FOB price fm Farm price
G Gasoline consumption I Inventory level
LL Local lettuce level mk Marketing costs
pr Producer price pw World price
Q Production rf Ex-refinery petroleum products price
rt Retail price S Stock of vehicles
sp Spot price st Stock level
tx Taxes ur Utilization rate
wp World import price ws Wholesale price
x Generic variable y Generic variable
Z Income

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 351

Table 2. List of Countries Analyzed in the Surveyed Literature.

A Argentina Au Australia B Brazil C Canada

D Denmark E European Union Fi Finland F France
G Germany Gn Ghana Ir Ireland I Italy
j Japan N Netherlands Pe Peru Ph Philippines
Sp Spain S Switzerland SA South Africa Sw Sweden
UK United Kingdom US United States World Worldwide

(Table 3). As a consequence, the empirical findings are not always unique, making it
difficult to determine whether prices do behave in an asymmetrical way or consumers
are wrong. The proliferation of many different contributions in this area provides
a valid motivation for this survey paper, which proposes a detailed classification
of the various studies according to the different categories of asymmetries and
a critical evaluation of the relative merits of the most popular econometric
models.
Given the relevance of this topic, it is not surprising that our review of the
empirical literature on asymmetric price transmission is not the only one available to
the interested reader. Recently, two different surveys have been published – Geweke
(2004) and von Cramon-Taubadel and Meyer (2004). Geweke (2004) concentrates
on the gasoline price literature, which addresses the question of whether there is a
systematic tendency for downstream prices in the oil well-to-service station gasoline
industry to respond to increases in upstream prices more rapidly than downstream

Table 3. List of Journals and Reviews.

Ag Agribusiness
AE Applied Economics
AgE Agricultural Economics
AJAE American Journal of Agricultural Economics
CJAE Canadian Journal of Agricultural Economics
CJE Canadian Journal of Economics
ERAE European Review of Agricultural Economics
EE Energy Economics
EP Energy Policy
IJER International Journal of Energy Research
JA Journal of Agribusiness
JDE Journal of Development Economics
JPE The Journal of Political Economy
OR OPEC Review
PRE Philippine Review of Economics
QJE The Quarterly Journal of Economics
RES The Review of Economics and Statistics
RIO Review of Industrial Organization
SJE Scandinavian Journal of Economics
WP Working Paper

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
352 FREY AND MANERA

prices respond to decreases in upstream prices. Geweke surveys 18 papers, and

excludes the contributions which have to do with the gasoline industry in Europe,
as well as the related literature in economic theory on possible connections between
imperfect competition and asymmetric movements in price. Von Cramon-Taubadel
and Meyer (2004) classify asymmetries according to their type (either vertical or
spatial), magnitude and sign. They concentrate on the economic explanations of
asymmetric price behaviour, with particular reference to market power, adjustment
and menu costs. They divide the econometric methods used to identify the
presence of asymmetric price transmission into pre-cointegration approaches and
cointegration-based models. They discuss a number of methodological issues
which are relevant to evaluate existing contributions and to describe patterns for
future research in this area. Meyer and von Cramon-Taubadel survey 40 papers,
80% of which are on agricultural markets, while only 10% are on the gasoline
market.
In this paper we review the existing empirical work on asymmetric price
transmission in commodities markets, suggesting a multi-dimensional approach
to classify and compare different studies that are highly heterogeneous in terms
of econometric models, type of asymmetries and empirical findings. Our survey
answers three crucial questions: 1) If the empirical evidence in favour of (or against)
price asymmetries is, as one would expect, not model invariant, what are the most
popular models used in the existing literature to investigate the input–output price
transmission mechanism? 2) Does the term ‘price asymmetry’ define a homogeneous
concept, or should alternative types of asymmetries be identified and introduced?
3) Would it be possible to classify the models currently used in the empirical work on
price asymmetries in terms of their ability to describe specific types of asymmetries?
Our survey complements and extends the contributions of Geweke (2004) and
von Cramon-Taubadel and Meyer (2004) in several respects. First, it presents a
comprehensive and updated review of the existing empirical contributions on the
presence of price asymmetries in the transmission mechanism linking input prices
to output prices. Our review includes 70 papers, which are equally distributed
between agricultural/alimentary and gasoline markets. Second, we introduce new
categories of asymmetries, namely: contemporaneous impact, distributed lag effect,
cumulated impact, reaction time, equilibrium and momentum equilibrium adjustment

Table 4. Definition of Asymmetries (Symmetries).

COIA (COIS) Contemporaneous impact

DLEA (DLES) Distributed lag effect
CUIA (COIS) Cumulated impact
RTA (RTS) Reaction time
EAPA (EAPS) Equilibrium adjustment path
MEAPA (MEAPS) Momentum equilibrium adjustment path
REA (RES) Regime effect
REAPA (RTS) Regime equilibrium adjustment path


C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 353

Table 5. Econometric Models of Asymmetric Price Transmission.

ARDLpp Autoregressive Distributed Lag Model based on period-to-period price

variations
ARDLcu Autoregressive Distributed Lag Model based on cumulative price
variations
ECMeg Error Correction Model estimated with Engle and Granger’s method
ECMsw Error Correction Model estimated with Stock and Watson’s method
ECMth Error Correction Model with threshold cointegration
PAM Partial Adjustment Model
RSM Regime Switching Model
DRS Deterministic Regime Switching Model
SRS Stochastic Regime Switching Model
VECM Vector Error Correction Model
VAR Vector Autoregressive Model
VARcu Vector Autoregressive Model based on cumulative price variations
VRS Vector Regime Switching Model
VRSeg Vector Regime Switching Model estimated with Engle and Granger’s
method
VRShs Vector Regime Switching Model estimated with Hansen and Seo’s method

path, regime effect and regime equilibrium adjustment path (Table 4). We then
discuss each empirical study in the light of this new classification of asymmetries.
Third, we evaluate the relative merits of the most popular econometric models for
price asymmetries, with particular reference to autoregressive distributed lags, partial
adjustments, error correction models, regime switching and vector autoregressive
models (Table 5). Finally, we use the powerful tool of meta-regression analysis to
investigate whether the results of asymmetry tests are not model invariant and find
which additional factors (e.g. types of asymmetries, sample size, etc.) systematically
influence the rejection of the null hypothesis of symmetric price adjustment.
The plan of the paper is as follows. Section 2 discusses the new, as well as
the more traditional, definitions of asymmetry. Section 3 presents the empirical
work based on early econometric models, namely autoregressive distributed lag
specifications. Section 4 is dedicated to the equilibrium correction approaches,
that is, partial adjustment, error correction and threshold autoregressive models.
In Section 5 the more recent econometric models are illustrated, such as regime
switching and vector autoregressive models. The results from the meta-regression
analysis of price transmission asymmetries are presented in Section 6. Section 7
concludes the study.

2. Price Asymmetries
Prior to answering the question of whether the relationship between the price of
an input and the price of one (or more) output(s) is symmetric or asymmetric, it is
crucial to understand that the word ‘asymmetry’ does not have a unique meaning
and, consequently, to distinguish among different types of asymmetries. A widely

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
354 FREY AND MANERA

used classification is between short-run (SR) and long-run (LR) asymmetries, since,
in general, an SR analysis is more indicated to compare the intensity of output price
variations to positive or negative changes in input prices, whereas an LR perspective
is needed if the empirical investigation concentrates on the computation of reaction
times, length of fluctuations, as well as speeds of adjustment towards an equilibrium
level.
Specific econometric models focus on different aspects of the relation between
input and output prices, or, equivalently, on different types of asymmetries. In
this paper we identify five major classes of econometric models, namely the
autoregressive distributed lag (ARDL) model, the partial adjustment model (PAM),
the error (or equilibrium) correction model (ECM), the regime switching model
(RSM) and, finally, their multivariate extensions. Since the different concepts
of asymmetry which are defined in this section apply to both univariate and
multivariate models, for the sake of simplicity we discuss asymmetries using
single-equation specifications. In particular, we define eight types of asymmetries
(A)/symmetries (S), namely: contemporaneous impact (COIA/COIS), distributed lag
effect (DLEA/DLES), cumulated impact (CUIA/CUIS), reaction time (RTA/RTS),
equilibrium adjustment path (EAPA/EAPS) and momentum equilibrium adjustment
path (MEAPA/MEAPS), regime effect (REA/RES) and regime equilibrium adjust-
ment path (REAPA/ REAPS).
In an ARDL, a variable y t ; t = 1, . . . , n, depends on its own lags (autoregressive
part, or AR) and on a vector of variables X, both contemporaneous and lagged
(distributed lag part, or DL).
If with x we indicate a single explanatory variable, i.e. an element of X, a typical
ARDL can be specified as:
 r s
yt = φh yt−h + αi xt−i +  t (1)
h=1 i=0

where  t is a white noise.

Model (1) can be generalized to incorporate asymmetries by assuming that x has
a different impact on y, according to whether its sign is positive (+) or negative (−):

r 
s 
q
yt = φh yt−h + αi+ xt−i
+
+ α−j xt−
−
j + t (2)
h=1 i=0 j=0

Clearly, the above specification supports various types of asymmetries, which can
be classified in four main categories.
First, a test of the null hypothesis α+ −
0 = α0 provides information about the
+ −
contemporaneous impact of x and x on y, which is defined to be asymmetric
(COIA) or symmetric (COIS) according to whether the null is rejected or not.
Second, it is easy to check whether the impact of x + and x − is the same at any
lag by testing the null hypothesis αi+ = α−j , i = 1, . . . , s, j = 1, . . . , q, which, if
rejected (not rejected), will denote an asymmetry (symmetry) due to a distributed
lag effect (DLEA/DLES). It is worth noting that s = q implies a DLEA, while the
vice versa is clearly false.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 355

Another kind of asymmetry which is linked to DLEA is the mean lag asymmetry.
A mean lag is generally defined as a weighted average of the lags on x, with weights
being the coefficients of the model. Mean lag asymmetry occurs when the mean lags
for positive and negative variations of x are different. If the mean lags are symmetric,
we cannot reach any reliable conclusion on the presence of DLEA, as the distributed
lag effect can be either symmetric or asymmetric. Conversely, when the mean lags
are asymmetric, then we can support a DLEA.
COIA and DLEA are SR asymmetries, since they are comparing the impacts on y
of x + and x − at a given instant in time; however, it is evident that specification (2)
can also incorporate LR asymmetries. A third possibility is whether the cumulated
effect of x + and x −at lag t − k is symmetric. This can be checked by testing
αi+ = j=k α−j , with k ∈ [0, min(s, q)]; hereafter, we
s q
the null hypothesis i=k
will denote this asymmetry as CUIA. Note that testing this hypothesis for all k ∈
[0, min(s, q)] is equivalent to jointly testing the two hypotheses αi+ = α−j , α+0 = α0
−

described above. Moreover, the joint existence of DLES and COIS is a sufficient,
although not necessary, condition for CUIS, while the coexistence of DLEA and
COIA does not imply either CUIA or CUIS.
Finally, impulse response or cumulative adjustment functions are used to compute
the number of periods needed by the dependent variable to (re-)adjust to an
equilibrium level once an asymmetric shock to x + and/or x − has occurred. Clearly,
this kind of asymmetry is related to the persistence of the effects of x + and x − ,
and, unlike a test on s = q, it also accounts for the impact of the other variables
in the model. This asymmetric (symmetric) response will be hereafter referred to as
reaction time asymmetry (RTA) (reaction time symmetry, or RTS).
A PAM assumes that there exists a target level for y (say y ∗ ) and relates the actual
value of y to its value at time t − 1 and to the deviation of y t−1 from the actual
target level y ∗t :
 
yt = β yt−1 + (1 − φ) yt∗ − yt−1 +  t (3)

Since φ expresses the speed of convergence of y to y ∗ , if φ = 0 the adjustment

to the equilibrium level is instantaneous, while φ = 1 implies an infinite adjustment
process.
This model can be generalized to incorporate asymmetries by assuming that the
adjustment depends on whether y t−1 is above or below the equilibrium level:
 ∗ +  ∗ −
yt = β yt−1 + φ + ϕ yt−1 − yt−1 + φ − ψ yt−1 − yt−1 +  t (4)
where ϕ(.) and ψ (.) are functions of the disequilibrium, which may be equal to
the identity function. Henceforth, we will denote this kind of asymmetry as an
equilibrium adjustment path asymmetry (EAPA).
If the considered series are stationary, then ARDL and PAM can be consistently
estimated with OLS. On the contrary, if the series are non-stationary, then, as
shown by Granger and Newbold (1974), standard linear regression analysis can lead
to spurious results, that is, the relation between two variables is only apparently
significant. Economic variables are often integrated of order one, or I(1), meaning

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
356 FREY AND MANERA

that they can be made stationary by first differencing; for this reason, a solution for
non-stationarity that has been widely used in the past is to estimate a model in first
differences.
Modern econometric analysis proposes a different framework for modelling non-
stationary data. In their seminal paper, Engle and Granger (1987) point out that,
given a pair of I(1) series, if there exists a linear combination between them which is
stationary, the two processes move together in the LR and are said to be cointegrated.
In order to exploit the concept of cointegration, Engle and Granger develop an
equilibrium correction representation (ECM), which, given two I(1) variables y and
x which are cointegrated with cointegrating vector (1 − θ), can be written as follows:

yt = αxt + λ(yt−1 − θxt−1 ) +  t (5)

Lagged variables and autoregressive effects can be added to this model, which is
also able to incorporate asymmetries as proposed by Granger and Lee (1989):

r 
s 
q
yt = βh yt−h + αi+ xt−i
+
+ α−j xt−
−
j
h=1 i=0 j=0
+ −
+ λ+ EC Tt−1 + λ− EC Tt−1 + t (6)
where ECT t−1 = (y t−1 − θx t−1 ).
Model (6) considers all the asymmetries which are testable within the ARDL
specification and also supports a test for symmetric equilibrium adjustment path. As
a matter of fact, if λ+ = λ− , the convergence process is different depending on the
direction of the deviation from the equilibrium level. Model (6) can also be extended
as suggested by Enders and Granger (1998):

s
yt = αi xt−i + γ + EC Tt−1 It + γ − EC Tt−1 (1 − It ) +  t (7)
i=0

where

1 if EC Tt−1 ≥ 0
It = (8)
0 if EC Tt−1 < 0

In this case, asymmetries arise depending on whether the deviation from the
equilibrium is increasing or decreasing. These asymmetries are known as momentum
equilibrium adjustment path asymmetries (MEAPA).
The models described so far are based on the idea that some of, or even all, the
explanatory variables X may have a non-linear impact on y. However, the analysis
can be extended to incorporate the possibility that the relationship between y and X
as a whole depends on the state of a variable v, which can be one of the explanatory
variables. Generally, it is said that the level of v, relative to a threshold δ, describes
different states of the world, or regimes, hence the name regime switching models
(RSM). Specifically, we can define two different models, depending on the nature
of the threshold variable, the deterministic RSM and the stochastic RSM. In the first

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 357

model, we know which regime prevails in each instant of time; in the second model,
the shift from one regime to another is random.
According to the value of the threshold variable, we can consider a more detailed
classification of RSM, based on whether or not v belongs to X. If v does not belong
to X, then a general RSM, with p + 1 states of the world, can be written as:
⎧
⎪
⎪ yt = f (X ) + u t if v < δ1
⎪
⎪
⎪
⎪ y = f 
(X ) + u 
if δ1 ≤ v ≤ δ2
⎨ t t
...... (9)
⎪
⎪
⎪
⎪ ∗
yt = f (X ) + u t ∗
if δ p−1 ≤ v ≤ δ p
⎪
⎪
⎩ y = f ∗ (X ) + u ∗ if v > δ
t t p

where, within each regime, the relationship between y and X can assume an ARDL,
PAM or ECM form. If we consider the general case of an asymmetric ECM, when v
belongs to X, the threshold can be defined in terms of either x or the error correction
term ECT, yielding:
⎧  s q
⎪ yt = φ0 + ri=1 βi yt−i + i=0 αi+ xt−i
+
+ i=0 αi− xt−i
−
⎪
⎪
⎪
⎪ +
+ λ+ EC Tt−1 −
+ λ− EC Tt−1 + ut
⎪
⎪
⎪
⎪
⎪
⎪ if xt < δ1
⎪
⎨
......
⎪
⎪y = φ ∗ + r β ∗ y + s α∗ + x + + q α∗ − x −
⎪
⎪
⎪
⎪ t 0 i=1 i t−i i=0 i t−i i=0 i t−i
⎪
⎪
⎪
⎪ +∗ + −∗ −
+ λ EC Tt−1 + λ EC Tt−1 + u t ∗
⎪
⎪
⎩
if xt > δ p
(10)
or
⎧  s q
⎪
⎪ yt = φ0 + ri=1 βi yt−i + i=0 αi+ xt−i+
+ i=0 αi− xt−i
−
⎪
⎪
⎪
⎪ +
+ λ+ EC Tt−1 + λ− EC Tt−1 −
+ ut
⎪
⎪
⎪
⎪
⎪
⎪ if EC Tt−1 < δ1
⎪
⎨
......
⎪
⎪ r s q
⎪
⎪ y = ∗
+ ∗
y + ∗ +
x +
+ ∗ − −
⎪
⎪ t φ0 i=1 βi t−i i=0 αi t−i i=0 αi x t−i
⎪
⎪
⎪
⎪

+ λ+∗ EC Tt−1+ 
+ λ−∗ EC Tt−1 −
+ u ∗t

⎪
⎪
⎪
⎩
if EC Tt−1 > δ p
(11)

In this case, not only do the values of x and ECT directly affect y (as in the
ARDL, PAM and ECM), but also they indirectly influence y through their effect
on the other explanatory variables.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
358 FREY AND MANERA

Models (10) and (11) allow us to define two new concepts of asymmetry.
Specifically, we identify the regime effect asymmetry (REA) as the situation where
the existence of more than one regime defined by the variable x is significant; if
it is not, we have a regime effect symmetry (RES). When the threshold variable
is given by the ECT, instead, we define a regime equilibrium adjustment path
asymmetry/symmetry (REAPA/REAPS).
Finally, since the vectorial models we are presenting in this work are multivariate
extensions of the uniequational specifications previously discussed, the same
asymmetries are supported by the multiequational models. In our survey we will
concentrate on Vector AutoRegressive (VAR), Vector Error Correction (VEC) and
Vector Regime Switching (VRS) models.

3. Early Asymmetric Models

3.1 Autoregressive Distributed Lags Models

The empirical literature on asymmetric price transmissions goes back to Farrel
(1952). This is the first attempt to empirically investigate the irreversibility behaviour
of the demand function of some habitual consumption goods. Farrel analyzes the
demand (Y) for tobacco, beer, wines and spirits in response to income (Z) and price
(X) variations, using a functional form of the type:
  a    a    b   b
Yt Zt Zt Xt Xt
=e c
et (12)
Yt−1 Z t−1 Z t−1 X t−1 X t−1
where
⎧ Z ⎧
⎨ Z t−1
t
= Zt
Z t−1
if Zt
Z t−1
≥1 ⎨Z t
= Zt
if Zt
<1
Z t−1 Z t−1 Z t−1
Z t (13)
⎩ Z t−1 = 1 otherwise, ⎩ Z t = 1 otherwise
Z t−1

with a, b and c indicating unknown parameters.

The hypothesis to be tested is whether an income increase, or a price decrease,
leads to an increase in consumption of habitual goods, while an income decrease, or
price increase, leaves the levels of consumption virtually unaltered. The empirical
findings seem to be inconclusive, but, in Farrel’s words, ‘they do suggest that
irreversibility may be quite an important factor in the change of tastes’ (p. 186).
It is worth noting that the model proposed by Farrell applies to the levels of
demand, income and price. If, instead, we consider its logarithmic transformation,
equation (12) takes the following form:

yt = c + a + z t+ + a − z t− + b+ xt+ + b− xt− +  t (14)

which is the typical asymmetric specification used in the literature.

At the end of the 1960s and during the 1970s, most of the studies on asymmetric
price transmission concentrate, not surprisingly, on agricultural goods. Tweeten and

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 359

Quance (1969) investigate the relationship between the level of output (y) and the
ratio between input and output prices (x) in the agricultural sector, using an indicator
variable to discriminate between positive and negative variations of x. Two different
models are estimated. The first is a two-equation system, one referring to the years
of price increases, the other to the years of decreasing prices:

yt = α0 + α+
xt+ +  t (15)

yt = α0 + α−
xt− +  t (16)

where x + is equal to x if its value has increased over the last year and zero otherwise
(vice versa for x − ).
Asymmetry is present if the null hypothesis of α+ = α− is rejected. The second
model combines the effects of increasing and decreasing prices in a single equation:

yt = α0 + α+
xt+ + α−
xt− +  t (17)

Using annual data for the period 1921–1966, the authors find some evidence of
asymmetry in the empirical results produced by the first model only.
Wolffram (1971) shows that the approach followed by Tweeten and Quance to
distinguish between periods of expansion and periods of reduction of the input/output
price ratio can lead to biased estimates. As a solution, Wolffram suggests redefining
xt+ and
the variables xt− as:
+


x 1 = x1

+

x2 =

x 1 + D(x2 − x1 )

+

x3 =

x 2 + D(x3 − x2 )
......
−


x 1 = x1
− −

x2 =

x 1 + (1 − D)(x2 − x1 )
− −

x3 =

x 2 + (1 − D)(x3 − x2 )
......

where D is a dummy variable which takes the value of 1 if the variation of the
input/output price ratio is positive, while it is equal to 0 otherwise.
The main difference between Wolffram’s approach and Tweeten and Quance’s
model is that the former explicitly considers the effect of cumulative variations in
the variable x, while the latter takes into account the direct impact of period-to-period
variations. Thus, it can be useful to divide the literature into two broad categories,
depending on whether the explanatory variable is defined according to Tweeten and
Quance or to Wolffram.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
360 FREY AND MANERA

3.1.1 Cumulative price variations

In 1977 Houck proposed a work on inventories and prices of milk and beans in the
US, which improves Wolffram’s approach in the following direction:

t 
t
+ −
yt − y0 = α0 t + α+ xt−i + α− xt−i + t (18)
i=0 i=1

It is immediately noticeable that the dependent variable in model (18) is no longer

y t , as in the previous studies, but the deviation of y t from its starting value y 0 .
Moreover, the model directly considers the impact of positive and negative variations
of x on y, cumulated from the first period
+ t (i = +t) up to−the current
t period (i = 0).
Hereafter, we define with Cxt−k = i=k xi , Cxt−k = i=k xi− , k ∈ [0, t],
the cumulative sums of price variations used in Wolffram’s approach models as the
independent variable.
When testing the null hypothesis α+ = α− using annual data, Houck finds that,
for the milk market only, the variation of the level of inventories over the sample
period asymmetrically depends on the contemporaneous impact of cumulative price
changes (COIA).
Young (1980) analyzes the specification of the consumer’s demand function
in response to price changes. In particular, he proposes three different kinds of
asymmetric behaviour. The first is a reparametrization of Hook’s (1977) model,
which accounts for the fact that a generic variable x can be expressed as the sum of
its initial value (at time t = 0) and
t all the+ positive
t and negative deviations observed
−
up to current time t(xt = x0 + i=0 xt−i + i=0 xt−i ):
yt − y0 = α0∗ + α+ xt + α∗ cxt− +  t (19)
where α∗ = α+ − α− . Model (19) clearly assumes that the demand may react
asymmetrically to the total impact of cumulative price increases and decreases.
However, Young suggests that the demand curve would rather show an asymmetric
behaviour when prices are ‘at unprecedently low or high levels’ (p. 178). The
definition of xm i and xM i as the minimum and the maximum price at time i yields
the two alternative models:

t
yt − y0 = α0∗ + α+ xt + α∗ xm t−i +  t (20)
i=0

t
yt − y0 = α0∗ + α+ xt + α∗ x Mt−i +  t (21)
i=0

Models (20) and (21) represent the first attempt to describe asymmetry in terms
of the response of y to the deviation of x above or below a given threshold, rather
than to the sign of x itself. This approach has become very popular in the empirical
literature.
All studies discussed up to this point analyze the asymmetric effects of prices
on the demand levels of different goods. Ward (1982) is the first contribution

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 361

which shifts the attention to the core topic of this survey, that is, the transmission
mechanism between prices (Table 6).
Specifically, Ward models the impact of wholesales prices (ws) on retail (rt) and
FOB prices (fb), using monthly data of different types of fresh vegetables in the US
market, and dividing the sample in l seasons of t observations each. With respect to
Houck (1977), Ward does not simply express the current value of rt or fb as functions
of the positive and negative sum of the variation of ws over the observation period;
rather he allows the effect of those variables to persist over time:

3 
3
r tlt = φ + α+jlt Cwslt−
+
j D jlt + α−jlt Cwslt−
−
j D jlt + lt (22)
j=0 j=0

where D jlt are dummy variables used to identify the existence of Cw lt−J , as
the required lagged quantities are not available for the first three observations of
each season l. Hence, equation (22) is a generalization of Houck’s model, since it
defines a new kind of asymmetry, namely the distributed lag effect of the cumulative
variations.
Since equation (22) is defined over a very large parameter space, Ward actually
analyzes a simplified version of it, along Young’s line:

3 
3
 
r tlt = φ + α+jl (wslt− j − wsl0 )D jl + α−jl − α+jl Cwslt−
−
j D jl + lt
j=0 j=0 (23)
The actual value of rt is now expressed as a function of the deviation of ws from
its initial value and of the cumulative impact of its negative variations; the effect of
positive deviations is not made explicit.
Starting from equation (23), the parameter space can be easily reduced if the
impact of each variable depends on the time considered:
α+jl = γ1 + γ2 ξ j
(24)
α−jl − α+jl = γ3 + γ4 ξ j
√
where ξ j is a known coefficient, equal to 3 j, which forces the impact of cumulative
prices ws to decrease with the time lag. Substituting model (24) into equation (23),
we eventually obtain Ward’s specification:

3 
3
r tlt = φ + γ1 (wslt− j − ws0 )D jl + γ2 (wslt− j − ws0 )ξ j D jl
j=0 j=0

3 
3
− −
+ γ3 Cwslt− j D jl + γ4 cwslt− j ξ j D jl + lt (25)
j=0 j=0

Model (25) depends on four parameters only; the significance of γ 3 and γ 4 is

supportive of asymmetry in the distributed lag effect of cumulative price variations.
It is important to point out that the biunivocal correspondence between the
coefficients α−jl , α+jl , γ 1 , γ 2 , γ 3 and γ 4 implies α−jl = α+jl if and only if γ 3 +
γ 4 ξ j = 0; DLEA occurs when, for at least one j ∈ [0, 3], γ 3 + γ 4 ξ j = 0. If

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 C

362

Table 6. Summary of Asymmetries by Model – ARDL.

2007 The Authors

Journal compilation 
Asymmetries
Data
Year Authors Journal Model COI DLE CUI RT EAP Country Product Sample Frequency

1982 Ward AJAE ARDLcu Y1 Y1 N1 – – USA Agricultural – Monthly

1987 Kinnucan, Forker AJAE ARDLcu – Y Y1 – – USA Alimentary 1971–1981 Monthly
1991 Karrenbrock WP ARDLpp Y1 Y1 N1 – – USA Gasoline 1983–1990 Monthly
1991 Punnyawadee et al. CJAE ARDLpp N1 N1 N1 – – C Alimentary 1965–1989 Weekly

C 2007 Blackwell Publishing Ltd

1991 Punnyawadee et al. CJAE ARDLpp Y Y N1 – – C Alimentary 1965–1969 Weekly
1993 Balabanoff OR ARDLpp – Y1 N1 – – F, I, J, UK Gasoline 1985–1992 Monthly
1993 Balabanoff OR ARDLpp Y1 Y1 N1 – – G, US Gasoline 1985–1992 Monthly
1993 GAO WP ARDLpp N1 N1 N1 – – USA Gasoline 1984–1991 Weekly
1993 GAO WP ARDLpp Y1 N1 N1 – – USA Gasoline2 1984–1991 Weekly
1993 GAO WP ARDLpp N1 N1 N1 – – USA Gasoline3 1984–1991 Weekly
FREY AND MANERA

1994 Griffith, Piggott AgE ARDLcu Y1 Y1 N1 – – USA Alimentary4 1971–1988 Monthly

1994 Griffith, Piggott AgE ARDLcu Y1 Y1 Y1 – – USA Alimentary5 1971–1988 Monthly
1994 Shin OR ARDLpp N1 – – – – USA Gasoline 1986–1992 Monthly
1995 Mohanty et al. CJAE ARDLcu Y Y Y1 – – C, A, Au, E Agricultural 1980–1990 Monthly
1995 Powers Ag ARDLcu – Y1 Y1 Y1 – USA Agricultural6 1986–1992 Weekly
1995 Powers Ag ARDLcu – Y1 Y1 N1 – USA Agricultural7 1986–1992 Weekly
1995 Powers Ag ARDLcu – – N1 N1 – USA Agricultural8 1986–1992 Weekly
1995 Zhang et al. Ag ARDLcu – Y1 N1 – – USA Agricultural 1984–1992 Monthly

14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 
1996 Duffy-Deno EE ARDLpp Y1 Y1 Y1 – – USA Gasoline 1989–1993 Weekly
1996 Duffy-Deno EE ARDLpp Y1 Y1 N1 – – USA Gasoline9 1989–1993 Weekly

C 2007 The Authors

Journal compilation 
1998 Balke et al. WP ARDLpp Y Y – – – USA Gasoline 1987–1996 Weekly
1999 EIA WP ARDLpp Y1 Y1 – – – USA Gasoline 1992–1998 Weekly
2000 Worth WP ARDLcu – Y1 Y1 – – USA Agricultural10 1980–1999 Monthly
2000 Worth WP ARDLcu – Y1 N1 – – USA Agricultural11 1980–1999 Monthly
2001 Parrott JA ARDLcu N1 N1 N1 – – USA Agricultural 1988–1993 Weekly
2002 Aguiar, Santana Ag ARDLpp – Y1 Y1 – – B Agricultural12 1987–1998 Monthly
2002 Aguiar, Santana Ag ARDLpp Y1 Y1 Y1 – – B Agricultural13 1987–1998 Monthly
2002 Aguiar, Santana Ag ARDLpp Y1 Y1 N1 – – B Agricultural14 1987–1998 Monthly

C 2007 Blackwell Publishing Ltd

2002 Aguiar, Santana Ag ARDLpp – Y1 N1 – – B Agricultural15 1987–1998 Monthly
2002 Aguiar, Santana Ag ARDLpp Y1 N1 Y1 – – B Agricultural16 1987–1998 Monthly
2003 London Economics WP ARDLpp Y Y – – – Au, D, F, G, Ir Alimentary 1985–2003 Monthly
2003 N, Sp, UK
2003 Bunte, Zachariasse WP ARDLpp – – – N1 – N Alimentary17 1990–1997 –
2003 Bunte, Zachariasse WP ARDLpp – – – Y1 – N Alimentary18 1990–1997 –
2004 Girapunthog et al. WP ARDLcu N1 N1 N1 – – USA Agricultural19 1975–1998 Monthly
2004 Girapunthog et al. WP ARDLcu Y1 Y1 Y1 – – USA Agricultural20 1975–1998 Monthly
2004 Girapunthog et al. WP ARDLcu Y1 Y1 N1 – – USA Agricultural21 1975–1998 Monthly

Y (N) indicates the presence (absence) of asymmetry.

1
denotes that the presence or absence of asymmetry is supported by a statistical test.
2
Crude–wholesale relationship during a market shock; 3 Wholesale–retail relationship during a market shock; 4 Pork; 5 Beef and lamb; 6 Wholesale retail;
7
FOB retail; 8 FOB wholesale; 9 During a market shock; 10 Carrots, tomatoes; 11 Celery, lettuce, onions, potatoes; 12 Beans, tomatoes; 13 Milk; 14 Rice;
15
Onions; 16 Coffee; 17 Farm–retail transmission for beef, poultry and chips; Farm wholesale transmission for pork, beef, potatoes and chips; 18 Farm retail
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION

transmission for pork and potatoes; Farm wholesale transmission for poultry;19 Producer retail; 20 Producer wholesale; 21 Wholesale retail.
363

14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
364 FREY AND MANERA

just one of the two coefficients γ 3 and γ 4 is different from zero, DL asymmetry is
clearly implied; but if γ 3 and γ 4 are both significant and have opposite sign, there
can be a value of ξ j such that γ 3 + γ 4 ξ j = 0. However, this result can occur
for just one particular value of ξ j and, consequently, the significance of γ 3 and γ 4
always implies DLEA. Note that relationships (24) also imply that COIA occurs
when γ 3 = 0.
To summarize, Ward finds out the existence of asymmetry in the contemporaneous
and distributed lag effects of cumulative wholesale prices variations on both FOB
and retail prices. Results concerning the presence of DLEA are also supported
by the calculated mean lags, which are asymmetric in most of the cases. Finally,
Ward tests the hypothesis of symmetry in the cumulative impact of cws+ and
cws− , finding that CUI asymmetry affects only the wholesale–retail transmission
mechanism.
Ward’s approach has been extended by several authors. For instance, in 1987
Kinnucan and Forker analyze the farm(fm)–retail price transmission for major dairy
products in the US, using monthly data over the period January 1971–December
1981. In order to make explicit the distributed lag impact of cumulative price
variations, they choose not to use Ward’s final specification (25), but to directly
estimate equation (22), where they also introduce the role of marketing costs (mk):

s 
q
r tt − r t0 = φt + αi+ C f m i+ + α−j C f j− + mkt − mk0 +  t (26)
i=1 j=1

Although the authors estimate the coefficients of the cumulative variables Cfm+j
and Cfm−j , they do not test the hypothesis αi+ = α−j , ∀i, j; instead, they investigate
the presence of DLEA by simply computing the mean lags, which appear to be
different depending on the sign of farm price variations. However, this result should
not be interpreted as reliable evidence in favour of asymmetry, especially given
that no statistical test is provided. As a consequence, any consideration about the
presence of distributed lag asymmetry is inconclusive, as the number of significant
lags is symmetric, which does not exclude  either DLEA or DLES. Kinnucan and
αi+ = j=1 α−j , which is rejected
s q
Forker formally test only the null hypothesis i=1
for all considered goods in favour of CUIA.
The specifications proposed by Houck (1977), Ward (1982) and Kinnucan and
Forker (1987) constitute a complete overview of the types of models proposed in
the literature to evaluate the impact of cumulative variations of a price x on a price
y. As a result, the empirical work based on Wolffram’s approach can be related to
either one of these three models.
Griffith and Piggott (1994) use Kinnucan and Forker’s specification to analyze
the relationships between retail–wholesale prices, farm–wholesale prices and farm–
retail prices for the Australian beef, lamb and pork markets, using monthly data
from January 1971 to December 1988. s +
As in Kinnucan and Forker, the authors test the null hypothesis i=1 αi =
q −
j=1 α j , which is never rejected for the pork market, whereas it is not rejected only
for the farm–wholesale price transmission and for the farm–retail price relationship

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 365

in the beef and lamb markets, respectively. In the light of these findings, Griffith
and Piggott suggest that ‘asymmetrical price response is a strategy used by beef and
lamb retailers and wholesalers to adjust to changing input prices but not by pork
retailers and wholesalers’ (p. 307).
However, a closer analysis of these results leads to a different conclusion, namely
that CUIA is not really relevant in the Australian markets for beef, lamb and pork,
while DLEA is. As a matter of fact, except for the farm–wholesale price relationship
in the lamb market, all models exhibit an asymmetric number of lags for the positive
and the negative cumulative price variations, which implies an asymmetric behaviour
of their distributed lag effects. Furthermore, the asymmetric lag structure leads
to conclude that all three price relationships in the beef market, the farm–retail
transmission in the markets for beef and pork, as well as the wholesale–retail price
relation for the pork market are characterized by COIA. Actually, in the first four
cases only either the positive or the negative contemporaneous impact of upstream
prices results are statistically significant. In the latter case, the presence of CUIA,
together with the structure of the significant lags, shows that the cumulative impact
of the cumulative positive variations of farm prices up to time t − 2 is equivalent to
the contemporaneous impact of the negative variations, which consequently supports
the presence of COIA.
Kinnucan and Forker’s (1987) approach is also employed by Powers (1995), who
studies the impact of FOB prices on retail and wholesale prices, as well as the effect
of wholesale prices on retail prices, together with the impact of hauling costs on
both wholesale and retail prices, in 12 US cities. The analysis focuses on iceberg
lettuce and relies on weekly data from 9 March 1986 to 30 August 1992. The effect
of hauling costs is tested for the presence of CUIS, which, as a whole, cannot
be rejected in both markets. As far as the price transmission is concerned, the
author provides detailed information about three types of asymmetries. First, he
analyzes the cumulative impact of the cumulative price variations and finds out that
CUIA generally characterizes the wholesale–retail and the FOB–retail transmission
mechanisms, but not the FOB–wholesale price relation. Second, Powers looks for
asymmetries in the reaction time by evaluating the median lags, which support RTA
only in the wholesale–retail market. Finally, the author shows that the lag structure
of the retail–wholesale and retail–FOB relationships is often asymmetric across the
12 cities considered, suggesting the existence of DLEA in both markets.
Zhang et al. (1995) apply Kinucan and Forker’s model to the wholesale
prices of peanuts and the price of peanut butter in the US, using monthly data
over the period January 1984–July 1992. In this market the hypothesis of CUI
symmetry of the cumulative wholesale price variations cannot be rejected. As
for the DL impact, the mean lags computed by the author look symmetric and
do not provide any valid information about the presence or absence of price
asymmetries, as shown in Section 2. However, the number of significant lags in
the model differs between regimes of rising and falling prices, which clearly implies
DLEA.
Worth (2000) studies the relationship between FOB shipping point prices and
retail prices of some fresh vegetables. In particular, he applies Kinnucan and Forker’s

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
366 FREY AND MANERA

model to monthly data from January 1980 to May 1999. It is notable that Worth
chooses not to use ‘earlier data, though available, because of changes in agricultural
markets since 1960’ (p. 6), that might bias the analysis. The author tests only CUIS,
which leads to the conclusion that ‘carrots and tomatoes are the only commodities
which show evidence of price asymmetry’ (p. 8). In the light of our taxonomy
of asymmetries, however, Worth’s results draw a different picture. CUIA, in fact,
characterizes only a subset of vegetables (i.e. carrots and tomatoes), while all goods
under scrutiny are affected by DLEA since, as the author points out, the number of
significant lags for price increases and decreases is always asymmetric.
Parrott et al. (2001) analyze the transmission mechanism between retail and FOB
shipping point prices, weighted by the volume of shipments, in the US fresh tomato
market, using weekly data over the period June 1988–December 1993. By estimating
Kinnucan and Forker’s model, the authors evaluate the presence of CUI, COI and
DLE symmetry, all of which cannot be rejected.
The US fresh tomato market is also studied by Girapunthong et al. (2004)
who, unlike Parrott et al., use Ward’s specification. In particular, they focus on
the producer–retail, producer–wholesale and wholesale–retail price relationships
between May 1975 and February 1998, using monthly data. The authors test
for CUIS and for both COIS and DLES, which, as we know, are biunivocally
linked to the significance of γ 3 and γ 4 in Equation (25). The empirical findings
suggest that no asymmetry occurs in the producer–retail market, while all types
of asymmetries characterize the producer–wholesale price transmission. Finally, the
different wholesale–retail price relationships exhibit both COIA and DLEA. Note
that the estimated mean lags confirm the existence of DLEA in the producer–retail
market but appear to be symmetric at both the producer–wholesale and wholesale–
retail stages. This result, however, is not surprising, since mean lag symmetry does
not imply either DLEA or DLES.
Within the class of Wolffram’s types of models, Mohanty et al.’s (1995)
contribution forms a separate category, as it deals with spatial asymmetry. So far
we have analyzed the price transmission between the different levels of a market
chain, although it is also possible to study the relationship between the same price
in different countries. The authors investigate the relationship between US FOB
wheat prices and the correspondent prices in Canada, Australia, Argentina and the
European Union and, for each pair of countries, estimate a Kinnucan and Forker’s
(1987) type of model. In particular, the authors, who consider monthly data from
January 1980 to June 1990, propose two symmetry tests. The first is directed to
test CUIS, while the second aims at testing the joint null hypothesis of COIS and
DLES, which, if refused, denotes the presence of at least one of the two. The results
show that both hypotheses can be rejected, that is, CUIS and at least either DLES
or COIS are present in the data.

3.1.2 Period-to-period price variations

Unlike Wolffram’s approach, which evaluates the effect of cumulative price
variations, Tweeten and Quance (1969) study the impact of period-to-period

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 367

price increases and decreases. For this reason, the results on DL, CUI and COI
asymmetries obtained by the papers which follow this approach cannot be directly
compared with those described in Section 3.1.1.
The most effective way to introduce Tweteen and Quance’s (1968) model is
to present the generalization proposed by Balke et al. (1998), which includes
simple distributed lag effects. In their work, the authors study the transmission
mechanism between prices at different levels of the gasoline distribution chain using
a specification which, for the relationship spot(sp)–wholesale prices, can be written
as:
s  s 
r 
r
wst = φ + αi spt−i + αi+ sp
t−i + βi wst− j + βi+ w
s +
t− j +  t (27)
i=0 i=0 j=1 j=1

where, as in Tweeten and Quance’s model, sp is equal to sp if the value of sp has

increased over the last period and it is zero otherwise (identical considerations apply
to ws).
In the context of Equation (27), it is possible to test for symmetry by checking
the significance of specific coefficients. For instance, the impact of negative spot
price variations is described by the αi coefficients, while the effect of positive spot
price changes is measured by the sum αi + αi+ .
If the αi+ coefficients are statistically equal to zero, then the impact of spot price
increases and decreases is the same (say α); otherwise, the effect of spot prices
depends on their sign through αi+ .
The authors use US weekly data from January 1987 to August 1996 and consider
all possible upstream–downstream price transmission mechanisms among crude oil
and gasoline prices (namely, spot, wholesale and retail), with and without taxes.
In particular, they test the joint null hypothesis α+ +
1 = 0, . . . . , αs = 0, which, if
rejected, denotes the existence of, at least, either COIA or DLEA. The results show
that crude prices have an asymmetric impact on retail prices (with and without
taxes), spot prices asymmetrically affect wholesale and retail prices with taxes and,
finally, asymmetry occurs also in the response of ex-taxes retail prices to wholesale
costs.
Karrenbrock (1991) proposes a specification to test for symmetry in the impact
of period-to-period wholesale price variations on retail prices:

s 
q
r tt = φ + αi+ wst−i
+
+ αi− wst−
−
j + t (28)
i=1 j=1

Using US monthly data over the period January 1983–December 1990, Kar-
renbrock tests the presence of CUI, COI and DL asymmetries in themarket of
premium,
q regular leaded and unleaded gasoline. The null hypothesis i=1 s
αi+ =
−
j=1 αi is never rejected, that is, the cumulative effect of wholesale price variation
is symmetric. On the contrary, COIA characterizes the response of the premium
gasoline price. As for the distributed lag impact, Karrenbrock provides both a test of
persistence and a direct test of equality among the coefficient at each time lag. In gen-
eral, Karrenbrock shows that, although the effects of wholesale price increases and

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
368 FREY AND MANERA

decreases exhibit the same degree of persistence,1 their distributed lag effect is
clearly asymmetric.
The distributed lag structure of the impact of a price on another price may also be
estimated by a second model, described in a report of the General Accounting Office
(GAO) (1993). GAO investigates the crude–wholesale and the wholesale–retail price
transmission in the US on weekly data from January 1984 to March 1991. If we
concentrate on the relationship between wholesale and retail prices, GAO proposes
an extension of the Balke et al. (1998) model where the impact of positive and
negative wholesale price variations is explicitly considered:
s
 
r tt = φ + αi wst−i + αi+ wst−+
j + t (29)
i=1
The empirical findings suggest that both crude–wholesale and wholesale–retail
markets do not exhibit any asymmetric behaviour to upstream price variations,
irrespective of the type of impact (i.e. contemporaneous, lag distributed and cumula-
tive), as none of the αi+ is statistically significant. GAO exploits this result to argue
that, even though the price transmission in the US gasoline market is, on average,
symmetric, nevertheless, it could well be asymmetric in response to market shocks.
According to the author’s definition, a ‘market shock’ is ‘an event or rumour that
substantially alters the actual or expected supply of and demand for crude oil or
petroleum products’ (p. 60). In this sense, price asymmetries may arise since, after
a shock, existing inventories are generally sold at their stock-induced market value,
independently of their acquisition costs. Consequently, GAO extends this model by
including additional economic variables which can help explain the different price
adjustment processes during periods characterized by the presence or the absence
of economic shocks. A list of these variables includes fuel stocks, the petroleum
refinery capacity utilization rate and a logistic time trend, which accounts for the
increased flow of market information over time. Clearly, all these variables affect
the crude–wholesale price transmission, while only fuel stock has an impact on the
wholesale–retail relationship.
If we concentrate on the wholesale–retail transmission and introduce a pair of
dummy variables to account for shock-induced price increases and decreases:

1 if wst > δ
D1t =
0 otherwise
 (30)
1 if |wst | > δ and wst < 0
D2t =
0 otherwise
the modified version of model (29) is:
s

r tt = φ + αi wst−i + αi+ wst−i
+
+ γ1i (D1t + D2t ) f st−i wst−i
i=1

+ γ2i D1t f st−i wst−i +  t (31)
The empirical findings suggest that, after a market shock, wholesale prices
respond asymmetrically to crude prices, while symmetry characterizes the

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 369

wholesale–retail relation. In particular, the crude–wholesale market seems to be

characterized by COIA.
A very interesting approach is proposed by Asplund et al. (2000). The authors
analyze the Swedish gasoline market, focusing on how Rotterdam spot gasoline
prices (sp), exchange rates (er) and taxes (tx) are transmitted to the local retail prices
of leaded premium gasoline. All previous models have considered the dependent
variable as function of a set of explanatory variables which are split into positive
and negative values. Asplund et al. reverse this approach by distinguishing between
positive and negative values of the dependent variable. They also investigate the
effects of using the cost of gasoline expressed in local currency as explanatory
variable, rather than decomposing it into variations of the spot price and of the
exchange rate. To account for all these factors, the authors propose the following
two equations, which are independently estimated for increasing and decreasing
retail prices:

s
r tt |r tt >,<0 = φ + αi (spt−i ert−i ) + γ t xt +  1t (32)
i=1


s 
q
r tt |r tt >,<0 = φ + αi spt−i ert−i + α j spt− j ert− j + γ t xt +  2t (33)
i=1 j=1

Daily data from January 1980 to December 1996 highlight that the impact of cost
variations is different depending on whether they are the consequence of spot price
changes or exchange rate changes. As far as asymmetry is concerned, the results
confirm that rising and falling retail prices respond differently to spot prices and
exchange rate variations.
The studies by Balke et al., Karrenbrock, GAO and Asplund et al. provide a
complete overview of the asymmetric ARDL models proposed to date. Starting from
these contributions, many other authors have studied the transmission mechanism
between prices of different goods.
Balabanoff (1993) investigates the transmission mechanism between CIF crude
oil and the composite barrel of retail prices in France, Germany, Italy, Japan, the
UK and the US. Data are monthly and cover the period 1985–1992. The author
uses Karrenbrock’s (1991) model and tests for symmetry in the total cumulative
effect of crude prices, which is never rejected. Conversely, the persistence of crude
price variations is clearly asymmetric for all considered nations, and it implies
the presence of DLEA. Even if the author tests directly only CUIA and DLA,
the estimation results provide some indirect information about COIA. Actually,
the US and Germany show significant lags for crude price decreases only; as a
consequence, the symmetry in the cumulative impact clearly suggests asymmetry in
the contemporaneous effect of the crude price variations in both countries.
Shin (1994) again applies Karrenbrock’s model to the analysis of the US crude–
wholesale price transmission, using monthly data over the period 1986–1992. In
particular, he starts with a distributed lag structure, but, since it is not statistically
significant, he finally estimates a model with only contemporaneous effects. His

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
370 FREY AND MANERA

empirical findings show that crude oil price variations have a symmetric impact on
the wholesale market (i.e. they are characterized by COIS).
Another interesting work is due to Duffy-Deno (1996), who combines
Karrenbrock’s (1991) model with GAO’s market shock analysis. In particular, the
consequences on asymmetry detection of considering or ignoring the presence of a
market shock are investigated. The author focuses on the Salt Lake City gasoline
market, using weekly data for the period 1989–1993 and shows that, when market
shocks are ignored, COIA, DLEA and CUIA generally emerge. Conversely, when
the model is extended to incorporate the effect of unusual price variations, wholesale
price decreases exhibit a longer persistence and the cumulative impact of wholesale
prices is now symmetric.
In 1999, the Energy Information Administration (EIA) investigates the upstream–
downstream price transmission at the different levels of the gasoline distribution
chain in the US Midwest, using weekly data for the period October 1992–June
1998. Though they do not consider the presence of market shocks, the authors apply
GAO’s asymmetry specification and, for each price relationship, estimate a model
similar to Equation (29). A joint and a single test for the significance of the α+
coefficients provide strong evidence of asymmetries. Although the DLE of crude
prices on each downstream market is always symmetric, COIA affects the response
of pipeline and rack prices. Furthermore, COIA is also detectable in the response
of rack prices to pipeline costs, while DLEA affects the response of retail prices to
all upstream markets (with the exception of crude oil) and the Gulf Coast–Chicago
pipeline transmission.
Aguiar and Santana (2002) follow Karrenbrock’s (1991) approach to analyze the
transmission mechanism between farm and retail prices of some agricultural products
in Brazil. Monthly data over the period January 1987–June 1998 show that DLEA
is a relevant issue in all markets, coffee excluded, where farm costs show only
an instantaneous effect on retail prices. Cumulated Impact Asymmetry (CUIA) is
another relevant topic in the Brazilian agricultural market, since the null of CUIS is
not rejected in two cases only (namely, onions and rice). Finally, COIA affects milk,
rice and coffee markets, as the estimated coefficients do not support any reliable
conclusion for the remaining products.
In a report by London Economics (2004) it is possible to find another application
of Karrenbrock’s model. The authors investigate the mutual relationship between
retailer and producer prices of different types of vegetables, fruits, meat and dairy
products in Austria, Denmark, France, Germany, Ireland, the Netherlands, Spain and
the UK. In particular, the prices of apples, carrots, potatoes, beef, lamb, flour, eggs
and chicken are investigated. Results show that asymmetry (either COIA or DLEA, at
least) mainly occurs in the dairy market, where it affects the milk price transmission
in the UK, France and Denmark, the cheese prices in France and Denmark and,
finally, the prices of butter in France and the UK. However, asymmetry is also
evident in the markets of Dutch beef, French bread and Danish chicken.
In 2003, Bunte and Zachariasse analyze how farm prices are transmitted to
wholesale and retail prices in the Netherlands. In particular, they apply Karrenbrock’s
approach to evaluate the effects of positive and negative price shocks at the producer

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 371

level in the markets of pork, beef, poultry, table potatoes and chips. Data from
1990 to 1997 show that RTA characterizes the farm–retail relationship in the pork
and table potato markets, as well as the farm–wholesale transmission of poultry
prices.
Finally, Punyawadee et al. (1991) propose a model for spatial asymmetry applied
to Canadian data. This paper investigates whether the price of pork in Ontario has an
asymmetric impact on the correspondent price in Alberta, over the period January
1965–December 1989. The authors use weekly data and consider six subperiods
to capture potential changes in the price relationships. The model proposed follows
Karrenbrock’s specification and allows to test for COIS, DLES and CUIS. The results
clearly show that the cumulated impact of Ontario price variations on the Alberta
pork market is always symmetric. On the contrary, a joint test for the equivalence of
the effects of price increases and decreases at each lag shows that COIS and DLES
are present in all subsamples, except the time interval from January 1965 to October
1969.

4. Equilibrium Correction Approaches

4.1 Partial Adjustment Models

During the 1990s, the analysis of price asymmetries was enriched by the
partial adjustment models (PAM), which describe the adjustment process
of a price variable in response to deviations from a given target level
(Table 7).
A widely-cited example is provided by Bacon (1991), who studies the relation
among ex-refinery petroleum product prices (rf ), retail gasoline prices net of taxes
and exchange rate (er) using a quadratic PAM model:
 2  
r f t−1 r f t−1
r tt = r tt−1 + φ1 γ + δt + − r tt−1 + φ2 γ + δt + − r tt−1 + ηt
ert−2 ert−2
(34)

which implies the following LR relationship:

r f t−1
r tt∗ = γ + δt + + t (35)
ert−2

Using fortnightly UK data from 1982 to 1989, Bacon shows that retail prices
rise much more slowly after an increase of ex-refinery prices than they fall after a
decrease. Moreover, Bacon also provides detailed information about the mean lags
which are asymmetric and support the existence of DLEA.
Another interesting application is found in Salas (2002). Using weekly data for the
period January 1999–February 2002, the author examines the relationship between
retail and crude (cr) prices in the Philippine market and introduces a specification

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 C

372

Table 7. Summary of Asymmetries by Model – PAM, ECM.

2007 The Authors

Journal compilation 
Asymmetries
Data
Year Authors Journal Model COI DLE CUI RT EAP MEAP Country Product Sample Frequency

1991 Bacon EE PAM – Y1 – – Y1 UK Gasoline 1982–1989 Biweekly

1991 Manning AE ECMeg – Y1 – N1 – – UK Gasoline 1973–1988 Monthly
1991 Manning AE ECMsw – Y1 – N1 – – UK Gasoline 1973–1988 Monthly
1991 Norman, Shin WP PAM – – – – N1 – USA Gasoline 1984–1992 Weekly

C 2007 Blackwell Publishing Ltd

1994 Shin OR PAM – – – – N1 – USA Gasoline 1986–1992 Monthly
1996 Borenstein, Shepard WP ECMsw – – – Y1 – – USA Gasoline 1982–1991 Monthly
1997 Arden et al. WP ECMeg – – – – N1 – UK Manufacturing 1970–1996 Quarterly
1997 Arden et al. WP ECMsw – – – – Y1 – UK Manufacturing 1970–1996 Quarterly
1997 Borenstain et al. QJE ECMsw Y – – Y1 – – USA Gasoline2 1986–1992 (Bi/)Weekly
1997 Borenstain et al. QJE ECMsw Y – – N1 – USA Gasoline3 1986–1992 (Bi/)Weekly
FREY AND MANERA

1998 Balke et al. WP ECMsw Y1 Y1 – – – – C Gasoline 1987–1996 Weekly

1998 Eltony IJER ECMsw Y1 – – – – – UK, USA Gasoline 1980–1996 Monthly
1998 Reilly, Witt EE ECMsw Y1 – – – – – UK Gasoline 1982–1995 Monthly
1998 Von Cramon-Taubadel ERAE ECMeg – – – Y1 Y1 – G Alimentary 1990–1993 Weekly
2000 Abdulai JDE ECMth – – – Y1 – Y1 Gn Agricultural 1980–1997 Monthly
2000 Asplund et al. SJE ECMeg Y1 Y1 – – – – Sw Gasoline 1980–1996 Monthly
2000 Berardi et al. WP ECMeg Y Y – Y1 N1 – I Gasoline 1996–2000 Weekly
2000 Peltzman JPE ECMsw – – Y1 – – – USA Various 1978–1996 Monthly

14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 
2001 Hassan, Simioni WP ECMth – – – – Y1 – F Agricultural4 – –

C 2007 The Authors

Journal compilation 
2001 Hassan, Simioni WP ECMth – – – Y1 – Y1 F Agricultural5 – –
2002 Abdulai AE ECMth – – – Y1 – Y1 S Agricultural 1988–1997 Monthly
2002 Salas PRE PAM – – – – Y1 – Ph Gasoline 1999–2002 Weekly
2002 Salas PRE ECMeg – – Y1 – – – Ph Gasoline 1999–2002 Weekly
2002 Eckert CJE ECMsw Y1 – – Y1 – – C Gasoline 1989–1994 Weekly
2003 Bachmeier, Griffin RES ECMeg N1 N1 N1 N1 N1 – USA Gasoline 1985–1998 Daily
2003 Bachmeier, Griffin RES ECMsw – – – Y1 – – USA Gasoline 1985–1998 Weekly
2003 Bachmeier, Griffin RES ECMsw – – – Y1 – – USA Gasoline 1985–1998 Daily

C 2007 Blackwell Publishing Ltd

2003 Bettendorf et al. EE ECMeg Y1 Y1 – Y1 – – N Gasoline6 1996–2001 Weekly
2003 Bettendorf et al. EE ECMeg N1 N1 N1 N1 – – N Gasoline7 1996–2001 Weekly
2003 Conforti et al. WP ECMeg – – – – Y1 – World Agricultural 1989–2001 Monthly
2003 Conforti et al. WP ECMeg N1 N1 N1 – N1 – World Agricultural 1990–2001 Monthly
2003 Galeotti et al. EE ECMeg Y1 – – N1 Y1 – I, F, Sp, Gasoline 1985–2000 Monthly
G, UK
2003 Gonzales et al. WP ECMth – – – N1 N1 N1 F Alimentary 1988–1999 Monthly
2003 V. Cramon Taubadel, WP ECMeg – – – – N1 – G Alimentary8 1995–2000 Weekly
Meyer
2003 V. Cramon Taubadel, WP ECMeg – – – – Y1 – G Alimentary9 1995–2000 Weekly
Meyer
2003 London Economics WP ECMeg – – – – Y1 – D, UK Agricultural 1985–2003 Monthly
2003 NDA WP ECMeg – – – Y1 – – SA Agricultural 2000–2003 Monthly
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION
373

14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 C

Table 7. Continued.
374

Asymmetries
Data

2007 The Authors

Year Authors Journal Model COI DLE CUI RT EAP MEAP Country Product Sample Frequency

Journal compilation 
2004 Conforti WP ECMeg – – – – Y1 – World Agricultural 1969–2001 Annual
2004 Contin et al. WP ECMsw Y1 Y1 – Y1 – – Sp Gasoline 1993–2002 Weekly
2004 Deltas WP ECMeg Y Y – Y1 – – USA Gasoline 1998–2002 Monthly
2004 Deltas WP ECMsw Y Y – Y1 – – USA Gasoline 1998–2002 Monthly
2004 Krivonos WP ECMeg N1 – – N – – Africa Agricultural 1984–1990 Monthly
2004 Krivonos WP ECMeg Y1 – – Y – – Africa Agricultural 1990–2003 Monthly

C 2007 Blackwell Publishing Ltd

2004 Verlinda WP ECMsw – – – Y1 – – USA Gasoline 2002–2003 Weekly
2005 Grasso, Manera WP ECMeg Y1 Y1 – – Y1 – F, G, I, Gasoline 1985–2003 Monthly
Sp, UK
2005 Grasso, Manera WP ECMth – – – – – Y1 F, G, I, Gasoline10 1985–2003 Monthly
Sp, UK
2005 Grasso, Manera WP ECMth – – – – – N1 F, G, I, Gasoline11 1985–2003 Monthly
FREY AND MANERA

Sp, UK
2005 Kaufmann, Laskowski EP ECMeg – – – – N1 – USA Gasoline12 1986–2002 Monthly
2005 Kaufmann, Laskowski EP ECMeg – – – – Y1 – USA Gasoline13 1986–2002 Monthly
2005a Radchenko EE ECMeg – – – Y1 – – USA Gasoline 1991–2003 Weekly
2005b Radchenko EE ECMeg – – – Y1 – – USA Gasoline 1991–2003 Weekly

Y (N) indicates the presence (absence) of asymmetry.

1
denotes that the presence or absence of asymmetry is supported by a statistical test.
2
cr → r t, cr → s p, ws → r t; 3 s p → ws; 4 Chicory; 5 Tomatoes; 6 Weekly data refer to Monday, Thursday, Friday; 7 Weekly data refer to Tuesday,
Wednesday; 8 Aggregated data; 9 Non-aggregated data; 10 Crude–retail transmission in F, G, I, Sp, UK; Crude–spot relationship in I, Sp; 11 Spot-retail
transmission in F, G, I, Sp, UK; Crude–spot relationship in F, G, UK; 12 Motor gasoline and heating oil crude-refinery transmission in US; Motor gasoline
refinery–retail transmission in US, except California, Lousiana and Idaho; 13 Heating oil refinery–retail transmission in US; Motor gasoline refinery–retail
transmission in California, Lousiana and Idaho.

14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 375

which is quite different from Bacon’s model:

 +  −
r tt = φ + r tt∗ − r tt−1 + φ − r tt∗ − r tt−1 +  t (36)

The effect of positive and negative deviations from the target level is now made
explicit, and the presence of asymmetry can be detected by testing the null hypothesis
φ + = φ − . In the definition of the equilibrium level, r ∗t , Salas considers an 8-week
lag between retail and crude price variations, which is translated into a moving
average structure:

r tt∗ = γ + α1 crt +  t + θ1  t−1 + · · · · +θ8  t−8 (37)

By introducing the estimated target level into the adjustment equation (36), Salas
shows that the adjustment speeds are asymmetric.
Additional applications of PAM to the analysis of price asymmetries are provided
by Norman and Shin (1991) and by Shin (1994). The former paper applies Bacon’s
model to the transmission mechanism between crude, wholesale and retail gasoline
prices in the US. Using two different samples of weekly data, one from January 1984
to March 1991, the other from January 1984 to July 1992, the authors find that retail
prices respond symmetrically to wholesale and crude oil price variations. The latter
contribution again employs Bacon’s model and analyzes the transmission mechanism
between wholesale gasoline price and crude price variations in the US. Monthly data
over the period January 1986–May 1992 provide no evidence of EAPA.

4.2 Error Correction Models

The error correction model (ECM) proposed by Engle and Granger (1987), and
subsequently modified by Granger and Lee (1989), allows to test all the asymmetries
supported by ARDL specifications and PAM. Two different approaches have been
proposed for ECM estimation, namely Engle and Granger (1987) and Stock and
Watson (1993).

4.2.1 Engle and Granger’s estimation method

Engle and Granger (1987) suggest a two-step procedure: i) estimate the equilibrium
relation and test for cointegration; ii) estimate the ECM, that is, a regression where
all variables are expressed in first differences, apart from the stationary residuals
from step i), the so-called error correction term (ECT), which can be interpreted as
the deviations from the equilibrium level.
The application of the error correction approach to the analysis of price
asymmetries goes back to Manning (1991), who studies the relationships between
retail prices, excise duties (tx) and crude oil prices (cr) in the UK, using monthly
data over the period 1973–1988. The analysis starts from a LR relationship, where
retail prices are expressed as a function of crude oil prices and taxes:

r tt = φ0 + φ1 crt + φ2 t xt +  t (38)

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
376 FREY AND MANERA

In the presence of cointegration, the residuals from regression (38) are stationary
and can be introduced as an additional regressor into the following asymmetric
ECM:
s  q
r tt = φ + ϕt+ + αi crt−i + α+j crt−
+
j
i=0 j=0

r 
z
+ βh r tt−h + θm t xt−m + λEC Tt−1 + u t (39)
h=1 m=1

where ECT t−1 are the lagged residuals from equation (38), and ϕ+ t is an intercept
dummy which is equal to one if cr t > 0.
Potential asymmetries implied by variations of crude prices can be found by
testing the significance of the coefficients α+j and ϕ+ + +
t ; actually, when α j and ϕt are
not statistically different from zero, the effect of positive and negative variations of
cr is the same (i.e. αi ). The results support the existence of DLE asymmetry, while
the cumulative adjustment function provides evidences of RTS.
While Manning proposes a model to test for asymmetries in the direct impact
of a price increase and decrease, von Cramon-Taubadel (1998) focuses on asym-
metries in the adjustment to the equilibrium. The author studies the relationship
between retail and wholesale prices of pork in the German market, using weekly
data over the period January 1990–October 1993 and proposes the following
specification:

k 
r
+ −
r tt = φ + αi wst−i + βh r tt−h + λ+ EC Tt−1 + λ− EC Tt−1 + u t (40)
i=0 h=1
where, even if a distributed lag structure is considered, the presence of asymmetry
is tested only on the ECT. The estimated coefficients and the impulse response
functions show the existence of both EAPA and RTA.
A different approach is proposed by Asplund et al. (2000). Their analysis focuses
on the spot–retail relationship in the Swedish gasoline market and relies on monthly
data from January 1980 to December 1996. The authors propose two different
adjustment equations, the first one where the impact of increases and decreases
in the marginal costs, measured in local currency, is accounted for:
s 
q
r tt = αi+ (sp × er )+t−i + α−j (sp × er )−
t− j + λEC Tt−1 + u t (41)
i=0 j=0

while in the second equation the effects of increases and decreases in the spot price
and in the exchange rate are made explicit:
s s 
q
r tt = αi+ spt−i ert−i
+
+ α∗+
j sp +
er
t− j t− j + βk− spt−k ert−k
−

i=0 j=0 k=0


q
+ βl∗− spt−l
+
ert−l + λEC Tt−1 + u t (42)
l=0


C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 377

The data show that, when specification (41) is used, the presence of CUIS
cannot be rejected, but the hypotheses of COIS and DLES can. Specification (42)
provides additional information on Equation (41), since the adjustment to exchange
rate variations is instantaneous and asymmetric, while the response to spot price
increases and decreases is distributed over two periods and is affected by COIA and
DLEA.
A slight modification of this model is proposed by Bettendorf et al. (2003), who
examine the Dutch gasoline market using weekly data for the relationship between
exchange rate, retail and spot prices for the period January 1996–December 2001. In
this work, asymmetries in the response to exchange rate variations are not considered,
leading to the following simplified form:

s 
q
r tt = αi+ spt−i
+
+ α−j spt−
−
j + γ ert + λEC Tt−1 + u t (43)
i=0 j=0

Despite the availability of daily data, the authors find that a daily analysis is not
significant and choose to estimate five distinct weekly models, each referred to a
different day of the week. Unexpectedly, the results in terms of asymmetries turn
out to depend on the selected day. The usual test for COIS and for DLES (i.e. αi+ =
α−j ) rejects the null hypothesis only for the Monday, Thursday and Friday models. It
is worth noting that these results contrast with the empirical evidence implied by the
number of significant lags, which are different in all models. However, as the authors
point out, the Akaike Information Criterion used to select the lag structure is almost
identical for each pair of lags, meaning that the difference between s and q cannot
be supportive of DLEA. Since DLES and COIS imply CUIS, while the reverse is
not necessarily true, we can infer that the models for Tuesday and Wednesday data
also exhibit CUIS. Finally, the authors estimate the cumulative adjustment functions,
which show RTA only for Monday, Thursday and Friday. The study by Bettendorf
et al. (2003) demonstrates how data selection may influence the statistical tests for
asymmetries in the price transmission mechanism.
This issue is explicitly tackled by Galeotti et al. (2003) from two perspectives.
First, the analysis concerns five different countries (Italy, France, Spain, Germany
and the UK); second, the asymmetries are explored not only between spot and
retail prices (second stage), but also between crude and spot prices (first stage) and
between crude and retail prices (single stage). The authors use monthly data for
the period January 1985–June 2000 and focus on the market of leaded gasoline.
An asymmetric ECM is estimated for each of the three relationships; for the sake
of brevity we report the LR relationship and the ECM only for the transmission
between retail and spot prices:
r tt = γ0 + α1 spt +  t (44)

+ −
r tt = γ + α+ spt+ + α− spt− + λ+ EC Tt−1 + λ− EC Tt−1 + ut (45)
The asymmetries specified in equation (45) clearly affect the actual response of
retail prices to spot price variations and the adjustment to the equilibrium level.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
378 FREY AND MANERA

Furthermore, the authors calculate the number of weeks necessary to close the
gap between the current and the desired level of prices. The estimated models
have a penchant towards the null hypothesis of symmetry. A result of this kind
is not surprising, if we consider that the usual F-tests lose power when applied
to an asymmetric ECM, as illustrated by Cook et al. (1999). By bootstrapping the
F-statistics, the authors show that all countries under scrutiny are likely to support
both COI and EAP asymmetries, even though there is no evidence of RTA.
In particular, for France and Germany both asymmetries arise in the first and
single stages, while in Italy, Spain and the UK asymmetries affect the spot–retail
relationship. Furthermore, for Italy the crude–spot relationship also rejects the null
of EAPS, while in the UK the single-stage relationship seems to be characterized
by COIA. Consequently, though price asymmetry is a relevant issue in all countries
considered by this study, it assumes different features depending on which market is
analyzed. Finally, it is worth mentioning that, within the crude–retail relationship, the
responsiveness of retail prices is higher to exchange rate increases than to decreases.
Conforti et al. (2003) propose an extension of Manning’s approach to study the
link between world import prices (wp) and the domestic producer prices (dp) of
wheat in Egypt. Their specification is:
 s  r
+
w pt = φ + αi dpt−i + βh w pt−h + λEC Tt−1 + λ+ EC Tt−1 + u t (46)
i=0 h=1

In model (46) a dummy variable is used to account for asymmetry in the error
correction process, but not in the direct impact of producer price increases and
decreases. Monthly prices from January 1989 to May 2001 confirm the statistical
significance of the coefficient λ+ , that is, the existence of EAPA.
The paper by Berardi et al. (2000) provides a very general model which
encompasses many of the studies described above. Using weekly data from April
1996 to February 2000, the authors study the relationship between ex-refinery oil
prices and wholesale prices in the Italian market of leaded and unleaded gasoline
and diesel oil, looking for both SR and LR asymmetries:
s
 +   r
 + 
+
wst = φ0 + αi r f t−1 + αi− r f t−1
−
+ +
βi wst−1 + βi− wst−1
−

i=0 i=1
+ + − −
+λ EC Tt−1 +λ EC Tt−1 + ut (47)
where the LR equation takes the following form:
ws = γ0 + α1r f t + α2 mkt + α3 t +  t (48)
and mk is a proxy of the marketing costs.
A test on the joint null hypothesis αi+ = αi− , ∀i, λ+ = λ− , and the behaviour
of the cumulative adjustment functions show that the three products are affected
by the same asymmetries. In particular, RTA turns out to be a relevant issue in the
Italian market, in addition to COIA and DLEA, though it is not possible to conclude
in favour of both. Conversely, the adjustment path towards the equilibrium level is
symmetric. Model (48) is compared with a restricted form where the trend and the

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 379

marketing costs are not considered. The empirical results show that the restricted
model leads to over-reject the hypothesis of symmetry. Since the general model is
found to fit data better than the restricted specification, omitting relevant variables
may lead to spurious results in terms of asymmetries.
Kaufmann and Laskowski (2005) propose a modified version of von
Cramon-Taubadel’s (1998) model. The authors propose a different way of splitting
ECT, which is based on the level of the explanatory price, rather than on the sign of
the deviations from the LR equilibrium. The analysis focuses on the prices of heating
oil and motor gasoline, it investigates both the crude–refinery and the refinery–retail
relationships, and it takes into account the stock level (st) and the utilization rate
(ur). If we consider, for instance, the refinery–retail transmission mechanism, the
model is as follows:
r 
r r r
r tt = φ0 + αi r f t−i + βi r tt−i + γi stt−i + ϕi urt−i
i=1 i=1 i=1 i=1

11
+ λ+ EC Tt−1 + λ+ EC Tt−1
up dw
+ ζi Di + u t (49)
i=1

where D i are monthly dummies and ECT dw and ECT up are defined as:
 up
EC Tt = EC Tt if crt > 0
(50)
EC Ttdw = EC Tt if crt ≤ 0

Monthly data from January 1986 to December 2002 are used for 12 US regions,
showing that EAPA is not a relevant issue at the crude–refinery level. On the
contrary, EAPA affects the heating oil market across the US, while little evidence of
asymmetry is found in the motor gasoline market, as the null hypothesis of symmetry
can be rejected only for California, Louisiana and Idaho.
The contributions described so far estimate the different ECM with Engle and
Granger’s method. However, the recent literature is far richer in applications of this
type of model.
A representative selection of recent studies should start with Salas (2002),
who studies the Philippine retail gasoline market from January 1999 to February
2002. In addition to a PAM, whose structure has been described in the previous
section, he estimates an ECM along the lines of Asplund et al. (2000) to test the
cumulative impact of crude price variations on retail price, which results to be
asymmetric.
Conforti et al. (2003) also analyze the producer–import price relationship in the
Ethiopian, Rwandan and Ugandan coffee markets over the period January 1990–
December 2001, using a specification similar to Berardi et al. (2000) to test for
COIS, DLES and EAPS. Monthly data show that none of these three countries
exhibits any evidence of asymmetry.2
Von Cramon-Taubadel and Meyer (2003) investigate the link between retail and
wholesale prices of German lettuce and chicken, using weekly data for the period

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
380 FREY AND MANERA

May 1995–December 2000. In particular, two different sets of data are used, the
first based on individual store prices, and the second on average retail prices. Using
the model proposed by von Cramon-Taubadel (1998), the authors show that, for
both products and when individual data are used, the null hypothesis of symmetry
can be rejected, while aggregated data provide no evidence of asymmetric price
behaviour.
In a second paper, Conforti (2004) applies Conforti’s et al. (2003) model to test
the existence of EAPA in the adjustment of the local prices of different agricultural
products in response to world price variations, for a representative number of
countries.3 As a whole, annual data from 1969 to 2001 support the existence of
asymmetry.
The National Department of Agriculture in South Africa (NDA) (2003) uses the
model of von Cramon-Taubadel (1998) to study the farm–retail transmission in the
South African market of maize meal, bread, fresh and long life milk, cheddar cheese
and cooking oil. Monthly data over the period January 2000–July 2003 are used to
obtain the impulse response functions for farm price increases and decreases, which
suggest the presence of RTA for all considered cases.
London Economics (2004) analyzes the mutual relationship between retailer and
producer prices of a number of goods in Austria, Denmark, France, Germany, Ireland,
the Netherlands, Spain and the UK (see Section 3.1.2). In this study the authors also
employ a variation of the von Cramon-Taubadel (1997) ECM for the price series
which turn out to be cointegrated. Empirical evidence supports the presence of
EAPA in the producer–retail relationship for the markets of Danish carrots and UK
bread, as well as in the retail–producer transmission mechanism for the UK lamb
market.
Bachmeier and Griffin (2003) analyze the UK gasoline market from February
1985 to November 1998. They compare a symmetric ECM with an asymmetric
specification. Using daily data, they are able to find that the UK market is
characterized by COIS, DLES, CUIS, RTS and EAPS.
Deltas (2004) extends Manning’s model to a panel context. The author analyzes
the monthly relationship between wholesale and gasoline prices for 48 US states,
using a fixed effects model with time and state dummy variables. The impulse
response functions and a test for the joint null hypothesis of COIS and DLES prove
that, during the period 1998–2002, RTA and at least either COIA or DLEA affect the
US gasoline market. Interestingly, Deltas also notes that the degree of asymmetry
in a state depends on its average retail–wholesale margin.
Krivonos (2004) investigates the transmission mechanism between local and world
coffee prices before and after the structural reforms which affected the coffee
markets of Sub-Saharan Africa and Latin America during the late 1980s and the
early 1990s. Using monthly data from 1984 to 2003, the author finds no evidence
of asymmetries in the pre-reform period, since local prices were driven by the
local governments and were not directly influenced by world prices. In the post-
reform period, local prices started to react to world price variations, and three
of the 20 countries under study have shown an asymmetric price behaviour. In
particular, Kenya, Madagascar and Cameroon seem to be influenced by COIA and

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 381

RTA. Note that, while the former asymmetry is formally tested, the latter is only
indirectly inferred from the degree of adjustment of domestic prices after 6 and
12 months.
Another interesting paper is proposed by Radchenko (2005a), who analyzes the
link between oil price volatility and the asymmetric response of gasoline prices to
oil price variations in the US market. In his work, Radchenko uses both a VAR
model, which will be formally introduced in Section 5.2, and an error correction
specification similar to the model of Berardi et al. (2000). The author uses weekly
data from March 1991 to February 2003 to compute the impulse response functions
to crude price increases and decreases. The empirical results show that RTA affects
the response of retail prices.
In a different study, Radchenko (2005b) applies a similar model to the US gasoline
market, using weekly data from March 1991 to February 2003. Impulse response
function are used to investigate the crude–retail and the spot–retail price transmission
mechanisms, both of which turn out to be affected by RTA.
Finally, Grasso and Manera (2007) analyze the three typical stages of the oil–
gasoline price transmission mechanism, i.e. crude–spot, spot–retail and crude–retail,
in France, Germany, Italy, Spain and the UK. The authors follow Galeotti et al.
(2003) in the choice of the sample period, which goes from January 1985 to March
2003 with a monthly frequency, and in the use of bootstrapped F-tests for the
null hypothesis of symmetry. As for the model structure, they apply a specification
similar to the model by Berardi et al. (2000), though enriched with autoregressive
asymmetric effects and with asymmetric adjustments to exchange rate variations. The
estimations’ results are then used to test for symmetries of the type EAPS, COIS
and DLES. When the crude–spot relationship is considered, all countries exhibit
COIA in response to exchange rate variations. DLEA is found in all countries for
the response to crude prices, with the exception of the UK, while COIA affects only
the Spanish market. The spot–retail transmission mechanism is affected by EAPA in
all considered countries; Spain and the UK are also affected by COIA and DLEA,
while only DLEA is evident in the French market. Finally, in France, Italy and the
UK asymmetric autoregressive effects are also present.
With respect to the single-stage analysis, autoregressive asymmetric effects are
found only in the UK, which is also the only market where the hypothesis of DLES
in the response to exchange rate variations can be rejected. The contemporaneous
impact of the exchange rate is asymmetric not only in the UK market, but also in
Germany and Italy. Finally, EAPA occurs in Italy and France, which has a behaviour
similar to the UK and it is also affected by COIA in response to crude price
variations.

4.2.2 Stock and Watson’s method

The two-step Engle and Granger approach is not the only alternative to estimate
ECM. Stock and Watson (1993) propose a different method which essentially
requires the simultaneous estimation of the LR equilibrium and of the adjustment
process.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
382 FREY AND MANERA

Given the differences between the two approaches, an open question is whether
they lead to the same results in terms of asymmetries or, if this is not the case,
which is the more reliable.
In 1991 Manning published a paper which is one of the first applications of the
Stock and Watson procedure to the analysis of price asymmetries, and provides
a comparison with Engle and Granger’s approach. As described in Section 4.2,
Manning analyzes the UK gasoline market using monthly data over the period 1973–
1988. The method of Stock and Watson, which avoids imposing a pre-determined
equilibrium level on the relation among crude and retail gasoline prices and taxes,
produces the following model:
 s 
q r
r tt = φ + ϕt+ + αi crt−i + α+j crt−
+
j + βh r tt−h
i=0 j=0 h=1

z
+ θm t xt−m + c1r tt−1 − c2 crt−1 + c3 t xt−1 + u t (51)
m=1
The LR equilibrium, which in Engle and Granger procedure is pre-estimated, can
be identified from the estimation of the ECM. Given Equation (51), the implied LR
solution is:
 
φ c2 c3
rt = − + cr + t x (52)
c1 c1 c1
In this case, the results confirm the outcomes derived from Engle and Granger’s
procedure, that is, the null hypothesis of RTS is not rejected, while the null hypothesis
of DLES is rejected.
Arden et al. (1997) provide an example of how the two approaches can end up
with contrasting empirical findings. The analysis starts from the statement that it is
inappropriate to look for asymmetries using a symmetric relationship. Actually, the
first step of Engle and Granger’s procedure is based on a symmetric LR relation,
which may invalidate the identification of LR asymmetries. Some authors have
solved this problem using TAR (Threshold AutoRegressive) models (see Section sec:
ECMtar). Arden et al., instead, choose to jointly estimate the LR equilibrium and the
asymmetry parameters, as prescribed by Stock and Watson’s method. The authors
investigate how input prices x and labour costs lb affect output prices y in the UK
manufacturing sector, during the period 1970–1996 (quarterly data). The production
function is assumed to be Cobb-Douglas, and the ECM includes asymmetries in the
adjustment to the equilibrium:
 r
yt = φ0 + yt−i + α1 xt + α2 lbt + θ + (yt−1 − c1 xt−1 − c2lbt−1 − μ)+
k=1
+ θ − (yt−1 − c1 xt−1 − c2lbt−1 − μ)− +  t (53)
For comparison, the authors also estimate this model using the two-step Engle
and Granger procedure. Testing the null hypothesis θ + = θ + with both techniques
produces contrasting empirical results. Specifically, only Stock and Watson’s method
strongly supports the evidence of EAPA.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 383

Among the contributions which rely on the simultaneous estimation of the LR

equilibrium and the adjustment process, a widely cited paper is Borenstein et al.
(1997), which focuses on the US gasoline market for the period January 1986–
December 1992. Since the purpose of this article is to detect the presence of
asymmetries at the different levels of the distribution chain, the authors use four
different prices: crude spot prices, gasoline spot prices, wholesale prices and retail
prices. All data are weekly, except retail prices which are biweekly. The authors
consider first crude and retail prices, and define the LR relation as:
r tt = γ0 + α1 crt +  t (54)
Model (54) is then included in the ECM:
p
 +  p
 + 
+
r tt = −θ1 φ0 + αi crt−1 + αi− crt−1
−
+ +
βi r tt−1 + βi− r tt−1
−

i=0 i=1

n
− (θ1 η j D j,t ) + θ1r tt−1 − θ1 φ1 crt−1 − θ1 φ2 t +  t (55)
j=2

where D is a set of variables which capture seasonal effects. Each variable in D

indicates the period of the year the data are referred to; consequently, n is equal to
24 (54) for biweekly (weekly) data. This model accounts for both the contemporary
and lagged impact of crude price variations and asymmetric autoregressive effects,
but it imposes the presence of symmetry in the persistence of crude price variation,
as the number of lags is constant and equal to p. Since the contemporary variation
of upstream prices can be considered endogenous, Borenstein et al. use a Two Stage
Least Square (TSLS) estimator, with positive and negative variations of Brent and
Forties crude spot and futures prices as instruments. The large difference between the
estimated values of the coefficients α+ −
0 and α0 confirms the presence of asymmetry
in the contemporary effects of crude price variations, even though the authors do
not provide any statistical test to support their conclusion. In order to evaluate the
presence of RTA, they estimate the cumulative adjustment functions B +j and B −j ,
and test the significance of the variable A n defined as:
 n
 + 
An = B j − B−j dj (56)
j=0

The results confirm the existence of RT asymmetry. By extending their analysis

to the intermediate levels of the distribution chain, Borenstein et al. find that almost
all markets exhibit RTA, with the only exception of the spot–wholesale transmission
mechanism.
Bachmeier and Griffin (2003) question whether Borenstein et al. (1997) have really
proved the existence of asymmetry. In particular, they assert that the application of
Stock and Watson’s method to the model proposed by Borenstein et al. leads to
over-reject the null hypothesis of symmetry. To demonstrate their claim, the authors
analyze the same US crude refinery market analyzed by Borenstein et al., on a
larger data sample (February 1985–November 1998). Specifically, they estimate an
ECM using both Engle and Granger’s and Stock and Watson’s techniques, and

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
384 FREY AND MANERA

use an OLS estimator, rather than TSLS, in both cases. In accordance to their
claim, the authors show that the null of RTS is rejected only when the model is
estimated using Stock and Watson’s approach. These results support Arden et al.’s
(1997) conclusion that only Stock and Watson’s estimation method leads to strong
asymmetries.
The debate on the empirical robustness of the two approaches is wide. A selection
of contributions which use Stock and Watson’s method is reported below.
In 1996, Borenstein and Shepard analyze the US gasoline market from 1982 to
1991 and concentrate on retail, wholesale and crude oil prices. The authors analyze
how the retail margin, which is given by the difference between retail and terminal
prices, is affected by retail and terminal price variations, as well as by the volume
of gasoline consumption. They also investigate the link between terminal and crude
oil prices. The empirical results confirm the presence of RTA in both relationships.
Eltony (1998) studies the instant response of gasoline prices at the pump to positive
and negative variations of crude prices and exchange rate. The author uses monthly
data for UK and US markets during the period January 1980–June 1996. Empirical
evidence shows that both UK and US gasoline prices exhibit COIA to crude price
and exchange rate variations.
Reilly and Witt (1998) study the same problem analyzed in Eltony (1998) with
the same model. Using monthly data for the UK market during the period January
1982–June 1995, they show that retail prices respond much more strongly to crude
prices increases than decreases. Their results suggest that retailers change prices in
response to exchange rate reductions but not to exchange rate increases.
Balke et al. (1998), who study the gasoline price transmission mechanism at the
different levels of the distribution chain, use Manning’s model. They find that DLEA
and COIA occur in almost all upstream–downstream relationships between crude,
spot, wholesale and retail prices, with and without taxes. The only exception is
the spot–retail (ex-taxes) price relationship, where the null hypothesis of symmetry
cannot be rejected.
Another application of Manning’s model is provided by Peltzman (2000), who
offers a very detailed discussion of the issue of price transmission by investigating
how the output prices of 77 consumer and 165 producer goods in US react to
cost variations. The analysis relies on monthly data over the period 1978–1996 and
focuses on the cumulative impact of input prices. His results show that asymmetry
is a relevant factor in all considered markets.
Eckert (2002) investigates the response of weekly retail gasoline prices to
wholesale prices in Ontario, Canada, from November 1989 to September 1994.
The model follows Borenstein et al. and confirms that the Ontario market is affected
by both COIA and RTA.
In 2004, Contin et al. investigate the retail gasoline mark in Spain. They analyze
the relationship between crude and retail prices before and after the abolition of
the system of ceiling price regulation, which took place in 1998. Data are weekly
and the authors provide both a statistical test for the equality of contemporaneous
and distributed lag impacts of positive and negative crude price variations and the
cumulative adjustment functions, in response to crude price increases and decreases.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 385

The Spanish market results to be affected by COIA, DLEA and RTA, before
(i.e. January 1993–September 1998) and after (i.e. October 1998–December 2002)
the reform. However, it is worth noting that in the first period gasoline prices adjust
faster in response to crude price decreases than to crude price increases, while the
reverse is true for the second period.
Deltas (2004) analyzes the US gasoline market using monthly data from 1998 to
2002. The author uses both Engle and Granger’s and Stock and Watson’s approaches
and, as described in Section 4.2, he extends Manning’s model to a panel framework.
The model based on Stock and Watson’s procedure confirms the finding provided
by Engle and Granger’s method, that is, RTA, together with COIA and/or DLEA,
are present in the data.
Finally, a completely different solution to the problem of jointly estimating the
LR equilibrium and the error correction process is discussed in Verlinda (2004), who
uses a Bayesian estimation procedure. The following non-linear Bayesian regression,
with informative prior, is used to estimate the impact of spot gasoline prices on retail
ones:
s q  r
r tt = φ + αi wst−i + α+j wst−
+
j + βh r tt−h
i=0 j=0 h=1

r
+ βr+ r tt−n
+
+ θ(r tt−1 − c1 − c2 wst−1 ) + u t (57)
n=1

Specification (57) is a mixture of Bettendorf et al.’s (2003) model and Manning’s

(1991) approach. The author considers weekly station prices in Southern California
from September 2002 to May 2003 to study the price transmission mechanism at
station level and at an aggregated level. Cumulative impulse response functions for
spot price increases and decreases show that both average and local retail prices are
affected by RTA.
So far we have illustrated a significant number of relevant contributions which use
Engle and Granger’s and Stock and Watson’s approaches to model cointegrated vari-
ables and LR asymmetries. These studies, in general, suggest that the two approaches
can often produce different outcomes in terms of asymmetries. Unfortunately, the
available literature does not help the applied researcher to determinate which results
are more reliable. However, a third approach to asymmetric cointegration is available
in the literature and will be the subject of the next subsection.

4.2.3 TAR and M-TAR cointegration

As we have already pointed out, Engle and Granger’s (1987) approach is based
on a symmetric LR relationship and allows the introduction of asymmetries in the
estimation of the adjustment process of an ECM. However, some authors assert that,
if the true LR relationship between two prices is asymmetric, a test for cointegration
based on a symmetric LR equilibrium may lead to misleading results (see, among
others, Arden et al., 1997; Abdulai, 2002). As discussed in the previous section,
Stock and Watson’s estimation technique does not impose symmetric behaviour

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
386 FREY AND MANERA

when testing for cointegration. A different solution to this problem is proposed by

Enders and Granger (1998), who introduce Threshold AutoRegressive (TAR) and
Momentum-TAR (M-TAR) cointegration.
In an Engle and Granger’s framework, given the LR relationship:

yt = φ0 + φ1 xt +  t (58)
x and y are cointegrated if the null hypothesis ρ = 0 is rejected in the following
regression model:
 t = ρ t−1 + u t (59)
Enders and Granger (1998) suggest using the alternative TAR specification, where
the relation between  t and  t−1 is supposed to vary across two regimes, depending
on the value of  t−1 :
 t = It ρ1  t−1 + (1 − It )ρ2  t−1 + ηt (60)
I t is an indicator function defined as:

1 if  t−1 ≥ 0
It = (61)
0 if  t−1 < 0
If the null hypothesis ρ 1 = ρ 2 = 0 in Equation (60) is rejected, then x and y
are cointegrated and the asymmetric ECM which stems from the TAR specification
is:
 s
yt = αi xt−i + γ +  t−1 It + γ −  t−1 (1 − It ) + u t (62)
i=0

Model (62) is equivalent to the usual specification:


s
+ −
yt = αi xt−i + γ + EC Tt−1 + γ − EC Tt−1 + ut (63)
i=0

and, consequently, a test on the null hypothesis γ + = γ − is a test for EAPA.

Enders and Granger propose a second model for cointegration, known as M-TAR.
This name comes from the financial definition of ‘momentum’, which indicates the
rate of acceleration of prices. As the authors assert, M-TAR models are especially
valuable when adjustment is asymmetric in a way that the series exhibit more
momentum in one direction than in the other. In M-TAR models the threshold
is placed on the variation of  t−1 , rather than on  t−1 , and the indicator function in
Equations (60) and (62) takes the form:

1 if  t−1 ≥ 0
It = (64)
0 if  t−1 < 0
In this case, Equation (63) is not suitable to test for the presence of EAPA. As
a matter of fact, the term γ +  t−1 I t + γ −  t−1 (1 − I t ) is no longer equivalent to
the term γ +  + − −
t−1 + γ  t−1 , since γ
+
and γ − are now the weights of  t−1 when its

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 387

variation with respect to the previous period is positive or negative, respectively.

This model clearly introduces a new kind of asymmetry in the adjustment to
the equilibrium, which we define as Momentum Equilibrium Adjustment Path
Asymmetry (MEAPA).
TAR and M-TAR cointegration has become increasingly popular in the literature
on asymmetric price transmission during the last few years. One recent contribution
in this direction is Abdulai (2000), who studies the Ganan market of maize during the
period May 1980–October 1997. The major aim of this analysis is the transmission
mechanism between the wholesale price of the central market of Techiman and the
corresponding local prices formed in the Accra and Bolgatanga markets. Abdulai
tests for cointegration using both the linear Engle and Granger model and the
TAR and M-TAR specifications, finding the existence of a LR relationship in both
the Techiman–Accra and the Techiman–Bolgatonga markets. However, the M-TAR
specification seems to fit the data better than the others, leading the author to use
model (60) with the indicator function given by Equation (64). The estimation
results and the impulse response functions indicate that both markets are affected
by MEAPA and RTA.
Hassan and Simioni (2001) study the link between shipping point prices and retail
prices of French tomatoes and chicory, using both TAR and M-TAR cointegration.
Specifically, they investigate 22 relationships for the first market and 20 for second
market, where each relation is identified by the type of tomato/chicory and the
origin of production. The empirical findings suggest that, even if almost half of
the relationships under study for the two markets exhibit an asymmetric behaviour,
different products are affected by different kinds of asymmetries, as tomatoes display
MEAPA while chicory displays EAPA. This result confirms that M-TAR models
fit well in the presence of very quick adjustments of retail prices to wholesale
prices (as justified by the highly perishable nature of tomatoes). TAR cointegration,
instead, is appropriate when the less perishable nature of a good (e.g. chicory)
reduces the influence of supply fluctuation on retail prices (see Abdulai and Rieder,
1999).
Abdulai (2002) proposes a different contribution, which investigates the mutual
relationship between retail and wholesale prices of pork meat in the Swiss market,
from January 1988 to September 1997, using monthly data. In this case, a symmetric
cointegration test along Engle and Granger’s (1987) approach does not provide any
evidence of cointegration, while both TAR and M-TAR models support cointegration.
Standard information criteria indicate M-TAR as the best fitting specification. The
estimated coefficients and the impulse response functions show that there is no
impact of retail prices on producer prices, while the response of the retail market to
production costs is affected by both MEAPA and RTA.
Gonzales et al. (2003) apply TAR and M-TAR cointegration to the analysis
of the French value chain of cod and salmon, using monthly data ranging from
February 1988 to December 1999. The paper looks for EAPA, MEAPA and RTA
in the transmission mechanism between upstream and downstream prices, but the
estimated coefficients and the impulse response functions provide no evidence of
any asymmetry.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
388 FREY AND MANERA

Grasso and Manera (2007) analyze the gasoline market in Italy, France, Germany,
Spain and the UK and employ, among other models, a M-TAR specification. Their
results show that MEAPA occurs in the crude–retail price transmission for all
considered countries, as well as also in the crude–spot relationship for Italy and
Spain.

5. Modern Econometric Models

5.1 Regime Switching Models

Regime switching models started to be applied to the analysis of price asymmetries
in the mid-1990s with the pioneering work of Powers (1995), which we have
already described in the context of the ARDL models based on Wolffram’s approach
(Table 8). This class of models applies to situations where the relation between y
and a set of variables X depends on the state of a variable v, which can belong to
X.
Powers analyzes the Californian market for lettuce and argues that the relationship
between local retail prices and shipping point FOB prices (fb) may be different
depending on the volume of locally grown lettuce. The argument goes as follows:
since the FOB price of locally grown lettuce is influenced by both national and
local market characteristics, the link between retail price and the FOB price of
lettuce might weaken when the arrivals of locally grown lettuce increase. In order
to test this hypothesis, Powers proposes the following model:

⎧ s s
+ +
⎪
⎪r tt − r t0 = α C f b + α− C f bt−i
−
⎪
⎪ i=0 i t−i i=0 i
⎪
⎪
⎪
⎪ + βi+ Cmkt−i+
+ βi− Cmkt−i
−
+ t
⎪
⎪
⎪
⎨ if L L t ≤ δ1
⎪  s  s
⎪
⎪r tt − r t0 = α+ C f bt−i
+
+ α− C f bt−i
−
⎪
⎪ i=0 i i=0 i
⎪
⎪
⎪
⎪ + βi+ Cmkt−i
+
+ βi− Cmkt−i
−
+  t
⎪
⎪
⎩ (65)
if L L t > δ1

where the impact of FOB prices depends on the actual level of local lettuce (LL t ) and
within-regime asymmetries are allowed. Weekly data from March 1986 to August
1992 show that the levels of local lettuce influence the retail–FOB relationship and
that CUIA occurs in both states of the world.
Goodwin and Holt (1999) investigate the relationship between farm, wholesale
and retail beef prices in the US, using a three-regime error correction specification,
with a threshold placed on the ECT, rather than on a generic regressor as in the

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 
C 2007 The Authors
Table 8. Summary of Asymmetries by Model – RSM.

Journal compilation 
Asymmetries

Within Regime Between Regime

Data
Year Authors Journal Model COI DLE CUI RT EAP REA REAPA Country Product Sample Frequency

1995 Powers Ag DRS – – Y∗ – – – – USA Agricultural† 1986–1992 Weekly

C 2007 Blackwell Publishing Ltd

1999 Goodwin, Holt AJAE DRS – – – N∗ – – Y∗ USA Alimentary 1981–1998 Weekly
2000 Godby et al. EE DRS – – – – – N∗ – C Gasoline 1990–1996 Weekly
2001 Goodwin, Piggott AJAE DRS – – – N∗ – – Y∗ USA Agricultural 1992–1999 Daily
2002 Johnson RIO DRS – – – Y∗ – Y∗ – USA Gasoline 1996–1998 Weekly
2004 Lewis WP DRS Y∗ Y∗ Y∗ Y∗ – – Y∗ USA Gasoline 2000–2001 Weekly
2005 Grasso, Manera WP DRS – – – – – Y∗ – F, G, I, Gasoline 1985–2003 Monthly
Sp, UK
2005 Grasso, Manera WP DRS – – – – – N∗ – I Gasoline‡ 1985–2003 Monthly
2005b Radchenko EE SRS – – – Y∗ – – – USA Gasoline 1991–2003 Weekly

Y (N) indicates the presence (absence) of asymmetry.

∗
denotes that the presence or absence of asymmetry is supported by a statistical test.
†
wholesale–retail; ‡ Italy: crude–spot relationship.
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION
389

14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
390 FREY AND MANERA

model by Powers:
⎧ r s
⎪
⎪ r tt = φ0 + βi r tt−i + α wst−i
i=0 i
⎪
⎪ 
i=1
⎪
⎪
⎪
⎪ +
p
γ  f m t−i + λEC Tt−1 + u t
⎪
⎪ i=0 i
⎪
⎪
⎪
⎪ if EC Tt−1 ≤ δ1
⎪
⎪
⎪
⎪ r s
⎪ 
β  r tt−i + α wst−i
⎪r tt = φ0 +
⎪
⎨ i=1 i i=0 i
p
⎪ + γ i   f m t−i + λ EC Tt−1 + u t
⎪
⎪
i=0
⎪
⎪ if δ1 < EC Tt−1 < δ2
⎪
⎪
⎪
⎪  
⎪
⎪r tt = φ0 + r
β  r tt−i +
s
α wst−i
⎪
⎪ i=1 i i=0 i
⎪
⎪ p
⎪
⎪
⎪
⎪ + γ i   f m t−i + λ EC Tt−1 + u t
⎪
⎪ i=0
⎩
if δ1 EC Tt−1 ≥ δ2 (66)
Weekly data over the period January 1981–March 1998 show that the existence of
more than one regime cannot be rejected, which means that the dynamic relationship
between farm, wholesale and retail prices as a whole is different according to the
deviation from the LR equilibrium, which we define as REAPA. The authors also
look for RTA using impulse response functions, but in this case no evidence of
asymmetry is found.
Goodwin and Piggott (2001) apply model (66) to the daily price transmission
between the central market and three local markets of corn and soybeans in North
Carolina, from January 1992 to March 1999. Although the investigated market is
not comparable with the one studied by Goodwin and Holt (1999), the empirical
results again suggest the presence of REAPA and RTS.
A different specification is provided by Godby et al. (2000), who use a
deterministic error correction regime switching model to investigate the existence of
REA between crude and retail gasoline prices in 13 major Canadian cities:
⎧ s
⎨r tt = φ0 + α crt−i + λEC Tt−1 + u t if f (crt ) ≤ δ1
i=0 i
s (67)
⎩r t = φ  + α
cr + λ
EC T + u 
if f (cr ) > δ
t 0 i=0 i t−i t−1 t t 1

In this case the threshold is not placed on the ECT, but on a function f (.) of crude
prices; in particular, five different cases are considered, according to whether the
threshold is imposed on the mean of crude price variations, calculated on the most
recent eight lagged values (in this case f (.) is the mean function), or directly on
the lagged values at one, two, three and four lags respectively (f (.) is now the lag
operator). Using weekly data over the time span January 1990–December 1996, the
authors test the equality of all coefficients between the two regimes and are never
able to reject the null of symmetry.
Johnson (2002) uses a modified version of specification (67), where autoregressive
effects are considered and a zero-threshold is imposed on the input price variation. In

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 391

particular, the author analyzes the weekly transmission of wholesale price variation
on retail prices in the US, focusing on 15 gasoline and diesel markets over the
period July 1996–June 1998. In about half of the examined cases, the existence
of two regimes determined by the value of wholesale price variations is supported
by data, thus showing the presence of REA in both diesel and gasoline markets.
Johnson notes also that, for both fuels, the behaviour of the cumulative impulse
response functions depends on the sign of wholesale price shocks, which confirms
the existence of RTA.
If the models described above consider either between-regime or within-regime
asymmetries, Lewis (2004) proposes a specification which nests both:
⎧ r s
⎪
⎪
⎪ r tt = φ0 + βi r tt−i + α+ wst−i
i=0 i
+
⎪
⎪ q
i=1
⎪
⎪
⎪
⎪ + α− wst−i −
+ λEC Tt−1 + u t
⎪
⎪ i=0 i
⎪
⎨ if EC Tt−1 ≤ δ1
r s
⎪
⎪r tt = φ0 + β  r tt−i + +
α wst−i +
⎪
⎪ i=1 i i=0 i
⎪
⎪ 
⎪
⎪ q
⎪
⎪ + α− wst−i −
+ λ EC Tt−1 + u t
⎪
⎪ i=0 i
⎩
if EC Tt−1 > δ1 (68)
This author studies the spot–retail gasoline price transmission in California, using
weekly data over the period January 2000–December 2001, and, in order to test
for asymmetry, he compares the above model with the corresponding symmetric
specification. The results suggest that the Californian market is affected by REAPA
and, within each regime, by COIA, CUIA and DLEA. Furthermore, the cumulative
adjustment functions for wholesale price increases and decreases evidence the
existence of RTA.
Grasso and Manera (2007) use a model with between-regime asymmetries as in
Johnson (2002). Monthly data over the period January 1985–March 2003 show that
in the Italian, French, German, Spanish and English gasoline markets REA occurs
at all levels of the gasoline distribution chain, with the exception of the crude–spot
transmission in Italy.
Finally, Radchenko (2005b) proposes, along with the ECM described above, a
very interesting RS specification, where the transition from a regime to the other is
driven by a Markov chain, rather than by a deterministic process. The author, who
focuses on the US gasoline market, supposes that the relationship between crude
and retail prices and between spot and retail prices can depend on whether a price
shock in the upstream market is viewed as long- or short-lasting. He assumes that
the consumers’ beliefs about the occurrence of economic shocks can be described
by an unobserved state variable S which follows a Markov process with transition
probability matrix:
P [St = i|St−1 = j] = pi j (69)
where indices i and j, i, j = 1,2, denote a long-term and a short-term shock,
respectively.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
392 FREY AND MANERA

Another interesting feature of Radchenko’s model is that only the impact of

upstream prices is allowed to vary between regimes, while the effect of all other
variables is maintained fixed; for the sake of brevity, we report the crude–retail
model only, since all considerations on this model are equally valid for the spot–
retail specification:
⎧ r r s
+ + − −
⎪
⎪ r tt = φ0 + β r t + β r t + α+ crt−i
+
⎪
⎪ i=1 i t−i i=1 i t−i i=0 i
⎪
⎪ q
⎪
⎪ + α− crt−i
− +
+ λ+ EC Tt−1 + λ− EC Tt−1−
+ ut
⎪
⎪ i=0 i
⎪
⎪
⎨ if St = 1
⎪ r r s
⎪
⎪ r tt = φ0 + + +
β r tt−i + − −
β r tt−i + +
α crt−i+
⎪
⎪ i=1 i i=1 i i=0 i
⎪
⎪ 
⎪
⎪ q
⎪
⎪ + α− crt−i
− +
+ λ+ EC Tt−1 + λ− EC Tt−1−
+ u t
⎪
⎩ i=0 i
if St = 2
(70)
A Bayesian regression with informative prior on weekly data from March 1991
to February 2003 reveals that retail prices respond differently depending on the
nature of crude and spot prices shocks. However, this finding does not provide any
information about asymmetries, since, in this model, only within-regime asymmetries
are considered. In particular, Radchenko tests for RTS using a cumulative response
function for upstream price increases and decreases, which confirms the presence
of asymmetry in both states of the world.

5.2 Systems of Equations

The standard single equation analysis of price asymmetries has been generalized to
systems of equations in order to take into account the potential interdependences
among input and output prices and other exogenous variables. In the literature,
multivariate extensions of ARDL, EC and RS models have been proposed, which
are respectively known as Vector AutoRegressive (VAR), Vector Error Correction
(VEC) and Vector Regime Switching (VRS) models (Table 9).

5.2.1 VAR models

VAR models can be represented as:
Yt = 1 Yt−1 + 2 Yt−2 + · · · + q Yt−q +  t (71)
where Y t is an n × 1 vector of the variables of interest,  t is an n × 1 vector
of error terms, and  1 , . . ,  q are n × n matrices of coefficients. In model (71),
each element of the vector Y , {y 1 , . . . , yn }, is a symmetric function of its own
lagged values and of the lagged values of the other components of Y. However,
VAR models can be generalized to incorporate asymmetries, along the same lines
of the corresponding univariate specifications.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 Table 9. Summary of Asymmetries by Model – VAR, VECM, VRSM.


Asymmetries

Within regime Between regime

Data
Year Authors Journal Model COI DLE CUI RT EAP REA REAPA Country Product Sample Frequency

C 2007 The Authors

Journal compilation 
1992 Kirchgassner, Kubler EE VECM Y Y – Y1 – – – G Gasoline 1972–1980 Monthly
1992 Kirchgassner, Kubler EE VECM N N – N1 – – – G Gasoline 1980–1989 Monthly
1993 Capps WP VARcu Y1 – – – – – – USA Alimentary 1986–1988 Weekly
1997 Willett et al. Ag VAR N1 N1 N1 – – – – USA Agricultural2 1975–1991 Monthly
1997 Willett et al. Ag VAR N1 Y1 N1 – – – – USA Agricultural3 1975–1991 Monthly
1997 Willett et al. Ag VAR Y1 N1 Y1 – – – – USA Agricultural4 1975–1991 Monthly
1997 Willett et al. Ag VAR Y1 N1 N1 – – – – USA Agricultural5 1975–1991 Monthly

C 2007 Blackwell Publishing Ltd

2001 Miller, Hayenga AJAE VAR – Y1 – – – – – UK Alimentary 1981–1995 Weekly
2002 Chavas, Mehta AJAE VECM – Y1 – Y1 – – – USA Alimentary 1980–2001 Monthly
2002 Gomez, Koerner WP VECM Y1 Y1 – – – – – USA, F, G Alimentary 1990–2000 Monthly
2003 Aguero WP VRSeg – – – – – – Y1 Pe Agricultural6 1995–2001 Daily
2003 Aguero WP VRSeg – – – – – – N1 Pe Agricultural7 1995–2001 Daily
2003 Aguero WP VRShs – – – – – – N1 Pe Agricultural8 1995–2001 Daily
2003 Aguero WP VRShs – – – – – – Y1 Pe Agricultural9 1995–2001 Daily
2003 Goodwin, Serra AE VRS – – – Y1 – – N1 Sp Alimentary10 1994–2000 Monthly
2003 Goodwin, Serra AE VRS – – – Y1 – – Y1 Sp Alimentary11 1994–2000 Monthly
2003 Meyer WP VRS – – – – – – Y1 G, N Alimentary 1989–2001 Weekly
2004 Luoma et al. WP VRS – – – N1 – – N1 Fi Alimentary 1981–2003 Monthly
2004 Radchenko, Tsurumi WP VAR N1 – – – – – – USA Gasoline 1976–1997 Monthly
2004 Sheperd WP VAR – Y1 – – – – – World Agricultural 1982–2001 Monthly
2005a Radchenko EE VAR – – – Y1 – – – USA Gasoline 1991–2003 Weekly
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION

Notes: Y (N) indicates the presence (absence) of asymmetry;

1
denotes that the presence or absence of asymmetry is supported by a statistical test.
2
s p → r t West, North-east; s p → ws West; r t → ws North-centre; ws → s p, r t → s p West, North-east, North-centre; 3 s p → ws North-centre;
393

4
s p → r t North-centre, s p → ws North-east; 5 r t → ws North-east, West; 6 Potatoes, rice; 7 Tomatoes; 8 Potatoes, rice; 9 Tomatoes; 10 Blended cheese, cream
caramel and pasteurized milk; 11 Sterilized milk.

14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
394 FREY AND MANERA

A very general asymmetric version of model (71) is:

Yt = + + − − + + − −
1 Yt−1 + 1 Yt−1 + · · · + q Yt−q + q Yt−q +  t (72)

where all (or only some) elements of the vector Y are now split into positive and
negative values according to their sign.
Capps (1993) proposes a VAR specification for the wholesale–retail price
transmission of 15 meat products in the Houston market. The model is a restricted
form of the VAR described above, where direct cross-price effects are not considered,
but only contemporaneous correlation between errors in different equations is
accounted for. The mth product is specified according to Houck’s (1977) model:
+ + − −
r tmt − r tm0 = φm0 + φm1 Cwsmt + φm2 Cwsmt +  mt (73)

and the resulting system is estimated as seemingly unrelated regressions (SUR).

Weekly data, from September 1986 to November 1988, show asymmetry in the
contemporaneous impact of cumulative wholesale price variations.
Another interesting example is provided by Willett et al. (1997), who analyze the
transmission mechanism between wholesale, retail and shipping point prices of Red
Delicious apples in the West, the North-east and the North-central of the US, over the
period 1975–1991. Willett et al. formulate three region-specific equations for both
wholesale and retail prices, as well as a single equation for shipping point prices, and
then use SUR to estimate the corresponding parameters; in this way, for example,
each transmission mechanism in the North-west is affected by the behaviour of the
other regional markets. As the structure of the system is concerned, Willett et al.
propose a modified version of Ward’s (1982) model, where the dependent variable
is the actual price in first differences. That is, if we consider the effect of wholesale
on shipping point prices, Ward’s specification (25) becomes:

3 
3 
3
 −
spt − spt−1 = φ + γ1 wst− j D j + γ2 wst− j ξ j D j + γ3 Cwst− j
j=0 j=0 j=0

 3
 
− − −
− Cwst− j−1 D j + γ4 Cwst− j − Cwst− j−1 ξ j D j +  t .
j=0 (74)
Note that Cws− − −
t− j − Cwst− j−1 = wst− j , and hence the two relations:

π+
jlt = γ0 + γ1 ξ j
π− +
jlt − π jlt = γ3 + γ4 ξ j

define the impact of a period-to-period variation in wholesale prices. Consequently,

this model can be considered a vector extension of Tweeten and Quance’s model,
rather than of Wolffram’s, as it allows the evaluation of the effect of single price
variations. Using monthly data, Willett et al. find that shipping point prices respond
symmetrically to retail and wholesale prices in the three regions. Wholesale and
retail prices, instead, exhibit evidence of asymmetric behaviour. In particular, in the

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 395

West the contemporaneous impact of retail prices asymmetrically affects wholesale

prices. In the North-central region, COIA and CUIA are observed in the transmission
from shipping point prices to retail prices, while DLEA characterizes the shipping
point–wholesale price relationship. In the North-east, wholesale prices respond
asymmetrically to contemporaneous variations of both retail and shipping point
prices. Finally, CUIA is present in the response of wholesale prices to shipping
point prices.
A different approach is suggested by Miller and Hayenga (2001), who combine
the investigation of price transmission with spectral analysis, in order to highlight
whether asymmetries are linked to the frequency of price cycles. Their study focuses
on farm, retail and wholesale prices of pork in the UK and it is based on weekly data
from 1981 to 1995. Spectral analysis allows the conversion of a time series from
time domain to frequency domain. In particular, since a stationary and invertible
process is uniquely determined by its autocovariance function γ k , where k is the lag,
this function is translated, through the Fourier transformation, into a power spectral
density function hy(ω), which decomposes the total variance into contributions from
short-, medium- and long-run frequencies:

 
1 
∞
hy(ω) = γ0 + 2 γk cos(ωk) (75)
2π k=1

where ω is the frequency varying in the interval [0, π ].

After applying transformation (75), the authors select four frequency bands;4 then,
for each band, the VAR model (72) is appropriately adapted to the frequency domain
and estimated.5
The results show that retail prices exhibit DLA in the response to low-frequency
cycles in wholesale price, while farm–wholesale margins are affected by DLA at
all frequencies. These findings are even more interesting if we consider that, when
testing the same hypothesis in the time domain, the authors are never able to reject
the null of symmetry for the retail–wholesale margin. As the authors state, their work
shows that ‘traditional time domain methods can mask underlying asymmetries that
occur in subsets of the frequency domain’ (p. 561). However, it is worth noting that
these results can also be used to show that spectral analysis leads to over-rejection
of the null of symmetry.
Another example is provided by Shepherd (2004), who analyzes the world coffee
market. This author uses a multivariate model to explain the relationship between
producer and world prices in six exporting regions, namely Brazil, Colombia,
Guatemala, India, Mexico and Uganda, together with the relationship between world
and retail prices in two consumer countries, the US and Germany. The aim of this
work is to assess whether there are any differences in the transmission mechanism
of coffee prices prior to and after the liberalization process, which took place in
1989. If we indicate with pw the world price and with pr the producer price, the


C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
396 FREY AND MANERA

asymmetric VAR model used by Shepherd can be specified as follows:

⎧ k k
⎪
⎪  pw t = ϕw + φw89 D89t + φ pw + φ + prt−
+
⎪
⎪ j=1 w jw t− j j=1 w jr j
⎪
⎪ 
⎪
⎪ k
⎨ + φ − pr − +  w
j=1 w jr t− j t
k k (76)
⎪
⎪ + +
⎪
⎪  pr t = ϕr
+  r 89 D 89t + φ pw + φ − pw t−
−
⎪
⎪ j=1 r jw t− j j=1 r jw j
⎪
⎪ 
⎩ k
+  prt− j +  tr
j=1 r jr
where D 89t is an impulse dummy which is equal to one at the time of the
liberalization and zero elsewhere.
Using monthly data from January 1982 to December 2001, the author finds that,
before liberalization, all countries, except Guatemala, were affected by DLEA in the
transmission from world to local prices. Furthermore, in the US, DLEA occurred
both in the transmission from world to retail price and in the reverse direction as
well. After liberalization, asymmetries disappeared in India and Mexico, while in
the remaining regions DLEA characterizes both directions of the price transmission.
As anticipated in Section 4.2, Radchenko (2005a) uses a VAR model to test for
RTA. The author studies the relationship between crude and retail gasoline prices
in the US, during the period March 1991–February 2003, using a restricted form
of Equation (72), where only crude price variations are divided into positive and
negative. In accordance with the results derived from the EC specification, evidence
of RTA is found.
The asymmetries supported by ARDL models can also be tested in a multivariate
framework, as proposed by Radchenko and Tsurumi (2006). The authors present a
model where the transmission mechanism between crude and retail gasoline prices
is estimated together with gasoline consumption (G) per vehicle (S), inventory level
(I) and production (Q):
⎧ + −
⎪ r tt = ϕ11 + φ12 crt+ + φ12 crt− + φ13 It + φ14 Dst + φ15 Dwt +  1t
⎪
⎪ Gt
⎨ = ϕ21 + φ22r tt + φ23 Z t + φ23 Dsu + φ24 Dwi +  2t
St
(77)
⎪
⎪ I = ϕ + φ I + φ D + φ D + 
⎪
⎩
t 31 32 t−1 33 st 34 wt 3t
Q t = G t + It − It−1
where Z indicates income, D s and D w are seasonal dummy variables for summer
and winter, respectively. Radchencko and Tsurumi apply a Bayesian Markov Chain
Monte Carlo algorithm to estimate model (77), using the prior distribution defined
in Zellner (1988), Tsurumi (1985) and Dreze (1976), among others. Monthly data
for the US market, over the period January 1976–December 1997, show that the
contemporaneous impact of crude gasoline prices on retail prices is symmetric.

5.2.2 VEC models

The second group of vector models, which have been widely used for the analysis
of price transmission asymmetries, is known as Vector Error Correction (VECM).

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 397

The general form of a VECM is:


q
Yt = Yt−1 + i Yt−i +  t (78)
i=1
where Y is an n × 1 vector of observed variables,  and  are two n × n matrices
of coefficients. In particular, if the number of cointegrating vectors is equal to r, the
LR matrix  can be decomposed into:
 = αn×r βr ×n (79)
where β is the matrix of the r cointegrating vectors. This model can be generalized
to the asymmetric case along the same lines followed by the ECM described above.
Kirchgassner and Kubler (1992) use a VECM to analyze the price transmission
between German wholesale and retail prices of gasoline and light heating oil and
the corresponding spot prices formed in the Rotterdam market. The specification
proposed is a vector extension of Stock and Watson’s model, as the LR equilibrium
and the adjustment process are contemporaneously estimated. Considering, for
example, the gasoline market, the model works as follows:
⎧ k k
⎪
⎪ wst = ϕ1 + φ ws + φ + spt−i
+
⎪
⎪ 11 j t− j i=0 12i
⎪
⎪ k
j=1
⎪
⎨ + φ − spt−i
i=0 13i
−
+ (c1 wst−1 + c2 spt−1 ) +  1t
k k (80)
⎪
⎪ r tt = ϕ2 + r + +
sp +
⎪
⎪
φ21 j t t− j φ22i t−i
⎪
⎪ k
j=1 i=0
⎪
⎩ − −
+ φ spt−i + (c3r tt−1 + c4 spt−1 ) +  2t
i=0 23i
The authors use monthly data from January 1972 to December 1989 and seek to
verify whether there are any asymmetries in the price relationship before and after
1980, since the number of spot contracts during the 1980s was much larger than
during the 1970s. The empirical results show that, during the 1970s, gasoline retail
and wholesale prices and heating oil consumer prices exhibited RTA, DLEA and/or
COIA in response to spot prices, while the wholesale heating oil market was only
affected by either DLEA or COIA. During the second half of the sample period no
evidence of asymmetries is found.
Another example of asymmetric VECM is offered by Chavas and Mehta (2002).
The authors study wholesale and retail prices of US butter using monthly data over
the period January 1980–August 2001 and propose a regime switching VECM with
seasonal effects (D):
 r q  q
Yt = 0 + 1 t + Yt−1 + k Dts + i+ Yt−i
+
+ i− Yt−i
−
+  t (81)
k=1 i=1 i=1

A test on the hypothesis = i+ i− ,

∀i, clearly shows asymmetries in the
distributed lag effect of each price on the other; furthermore, impulse response
functions suggest the existence of RTA. Chavas and Mehta propose a new kind
of asymmetry, which captures the effects of expectations on the price transmission
mechanism. In particular, the authors suggest estimating the covariance between

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
398 FREY AND MANERA

retail (rt) and wholesale (ws) prices using the specification:

swr = ω 0 + ωr E(r tt ) + ω w E(wst ) +  t (82)
Asymmetry occurs when ω r = ω w , that is, when the covariance between ws and rt
has an asymmetric response to expected price increases or decreases. As ω r results to
be positive and ω w negative, the authors conclude that ‘the contemporaneous linkage
between retail and wholesale prices become weaker (stronger) when the wholesale
(retail) price is expected to increase’ (p. 11). It is worth noting, however, that the
asymmetry introduced by the authors does not provide any information about COIA
and, consequently, it does not fit into the classification proposed in our survey.
In 2002 Gomez and Koerner published another interesting paper, which analyzes
the transmission between international and retail coffee prices in the US, France and
Germany using Kirchgassner and Kubler’s (1992) model. The authors investigate
the mutual relationship between the two prices, taking into account the existence of
an asymmetric effect of the exchange rate on the local price and of the monthly rain
precipitations on the international price. Monthly data over the period January 1990
to December 2000 show that in all countries each price exhibits DLA with respect
to its own lagged values and to the lagged value of the other price, in response to
which each price also displays COIA.

5.2.3 VRS models

Finally, VRSM are used to account for the existence of multiple regimes in a VAR
or a VEC specification. A very interesting example is provided by Aguero (2003),
who uses a Vector Error Correction Regime Switching model (VECRSM) to study
the transmission between wholesale and retail prices of rice, tomatoes and potatoes
in Peru from January 1995 to July 2001. For each product, Aguero proposes the
following model to test for REAPS:
 q
Yt = α1 β  Yt−1 + i1 Yt−i +  1t if β  Yt−1 < δ
i=1
q (83)
Yt = α2 β  Yt−1 + i=1
i
2
Yt−i +  2t if β 
Yt−1 ≥ δ
Model (83) is estimated using both Enders and Granger’s (1998) and Hansen and
Seo’s (2002) approaches.
As we have seen in Section 4.2.3, Enders and Granger’s TAR and MTAR models
allow testing for cointegration without imposing symmetry, by means of a two-
step procedure which estimates the LR equilibrium using a two-regime model
with a known threshold. Conversely, Hansen and Seo propose a method to deal
with a regime switching cointegration model when the cointegrating vector β and
the threshold δ are unknown and should be estimated. The existence of different
states of the world can then be tested against the hypothesis of linear cointegration
with a Lagrange Multiplier statistic, suitably modified to account for the fact that,
under the null hypothesis, the value of δ is unknown. Using daily data, the author
finds that the Enders and Granger method is supportive of the null hypothesis of
symmetry for the most perishable products (i.e. tomatoes) but not for potatoes and

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 399

rice, while the converse is true when Hansen and Seo’s approach is used. Aguero
has a preference towards Hansen and Seo’s approach because this method does not
require a preliminary choice of the threshold value.
In 2003, Goodwin and Serra use a Threshold Vector Error Correction model
(TVECM) to investigate how farm prices of raw milk are transmitted to the retail
prices of blended cheese, cream caramel, pasteurized and sterilized milk in Spain.
In particular, the authors propose a three-regime version of model (83), which they
estimate using a sequential conditional iterated SUR in two steps, in order not to
impose specific values for the threshold parameters or cross-equation independence.
Monthly observations over the period July 1994–December 2000 point out that only
the sterilized milk model supports the presence of REAPA. RTA, instead, is a relevant
issue for all products, as shown by the impulse response function provided by the
authors.
A different threshold multivariate model is proposed by Meyer (2003). The author
observes that a single-threshold model, as in Aguero (2003), can be easily tested for
the existence of a threshold, although this type of model excludes the possibility of
a ‘band’ of non-adjustment. What the author criticizes is that, in a model as (83),
‘even very small deviations from the long term equilibrium will always lead to an
adjustment process’ (p. 2). Indeed, given, for instance, the existence of transaction
costs, there may be values of the error correction term for which no adjustment
occurs. To solve this problem, Meyer proposes the alternative specification:
 q
Yt = α1 + i1 Yt−i + λ1 EC Tt−1 +  1t if |EC Tt−1 | ≤ δ
i=1q (84)
Yt = α2 + 2 Yt−i + λ2 EC Tt−1 +  2t if |EC Tt−1 | > δ
i=1 i

where the adjustment is different depending on whether deviations from the

equilibrium take extreme or intermediate levels. Meyer applies model (84) to
investigate the mutual relationship between German and Dutch pig farm prices on a
weekly basis, from June 1989 to March 2001. The empirical results confirm that no
significant adjustment occurs for small deviations from the LR value, which implies
the presence of REAPA.
Finally, Luoma et al. (2004) re-examine the issue of not imposing specific
threshold values in a VRSM, which they solve using a Bayesian estimation
procedure. They study the transmission between producer and consumer prices in
the Finnish beef and pork markets, over the period January 1981–May 2003. The
model adopted is a three-regime specification with no lags:
⎧
1  
⎨Yt = α β Yt−1 +  1t if β Yt−1 ≤ δ1
⎪
Yt = α2 β  Yt−1 +  2t if δ1 < β  Yt−1 ≤ δ2 (85)
⎪
⎩
Yt = α3 β  Yt−1 +  3t if β  Yt−1 ≥ δ2

which is used to test for REAPS and RTS on a monthly basis. For both beef and pork,
the estimated coefficients and the impulse response functions provide no evidence
of asymmetry.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
400 FREY AND MANERA

Table 10. Number of Surveyed Studies Analyzing a Specific Country.

Country Number of Studies

Argentina 1
Australia 1
Brazil 1
Canada 4
Denmark 1
European Union 1
Finland 1
France 6
Germany 7
Ghana 1
Ireland 1
Italy 4
Japan 1
Netherlands 3
Peru 1
Philippines 1
Spain 4
Switzerland 1
South Africa 1
Sweden 1
United Kingdom 9
United States 34
Worldwide countries 3

6. Asymmetric Price Behaviour: A Meta-Regression Analysis

The presence of asymmetries in the price transmission mechanism has been
investigated during the last 20 years through a wide variety of models, countries and
commodities (Tables 10 and 11). The existing literature suggests that the presence of
asymmetry is more than a murmur. Among the 70 papers considered in this survey,
which provide a total of 87 estimated models, only 11 models show no evidence
of any kind of asymmetry (Table 13). However, are these figures really conclusive

Table 11. Number of Surveyed Studies Analyzing a Specific Market.

Market Number of Studies

Agricultural 18
Alimentary 16
Gasoline 34
Other 2


C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 401

Table 12. Testing the Null Hypothesis of Symmetry by Type.

Type of Asymmetry Number of Models Frequency of No Rejection

COIS 51 24%
DLES 55 24%
CUIS 39 59%
RTS 43 34%
EAPS 20 40%
MEAPS 6 33%
RES 4 50%
REAPS 11 36%

in favour of asymmetric price behaviour? Or, conversely, should they be interpreted

with particular care?
Researchers have generally not accounted for all possible sources of asymmetry.
Table 12 shows that a relevant number of studies concentrates on COIA, DLEA,
CUIA and RTA, while REA and REAPA have been scarcely considered so
far.
Various models and approaches have been proposed and the issue of which one
is the most reliable is still open. For instance, since ARDL and ECM have been the
most popular frameworks to investigate price asymmetries (Table 13), the debate
on which of the two is the best model has been the focus of many studies. It is
well known that ECMs are generalizations of the ARDL specifications, used to
account for the presence of cointegration; von Cramon-Taubadel (1998) discusses
the performance of ARDL model applied to cointegrated series and the consequences
of using an ECM with non-cointegrated data.

Table 13. Percentage of Surveyed Studies Not Supporting any Kind of Asymmetry
by Model.

Model Number of Studies Frequency of No Rejection

ARDLpp 13 9%
ARDLcu 9 10%
ECMeg 21 14%
ECMsw 13 0%
ECMth 5 20%
PAM 4 50%
RSM 8 25%
VAR 6 17%
VECM 3 0%
VRSM 5 20%
Total 87 13%


C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
402 FREY AND MANERA

Table 14. Causality Tests Between Input and Output Prices.

Test Number of Studies Percentage

Granger (1969) 17 24.3%

Granger (1969), Sims (1972) 1 1.4%
Sims (1980) 1 1.4%
No test 51 72.9%
Total 70 100%

If we concentrate on ECM, some authors argue that this model can be applied
even to stationary data. An interesting debate on this issue can be found in a series of
articles published in the review Political Analysis between 1992 and 1993 (e.g. Beck,
1992, 1993; Durr, 1993a, 1993b; Smith, 1993; Williams, 1993). This issue has been
recently revisited by Keele (2005), who shows that the empirical properties of ECM
are maintained even when stationary data are used. In summary, the literature does
not help in understanding whether ECM can be applied independently of the presence
of cointegration, rather it shows that modelling non-cointegrated data with either
ECM or ARDL can affect the results of symmetry tests.
What are the effects on statistical tests for price asymmetries of ignoring the
presence of structural breaks in the data? The presence of unobserved structural
breaks in ECM may lead to over-rejection of the cointegration hypothesis. This
drawback is generally solved with suitable modifications of the cointegration tests
(see, among others, Johansen et al., 2000 for details). It is also crucial to notice
that the presence of structural breaks in the cointegrating relationship is likely to
produce the false impression of asymmetry (see, for an extensive discussion, von
Cramon-Taubadel and Meyer, 2003, 2004).
A typical maintained hypothesis is that price formation goes from upstream to
downstream or, equivalently, input prices cause output prices. However, this causality
direction is not always supported by the data. In the considered literature, three
tests have been used to check the direction of causality: Granger (1969), Sims
(1972) and the variance decomposition as described by Sims (1980). Of the 70
papers dealing with price transmission that we have considered in this survey,
19 studies test for causality (Table 14). Specifically, 18 rely on Granger’s (1969)
test, while Sims’ (1972, 1980) approaches are used only by Aguiar and Santana
(2002), who also employ Granger’s method, and by Balke et al. (1998). It is also
important to notice that evidence of a ‘bottom-up’ transmission is found in seven
cases.
Empirical results can also be affected by frequency and aggregation characteristics
of the data used (see, among others, von Cramon-Taubadel, 1997; Bachmeier and
Griffin, 2003; Bettendorf et al., 2003). The literature has mainly focused on weekly
and monthly data and that the number of cases where symmetry is found does
not seem to differ significantly across the two frequencies (Table 15). As far as
aggregation is concerned, the empirical results point out that neither aggregated nor

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 403

Table 15. Testing the Null Hypothesis of Symmetry by Data Frequency.

Data Frequency Number of Studies Percentage

Daily 5 20%
Weekly 33 12%
Biweekly 2 0%
Monthly 42 12%
Quarterly 2 50%
Annual 1 0%
Missing 2 –

disaggregated data are likely to bias symmetry tests (see, for instance, Peltzman,
2000 and von Cramon-Taubadel et al., 2003).
Many factors are likely to affect the testing results for asymmetric price
transmission. But which are the actual factors having a systematic impact on the
F tests for the null hypothesis of symmetry? In this section we perform a meta-
regression analysis (see Roberts, 2005; Roberts and Jarrell, 2005; Stanley and Jarrell,
2005) to assess the relative impact on the calculated F tests for symmetric price
behaviour of a set of meta-explanatory variables. Our investigation is based on the
following regression model:

K
Ftest j = F + βk X jk + η j (86)
k=1

where j = 1, . . . , L. In Equation (86) the dependent variable Ftest j is the j-th

F test for the null hypothesis of symmetric price behaviour reported in the empirical
literature; f is the ‘true’ value of the F test; X jk is the k-th meta-explanatory
variable which measures relevant characteristics of an empirical F test and explains
its systematic variation from other F test calculated in the literature; β k is the
meta-regression coefficient which reflects the biasing effect of the k-th characteristic
relative to the j-th F test; η j is the meta-regression error term.
As pointed out by Stanley and Jarrell (2005), the explanatory variables in a meta-
regression analysis identify important study characteristics or model specifications.
In Table 16 we report the meta-explanatory variables used in our investigation. They
can be divided into four broad groups: 1) characteristics of the data (degrees of
freedom and number of restrictions for the F test, data frequency); 2) characteristics
of the asymmetries (dummy variables capturing the type of asymmetry to be tested);
3) characteristics of the econometric model (whether the estimated equation is an
asymmetric ECM, an asymmetric ECM with M-TAR cointegration, it uses a lagged
dependent variable or a distributed lag specification, etc.); 4) characteristics of the
market under scrutiny (geographical area, type of commodity, stage of the input–
output chain, etc.).
Among the 70 papers cited in this survey, we select the contributions which
provide complete information on the calculated F-statistic for price symmetry for a

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
404 FREY AND MANERA

Table 16. Meta-Regression Variables.

Meta Variable Description

F-test Value of the F-test statistic for the null hypothesis of symmetry
dof Degrees of freedom
nrest Number of restrictions
coiadlea =1 if the joint null hypothesis is COIS and DLES
coia =1 if the null hypothesis is COIS
dlea =1 if the null hypothesis is DLES
cuia =1 if the null hypothesis is CUIS
eapa =1 if the null hypothesis is EAPA
rea =1 if the null hypothesis is RES
stage 1 =1 if the input–wholesale relationship is analyzed
stage 2 =1 if the wholesale–retail relationship is analyzed
stage s =1 if the input–retail relationship is analyzed
ecm =1 if the model is an asymmetric ECM
ecm mtar =1 if the model is an asymmetric ECM with M-TAR cointegration
ardl =1 if the model is an asymmetric ARDL
pam =1 if the model is an asymmetric PAM
rsm =1 if the model is an asymmetric RSM
vector s =1 if the model is vectorial
lagged d =1 if the model uses lagged values of the dependent variable
lagged r =1 if the model uses lagged values of the explanatory variables
gasoline =1 if the gasoline market is analyzed
alimentary =1 if the alimentary market is analyzed
month =1 if the data frequency is monthly
week =1 if the data frequency is weekly
country eu =1 if a European country is studied
country na =1 if a North American country is studied

total of 29 articles and 462 observed F tests. The selected articles are representative
of the equal proportions between studies on agriculture food markets and studies
on gasoline markets which characterize the total number of articles surveyed in
our paper. Table 17 reports some descriptive statistics on the F tests and two
additional variables which are crucial to determine the critical values for the F

Table 17. Meta-Regression Analysis – Descriptive Statistics for the Dependent Variable.

Variable Mean Max Min 50% 95% Standard Obs

Percentile Percentile Deviation

F-test 6.54 217.27 0.0001 1.58 21.38 20.66 462

dof 216.81 1844 34 207 332 171.10 415
nrest 1.32 11 1 1 4 1.20 462


C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 405

distribution, i.e. the number of restrictions (nrest) and the degrees of freedom (dof).
The minimum and maximum values for F test range from virtually zero to 217.27,
with a mean value of 6.54 which denotes that, on average, the generic null hypothesis
of symmetry is rejected (the sample means of the degrees of freedom and the number
of restrictions are 216.81 and 1.32, respectively). Moreover, the 50th (median) and
95th percentiles of the F test are 1.58 and 21.38, which, combined with the maximum
value of 217.27, suggest the presence of outliers in the available observations.
The Robust regression analysis proposed by Berk (1990), Goodall (1983) and
Rousseeuw and Leroy (1987) is a valid alternative to a least squares regression model
when the sample size is small and the data are influenced by outliers. The Robust
regression algorithm we use performs an initial screening based on Cook’s distance
D (Bollen and Jackman, 1990) to eliminate any observation for which D > 1.
Then, it performs an ordinary least squares regression, calculates weights based
on absolute residuals and estimates a set of additional regressions using those
weights, until convergence is achieved. Formally, the final estimated coefficients
f̂ and βˆk , k = 1, . . . , K , are the solution of the following minimization problem:
L  
η̂ j
wj (87)
j=1
s
K
where η̂ j = Ftest j − f̂ − k=1 βˆk X jk is the j-th residual, η̂ j /s is the j-th scaled
residual and w j is the j-th weight. The weighting functions employed by the
Robust regression algorithm are Huber weights (Huber, 1964) and biweights (Beaton
and Tukey, 1974). Standard errors are calculated using the pseudovalues approach
described in Street et al. (1988).
The Robust regression results are presented in Table 18. Column 1 in the
table reports the estimated coefficients of the complete, or benchmark, model,
where the dependent variable F test is regressed on the four groups of meta-
independent variables. Quite surprisingly, several published papers on asymmetric
price transmission do not report precise indications on the available number of
observations and the number of regression parameters. This lack of information
reduces the number of observations from 462 to 415. The calculated F-statistic
for zero slopes is large, denoting a statistically significant joint effect of all meta-
regressors on the dependent variable. The regression shows a significant level for
the constant term, which captures the ‘true’ value of the F test and suggests
that, on average, the analyzed literature supports the idea of asymmetric price
transmission. If we concentrate on the data characteristics, the use of monthly
and weekly frequencies decreases the average level of the F test relative to
daily data. Conversely, the rejection of the null hypothesis of symmetry is only
slightly affected by the characteristics of the market under study. However, we
find that the F test decreases for the European countries and the North American
area, whereas it increases for the wholesale–retail relationship. The regression
output suggests also that the F test is not biased towards any type of asymmetry,
as none of the dummy variables capturing a specific type of asymmetry is
significant.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 C

406

2007 The Authors

Journal compilation 
Table 18. Meta-Analysis with Robust Regression.

Dependent Variable: (1) (2) (3) (4) (5) (6) (7)

F-test complete coiadlea=1 dlea=1 eapa=1 cuia=1 ecm=1 ardl=1

constant 52.04∗∗ (3.15) 48.14∗∗ (0.72) 183.98∗∗ (47.03) 5.63∗∗ (1.52) 9.21∗∗ (2.86) 22.17∗∗ (1.78) 2.76∗∗ (1.20)
dof −0.001 (0.001) 0.0007∗ (0.003) −0.61∗∗ (0.12) −0.004 (0.005)−0.003 (0.003)−0.004 (0.005)−0.001 (0.001)
nrest 0.02 (0.12) −0.08 (0.07) −1.47 (2.44) – – – −0.19 (0.17)

C 2007 Blackwell Publishing Ltd

coiadlea −0.13 (1.53) – – – – – 4.09∗∗ (1.20)
∗∗
coia −0.85 (1.50) – – – – −3.02 (1.11) –
dlea −0.47 (1.50) – – – −2.86∗∗ (1.15) 5.76∗∗ (1.86)
cuia −0.72 (1.46) – – – – – 2.85∗∗ (1.07)
∗∗
eapa −1.16 (1.48) – – – – −3.35 (1.07) –
stage 1 −0.67∗∗ (0.28) −8.05∗∗ (1.07) – – – −0.84∗∗ (0.33) –
FREY AND MANERA

stage 2 – −8.06∗∗ (1.01) 1.48∗ (0.75) 0.75∗ (0.39) −0.31 (1.18) – −0.13 (0.66)
stage s −0.30 (0.27) – −1.87∗ (0.95) 0.37 (0.39) −0.08 (1.06) −0.71∗∗ (0.33) 0.04 (0.63)
∗∗
ecm −0.99 (1.53) – −0.54.38 (22.54)−0.38 (0.47) – – –
ardl 0.22 (1.61) – 54.03∗∗ (23.04) – −3.50 (2.46) – –
pam 3.50∗ (2.08) – – 5.10∗∗ (1.87) −3.58 (4.18) – –
rsm 1.42 (1.51) – −55.78∗∗ (22.74) – 25.12∗∗ (3.02) – –

14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 
C 2007 The Authors
Journal compilation 
vector s 1.85∗∗ (0.80) – – – – – –
lagged d 0.84∗∗ (0.40) −0.47 (0.42) 1.04 (1.93) 0.56 (0.55) – 0.88∗ (0.45) 0.33 (0.70)
lagged r −0.81∗ (0.43) – – −0.22 (0.53) −2.66∗ (1.49) −0.72 (0.46) −3.01∗∗ (0.89)
gasoline −0.46 (0.43) – – 0.57 (0.53) −1.30 (1.17) −0.11 (0.51) −0.58 (0.59)
agriculture – – – – – −12.74∗∗ (1.77) –

C 2007 Blackwell Publishing Ltd

alimentary −0.51 (0.42) −7.22∗∗ (0.90) – – −0.08 (0.86) – −0.39 (0.50)
month −46.22∗∗ (2.90) – – – 0.66 (1.00) 0.22 (0.87) 0.62 (0.60)
week −45.27∗∗ (2.83) – – −0.26 (0.81) – – –
country eu −3.65∗∗ (0.54) −39.42∗∗ (1.13) – −4.67∗∗ (1.01) −1.24 (1.79) −17.06∗∗ (1.40) −1.88∗∗ (0.78)
country na −3.44∗∗ (0.54) −40.43∗∗ (0.97) – – −0.73 (1.15) – −0.73 (0.71)
Obs 415 36 61 90 101 145 153
F-Statistic 105.90∗∗ 842.56∗∗ 6.46∗∗ 9.40∗∗ 28.29∗∗ 16.50∗∗ 2.50∗∗
∗ ∗∗
( ) indicates 10% (5%) significance level.
Robust standard errors are reported in round brackets.
a ‘–’ indicates that an explanatory variable does not appear in a specific meta-regression.
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION
407

14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
408 FREY AND MANERA

Nonetheless, it is informative to investigate whether, for a given type of

asymmetry, the rejection of null hypothesis is related to the structure of the dataset,
the analyzed market and the type of econometric model.
For this purpose, in columns 2 to 5 of Table 18 we report the results of four
meta-regressions conditioned on different asymmetries. Although the number of
observations varies substantially across models, the F statistic for zero slopes is
always highly significant. On average, the null hypothesis of symmetry tends to be
rejected when a joint test for COIS and DLES (regression 2) or a single test for DLES
(regression 3) are run. In particular, the average level of the F test for the DLES
hypothesis is far above the average value of the complete case, that is, the rejection
of DLES is evident and does not depend on the particular study or the specific
econometric model. The rejection of the joint null hypothesis of COIS and DLES is
weakly related to the number of degrees of freedom and is slightly increased when
the asymmetric price relationship directly links the price of an input to the price of the
corresponding output (single stage). The role played by each econometric model in
rejecting a specific type of symmetry depends on the null hypothesis to be tested. The
dummy variables capturing a specific type of econometric specification are absent
from meta-regression (2), since the joint null of COIS and DLES is tested by virtually
all studies. Nevertheless, asymmetric ECM and RSM are complementary in reducing
the value of the F test when DLES alone is tested (regression 3), which seems to
be captured most easily by ARDL. Moreover, PAM and again RSM are important
in explaining the significance of the F test for EAPA (regression 4) and CUIA
(regression 5), respectively. As in the benchmark regression (1), the F test increases
in all meta-regressions when the examined countries are non-European or belong
to the North American area. Finally, the type of commodity is significant in reg-
ression (2) only (alimentary) and has the tendency to depress the value of the F test.
We also use meta-regression analysis to examine the effects of the F test of
different types of asymmetries for a given econometric model. The last two columns
of Table 18 report the meta-regression output which has been obtained when two
of the most popular econometric models, namely ECM and ARDL, are considered.
Column 6 shows that COIA, DLEA and EAPA tend to reduce the average value
of the F statistic calculated with the ECM. Additional features of ECM are: i) the
relevance on the F test of the specific stage of the price transmission chain; ii)
the (weak) positive impact on the F test of the presence of the lagged dependent
variable among the regressors, which contributes to mitigate residual autocorrelation;
iii) the (strong) negative impact of the dummy variable agriculture, suggesting that
ECM are less likely to reject the null hypothesis of symmetry when the asymmetric
price transmission involves agricultural goods and commodities; iv) the (strong)
negative effect on F test of considering European countries. In contrast with the
previous findings, column 7 shows that COIA and DLEA jointly, DLEA alone and
CUIA have a positive impact on the average value of the F statistic calculated
with ARDL. Other relevant characteristics of the ARDL specification are: i) the
(strong) negative impact on the F test of the presence of the lagged regressors;
ii) as in meta-regression (6), the negative effect on the F test of considering European
countries.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 409

In summary, our meta-regression results suggest that: i) each econometric model

is specialized to capture some specific asymmetries; ii) each asymmetry is better
investigated by a subset of econometric models; iii) the general significance of
the F test for asymmetric price transmission depends mainly on the characteristics
of the data (i.e. frequency), the dynamic specification of the econometric model
(e.g. presence of distributed lags and lagged dependent variable), and the market
characteristics (i.e. country and stage of the price transmission mechanism under
scrutiny). Overall, our empirical findings confirm that asymmetry, in all its forms,
is very likely to occur in a wide range of markets and econometric models.

7. Conclusions
Consumers often complain that retail prices increase more when input prices are
rising than they decrease when costs are falling. In response to this sentiment, a
wide range of empirical works have tried to clarify whether or not asymmetries
occur, proposing various definitions of asymmetries and using different econometric
models.
In this survey, we have classified the existing empirical literature on price
transmission according to the econometric models used in the empirical analysis
and to the asymmetries subject to statistical testing procedures. Specifically, we have
proposed an exhaustive classification of asymmetries into eight categories, namely
contemporaneous impact, distributed lag effect, cumulated impact, reaction time,
equilibrium and momentum equilibrium adjustment path, regime effect and regime
equilibrium adjustment path. This review has also evaluated the relative merits of
the most popular econometric models for price asymmetries, namely autoregressive
distributed lags, partial adjustments, error correction models, regime switching and
vector autoregressive models. The existing literature suggests that the presence of
asymmetry is more than a murmur. Among the 70 papers considered in this survey,
which provide a total of 87 estimated models, only 11 models show no evidence of
any kind of asymmetry.
Many factors are likely to affect the testing results for asymmetric price
transmission. However, which are the relevant factors having a systematic impact on
the F tests for the null hypothesis of symmetry? In order to answer this question we
have used a meta-regression analysis to assess the relative impact on the calculated
F tests for symmetric price behaviour of a set of meta-explanatory variables using a
selection of contributions which provide complete information about the calculated
F-statistic for price symmetry. The meta-regression results from the general model
confirm our conclusion. The calculated F-statistic for zero slopes is large, denoting a
statistically significant joint effect of all meta-regressors on the dependent variable.
The regression shows a significant level for the constant term, which captures the
‘true’ value of the F test and suggests that, on average, the analyzed literature
supports the idea of asymmetric price transmission. If we concentrate on the data
characteristics, the use of monthly and weekly frequencies decreases the average
level of the F test relative to daily data. Conversely, the rejection of the null
hypothesis of symmetry is only slightly affected by the characteristics of the market

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
410 FREY AND MANERA

under study. However, we find that the F test decreases for the European countries
and the North American area, while it increases for the wholesale–retail relationship.
The regression output suggests also that the F test is not biased towards any type of
asymmetry, as none of the dummy variables capturing a specific type of asymmetry
is significant.
In the light of our meta-regression investigation, the main results of our survey
can be summarized as follows: i) each econometric model is specialized to capture
a subset of specific asymmetries; ii) each asymmetry is better investigated by a
subset of specific econometric models; iii) the general significance of the F test
for asymmetric price transmission mainly depends on the characteristics of the data
(i.e. frequency of the data), the dynamic specification of the econometric model
(e.g. presence of distributed lags and lagged dependent variable), and the market
characteristics (i.e. country and stage of the price transmission mechanism under
scrutiny). Overall, our empirical findings confirm that asymmetry, in all its forms,
is very likely to occur in a wide range of markets and econometric models.

Acknowledgements
The authors would like to thank Marzio Galeotti, Alessandro Lanza, Anil Markandya,
Michael McAleer and Elisa Scarpa for insightful discussion, and seminar participants at the
Fondazione Eni Enrico Mattei, Milan and at the Department of Statistics of the University
of Milan-Bicocca for helpful comments. The paper has substantially benefitted from the
suggestions of Les Oxley and two anonymous referees.

Notes
1. The only exception is the leaded regular gasoline market, where persistence in
wholesale price decreases is slightly stronger than for price increases.
2. The presence of COIS and DLES also implies CUIS.
3. Namely, Argentina, Brazil, Chile, Costa Rica, Egypt, Ethiopia, Ghana, India,
Indonesia, Mexico, Pakistan, Senegal, Thailand, Turkey, Uganda and Uruguay.
4. ω ∈ [0, 0.2], [0.1, 0.3], [0.2, 0.4], [0.3, 0.5], which correspond to the week intervals
[0, 5], [3.33, 10], [2.5, 5], [2, 3.33] in the time domain.
5. A generic VAR Y = X  +  is transformed into ZY = ZX + Z , where Z is the
Fourier transform matrix. For details see Miller and Hayenga (2001).

References
Abdulai, A. (2000) Spatial price transmission and asymmetry in the Ghanaian maize
market. Journal of Development Economics 63: 327–349.
Abdulai, A. (2002) Using threshold cointegration to estimate asymmetric price transmission
in the Swiss pork market. Applied Economics 34: 679–687.
Aguero, J. M. (2003) Asymmetric price adjustment and behaviour under risk: evidence
from Peruvian agricultural markets. University of Wisconsin-Madison, Department of
Agricultural and Applied Economics.
Aguiar, D. R. D. and Santana, J. A. (2002) Asymmetry in farm to retail price transmission:
evidence from Brazil. Agribusiness 18: 37–48.

C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 411

Arden, R., Holly, S. and Turner, P. (1997) The asymmetric adjustment of prices: theory
and evidence from UK manufacturing. Cambridge Working Papers in Economics No.
9715, Department of Applied Economics, University of Cambridge.
Asplund, M., Eriksson, R. and Friberg, R. (2000) Price adjustment by a gasoline retail
chain. Scandinavian Journal of Economics 102: 101–121.
Bachmeier, L. J. and Griffin, J. M. (2003) New evidence on asymmetric gasoline price
responses. The Review of Economics and Statistics 85: 772–776.
Bacon, R. W. (1991) Rockets and feathers: the asymmetric speed of adjustment of UK
retail gasoline prices to cost changes. Energy Economics 13: 211–218.
Balabanoff, S. (1993) The composite barrel of retail prices and its relationship to crude oil
prices. OPEC Review 17: 421–449.
Balke, N. S., Brown, S. P. A. and Yucel, M. K. (1998) Crude oil and gasoline prices: an
asymmetric relationship? Economic and Financial Policy Review 1: 2–11.
Beaton, A. E. and Tukey, J. W. (1974) The fitting of power series, meaning polynomials,
illustrated on band-spectroscopic data. Technometrics 16: 146–185.
Beck, N. (1992) Comparing dynamic specifications: the case of presidential approval.
Political Analysis 3: 27–50.
Beck, N. (1993) The methodology of cointegration. Political Analysis 4: 237–248.
Berardi, D., Franzosi, A. and Vignocchi, C. (2000) Il prezzo dei carburanti in Italia:
asimmetrie e mispecificazion. Contributi di ricerca IRS No. 50.
Berk, R. A. (1990) A primary on robust regression. In J. Fox and J. S. Long (eds.), Modern
Methods of Data Analysis. Newbury Park, CA: Sage Publications.
Bettendorf, L. and van der Geest, S. A. and Varkevisser, M. (2003) Price asymmetry in
the Dutch retail gasoline market. Energy Economics 25: 669–689.
Bollen, K. A. and Jackman, R. W. (1990) Regression diagnostics: an expository treatment
of outliers and influential cases. In J. Fox and J. S. Long (eds.), Modern Methods of
Data Analysis. Newbury Park, CA: Sage Publications.
Borenstein, S. and Shepard, A. (1996) Dynamic pricing in retail gasoline markets. The
RAND Journal of Economics 27: 429–451.
Borenstein, S., Cameron, A. C. and Gilbert, R. (1997) Do gasoline prices respond
asymmetrically to crude oil price changes? The Quarterly Journal of Economics 112:
305–399.
Brown, S. P. A. and Yucel, M. K. (2000) Gasoline and crude oil prices: why
the asymmetry? Economic and Financial Policy Review, pp. 23–29. (http://ideas.
repec.org/a/fip/fedder/y2000iq3p23-29.html).
Bunte, F. and Zachariasse, V. (2003) How are farmers faring in the changing balance
of power along the food chain? Paper presented at the Conference on Changing
Dimensions of the Food Economy, The Netherlands, February 2003.
Capps, O. Jr. (1993) Uses of supermarket scan data in demand analysis. Food Demand and
Consumption Behaviour Regional Committee Working Paper No. s21693capp01.
Chavas, J. P. and Mehta, A. (2004) Price dynamics in a vertical sector: the case of butter.
American Journal of Agricultural Economics 86: 1078–1093.
Conforti, P. (2004) Price transmission in selected agricultural markets. FAO Commodity
and Trade Policy Research Working Paper No. 7.
Conforti, P., Hallam, D. and Rapsomanikis, G. (2003) Market integration and price
transmission in selected food and cash crop markets of developing countries: review and
applications. FAO Commodity Market Review, FAO Commodity and Trade Division
Working Paper.
Contin, I., Correljé, A. and Palacios, M. B. (2004) Competition and price asymmetry in
the Spanish retail gasoline market. EARIE.
Cook, S., Holly, S. and Turner, P. (1999) The power of tests for non-linearity: the case of
Granger-Lee asymmetry. Economics Letters 62: 155–159.


C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
412 FREY AND MANERA

Deltas, G. (2004) Retail gasoline prices dynamics and local market power. University of
Illinois, Urbana-Champaign.
Dreze, J. H. (1976) Bayesian limited information analysis of the simultaneous equations
model. Econometrica 44: 1045–1075.
Duffy-Deno, K. T. (1996) Retail price asymmetries in local gasoline markets. Energy
Economics 18: 81–92.
Durr, R. (1993a) An essay on cointegration and error correction models. Political Analysis
4: 185–228.
Durr, R. (1993b) Of forests and trees. Political Analysis 4: 255–258.
Eckert, A. (2002) Retail price cycles and response asymmetry. Canadian Journal of
Economics 35: 52–76.
Eltony, M. N. (1998) The asymmetry of gasoline prices: fresh evidence from an error
correction model for U.K. and U.S.A. International Journal of Energy Research 22:
271–276.
Enders, W. and Granger, C. W. J. (1998) Unit-root tests and asymmetric adjustment with an
example using the term structure of interest rates. Journal of Business and Economic
Statistics 16: 304–311.
Energy Information Administration (1999) Price changes in the gasoline market. Report n.
DOE/IEA-0626, February 1999.
Engle, R. F. and Granger, C. W. J. (1987) Co-integration and error correction: representa-
tion, estimation and testing. Econometrica 55: 251–276.
Farrell, M. J. (1952) Irreversible demand functions. Econometrica 20: 171–186.
Galeotti, M., Lanza, A. and Manera, M. (2003) Rockets and feathers revisited: an
international comparison on European gasoline markets. Energy Economics 25: 175–
190.
General Accounting Office US (1993) Analysis of the pricing of crude oil and petroleum
products. GAO/RCED91-97.
Geweke, J. (2004) Issues in the rockets and feathers gasoline literature. Report to Federal
Trade Commission.
Girapunthong, N., VanSickle, J. J. and Renwick, A. (2004) Price asymmetry in the United
States fresh tomato market. Journal of Food Distribution Research 34(3): 51–59.
Godby, R., Lintner, A. M., Stengos, T. and Wandschneider, B. (2000) Testing for
asymmetric pricing in the Canadian retail gasoline market. Energy Economics 22:
349–368.
Gomez, M. I. and Koerner, J. (2002) Do retail coffee prices increase faster than they fall?
Asymmetric price transmission in France, Germany and the United States. Mimeo.
Gonzales, F., Guillotreau, P., LeGrel, L. and Simioni, M. (2003) Asymmetry of price
transmission within the French value chain of seafood products. INRA Working Paper.
(http://netec.mcc.ac.uk/WoPEc/data/Papers/reainrawp49.html).
Goodall, C. (1983) M-estimators of location: an outline of the theory. In D. C. Hoaglin,
F. Mosteller and J. W. Tukey (eds.), Understanding Robust and Exploratory Data
Analysis. New York: Wiley.
Goodwin, B. K. and Holt, M. T. (1999) Price transmission and asymmetric adjustment in
the US beef sector. American Journal of Agricultural Economics 81: 630–637.
Goodwin, B. K. and Piggott, N. E. (2001) Spatial market integration in the presence of
threshold effects. American Journal of Agricultural Economics 83: 302–317.
Goodwin, B. K. and Serra, T. (2003) Price transmission and asymmetric adjustment in the
Spanish dairy sector. Applied Economics 35: 1889–1899.
Granger, C. W. J. (1969) Investigating causal relations by econometric models and cross-
spectral methods. Econometrica 37: 424–438.
Granger, C. W. J. and Lee, T. H. (1989) Investigation on production sales and inventory
relationship using multicointegration and non-symmetric error correction models.
Journal of Applied Econometrics 4: 145–159.


C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 413

Granger, C. W. J. and Newbold, P. (1974) Spurious regressions in econometrics. Journal

of Econometrics 2: 111–120.
Grasso, M. and Manera, M. (2007) Asymmetric error correction models for the oil-gasoline
price relationship. Energy Policy 35: 156–177.
Griffith, G. R. and Piggott, N. E. (1994) Asymmetry in beef, lamb and pork farm-retail
price transmission in Australia. Agricultural Economics 10: 307–316.
Hansen, B. and Seo, B. (2002) Testing for two-regime threshold cointegration in vector
error-correction models. Journal of Econometrics 110: 293–318.
Hassan, D. and Simioni, M. (2001) Price linkage and transmission between ship-
pers and retailers in the French vegetable channel. INRA Working Paper.
(http://netec.mcc.ac.uk/WoPEc/data/Papers/reainrawp49.html).
Houck, J. P. (1977) An approach to specifying and estimating nonreversible functions.
American Journal of Agricultural Economics 59: 570–572.
Huber, P. J. (1964) Robust estimation of a location parameter. Annals of Mathematical
Statistics 35: 73–101.
Johansen, S., Mosconi, R. and Nielsen, B. (2000) Cointegration analysis in the presence
of structural breaks in the deterministic trend. Econometrics Journal 3: 216–249.
Johnson, R. N. (2002) Search costs, lags and prices at the pump. Review of Industrial
Organization 20: 33–50.
Karrenbrock, J. D. (1991) The behaviour of retail gasoline prices: symmetric or not?
Federal Reserve Bank of St. Louis Review 4: 19–29.
Kaufmann, R. K. and Laskowski, C. (2005) Causes for an asymmetric relation between
the price of crude oil and refined petroleum products. Energy Policy 33: 1587–1596.
Keele, L. (2005) Not just for cointegration: error correction models with stationary data.
Nuffield College Working Papers in Politics No. 2005-W7.
Kinnucan, H. W. and Forker, O. D. (1987) Asymmetry in farm to retail price transmission
for major dairy products. American Journal of Agricultural Economics 69: 285–292.
Kirchgassner, G. and Kubler, K. (1992) Symmetric or asymmetric price adjustments
in the oil market: an empirical analysis of the relations between international and
domestic prices in the Federal Republic of Germany, 1972–89. Energy Economics 14:
171–185.
Krivonos, E. (2004) The impact of coffee market reforms on producer prices and price
transmission. The World Bank Policy Research Working Paper Series No. 3358.
Lewis, M. (2004) Asymmetric price adjustment and consumer search: an examination of
the retail gasoline market. Center for the Study of Energy Markets (CSEM) Working
Paper n. 120.
London Economics (2004) Investigation of the determinants of farm-retail price spreads.
Report, February 2004.
Luoma, A., Luoto, J. and Taipale, M. (2004) Threshold cointegration and asymmetric price
transmission in Finnish beef and pork markets. Pellervo Economic Research Institute
Working Papers No. 70.
Manning, D. N. (1991) Petrol prices, oil price rises and oil price falls: some evidence for
the UK since 1972. Applied Economics 23: 1535–1541.
Meyer, J. (2003) Measuring market integration in the presence of transaction costs. A
threshold vector error correction approach. Contributed paper selected for presentation
at the 25th International Conference of Agricultural Economists, Durban, South Africa,
August 16–22, 2003.
Miller, D. J. and Hayenga, M. L. (2001) Price cycles and asymmetric price transmission
in the US pork market. American Journal of Agricultural Economics 83: 551–562.
Mohanty, S., Peterson, E. W. and Kruse, N. C. (1995) Price asymmetry in the international
wheat market. Canadian Journal of Agricultural Economics 43: 355–366.


C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
414 FREY AND MANERA

National Department of Agriculture South Africa (2003) Market structure, asymmetry and
price transmission in the food chains. Food pricing monitoring committee final report
Vol. 5.4.
Norman, D. and Shin, D. (1991) Price adjustment in gasoline and heating oil markets.
American Petroleum Institute No. 60, Washington DC.
Parrott, S. D., Estwood, D. B. and Brooker, J. R. (2001) Testing for symmetry in price
transmission: an extension of the Shiller lag structure with an application to fresh
tomatoes. Journal of Agribusiness 19: 35–49.
Peltzman, S. (2000) Prices rise faster than they fall. The Journal of Political Economy 108:
466–502.
Powers, N. J. (1995) Sticky short run prices and vertical pricing: evidence from the market
for iceberg lettuce. Agribusiness 11: 57–75.
Punyawadee, V., Boyd, M. S. and Faminow, M. D. (1991) Testing for asymmetric pricing
in the Alberta pork market. Canadian Journal of Agricultural Economics 39: 493–501.
Radchenko, S. (2005a) Oil price volatility and the asymmetric response of gasoline prices
to oil price increases and decreases. Energy Economics 27: 708–730.
Radchenko, S. (2005b) Lags in the response of gasoline prices to changes in crude
oil prices: the role of short-term and long-term shocks. Energy Economics 27:
573–602.
Radchenko, S. and Tsurumi, H. (2006) Limited information Bayesian analysis of a
simultaneous equation with an autocorrelated error term and its application to the
US gasoline market. Journal of Econometrics 133: 31–49.
Reilly, B. and Witt, R. (1998) Petrol price asymmetries revisited. Energy Economics 20:
297–308.
Roberts, C. J. (2005) Issues in meta-regression analysis: an overview. Journal of Economic
Surveys 19: 295–298.
Roberts, T. D. and Jarrell, S. B. (2005) Meta-regression analysis: a quantitative method of
literature surveys. Journal of Economic Surveys 19: 299–308.
Rousseeuw, P. J. and Leroy, A. M. (1987) Robust Regression and Outlier Detection. New
York: Wiley.
Salas, J. M. I. S. (2002) Asymmetric price adjustments in a deregulated gasoline market.
Philippine Review of Economics 39: 38–71.
Shepherd, B. (2004) Trade and market power in a liberalised commodity market: preliminary
results for coffee. Paper presented at the 85th EAAE Seminar, Florence, 8–11
September.
Shin, D. (1994) Do product prices respond symmetrically to changes in crude oil prices?
OPEC Review 18: 137–157.
Sims, C. A. (1972) Money, income and causality. American Economic Review 62: 540–552.
Sims, C. A. (1980a) Comparison of interwar and postwar business cycles: monetarism
reconsidered. American Economic Review 70: 250–257.
Sims, C. A. (1980b) Macroeconomics and reality. Econometrica 48: 1–48.
Smith, R. (1993) Error correction, attractors and cointegration: substantive and method-
ological issues. Political Analysis 4: 249–254.
Stock, H. J. and Watson, M. W. (1993) A simple estimator of cointegrating vectors in
higher order integrated systems, Econometrica 61: 783–820.
Street, J. O., Carroll, R. J. and Ruppert, D. (1988) A note on computing robust regression
estimates via iteratively reweighted least squares. The American Statistician 42: 152–
154.
Tsurumi, H. (1985) Limited information Bayesian analysis of structural coefficient in a
simultaneous equation system. Communication in Statistics 14: 1103–1120.
Tweeten, L. G. and Quance, L. (1969) Positivistic measures of aggregate supply elasticities:
some new approaches. American Economic Review 59: 175–183.


C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd
 14676419, 2007, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-6419.2007.00507.x by Quaid-I-Azam University, Wiley Online Library on [17/07/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ECONOMETRIC MODELS OF ASYMMETRIC PRICE TRANSMISSION 415

Verlinda, J. A. (2004) Price-response asymmetry and spatial differentiation in local retail

gasoline markets. University of California-Irvine Working Paper.
von Cramon-Taubadel, S. (1998) Estimating asymmetric price transmission with the error
correction representation: an application to the German pork market. European Review
of Agricultural Economics 25: 1–18.
von Cramon-Taubadel, S. and Meyer, J. (2004) Asymmetric price transmission: a survey.
Journal of Agricultural Economics 50: 581–611.
von Cramon-Taubadel, S., Loy, J. P. and Meyer, J. (2003) The impact of data aggregation
on the measurement of vertical price transmission: evidence from German food prices.
American Agricultural Economics Association Annual Meeting in Montreal, Canada—
Contributed paper.
Ward, R. W. (1982) Asymmetry in retail, wholesale and shipping point prices for fresh
vegetables. American Journal of Agricultural Economics 64: 205–212.
Willett, L. S., Hansmire, M. R. and Bernard, J. C. (1997) Asymmetric price response
behaviour of Red Delicious apples. Agribusiness 13: 649–658.
Williams, J. T. (1993) What goes around comes around: unit root tests and cointegration.
Political Analysis 4: 229–236.
Wolffram, R. (1971) Positivistic measures of aggregate supply elasticities: some new
approaches – some critical notes. American Journal of Agricultural Economics 53:
356–359.
Worth, T. (2000) The FOB–retail price relationship for selected fresh vegetables.
(http://www.ers.usda.gov/briefing/foodmarketstructures/conferencepapers/worth.pdf).
Zellner, A. (1998) The finite sample properties of simultaneous equations estimates and
estimators Bayesian and non-Bayesian approaches. Journal of Econometrics 83: 185–
212.
Zhang, P., Fletcher, S. M. and Carley, D. H. (1995) Peanut price transmission asymmetry
in peanut butter. Agribusiness 11: 13–20.


C 2007 The Authors
Journal compilation 
C 2007 Blackwell Publishing Ltd