Anda di halaman 1dari 19

Empir Econ (2015) 48:11491167

DOI 10.1007/s00181-014-0818-6

The trade effect of price risk: a system-wide approach


Dengjun Zhang

Received: 7 August 2012 / Accepted: 14 March 2014 / Published online: 2 June 2014
Springer-Verlag Berlin Heidelberg 2014

Abstract In this study, we are concern with incorporating risk factors into the Almost
Ideal Demand System model. In the modified model, the risk factors affect trade
response to price changes (the indirect effect) and the reallocation of the import expenditure (the direct effect). This specification can be used to explain why the trade effect
of risk factors is ambiguous regarding the magnitude and sign. The empirical application to the US whitefish import market reveals the proportionality between the risk
and the price effects and further implies a general positive direct effect.
Keywords

Import demand Price risk Cointegration Whitefish

JEL Classification

D81 F14 F31

1 Introduction
The relationship between risk factors and trade flows has been widely documented.
The research has focused on either the components of price risk (e.g., exchange volatility, see Bahmani-Oskooee and Hegerty (2007) for a survey of the broad literature) or
the compounded price risk (Wolak and Kolstad 1991; Muhammad 2012; Seo 2001).1
Among the industries explored, the agricultural industry is expected to be more sensitive to price risk in that agricultural goods are less storable than manufactured products
and are traded with flexible prices (Wang and Barrett 2007). Moreover, agricultural

1 Fluctuation in exchange rates is one of the inherent risk sources in actual import prices (Wolak and
Kolstad 1991). Export product prices expressed in dollars should contain all the relevant information,
such as exchange-rate volatility. Therefore, in this paper we mainly focus on price risk.

D. Zhang (B)
Department of Industrial Economics, University of Stavanger, Ullandhaug, 4036 Stavanger, Norway
e-mail: dengjun.zhang@uis.no

123

1150

D. Zhang

product differentiation is weak and firms are more numerous (Carter and GunningTrant 2010). This has led to a number of studies particularly concerned with the
effect of risk factors on agricultural products (Langley et al. 2000; Cho et al. 2002;
Wang and Barrett 2007; Kandilov 2008; Muhammad 2012; Awokuse and Yuan 2006).
However, a clear consensus has not yet been reached. In most cases, the empirical
results are insignificant or confusing in terms of a small negative or positive effect.
Unlike much of the above literature, in which single-equation specification models
are typically applied, in this study, we build risk factors into a demand system. The
risk-augmented demand system adequately reveals importers preferences toward risk
factors and provides a good explanation for the ambiguity of the empirical findings in
the literature.
As Alessie and Kapteyn (1991, p. 404) state, one obvious omitted factor in microstudies is the interdependence of preferences. The trade impact of the risk factor of
one product can be offset by the risk of other goods within the same group through the
substitution effect. Although the risk factors of products from third countries (i.e., the
competing exporters) have been incorporated into the risk-augmented demand models
for a long time (Cushman 1986; Cho et al. 2002), the single-equation estimates typically employed are inefficient and may be inconsistent. The demand system, derived
formally from utility theory, can largely avoid the pragmatic approach to equation
specification. Another advantage of the demand system approach is that the full array
of price and expenditure effects is obtained as a by-product of the estimation of the risk
effects. These estimates can be used to assess the relevant importance of risk factors
when explaining the observed trade pattern.
Among the previous studies investigating risk factors and trade, Muhammad (2012)
is the unique exception based on a system-wide approach. The author develops a differential demand system inclusive of the import price risk and finds evidence that the UK
carnation-importing firms are, for the most part, risk-averse. Muhammads model is
close to the Rotterdam demand system model. The difference between Muhammads
paper and the present paper is threefold: (i) we build risk factors into the Almost
Ideal Demand System (AIDS) of Deaton and Muellbauer (1980) and use the original version as a special case;2 (ii) while Muhammad derives the import demand
system inclusive of risk factors from production theory, in this paper, we apply the
consumer demand theory to derive the model, and hence treat imports as final goods
that enter directly into consumers utility function;3 and (iii) when estimating the
empirical model, the non-stationary property of the data-generating process is explicitly taken into consideration by applying a vector error correction model (VECM)
approach.
The linearity of the linear approximation AIDS model is an attractive feature, while
incorporating risk factors into the specification. Moreover, the modified model can
2 Different from the conventional Rotterdam demand model (Theil 1980), in which the dependent variable

is the expenditure share, Muhammad (2012) specification treats the quantity share as the dependent variable.
3 Newberry and Stiglitz (1981, p. 120) derived a one-equation specification to explore the relationships

between price risk and products consumed at the same points in time, although most literature on risk and
consumption is about inter-temporal consumptions and price / income risk (see e.g., Deaton and Muellbauer
1980).

123

The trade effect of price risk

1151

adequately reveal how price risk affects the demand decision. For example, the trade
effect of risk consists of two components (direct and indirect effects) in the extended
AIDS model. The indirect effect is subject to the magnitude of the price effect, since it
plays a role via increasing the effective price.4 The direct effect may strengthen or
weaken the indirect effect. Those facts can partially explain why it is difficult to capture
the trade effect of risk factors in the literature. Although the AIDS model has been
extensively applied to study the impacts of health information, general advertising,
and selected demographics on consumption (Alessie and Kapteyn 1991; Green et al.
1991; Duffy 1995, 2003), to the best of our knowledge, no studies have incorporated
risk factors into the AIDS model.
After recognizing the non-stationary properties of the variables included in model
specification, several economists began to apply various approaches to examine
the trade effect of risk factors (Chowdhury 1993; Langley et al. 2000) or estimate the demand system (Duffy 2003; Nielsen et al. 2011; Pesaran and Shin
2002). In the research on demand systems, when the time series data are nonstationary, cointegration analysis is a useful tool for investigating the co-movement
of several markets in the system (Pesaran and Shin 2002). The cointegration technique can further improve the estimation of the theoretical model with regard to
the adjusting mechanism of the market (Reziti and Ozanne 1999). Accordingly,
in the present paper, the empirical model is estimated by applying the VECM
approach.
The US whitefish import demand represents a good case study for empirical application. Aquatic goods, and especially wild fish, are extremely susceptible to dynamic
weather conditions. Aquatic product prices thus tend to be more volatile and the import
price risk is potentially very significant in the import expenditure allocation. The data
on whitefish traded depend on the particular species that are included, but the most
imported whitefish is cod, haddock, and pollock (Asche et al. 2009). Canada and
Iceland have historically been the largest exporters to the US, but since 2004, China
has dominated this market. On average, the annual whitefish imports from these three
countries accounted for about 85 % of the total imports in value. Therefore, for this
study, the main concern is the expenditure allocation of US imported whitefish inclusive of cod, haddock, and pollock among Canada, China, and Iceland. Over the last
decades, US whitefish imports have increased steadily to reach a peak of about 602
million dollars in 2011. However, only China benefited from the rising total expenditure at the expense of Canada and Iceland. In addition to traditional variables, like
relative prices, we want to evaluate the extent to which price risk can explain the
observed trade pattern.
In the following section, we discuss the theoretical framework. Afterwards, an
empirical model is established upon the theory, followed by the measurement of price
risk and estimation methods. The latter includes the estimation results and analysis.
The final section consists of a summary and the implications of the study.

4 Effect price refers to the price importers respond taking into account the risk premium.

123

1152

D. Zhang

2 The theoretical model


2.1 The trade effect of risk factors
Empirical research on the trade effect of risk factors indicates that risk can affect the
trade response to import prices. Wolak and Kolstad (1991) postulate that the price risk
premium is the percentage above the current expected market price a firm would pay
for a riskless supply. In terms of elements of price risk like exchange-rate volatility,
Bergin (2004) and Balg and Metcalf (2010) posit that a risk-averse firm would attach
a risk premium as an extra markup to cover the costs of volatility. This indicates a
negative trade effect of risk factors via raising the effective price (the indirect effect).
However, Newbery and Stiglitz (1981) claim an ambiguous effect of risk on efforts
for risk averters. Accordingly, price risk is likely to affect trade flows through other
channels besides the indirect effect. Indeed, a negative indirect effect of risk factors
implies reallocation of the budget between products (Grauwe 1988). This means that
an increase in price variability can have both substitution and reallocation effects,
which can offset in opposite directions. Based on a structural model, Seo (2001)
shows theoretically that there is a conflict between the risk-diversification effect (akin
to the indirect effect) and the agents preference for a certain product. Different from
the indirect effect, the reallocation effect of risk factors is directly related to trade
flows, and therefore, is termed as a direct effect. Both De Grauwe and Seo confirm
that the direct effect may offset or even dominate the indirect effect, which make it
difficult to sign the total effect of risk factors a priori.
2.2 The conventional AIDS model
The incorporation of additional variables into a complete demand system has a long
history in the studies of effects on demand of demographics, healthy information, and
advertising (Brown and Lee 1993; Green et al. 1991; Duffy 1995, 2003; Xie et al.
2009). The scale and translation procedures, which originated with Pollak and Wales
(1981), are the most frequently applied methods. Starting from a conventional AIDS
model, we will demonstrate how to use those methods to incorporate the indirect and
direct effects of price risk into the demand system.5
According to Deaton and Muellbauer (1980), the derivation of a conventional AIDS
model begins with the following expenditure function, which defines the minimum
expenditure necessary to attain a specific utility at given prices:
log c(u, p) = a(p) + u b(p),

(1)

where c(. ) is a cost function; u is the utility function, which does not take risk into
account; p is a vector of prices, and
5 We derive the empirical model from consumer demand theory by assuming that imports are treated as
final goods. This assumption is not unreasonable, at least in terms of agricultural products imported by
developed countries. See Yang and Koo (1994); Feleke and Liu (2005) and Boonsaeng and Wohlgenant
(2009) for the applications of the conventional AIDS model based on the consumer demand theory.

123

The trade effect of price risk

1153

a(p) = log P = a0 +

ak log pk + 0.5

b(p) = 0


k

kl log pk log pl ,

(2)

pk k ,

(3)

where a0 , ak , 0 , k , and kl are parameters.


Applying Shephards lemma and defining kl = 0.5(kl + lk ), the conventional
AIDS model with the expenditure share (wi ) as the dependent variable can be derived
from Eqs. (1)(3):
wi = ai + i log(x/P) +

, i j log p j

(4)

where x is the total expenditure, hence x/P is the real expenditure, P is defined in Eq.
(2), and ai is the intercept and represents the budget share when all the logarithm prices
and real expenditure equal zero. The ai can be regarded as the baseline or necessary
portion of imports from the ith supplier.
2.3 Scaling approach and the indirect effect
Taking the risk factors into account means that the specification of the utility function
in Eq. (1) should be modified as
u = u(q ),

(5)

where q is a vector of the adjusted quantity qi (= qi m i ). Here, m i is a scaling factor


varying with the level of the risk factor (vi ) in a constant elasticity formulation, i.e.,
m i = vii . The constant elasticity (i ) is positive and greater than unity on the
grounds that variability reduces the equivalent quantity with a diminishing marginal
effect. Intuitively, a high level of variability should reduce the utility maximized. Since
the budget constraint is unaffected by the impact of risk, it is necessary to transfer the
realized price ( pi ) to a riskadjusted price (effective price) as pi = pi /m i . Noting
m i = vii , there is onetoone mapping between the effective price and the risk,
reflecting the indirect role risk plays. This approach is similar to the scaling procedure
defined by Pollak and Wales (1981), which has been used to measure the impacts of
demographics and advertising in the modified demand system (e.g., Green et al. 1991;
Duffy 1995).
Accordingly, the expenditure function is re-specified as:
log c(u , p ) = a(p ) + u b(p )
where p is a vector with components p j =

pj
j
vj

(6)

123

1154

D. Zhang

2.4 Translating approach and the direct effect


As we analyzed before, a significant indirect effect of price risk also implies reallocation of the total expenditure due to the unchanged budget constraint. This reallocating
process can be reflected by allowing the baseline expenditure share (ai in Eq. 4) to
depend on the price risk. Thus, importers are perceived to define the baseline plan due
to particular preferences and/or attitudes toward risk. For example, a more volatile
price may raise the baseline share for the relevant product in order to avoid the worst
possible outcome, reflecting importers strong preference for that product.6 Although
the indirect effect of risk factors can be presumed to be negative, the direction of the
direct effect is undetermined. This suggests an inconclusive direction of the combined
risk effect on trade flows. The direct effect of the price risk is incorporated into the
demand system via modification of the price index in the demand system, because
the baselines are originally related to this index. This approach is known as the translation procedure following Pollak and Wales (1981) and has been applied by Green
et al. (1991) and Duffy (1995, 2003) to incorporate advertising into the conventional
demand model.
Decomposing the price risk effect into indirect and direct effects results in a modified
price index defined by:
log P = a0 +

ak log pk +

+ 0.5


k


k

kl

log

k j log pk log v j

pk log

pl

(7)

where the modified price index (P ) is composed of the effective prices and interaction
terms between price and risk. These interaction terms demonstrate how the direct effect
of the price risk impacts on the baseline imports when deriving the updated demand
system.
2.5 Riskaugmented AIDS model
Using a methodology and procedure similar to Deaton and Muellbauer (1980), in
which the conventional demand system (Eq. 4) is the concern, the riskaugmented
model (henceforth RA-AIDS) is derived from Eqs. (6) and (7) as follows:7
wi = a i +

i j log v j + i log(x/P ) +

ij

i j (log p j + j log v j ),

(8)

6 As Grauwe (1988) illustrates, the positive trade effect of exchange risk is plausible given a general utility

function underlying the import demand equation.


7 This specification reveals that the direct effect reflects a shift in the demand curve due to risk preferences,

while the indirect effect reflects a movement along the demand curve due to a change in the effective price
induced by a change in the price variability.

123

The trade effect of price risk

1155

where the effective prices are replaced with the components of the realized price and
p
price risk using p j = j j . While only the intercept (ai ) represents the baseline import
vj

in the conventional AIDS model, the summation of (ai + i j log v j ) in the extended
ij

model represents the baseline budget share, noting that it is the solution when all the
logarithm prices and real expenditure equal zero.
Rearranging the terms in (8) yields a linear specification of the theoretical model:8
wi = ai + i log(x/P ) +

i j log p j +

i j log v j ,

(9)

ij

where i j is the total effect of the price risk and:


i j = i j + i j j

(10)

Equation (10) implies how the estimated coefficients from the linear specification, i.e.,
Eq. (9), reveal the correlation between the price and the risk effect and between the
direct and the indirect effect. In Eq. (9), when i = j, the coefficient of the own-price
risk is ii = (ii + ii i ). Given a negative own-price coefficient (ii < 0) and hence
a negative indirect effect, the necessary condition for a negative sign of the total risk
effect is a negative direct effect (ii < 0). However, if ii > 0, a negative direct effect
cannot guarantee a negative total risk effect, although it can offset some impact of
the direct effect. On the other hand, a positive direct effect leads to an inconclusive
direction of the total risk effect if ii < 0, and even causes a positive total risk effect if
ii > 0. Thereby, expectation about the direction of the total risk effect in the demand
system is less transparent a prior. For cross-price risk effects, the sign of i j (i = j)
depends on the corresponding direct effect (i j ) and substitutability between products
i and j (i j ), indicating an undetermined direction.
3 Empirical specification and data
The source-differentiated RA-AIDS model for the US imported whitefish is based on a
multi-step budgeting process. Various aquatic products, like salmon, tilapia, and whitefish, compose the target group in the first budgeting stage. The source-differentiated
demand system for whitefish can be derived from the sequential stage. The demand
system confined to whitefish can prevent estimate bias compared with the demand
system for individual species, since the latter ignores interactions between products
within a broader group (Yang and Koo 1994).9 Considering that common factors like
the exchange rate driving the price variability of products (cod, pollock, and haddock)
8 The linear specification, which facilitates us to use the cointegration estimation techniques, is used in

the empirical application.


9 Asche et al. (2004) test a well-integrated whitefish market in France. In this instance, the bundle of

individual wild whitefish prices can be characterized using a composite price index, and hence the demand
system for the aggregate whitefish can be established.

123

1156

D. Zhang

from the same country and noting that the price risk is our main concern, in this paper,
the whitefish at the aggregate level is the primary empirical case. As a comparison,
we also estimate the source-differentiated RA-AIDS model for cod in the US market.
The empirical specification of the US imported whitefish (or cod) RA-AIDS model
is in the form:
wi,t = ai + i log(xt /Pt ) +

3


i j log pi,t +

j=1

3


i j log v j,t + ei,t

(11)

j=1

where i represents the supplier (i = 1 for Canada, i = 2 for China, and i = 3 for
Iceland); t stands for the time subscript (monthly); ei is the i.i.d. error term; and P
is approximated by a modified Stone index:10
log Pt =

3


w j,t log p j,t +

j=1

3


w j,t log v j,t

(12)

j=1

The properties of the demand system imply the following general restrictions:
3


i j = 0

i = 1, 2, 3 (homogeneity)

(13a)

j=1

i j = ji i = j (symmetry)
3

i=1

i = 1,

3

i=1

i =

3

i=1

i j =

3


(13b)
i j =0,

j = 1, 2, 3 (adding-up) (13c)

i=1

After estimating the demand system represented by Eq. (11), the estimate parameters can be used to derive the unconditional demand elasticities with respect
to expenditure, price, and variance.E ixj = 1 + i /w i is the expenditure elasticity;
E ii = 1 + ii /w i w i i /w i and E i j = i j /w i w j i /w i measure the percentage responsiveness of imports to a one percentage point change in its own-price and
cross-price, respectively;E ivj = i j /w i denotes the variance elasticity.
Monthly whitefish import data from January 1995 to December 2011 are gathered
from the US International Trade Committee (USITC) and the US National Marine
Fishery Sources (NMFS). The import values are in US dollars on a costinsurance
freight (CIF) basis and the import quantities are measured in kilograms, resulting in
the unit of prices in US$ per kilogram (dividing value by quantity for each month).
For a particular month, the expenditure share of the ith country (wi ) is calculated
by dividing the import value for country i by the total import value in this month.
The whitefish demand model uses the whole sample period (19952011). For the
10 An expansion of the modified price index (Eq. 7) shows that it is composed of two parts (the procedure
is available upon request). One part is just the normal index (Eq. 3); the second part includes only risk
variables and has the same structure as the other part. Therefore, we modify the conventional Stone index
by adding a linear share-weighted price risk index.

123

The trade effect of price risk

1157

sake of comparison, the cod demand model uses a more recent data period (1999
2011), starting from the year in which Chinas market share exceeded 10%. Each price
variable is further normalized to one with respect to the base year, 1995 and 1999 for
the whitefish and cod demand models, respectively. This treatment can minimize the
estimation problems due to the Stone index approximation (Pashardes 1993; Moschini
1995).
4 Proxy for price risk
Before estimating the models, the moments of import price distribution must be
estimated to obtain the proxy for price risk. While a clearly dominant method for
estimating price variance has not yet emerged in the empirical research (BahmaniOskooee and Hegerty 2007), the generalized autoregressive conditional heteroskedasticity method (GARCH) has become more popular since the advent of time series
analysis. The GARCH model allows for time-varying conditional variance in the original variable and accounts well for the heavy tails of the distribution of the original
variable (Kandilov 2008). One condition for the implementation of the GARCH-type
model is the existence of an ARCH effect in the price dynamics. We first test the
individual and joint ARCH effects of the three price variables in each demand system
by applying the Lagrange Multiplier (LM) method.11 The sample value is larger than
the LM critical value at the 5% level of significance for the individual and joint tests
in each demand system. Therefore, the GARCH method is employed in the present
paper to obtain the proxy for price risk.12
In the GARCH method, the stochastic error is obtained from a moving-average
AR(1) process or an autoregressive moving-average ARMA(1,1) process.13 For a
typical AR(1)-GARCH(1,1) model, the conditional variance is obtained by jointly
estimating the following equations:
 log pt = b0 + b1  log pt1 + t
vt = a0 + a1 2t1 + a2 vt1 ,

(14a)
(14b)

where  log pt is the first difference of the logarithmic price of whitefish (or cod), t
is a white noise error term, and vt is the conditional variance of t .
Figures 1 and 2 present the estimated conditional variance of each price series. For
either whitefish or cod, there is an upward trend in the variance of Canadas price.
This trend is even stronger during the 2000s, when Canada was losing its dominant
role in this market. In the whitefish market, Chinas share reached a peak at over 50 %
11 For an individual ARCH test, first the price variable is regressed on its lags to obtain the residual, and

then we regress the residual on m lags. Joint significance of the estimated coefficients of the lagged residuals
indicates that the null hypothesis of conditional homoskedasticity can be rejected.
12 We apply the univariate- instead of the multi-GARCH model to estimate the price volatility because

the preliminary estimates reflect the lack of volatility transmission. This can be explained by independence
between the causes underlying the price volatility, e.g., exchange rates and transportation methods.
13 As Pattichis (2003) pointed out, the order of the AR process has little impact on the estimated conditional

variances.

123

1158

D. Zhang

Fig. 1 GARCH estimated conditional variance of whitefish price

in 2004. In order to reflect the changes in price variance, we compute the average
annual growth of the price variances for 20002003 and 20042007. A comparison
of those two periods reveals that the highest growth rate of the price variance is 43 %
in the case of Canada and the lowest is 5.6 % for China. The moderate movement of
Icelands price occurred with a growth rate of the value of 14 %.
5 Econometric procedure
Using cointegration techniques to estimate the demand system requires the individual
data series in the system to be a unit root process. The outcomes of the augmented
DickeyFuller (ADF) test suggest that, at the conventional significance level, the null
hypothesis of non-stationarity cannot be rejected for all the variables, indicating that
all the variables are integrated of order one, i.e., I(1).
The properties of the data-generating process dictate the methods needed to test
cointegration between variables within a system. As the variables included in the model
are I(1), the Johansen procedure is the appropriate method for testing for cointegration
among the variables. In addition, one market share variable is excluded from the

123

The trade effect of price risk

1159

Fig. 2 GARCH estimated conditional variance of cod price

vector of variables in the demand system for singularity. This method is based on an
unrestricted vector autoregressive regression (VAR) approach. Both the trace and the
eigenvalue test indicate two cointegration relationships between the variables in each
demand system, consistent with the predictions from the theoretical framework (the
test results are available upon request).
Since the Johansen test is based on an unrestricted VAR model, only the cointegration space is determined consistently and not the economic parameters. This means
that the long-run parameters are not unique unless only one cointegration is identified. In the case with more than one cointegrating relation, restrictions motivated by
economic arguments need to be imposed to discriminate between different parameter
matrices. For a well-defined economic model with r linearly independent stationary
long-run relations, the r 1 just-identifying restrictions and one normalization on
each vector are required to identify each cointegration (Johansen and Juselius 1994;
Pesaran and Shin 2002). Considering the existence of two cointegrating relationships
in the unrestricted VAR model, we need 2 2 = 4 restrictions. Hence, in addition to
the two normalizing constraints with respect to w1 in Eq. (1) and w2 in Eq. (2), we
set another share variable in each equation to zero in order to mirror the theoretical
demand system. Subsequently, we restate the demand system in a VCEM form:

123

1160

D. Zhang

Table 1 Hypotheses test results


Model

Log-likelihood value

LR statistic

p value

Whitefish demand model


Unrestricted

2318.755

Homogeneity

2315.610

0.671

0.412

Symmetry

2318.429

6.290

0.043

Homogeneity and symmetry

2315.274

6.962

0.073

Risk neutrality

2296.285

44.939

<0.001

0.020

Cod demand model


Unrestricted

1169.176

Homogeneity

1163.458

5.370

Symmetry

1164.690

11.436

0.003

Homogeneity and symmetry

1162.782

12.786

0.005

Risk neutrality

1145.941

46.469

< 0.001

Yt =  Yt1 + 1 Yt1 + + p Ytg + Ut ,

(15)

where Yt is the 91 vector of all the variables in the demand system (i.e.,
w1,t , w2,t , log(xt /Pt ), log p1,t , log p2,t , log p3,t , log v1,t , log v2,t , and log v3,t ) and
g represents the number of lags for the first-difference variables.14 Here, is the
loading matrix (92), matrix (92) represents the long-run coefficients, and i
(99) is the short-run parameters matrix. As usual, Ut (91) is the vector of the i.i.d.
error terms.
6 Regression results
6.1 Hypothesis tests
Before proceeding to the empirical results analysis, we first test the general restrictions
from demand theory (i.e., homogeneity and symmetry) and a particular constraint (risk
neutrality). These additional restrictions in the VECM can be tested on the basis of
comparisons of the likelihood ratio values from the just- and over-identified models
(Johansen and Juselius 1994). The test results (Table 1) indicate that homogeneity
and symmetry cannot be rejected individually or jointly for the whitefish model, but
this does not apply to the cod model, in which only homogeneity cannot be rejected
at the 1% level of significance. Thus, only the whitefish demand system confirms
the proposition that the introduction of dynamics into an equilibrium model leads
to clear improvements regarding the test results of the regularity properties implied
by microeconomic theory (Reziti and Ozanne 1999). After imposing homogeneity
and symmetry on the demand model, the likelihood ratio value is computed from the
14 When estimating the model, the Schwarz information criterion is applied to select the number of lags

of the first-difference variable.

123

The trade effect of price risk

1161

Table 2 VECM estimates of whitefish import demand system in the US market (1 = Canada, 2 = China,
and 3 = Iceland)
Variable

Canada

China

Iceland

Constant

0.267*** (0.015)

0.260*** (0.014)

0.473*** (0.025)

log (x/P )

0.046 (0.046)

0.478*** (0.043)

0.432*** (0.076)

log p1

0.260*** (0.077)

0.591*** (0.06)

0.851*** (0.12)

log p2

0.591*** (0.06)

0.109 (0.087)

0.700*** (0.127)

log p3

0.851*** (0.12)

0.700*** (0.127)

1.551*** (0.223)

log v1

0.144*** (0.038)

0.112*** (0.033)

0.032 (0.127)

log v2

0.424*** (0.154)

0.372*** (0.137)

0.796*** (0.254)

log v3

0.232*** (0.04)

0.258*** (0.04)

0.490*** (0.069)

Numbers in parentheses are asymptotic standard errors


*, **, and *** indicate significance at the level of 0.10, 0.05, and 0.01, respectively
Table 3 VECM estimates of cod import demand system in the US market (1 = Canada, 2 = China, and
3 = Iceland)
Variable

Canada

China

Iceland

Constant

0.355*** (0.067)

0.046 (0.141)

0.598*** (0.078)

log (x/P )

0.290*** (0.088)

0.190 (0.186)

0.099 (0.105)

log p1
log p2

0.940*** (0.151)

0.930*** (0.296)

0.010 (0.166)

0.930*** (0.296)

1.091* (0.635)

0.162 (0.357)

log p3
log v1

0.010 (0.166)

0.162 (0.358)

0.172 (0.223)

0.483*** (0.106)

0.782*** (0.225)

0.299*** (0.357)

log v2

0.158*** (0.06)

0.388*** (0.126)

0.230*** (0.07)

log v3

0.363*** (0.059)

0.852*** (0.125)

0.489*** (0.07)

See notes to Table 2

constrained models with and without risk neutrality. The test results indicate that the
hypothesis of risk neutrality can be rejected at a high significance level for both the
whitefish and the cod models (p value 0.001). This means that the conventional
demand system exclusive of price risk is likely to be the subject to misspecification.
6.2 Estimation results
Tables 2 and 3 report the estimated long-run coefficients ( s in Eq. 14) of the
US whitefish and cod import demand systems.15 The homogeneity and symmetry
restrictions are imposed on the models. The regression results overall are satisfactory
in that most of the estimated parameters (especially those related to variance variables)
are statistically significant. All the own- and cross-price variance coefficients in those
15 When estimating the demand system, one equation is dropped to avoid singularity. The relevant coeffi-

cients for the dropped equation are recovered on the basis of demand constraints.

123

1162

D. Zhang

two models are significant with the exception of one cross-price variance coefficient in
the whitefish demand system. Turning to the coefficients of expenditure (the modified
Stone price index) and the price variables, for the whitefish demand model, four out
of six expenditure and own-price effects are significant, while only one expenditure
and two own-price coefficients in the cod model are statistically different from zero.16
Regarding the cross-price effects, the specification of the whitefish fits the data at a
more satisfying level; note that all the relevant coefficients are significant.
For the whitefish model, the own-price variance coefficients in Canadas and Icelands equations are negative and significant, with a value of 0.14 and 0.49, respectively. These types of risk preferences are consistently verified by the estimates from
the cod demand model, considering the significant own-price variance coefficients
with a magnitude between 0.39 and 0.49. Those negative signs of the total risk
effect indicate a generally inverse relationship between price risk and import expenditure. However, Chinas price risk plays a positive role in trade flows in terms of the
positive coefficient of own-price variance (0.37). Next, using Eq. (10), we analyze the
components of the price risk effects and the interactions between those risk effects
and the price effects.
For Canada and Iceland in the whitefish model, the own-price coefficients are
negative and greater than the own-price variance coefficients in absolute terms (0.26
vs. 0.14 and 1.55 vs. 0.49), thus indicating a positive direct effect of own-price
variance according to the decomposition of the own-price variance coefficient (see Eq.
10).17 This claim can be extended to the cases of Canada and China in the cod model,
considering the pairwise own-price and own-price variance coefficients, 0.94 vs.
0.48 and 1.09 vs. 0.39, respectively. For these four cases, the general consensus
is that the great negative own-price effect is associated with a strong trade effect of
own-price variance, in accordance with the hypothesis that the indirect effect of the
price risk is weighted by the corresponding price effect. There are two exceptions.
First, the own-price coefficient in Icelands cod equation is negative and smaller than
the own-price variance coefficient in absolute terms (0.17 vs. 0.49); however, the
own-price effect in this equation is not significant. Second, the positive own-price
variance effect is captured in Chinas whitefish equation (0.37), and this effect is
greater than the corresponding own-price effect (0.11) in absolute values. Of the
six product types, Chinas whitefish has the smallest responsiveness to the own-price
changes. This implies a small indirect own-price variance effect, since this effect is
proportional to the corresponding own-price effect. Consequently, the positive direct
risk effect outweighs the negative indirect effect and results in a positive compounded
risk effect. Since China has become the major supplier of whitefish in the US market,
the positive own-risk effect for China is probably related to the precautionary behavior
of the US importers. It means that the importers tend to protect themselves against the
risk of a higher price in the future by buying more whitefish.
16 The insignificant parameter of the modified Stone index implies a homothetic preference for the relevant
product (Xie et al. 2009).
17 According to Eq. (10), the coefficient of the own-price risk is (= + ). Considering a negative
ii
ii
ii i
one-price parameter and further noting i >1, a greater parameter of the own-price risk (ii ) compared
with the own-price coefficient (ii ) in absolute values should imply a positive direct effect.

123

The trade effect of price risk

1163

In the two demand models, both positive and negative coefficients of cross-price
variances are captured. Moreover, the sign of the cross-price variance is not always
the same as the sign of the corresponding cross-price. In the whitefish model, China
(Canada) responds negatively to changes in Canadas (Chinas) price; however, the
coefficients of the two cross-price variances are positive. This indicates the existence of
a positive direct cross-price risk effect, which offsets the negative indirect cross-price
risk effect and even dominates the sign of the total cross-price risk effect.
6.3 Elasticities and simulation
Since the estimation results from the whitefish model are somehow more appealing
in terms of tests of demand restrictions and the demand parameters, the remaining
discussion will rely on the whitefish demand system. The derived expenditure elasticities (E ix ), price elasticities (E i j ), and volatility elasticities (E ivj ) are displayed in
Table 4.18
Imports from China and Canada are more sensitive to changes in total imports (E ix
= 2.78 and 0.78, respectively). Hence, Chinas dominating market role is first due to
the rising import expenditure of whitefish during the sample period. In general, the
US whitefish imports are less sensitive to changes in Chinas price than to Canada
and Icelands prices. Chinas whitefish has the lowest price elasticity (E 22 = 1.88),
compared with Canada (E 11 = 2.22) and Iceland (E 33 = 3.52). Thus, Chinas
expansion in this market can be explained partially by the weak responsiveness of
importers to changes in Chinas price. The derived own-price variance elasticities are
0.70 for Canada and 1.38 for China. For cross effect, a 1 % change in Canadas
price variance would raise Chinas market share by 0.42 %. However, the imports of
Canadas whitefish are more sensitive to Chinas price variance; note the cross-price
variance elasticity at the value of 2.05.
From 20002003 to 20042007, average per year, the imports from China increased
by 37 % in quantity, while the imports from Canada dropped by 35 % in quantity. To
assess the extent to which the price variance variables contributed to the observed trade
pattern, we simulated the model using the elasticities in Table 4. As we mentioned
before, from 20002003 to 20042007, average per year, Canadas price variance
increased by 43 %; however, Chinas price was relatively stable, as the magnitude of
the variance only grew by 5.6 %. Considering the estimated elasticities, changes in
the own-price variance contributed to 30.4 and 7.8 % changes in the market shares
of Canada and China, respectively. Chinas increased price variance further raised
the imports from Canada by 11.6 %, while the growth in Canadas price variance
contributed to an 18.1 % rise in Chinas market share. The net contributions of the risk
factors are 2.9 and 39.5 % for Canadas and Chinas whitefish exports, respectively.
Obviously, ignoring the impact of risk factors may result in inaccurate description
about the market development.
18 The risk elasticity is derived by holding price constant. This is probably in contradiction to the existence

of causality between price and its volatility. However, there is no dichotomy of endogeneity and exogeneity of
variables in VECM, which is applied in this study. See Johansen (2005) for the interpretation of cointegration
coefficients in VECM.

123

123

See notes to Table 2

0.177*** (0.076)

1.791*** (0.259)

1.555*** (0.281)

1.884*** (0.365)

2.803*** (0.351)

2.562*** (0.256)

2.777*** (0.16)

2.215*** (0.418)

0.775*** (0.221)

E i2

E i1

E ix

Marshallian price elasticities

Expenditure elasticities

Model

0.417*** (0.122)
0.061*** (0.242)

3.524*** (0.501)

0.699*** (0.183)

v
E i1

1.669*** (0.555)

4.243*** (0.697)

E i3

1.517*** (0.484)

1.383*** (0.509)

2.054*** (0.748)

v
E i2

Price volatility elasticities

Table 4 Demand elasticities derived from whitefish import demand system (1 = Canada, 2 = China, 3 = Iceland)

0.935*** (0.131)

0.960*** (0.148)

1.125 (0.194)

v
E i3

1164
D. Zhang

The trade effect of price risk

1165

7 Concluding remarks
It is commonly argued that risk factors have a negative influence on trade flows.
However, the literature has not provided conclusive evidence for significant risk effects
in the demand models. As McKenzie (1999, p. 72) states, In general, the results of
this empirical work have been insignificant or where significant, conflicting. Most of
the previous studies on risks have used single-equation models. The estimation results
from a single-equation specification are probably biased toward finding a statistically
and/or economically insignificant relationship between risk factors and trade flows.
In the present paper, we developed a risk-augmented Almost Ideal Demand System
(RA-AIDS) model to explore how risk factors play a role in expenditure allocation.
In this model, the risk factors have direct and indirect effects on trade flows. The
indirect effect is weighted by the corresponding price effect. In the situation in which
the price effect is not strong, the minimal magnitude of the indirect effect is not
beyond expectation. There is no a priori presumption that the direct effect is negative.
A strong preference for a product can result in an increase in volume of the product
traded following increased variability of its price. Those facts can explain why the
empirical evidence in research on risk is always ambiguous regarding the magnitude
and sign of the risk factors.
The resulting model is applied to the US import whitefish market. Taking nonstationarities in the data into account, we employ a cointegrating-based vector error
correction model (VECM) approach to estimate the demand system. The significant
cross-price variance effects confirm the plausibility of using a demand system rather
than a single-equation specification. Focusing on the cases in which the own-price
effect is negative and significant, we can conclude that (i) the total effect of the ownprice variance variable is negative, indicating that the indirect effect dominates the
direction of the total risk effect; (ii) the comparisons between own-price and ownprice variance effects (see Eq. 10) indicate that the direct effect is generally positive,
which weakens the corresponding indirect effect; and (iii) the positive relationships
between the absolute magnitudes of the own-price effect and the own-price variance
effect are consistent with the theoretical proposition that the risk effect is weighted by
the corresponding price effect.
Evidence of the significant trade effects of risk factors also has implications for
econometric policy analysis, which in most cases is essentially based on price analysis.
The effectiveness of using the estimated demand parameters in policy formulation
hinges on the appropriate conceptual and empirical specification of the underlying
demand model (Davis and Jensen 1994). Taking no account of risk factors in the
demand analysis would lead to significant information loss. However, the estimated
risk effects from single-equation specification models are probably less credible. The
complete demand system can further track how risk factors play a role in the allocation
of expenditure. From the traders perspective, the implications are likely to seem
straightforward. Exporting countries interested in a target market should attempt to
stabilize the price. However, much of the price risk is induced by factors that are
beyond the control of suppliers (Wolak and Kolstad 1991). Further research is needed
to disentangle the impacts of multiple risks (e.g., exchange rate, ocean freight cost,
and exporting price risks) and provide suggestions concerning risk management.

123

1166

D. Zhang

Acknowledgments I would like to thank Henry Kinnucan, Frank Asche, Henry Thompson, and Andrew
Muhammad for insightful comments on an earlier draft of this paper. The usual disclaimer applies.

References
Alessie R, Kapteyn A (1991) Habit formation, interdependent preferences and demographic effects in the
Almost Ideal Demand System. Econ J 101(406):404419
Asche F, Gordon DV, Hannesson R (2004) Tests for market integration and the law of one price: the market
for whitefish in France. Mar Resour Econ 19:195210
Asche F, Roll KH, Trollvik T (2009) New aquaculture species: the whitefish market. Aquac Econ Manag
13(2):7693
Awokuse TO, Yuan Y (2005) The Impact of exchange rate volatility on U.S. poultry exports. Agribusiness
22(2):233245
Bahmani-Oskooee M, Hegerty SW (2007) Exchange rate volatility and trade flows: a review article. J Econ
Stud 34(3):211255
Balg BA, Metcalf H (2010) Modeling exchange rate volatility. Rev Int Econ 18(1):109120
Bergin P (2004) Measuring the costs of exchange rate volatility. FRBSF Econ Lett 13
Boonsaeng T, Wohlgenant MK (2009) A dynamic approach to estimating and testing separability in US
demand for imported and domestic meats. Can J Agric Econ 57(1):139157
Brown MG, Lee JY (1993) Alternative specifications of advertising in the Rotterdam model. Eur Rev Agric
Econ 20(4):419436
Carter CA, Gunning-Trant C (2010) US trade remedy law and agriculture: trade diversion and investigation
effects. Can J Econ 43(1):97126
Cho G, Sheldon IM, McCorriston S (2002) Exchange rate uncertainty and agricultural trade. Am J Agric
Econ 84(4):931942
Chowdhury AR (1993) Does exchange rate volatility depress trade flows: evidence from error-correction
models. Rev Econ Stat 75(4):4563
Cushman DO (1986) Has exchange risk depressed international trade? The impact of third-country exchange
risk. J Int Econ Financ 5:361379
Davis CG, Jensen KL (1994) Two-stage utility maximization and import demand systems revisited: limitations and an alternative. J Am Agric Resour 19(2):409424
De Grauwe P (1988) Exchange rate variability and the slowdown in growth of international trade. Staff Pap
Int Monet Fund 35:6384
Deaton A, Muellbauer J (1980) Economics and consumer behavior. Cambridge University Press, Cambridge
Duffy M (1995) Advertising in demand systems for alcoholic drinks and tobacco: a comparative study. J
Policy Model 17(6):557577
Duffy M (2003) On the estimation of an advertising-augmented, cointegrating demand system. Econ Model
20(1):181206
Feleke S, Liu H (2005) Aggregate demand for imported whole milk in Spain: implications for the European
Union (EU). J Food Distrib Res 36(2):2028
Green RD, Carman HF, McManus K (1991) Some empirical methods of estimating advertising effects in
demand systems: an application to dried fruits. West J Agr Econ 16(1):6371
Johansen S (2005) Interpretation of cointegrating coefficients in the cointegrated vector autoregressive
model. Oxford B Econ Stat 67(1):93104
Johansen S, Juselius K (1994) Identification of the long-run and the short-run structure: an application to
the ISLM model. J Econom 63:713
Kandilov I (2008) The effects of exchange rate volatility on agricultural trade. Am J Agric Econ 90(4):1028
1043
Langley SV, Giugale M, Meyers WH, Hallahan C (2000) International financial volatility and agricultural
commodity trade: a primer. Am J Agric Econ 82(3):695700
McKenzie MD (1999) The impact of exchange rate volatility on international trade flows. J Econ Surv
13(1):71106
Moschini G (1995) Units of measurement and the Stone index in demand system estimation. Am J Agric
Econ 77(1):6368
Muhammad A (2012) Source diversification and import price uncertainty. Am J Agic Econ 94(3):801814
Newbery DM, Stiglitz JE (1981) The theory of commodity price stabilization: a study in the economics of
risk. Clarendon Press, Oxford

123

The trade effect of price risk

1167

Nielsen M, Jensen F, Setl J, Virtanen J (2011) Causality in demand: a co-integrated demand system for
trout in Germany. Appl Econ 43:797809
Pashardes P (1993) Bias in estimating the almost ideal demand system with the Stone index approximation.
Econ J 103(419):908915
Pattichis C (2003) Conditional exchange rate volatility, unit roots, and international trade. Int Trade J
17(1):117
Pesaran MH, Shin Y (2002) Long-run structural modeling. Econ Rev 21(1):4987
Pollak RA, Wales TJ (1981) Demographic variables in demand analysis. Econometrica 49(6):153351
Reziti I, Ozanne A (1999) Testing regularity properties in static and dynamic duality models: the case of
Greek agriculture. Eur Rev Agric Econ 26(4):461477
Seo JJY (2001) Demand diversification under uncertainty and marketpower. Asian Econ J 15(4):425450
Theil H (1980) The system-wide approach to microeconomics. The University of Chicago Press, Chicago
Wang K, Barrett CB (2007) Estimating the effects of exchange rate volatility on export volumes. J Agric
Resour Econ 32(2):225255
Wolak FA, Kolstad CD (1991) A model of homogeneous input demand under price uncertainty. Am Econ
Rev 81(3):514538
Xie J, Kinnucan HW, Myrland (2009) The effects of exchange rates on export prices of farmed salmon.
Mar Resour Econ 23(4):43957
Yang S, Koo WW (1994) Japanese meat import demand estimation with the source differentiated AIDS
model. J Agr Resour Econ 19(2):396408

123

Anda mungkin juga menyukai