British Food

Journal
Welfare implications of intertemporal marketing margin manipulation
Thomas Kopp, Bernhard Brümmer, Zulkifli Alamsyah, Raja Sharah Fatricia,
Article information:
To cite this document:
Thomas Kopp, Bernhard Brümmer, Zulkifli Alamsyah, Raja Sharah Fatricia, (2017) "Welfare
implications of intertemporal marketing margin manipulation", British Food Journal, Vol. 119 Issue:
8, pp.1656-1671, https://doi.org/10.1108/BFJ-11-2016-0572
Permanent link to this document:
https://doi.org/10.1108/BFJ-11-2016-0572
)

T
Downloaded on: 18 November 2017, At: 18:35 (PT)
P

7
References: this document contains references to 32 other documents.
1

2
To copy this document: permissions@emeraldinsight.com
r

b
The fulltext of this document has been downloaded 47 times since 2017*
m

v
Users who downloaded this article also downloaded:
o

1
(2017),"Farmers’ self-reported bargaining power and price heterogeneity: Evidence from the dairy
5

3
supply chain", British Food Journal, Vol. 119 Iss 8 pp. 1672-1686 <a href="https://doi.org/10.1108/
:

1
BFJ-11-2016-0570">https://doi.org/10.1108/BFJ-11-2016-0570</a>
t

a
(2017),"Governance and quality disclosure: the palm oil issue", British Food Journal, Vol. 119 Iss
d

r
8 pp. 1718-1731 <a href="https://doi.org/10.1108/BFJ-11-2016-0566">https://doi.org/10.1108/
o

F
BFJ-11-2016-0566</a>
f

o
 y

s
Access to this document was granted through an Emerald subscription provided by emerald
r

i
srm:305060 []
n

y
For Authors
b

a
If you would like to write for this, or any other Emerald publication, then please use our Emerald
o

n
for Authors service information about how to choose which publication to write for and submission
w

o
guidelines are available for all. Please visit www.emeraldinsight.com/authors for more information.
D

About Emerald www.emeraldinsight.com

Emerald is a global publisher linking research and practice to the benefit of society. The
company manages a portfolio of more than 290 journals and over 2,350 books and book series
volumes, as well as providing an extensive range of online products and additional customer
resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the
Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative
for digital archive preservation.

*Related content and download information correct at time of download.

The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/0007-070X.htm

BFJ 7

119,8
1

1656 v

Received 28 November 2016 Revised 6 March 2017 8

Accepted 20 March 2017 1

1
)

t
T

P A

(
a

d
Göttingen, Germany and
Centre of Biodiversity and Sustainable Land
i

Use, University of Goettingen, Göttingen,

o

Germany
F

Welfare implications of Zulkifli Alamsyah

Fakultas Pertanian, Universitas Jambi, Jambo,
intertemporal Indonesia, and Raja Sharah Fatricia
Fakultas Ekonomi, Universitas Jambi, Jambo,
marketing Indonesia

margin manipulation Abstract

Thomas Kopp
Department for Agricultural Economics and
Rural Development, University of Goettingen,
Göttingen, Germany
Bernhard Brümmer
Department for Agricultural Economics and
Rural Development, University of Goettingen,
Purpose – In Indonesia, rubber is the most valuable export
crop produced by small scale agriculture and plays a key role
for inclusive economic development. This potential is likely to
be not fully exploited. The observed concentration in the crumb
rubber processing industry raises concerns about the
distribution of export earnings along the value chain.
Asymmetric price transmission (APT) is observed. The paper
aims to discuss these issues. Design/methodology/approach –
This study investigates the price transmission between
international prices and the factories’ purchasing prices on a
daily basis. An auto-regressive asymmetric error correction
model is estimated to find evidence for APT. In a subsequent
step the rents that are redistributed from factories to farmers
are calculated. The study then provides estimations of the size
of this redistribution under different scenarios.
Findings – The results suggest that factories do indeed
transmit prices asymmetrically, which has substantial welfare
implications: around USD3 million are annually redistributed
from farmers to factories. If the price transmission was only
half as asymmetric as it is observed, the majority of this
redistribution was re-diverted. Originality/value – This study
combines the approaches of non-parametric and parametric
estimation techniques of estimating APT processes with a
welfare perspective to quantify the distributional consequences
of this intertemporal marketing margin manipulation. Especially
the calculation of different scenarios of alternative price
transmissions is a novelty. The data set of prices on such a
disaggregated level and high frequency as required by this
approach is also unique.
Keywords Indonesia, Error correction model, Rubber, More than 15 million people generate their main
Asymmetric price transmission, Intertemporal marketing income from rubber cultivation, the most valuable
margin manipulation export crop. With an annual output of three million
Paper type Research paper metric tons, Indonesia is the second largest producer
of natural rubber in the world, accounting for 27 per
1. Introduction cent of global production[2] (Fathoni, 2009).
The agricultural sector is of great importance in It is likely that the importance of rubber will increase
Indonesia. In 2011 it contributed 15 per cent to the in Indonesia for two reasons.
GDP and employed 36 per cent of the workforce[1].
British Food Journal First, economic growth in emerging economies; and second the rising price of crude oil, Vol.
119 No. 8, 2017
pp. 1656-1671
© Emerald Publishing Limited 0007-070X small scale farmers and their families. It should
DOI 10.1108/BFJ-11-2016-0572
Thomas Kopp thanks Deutsche therefore be a primary policy target to
b

Forschungsgemeinschaft (DFG) for generous support m

through project KO 5269/1-1. e

ensure that these markets function properly.

which will make synthetic rubber more expensive v

thus increasing the demand for its substitute,

o

However, this does not seem to be the case. The

natural rubber on the margin.
Jambinese rubber sector is characterized
The Jambi province (Sumatra) depends crucially N

on its agricultural sector and also represents a 8

typical rubber production area. In total, 52 per cent by strong oligopsonistic market power. On the
of the workforce is employed in the agricultural processing side we observe a strong
sector and 0.6 million hectare (of 1.4 million
1

hectare) are dedicated to rubber production, of

5

concentration of demand for farmer produce, as

which 99.6 per cent are cultivated by smallholders
there are only nine crumb rubber factories
(Regional Account and Statistical Analysis 3

Division, 2012). Although Jambi is on average not 8

an exceptionally poor province, the rural in Jambi, vis-à-vis 252,000 farmers. These
population is still disadvantaged compared to factories, far from being in tight competition, are
populations in other parts of Indonesia. The
1

average income is 17.5 million Indonesian Rupiah

(1,600 USD)[3] per year (Arifin, 2005; Regional collaborating rather closely and are organized in
Account and Statistical Analysis Division, 2012), the association of the rubber processing
A

far below the national average of 26.8 million a

Rupiah (2,450 USD) per year [4]. Other d

sector, Gapkindo. A report prepared for USAID

development indicators show a similar picture; life
(Peramune and Budiman, 2007) found that
expectancy at birth is 71 years in Jambi, compared i

to 76 years in Jakarta (Regional Account and o

Statistical Analysis Division, 2012). Gapkindo is an efficiently organized and powerful

As rubber is predominantly cultivated by lobbying-institution representing the
l

smallholders, it does have the potential for

)
F

f
T

P interests of rubber processors. There are strong

economic and social development in rural areas. indications that some individual firms
In total, 252,000 Jambinese households (out of
o

( y

7 exploit their network in a way reminiscent of a

619,000) depend on rubber cultivation (Regional cartel or oligopsony (Peramune and
Account and Statistical Analysis Division, 2012).
t

1 s

r
0

2
Budiman, 2007). The link between the evidence
Therefore, malfunctions in this market can have a for asymmetric price transmission (APT)
tremendous effect on the livelihoods of e

i
r

e
and the occurrence of market power is based on manipulation 1657
Meyer and von Cramon-Taubadel (2004).
n

BFJ
In order to shed light on the price formation 119,8
process in the rubber value chain, we employ
y

a price transmission approach. As

Ahmadi-Esfahani (2009) shows, market power has
the 1658
d

potential to affect price transmission. We study the

vertical transmission between the
a

output and input prices of the five crumb rubber

factories in Jambi City from 1 January 2009
n
)

P
w
(

to 31 December 2012 via an asymmetric error 7

correction model (AECM), as done so by von

o
1

D 2

Cramon-Taubadel (1998). Previous research has r

shown that parametric estimations of price b

transmission are sensitive to asymmetries in the m

error correction process (Meyer and von v

Cramon-Taubadel, 2004). So in order to specify

o

the error correction model correctly without having 8

to rely on a priori assumptions, we employ the 1

non-parametric technique of penalized splines 5

before estimating a set of candidate parametric :

models to test which one represents the data best,

8

as suggested by Serra et al. (2006). In addition to t

demonstrating the existence and extent of market A

power, we also quantify a part of the resulting

a

redistribution of welfare from the suppliers to the i

factories. These welfare implications are shown to o

be substantial. Subsequently we calculate the F

potential for redistributing forgone profits from f

factories to farmers under different scenarios of y

altered error correction processes. t

To the best of our knowledge, this is the first paper s

combining the approaches of non-parametric and e

parametric estimation techniques of estimating i

APT processes with a welfare perspective to U

quantify the distributional consequences of this y

intertemporal marketing margin manipulation. The b

calculation of different scenarios of alternative d

price transmissions and data set of daily prices on d

such a disaggregated and local level required by o

this approach are also unique.

l

Intertemporal marketing margin w

The paper is organized as follows: Section 2

provides background on the rubber market in the village-level the traders’ market power seems to
Jambi province and introduces the typical be based on the farmers’ credit constraints as well
marketing chain for natural rubber originating from as asymmetric information between traders and
smallholder production in this area. In Section 3 farmers (Kopp and Brümmer, 2015).
the rationale behind APT is discussed. Section 4 is In this paper, however, we are focussing on the
dedicated to model development, and Section 5 market power at the next stage: the factory gates.
presents the statistical results. Section 6 derives The incriminating indicators are strong. During our
the resulting welfare implications before the traders survey in 2012 some respondents claimed
concluding Section 7. that they were victims to the market power of
downstream stakeholders (other traders,
2. Background warehouses, and factories). It seems possible that
2.1 Rubber marketing in the Jambi province the critique they express is justified to some
The structure of the Jambinese rubber sector is extent. Each of the five factories located in Jambi
displayed in Figure 1[5]. Most farmers sell to a City report the price that they are paying each day
village trader who then has the choice between for their main input – slabs of raw rubber – to one
three different kinds of stakeholders: a factory, a central agent (their association). They also have
warehouse or another trader, for example at the the option to attain information on their
district level. This choice is influenced by various competitors’ prices from this agent. This enables
factors, including the remoteness of the trader, his each factory to control its competitors’ pricing.
or her capital, access to credit and information, Another piece of evidence for the power of the
etc. (Sujarwo et al., 2014). factories is the standard procedure that follows
There are approximately 16,000 traders in the when an external investor wants to construct a
Jambi province and nine factories, five of which new crumb rubber factory. Before getting the
are located in Jambi City (Regional Account and required permission by the government, officials
Statistical Analysis Division, 2012 and own survey will first consult with the rubber processors’
data, 2012). association on whether to give the permission or
not ( Jambi Provincial Government Office for Trade
2.2 Market concentration at the processing stage and Industry, personal communication, 2012).
It appears that oligopsonistic market power occurs
at several stages in this value chain. On the
(39)
Village
Farmers District
Traders Jambi-City
(251,000) 17% Traders
(16,000) 55%
61% 32% 100% Factories (9)
50% Warehouses
3%

Figure 1.
Marketing channels for rubber

29% 33% Note: The percentages indicate which marketing

channel is employed and for how often Sources: Authors’
calculations, based on own survey data (2012) and Euler et
al. (2012)
Anwar (2004, cited in Arifin, 2005) argues that the
margin of Jambinese rubber factories is much Organization provides a direct
higher than those in other provinces. While Anwar t

argues this to be the result of close geographic A

approach to finding indications of market power

proximity to one of the most important export
(Bresnahan, 1982), these approaches
market ports (Singapore), it is much more likely
that this observed increased margin stems from
a

the oligopsony power of the rubber factories require detailed information on the firms’ demand
(Arifin, 2005). and supply structures, data we do not
This market power has a tremendous effect on the
i

distribution of welfare, both for the rubber farmers

o

and Jambinese society in general. Welfare loss have access to. One alternative way of finding
experienced by the farmers to the factories is due empirical evidence for the existence of
F

to lower pricing: the factories cause a transfer of f

income from the farmers to themselves by paying market power is by testing for a non-constant
prices below the competitive level. The welfare transmission of price changes
loss to society in general stems from the farmers’ o

total production below the competitive level y

because of the reduced raw rubber prices. In the (Kinnucan and Forker, 1987; McCorriston et al.,
long run it is reasonable to argue that the farmers 2001; Meyer and von Cramon-Taubadel,
could increase their rubber output, for example by
i

shifting their production from palm oil to rubber.

r

After 20-25 years an oilpalm plantation has to be 2004). Weldegebriel (2004) derive from a
replanted and the investments required for e
theoretical model that the adjustment coefficient
replanting oilpalm or rubber are similar. v

However, as the supply function of rubber farmers alone cannot be used as an indicator for market
is unknown, it is not
)
power. However, their argumentation
n

T
U
P

possible to derive how much the supplied quantity,refers to the absolute level of the adjustment
and thus total welfare loss, would
(
coefficient while our study focusses on
y

b
7

be in the case of a general improvement of the differences between adjustments to positive and
price level. Therefore, we will concentrate
1 negative changes of the leading price.
d
0
e
2

on the farmers’ welfare loss based on a below

d

The asymmetry that we are discussing here

competitive market price in times of implies that positive changes of an agent’s
a
r

e o

price hikes. We show that this oligopsonistic selling price are passed on to the provider at a
market power is exercised and how large
b lower speed than negative price changes.
n

m
w

the welfare loss to the farmers is, which results This means that when the agent’s margin
from intertemporal marketing
e
increases – that is in times of international price
o
v

D
o

margin manipulation. hikes – the buying price is corrected slower than

when the margin decreases, enabling the
N

processors to enjoy excess profits.

8

5 If asymmetries in the short-term dynamics occur

3. Methodology (not only in the adjustment parameter) it would
3

: also be interesting to analyse these dynamics via

8

1
impulse response functions. As we will see
While the literature of the New Empirical Industrial however, there are no asymmetries in the
short-run dynamics, so the generation of impulse A

response functions would not increase the quality a

of information.
d

The assumption behind the APT between the o

international rubber price and the Jambinese price

l

for raw rubber is that the factories are price takersf

at the international market. While Indonesia as a

o

whole can be assumed to have an influence on

y

international rubber prices, this is unlikely to be

i

true for the five Jambinese factories under

r

consideration. Mundlak and Larson (1992) finds v

that world market prices are the main reason for n

variations in domestic prices, also in the presence

U

of policy intervention. In the domestic market,

y

however, it can be assumed that they are price d

setters. (Both these assumptions are tested

e

below). One can therefore understand the shocks a

that arise in the first one as exogenous and the l

ones arising in the latter as reactions to that 3.1 Non-stationarity and co-integration
shock. Given that we are working with prices, a
non-stationarity nature of the data is expected
Intertemporal marketing margin which is tested via the Augmented Dickey-Fuller
manipulation 1659 (ADF) test with both variables of interest (ln_ pSell
and ln_ pBuy). As will be shown, they are indeed
BFJ non-stationary, which we address by taking the
119,8 first differences. We will then test whether the two
series are co-integrated which is done by
employing both the Johansen test ( Johansen,
1998) and the Engle-Granger Two-Step method
(Engle and Granger, 1987). For both tests we
1660 need to find the optimal lag-length. As we are
using daily data, it is likely that the price of one
day depends also on past shocks. To select the
optimal number of lags we consider the Akaike’s
information criterion (AIC), Schwarz’s Bayesian
information criterion (SBIC), and the Hannan and
Quinn information criterion (HQIC).
)

P 3.2 Error correction model

We assume a multiplicative markup model. The
(

rationale for using a multiplicative instead of

7

2
additive model is that the margin is assumed to be
r a percentage markup. This has been concluded
e

b
from qualitative interviews with representatives at
m the rubber factories. We tested both approaches,
and the results confirmed that taking the logarithm
e

N
represents the data better. pBuy
8

t
t refers to the buying price at time t and pSell
t to the selling
price. The long-run “co-integrating” relationship in

its logarithmic form is given by: lnpBuy

Sell
t b^0 b^1lnp
ectt ¼ lnpBuy
t (2)

In the case of a positive price shock on the

international level (i.e. a positive deviation from
the long-run equilibrium in which the factories’
margin increases) the ect will be o0 and if the
price is shocked negatively, the ect isW0. The
error correcting process (symmetric case) is
expressed as:

t¼ b0 þb1lnpSell
t þe (1)

which we estimate with the Johansen method.

The reason for doing so (despite our general
approach of the Engle-Granger two-step method)
is that the Johansen approach delivers better
results when estimating the co-integrating
relationship (Gonzalo, 1994). From the residuals
of this relation we can generate the error
correction term (ect) which is defined as follows:
 M

t¼ x0þaectt 1 þ X
DlnpBuy goDlnpBuy
n

w
to
Sell
t oþloDlnp þe (3)
o

o¼1

M is the number of lags, ξ0 a constant, and γω and λω the coefficients of short-run

dynamics. ε represents an error term.
For the thoughts laid out in the theoretical section above, the model is extended
to a threshold error correction process, which is the generalization of a simple
asymmetric adjustment. The existence of any threshold is tested for with a SupLM
test as suggested by Hansen and Seo (2002). Based on Model (3) the ect is split
up into N regimes by N−1 thresholds, which are located at Ψλ for λ∈[1, …, N−1]
and ectBt :¼ ectt if Ψς−1oectt⩽Ψς for ς∈[1, …, N ]:
goDlnpBuy
¼ x XN aBectBt 1þ X þe (4)
t 0þ
M Sell
Dlnp Buy þl
to o Dlnp
to
B¼1 o¼1
 An “asymmetric” process, which is the simplest form of a threshold error
correction, is characterized by the parameters N ¼ 2 and Ψ1 ¼ 0, i.e. the ect is
split into two variables, one for negative and one for positive price changes[6].
3.3 Non-parametric estimation To start with, we estimate a simple linear error
Most authors in the literature on threshold error correction model (M1) which corresponds
correction models use a parametric estimation
:

technique (Hansen and Seo, 2002; Lloyd et al., 2006; 1

Ihle et al., 2012). The drawback of this procedure is that to the model described in Equation (3). The second
one has to make certain a priori assumptions for model (M2) is an AECM which
specifying the model, such as the number of thresholds.
t

In order to overcome this limitation, we employ a

corresponds to Equation (4) with the specifications N ¼
non-parametric approximation. While using
non-parametric estimation techniques to detect 2 and Ψ1 ¼ 0. For the third model
a

unknown relationships is a widely used technique in the d

statistical and financial literature (Krivobokova et al., (M3) we assume a one-threshold model with no
2010; Escribano, 2004), in the agricultural economics restriction on the location of the threshold.
literature this has not been often employed. One r

exception is the work by Serra et al. (2006) who use a l

local polynomial fitting approach in order to understand The rationale behind model three (M3) is that the price
the error correction process without the need for F
gets corrected quickly during price
restrictive a priori assumptions. Contrary to that, we f

work with penalized splines (Eilers and Marx, 1996). drops (regime 3) and moderate hikes (regime 2) when
Regression splines consist of the sum of a number of the factories generate a normal
polynomial functions. The spline is fitted to match the o

data by giving each of these functions an individual y

shape. Penalizing the splines refers to the method of margin. In times of large price increases (regime 1)
including a penalty-term, which smoothes the spline by however, the prices get corrected at a
penalizing excessive zigzagging (i.e. big
)
i

T e

P
much slower rate; the factories generate a greater
differences between neighbouring values) of the spline
margin. M3 corresponds to Equation (4)
(Wood, 2003).
(
v

1 with N ¼ 2 and an unknown value of Ψ1[7].

0 U

r The exact location of this threshold can be found via a

3.4 Candidate models for parametric estimation
e
grid search approach. We test
y

b
b

In order to get to know the exact slope-coefficients d

necessary for calculating the each possible value of the ect as the threshold value
m

e
Ψ1, estimate the model and save the
v e

distributional effects, we continue with a parametric d

log-likelihood value. We then select the model with the

regression approach. Several
highest log-likelihood.
o

N
a

approaches are employed to model the error correction o

process before the model that

l

w
8
o
1

represents the data best is chosen via a testing 3.5 Threshold determination and model choice
procedure described below.
D

3
We find the threshold of model M3 via the grid search
following the method laid out above. No assumptions smallholders due to slower price transmission in times
are made about the location of the threshold until then. of tremendous price hikes, compared to a baseline
After estimating the different models described (M1-M3), scenario of the fastest adjustment possible which is
we test which of them represents the data best. As we assumed to be the adjustment that occurs in times of
compare models with different specifications concerning price decreases (see Figure 2). As discussed above, we
the number of regimes (one and two), we rely on do not focus on the total welfare effect because the
information criterion again. We employ the AIC which is price elasticities of the supply and demand are
in this case superior to other information criteria as laid unknown. The part of the welfare effect which stems
out by Burnham and Anderson (2002). from the intertemporal marketing margin
Intertemporal marketing margin
3.6 Distributional effects manipulation 1661
The quantification of the distributional effects stemming
from APT is based on the forgone income of
BFJ (a) (b)
119,8
price
supply
Price t =100 (99% corrected) Price t =100 (99% corrected)
price
supply
Redistribution

1662 Figure 2. Price t =0 (before shock)

Price t =1 (quick correction)

Price t =1 (slow correction) Price t =10 (quick correction)
Redistribution
Price t =0 (before shock) Price t =10 (slow correction)
T

710
Notes: (a) Time=t+1; (b)
Welfare effect during time=t+10 Source: Authors’
adjustment process after draft
shock at t ¼ 0 quantity quantity
)

2
manipulation is calculated as the difference between the price that is theoretically possible in
r

e
times of price hikes and the price that is actually paid, multiplied by the quantity.
b

m
In order to quantify the effect that the intertemporal marketing margin manipulation had
e

on all Jambinese farmers, we calculate the differences between two hypothetical scenarios of
o

8
local price development after 14 periods (the time after which a farmer sells his/her produce
1

5
is around two weeks) following each shock to the global price during 2009-2012. The two
3

8
scenarios differ in the assumed adjustment parameter, following the results from the AECM.
1
 t
We start with the following equation:
A

lnpBuy
t¼ lnpBuy
Buy
t 1 þDlnp

r
t þe (5)
o

in which we substitute DlnpBuy

o
t from a simplified version (without lagged prices)[8] of

i
Equation (3) and then ectt from Equation (2) in order to calculate the adjusted price after one
s

v
period[9]:
i

lnpBuy t1 b^0 b^1lnp

Sell
Buy
t ¼ lnp
þe (6)
y da

t 1 þa^ lnpBuy t1
b

de

o
Iterating this procedure 14 times generates the price after 14 periods after the shock in
l

w
period 1. In the computation the error term is set to zero. The difference pdiff between the two
o

scenarios is given as:

D

Þ t þ14
(7)
pdiff ¼ exp lnpBuy a þ ð exp lnpBuy a ð Þ t þ14

The total redistribution (RED) based on intertemporal marketing margin

manipulation is then the sum of all price differences, multiplied by the quantity sold
at time t+14:

T
RED ¼ X 4. Data

t qt þ14 (8) pdiff

t¼1

The daily buying prices of the five factories in Jambi City were provided by
Gapkindo. There is one price for each factory available for each day between 1
January 2009 and 31 December 2012, except for Thursdays, public holidays and
one holiday week in August or September.
Out of these five series, an unweighted average for the 1

Jambi-buying price was generated. The selling prices buying price (α¼ −0.0152704, p-value¼ 0.511), while
were drawn from PT. Kharisma (2013), a marketing the reaction of the buying price is strong
t

company located in Jakarta. These prices represent the A

average results of the auctioning of Standard and highly significant (α¼ −0.0593225, p-value¼
Indonesian Rubber (SIR20) on each day rubber was 0.001)[10]. Using the Engle-Granger two-step
sold (maximum four days per week, except for two a

weeks of holidays in December). In combination, this

d

gives us 701 days for which we have both selling and approach results in a very similar adjustment parameter
buying prices. The price series are graphed in Figure 3. of −0.0582281 for the buying price
r

5. Results and is also highly significant ( p-value¼ 0.001). Hence,

5.1 Non-stationarity and co-integrating relationship the use of the Engle-Granger two-step
The initial suspicion is confirmed. The series are indeed
F

both non-stationary (the H0 of non-stationarity cannot

f

be rejected at a confidence level of 10 per cent). To y

avoid the problem of spurious regressions, we take the

t

first differences. As the results of the ADF test show (H0 11

s

can be rejected at a 1 per cent confidence level), this

r

solves the problem. The SBIC suggests a lag-length of v

the order two, the HQIC three lags, and the AIC opts for n

four lags.
U

Following Ivanov and Kilian (2005), who suggests we

trust the AIC in situations of large
T

sample sizes (W250) and data of relatively high

frequency (Wweekly), we use four lags.
7

The second reason for choosing the lag order 10.5

y

suggested by the AIC is the danger of biasing

0
b

d
2
e

r d

the results by under-parametrizing the model; 10

a

over-parametrizing does not cause too much

e
o

b n

damage (Gonzalo, 1994). So the lag length was

w

specified as four periods in each case,

m
9.5
D

including a constant and without trend. Test results are 9

available on request.
o

N
8.5
From the test for a simple (i.e. non-threshold) ARVECM 1 January 2009 1 January 2010 1 January 2011 1 January 2012
with the Johansen method we can
8

confirm our assumption that the factories are clearly

price-takers on the international market
5

and price setters on the domestic market. The selling

price does not react significantly to the
:

8
 Intertemporal marketing margin
manipulation 1663

date
 data_availability
In_pBuy

In_pSell
Source: Authors’ production
Notes: Values are the logarithm of the prices in
Indonesian Rupiah. 1.00 USD=0.93 Indonesian
Rupiah (December 2013). The green bar indicates the Figure 3.
existence of data, so the holes in the green bar Time series of buying and selling prices
represent days without data. In the graph, the last point
before a gap was connected with the first one after it
T di
r
P
o
(
l
7
F
10
2
r f

e
o

y
m
t
e i

v s

o r

e
N
v

in
8
U
1
y
5
b
3:
de
8 da
o
1
t ln

w
A
o

) a
D
BFJ
119,8
1664
Table I.
Estimates of long-run relation
approach seems appropriate. The
co-integrating relationship is
presented in Table I. We continue gets corrected more rapidly in
the analysis using the residuals of cases of negative price-shocks
the co-integrating relationship
generated with the Johansen
method ( pBuy¼ 0.45( pSell)1.07), changes in their specification[12],
following the results of Gonzalo the confidence intervals widen
(1994) who finds that the Johansensubstantially at the rather extreme
method delivers the best results values in [−∞; −0.1[∪]0.1; ∞ ],
when estimating long-run which is caused by the small
relationships. An F-Test confirms number of observations in those
that the constant is significantly (1 areas (Figure 4).
per cent level) different from the
value one. Testing the residuals 5.3 Model choice
with the ADF test yields a test Table II presents the AIC values of
statistic of −6.980, with which we the models M1-M3. Following this
can reject the H0 of criterion, M3 represents the data
non-stationarity at the 1 per cent best. Executing an F-Test indicates
level. The results of Hansen and that the two slope-coefficients of
Seo’s (2002) SupLM test indicate Model 3 are different from each
the presence of a threshold, as the other with a significance of 6.58
H0 of an error correction process per cent. The following discussion
without a threshold can be rejected is therefore based on the
at a 10 per cent level (robust two-regime model with one
SupLM), and respectively a 1 per threshold at −0.038 (M3).
cent level (standard SupLM) of
significance. 5.4 Parametric regressions
The estimation results are
5.2 Penalized splines presented in Table III. The
Figure 2 shows the penalized specification of M3 stems from a
spline (blue line). The dotted lines one-dimensional grid search. Its
represent the 5 per cent results are shown in Figure 5. The
confidence intervals[11]. In order to display of the likelihood values
deal with the small numbers of shows two peaks which indicate
observations at both ends of the possible locations for the
population, we add a thin plate threshold, one at the ect value of
penalized spline for comparison −0.038 (splitting up the ect into one
(bronze line) (Wood, 2003). The regime of 135 observations and
thin plate regression splines one of 571 observations) and one
penalize by compiling the spline of at the value of 0.052 (662 and 44
the group of functions which are observations per regime).
the most relevant. These are Considering that the likelihood
chosen via an eigenvalue values are nearly identical
decomposition. (2,226.714 with the threshold at
The splines exhibit narrow the 135th observation vs 2,226.863
confidence intervals in the area of at the 662nd observation) but the
many observations and show one latter value produces one regime
threshold in the region [−0.05; 0], of only 44 observations, we chose
thus indicating at least two the first possibility[13].
regimes. The two regimes can be
characterized as follows: the slope ln_ pBuy OLS Johansen
is steeper for positive values of
ln_ pSell 1.067*** (0.0071) 1.067***
ectt−1, which means that the shock (0.0186) Constant −0.811*** (0.0723)
−0.800 Observations 701 701 R2 0.982
Notes: Since the VEC is not linear, it does
than in the case of positive not report t-statistics. The Johansen
price-shocks. While in the area results have four observations less,
[−0.1; 0.1] the splines are robust to because they include lags, while the first
step of the Engle-Granger method does
not require the inclusion of lags. Standard
errors in parentheses. ***po0.01
–0.05

–0.10

t
(

–0.2 –0.1 0.0 0.1

Intertemporal marketing
) margin
manipulation 1665
T

0.05

0.00

2
r

Source: Authors’ production

I_ect (t–1) Figure 4. Penalized splines
y

(M1) (M2) (M3)

Model ln(L) k AIC Rank e

5 v

:
d_ln_ pBuy Regular OLS One threshold One threshold
i
8
n
1

M1 2,223.7814 10 −4,427.5628 3
t

M2 2,224.8331 11 −4,427.6662 2
A

M3 2,226.7141 11 −4,431.4282 1
d

f
(at zero) (at −0.0383844)
U

y
b
Table III displays the results of the three models of the
L.ect −0.0583*** (−4.234) parametric estimation. On average (column M1), 5.83
per cent of a price shock is corrected per day. If the
d

L.ect_ pos −0.0875*** (−2.954) −0.0935*** (−4.284)

buying price deviates
a

L.ect_neg −0.0473*** (−2.601) −0.0438** (−2.561) Table II.

l

n
Results of Akaike information criterion
LD.ln_ pSell 0.156*** (5.676) 0.149*** (5.055) 0.145*** (4.882)
w

L2D.ln_ pSell 0.139*** (4.535) 0.136*** (4.385) 0.134*** (4.289)

D

L3D.ln_ pSell 0.109*** (4.078) 0.110*** (4.115) 0.110*** (4.124)

L4D.ln_ pSell 0.0364 (1.136) 0.0360 (1.121) 0.0357 (1.113) LD.ln_
pBuy 0.0544 (1.081) 0.0541 (1.070) 0.0543 (1.069) L2D.ln_ pBuy
−0.0192 (−0.371) −0.0211 (−0.411) −0.0222 (−0.433)
L3D.ln_ pBuy 0.0365 (0.893) 0.0330 (0.817) 0.0308 (0.772) L4D.ln_
pBuy 0.130** (2.057) 0.125** (1.971) 0.124** (1.969) Constant
4.98e-05 (0.125) 0.000646 (1.112) 0.000529 (1.260) Observations
701 701 701
R2 0.387 0.389 0.392 Notes: Robust t-statistics in parentheses.
**po0.05; ***po0.01

6. Discussion Table III.

6.1 Interpretation of coefficients Results of all models discussed
e

k
i
l

2,225
_

1666 o
l

2,224

0 200 400 600 800 ID_ect

Figure 5.
Results of
one-dimensional grid search
)
2,227
T

71

BFJ
2,226
119,8 d

h
i
l

Source: Authors’ production

0

2
 r

from the long-run equilibrium price by 100 per cent for example (i.e. it is half of what it
v

N
should actually be in the long-run), 5.83 per cent of that shock is, on average, corrected the
8

1
following day. This is equivalent to an average half-life of a price shock of 11.4 days.
5

3
Reasons for these deviations include a shock to the international price, or past shocks which
:

1
have not been fully corrected.
t

A
When accounting for the asymmetric price adjustment, the picture looks different.
a

d
During the last four years, after 135 out of 390 price hikes (positive shocks to the price,
i

i.e. ect o 0), which is roughly one-third of these cases, the price was corrected significantly
l

o
slower than during price declines. More specifically, these 135 cases occurred at times of
y

t
extreme price hikes, i.e. ect o−0.038. It takes 16.5 days to correct half of a strong positive
i

e
price change and only 7.5 days in the case of a negative or small positive shock (see Figure 6.
v

n
The simulations are based on Equations (6) and (7)). The sign of the threshold value is
U

30,000
a

o
l

28,000

26,000

24,000

22,000
0 20 40 60 80 100
periods
target_pBuy pSell_shocked
pBuy_corrected_slowly pBuy_corrected_quickly
Figure 6.
Correction of shocks over time Indonesian Rupiah Source: Authors’ calculations
Note: In Indonesian Rupiah; 1.00 USD=10.93
counterintuitive (negative ects refer to positive a shock is 49 days in the case of a negative shock
price changes) because the ect is defined as the and 107 days in the case of a strong positive
long-run equilibrium price minus the actual price in shock. t are positive and significant while the lags
that period. This means that when the international of pBuy
price sinks, the factories’ buying prices decrease
twice as fast as when the international price rises
Intertemporal marketing margin
strongly. The time needed to correct 99 per cent of manipulation
The lagged values of pSell t are insignificant
which supports the results from the Johansen test v

(subsection non-stationarity) above that pSell market power (von Cramon-Taubadel, 1998), this kind of
analysis cannot provide a definite
t is the leading price.
o

There are two explanations as to why the price shocks

N

are not transmitted in an instant (9.4 per cent per period “proof” of market power. Meyer and von
is a very quick error correction, considering that we are Cramon-Taubadel (2004) show that APT is not
8

working with daily data). First, technical reasons in the 1

factories, such as communication between the selling necessarily caused by market power. In their literature
and buying departments. The second reason is more of review, they present an overview of
a methodological issue. For the analysis, the average
5

prices of five crumb rubber factories was generated. :

reasons for APT other than market power, arguing that

There are always small differences between the five
prices. These small differences impact the average in a evidence for APT is not equivalent to
8

way that leads to an apparent short delay in the 1

transmission time that is a proof of market power. For the case of our study
however, the alternative explanations that
)

T
t
P

the average between the firms.

A

can lead to APT discussed in the literature can be ruled

(

out. This leads us to the conclusion

7

1
a
0
d
2
i

r
r

6.2 Market power or not? that the APT observed in the rubber processing sector
e

b
in Jambi is indeed most likely caused
o

As asymmetries in price adjustments can result from

by market power, based on cartel-like behaviour or an
different kinds of processes other than
m oligopsonistic market structure:
F
e
 119,8
f

(1) “Menu costs” or the costs associated with changing

the price: the prices the factories
t

1668
s

pay to their suppliers change every day. There is no

reason to believe that the costs
e

of changing the price depend on the direction of the

price change.
n

U
)

T
y

(2) Fixed costs forcing a firm to operate close to its

P

production capacity: as the agricultural

b
7

0
d

input (the slabs of coagulated rubber) is extremely

2

durable; the factories always have a

e
e

d
m
a

stock available big enough to keep the factory running

e

for more than a week.

o

o N

8
n
1

(3) Perishability generates an incentive to sell the 5

product quickly: processed crumb

w
3

8
o
1

rubber is not perishable.

t

(4) A strong inflation in times of rising prices leads to a

data that exhibit asymmetry: while the inflation of the

d

Indonesian Rupiah is greater than that of the US dollar,

r

for example, it is not great enough to have any impact

l

on a daily basis which is the horizon of our data. f

(5) Policy interventions, price support, etc. can also lead y

to APT, but this has not occurred in Jambi (or on a t

national level in Indonesia) during the timeframe under s

consideration. Neither were new factories built or a new e

national/provincial government elected. Other input i

prices have also stayed constant (energy has been

n

subsidized at a constant level, and the minimum wage y

did not change during the time of analysis). Hassouneh b

et al. (2010) suggest searching for a regime-switch d

which was not found in the data. d

(6) Processing time: though a delayed reaction (caused o

by processing time) in combination with high inflation n

can show misleading signs of APT, this does not apply

w

here for two reasons: while it is true that inflation is high D

in Indonesia (4.3 per cent in 2012)[14], we work with daily data.

During the typical reaction times price hikes due to
1667 inflation are close to zero. Besides that, we are
BFJ observing a potentially mono- or oligopsonistic setting,
implying that the shock that hits the leading (selling)
price occurs after the processing. If factories who set the below free-market prices can be assumed to be
their buying price (and take their selling price) would substantial, too. For a single farmer, this amount
want to set the buying price according to what they represents 2.25 per cent of his or her annual revenue.
receive for that specific load of rubber after processing, Considering that around 32 per cent of the revenue
they would have to anticipate the price after the turns into income (calculation based on Euler et al.,
processing already at the time of purchasing. This is 2012), the calculation of the forgone income is based on
impossible. the following: profit π is equal to rs with r being the
(7) Non-cooperative game: there are cases where it revenue, and s the profit share of the revenue (32 per
looks like price-fixing has happened, while in fact there cent). π ¼ r−c with c representing the costs. The
was no outspoken agreement. It occurred in situations possible increased revenue (if the price transmission
in which firms could threaten to punish another firm was symmetrical) r′ is equal to r(1 + x) with x being the
which deviated from the cartel-solution (Perloff et al., percentage share of the possible increase of the
2007). However, only rarely could it be argued that revenue (2.25 per cent). Then π′ ¼ r′−c. The potential
these companies would have had an agreement that increase of income can be calculated as:
was not the subject of debate, especially given the fact 0
that in all other respects they are such close p ð Þ p =p ¼ ð Þ rð Þ 1þx c ð Þ rþc =ð Þ rs (9)
companions. Besides, even if there were indeed no
According to this analysis, increasing the revenue by
explicit agreement on pricing, the oligopsony-hypothesis
2.25 per cent would have led to an increased income of
would still hold.
7.03 per cent. So effectively each farmer could have
An explanation as to why cartels adjust (increase) their generated 6.97 per cent more income when the prices
buying prices at all – i.e. why they do not always pay a were increasing by more than the threshold value.
low price to the farmers – is that even cartels face In the second scenario we find an estimated forgone
restrictions concerning their price setting. There is income of only 13.5 billion IDR (1.2 million USD) for the
always one margin that cannot be exceeded without farmers due to the factories’ price rigging. Due to the
risking government interference. This is the margin that logarithmic nature of the price transmission effect, this is
is realized in times of constant or falling prices but significantly less than half of the amount calculated in
temporarily increased when prices rise. scenario 1, although only half of the asymmetry is
balanced out.
6.3 Distributional consequences
In order to account for the distributional consequences 7. Conclusions
of the intertemporal marketing margin manipulation we Indications that the five rubber processing businesses in
calculate the difference between the observed Jambi City, Sumatra, possess market power and use it
distribution of rents to hypothetical distributions in two to rig the prices they pay to their suppliers are strong. In
alternative scenarios (Equations (6) to (8)). this paper we found evidence for an asymmetric
The first scenario assumes perfect symmetry by setting transmission of prices, which has led to a great
α− equal to α + . This means that the asymmetry is fully redistribution of revenue from the farmers to the
eradicated. In the second scenario half of the processors during the four years of observation.
asymmetry remains. This is operationalised by setting Compared to a non-monopsonistic market situation, the
rubber farmers have missed out on an income of 7 per
α− to the arithmetic mean between the two estimated
cent. The net welfare loss generated in the process
coefficients (6.9 per cent). could not be quantified in this analysis but can be
The 252,000 Jambinese rubber producing smallholders assumed to be substantial. It is likely that these kinds of
produce 281,000 tons of rubber per year on average processes occur all over Indonesia.
(Regional Account and Statistical Analysis Division, The group has achieved its advantage by correcting
2012) and we assume them to sell, on average, the price changes on the international
same amount every day at which they sell. )

The results for scenario 1 yield an estimated forgone

T

revenue of 32.4 billion IDR (3.0 million USD) for the market (where its members act as price takers)
Jambinese rubber farmers in times of rising prices in asymmetrically. If the international price
P

every year. This is only the amount that was

(

redistributed from farmers in Jambi to the factories, due

7

drops, the buying price decreases much quicker than in

to the APT of the factories. The total welfare loss due to
times of great price hikes.
1
0 d

All alternative explanations for APT – other than market

e

power – could be ruled out for the

a

o
2
l

r Notes
n
e

rubber processing sector in Jambi. Risk managing w

strategies would lead to a generally lower 1. WorldBank World Development Indicators. Available at:
b
http://databank.worldbank.org
o
m

price level but not to a different reaction, depending on D

(accessed March, 2015).

the direction of price changes.
2. FAO FAOSTAT. Available at: http://faostat3.fao.org/home/E
e

One policy recommendation that could be drawn from (accessed March, 2015). 3. Exchange rate (December 2013)
our results is to involve all from Oanda Corporation.
N

4. WorldBank: WorldBank World Development Indicators.

stakeholders (including farmer representatives) in Available at: http://databank. worldbank.org (accessed March,
consultations before deciding to grant
8
2015).
5. Farmers’ marketing channels do not add up to 100, because
1

5
they sell a minor share on auction markets (6 per cent) where
permits for the construction of new crumb rubber
the buyer is unknown, as well as to farmers’ associations (1
factories. If more factories were
3 per cent). The missing 13 per cent of the district traders stem
from the fact that they can also sell to another trader, which
:

competing for the input of raw rubber, the general price was omitted from this graph.
level would be expected to increase.
1 6. I was checked whether the non-linearities in the error
t correction do not only occur in dependence of the direction of
The calculation of an alternative scenario shows that if the price change, but also in the form of structural breaks
the APT was halved, the farmers’ between regimes such as high-or low-price phases as done so
A
by Rajcaniova and Pokrivcak (2013). It was also checked for
a
asymmetries in the short term adjustments. Since none of
loss of income would be reduced significantly.
d
these have altered the results they were omitted from the
i

r
tables to save space. They are available upon request.
Another issue that has been touched upon only briefly is Intertemporal marketing margin
the behaviour between
o
manipulation 1669
l

Jambinese rubber processors. It would be interesting to BFJ

know if there is a rather random
f
119,8
o

section of stakeholders who apply price changes first, or

one clear Stackelberg leader
y

i 1670
s

determining the price. With this sort of game-theoretical

approach one would be able to get a
r

clearer picture of the roles of different stakeholders

within the group, and the functioning of
i )

n T

U P

it as a whole. This calls for research at a more (

disaggregated level.
7

0
y
2
b
r

e
per cent. All from World Bank Database, 2015, data set
b
Inflation, consumer prices (annual per cent).
m

N References
8

1
Ahmadi-Esfahani, F. (2009), “Estimation of trade elasticities in
the presence of trade barriers, multinationals and imperfect
competition”, Australian Journal of Agricultural and Resource
5

Economics, Vol. 53 No. 3, pp. 385-392.

:

Arifin, B. (2005), “Supply-chain of natural rubber in Indonesia”,

1

Jurnal Manajemen Agribisnis, Vol. 2 No. 1, pp. 1-16.

t

a Bresnahan, T.F. (1982), “The oligopoly solution concept is

identified”, Economics Letters, Vol. 10 No. 1, pp. 87-92.
d

l
Burnham, K. and Anderson, D. (2002), Model Selection and
F Multimodel Inference: A Practical Information-Theoretic
f Approach, Springer, Heidelberg.
o

y
Eilers, P.H.C. and Marx, B.D. (1996), “Flexible smoothing with
t

i
B-splines and penalties”, Statistical Science, Vol. 11 No. 2, pp.
s

r
89-102.
e

v
Engle, R.F. and Granger, C.W.J. (1987), “Co-integration and
i

n
error correction: representation, estimation, and testing”,
U
Econometrica, Vol. 55 No. 2, pp. 251-276.
y
Escribano, A. (2004), “Nonlinear error correction: the case of
money demand in the United Kingdom (1878-2000)”,
b

e
Macroeconomic Dynamics, Vol. 8, pp. 76-116.
Euler, M., Krishna, V., Fathoni, Z., Hermanto, S., Schwarze, S.
d

l
and Qaim, M. (2012), “Ecological and socioeconomic functions
n
of tropical lowland rainforest transformation systems (Sumatra,
w

o
Indonesia), household survey 2012”, Georg-August University
D of Goettingen, Bogor Agricultural Uni, Göttingen and Jambi
7. It was also experimented with a smooth-transition type of and Bogor.
model as employed by Hassouneh et al. (2012). The results
Fathoni, Z. (2009), “Evaluation of market system and market
did not show statistical significance, but can be made available
integration for rubber cultivation in Jambi Province –
on demand.
Indonesia”, master’s thesis, Bogor Agricultural Institute, Bogor.
8. We can make this simplification of Equation (3) since the Gonzalo, J. (1994), “Five alternative methods of estimating
short-run dynamics are not asymmetric.
long-run equilibrium relationships”, Journal of Econometrics,
9. The adjustment of pSell to pBuy is close to zero, since pSell Vol. 60 No. 1, pp. 203-233.
was shown above to be clearly the leading price, and not Hansen, B.E. and Seo, B. (2002), “Testing for two-regime
reacting to pBuy. threshold cointegration in vector error-correction models”,
10. The full results of the ARVECM can be made available Journal of Econometrics, Long Memory and Nonlinear Time
upon request. Series, Vol. 110 No. 2, pp. 293-318.

11. These calculations were carried out with the software R Hassouneh, I., Radwan, A., Serra, T. and Gil, J.M. (2012),
“Food scare crises and developing countries: the impact of
3.0.1 and version 1.7-22 of R package MGCV. 12. Available on avian influenza on vertical price transmission in the Egyptian
request. poultry sector”, Food Policy, Vol. 37 No. 3, pp. 264-274.
13. For model tests see below. The results of the estimation Hassouneh, I., Serra, T. and Gil, J.M. (2010), “Price
that assumes the other threshold can be made available on transmission in the Spanish Bovine sector: the BSE effect”,
demand. We also executed a two-dimensional grid-search and Agricultural Economics, Vol. 41 No. 1, pp. 33-42.
estimated a three-threshold model, whose results can also be Ihle, R., Brümmer, B. and Thompson, S.R. (2012), “Structural
made available on demand. change in European calf markets: decoupling and the blue
14. Inflation in 2009: 4.8 per cent, 2010: 5.1 per cent, 2011: 5.4 tongue disease”, European Review of Agricultural Economics,
Vol. 39 No. 1, pp. 157-179. Ivanov, V. and Kilian, L. (2005), “A Perloff, J.M., Karp, L.S. and Golan, A. (2007), Estimating
practitioner’s guide to lag order selection for VAR impulse Market Power and Strategies, Cambridge
response analysis”, Studies in Nonlinear Dynamics &
d

Econometrics, Vol. 9 No. 1, pp. 1-36. r

University Press, New York, NY.

Johansen, S. (1998), Likelihood-Based Inference in o

Cointegrated Vector Autoregressive Models, Oxford University F

Press, Oxford.
Rajcaniova, M. and Pokrivcak, J. (2013), “Asymmetry in price
Kinnucan, H.W. and Forker, O.D. (1987), “Asymmetry in transmission mechanism: the case of
farm-retail price transmission for major dairy products”, f

American Journal of Agricultural Economics, Vol. 69 No. 2, pp.

285-292. Slovak potato market”, Review of Agricultural and Applied
Kopp, T. and Brümmer, B. (2015), “Moving rubber to a better Economics, Vol. 16 No. 2, pp. 16-23.
y

place – and extracting rents from credit constrained farmers

t

along the way”, Efforts Discussion Paper No. 9, CRC990 s

Discussion Paper Series, Göttingen. Regional Account and Statistical Analysis Division (2012),
Krivobokova, T., Kneib, T. and Claeskens, G. (2010), Jambi in Figures 2011, BPS Statistics of
e

“Simultaneous confidence bands for penalized spline v

estimators”, Journal of the American Statistical Association,

i

Jambi Province, Jambi.

Vol. 105 No. 490, pp. 852-863. n

Lloyd, T.A., McCorriston, S., Morgan, C.W. and Rayner, A.J.

(2006), “Food scares, market power and Serra, T., Gil, J.M. and Goodwin, B.K. (2006), “Local
polynomial fitting and spatial price relationships:
)

price transmission: the UK BSE crisis”, European Review of y

Agricultural Economics, Vol. 33 b

price transmission in EU pork markets”, European Review of

T

(
Agricultural Economics, Vol. 33
No. 2, pp. 119-147. d

1
No. 3, pp. 415-436.
d

McCorriston, S., Morgan, C.W. and Rayner, A.J. (2001), “Price

a

transmission: the interaction between market l

0
Sujarwo, R.M., Kopp, T., Nurmalina, R., Asmarantak, R.W. and
Brümmer, B. (2014), “Choice of
2

r n

power and returns to scale”, European Review of Agricultural w

Economics, Vol. 28 No. 2, pp. 143-159. marketing channels by rubber small traders in the Jambi
province, Indonesia”, Tropentag 2014:
e

b
o

Meyer, J. and von Cramon-Taubadel, S. (2004), “Asymmetric D

price transmission: a survey”, Journal of Conference Proceedings, Czech Republic, Prague.

von Cramon-Taubadel, S. (1998), “Estimating asymmetric price
m

v
transmission with the error correction representation: an
Agricultural Economics, Vol. 55 No. 3, pp. 581-611.
o
application to the German pork market”, European Review of
N Agricultural Economics, Vol. 25 No. 1, pp. 1-18.
Mundlak, Y. and Larson, D.F. (1992), “On the transmission of Weldegebriel, H.T. (2004), “Imperfect price transmission: is
world agricultural prices”, World Bank market power really to blame?”, Journal of Agricultural
Economics, Vol. 55 No. 1, pp. 101-114.
8

Economic Review, Vol. 6 No. 3, pp. 399-422. Wood, S.N. (2003), “Thin plate regression splines”, Journal of
5

3
the Royal Statistical Society: Series B (Statistical
:
Methodology), Vol. 65 No. 1, pp. 95-114.
Peramune, M.R. and Budiman, A. (2007), “A value chain
assessment of the rubber industry in Corresponding author
Thomas Kopp can be contacted at:
8

t thomas.kopp@agr.uni-goettingen.de
Indonesia”, project report, US Agency for International
Development, Jakarta.
A

a
For instructions on how to order reprints of this article, please
visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details:
permissions@emeraldinsight.com
Intertemporal marketing margin
manipulation 1671