The extent of price transmission between international and regional cereal markets in Africa

(1)

Faculty of Economics and Business

The Extent of Price Transmission between

International and Regional Cereal Markets in

Africa

MSc. thesis International Economics

Author: Supervisor:

Henriëtte van der Kwast dr. Boe Thio

student number: 6102735 Second reader:

(2)

Abstract

This thesis investigates the extent of cereal price transmission between international and regional African markets by estimating a VECM model. The data consist of monthly cereal prices from 57 regions in thirteen countries during the time span January 2001- December 2011. The findings indicate that international price transmission during this period was limited, occurring only in 23% of the included price series and mainly for import rice prices. The estimated long-run transmission elasticity was 0.63, which is lower than the average of 0.75 reported in the literature, but for Africa a lower transmission elasticity is consistent with the literature. Regional price transmission within a country occurred more often and to

a larger extent: 47% of the regional price series had a long-run relation and the average transmission elasticity was 0.99. These findings imply that policies aimed at evading future

food price increases should focus at the regional rather than the international causes. The differences between different regions and countries indicate that country-specific research

(3)

Acronyms

ADF - Augmented Dickey Fuller

ADRL – Autoregressive Distributed Lag Model FAO- Food and Agricultural Organization OLS –Ordinary Least Squares

(V)ECM – (Vector) Error Correction Model

List of Figures

Figure 1: International commodity prices 1

Figure 2: Share of macronutrients to total dietary energy supplies 3 Figure 3: Conceptual framework for assessing price transmission and market integration 16

List of Tables

Table 1: Descriptive statistics 13

Table 2: Evidence for cointegration 17

Table 3: Estimated parameters for the ECM model 19

Table 4: Cointegration between regional prices 23

Table 5: Estimated parameters for rice prices in Mozambique 27 Table 6: Cointegration coefficients for rice in Cameroon 29 Table 7: Adjustment coefficients for rice in Cameroon 29 Table 8: Cointegration coefficients for maize in Niger 30 Table 9: Adjustment coefficients for maize in Niger 31 Table 10: Cointegration coefficients for rice in Niger 31 Table 11: Adjustment coefficients for rice in Niger 32

(4)

Table 12: Cointegration coefficients for rice in Senegal 33 Table 13: Adjustment coefficients for rice in Senegal 34 Table 14: Cointegration coefficients for rice in Somalia 34 Table 15: Adjustment coefficients for rice in Somalia 35 Table 16: Cointegration coefficients for sorghum prices in Togo 36 Table 17: Adjustment coefficient for sorghum prices in Togo 36

(5)

Introduction

The large increase of international food prices since 2002 and the major price spikes in 2006 and 2007 led to food crises in many developing countries. Figure 1 below shows the development of five of the main international food commodity prices in the US between 2000 and 2014 and illustrates the food price increase since 2002.

Figure 1: International commodity prices

Note: USDA = United States Department of Agriculture Source: Food and Agricultural Organization (FAO), (http://www.fao.org/giews/pricetool/)

The impact of these rising food prices is severe, not only in the short run, but also in the long run. Estimates from the World Bank suggest that the food crisis in 2007-2008 resulted in an additional 130 million people that fell below the poverty line, which is set at 1.25 US dollar a day (Lee and Ndulo 2008: 8). Moreover, rising food prices have a long-term impact, since they affect not only the quantity and quality of food items consumed, but also result in reduced expenditures on education and health care (ibid.: 9). This is not surprising considering that most poor people spend 75% of their income on staple foods (Cranfield et

(7)

al. 2007). While it is sometimes assumed that price increases can also reduce poverty rates, by increasing the incomes of poor farmers, most research has shown that on average the effect of higher food prices is negative (Ivanic and Martin 2008, Zezza et al. 2008, Wodon and Zaman 2008, Wodon et al. 2008).

The large impacts of food price increases explains why, in the aftermath of the food crisis, much research has been done towards the reasons for these sudden price spikes. Factors that have been identified to have contributed to the price rises are, among others, higher oil prices, increased demand for biofuels and a weak dollar (Minot 2011, Abbot and Borot de Battisti 2011). Subsequent research focused on appropriate policy responses to prevent future food crises. Logically, these policy responses were based on the broad set of identified factors to have contributed to the crisis. It is important to take into account however, that for consumers it is not the world price that is relevant, but the domestic price. Therefore it is important to know to what extent the increase in world prices was transmitted to local markets. While much research has been done on the extent of price transmission before the crisis (Conforti 2004), there has not been a lot of research on this effect during and in the immediate aftermath of the last food crisis of 2007/2008. In fact, I know of only four studies that have investigated this effect (Abbot&de Battisti 2011, Esposti &Listorti 2013, Greb et al. 2012 and Minot 2012). In addition, there are very few articles that have looked at regional differences in price transmission, in spite of the large regional disparities that are often present in (large) African countries (Cudjoe et al. 2010).

Therefore, this thesis aims to further investigate the following two questions: 1)To what extent was the rise in world food prices transmitted to African domestic regional cereal prices during and after the recent food crisis? and 2) What is the extent of price transmission between regional markets within the same country? The second question is important because some analysts have suggested that regional and local factors are more important in determining food prices than international factors (Benson et al. 2008). Together, the answers to these questions can say something about the appropriate focus of policy measures to ensure food security in African countries.

The thesis focuses on cereal prices, since cereals make up an important part of the diets of the poor in developing countries, almost 70%, as can be seen in Figure 2below.

(8)

Figure 2: Share of macronutrients to total dietary energy supplies

Source: FAO, the state of food security in the world 2012

Furthermore, the choice to focus on African countries is made because it is estimated by the World Food Programme that 22 of the 30 countries that are most vulnerable to food price increases, are located in Africa (Sanogo 2009). This thesis builds upon the IFPRI discussion paper ‘Transmission of World Food Price Changes to Markets in Sub-Saharan Africa’ (Minot 2011), but investigates a different and wider range of countries and also investigates the extent of regional price transmission. In addition, this thesis can be related to the broader economic literature on market integration and the law of one price.

To analyse the extent of price transmission, an error correction model (ECM ) will be employed. This model is employed by Minot, as well as in most other literature on price transmission before the food crisis. The advantage of this model is that it separates the short-run dynamics from the long-run equilibrium relationship between domestic and world prices (Greb et al. 2012), or between different regional prices.

The remainder of the thesis is structured as follows. The first chapter summarizes the economic theory on the law of one price and provides an overview of the main causes for imperfect price transmission. The second chapter provides a summary of the literature on

(9)

price transmission in agricultural markets and presents its main findings. Chapter three describes the data and methodology, while chapter four presents the results. Chapter five discusses the findings in chapter four and relates these to the findings in the literature and to the two research questions. The final chapter concludes.

(10)

1. Theoretical Framework

1.1 Price transmission within the economic theory

The concept of price transmission, sometimes also called price linkages (Baffes and Ajwad 2001), refers to the extent to which price changes for a specific good in one market affect price changes in another market (Minot 2011). Price transmission can take place vertically or spatially (horizontally), where vertical price transmission refers to transmission of prices across different points in the supply chain and spatial price transmission refers to transmission of prices across different locations. Usually price transmission is measured in terms of the transmission elasticity, which is defined as the percentage change in the price in one market, given a one percent change in the price in another market (ibid.).

The concept of price transmission is thereby strongly related to the concept of market integration and the law of one price. Full market integration would imply complete price transmission and thus one price for the same commodity in different markets. Therefore, the extent of price transmission is often used as an indicator of efficient markets (Conforti 2004, Fackler and Goodwin 2001) and of the existence of the law of one price (Baffes 1991).

An important condition for price transmission to take place, is the possibility of spatial arbitrage, which implies that trade should be possible, and therefore arbitrage will ensure that the prices of a homogenous good at two locations will differ by, at most, the transport and transaction costs between these two locations. (Fackler and Goodwin, 2001: 977). This is equivalent with the notion that a good should be both tradable and indeed traded in order for price transmission to take place. Fackler and Goodwin (2001) point out however, that this specific condition in which the maximum price difference is equal to the transaction costs, is an equilibrium concept and that temporary deviations are possible (ibid.).

1.2 Explanatory factors for imperfect price transmission

If full price transmission is an equilibrium concept, then there should be factors that can explain deviations from this equilibrium. According to Conforti (2004), the literature on price transmission has identified six types of factors that affect the extent of price

(11)

transmission: transport and transaction costs, market power, increasing returns to scale, product homogeneity and differentiation, exchange rates and border and domestic policies. This section will briefly discuss these factors.

Firstly, transport and transaction costs, relate to the costs involved in trading between different markets. Whereas transport costs are quite straight-forward, transaction costs are more complex. Dahlman distinguishes three different types of transaction costs: search and information costs, bargaining and decision costs and policing and enforcement costs (1979: 148). Nevertheless, these three different types of costs are all due to the same problem: imperfect information about the transactions possible, the appropriate price and the reliability of the trading partner (ibid.).

Secondly, market power, refers to the extent to which agents are price takers or price makers, often depending on the degree of concentration in the market. If for instance, there is only one firm in a specific country importing from the world market, it is possible that this firm chooses to pass on price increases from world prices but includes world price decreases in its profit margin (Meyer and von Cramon-Taubadell 2004, Conforti 2004).

Thirdly, increasing returns to scale in production are more important for vertical price transmission than for spatial price transmission. Nevertheless, in spatial price transmission they may play a role in the sense that for the import of large quantities transport may be relatively cheaper than for small quantities.

Fourth, product homogeneity and differentiation, can affect the extent of price transmission because the degree of substitutability in consumption between similar goods in different countries is likely to have an effect on market integration. For instance, if sorghum is considered a perfect substitute for wheat, then an increase in the world price for wheat could result in a local price increase of sorghum, because demand for sorghum increases as a result of the price increase of wheat.

Fifth, exchange rates, can affect the extent of price transmission because not all exchange rate changes need to be transferred to imported goods prices. This is often referred to as the ‘pass-through’ effect and this has been studied to investigate the ability of firms to charge different prices in different markets (pricing-to-market behaviour). In addition, official exchange rates could be under-or overvalued, resulting in wrong conversion rates which do not accurately reflect the value of domestic prices

(12)

(Mundlak&Larsson 1992).

Sixth, border and domestic policies, can refer to trade policies, which have a direct effect on the extent of price transmission, or to domestic policies that are affecting prices in general. Trade policies, both tariff as well as non-tariff barriers are likely to have a strong impact on price transmission. Fixed tariffs and ad valorem taxes can be considered to have the same impact on price transmission as transaction costs. An example of a domestic policy that is likely to affect prices is subsidizing a particular sector.

The majority of the articles on price transmission do not explicitly take these factors into account. Most of them start from the standard perfect competitive framework and then attribute deviations from full price transmission to one of the specified factors (Conforti 2004). This is likely due to the fact that many of the described factors are difficult to measure. One of the factors that has been taken into account by some scholars is the factor border and domestic policies, which is often related to the extent of trade liberalization and regime change in (mainly) developing countries (Baffes & Gardner 2003). Given the broad scope of this thesis, it is not feasible to investigate the influence of each factor on the extent of price transmission in the specific regions. For country studies and policy purposes this would be advisable however. Furthermore, in this study the focus is on cereal prices denominated in US dollars, which is likely to make the influence of product differentiation and exchange rate changes insignificant.

(13)

2. Literature Review

Since the concept of perfect competition and the notion of the law of one price play such a crucial role in most economic models, there has been a lot of research on this subject. This chapter will describe the main findings of the articles that are specifically related to the price transmission mechanism.

2.1 Simple OLS regression models

The early studies on price transmission often used simple OLS regression models in which simultaneous prices of different markets were regressed on each other (Minot 2011, Fackler &Goodwin 2001). Mundlak and Larson (1992) for example, estimated an equation in which domestic prices were regressed on world prices, taking into account the exchange rate and tax policy, where the latter refers to tariff and non-tariff barriers . This results in the following regression: 𝑝𝑖𝑡 = 𝛼 + 𝛽𝑝𝑖𝑡∗ + 𝛾𝑒𝑡+ 𝜖𝑖𝑡 where 𝑝𝑖𝑡 and 𝑝𝑖𝑡∗ refer to the domestic and world price price for good i and time t respectively, 𝑒𝑡 refers to the exchange rate and 𝛼, the intercept, is assumed to contain the effect of the tax policy. This regression is computed for 58 countries over a ten-year period (1968-1978), for 60 food commodity products. They find a transmission elasticity between .74 and 1.24 with a median of 0.95, which is surprisingly large: it implies that a one percent increase in the world price would lead to a median increase of 0.95% in the domestic price. It is important to note however, that the authors do not estimate transmission elasticities separately for different commodities, but compute an average cross-commodity price. This seems a bit odd, considering the likelihood of differences in transport and transaction costs between different goods, as well as different degrees of substitutability.

A more recent study that used the same type of model is the study by Abbot and Borot de Battisti (2011). They assess the extent of price transmission in eleven African countries over the period June 2006-June 2008 and June 2008-December 2008 for four types of agricultural commodities: rice, maize, sorghum/millet and wheat. They find transmission elasticities varying between .03 (rice in Mali) and 2.25 (maize in Malawi). Their estimates clearly exhibit a wider range than the estimates of Mundlak and Larsson, because the latter calculated a cross-commodity price instead of individual commodity prices. This suggests that the transmission elasticity can differ substantially between different

(14)

commodities and therefore implies that the method employed by Mundlak and Larsson is inappropriate. Furthermore, Abbot and Borot de Battisti conclude that import-dependent countries have higher transmission rates, a finding that is consistent with earlier studies (ibid. p. i31). This is due to the fact that import-dependent countries cannot easily switch to domestically produced products if import prices rise. Nevertheless, some authors have argued that import-dependent countries have a larger incentive to insulate their domestic market from world price shocks (Greb et al. 2012: p.18), which would result in a lower transmission elasticity.

These simple regression models have been criticized for two main reasons. First, the models described above, assume that markets respond instantaneously to changes in other markets. Since it is more likely that domestic prices adjust slowly and respond with a lag to changes in world prices (Minot 2011), a dynamic model in which lagged world prices are included is considered to be more realistic (Ravallion 1986).

Secondly, standard regression models assume that the mean and the variance of variables are constant over time, or stationary. In time-series regression however, prices and other variables are often non-stationary: they tend to follow a random walk rather than return to a mean value. Although the estimated coefficients in a standard regression remain unbiased when variables are non-stationary, the error term is no longer normally distributed which implies that standard significance tests are no longer valid (Minot 2011, Baffes and Ajwad 2001). To solve the issue of non-stationarity and dynamic effects, most recent articles employed an error correction model (ECM) instead of standard regression. The following section will describe the main findings of authors who employed such a model.

2.2 Results of articles that employed an ECM

Greb et al. (2012) provide an extensive analysis of price transmission estimates for cereal prices from international markets to the domestic markets of developing countries. They consider the estimates from 31 published papers and provide an analysis of their own estimates based on an FAO dataset, using monthly data between 1995 and 2011, although some price series start later. They conclude that most of the published papers consider a single cereal product and one or a few markets and typically use monthly price series. On average, all these estimates point to price transmission rates of roughly 0.75 in the long run

(15)

and short-run adjustment rates of 0.09-0.11. They also try to explain the variation in the transmission elasticities in a meta-regression analysis. For this analysis, they regress the estimated price transmission rate on variables that are likely to influence the extent of price transmission, including also commodity and country-specific factors. Examples of included variables are a variable that indicates the trade openness, the logistics performance and whether a country is landlocked. However, they fail to find a clear conclusion based on this meta-regression, in fact some findings are counterintuitive. For example, the results of the meta-regression suggest that improvements in logistics lead to smaller adjustment coefficients and hence slower price transmission. Other authors did find some potential explanatory factors for these differences. Conforti (2004) for instance, who also employed an ECM regression, argues that the price transmission to African markets is relatively lower than to other markets, mainly because of physical barriers and limited market size. Minot (2011) finds that within Africa, transmission elasticities for rice are larger than for maize, because most African countries are dependent on imports for their rice consumption but close to self-sufficient in maize. The latter argument is in line with the findings by Abbot and Borot de Battisti (2011).

Greb et al. also check whether there is a difference between price transmission before and after July 2007, by including a structural break in their ECM. A structural break allows for a change in the estimated parameters before and after a specific event. They find that the long-run transmission rates for maize have decreased after the price spikes, while the long-run transmission rates for rice and wheat have increased. However, they fail to provide an explanation for this difference. Baffes and Gardner (2003) consider the existence of a structural break in the presence of policy reforms. They examine whether the implementation of policy reforms in developing countries led to an increase in the degree of price transmission. Although they find an increase in the extent of price transmission for some country/commodity pairs, overall the existence of a structural break is rejected. Therefore, the effect of policy reforms is not found to have a significant change in the extent of price transmission. This could be due to the fact that price transmission can also occur in the presence of government intervention, as argued by Conforti (2004). If Conforti is right, this would imply that a policy reform need not have a large effect on the extent of price transmission.

In addition, Greb et al. find evidence of asymmetric price transmission: while world 10

(16)

price decreases are transmitted to domestic prices in 50% of the cases, increases in world prices are transmitted in about 60% of the cases. This last aspect is also found in other studies, such as the one by Morisset (1998), who argues that asymmetric price transmission is the main reason for increasing spreads between international and domestic price levels.

Finally, the paper by Minot (2011), which is the basis for my research, investigates the transmission of international food price increases to African domestic food prices during the food price spikes. This is done by comparing the African domestic price increases with the international price increases during 2007-2008 and by examining the existence of a long-term relationship using an ECM. This results in the puzzling finding that although African domestic prices did increase significantly in response to the international price spikes, a statistically significant relationship between local and international markets could only be established in 6 of the 62 cases. A possible explanation provided by Minot is that the simultaneous increase in oil prices had a large effect on transport costs which can make up a large part of import costs. This could mean that there is a relationship between oil prices and domestic food prices, rather than between international and domestic food prices. In addition, not all the price series in his ECM did include the years 2007-2008. Therefore he recommends that further research is necessary.

To sum up, the average long-run international price transmission rate in the literature is found to be 0.75, but the other articles described show that there are large differences across commodities and countries. Furthermore, considering the evidence of asymmetric price transmission, it seems likely that the price increases of the food price spikes in 2007-2008 resulted in a larger price transmission rate than the long-run average. However, the paper by Minot could not find such a clear relationship between world price increases and domestic price increases.

(17)

3. Data and methodology

This section starts with a description and summary of the data. Subsequently, the methodology will be introduced, describing the workings and interpretation of the variables in an ECM.

3.1 Data description

All data are monthly retail price data of cereal commodities (wheat, rice, sorghum and maize), expressed in USD/tonne. The world price data are obtained from the website of the FAO1_{and are defined as follows:}

World rice price = price of Thailand 100%B 2nd_grade World wheat price = US no. 2 hard red winter World sorghum price= US no. 2 yellow

World maize price = US no 2 yellow

Although there is no official definition of what constitutes ‘the’ world price (Greb et al. 2012) most of the literature uses the commodity prices published by the FAO. The FAO publishes the prices of the commodities that have the largest export volume.

Local prices are obtained from the website opendataforafrica.org and are available for 57 regions in thirteen countries: Benin, Cameroon, Chad, Congo, Malawi, Mauritania, Mozambique, Niger, Senegal, Somalia, Togo, Zambia and Zimbabwe; the regions can be found in appendix A1. The total time period under investigation is January 2001 until October 2011, although the time period and the commodities available differ across regions. However, since I estimated separate regressions for each region and commodity this should not be a problem. The minimum time span in the data set is 50 months for Bunia, a region in Congo, which is from July 2007 until August 2011 and this is also the common period for all regions under investigation. An overview of the time period available for each country-commodity pair can also be found in appendix A1. Price series that had more than two missing values in a row were excluded, and individual missing values were filled in by taking the average of the value before and after the missing value. For a more intuitive

1_{http://www.fao.org/economic/est/statistical-data/en/}

12

(18)

interpretation of the results, all price levels were converted in natural logarithms, which means that the coefficients can be interpreted as transmission elasticities.

The variable yearmonth describes the year and month for which the price was taken, the variable WorldpMaize describes the world price of maize, whereas pMaize describes the local price of maize. Similar notations were used for sorghum, wheat and rice. For rice, some countries distinguished between prices of locally produced rice (pRicelocal) and prices of imported rice (pRiceimported), if this was the case then separate regressions were run for both types of rice.

Table 1 below provides an overview of the number of observations, the mean and the standard deviation of all prices. The table demonstrates that the average local prices are more than two times as high as the average world price, with the exception of rice. This suggests that local prices have increased more than proportionally with world prices. In addition, the low number of observations for wheat and local rice prices is likely to present some problems.

Table 1: Descriptive statistics

Variable Observations Mean Standard deviation

WorldpMaize 131 150.64 63.65 pMaize WorldpRice pRice pRicelocal pRiceimport WorldpSorghum pSorghum WorldpWheat pWheat 1857 131 1451 328 3347 131 1914 131 283 365.13 376.38 655.21 851.9 658 151.86 327.86 215.67 908.48 135 183.7 234 331 259.3 58.6 129.57 79.8 519.84

Source: own calculations

3.2 Methodology

As was briefly mentioned in the literature review, recognition regarding the non-stationarity of prices led to a new model to estimate price transmission: the ECM. This section discusses workings of this model and describes the equations to be estimated for this paper.

(19)

Stationarity refers to the concept that the probability distribution of a time series is constant over time and the time series is said to be integrated of order zero. If time-series are stationary, an Autoregressive Distributed Lag Model (ARDL) can be estimated, in which a time series is regressed on lagged values of the dependent variable itself and of the independent variables. In many cases however, prices are not stationary and follow a random walk (with or without drift) instead. In this case, the errors of a simple regression model are no longer normally distributed, and standard significance tests are no longer valid. If price series are cointegrated however; that is if there exists a linear relationship between two non-stationary series which is stationary, then the first difference of the series is stationary and can be estimated with an ECM (Greb et al. 2012). One of the biggest advantages of an ECM is that the short-run dynamics are separated from the long-run equilibrium relationship.

3.2.1 An ECM for price transmission from world to domestic prices

Following Greb et al. (ibid. p.7) , a simple ECM with one lag, that captures the extent of price transmission from world to domestic prices is the following2_:

∆𝑝𝑡𝑑 = 𝜑1+ 𝛼1�𝑝𝑡−1𝑑 − 𝛽1𝑝𝑡−1𝑤 � + 𝛿1∆𝑝𝑡−1𝑤 + 𝜌1∆𝑝𝑡−1𝑑 + 𝜀1𝑡 (1)

Where 𝑝𝑡𝑑 is the log of the domestic price at time t, 𝑝𝑡𝑤 is the log of the world price at time t and 𝜑, 𝛼, 𝛽, 𝛿 𝑎𝑛𝑑 𝜌 need to be estimated.3

To allow for a trend in the cointegrating relationship, a constant term γ will be added to equation 1, resulting in equation 2 that will be estimated in the empirical analysis. This constant term can be interpreted as transport and transaction costs that are non-proportional to prices, and it is thereby influencing the relative price transmission (Sharma 2003:143).

∆𝑝𝑡𝑑 = 𝜑1+ 𝛼1�𝑝𝑡−1𝑑 − 𝛽1𝑝𝑡−1𝑤 + 𝛾� + 𝛿1∆𝑝𝑡−1𝑤 + 𝜌1∆𝑝𝑡−1𝑑 + 𝜀1𝑡 (2)

In this model, �𝑝𝑡−1𝑑 − 𝛽1𝑝𝑡−1𝑤 + 𝛾� is the error correction term: it represents the deviation of prices in the previous period from the long-term, or cointegrating relationship between

2

Greb et al. number the parameters (i.e. 𝜑1) because they also describe the equation for world prices.

3_{Logs of prices are used so that the coefficients can be interpreted as elasticities}

14

(20)

𝑝𝑡𝑑 and 𝑝𝑡𝑤. The most important parameters in this model are 𝛽1 and 𝛼1, where 𝛽1describes how domestic prices respond to changes in world prices in the long run, it is the long-run elasticity of domestic prices with respect to world prices.

It is important here to make a distinction between absolute price transmission and relative price transmission. Absolute perfect price transmission occurs when a price increase of 10 US dollar in the international price, results in a price increase of 10 US dollar in a domestic regional price as well. However, assuming that the international price starts at a level of 100 US dollar and the domestic price starts at a level of 200 US dollar, relative perfect price transmission occurs when both prices increase by the same percentage of their initial level. Hence, an increase of 10 US dollar in the international price, would need to lead to an increase of 20 US dollar in the domestic price in order to speak of perfect relative price transmission and thus to result in a transmission elasticity of exactly 1, represented by β=1. Moreover, the larger the constant term 𝛾 is, or the higher the transport and transaction costs, the lower is the elasticity from unity for an import good, even though in absolute terms full price transmission takes place. This is due to the fact that a large constant requires a more than proportional absolute price increase in order to realize a complete relative transmission. As a result, a larger constant also leads to a larger difference between absolute and relative price transmission.4

Furthermore, 𝛼1 indicates the speed of adjustment of prices towards their long-run equilibrium. This implies that the larger is 𝛼1 , the faster the domestic price will return to the value that is suggested by the long term relationship with the world price. The coefficient 𝛿1 describes how the domestic price responds to a change in the world price in the previous period, it the therefore the short-run elasticity of the domestic price with respect to the world price. Finally, the coefficient 𝜌1 describes how the domestic price responds to a change in the domestic price in the previous period.

To determine whether the ECM is the appropriate model for the data, the analysis starts by testing for stationarity and cointegration. The first step is to test whether both variables have a unit root, that is to test whether they are stationary when they are differenced once. For this purpose, an Augmented Dickey Fuller (ADF) test will be used. If the two variables

4_{For an extensive description and explanation of this mechanism, see Sharma (2003)}

15

(21)

are integrated of a different order, then there is no cointegration and no long term relationship between the two variables. If however a unit root is found for both variables, then the next step is to perform a test for cointegration. The test that will be used is the Engle-Granger Augmented Dickey Fuller test and Johansen test. If the two variables are found to be cointegrated, an ECM of the form above can be estimated. If there is no evidence for cointegration, an ARDL model needs to be estimated instead. These steps are summarized in Figure 3 below.

Figure 3: Conceptual framework for assessing price transmission and market integration

Source: Figure 1 from Greb et al. (2012)

The ECM will be estimated for a number of lags, and the appropriate number of lags will be determined using the Akaike information criterion (AIC).

(22)

4. Results

4.1. Transmission from world prices to local prices

As mentioned in section 3.1, an ADF test was used to test whether the price variables contained a unit root, that is whether they were non-stationary. First the AIC was used to determine the appropriate number of lags. While the selected number of lags for all rice prices was one, the number of lags for maize, the number of lags for wheat and sorghum was all larger than one, and most were three or four. A unit root was found for most variables, which is consistent with the overall findings in the literature.

Subsequently, tests for cointegration were performed using the two step method proposed by Engle and Granger(1987). This method implies that local prices were first regressed on world prices and a constant. Thereafter, the resulting residuals were tested for a unit root, using an ADF test. If a unit root is found, then the residuals are non-stationary and the null hypothesis of no cointegration is not rejected. However, if no unit root is found, then the residuals are stationary and there is evidence for cointegration between the two variables of interest. Table 2 below presents the main results of this procedure, complete results of the selected lag length, unit root tests and cointegration tests can be found in appendix A2.

Table 2: Evidence for cointegration

Commodity price Observations Number of

cointegrated series Percentage

pRice 13 7 54% pRiceimp 32 12 37.5% pRicelocal 4 1 25% pMaize 20 1 5% pWheat 4 0 0% pSorghum 17 0 0% Total 90 21 23.3%

Source: Own calculations

Table 2 shows that only rice prices reveal clear evidence for cointegration, that is have a long-run relationship with world rice prices. More specifically, rice prices in Cameroon (regions: Bamenda, Douala and Garou), Benin (region: Natitingou), Chad (regions:

(23)

Moussoro and N’Djamena), Mozambique (regions: Angonia, Gorongosa, Manica and Nampula), Niger (regions: Zinder, Dosso, Niarney, Maradi and Agadez ), Somalia (regions: Buale and Hudur) and Togo (regions: Kara and Cinkasse) were found to be cointegrated with international rice prices. In addition, cointegration was found for local rice and maize prices in Bunia (Congo), which is rather surprising, given that no cointegration was found for these commodity prices in any other region. It is possible however that the presence of the UN mission MONUSCO in Bunia had an influence on the prices there. Therefore, the findings for Bunia should be considered as exceptional.

Next, an ECM was estimated for all regional prices for which cointegration with world prices was found. Since Stata uses the Johansen method to estimate a VECM, the Johansen test was performed before the estimation of the VECM to confirm the existence of cointegration. In some cases this led to a different conclusion than based on the Engle Granger test. The Engle Granger test is often considered more robust if it is easy to identify the dependent variable(Gonzalo and Lee 1997). In this case it is reasonable to assume that local African prices do not influence the world price, and hence that the dependent variable is the local African price. Therefore, an ECM was still measured, even when the Johansen test did not confirm the existence of cointegration. Table 3 summarizes the estimated parameters and also indicates whether the Johansen test resulted in a different finding than the Engle Granger test.

(24)

Table 3: Estimated parameters for the ECM model:

Country Region Cereal Adjustment

coefficent Α Cointegration coefficient β Constant in ECM Γ Short run elasticity w.r.t. world price δ Short run elasticity w.r.t. regional price ρ Cameroon Cameroon Cameroon Chad Chad Somalia Somalia Mozambique Mozambique Mozambique Mozambique Niger Niger Niger Niger Niger Togo Togo Benin Congo Congo Bamenda Douala Garoua Moussoro N’Djamena Buale Hudur Angonia# Gorongosa Manica Nampula Zinder# Niarney Dosso# Maradi# Agadez# Kara Cinkasse Natitingou Bunia# Bunia# Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Rice Local Rice Maize -0.66* -0.08 -0.24* -0.29* -0.52* -0.54* -0.3* -0.17* -0.30* -0.48* -0.38* -0.04 -0.13* -0.09* -0.07 -0.02 -0.20* -0.23* -0.11* -0.10 -0.31* 0.58* 0.51* 0.50* 0.58* 0.36* 0.59* 0.63* 0.71* 0.85* 0.72* 0.86* 0.69* 0.66* 0.66* 0.78* 0.69* 0.54* 0.62* 0.61* 0.69* 0.49 3 3.42 3.67 3.43 4.6 3.07 2.69 2.4 1.36 2.27 1.24 2.49 2.6 2.64 2.01 2.51 3.36 2.87 3.05 2.4 3.6 0.19 0.11 0.24* -0.08 0.23 0.03 0.08 -0.07 -0.01 0.08 -0.21 -0.07 0.06 -0.02 -0.07 -0.02 0.24 0.15 0.17* -0.03 -0.40 0.20 0.19 -0.05 -0.22 -0.08 -0.05 0.25* -0.02 -0.07 -0.06 0.17* 0.11 -0.06 0.17 0.02 0.06 -0.20* -0.15 0.14 0.11 0.19 Average -0.33 0.64 2.79 0.21 0.07

Note: averages are calculated using only the significant coefficients *indicates significance at a 5% significance level

#_{indicates that these regions were not cointegrated according to the Johansen test}

Source: Own calculations

The results in Table 3 indicate that in most cases only the adjustment coefficient and the cointegration coefficient are significant. The significant adjustment coefficients vary between -0.24 and -0.86, with an average of -0.33. Since the prices were converted to natural logarithms, this implies that on average 33% of any deviation from the long-run relationship will be corrected within one month. This is relatively large compared to reported estimations in the literature, which tend to report an adjustment coefficient of

(25)

around -0.11 (Greb et al. 2012). It is likely that the concept of asymmetric price transmission plays a role here. Since international prices were rising rapidly in the period under investigation, and the theory of asymmetric price transmission argues that price increases are more likely transmitted than price decreases, it is not surprising that the estimated adjustment coefficient is higher in this analysis than in the literature.

The cointegration coefficient has an average of 0.64 and varies between 0.36 and 0.86, which implies that changes in international prices are transmitted to regional prices by 36-86%. The average is lower than the average found by Greb et al. (0.75), but this could well be due to the fact that this thesis only uses African data. In most literature, African prices are indeed found to be less likely to be cointegrated with international prices than other regions.In addition, more than half of the prices included were import prices (see Table 2) which implies that in the existence of transport and transaction costs, as captured by the constant term, the transmission elasticity will be smaller than 1 even in the case of perfect absolute price transmission.5_{The influence of transport and transaction costs, as} captured by the constant term is very clear: the larger these costs, the lower the cointegration coefficient. This has to do with the fact that the larger the constant term γ is in the long-run equilibrium domestic price, the smaller the effect of an absolute change in the world price. The average of this constant term is 2.79, indicating that there is a clear trend in the long-term relation, which can be interpreted as evidence for the existence of transport and transaction costs. The constant in the current price changes, i.e. 𝜑1 in equation 2, is mostly insignificant and consistently close to 0.01. Since this proved to be the case for all other estimated equations as well, they will not be discussed in the remainder of this thesis. The short-run elasticities are significant in only a few cases and of a small magnitude: a short-run elasticity of 0.24 w.r.t. the world price in Garoua (Cameroon) for instance, implies that a 1% increase in the world price in the past month would lead to an increase in the regional price in Garoua this month by 0.24%.

5_{See the example in section 3.2.1}

20

(26)

4.2 Price transmission between regions within a country

Given the limited evidence for cointegration between international prices and regional African prices, as well as the fact that some analysts consider regional factors more important than international factors for the determination of food prices, this section will consider the extent of cointegration between different regions in a country. Since in this case it is not evident anymore which region is dependent and which is independent, a Vector Error Correction Model (VECM) will be estimated, in which the first difference of each regional price is regressed on the first differences of other regional prices. For three regions, this results in a model of the following form:

∆𝑝_𝑡𝑟𝑒𝑔𝑖𝑜𝑛1 = 𝜑1+ 𝛼1�𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛1− 𝛽11𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛2− 𝛽12𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛3+ 𝜆1� + 𝛿1∆𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛1+ 𝜌1∆𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛2+ 𝛾1∆𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛3+ 𝜀1𝑡 (3) ∆𝑝_𝑡𝑟𝑒𝑔𝑖𝑜𝑛2 = 𝜑2+ 𝛼2�𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛2− 𝛽21𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛1− 𝛽22𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛3+ 𝜆2� + 𝛿2∆𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛2+ 𝜌2∆𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛1+ 𝛾2∆𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛3+ 𝜀2𝑡 (4) ∆𝑝_𝑡𝑟𝑒𝑔𝑖𝑜𝑛3 = 𝜑3+ 𝛼3�𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛3− 𝛽31𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛1− 𝛽32𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛2+ 𝜆3� + 𝛿3∆𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛3+ 𝜌3∆𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛2+ 𝛾3∆𝑝𝑡−1𝑟𝑒𝑔𝑖𝑜𝑛1+ 𝜀3𝑡 (5)

In addition, since there are multiple dependent variables included in the VECM now, there can also be multiple cointegrating independent relationships. This implies that the methodology needs to be slightly altered. After testing for the number of lags and the existence of a unit root, the Johansen test is used to test for the number of cointegrating equations in the model, or the cointegrating rank of the VECM. In this case the Johansen test is preferred to the Engle Granger test, since the latter only allows for one cointegrating relationship. Subsequently, the VECM is estimated using the found rank and lags.

For the analysis, the number of regions, (or included variables in general) is relevant 21

(27)

in a VECM since the closer the rank is to the number of regions included, the higher the extent of market integration, or price transmission between the different regional markets in a country(Goodwin and Grennes 1994). This implies that a cointegrating rank of 2 indicates higher integration if there are three regions included than if there are 6 regions included. Furthermore, the interpretation of a VECM with one cointegration relation is similar to the interpretation of an ECM (with one equation). For example, a VECM with three equations and one cointegrating relation can be rewritten as three separate ECM equations. However, a complete interpretation of a VECM with multiple cointegration equations is complicated if there is no prior economic knowledge about the direction of causality. This is the result of the independence condition of each cointegrating equation which implies that one region has to be excluded in a VECM with rank 2, and two regions in a VECM with rank 3. If no region would be excluded, the cointegration relations would not be independent anymore and hence contain little useful information. When there is no clear cut theoretically based restriction regarding the regions to be excluded however, interpretation of the error correction term does not make sense since Stata then randomly excludes regions. Therefore, the only way to further examine the direction and extent of integration is by estimating a bivariate VECM, including only two regions, since the bivariate model does not require any prior knowledge regarding the direction of transmission. The next section will start by providing an overview of the general results regarding regional cointegration in the different countries. Subsequently, the countries that exhibited cointegration will be investigated further.

4.2.1 General results

For the analysis of regional price transmission, there was data available for 9 countries on the same commodity price in multiple regions. Table 6 describes the main results of the lag length selection test, the test for cointegration, the rank of the VECM and the number of regions that were included in the VECM. In some cases not all regions for which prices were available could be included in the VECM since these prices were stationary.

(28)

Table 4: Cointegration between regional prices

Country Commodity Lag length Cointegration? Rank Number of

regions Cameroon Chad Congo Malawi Mozambique Niger Senegal Somalia Togo Maize Rice Maize Rice Sorghum Maize Rice Maize Rice Maize Rice Sorghum Rice Sorghum Rice Rice Sorghum 2 2 4 4 4 4 4 4 4 4 4 4 3 4 4 4 4 Yes Yes No No No No No No Yes Yes Yes No Yes No Yes No Yes 1 2 - - - - - - 1 3 2 - 2 - 3 - 2 4 5 - - - - - - 7 6 5 - 7 - 8 - 4

Note: cointegration was determined based on a significance level of 5% Source: Own calculations

The results in Table 4 show some interesting differences with the results of the tests for cointegration between world prices and regional prices. Firstly, it is noteworthy that the selected lag length for regional prices is almost consistently 4, with the exception of prices in Cameroon and rice prices in Senegal. Unlike the selected lag lengths in the international cointegration tests, there are no clear differences between commodities anymore. The longer lag selection suggests that it takes more time to transmit price changes between regions within a country than between international and regional markets. A second obvious difference with the results for international cointegration, is that Table 6 reveals cointegration for sorghum and maize prices as well, although rice prices remain clearly dominant with respect to the number of cointegrated series: five of the eight series are rice prices. This indicates that crops like sorghum and maize are traded within a country, but less so on international markets.

From the nine countries that were analysed, six had one or more cointegrating relations and hence exhibited some extent of market integration. The three countries that

(29)

do not reveal any market integration are Chad, Congo and Malawi. In the other six countries, comparing the rank of the VECM with the number of regions included provides a rough indication of the extent of market integration for the specific commodities. This implies that in Cameroon the rice market is more integrated than the maize market, while in Niger the maize market appears to be more integrated than the rice market.

4.2.2.Country results

Single cointegration: maize in Cameroon and rice in Mozambique

This section first describes the VECM for maize in Cameroon and rice in Mozambique, since these country/commodity combinations led to a VECM of rank 1, meaning that one cointegrating relation was found between all regions.

Maize prices in Cameroon

Below are three equations of the VECM for maize in Cameroon. An asterisk indicates significance of the coefficient at the 5% level. The average cointegration coefficient, which is calculated using only the significant variables, is 1.27 in absolute value, the average adjustment coefficient is -0.33. The equation for the fourth region, Yaundé has been omitted here, since the estimated coefficients were extremely large and unlikely to be true. The ECM equation for Yaundé has been included in appendix A3 however.

∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝑩𝒂𝒎𝒆𝒏𝒅𝒂 = −𝟎. 𝟏𝟕�𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝑩𝒂𝒎𝒆𝒏𝒅𝒂_{− 𝟏. 𝟓}∗_{𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏}𝑫𝒐𝒖𝒂𝒍𝒂_{+ 𝟎. 𝟓𝟔}∗_{𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏}𝑮𝒂𝒓𝒐𝒖𝒂 + 𝟎. 𝟎𝟒𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝒀𝒂𝒖𝒏𝒅é− 𝟎. 𝟏𝟗� + 𝟎. 𝟏𝟑∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝑫𝒐𝒖𝒂𝒍𝒂+ 𝟎. 𝟎𝟒∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝑮𝒂𝒓𝒐𝒖𝒂 + 𝟎. 𝟒∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝒀𝒂𝒖𝒏𝒅𝒆− 𝟎. 𝟏𝟗∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝑩𝒂𝒎𝒆𝒏𝒅𝒂 ∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝑫𝒐𝒖𝒂𝒍𝒂 = −𝟎. 𝟒𝟑∗_{�𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆} 𝒕−𝟏 𝑫𝒐𝒖𝒂𝒍𝒂_{−𝟎. 𝟔𝟕}∗_{𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆} 𝒕−𝟏 𝑩𝒂𝒎𝒆𝒏𝒅𝒂_{−𝟎. 𝟑𝟕}∗_{𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆} 𝒕−𝟏 𝑮𝒂𝒓𝒐𝒖𝒂 − 𝟎. 𝟎𝟐𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝒀𝒂𝒖𝒏𝒅é− 𝟎. 𝟏𝟑� + 𝟎. 𝟏𝟗∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝑩𝒂𝒎𝒆𝒏𝒅𝒂− 𝟎. 𝟎𝟗∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝑮𝒂𝒓𝒐𝒖𝒂 + 𝟎. 𝟐𝟔∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝒀𝒂𝒖𝒏𝒅𝒆+ 𝟎. 𝟎𝟗∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝑫𝒐𝒖𝒂𝒍𝒂 ∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝑮𝒂𝒓𝒐𝒖𝒂 = −𝟎. 𝟐𝟐∗_{�𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏}𝑮𝒂𝒓𝒐𝒖𝒂_{+ 𝟏. 𝟖}∗_{𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏}𝑩𝒂𝒎𝒆𝒏𝒅𝒂_{−𝟐. 𝟔𝟗}∗_{𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏}𝑫𝒐𝒖𝒂𝒍𝒂 + 𝟎. 𝟎𝟔𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝒀𝒂𝒖𝒏𝒅é_{− 𝟎. 𝟑𝟒� + 𝟎. 𝟑}∗_{∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏}𝑩𝒂𝒎𝒆𝒏𝒅𝒂_{− 𝟎. 𝟑∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏}𝑫𝒐𝒖𝒂𝒍𝒂 + 𝟎. 𝟑𝟑∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏𝒀𝒂𝒖𝒏𝒅𝒆_{− 𝟎. 𝟎𝟗∆𝒍𝒏𝒑𝑴𝒂𝒊𝒛𝒆𝒕−𝟏}𝑮𝒂𝒓𝒐𝒖𝒂 24

(30)

From these equations, it appears as if there is a long-run significant relationship between Bamenda, Douala and Garoua, since their cointegrating coefficients are all significant. The adjustment coefficients for Douala and Garoua were also significant: in Douala 43% of a deviation from the long-run equilibrium was corrected within one month, in Garoua this was 22%. The constants in the cointegrating relations of Cameroon are rather small and all negative. While it is difficult to interpret the sign of the constant, the small size does suggest that transport and transaction costs are relatively low. Furthermore, lagged prices have no significant effect on current prices, with the exception of Bamenda maize prices for Garoua, which shows a positive relation of 0.3.

The interdependencies between the different regions are easier interpreted from behavioural equations, which describe to what extent regional prices respond to changes in other regional prices6_{in the long run. In order to create a behavioural equation for each} region, one can take the error correction term out of the VECM equation, and subsequently take the normalized variable to the left side of an equation, while reversing the signs of the coefficients in the error correction term (Toppinen 1998). This then results in the following behavioural equations for maize prizes in Bamenda, Douala and Garoua:

𝑙𝑛𝑝𝑀𝑎𝑖𝑧𝑒𝐵𝑎𝑚𝑒𝑛𝑑𝑎 _{= 1.5𝑙𝑛𝑝𝑀𝑎𝑖𝑧𝑒}𝐷𝑜𝑢𝑎𝑙𝑎_{− 0.56𝑙𝑛𝑝𝑀𝑎𝑖𝑧𝑒}𝐺𝑎𝑟𝑜𝑢𝑎_{+ 0.19} ₍₆₎ 𝑙𝑛𝑝𝑀𝑎𝑖𝑧𝑒𝐷𝑜𝑢𝑎𝑙𝑎 _{= 0.67𝑙𝑛𝑝𝑀𝑎𝑖𝑧𝑒}𝐵𝑎𝑚𝑒𝑛𝑑𝑎_{+ 0.37𝑙𝑛𝑝𝑀𝑎𝑖𝑧𝑒}𝐺𝑎𝑟𝑜𝑢𝑎_{+ 0.13} ₍₇₎ 𝑙𝑛𝑝𝑀𝑎𝑖𝑧𝑒𝐺𝑎𝑟𝑜𝑢𝑎 _{= −1.8𝑙𝑛𝑝𝑀𝑎𝑖𝑧𝑒}𝐵𝑎𝑚𝑒𝑛𝑑𝑎_{+ 2.69𝑙𝑛𝑝𝑀𝑎𝑖𝑧𝑒}𝐷𝑜𝑢𝑎𝑙𝑎_{+ 0.34} ₍₈₎

These results suggest that price changes in Douala lead to particularly strong price changes in Bamenda and Garoua. Similar estimates in the literature have been mostly discussed in terms of vertical price transmission, where coefficients of this size suggest a high degree of market power (Amikuzuno and Ogundari 2012:14). However, in the case of spatial price transmission, estimates of this size could also imply that the good is an export good(Brooks and Melyukhina 2005). In the situation above, this would mean that Bamenda and Garoua are exporting maize to Douala. This suggestion is supported by the fact that coefficients on maize prices in Douala and Garoua in equation 7 are rather small, indicating that they could be import prices. More specifically, in this case, it is possible that an exogenous decrease in

6_{The use of logarithms implies that the cointgration coefficients should be interpreted as elasticities}

25

(31)

the supply of maize in Douala leads to an increase in exports to this region, from Garoua and Bamenda. Assuming that maize from the different regions is perceived as a homogeneous good, i.e. that the elasticities of demand and supply are the same, the effect of the decrease in supply in Douala is larger, the larger the volume of Douala is, and the smaller the transaction costs. Since the transaction costs, as captured by the constant, are rather small, this suggests that the volume of maize in Douala is relatively large compared to the volume in other regions. This explains the large effect of price changes in Douala on the other two regions.

A second interesting finding is the negative relationship between maize prices in Bamenda and Garoua. The literature scarcely discusses the interpretation of negative price transmission coefficients, especially in the context of spatial price transmission. This is rather unfortunate since negative estimates are not uncommon (Amikuzuno and Ogundari 2012: 4,Barret 1996: 827). Suggestions that have been made are related to competitive pricing strategies, where traders choose to lower their own price despite higher international prices, in order to secure their market share (Meyer and von Cramon Taubadel 2004, Quagrainie and Engle 2002). Alternatively, external factors such as geographical differences and differences in rainfall seasons could have an indirect effect on prices and thereby suggest a long-term relation which is not based on the same fundamentals but rather a coincidence. Most important however, is that a negative cointegrating relation implies that prices move in opposite directions instead of moving together (Barrett 1996: 827). This suggests that cointegration does not always imply integrated markets. In this case, it is possible that the cointegration relation between Garoua and Bamenda only exists through both of their interactions with Douala. However, it remains extremely difficult to verify this explanation without having more information about the regions.

Rice prices in Mozambique

In Mozambique there were seven regions with one cointegrating relationship for rice prices. Table 5 reports the estimations of the adjustment coefficient and the cointegration

coefficient on the different regions, as well as the long-run constant. To save space and because no clear pattern emerges, the lagged variables are not included in this table, the full equations can be found in appendix A4.

(32)

Table 5: Estimated parameters for rice prices in Mozambique Region Α Adjustment coefficient: β Angonia Gorongosβ a β

Manica Maputo β Maxixe β Nampula Β Ribaue β Constant γ of long-run relation Angonia -0.37* - -0.07 0.67* _-0.31 _0.89* _0.23 _-0.45* _-0.55 Gorongosa -0.03* _-13.97* _- _9.4* _-4.4* _12.4* _3.17 _-6.27* _-7.62 Manica -0.21* _1.49* _0.11 _- _0.47 _-1.32* _0.34 _0.67 _-0.81 Maputo -0.02 -3.16* _-0.22 _2.12* _- _2.81* _0.72 _-1.42* _-1.73 Maxixe 0.08 1.13* _0.08 _-0.75* _0.36* _- _-0.26 _0.51* _-0.62 Nampula 0.04 4.4* _0.31 _-2.96* _1.39 _-3.9* _- _1.97* _-2.4 Ribaue -0.18* _-2.23* _-0.16 _1.5* _-0.71 _1.98* _0.51 _- _-1.22

Note: * indicates significance at the 5% level Source: Own calculations

From Table 5 it appears as if there is a long-run relationship between Angonia, Manica, Maxixe and Ribaue. The average cointegration coefficient for rice prices in Mozambique is extremely large: 1.8 if Gorongosa is excluded and 3.3 if this region is included. Looking at the size of the specific cointegration coefficients, one can say something about the possible directions of trade: Manica, Maxixe and Nampula seem to be exporting to Angonia, considering the large coefficients on the latter region. In addition, Nampula could be exporting to Ribaue given the coefficient of 1.97 on this region. Finally, Maputo and Ribaue appear to export to Manica and Maxixe, since these coefficients are all larger than 1.5. Adjustment coefficients are significant in Angonia, Gorongosa, Manica and Ribaue, with an average of about 20%.7 The differences between the adjustment coefficients are very large however, varying between -0.03 for Gorongosa and -0.37 for Angonia. Overall, the constants are rather small, with the exception of the constant of Gorongosa. This suggests that transport and transaction costs for Gorongosa are very high, hindering market integration for this region. With respect to the signs of the estimated parameters, some points stand out. Firstly, there are 16 negative cointegration coefficients, of which 9 are

7_{The average is calculated using only the significant values}

27

(33)

significant.8 As discussed in the previous section, these negative estimates would mean prices drift apart instead of moving together. This suggests that market integration for these markets is unlikely. Secondly, although the positive adjustment coefficients for Maxixe and Ribaue are not significant, they are rather strange; it would imply that deviations from the long-run equilibrium are amplified rather than corrected (Greb et al. 2012: p.16).

Multiple cointegration

As mentioned in section 3.2.2., it is impossible to give a plausible interpretation of the cointegrating relationships if there is more than one relationship and if there is no prior economic knowledge about the mechanics at work. Therefore, this section further investigates the cointegration between different regions in one country, by estimating a VECM for only two regions at a time.

Rice prices in Cameroon

Table 6 presents the estimated cointegration coefficients for rice prices in Cameroon. The left column is the dependent variable, the upper row shows the independent variables. For example, the first row would result in the following behavioural equations for rice prices in Bafoussam:

ln𝒑𝑹𝒊𝒄𝒆_𝒕−𝟏𝑩𝒂𝒇𝒐𝒖𝒔𝒔𝒂𝒎= 𝟎. 𝟗𝟐𝒍𝒏𝒑𝑹𝒊𝒄𝒆𝒕−𝟏𝑩𝒂𝒎𝒆𝒏𝒅𝒂_{+ 𝒄𝒐𝒏𝒔𝒕𝒂𝒏𝒕} ₍₉₎

𝒍𝒏𝒑𝑹𝒊𝒄𝒆_𝒕−𝟏𝑩𝒂𝒇𝒐𝒖𝒔𝒔𝒂𝒎= 𝟎. 𝟗𝟗𝒍𝒏𝒑𝑹𝒊𝒄𝒆𝒕−𝟏𝑫𝒐𝒖𝒂𝒍𝒂+ 𝒄𝒐𝒏𝒔𝒕𝒂𝒏𝒕 (10)

𝒍𝒏𝒑𝑹𝒊𝒄𝒆_𝒕−𝟏𝑩𝒂𝒇𝒐𝒖𝒔𝒔𝒂𝒎= 𝟏. 𝟎𝟓𝒍𝒏𝒑𝑹𝒊𝒄𝒆𝒕−𝟏𝑮𝒂𝒓𝒐𝒖𝒂+ 𝒄𝒐𝒏𝒔𝒕𝒂𝒏𝒕 (11)

𝒍𝒏𝒑𝑹𝒊𝒄𝒆𝒕−𝟏𝑩𝒂𝒇𝒐𝒖𝒔𝒔𝒂𝒎= 𝟎. 𝟖𝟗𝒍𝒏𝒑𝑹𝒊𝒄𝒆𝒕−𝟏𝒀𝒂𝒖𝒏𝒅𝒆+ 𝒄𝒐𝒏𝒔𝒕𝒂𝒏𝒕 (12)

8_{More specifically, there appears to be a negative relation between Angonia and Gorongosa, Angonia and}

Ribaue, Angonia and Maputo, Maputo and Ribaue, Gorongosa and Ribaue, Gorongosa and Maputo, Maxixe and Manica and Maxixe and Nampula.

28

(34)

The constants varied between -0.71 and 1.53 and there were no significant coefficients on the lagged variables. The full equations, including the constants and lagged variables can be found in appendix A5.

Table 6: Cointegration coefficients for rice in Cameroon

Region Bafoussam Bamenda Douala Garoua Yaundé

Bafoussam — 0.92* _0.99* _1.05* _0.89*

Bamenda 1.08* _— _1.12* _1.16* _0.95*

Douala 1.01* _0.89* _— _1.02* _0.95*

Garoua 0.96* _0.86* _0.98* _— _0.82*

Yaundé 1.13* _1.06* _1.06* _1.22* _—

Note: * indicates significance at the 5% level Source: own calculations

All cointegration coefficients are significant and positive and have an average of exactly 1, indicating that rice prices in Cameroon tend to follow the same path over the long run. The size of the coefficients of Garoua suggest that this region is importing its rice from other regions in Cameroon: the coefficients are consistently below unity. Yaundé on the other hand, seems to be one of the main exporting regions, since this region’s coefficients are all larger than 1.

Table 7 demonstrates the results for the adjustment coefficients. The average adjustment coefficient, again including only the significant variables, is -0.32. All regions adjust to a deviation from their equilibrium relationship with Douala, ranging from an adjustment of 53% within one month in Bamenda to an adjustment of 12% within one month in Yaundé. Deviations from the long-run relationship with Garoua are also mostly adjusted; except for Douala all regions have a significant adjustment coefficient on this region. Furthermore, no region has a significant adjustment coefficient for Bamenda, suggesting that Bamenda is not an important determinant for regional price adjustments.

Table 7: Adjustment coefficients for rice in Cameroon

Region Bafoussam Bamenda Douala Garoua Yaundé

Bafoussam — -0.09 -0.35* _-0.31* _-0.15

Bamenda -0.31* _— _-0.53* _-0.53* _-0.29*

Douala -0.01 0.05 — -0.02 -0.02

Garoua -0.17 -0.15 -0.49* _— _-0.24*

Yaundé -0.21* _-0.16 _-0.12* _-0.13* _—

(35)

Maize prices in Niger

Table 8 displays the estimated cointegration coefficients that were found in the bivariate ECM models for maize prices in Niger. Like rice prices in Cameroon, all estimates are close to 1, the exact average is 1.01. From the six regions that were included in this analysis, it seems that Agadez is the largest importing region, considering that all cointegration coefficients of Agadez are smaller than 1. Similarly, the estimates in Table 8 suggest that Maradi is an exporting region, given that all cointegration coefficients of that region are larger than 1. The regions Dosso and Niarney and Tillaberi and Zinder, exhibit price transmission of exactly 1, which would imply that these markets are perfectly integrated with each other, assuming that transport and transaction costs are not too large. Given the size of the constants, which are 0.05 and 0.12 respectively, this is not a strange assumption. An overview of all the equations, can be found in appendix A6; the lagged variables were hardly significant and constants range from -0.89 to 1.05.

Table 8: Cointegration coefficients for maize in Niger

Region Agadez Dosso Maradi Niarney Tillaberi Zinder

Agadez — 0.93* _0.82* _0.93* _0.89* _0.86* Dosso 1.08* _— _0.86* ₁* _0.97* _0.90* Maradi 1.22* _1.17* _— _1.14* _1.06* _1.06* Niarney 1.08* ₁* _0.88* _— _0.97* _0.92* Tillaberi 1.12* _1.03* _0.94* _1.03* _- ₁* Zinder 1.17* _1.12* _0.95* _1.09* ₁* _—

The average adjustment coefficient for Maize in Niger is -0.26, indicating that 26% of a deviation from the long-run equilibrium is corrected within one month. It is worth noting that the adjustment coefficients on Maradi are all significant and relatively large. This suggests that Maradi is indeed an important internal export market, since it indicates that in the short run, prices respond more quickly to deviations from this equilibrium than from the equilibrium with other regions. In addition to Maradi, adjustment coefficients on Zinder are also quite large. With respect to the latter region it is interesting to consider that there does seem to exist a leader and a follower role between the regions that exhibited long-term cointegration of exactly 1. Since the adjustment coefficient of Tillaberi on Zinder is

The extent of price transmission between international and regional cereal markets in Africa