The Impact of World Oil Price Shocks on Maize Prices in East Africa

Page created by Allen Gutierrez
 
CONTINUE READING
The Impact of World Oil Price Shocks
            on Maize Prices in East Africa

                           Brian M. Dillona,b and Christopher B. Barretta,c,d *

                                               May 5, 2013
                                      First public draft for comments

Abstract:

This paper estimates the impacts of changes in global crude oil prices and global maize prices on
local maize prices in east Africa. We estimate the inter-commodity price transmission between
crude oil and maize prices in global markets, and then the spatiotemporal price transmission of
fuel and maize prices between global markets, port-of-entry markets, and sub-national markets
within the four major east African economies. Although global oil prices exert no identifiable
effect on global maize prices, they do affect sub-national maize market prices through their
impacts on transport fuel prices. The net effects are significant. The average elasticity of local
maize prices with respect to global oil prices is 0.29. The average elasticity of local maize prices
with respect to global maize prices is 0.44. In the four study markets that are farthest from
coastal ports, changes in transport costs have larger effects on maize prices than do changes in
the global market price of the grain itself. Also, oil price shocks on the world market transmit
much more rapidly throughout fuel markets in the region than do global maize price shocks. The
policy implications are clear: when oil and maize prices co-move on global markets, policies to
intervene directly in grain markets will not have the desired impact if they are not coupled with
policies to address rising transport costs.

a
  Charles H. Dyson School of Applied Economics and Management, Cornell University
b
  Sustainability Science Program, Kennedy School of Government, Harvard University
c
  Department of Economics, Cornell University
d
  David R. Atkinson Center for a Sustainable Future, Cornell University

*
  We thank Joanna Barrett and Shun Chonabayashi for research assistance, Chris Adam, Harry DeGorter, Oliver
Gao, Miguel Gómez, Yossie Hollander, Wenjing Hu, David Lee, Tanvi Rao and participants at a Cornell University
workshop for helpful comments on preliminary results, and the Bill and Melinda Gates Foundation for financial
support of this project. Assistance in gathering data from government sources was provided by Mujo Moyo in
Tanzania, Bjorn Van Campenhout, Todd Benson, and Andrew Mutyaba in Uganda, Pasquel Gichohi in Kenya, and
Bart Minten, Guush Tesfay, and Temesgen Mulugeta in Ethiopia. This project was partly undertaken through a
collaborative arrangement with the Tanzania office of the International Growth Center. Barrett thanks Monash
University and the University of Melbourne for their hospitality while he worked on this paper, and Dillon thanks
the Harvard Kennedy School for ongoing support through the Sustainability Science program. Any errors are the
authors’ alone.
1. Introduction

The global food price crises of 2008 and 2011 attracted widespread, high-level attention to the
adverse prospective effects of price shocks on poverty, food insecurity and macroeconomic
stability in the developing world. In subsequent analyses of what caused those grain price spikes
(e.g., Abbott et al. 2008, Headey and Fan 2008, Mitchell 2008, Rosegrant et al. 2008), one
prominent thread emphasizes the connections between crude oil prices and food prices. Yet there
is a notable absence of empirical analysis of if and how these connections actually operate at the
level of local markets within developing countries. Do global oil price shocks impact local food
prices, particularly in countries with high levels of subsistence food production? If so, how
much, and by what mechanism(s)?

This paper tackles those questions, focusing on maize markets within the four major east African
economies: Ethiopia, Kenya, Tanzania and Uganda. We use a newly assembled data set of local,
monthly average maize and petrol prices (at the pump) from 17 sub-national markets, covering
most of the period January 2000 – November 2012, to study the impacts on local maize prices of
changes in the prices of oil and maize on world markets. We use the estimated models to
estimate the cumulative impact of shifts in global commodity prices on food prices across the
region.

These questions harken back to an earlier literature in which economists worried about
commodity price dynamics and global-to-local price transmission because of their economic
impacts on developing countries, perhaps especially in Africa (Ardeni and Wright 1992, Deaton
and Laroque 1992, 1996, Deaton 1999). As Deaton (1999, p.24) lamented then, emphasizing
commodity exports rather than imports on which we focus, “the understanding of commodity
prices and the ability to forecast them remains seriously inadequate. Without such understanding,
it is difficult to construct good policy rules.” The same concern applies today. As we
demonstrate below, in the absence of careful empirical analysis of commodity price behavior,
even very thoughtful commentators may misunderstand the drivers of observed patterns, with
important implications for public policy.

Our focus is on maize because it is the primary staple grain in east Africa, and the most
important source of both calories for consumers and income for many small farmers. Maize is
also the world’s primary ethanol feedstock due to the US ethanol industry’s reliance on corn. We
study multiple markets in several countries so as to ensure that country-specific infrastructure or
policies do not drive results; our core findings are qualitatively identical and of similar order of
magnitude across the full set of sub-national markets.

Oil prices can influence grain prices through three main channels. First, higher oil prices can
drive up the cost of inputs such as inorganic fertilizer and fuel for running tractors or pumps.
Second, higher oil prices can induce increased conversion of corn into ethanol, thereby reducing
maize supplies and driving up global prices, which are then transmitted to local markets through
trade linkages. Third, oil price increases can drive up the cost of transport, which in turn affects
the prices of commodities that are traded across space.

                                                 1
The first channel is of little significance for three of the four countries in our study. Farmers in
Tanzania, Uganda, and Ethiopia use very little inorganic fertilizer or machinery. In these
countries, induced changes in input prices caused by rising oil prices are unlikely to have
appreciable effects on domestic maize output under the production arrangements that generated
our data. In Kenya, however, the government has actively promoted the import and distribution
of inorganic fertilizer for the last two decades, and fertilizer application rates are well above the
regional average (Ariga and Jayne, 2009). While we would not expect local production costs to
have a long run impact on the output price for a price-taking economy, seasonal impacts are
possible. Because this channel is only relevant for one country, we set it aside initially. In
Section 6 and Appendix A we extend our core results to allow for changes in the price of
inorganic fertilizer in Nairobi to impact sub-national maize prices in Kenya.1 Although our
findings are based on a relatively short time series, we find no evidence that fertilizer prices put
upward pressure on maize prices in Kenya, in the short or the long run, consistent with the
observation that Kenya is a price-taker on international maize markets and thus its local costs of
production affect only its producers’ profits and output levels, not prices, in equilibrium.

The second channel rests on the premise that biofuel production creates a systematic, causal
relationship between oil prices and maize prices. This potential food and energy price link has
received substantial attention in recent years, particularly since the passing of the ethanol
mandate as part of the US Energy Policy Act of 2005 (Krugman 2008, Mitchell 2008).
However, historically oil prices and basic grains prices have been only weakly linked,2 and the
recent literature finds little empirical support for the hypothesis that oil price shocks transmit
strongly to maize prices (e.g., Zhang et al. 2007, 2009, 2010, Gilbert 2010, Serra et al. 2011,
Zilberman et al. 2013). We test a variety of models linking oil and maize prices on global
markets. We find no evidence of a systematic, causal link, and therefore do not emphasize this
channel when we estimate the cumulative impact of global oil price changes on local maize
prices.

It is the third channel – the link through transport costs – on which we focus. Transport costs
loom large in African maize markets, because of the low value-to-weight ratio of grains, long
distances between population centers, and rudimentary transport infrastructure dependent
primarily on trucks (lorries). Even though subsistence agricultural production is still widespread
across east Africa, all four study countries trade maize on world markets. More importantly, the
food supply to non-farming households in cities and towns relies heavily on lorry-based grain
shipments from the breadbasket regions. As we show, although global oil prices exert no
identifiable effect on global maize prices, they do affect sub-national maize market prices
through their impacts on transport fuel prices.

The link between transport costs and economic development has been extensively studied. A
very large literature focuses on the impacts of changes to the fixed cost components of transport
– roads, railways, etc. – on growth, prices, migration, sectoral composition, and other economic

1
  Of course, rising production costs can influence the price of maize on global markets, which in turn affects the
price of maize in our study countries. We directly use observed prices of oil and maize at world markets, which
suffices to capture any underlying impacts mediated through production costs in non-study countries.
2
  Using simple regression techniques, Baffes (2007) estimates a 0.19 elasticity of maize prices with respect to crude
oil prices changes in global markets using annual data from 1960-2005.

                                                          2
outcomes (see, for example, Fogel 1964, Chandra and Thompson 2000, Jacoby 2000, Limão and
Venables 2001, Baum-Snow 2007, Jacoby and Minten 2009, Atack et al. 2010, Buys et al. 2010,
Bell 2012, Storeygard 2012, Duranton et al. 2013, Donaldson forthcoming).3 In this paper, we
estimate the impact of changes to the key variable transport cost on the price of a key staple food
commodity. While the fuel-food price connection is often discussed in the popular press,
particularly in the US,4 the thread of the spatial price transmission literature that emphasizes the
importance of time-varying transport costs is relatively new (Barrett 1996, Baulch 1997, Fackler
and Goodwin 2001, Barrett and Li 2002, Meyer and von Cramon-Taubadel 2004, Moser et al.
2009, Stephens et al. 2012). This is likely due to the scant availability of data on the price of
transport services (World Bank, 2009, p. 175).

The empirical approach in this paper involves stepwise application of standard time series
techniques to sequences of price pairs. First, we separately estimate the impact of global price
movements in oil and maize on petrol and maize prices in the port-of-entry5 (POE) markets for
the four study countries, directly allowing changes in transport costs to impact the price margin
between POE maize and global maize. We then estimate the link between POE petrol prices and
petrol prices in geographically dispersed sub-national markets. Not surprisingly for these strict
oil-importing countries, we find rapid, nearly complete price pass-through in fuel markets. Next,
we estimate the pass-through rates of maize prices from the POE markets to other sub-national
markets, allowing changes in local fuel prices to impact the maize price spread. Finally, we use
the empirical results to calculate cumulative pass-through elasticities, which measure the impacts
on local maize prices of increases in global oil and maize prices. We focus on long run impacts,
because in the empirical sections we find evidence of relatively rapid pass through in fuel
markets, suggesting that the “long run” is typically only a matter of months.

We find a significant, direct impact of fuel price changes on local maize prices. Across the 17
markets in our study, a 1% increase in nominal global oil prices is associated with an average
long run local nominal maize price increase of 0.29%, even in the absence of a correlated
increase in global maize prices. This finding is remarkably stable across study markets: all but
one of the estimated global oil price pass-through rates lies in the range 0.20-0.41%.

Similarly, a 1% rise in global maize prices, absent a corresponding change in oil prices, leads to
a 0.44% average local maize price increase. When global oil and maize prices simultaneously
increase by 1%, the average increase in local maize prices is 0.73%. Any remaining maize price
adjustments operate via the exchange rate (more on that in a moment). These estimated rates of
international price transmission are substantially greater than those reported in much of the
current literature, which does not account for the prospective role of fuel prices in changing
transport cost margins (Benson et al. 2008, Abbott and Borot de Battisti 2011, Baltzer 2013).

The four most geographically remote markets in our study (Gulu and Mbarara, Uganda, and
Kigoma and Mbeya, Tanzania) offer the best opportunities to explore the pass-through effects of
commodity price shocks in landlocked areas of Africa. Strikingly, we find that for these

3
  Storeygard (2012) uses global oil prices as a proxy for local fuel prices in a paper that considers both fixed and
variable transport costs in a study of the link between transport infrastructure, urbanization, and growth.
4
  For example, see Lesiew (2013), Bjerga (2012), Rooney (2011).
5
  “Port” is symbolic here, because Uganda and Ethiopia are land-locked.

                                                           3
landlocked areas, increases in transport costs are arguably more important to long run food
security than are the costs of the traded grains. For these markets, the elasticities of local maize
prices to global oil prices are greater in magnitude than are the corresponding elasticities of local
maize prices to global maize prices.

Causal identification in our nested series of estimates is largely straightforward. The four study
countries are all price-takers in both oil and maize markets. Domestic production of maize can
slow the transmission of global maize price changes, as we find, but in the long term prices
adjust to world market prices. Likewise, while fuel prices can influence maize prices via
transport costs, there is little reason to expect an effect of maize prices on local fuel prices,
especially given the absence of any large-scale maize-based ethanol production in the region.
Within each country we impose the identifying assumption that any disequilibrium between
prices at the port-of-entry – the major hub for international trade – and prices in another market j
is resolved through adjustment in market j. While this may be a simplification in the very short
run, it is surely a benign assumption in the medium and long term, since trade with international
markets is mediated through the major port-of-entry in each country.

If there is an identification challenge in the paper it relates to the exchange rate. We include
monthly exchange rates in the long run equations of all models that link global prices to domestic
prices. In Appendix B we provide evidence that in nearly all cases, exchange rate changes are at
least weakly exogenous to changes in the global prices of oil and maize. Nevertheless, over a
long enough time horizon, exchange rate movements are likely endogenous to major shifts in the
prices of traded commodities. In the interpretation of results, we allow for this possible exchange
rate endogeneity by separately assuming both zero and complete exchange rate adjustment to
increases in oil and maize prices, and using the estimates to bound cumulative pass-through
elasticities.

Our findings have immediate relevance for policymakers concerned about a global commodity
price spike. Increased high-level attention to global food security tends to focus on farm-level
productivity growth and on safety nets for poor consumers. Although these are clearly high
priorities, so too is it essential to increase efficiency in the post-harvest systems – including
liquid fuels and transport – that deliver food to rapidly urbanizing populations from both
domestic farmers and international markets (Gómez et al. 2011). Because commodity prices
tend to co-move on global markets, policymakers who ignore the correlated movement of grain
and oil prices may over-attribute domestic food price spikes to the direct effect of changes in
global grain prices. To the extent that fuel price rises are partially responsible for increases in
food prices, policies to mitigate food price spikes by intervening exclusively in grain markets
(such as export bans or releases of grain reserves) will be less effective than expected.

The rest of the paper proceeds as follows. In Section 2 we describe the basic features of maize
and oil markets in east Africa, as well as the relevant policy backdrop in the study countries.
Section 3 describes the data. In Section 4 we outline the empirical framework. In Section 5 we
present the empirical results. Section 6 discusses findings and presents the cumulative pass-
through elasticities. Section 7 concludes.

                                                 4
2. Maize and Oil in East Africa

The price of staple foods, like maize in the four countries we study, is always a serious economic
and political issue in developing countries. Price spikes can have pronounced effects on poverty
rates, inflation, external terms of trade, fiscal balances and lead to political instability (Barrett
2013). The specter of such consequences commonly induces significant policy interventions in
an attempt to dampen pass-through from international markets (Ivanic et al. 2012).

In the 1990s, many commodity markets in Africa were liberalized as part of a general shift in the
developing world away from central planning and toward market determination of prices and
trade flows. Since that time, governments in the four study countries have for the most part
withdrawn from direct participation in the production or distribution of food and fuel.
Nevertheless, there are policies in each study country that provide important context for the
analysis to follow. In this section we give a brief overview of the relevant policy environment,
and the role of maize in supporting food security, for each study country.6

Ethiopia

According to the FAO’s Food Balance Sheet (FBS) data, maize is the most important source of
calories in Ethiopia. In 2009, the average Ethiopian consumed 418 kcal/day per capita of maize,
accounting for 20% of total dietary energy intake. Table 1 shows the allocation of land to crops
in the study countries over the period 2007-2010. In Ethiopia, “Other cereals” – including teff,
wheat, barley and sorghum – are extensively cultivated. However, the number of acres allocated
to maize, the number of households growing maize, and total maize output are all greater than
the comparable figures for these other cereals (Rashid 2010).7 Figures 1 and 2 show the time
path of cultivated maize acreage and total maize output, respectively, for the four study
countries.8 Ethiopia shows a relatively steady upward trend in maize acreage and output over the
study period. Most maize in Ethiopia is consumed and produced domestically. Table 2 shows
annual maize net import (imports – exports) statistics for the four study countries over the period
2000-2010. Ethiopia is a marginal net importer, on average, though volumes of both exports and
imports are low.

The government of Ethiopia withdrew nearly all controls from maize markets during the period
1999-2002 (Rashid et al. 2010). The Ethiopian Grain Trade Enterprise (EGTE) still maintains
strategic grain reserves that act as a buffer stock in the event of price spikes, but the price
impacts of EGTE procurement are considered negligible (Tadesse & Guttormsen 2011). There is
no set of well-documented import or export policies for the international maize trade. However,
from the period 2008-2010 the government put in place a ban on exports, in response to global
food price spikes.

6
  For a more extensive discussion of the relevant policy environment, see Rao and Lee (2013).
7
  Over 8 million households cultivated maize in 2007-2008, as compared to 5.8 million households growing teff, and
4.2 million growing wheat. Total maize output in that same year was 4.2 million tons, compared to 3 million tons of
teff and 2.7 million tons of sorghum (Rashid 2010).
8
  While the FAO output statistics are the best available, we suggest that the numbers in Figures 2 and 3, particularly
those for output, be read with caution. Recent work on the quality of African economic data suggests that
measurement error is significant (Jerven 2013).

                                                          5
Government control of the oil and fuel sector in Ethiopia is by far the most significant form of
state intervention in any of the markets under study in this paper. The parastatal Ethiopian
Petroleum Enterprise (EPE) is the exclusive importer petrol and diesel, and the pump prices of
both commodities are fixed by the Ministry of Trade and Industry (MoTI). This is potentially
problematic for the analysis, because it suggests that observed fuel prices in Ethiopia are choice
variables rather than the product of market forces. However, the government of Ethiopia lacks
the resources to heavily subsidize a substantial fuel price subsidy for an extended length of time.
Indeed, as we show below, fuel prices in Ethiopia generally track global prices. We therefore
take the system of price determination as given, rather than attempt an explicit model of
objectives and price-setting mechanism of MoTI.

Kenya

Maize is more central to food security in Kenya than in any of the other study countries. The
2009 FBS data indicates average daily consumption of 672 kcal/day from maize, accounting for
32% of total dietary energy intake. Over the period 2007-2010, maize accounted for a greater
share of cultivated land than all other categories of crops in Kenya (Table 1). Figures 1 and 2
show steady increases in maize acreage and output over the study period. The government is a
significant player in the market, through the National Cereals Produce Board (NCPB, which
buys and sells maize to address government food security objectives, although the private market
remains highly competitive (Jayne et al. 2008).

Kenya is the only study country to exhibit significant volumes of international trade in maize.
However, in most study years maize net imports into Kenya are just a small fraction of domestic
production. Figure 3 shows maize production and imports in Kenya for the period 1997-2010.
Only in 1997 and 2009 do net maize imports account for more than 20% of consumption, and in
most years the proportion is far lower. Imports from other east African countries comprise a
negligible percentage of total imports in any year, indicating that the adjustment in import
volumes comes from outside the immediate region.

Kenya is also the only study country with noteworthy maize import tariffs. In the early years of
our study period, tariffs ranged from 20-30% (2000-2004). For the last five years, the
government of Kenya has not taxed maize imports from Uganda and Tanzania, but has imposed
50% tariffs on imports from elsewhere.

Finally, although Kenya like the other three countries is a strict importer of oil products, it is the
only study country with a domestic oil refinery. By mandate, domestic refining of imported
crude oil supplies approximately 50% of consumer fuel products to Kenyan markets (Kojima et
al. 2010). A pipeline carries oil from Mombasa to Nairobi, while road transport accounts for all
other fuel distribution within the country. In 2011 the government of Kenya instituted a system
of price controls in petrol and diesel markets, with the central government determining the
maximum price of each product in each major market. However, over the period of our data,
prices were market-determined throughout the country.

Tanzania

                                                  6
The FBS statistics for maize in Tanzania are similar to those from Ethiopia and Kenya: in 2009
the average Tanzanian consumed 519 kcal/day per capita of maize, comprising 23% of total
energy intake. Table 1 shows that maize cultivation in Tanzania accounts for greater acreage
than that of any other crop. Figure 1 shows a near doubling of maize cultivated acreage from the
1990s to the 2000s. However, the drastic change in acreage in 2002, along with the
corresponding sharp change in Tanzania’s maize output in 2002 (Figure 2), should be interpreted
with caution, as these are likely due to measurement error in the entire series that was corrected
during the 2002 agricultural census (Jerven 2013 and footnote 8). Regardless, maize production
is one of if not the most significant agricultural activity in Tanzania.

As in most of the region, the majority of the maize in Tanzania is consumed and produced
domestically. Table 2 shows that the country is a marginal net importer, though trade volumes
are relatively small, averaging only 0.5-1.6% of domestic production.

Maize prices in Tanzania are primarily determined by market forces. The government is not
heavily involved in the maize trade, although the Food Security Department of the Ministry of
Agriculture maintains a strategic grain reserve that it attempts to use to mitigate the effects of
large price shocks. The most significant maize policy in recent years has been a series of ad hoc
export bans, implemented periodically since 2008, purportedly to drive down prices during
periods of re-stocking reserves (Rao and Lee 2013).

From 2000-2005 prices of fuel products in Tanzania were determined competitively. Since
2006, the Energy and Water Utilities Regulatory Authority (EWURA) has issued a cap on the
prices of petrol, diesel, and kerosene, based on a publicly available formula. Beginning in 2011,
a competitive auction was established to assign bulk procurement rights (an exclusive importing
contract) to a single company for a two month period. This was primarily intended to reduce
congestion at the ports by sequencing the activities of the major fuel importers. Most of our data
pre-dates this policy change.

Uganda

Although maize is critical to food security in Uganda, it is the only study country in which other
crops account for a greater percentage of average daily caloric intake. In 2009 maize
consumption amounted to an average of 190 kcal/ day per capita, or 9% of total dietary energy
intake, which was less than the average intake of plantains (matoke) and cassava. Dependence
on maize varies regionally, with greater maize consumption in the north and east than in the
southwest. During the years 2007-2010, maize acreage in Uganda was greater than that of any
other crop (Table 1). Over the full study period (2000-2012), both acreage and output of maize
increased steadily (Figures 1 and 2). Uganda is the largest maize exporter among these four
countries (Table 2), and acts as a trading hub for the regional maize trade. However, import and
export volumes are still low relative to consumption or production.

Uganda is arguably the most liberalized market economy in east Africa. There are no price
controls on maize, and no government organizations involved in production or distribution.
There are no noteworthy import or export controls. In recent years, the World Food Program has

                                                7
procured substantial amounts of maize from Uganda for re-distribution within the region as food
aid, which has occasionally put some upward pressure on prices.

Likewise, the market for petrol and diesel in Uganda is less regulated than in the other countries.
Pump prices are competitively determined. Fuel products are imported via trucks from Kenya
and Tanzania, and retail prices are generally higher than in those two countries due to the
additional transport costs and costs associated with management of imports. There are no import
tariffs on oil products. Despite the lack of regulation and relative ease of doing business in
Uganda, Kojima et al. (2010) report substantial concentration in the fuel sector.9

3. Data

Figure 4 shows the location of all the sub-national markets for which we could match fuel and
maize price series.10 All are urban areas, but of varying size and remoteness from global markets.
The POE markets – Mombasa, Kenya; Dar es Salaam, Tanzania; Kampala, Uganda; and Addis
Ababa, Ethiopia – are indicated with a larger font size.

The highest frequency price series we could assemble for petrol and maize prices at the sub-
national market level was monthly. We match these data with readily available monthly time
series price data for crude oil and maize on global markets over the period January 2000 –
November 2012 (with only partial coverage of some series, as described below).

Global market price data

Global market price series come from the World Bank Global Economic Monitor (GEM)
database of commodity prices.11 Crude oil prices are expressed in nominal USD per barrel, and
are the equally weighted average spot price of Brent, Dubai, and West Texas Intermediate crude,
as calculated by the World Bank. Maize prices are expressed in nominal USD per metric ton for
number 2 yellow maize, f.o.b. at US Gulf ports.12, 13 In regressions that involve both global and
national prices, we use the monthly CPI and US dollar exchange rate for each study country, as
extracted from the IMF International Financial Statistics database.

Figure 5 shows the two global oil and global maize price series over the study period and the
decade previous. No obvious causal relationship presents itself. The correlation coefficient

9
 Oil was discovered in Uganda in 2006, though domestic extraction has not begun. It is expected that in the next 5-
10 years domestic oil production will come on-stream. However, we expect Uganda to remain a price-taker in oil
markets unless domestic storage and refining capacity expands dramatically.
1010
     In one location we have to merge series from proximate towns. We could assemble fuel price data from Nakuru
but not from Eldoret, while the maize price series from Eldoret was complete and that from Nakuru truncated. Since
these are quite proximate and the two main urban concentrations of Rift Valley Province in Kenya, we merge them
into a synthetic series, using Eldoret’s maize prices and Nakuru’s fuel prices.
11
   http://data.worldbank.org/data-catalog/commodity-price-data
12
   These are the same crude oil and maize price series that Baffes (2007) and Zhang et al. (2010) use.
13
   We do not deflate the global price series in the estimated models, for reasons discussed below.

                                                        8
between the nominal price series is 0.83 over the entire study period. However, much of this is
due to common inflationary trends. In real 2005 prices, the correlation coefficient is 0.45.14

Sub-national maize price data

Wholesale prices of white maize for markets in Kenya are from the USAID-funded Famine Early
Warning System (FEWS). Average wholesale maize prices for Tanzania were provided by the
Ministry of Agriculture, via the Bank of Tanzania and the International Growth Center (IGC).
Wholesale maize prices for markets in Ethiopia were downloaded from the website of the
Ethiopia Grain Trade Enterprise (EGTE). Retail white maize prices for Uganda markets are
from the Regional Agricultural Trade Intelligence Network (RATIN) of the East Africa Grain
Council (wholesale prices were not available).15

Table 3 shows the average price of maize at sub-national markets in the sample data, along with
the number and percentage of study months in which the price in each market was the lowest in
the country. Not surprisingly, the lowest average prices are not in the POE markets, but in the
trading centers closest to the maize breadbasket regions in each country (Bahir Dar, Ethiopia;
Eldoret/Nakuru, Kenya; Mbeya, Tanzania; Gulu, Uganda). On net, within-country transport
costs are incurred when grain is shipped from these regions to the rest of the country.

Perhaps the only surprise in Table 3 is that maize prices in Mombasa, Kenya, tend to be lower
than in Kisumu and Nairobi. Very little maize is grown in the coastal areas around Mombasa,
Kenya’s main port and second largest city. This finding may indicate that maize markets in the
eastern coastal regions are primarily served by imports that enter via Mombasa, rather than by
trade of domestically produced maize, so that the net maize transport costs to Mombasa are
lower for international exporters than those to more centrally located cities.

Figures 6-9 show the POE maize prices for each country, plotted against global maize prices
over the study period. In each figure, global prices in month t are converted to local, nominal
currency units using the exchange rate in month t. While intra-annual seasonality related to the
harvest cycle is clearly visible in each graph, the long-run trajectory of prices in each market
tracks the major shifts in global prices.

Sub-national fuel price data

Limited data availability requires that we use retail prices of petrol at sub-national markets in the
study countries. Data on prices and on the exact dates of all price changes in Ethiopia were
provided by the Ethiopia Ministry of Trade. Petrol prices in Uganda are monthly averages
provided by the Consumer Price Index division of the Uganda Bureau of Statistics. Data for

14
  Nominal prices are deflated using the World CPI from the IMF International Financial Statistics database.
15
  Price series for Uganda are less complete than those for other markets, with very few observations after 2008 and
spells of missing data in other years. However, we have numerous alternative sources of prices for Uganda that
overlap the RATIN price series in some periods, including FEWS, Uganda FoodNet, and the FAO. To
accommodate missing values in the Uganda RATIN series, we predict prices using least squares estimates based on
regressions of RATIN prices on these other price series, and on RATIN prices in non-study markets. Details are
available upon request.

                                                         9
Kenya were provided by the Kenya National Bureau of Statistics. Monthly average retail petrol
prices for Tanzania were provided by the Bank of Tanzania and IGC.16

Table 4 shows the average price of fuel at sub-national markets in the sample data, along with
the number and percentage of study months in which the price in each market was the lowest in
the country. For an imported good such as petrol, we expect prices to generally be lower at the
port-of-entry than elsewhere in the country. In Ethiopia, Kenya, and Tanzania, the fuel prices in
the POE market are indeed the lowest on average, in line with expectations. In Kenya and
Tanzania, the POE price is also most frequently the lowest among national market prices over
the study period. The Mombasa price is the lowest price in 40.4% of months, and the Dar es
Salaam price is the lowest price in 61.7% of months. In Uganda, retail fuel prices in the POE
Kampala are systematically higher than in Mbale, which is near the Kenya border. It is possible
that some fuel imports are diverted directly to Mbale and other border markets without first
passing through Kampala.17

In Ethiopia, although the price of petrol in the POE Addis Ababa is lowest on average, prices are
lowest in Dire Dawa over 70% of the time. There is a clear temporal pattern underlying these
data. Dire Dawa is the low-price market in almost every month from July 2000 to July 2009;
from August 2009 onwards the fuel price in Addis Ababa has been lowest in all but 6 months.
While this likely reflects some change in the underlying formula used to calculate market-
specific prices, it may be that fuel supplies which enter via Djibouti, go directly to Dire Dawa,
which is closer to the border, without first passing through Addis Ababa.

Figures 10-13 show the time path of POE fuel prices along with global oil prices. Visual
inspection suggests that the each pair of series closely co-move, with changes in the POE price
tending to lag global price changes. Infrequent updating of the Addis Ababa petrol price,
especially in the early years of the data set, is clear in Figure 10.

4. Empirical Framework

Our empirical approach rests on the identifying assumption that prices are transmitted from
global markets to local markets via the port-of-entry. For fuel prices this is intuitive, because oil
imports are channeled through the POE. Likewise, although domestic maize supply shocks are
realized in the breadbasket regions, domestic maize prices are tethered to world markets via the
primary hub of international trade. To best reflect this system of price transmission, we estimate
a sequence of models in which at each step of the procedure we treat the larger market (global in
one stage, POE in the next) price as weakly exogenous to changes in the smaller market price.
By separately estimating the link between the POE price and the global price in a first stage, we
allow for country-specific tariffs and import policies. Second-stage equations linking the POE
16
   Although most lorries use diesel fuel, we use petrol prices because they are available for a greater number of
markets. This is of minimal concern: level differences are absorbed by constants in the estimated models, and the
two price series are very highly correlated in those markets for which we have both.
17
   To accommodate this possibility, we re-estimated models for Uganda treating Mbale as the port-of-entry, rather
than Kampala. Results were essentially unchanged. Therefore, we only report results that treat Kampala as the port
of entry, since it is by far the largest market and was designated as the port of entry by business owners and
economists in Kampala with whom we discussed the project.

                                                       10
price to each sub-national market allow for distance, infrastructural differences – for example,
mobile telephony entered Kenya much earlier and spread far more quickly than in Ethiopia, with
possible effects on price transmission rate (Aker and Mbiti 2010) – and possible local market
effects to differentially affect the rate at which global prices transmit within national markets.18

Figure 14 summarizes our empirical approach. We begin by testing for a stable long-run
relationship between the global prices of oil and maize (step 1). We then estimate the impact of
changes in global market prices of each commodity on the POE prices in each country,
separately, treating the global price as weakly exogenous (step 2). Changes in global oil prices
are included in the POE maize equations in order to allow for variable transport cost margins at
the border. We then estimate the link between POE petrol prices and petrol prices in
geographically dispersed sub-national markets (step 3). Finally, we estimate the pass-through
rates of maize prices from the POE markets to other sub-national maize markets, allowing
changes in local fuel prices to impact the maize price spread (step 4).19

Step 1. Global Oil - Global Maize Price Linkages

Both nominal oil prices and nominal maize prices are nonstationary, I(1) series, as established
using either the modified Dickey-Fuller test (Elliott et al. 1996) or the Phillips and Perron (1988)
test for stationarity. If there is a clear causal link between these series, it will be reflected by the
presence of a cointegration relationship. However, Johansen (1991) tests for the maximum
number of cointegrating vectors indicate that the two series are not cointegrated at conventional
levels of significance (results available upon request). This result does not change if we include
a trend in the cointegrating equation, suppress the constant in the cointegrating equation, or use
real 2005 prices rather than nominal prices. Using different, US price series, Zhang et al. (2009)
and Serra et al. (2011) similarly find no evidence of cointegration between monthly crude oil and
corn price series. Zhang et al. (2010) find precisely the same result using the same data series,
but with somewhat earlier dates.

This finding does not account for the ethanol mandate that took effect in October 2006, under the
United States Energy Policy Act of 2005, which may have fundamentally changed the
relationship between fossil fuel prices and maize prices. In Figure 5 it does appear that co-
movement between oil and maize prices is stronger after 2006 than before. Indeed, the
correlation between the real prices of oil and maize jumps to 0.74 from October 2006 to
November 2012, much higher than over the longer period. However, Johansen tests on data
from October 2006 onwards still do not show evidence of cointegration between the series. This

18
   For multiple reasons, we do not try to control for policy changes by using dummies for prospective structural
breaks. First, the time series are relatively short, and many policy changes (e.g., fuel price caps) were concentrated
near the very start or end of the time period we study, leaving few observations to identify any effect. Second, there
are few clear, discrete policy changes that can be confidently assigned to specific months and used as structural
breaks, as many policy instruments (e.g., export bans) have been episodically implemented and not formally
recorded in any sources we can identify. Third, many relevant policies are expressly endogenous to market
conditions; for example, export bans have been introduced occasionally explicitly in response to rising market
prices.
19
   Enders (2010) lays out the general approach to VAR and error correction models that we use in this paper.

                                                          11
result is consistent across specifications (including a trend, suppressing the constant), and holds
for both nominal and real prices (results available upon request).

Failure to demonstrate cointegration suggests that no clear, stationary long-run equilibrium
relationship has emerged between these global crude oil and maize spot market price series over
the last six years or over the past dozen years, even following a major policy shock that
seemingly linked the two more tightly. This is perhaps evident in Figure 15, which shows the
nominal prices of oil and maize from October 2006 to November 2012. While the prices appear
to follow similar trends, it is not apparent that one series regularly leads the other, nor that they
maintain some fixed additive or proportional relationship.

Therefore, in order to formally model the observed co-movement between global oil prices and
global maize prices without imposing an unsubstantiated long-run stationary relationship, we
estimate a reduced form vector autoregression (VAR), in first differences, separately for the
entire sample and for the period from October 2006 onwards. A lag length of one month was
selected for both specifications, based on both Akaike and Schwarz-Bayesian information
criteria.20

Step 2. Global-POE Price Linkages

For all four countries, Johansen tests indicate the presence of a single cointegrating vector
between global oil prices, POE fuel prices, and the exchange rate, with a constant in the long-run
equation.21 Therefore, for each country we test a variety of fuel price specifications (varying the
lag length K and the inclusion of a trend in each equation) based on the following two-stage
asymmetric error-correction model (ECM):

     (1)

     (2) ∆                                         ∆        ∑            ∆              ∆              ∆
               ∆

where         is the POE fuel price in month t, and    is the global oil price,     is the US dollar
exchange rate (local currency over USD),          is the consumer price index, and and             are
statistical error terms. Equation 1 represents the co-integrating vector, i.e., the long run
equilibrium relationship between the variables. Equation 2 captures the short-run dynamics. The
error correction term,                                                     , is the residual from
equation 1, which measures period t-1 deviations from the long run stationary relationship. The
neg and pos superscripts indicate negative and positive first-stage residuals, respectively (i.e., the
variables                   if         0, = 0 otherwise, and complementarily for               ). The
asymmetric structure is important. The      and parameters can be interpreted as the speed-of-
adjustment parameters for negative and positive deviations from the long-run equilibrium,
respectively. We expect those parameter estimates to be negative. The absolute value of the
20
   The residuals from the two equations look like white noise, and we cannot reject the null hypothesis of a unit root
in either residual series.
21
   Johansen tests are based on maximum likelihood estimation of the three-equation VAR system, in levels. These
and all other results not provided in the main body of the paper are presented in Appendix C.

                                                          12
reciprocals of those rate of decay estimates, |1/ | and |1/ |, provide an estimate of the speed of
adjustment back to long-run equilibrium, measured in months.

Asymmetries in adjustment to long-run equilibrium may arise from firm-level market power,
government policy interventions, or infrastructural or institutional bottlenecks that make
adjustment more difficult in one direction than in the other, for example if increasing imports
may be limited by port capacity (Meyer and von Cramon-Taubadel 2004). Because our interest
will be in estimating the pass-through effects of long-run price increases, the asymmetric
structure is important for ensuring that the “long run” is defined using the average response to
negative ECT terms (i.e., month in which the POE price is too low relative to its stationary
relationship with the global price).

We estimate a similar series of ECM models for maize. The primary modification is that we
include the global oil price in the maize ECM system, to allow for changes in fuel costs to
impact the relationship between POE maize prices and global maize prices:

       (3)

       (4) ∆                                          ∆          ∑        ∆             ∆          ∆
                  ∆                 ∆

where         is the POE maize price in month t,                     is the global maize price in month t, and
other variables are as before.

The relationship between domestic maize prices and global maize prices is complicated by the
twin facts that all of the study countries are major maize producers, and that they mainly produce
and consume white maize, an imperfect substitute for the yellow maize that prevails on global
markets. While domestic prices in these four countries are necessarily linked to global price
movements through near-constant cross-border trade, policymakers have supply-side tools for
stabilizing maize prices (e.g., export bans and input subsidy programs) not available to them for
influencing fuel prices. Indeed, a thread of the literature exploring the impacts of the 2008-2011
global food price shocks on local food economies emphasizes the extent to which national
governments were able to use a range of policy instruments to buffer their constituents against
price movements.22 Nonetheless, global maize prices are necessarily transmitted to domestic
maize prices in these small economies through ports-of-entry. Transport costs to and from the
POE, combined with transport costs to and from the global market, determine the profitability of
trade with international markets. In the absence of an export ban, the return to transporting maize
from the breadbasket region(s) to any other market is always constrained by the available return
from transporting the crop to the port for export and the cost of exporting the crop from that
country to prospective importers on the global market.

In treating (2) and (4) as single equation systems rather than vector error-correction models, we
implicitly treat the exchange rate as weakly exogenous. In Appendix B we provide evidence that
the exchange rates in Tanzania, Uganda, and Ethiopia do not respond in a statistically significant

22
     Baltzer (2013) offers a nice summary of that literature.

                                                            13
manner to deviations from the long run fuel and maize price equilibrium, and that the exchange
rate in Kenya does so with only marginal statistical significance.

Step 3. Fuel price transmission from POE to other sub-national markets

We expect fuel prices in sub-national markets other than the POE to reflect POE prices plus
domestic transport costs. Deviations from this relationship – due to supply chain disruptions,
localized fuel demand shocks related to seasonality, or other forces – should not persist for long
under reasonably competitive conditions. Not surprisingly, Johansen tests clearly indicate the
presence of a single cointegrating vector between the POE market price of fuel and the fuel price
in each non-POE market in the sample. In all cases the Schwarz Bayesian criterion indicates an
optimal lag length of two months in levels (therefore 1 month in differences). Accordingly, for
each POE/other market pair, we estimate the following ECM:

   (5)

   (6) ∆                                   ∆            ∆

where      is the fuel price in “other market” j, in month t, and all other variables are as described
above. For the within-country specifications we work entirely in nominal, local currency terms,
so there is no need to account for inflation or exchange rate changes in these equations. While
feedback from fuel prices in market j to POE fuel prices is possible, if localized demand shocks
change the relative returns to supplying fuel to the POE and market j over a short period of time,
we abstract from this possibility and treat POE fuel prices as weakly exogenous.

Step 4. Maize price transmission from POE to other sub-national markets

The final relationships of interest are those between POE maize prices and maize prices at sub-
national markets. Because crop transport in this region is conducted primarily via lorry, we use
an error correction framework that allows fuel prices to affect maize price spreads within each of
the study countries. As in the above sections, Johansen tests show that in all specifications there
is a single cointegrating vector between POE maize prices, other market maize prices, and other
market fuel prices, with an optimal lag of length of two months in price levels.

The error-correction framework takes the following form:

   (7)

   (8) ∆                                       ∆            ∆           ∆

where      is the price of maize in market j and all other variables are as before. The hypothesis
  :        0 captures the expected effect of fuel prices on long-run maize price spreads.

                                                   14
We estimate all of the equations in Steps 2-4 using two-stage OLS.

5. Results

Global Price Linkages

Table 5 shows the results of the reduced form VAR linking changes in oil and maize prices on
global markets. We show separate results for the periods Jan 1990 – Nov 2012 and October
2006 – Nov 2012 in case the change in US ethanol policy affects the underlying inter-commodity
price relationship. Coefficient estimates are generally similar over the two periods. In the more
recent period maize prices exhibit no response to lagged changes in oil prices. Over the full
sample, the point estimate of the effect of lagged oil prices on maize prices is negative and
statistically insignificantly different from zero. On the other hand, positive changes in maize
prices tend to drive up oil prices. This is consistent with previous findings by Serra et al. (2011)
that corn price shocks cause increases in ethanol prices, which in turn induce adjustments in
gasoline prices, which feedback to crude oil markets. While the estimates in Table 5 cannot be
interpreted as causal, they do suggest that we can reject a model in which global oil price
movements are absorbed directly into future maize price movements on the main international
market. This finding persists if we increase or decrease the number of lags, or if we estimate a
structural VAR in which we allow for contemporaneous impacts of oil prices on maize prices
(but not vice versa).

Because our interest lies in understanding the impact of exogenous oil price changes on maize
prices, we take this as evidence that no meaningful causal relationship can be identified for either
period. Although they are strongly positively correlated, oil price changes appear only to exert
little (and no statistically significantly) influence on global maize prices, calling into question
popular claims that global oil prices shocks trigger global maize market adjustments. Any effects
of oil price shocks on local maize prices must either occur through transport costs, or be merely
due to correlated global commodity price shocks due to common underlying factors, as other
recent studies have found (Gilbert 2010, Byrne et al. 2013).

Global-POE price transmission

Table 6 shows the estimates of equation 1, for all four countries. The final row of the table shows
the mean POE price over the study period, to aid in interpreting magnitudes. POE retail fuel
prices are increasing in both global oil prices and the exchange rate, as expected. Standard
hypothesis testing on these equations is not possible because the error term is nonstationary,
therefore we do not report statistical significance.

                                                15
The key findings in Table 6 are summarized in the average pass-through elasticities for oil price
changes and exchange rate changes, which are listed in the lower half of the table.23 Estimates of
POE petrol price elasticities with respect to the global oil price are remarkably similar across
countries. On average, a 1% increase in the price of oil on world markets leads to an increase in
the POE petrol price of 0.38-0.46%. Petrol price elasticities with respect to the exchange rate are
higher, ranging from 0.85 in Kenya to 1.52 in Ethiopia. These findings indicate that over the
study period, slightly less than half of the increase in nominal fuel prices in the POE markets is
due to changes in nominal prices of global oil. The remainder of the increase is driven by
exchange rate depreciation.

Table 7 shows the estimates of equation 2. All coefficient estimates have the expected sign,
when significant. Adjustment back to the long run equilibrium is not instantaneous, but is still
reasonably fast on average, ranging from 2-7 months. Increases in global oil prices transmit
faster than decreases. However, F-tests for asymmetric adjustment indicate that only in
Tanzania, where price increases transmit five times faster than decreases, is the difference
statistically significant. This is consistent with the existence of bottlenecks to importing
additional oil quickly, whether due to port or foreign exchange constraints or contracting lags, or
of imperfect competition with importers adjusting prices upwards faster than downwards.

Estimates of equation 3, the cointegrating vectors linking global maize prices to POE maize
prices, are given in Table 8.24 POE maize–global maize pass-through elasticities exhibit greater
heterogeneity than did the analogous POE petrol–global oil elasticities, ranging from 0.22 in
Kenya to 0.82 in Ethiopia. These results correspond with an ordering of the degree to which
central governments intervene in maize markets, as Kenya’s tariff and parastatal marketing board
policies translate into weaker transmission of global maize prices to the national market than in
neighboring countries. Pass-through elasticities of POE maize with respect to global oil prices
are also substantial, lying in the range 0.20-0.36, even after accounting for the direct impact of
maize price changes. In Kenya, by far the biggest maize importer in the region, a 1% increase in
global oil prices exhibits greater upward pressure on Mombasa POE maize prices than does a 1%
increase in global maize prices, underscoring the importance of transport costs to the pricing of
bulk grains. As with fuel prices, exchange rate elasticities vary widely, and are responsible for
any remaining changes in nominal POE maize prices after accounting for the direct impact of
global maize and global oil price changes.

In Figure 16 we plot the monthly average residuals from estimates of equation 3, normalized by
the overall mean POE maize price in each respective country. In each country, clear seasonal
patterns are apparent in the deviations from long-run equilibrium. These are consistent with
intra-annual fluctuations in domestic supply, due to the agricultural production cycle. For
example, maize harvests in Ethiopia are concentrated in the months September-November,

                                                                                                           ̅
23
     The average pass-through elasticity of price pj to price pi, designated ji, is calculated as ̂   ̅
                                                                                                               ,   where ̅ is
the average of price k over the observations used in the relevant regression, for     , , and is the coefficient on
price i in the relevant regression. In Table 6,  {Global oil price, Exchange rate}, and j is POE petrol.
24
  The constant is excluded from the Tanzania equation, in accordance with the results of a Johansen test for
cointegration.

                                                             16
which coincides with a drop in the Addis Ababa maize price vis-à-vis its long-run relationship to
the world price.

Second stage ECM results for the global-to-POE maize price relationship, based on equation 4,
are shown in Table 9. The error correction terms are highly significant in the wake of a positive
deviation from long-run equilibrium price (ECTneg), exhibiting the opposite pattern from the
asymmetric models of POE fuel prices. One interpretation of the asymmetric adjustment is that
price arbitrage via exports is logistically difficult due to port queues, regulatory barriers, the
absence of short-term forward contracting, and storage bottlenecks. On the other side, rapid
recovery from higher prices may reflect the roles of food aid, explicit export bans, and strategic
release of grain reserves in mitigating the pace of food price increases in the POE markets.
Higher-than-equilibrium POE maize prices never persist beyond the next harvest, disappearing in
5-6 months in each country. Lower-than-equilibrium prices persist far longer. However, only in
Ethiopia is the asymmetry statistically significant at the 10% level. Coefficients on lagged
differences in global oil prices are not significant in any of the equations, suggesting that changes
in transport costs matter more for the long-run equilibrium (Table 8) than for short-run price
dynamics.

Within-country petrol price transmission

Table 10 shows the estimated co-integrating vectors based on equation 5, which link fuel prices
in sub-national markets to the POE fuel price. Nearly all of the estimated models fit the data
almost perfectly. Fuel markets are very well integrated within the study countries. The β
coefficient estimates from equation 3 are all very close to unity, as are the estimated pass-
through elasticities, corresponding to the law of one price.

Second-stage ECM estimates, based on equation 6, are provided in Table 11. In all markets, POE
price increases transmit faster than POE price decreases. However, we can reject the null of
symmetric adjustment for only 5 of 13 markets. Faster pass-through of price increases could be
consistent with the existence of structural impediments to moving additional fuel quickly to non-
POE markets, or with imperfect competition among fuel distributors. Overall, equilibrium is
restored very rapidly when POE prices increase. Adjustment occurs within two months for all
markets in Ethiopia, Tanzania, and Uganda. In Kenya, adjustment speeds range from three
months in Eldoret/Nakuru to nine months in Kisumu.

Within-country maize price transmission

Table 12 shows the estimates of equation 7 for each of the sub-national markets. For markets in
Ethiopia and Kenya, as well as for Arusha, Dodoma, and Mbale, both the point estimates of
and the POE maize price pass-through elasticities are close to unity, indicating conformity with
the law of one price. Within-country maize price elasticities are lower for the other four markets
in Tanzania and Uganda – Kigoma, Mbeya, Gulu, and Mbarara. It is notable that these markets
are the greatest distance from the POE markets and the coast (Figure 4). It is also these four
markets that exhibit the largest positive pass-through elasticities with respect to local fuel prices,
ranging from 0.29 in Mbeya to 0.76 in Mbarara. In Mbarara the petrol price elasticity is higher
than the POE maize price elasticity, and in Gulu and Kigoma the petrol elasticity is

                                                 17
You can also read