1. Introduction

Although the poverty-reducing potential of growth in agricultural productivity has long been recognized, the extent of the impact remains disputed in the development literature ever since the advent of the Asian green revolution (e.g., Freebain, 1995). Other things being equal, higher yields increase output and farm income, thereby reducing poverty. However, the issue is more complicated as there are a broader spectrum of indirect mechanisms by which higher yield impact on poverty. In addition to direct effects on output and farm incomes, higher productivity impacts on the poor indirectly through food grain prices, labour markets and farm-non-farm linkages. Indeed the magnitude of these indirect effects is one of the contentious issues between the defenders and opponents of the Asian green revolution, and empirical studies rarely quantify the magnitudes of these indirect effects. Also, the ultimate impact on poverty depends on local terrain, and it is not clear to what extent the Asian green revolution can be replicated to Africa.

The recent effort by donors and African governments in reducing poverty in the 1990s has re-ignited the debate about the role of agriculture (see DfID 2005, policy paper on agriculture for poverty reduction). Several African governments (e.g., Ethiopia) have long adopted strategies to reduce poverty through agricultural growth, including the spread of improved varieties. These strategies have often been critisized (especially in Ethiopia) for failing to reduce poverty, particularly urban poverty. On other hand some

¹ This study is part of an ongoing research project "Agricultural Strategies for Chronic Poverty Reduction" funded by ODI/USAID (No. CPR003), which investigates the role of agriculutre in reducing poverty focussing in three African countries, Ethiopia, Uganda and Zimbabwe.

² Suleiman and Mosley: University of Sheffield, 9 Mappin Street, Sheffield S1 4DT, UK. Mosley: p.mosley@sheffield.ac.uk, tel: 44(0)114-222-3397; Suleiman (corresponding author): a.suleiman@sheffield.ac.uk tel: 44(0)114-222-3414. We are grateful to Bereket Kebede and Stefan Dercon for kindly providing their set of data and variables from ERHS and the poverty estimates used in this paper. We also acknowledge the support of Andrew McKay at various stages of this research.

studies suggest that there has been a decrease in overall poverty over the last decades though inequality has increased (Bigsten et al, 2003). However, it is not clear how and why poverty has reduced, if it did, particularly the role of agriculture in that process.

Recent research on Ethiopian economy has focused on the sources of agricultural productivity (e.g., Abrar et al, 2004; Croppenstedt et al, 2003), or on measuring extent and profile of poverty (e.g., Bigsten et al, 2003; Dercon and Krishnan, 2003). Generally, the poverty impacts of agricultural policies are not systematically investigated in Africa, and more often than not, the link between higher productivity and poverty is simply assumed, rather than empirically established. A number of recent contributions, using a global cross-country data, find a significant correlation between growth of agricultural productivity and poverty (e.g., Irz et al, 2002; Mosley and Suleiman, 2004, 2006), but country specific studies are rare in the context of Africa.

This paper primarily seeks to assess the poverty-reducing impacts of increased productivity in agricultural food crops in three African countries, Ethiopia, Uganda and Zimbabwe, as may result from the adoption of new varieties. A multi-market model with significant disaggregation of food crops is used to estimate the likely impacts of productivity directly on farm incomes, as well as through indirect routes, notably price and labour markets.

As in many African countries, food crop producers are disproportionately represented among the poor, hence extent to which improved varieties and other practices that increase food productivity impact on poverty largely depends on whether or not they reach the poorest. We seek to quantify the likely impacts of productivity in key food crops if the benefits were to reach the poor in the same proportion as other groups. Of still greater relevance here is spatial variation in consumption patterns, not only among food crop producers but also consumers in both rural and urban areas. In the case of urban poor, the effect coming from prices is of particular importance. Based on the estimated price and income changes and information about historical changes in productivity, household survey data are then used to predict the likely impacts on household groups distinguished by regions of residence (distinguishing between rural and urban households) and poverty status.

Section 2 explains the multi-market model used to estimate the price effects associated with increased productivity, and presents the initial estimates for the case of maize. The nature of data and the characterization of household groups considered in this paper is presented in Section 3. Section 4 then considers the direct and indirect effects of increased productivity on the different household groups, as well as the poverty effects at the household level.

2. The multi-market model and initial estimates

The effects of increased productivity in food crop production affects producers of these crops directly (output effect), and affects producers and consumers indirectly through changes in prices of these and other commodities resulting from the productivity increases. The direct effects can be estimated directly from survey data given information about the level of production of each commodity and its relative importance as a source of income). But the indirect effects are general equilibrium in nature, and are much more difficult to quantify, and the results of will be sensitive to the assumptions underlying the model. But these effects can be quite large in practice and can not be ignored.

What is critically important is whether commodities are modelled as tradables or non-tradables. Changes in productivity of tradable good will only affect external transactions. The prices of these commodities will not be affected if domestic produced commodities can be regarded as perfect substitutes for improved variety of the same of commodity, and if producers are indifferent between domestic sales and export. But the indirect effects from a non-tradable good require a general equilibrium framework. Increased productivity of a non-tradable commodity will tend to drive its price down, other things being equal, though in reality ceteris paribus does not apply as increased productivity also affects incomes, as well as the prices of other commodities that are substitutes/complements in production or consumption.

Figure 1: Conceptual Framework: Link from reversible technology adoption to poverty

Crop-specific Improved variety, or other modern

Ideally, such effects can be investigated using a full-fledged CGE models. But the SAM for Ethiopia, itself a recent one, is very limited in its disaggregation of the agricultural sector and is not suitable for the prupose at hand. Thus, we construct a multi-market model focusing on the agricultural sector to determine the indirect (first price) effects. The model is set out based on the summary by Sadoulet and de Janvry (1995). As regards the specific model to be developed for Ethiopia and Uganda, we closely follow the multi-market model developed by Mckay and Mosley (2001) for Uganda, where: ƒ The equations are log-linearized (formed in percentage forms), and solved by matrix

inversion. ƒ Factor prices are fixed (most are imported hence, little effect on prices), thus factor

markets will have no impact on solution. ƒ Producer and consumer prices same ƒ Our interest is in the repsponse of prices of non-tradables and net imports of food

crops to a separate (autonomous) increases in productivity of specific crops, hence (reversible) technology is specified in the respective product supply equations as percentage increase in production due to improved technology, ceteris paribus

There are several issues related to specification and solution of the model. First is the choice commodities to be modeled and assumptions of whether or not they are tradable/non-tradable. Given the focus of this study and the characteristics of the Ethiopian agriculture, the commodities and their assumed characteristics are set out in Table 1.

Table 1 about here

The log-linearized form of the equations of the Ethiopian model are represented as:

,

_D dy

+

_C dp

py

si

⎟⎟

i si

⎜⎜

di

dq

dp

⎞

si

si

dq

=

di

dq

si

q

⎛

=

di

q di

q

s

dq

s

q

=

dq^d

d

dq

py

i

q p

dz

B

+

3. Data and pattern of agricultural production in Ethiopia

The data we use is the Ethiopian Rural Household Survey (ERHS), a nation-wide panel data of rural households conducted in six waves during 1994-2004. The survey was undertaken in 15 villages across the country from which nearly 1500 households were selected randomly. The villages were deliberately selected to account for the diversity of agro-climatic and farming systems in the country, and the two major farming systems and technologies (grain-plough and enset³-hoe systems) are fairly represented. For this study, we use data from the 1994 wave.

³ Enset is a perennial and major staple for an estimated 15-20 percent of the population, mostly in South.

There are several reasons for using this Survey for such analysis, despite the fact that it is not nationally representative. First, this is the only multi-purpose (integrated) household survey, collecting information not only on different characteristics but also detailed in terms of the consumption and production activities and incomes sources. Thus, it enables a detailed characterization of household groups, in particular focusing on the extent to which they produce and consume major food commodities, which is crucial for such analysis. Particularly, the data permits to simulate poverty effects at the level of individual households, thus avoiding the need to link to agricultural surveys, unlike other survey such as the HICE.

Secondly, there have been numerous studies on poverty in Ethiopia using this survey than any other, hence easier to obtain some important estimates from the literature, and from other researchers, particularly on poverty. Among other things, a consumption-based standard of living measure was obtained from other researchers, and this was used to determine the poverty line. This defines standard of living as household consumption per capita, adjusted for temporal and spatial variations in the cost of living (see Bigsten, et al 2003; Dercon, __, for further details, as well as a thorough investigation of consumption/income poverty and inequality in Ethiopia based on this measure). However, given the small number of housholds, care need to be taken in interpretating the results.

In light of the significant variation in production and consumption patterns observed in Ethiopia, and also given our focus on poverty, we have disaggregated househods by income group and locality. Variation in production is taken into account by disaggregatng houshelods into three agro-climatic zones, North, Central and South. The diversity of standard of living is also taken into account, by disaggregating households into four quartile groups based on the consumption-based standard of living measure mentioned above, with all those in the lowest two quartiles and some of the least well-off in the third quartile (about 5% of sample) are identified as poor relative to the poverty line.

Table 3 reports the distribution of households by living standard quartile and agro-zone.

It is evident from the table that poverty is disproportionately high in the Southern region, followed by the North. In other words, nearly 40% of the poor people live in the South. However, it should be noted that this is partly due to the high population density in the region, hence the importance of adjusting poverty measures for adult equivalent units. There is a diversity of living standards within each locality, with more than half of the households in the Cereal areas belonging to the two upper quartiles, and nearly a third in the richest quartile. This is in sharp contrast with the Sourthern region only 20% of the households in the richest quartile. Note that inequality seems to be the lowest in the Northern region.

Table 3 about here

Of greater importance for this study is variation in production pattern, which is summarized in Table 4, reporting the share of major commodities in agricultural revenue. The commodity groups are delibrately chosen to identify the food crops for which the effects of productivity are to be considered.

Table 4 about here

The first panel of Table 4 relates to the national pattern. As expected, maize and teff account for nearly a third of total crop revenue. In fact, these two crops account for a much higher proportion of the revenue of the poorest quartile (about 50%), these then are crops of disproportionate importance to the poorest groups. What is more, this is the case for all regions except the North, with the proportion in the Central as high as 70%. In addition, the revenue shares for these crops monotonically decline with the standard of living quartile in all regions, indicating that they are more important source of crop revenue for the poor, although in absolute terms, the higher quartiles still probably produce more. A peculiar feature of maize is that even in the Northern region where the share of maize is a mere 5%, farmers in the poorest quartile produce nearly 80% percent of maize. Wheat on the other hand is the only crop with its share of revenue consistently increasing with quartiles, hence least important for poorest farmers.

The relative importance of the different crops to the poorest households varies from region to region. In this region, unlike the Central and South, Barley and Sorghum are major sources of crop revenue (60%), accounting for nearly 65% percent of revenue for the poorest quartile. Sorghum particularly is more important for the poorest than the higher quartiles, which is not the case for barley. Farmers in the cereal regions (particularly the Central zone) are generally more diversified in terms of crop production, whereas there is little or no production of wheat, barley and sorghum in the South. Instead farmers, including the poorest, derive a significant share of revenue (just over half) from coffee and other cash crops.

On the other hand, farmers in the Northern region, particularly the poorest, seem to derive a significant proportion of their agricultural income from livestock and other off-farm activities (see Table 6 and 7, to be reported later). In general, it is evidient that the distributional effects of increased productivity would vary from region to region, with maize (and teff and sorghum to some extent) seemingly resultingly the greatest impact on poverty.

4. Simulation Results of a 30% increase in the productivity of maize

Two types of simulations are considered: ƒ Impact on agricultural revenue of households groups by income quartile ƒ Poverty impacts at the household level.

5. Conclusion and Extensions

Therefore, the results generally show that:

To be done later:

ƒ Simulations for other crops (sorghum and wheat?) ƒ include urban sector ƒ sensitivity analysis for elasticites ƒ estimation for Uganda Future extensions: ƒ Incorporate consumption consequenses of productivity increases ƒ Labour market and other non-farm linkages