Distortion of agricultural incentives in East Africa: effects on agricultural value added

Abstract

This study examines the effects of distortion of agricultural price incentives on agricultural value added in East Africa. The World Bank, IFPRI, FAO, and CSP are the sources of data. The dataset ranges from 1981 to 2018, and the error-corrected LSDV model is used to analyze the data. The results indicate that agricultural price incentives have positive and significant effects on agricultural value-added. Aggregate nominal assistance coefficient, exportable agricultural products nominal assistance coefficient, and nominal rate of protection have increased agricultural value-added significantly. Agricultural price incentives targeting different levels of value addition have larger effects than those targeting aggregate outputs. This implies that agricultural incentive policies and market conditions in support of local producers are vital to enhancing AVA in East Africa. Besides, larger areas of arable land, lower agricultural employment, a smaller population size, a larger GDP, less spending on education, and a better-performing polity contribute to a significant increase in the regional agricultural value added. The results generally imply that agricultural price incentives are vital to accelerating agricultural value addition in East Africa. Governments in this region should thus consider revising agricultural policies in a pro-agricultural way to further accelerate regional growth in agricultural value-added. Enhancing agricultural price support needs to be a crucial element of policy revisions in the region.

Impact Statement

In East Africa, agriculture is the main source of employment for a large section of the population. However, agricultural incentives have been reportedly distorted against agriculture, and sectoral income has been low. Consequently, farmers’ income from agricultural value addition has been low. This study reports the effects of the distortion of agricultural incentives on agricultural value added in East Africa. The study shows that favorable agricultural incentives enhance agricultural value-added. The findings have strong implications for the region’s smallholders, who are the subject of heavy taxation, either directly or indirectly. It will have far-reaching consequences for the poor, who rely on agriculture for a living. In particular, the findings influence regional anti-agricultural policy design, which is vital for the regional goal of achieving inclusive growth and structural transformation.

Keywords:

• Agricultural protection
• price incentives
• price distortion
• NAC
• NRP
• LSDVC

REVIEWING EDITOR:

• Aye Goodness, University of Agriculture, Benue, Nigeria

SUBJECTS:

• Development Studies
• Development Policy
• Politics & Development

1. Introduction

In Africa, agriculture has continued to play a strategic role in the process of economic development. Despite its dominant role, public spending in this sector appears inadequate on this continent when compared to the wealthiest countries (World Bank, 2022). The Anderson & Masters (2009) report shows that during the 1960s and 1970s, many African countries implemented pro-urban, anti-agricultural, and anti-trade policies, while many high-income countries restricted agricultural imports and subsidized their farmers. The economic policy landscape of developing countries has long included distorted trade incentives. In most of these countries, the provision of agricultural incentives has continued to be unfavorable (Baliño et al., 2019; Mukasa et al., 2020), though progress has been observed in recent years (Mukasa et al., 2020).

Theories note that government intervention distorts agricultural markets (Jetha, 1982). Effective agricultural protection policies are based on what is politically acceptable (Masters, 1993), while their agricultural growth implications deviate from the predetermined outcomes. Kydland & Prescott (1982) argue that output growth variation is an effective reaction to external changes in the actual economic environment. They stress that intervention through fiscal or monetary policy to smooth growth fluctuations would be less effective, particularly in the short run. An integrated model of fiscal policy instruments notes that raising taxes is beneficial only if public spending is effectively directed to productive sectors (Bénabou, 1996). Otherwise, it would hamper growth. This is well explained in the works of Grossman & Kim (1997); Alesina & Perotti (1996); Fay (1993). According to Stiglitz (1987), agricultural taxes and subsidies occur when governments try to control the sector. However, while subsidies would enhance agricultural production (Seitov, 2023), taxing the sector would be economically unprofitable and destructive, particularly in low-income countries.

In Africa, to handle the expanding continental food demand, several continental frameworks and declarations have been initiated in the past ten years. Some of the key areas include removing unfair trade obstacles to encourage sustainable agricultural growth, ensuring food security, and eliminating poverty and hunger (AU, 2015; UN, 2023). Nevertheless, the majority of the nations’ policies remained anti-agricultural, especially the agricultural policies of the East African countries (Mukasa et al., 2020; World Bank, 2023). For most of the impoverished in this region, agriculture is the primary source of employment. Nevertheless, the distortion of agricultural incentives persisted until recently (Mukasa et al., 2020), and it is still unclear how such distorted policies would affect regional agricultural value added (AVA). Related studies note that agricultural protection influences crop production, livestock production, and household incomes positively (Dastagiri & Vajrala, 2018; Hemming et al., 2018; Petreski, 2014; Swinnen, 2021). Bollman & Ferguson (2018) argue that subsidies increase farm income. They note that the reductions in subsidies result in lower farm value-added and farm asset values. Seitov (2023) similarly argues that subsidies safeguard food security. Nevertheless, according to Omeje et al. (2019), a high level of agricultural protection has a negative effect on the growth of agricultural outputs. In this line, Chandel et al. (2019) argue that subsidies only serve short-term objectives and would have adverse effects on economic efficiency in the long run.

Some studies generally advocate for agricultural protection (Bollman & Ferguson, 2018; Seitov, 2023; Ye et al., 2023), and others oppose such incentives for increased agricultural growth (Chandel et al., 2019; Heyl et al., 2022; Omeje et al., 2019; Swinnen, 2021). Numerous studies that employ subsidies (Bollman & Ferguson, 2018; Seitov, 2023; Ye et al., 2023) paid little attention to how complementary tax and subsidy policies affect agricultural growth. For those studies that utilize complementing tax and subsidy programs, the outcomes are incongruous (Chandel et al., 2019; Heyl et al., 2022; Omeje et al., 2019; Swinnen, 2021). None of them, however, attempt to examine the effect of distorted agricultural price incentives on AVA. To contribute to filling this gap, the current study explores the effects of distortion of agricultural incentives on AVA in East Africa.

Using alternative cross-sectional time series of distorted regional agricultural policy datasets, namely the nominal assistance coefficient (NAC) and the nominal rate of protection (NRP), the study provides novel insights into the role that distorted agricultural price incentives play in agricultural value addition. These measures, of course, account for distortions of sectoral price incentives, which provide unique insights in this area, unlike many previous related studies that rely exclusively on either subsidies or tax measures (Bollman & Ferguson, 2018; Seitov, 2023; Ye et al., 2023). The findings thus extend the existing knowledge in the political economy literature by providing region-specific, robust findings that use alternative agricultural incentive measures. Furthermore, the positive response of agricultural value addition to rising food prices would have useful implications for food production. In this line, the findings provide insight into Malthusian fears about the future prospects of the food supply.

2. Literature review

Growth theories presume that advances in productivity and technology are what drive output expansion (Harris, 2017). Nevertheless, mainstream growth models are widely criticized for their failure to include political economy perspectives in economic analysis. Stiglitz (1987) notes the importance of understanding the political economy in the process of output growth. Politicians determine the best level of intervention based on their political interests (Jetha, 1982). Effective agricultural protection policies are based on what is politically acceptable (Masters, 1993), while their growth implications would deviate from the predetermined outcomes. For instance, in developing countries, government intervention in the agricultural sector through subsidies or taxes influences the choice of production resources, including the uptake of farm technologies (Hemming et al., 2018). In this sector, as the process of value addition becomes complex or as the sector loses comparative advantage, the need for protection increases (Olper et al., 2013). Countries with lower agricultural protection earn low incomes from this sector. Farm incentives facilitate farmers’ assimilation of novel technologies, a crucial factor in advancing agricultural production and value addition. It augments or impedes value-added pursuits therein.

Empirically, studies note that distortionary agricultural policies influence agricultural growth either in a negative or positive way. Using a quasi-natural experiment of China’s maize purchasing and storage policy reform, Ye et al. (2023) examine the impact of agricultural subsidy market-oriented reform on agricultural green development using the difference-in-difference model. The study showed that maize purchasing and storage policy reform promoted the improvement of the green total factor productivity of maize with a hysteresis effect. Similarly, a study by Seitov (2023) indicates that in India, subsidies safeguard agricultural food security, although a substantial portion of them support farmers in wealthier regions, distorting interstate agricultural growth. Targeted subsidies are noted to be important for long-term sustainable development. In this study, the direct effect of subsidies on agriculture is positive, although it was argued to be non-sustainable.

Aiming at examining agricultural subsidies in the EU, a study by Heyl et al. (2022) notes that agricultural subsidies need to be substantially downscaled and implemented as complementary instruments along with other policy instruments such as quantity control. A review of livestock production systems, subsidies, and their implications by Chandel et al. (2019) indicates that subsidies serve short-term objectives, stating that they would have adverse effects on economic efficiency in the long run. Bollman & Ferguson (2018) estimate the impact of removing export subsidies on the local economies of Alberta, Saskatchewan, and Manitoba using a generalized difference-in-difference analytical method. In this study, the loss of the subsidy resulted in significantly lower farm value-added and farm asset values.

Hemming et al. (2018) argue that fertilizer and seed subsidies have positive effects on consumer welfare and overall economic growth. Agricultural input subsidies increase the use of farm inputs and thus enhance agricultural yields. However, subsidy schemes are prone to inefficiency, bias, and corruption. According to Omeje et al. (2019), protection policy instruments such as subsidies and direct transfers have been widely used to enhance agricultural production. However, such distortionary policies would reduce a society’s welfare by causing inflation (Swinnen, 2021).

According to Asano & Kosaka (2017), partial tariff reduction policies tend to make production and consumption activities inefficient. The study notes that a complete removal of the production subsidy has no effect on changing the negative welfare effects of tariff policy in Japan. According to Dastagiri & Vajrala (2018), whenever agricultural and food policies are in place, redistributive effects are always expected. Redistribution through supporting producers would raise inflation, particularly when it increases aggregate demand. Controlling for such types of adverse effects, Dastagiri & Vajrala (2018) argue that in developing countries, stimulus packages are vital policy options for agricultural development.

Malan (2015) examines the determinants and effects of agricultural price distortions in Africa. By applying a linear panel model to agricultural distortion data obtained from 22 African countries, the study notes that the nominal rate of assistance (NRA) for agriculture does not have a significant effect on total agricultural production. It affects cocoa and cotton yields significantly. The study further argues that a more democratic country tends to have a more effective policy, and thus the effect of the NRA on yield tends to be stronger in more democratic countries. When democracy is controlled, the NRA has a significant and negative effect on wheat. The NRAs are positive for the traditional cash crops that are exportable and negative for the import-competing food crops.

According to Riesgo et al. (2016), in sub-Saharan Africa (SSA), fertilizer subsidies are among the most common and politically sensitive policies. Mukasa et al. (2020) note that the amount of taxes paid by African farmers has significantly decreased during the past 40 years. Yet, this has not been buoyant in some countries. For instance, in Ethiopia, Hassen (2016) argues that agricultural income tax and land use fees are not buoyant. He underscores that the growth of the Ethiopian agricultural sector has no significant relationship with agricultural income tax resilience, highlighting the importance of further study on agricultural protection dynamics in relation to agricultural growth. Examining the impacts of different subsidy programs on rice production in Ghana, Badu & Lee (2020) note that fertilizer is more effective in rice production after subsidies.

Examining the impacts of removing fertilizer subsidies on production, prices, income, and consumption, as well as fertilizer demand in Lesotho, Ratii (2016) notes that decreasing fertilizer subsidies reduces crop production and increases livestock production. The decrease in subsidies increases urban households’ income while reducing rural households’ income. This policy also increases the prices of internationally non-tradable products, while it does not influence the prices of internationally tradable goods. Consumption demand for tradable products increased, while consumption of non-tradable products decreased. The study used a multi-market model based on secondary data obtained in 2012/13.

A study by Omeje et al. (2019) notes that agricultural protection has a negative effect on agricultural growth in Nigeria. The study investigates how macroeconomic factors, including agricultural protection, affect agricultural growth in Nigeria. It employs a multiple regression model and Granger causality tests on a dataset spanning between 1980 and 2016, and the authors argue against protected agriculture in favor of liberalized agriculture for sustained output growth. This clearly contradicts the evidence in much political economy literature. For instance, Vincent & Lee (1997) note that farm incomes of cereal producers increase by more than 40% due to protection rendered to agriculture in Korea.

The other strands of literature considered for this study deal with factors influencing AVA other than agricultural price incentives. Nugroho et al. (2022) explore the impacts of exchange rates, foreign direct investment (FDI) inflows, total agricultural export values, agricultural import duties, and fertilizer imports on AVA in developing countries. Using a fixed effect (FE) model on panel data obtained from 17 developing countries during 2006–2018, the study showed that FDI inflows and agricultural export values increase AVA in developing countries.

Onoja et al. (2017) examined the effects of trade openness, electricity consumption, education, and technology on agricultural value addition growth in Africa. Using an autoregressive distributed lag model on a dataset spanning from 1971 to 2011, the existence of a steady-state long-run relationship between agricultural value addition and education, trade openness, electricity consumption, and technology was additionally investigated. The study showed that technology and electricity consumption are the long-run determinants of the growth of agricultural value addition. In the short run, education, technology, and electricity consumption explained the variation of agricultural value addition in Africa.

Muyanga & Jayne (2012) investigated the implications of increasing population density in Kenya’s rural areas on smallholder production and commercialization. Employing a correlated random effects (RE) estimator on a dataset obtained from surveys of panel data spanning from 1997 to 2010, they show that a rising population density is associated with a decline in farm productivity. They also note that smaller farm sizes reduce the potential to produce surpluses, limiting demand for purchased inputs and new technologies, which would further discourage AVA.

Ben Jebli & Ben Youssef (2017) examined short- and long-run relationships between per capita carbon dioxide emissions, real gross domestic product (GDP), renewable and non-renewable energy consumption, trade openness ratio, and AVA in Tunisia. The study applies the vector error correction model and Granger causality tests to a dataset spanning from 1980 to 2011. Cointegration was observed between GDP and AVA, where short-run unidirectional causalities were running from GDP to AVA. In the long run, bi-directional causalities were observed.

Badri et al. (2017) explore the effects of human development on the value added of the agriculture sector in selected developing countries. Applying the OLS model to a panel dataset spanning from 2006 to 2014, they show that human development has a positive and significant effect on AVA. They further note that education improves labor productivity by allowing one to understand, predict, recognize, and address business needs, which enhances participation in agricultural value addition. The study also shows that increased investment in physical capital leads to a higher capital stock in agriculture, which supports long-term growth in AVA.

Allcott et al. (2006) regarded GDP as a crucial component of agricultural income. Using a dataset of rural public expenditures in a panel of Latin American economies, the study notes that non-social subsidies reduce agricultural GDP. They further note that political and institutional factors are important in dictating the size and structure of rural public expenditures, through which they have large effects on agricultural GDP. According to Hendricks et al. (2023), policy changes in Sub-Saharan Africa are driven in part by external political shocks. Agricultural policies subjected to such external shocks are likely to influence agricultural production, which in turn would influence value addition in this sector. In general, although a majority of the reviewed studies present a number of factors that have potential effects on the growth of AVA, none of them attempt to quantify the implications of agricultural price distortion for AVA. This study thus primarily attempts to contribute towards filling this gap.

3. Methods

3.1. Data sources and description

This study was conducted in East Africa. Based on the availability of a number of aggregate agricultural market distortion statistics, seven countries, namely Burundi, Ethiopia, Kenya, Rwanda, Sudan, Tanzania, and Uganda, were purposefully selected for the study. The economy of the region is largely dominated by agriculture and has been flourishing to support continental free trade to make mutual trade cheaper and quicker (UNCTAD, 2022), motivating the choice of this region as a unit of analysis. The study uses secondary data spanning from 1981 to 2018. The choice of this timeframe is guided by the availability of the time series of the distorted agricultural price incentive dataset. We use the NAC and NRP as measures of distorted agricultural price incentives. The NAC’s data covers the years 1981 to 2010, whereas the NRP’s data covers the years 2005 to 2018. The AVA data for each of the covered years is then obtained.

The NAC is the ratio of domestic prices that producers effectively receive given the distortions in country i at year t ($P_{\mathit{\text{it}}}^e$) and the undistorted border price ($P_{\mathit{\text{it}}}$) (Hendricks et al., 2023). That is $\mathit{\text{NAC}}=P_{\mathit{\text{it}}}^e/P_{\mathit{\text{it}}}.$ A NAC less than one represents an anti-agricultural bias, and a NAC greater than one represents a pro-agricultural bias. We compute the NAC from the NRA. According to Hendricks et al. (2023), the NAC is simply a NRA + 1. The NRA data is extracted from the World Bank’s updated estimates of distortions to agricultural incentives by Anderson & Nelgen (2013). Limited by the availability of non-tradable and importable products in NRA data for some of the considered countries, we rely on tradable (specifically, exportable) products NAC¹ along with the overall NAC in our model estimation. Of course, the recent findings of Hendricks et al. (2023) note the importance of exportable products NAC for agricultural productivity growth in SSA. Another measure of distorted agricultural incentives in this study is the NRP. The NRP is a proportional difference between producer prices and border prices adjusted for distribution, storage, transport, and other marketing costs (Tokgoz et al., 2016) (Equation 1(1) $$\mathit{\text{NRP}}=\frac{(\mathit{\text{Domestic price}}-\mathit{\text{Reference price}})}{\mathit{\text{Refernce price}}}\mathrm{*}100\mathrm{\%}$$(1) ). It compares the producer price of a locally produced good with a similar internationally traded good. That is, (1) $$\mathit{\text{NRP}}=\frac{(\mathit{\text{Domestic price}}-\mathit{\text{Reference price}})}{\mathit{\text{Refernce price}}}\mathrm{*}100\mathrm{\%}$$(1)

The price gap between the domestic and international prices is considered at two points on the commodity value chain. The initial point of competition for importable goods is at the wholesale level, while for exportable goods, it is at the exit border. A positive price gap results when the domestic price exceeds the reference price, indicating that domestic policies and market conditions support local producers. If a negative price gap emerges, it is an indication that domestic policies and market conditions penalize local producers.

In the econometric specification, we also include additional control variables that are likely to affect AVA, as suggested by many prior studies (Badri et al., 2017; Barbero & Rodríguez-Crespo, 2020; Ben Jebli & Ben Youssef, 2017; Hendricks et al., 2023; Muyanga & Jayne, 2012; Nugroho et al., 2022). Arable land, agricultural employment, education expenditure, gross fixed capital formation (GFCF), GDP, population size, polity, and year dummies of major global economic shocks are considered.

Land has been widely expounded as an output growth attribute since the inception of classical growth models (Arrow, 1962; Lucas, 1988; Romer, 1986; Solow, 1956). Larger farms encourage the use of modern technologies, which in turn influences AVA. The area of cultivable land may reflect the country’s potential to produce agricultural outputs that would have a direct influence on AVA through the supply of outputs that serve as inputs in the process of value addition. Larger farm sizes may be associated with economies of scale in input acquisition, which encourages farmers to participate in AVA. According to Muyanga & Jayne (2012), smaller farm sizes reduce the potential to produce surpluses, which may in turn cause capital constraints that impede the demand for purchased inputs and new technologies. Ben Jebli & Ben Youssef (2017) note that larger agricultural land increases agricultural production, allowing AVA to rise. This also supports the findings of Singariya & Sinha (2015), who claim that arable land has a long-term beneficial and consistent impact on AVA growth. In this study, land is referred to as the percentage of land that is arable.

Population size is proxied by the population aged 15–64 years, as presented in Beyene (2022) and Garedow (2022). A larger population provides a larger market base, which encourages competition and induces innovations as well as technological advancements (Furuoka, 2009) that influence AVA. The rural population is controlled, following Muyanga & Jayne (2012). Populations in rural areas primarily provide labor for agricultural production and value-adding processes.

Human capital has been widely acknowledged as a key output growth component, approximated by different indicators bounded by data availability. For instance, Beyene (2022) uses school enrollment as a measure of human capital, while Cole & Chawdhry (2002) use human capital investment. In the current study, this variable is represented by education spending as a percentage of gross national income (GNI). According to Badri et al. (2017), education improves labor productivity by allowing one to understand, predict, recognize, and address business needs, which enhances participation in agricultural value addition. Educated labor takes advantage of various opportunities in the process of AVA (Nugroho et al., 2022). Given the growing number of agro-industries engaged in agricultural value addition in developing countries, the importance of educated labor is crucial to meeting the already rising food demands. Education, in this case, plays a vital role in creating new capacities, both quantitatively and qualitatively. The more decent and efficient the education is, the more it directs the labor employed in the AVA to be more successful in undertaking value addition. According to Onoja et al. (2017), education is important in agribusiness as it enables one to apply efficient technologies and skills to bring about a quantum leap in the level of AVA.

GDP is considered a potential attribute of AVA in this study, measured in the per capita constant of 2015 USD. According to Gelgo et al. (2023), per capita GDP has a positive effect on AVA. An increase in GDP increases a country’s nominal expenditure in agriculture. Expenditure in agriculture enhances AVA by providing better opportunities for farmers to use farm technologies and other inputs. Allcott et al. (2006) and Singariya & Sinha (2015) regarded GDP as a crucial component of agricultural income. Barbero & Rodríguez-Crespo (2020) also note that countries with higher levels of economic development participate more in AVA.

Following Badri et al. (2017), GFCF is considered a potential AVA attribute. Increased investment in physical capital leads to a higher capital stock in agriculture, which would support long-term growth in AVA. This is the net addition to the stock of fixed capital assets, including machinery and infrastructure, in the economy. A larger GFCF would thus improve farm infrastructure, encourage the use of machines and equipment, and support the process of value addition in agriculture. According to Hendricks et al. (2023), policy changes in Sub-Saharan Africa are driven in part by external political shocks. In a highly complex, interlinked, and dynamic global economic arena, policies at the country level are usually motivated by political interests. Agricultural policies subjected to such external shocks are likely to influence agricultural products, which in turn would influence value addition in this sector. We use the polity variable to account for such political dynamics. As highlighted by Kose et al. (2020), year dummies (the year 1982, 1991, 2007, 2008, and 2009), that indicate major global economic shocks are finally taken into consideration to forecast AVA in this study. Table 1 provides a summary of the variables used in this study, their sources, and the anticipated effects.

Table 1. Data description, data sources, and hypotheses.

Variables	Description	Hypothesis	Sources
Dependent variable
AVA	Agricultural value added (Million USD)	–	World Bank
Independent variables
NAC	Nominal assistance coefficient; aggregated and for exportable	+ve	World Bank
NRP	Total NRP (%)	+ve	IFPRI
Land	Arable land (1000 hectares and % of total area of land)	+ve	FAO
PopAgric	% of people depending on farming	−ve	World Bank
AgEmp	% of people employed in agriculture	−ve	FAO
Educ	Education expenditure (% of GNI)	+ve	World Bank
GFCF	Gross fixed capital formation (% of GDP)	+ve	World Bank
PopTotal	Total population aged 15-64 years	−ve	World Bank
GDP	Per capita real GDP (in USD)	+ve	FAO
Polity	Institutionalized democracy and autocracy index ranging from 0 (worst) to 1 (best)	+ve	CSP

3.2. Model specification

Some studies use simple linear relationships to model AVA (Ben Jebli & Ben Youssef, 2017; Melembe, 2021; Nugroho et al., 2022), while others, for instance, Epaphra & Mwakalasya (2017), use a log-log function to reduce the severity of the regressors’ heterogeneity. For the present study, considering the merits of accounting both for log-linear and log-log relationships, we rely on the works of Rebelo (1992) while considering Badri et al. (2017) and Epaphra & Mwakalasya (2017) to build an empirical model of the following form (Equation 2(2) $${\mathit{\text{lny}}}_{\mathit{\text{it}}}=\alpha +\phi G_{\mathit{\text{it}}}+\sum _k{\beta }^k{\mathit{\text{lnZ}}}_{\mathit{\text{it}}}^k+{\epsilon }_{\mathit{\text{it}}}$$(2) ). (2) $${\mathit{\text{lny}}}_{\mathit{\text{it}}}=\alpha +\phi G_{\mathit{\text{it}}}+\sum _k{\beta }^k{\mathit{\text{lnZ}}}_{\mathit{\text{it}}}^k+{\epsilon }_{\mathit{\text{it}}}$$(2)

Where ∑ is a summation operator, ${\mathit{\text{lny}}}_{\mathit{\text{it}}}$ is log of AVA in country i at year t, $G_{\mathit{\text{it}}}$ is a measure of Barro (1990)’s public intervention (measures of distorted agricultural price incentives in our case) in country i at year t with its coefficient, and ${\mathit{\text{lnZ}}}_{\mathit{\text{it}}}^k$ is a vector of log-transformed k control variables in country i at year t. Vectors $\alpha ,\phi$ and ${\beta }^k$ are parameters to be estimated. ${\epsilon }_{\mathit{\text{it}}}$ is the error term with two orthogonal components: the fixed effect denoted by ${\mu }_i$ and the idiosyncratic shocks (${\nu }_{\mathit{\text{it}}})$ (Equation 3(3) $${\epsilon }_{\mathit{\text{it}}}={\mu }_i+{\nu }_{\mathit{\text{it}}},\text{where}E\left[{\mu }_i\right]=E\left[{\nu }_{\mathit{\text{it}}}\right]=E\left[{\mu }_i\mathrm{ }{\nu }_{\mathit{\text{it}}}\right]=0$$(3) ): (3) $${\epsilon }_{\mathit{\text{it}}}={\mu }_i+{\nu }_{\mathit{\text{it}}},\text{where}E\left[{\mu }_i\right]=E\left[{\nu }_{\mathit{\text{it}}}\right]=E\left[{\mu }_i\mathrm{ }{\nu }_{\mathit{\text{it}}}\right]=0$$(3)

Singariya & Sinha (2015) note that AVA in the current year is a function of the previous year’s AVA. With this concept, the dynamic expression of the response variable in the above specification (Equation 1(1) $$\mathit{\text{NRP}}=\frac{(\mathit{\text{Domestic price}}-\mathit{\text{Reference price}})}{\mathit{\text{Refernce price}}}\mathrm{*}100\mathrm{\%}$$(1) ) takes the following form (Equation 4(4) $${\mathit{\text{lny}}}_{\mathit{\text{it}}}=\alpha +\theta {\mathit{\text{lny}}}_{i,t-1}+\phi G_{\mathit{\text{it}}}+\sum _k{\beta }^k{\mathit{\text{lnZ}}}_{\mathit{\text{it}}}^k+{\epsilon }_{\mathit{\text{it}}}$$(4) ): (4) $${\mathit{\text{lny}}}_{\mathit{\text{it}}}=\alpha +\theta {\mathit{\text{lny}}}_{i,t-1}+\phi G_{\mathit{\text{it}}}+\sum _k{\beta }^k{\mathit{\text{lnZ}}}_{\mathit{\text{it}}}^k+{\epsilon }_{\mathit{\text{it}}}$$(4)

Where ${\mathit{\text{lny}}}_{i,t-1}$ is the lagged response variable in country i at year t-1 along its coefficient $\theta$ and all others are as defined earlier. Building on the works of Hendricks et al. (2023) and Singariya & Sinha (2015), the following reduced econometric model is estimated in this study (Equation 5(5) $${\mathrm{ }\mathit{\text{lnAVA}}}_{\mathit{\text{it}}}={\beta }_0+{\beta }_1{\mathit{\text{lnAVA}}}_{i,t-1}{+\beta }_2{\mathit{\text{Dist}}}_{\mathit{\text{it}}}+{\beta }_3{\mathit{\text{lnLand}}}_{\mathit{\text{it}}}+{\beta }_4{\mathit{\text{lnAgEmp}}}_{\mathit{\text{it}}}+{\beta }_5{\mathit{\text{lnEduc}}}_{\mathit{\text{it}}}+{\beta }_6{\mathit{\text{lnGFCF}}}_{\mathit{\text{it}}}+{\mathrm{ }\beta }_7{\mathit{\text{lnPop}}}_{\mathit{\text{it}}}+{\beta }_8{\mathit{\text{lnGDP}}}_{\mathit{\text{it}}}+{\beta }_9{\mathit{\text{lnPolity}}}_{\mathit{\text{it}}}+{\beta }_{10}\mathit{\text{Dy}}82+{\beta }_{11}\mathit{\text{Dy}}91+{\beta }_{12}\mathit{\text{Dy}}07+{\beta }_{13}\mathit{\text{Dy}}08+{\beta }_{14}\mathit{\text{Dy}}09+{\epsilon }_{\mathit{\text{it}}}$$(5) ). (5) $${\mathrm{ }\mathit{\text{lnAVA}}}_{\mathit{\text{it}}}={\beta }_0+{\beta }_1{\mathit{\text{lnAVA}}}_{i,t-1}{+\beta }_2{\mathit{\text{Dist}}}_{\mathit{\text{it}}}+{\beta }_3{\mathit{\text{lnLand}}}_{\mathit{\text{it}}}+{\beta }_4{\mathit{\text{lnAgEmp}}}_{\mathit{\text{it}}}+{\beta }_5{\mathit{\text{lnEduc}}}_{\mathit{\text{it}}}+{\beta }_6{\mathit{\text{lnGFCF}}}_{\mathit{\text{it}}}+{\mathrm{ }\beta }_7{\mathit{\text{lnPop}}}_{\mathit{\text{it}}}+{\beta }_8{\mathit{\text{lnGDP}}}_{\mathit{\text{it}}}+{\beta }_9{\mathit{\text{lnPolity}}}_{\mathit{\text{it}}}+{\beta }_{10}\mathit{\text{Dy}}82+{\beta }_{11}\mathit{\text{Dy}}91+{\beta }_{12}\mathit{\text{Dy}}07+{\beta }_{13}\mathit{\text{Dy}}08+{\beta }_{14}\mathit{\text{Dy}}09+{\epsilon }_{\mathit{\text{it}}}$$(5)

Where, ln is log operator and Dist is NRP and NAC (aggregated as well as only for exportable products). Others, namely AVA, Land, AgEmp, Educ, GFCF, Pop, GDP, and Polity², are as explained earlier (Table 1). Dy82, Dy91, Dy07, Dy08, and Dy09, are year dummies showing the year 1982, 1991, 2007, 2008, and 2009, respectively. t, i, and ε retain the earlier explanations. β₀ is constant, and β₁-β₁₄ are coefficients of explanatory variables.

3.3. Estimation method

Equation (5)(5) $${\mathrm{ }\mathit{\text{lnAVA}}}_{\mathit{\text{it}}}={\beta }_0+{\beta }_1{\mathit{\text{lnAVA}}}_{i,t-1}{+\beta }_2{\mathit{\text{Dist}}}_{\mathit{\text{it}}}+{\beta }_3{\mathit{\text{lnLand}}}_{\mathit{\text{it}}}+{\beta }_4{\mathit{\text{lnAgEmp}}}_{\mathit{\text{it}}}+{\beta }_5{\mathit{\text{lnEduc}}}_{\mathit{\text{it}}}+{\beta }_6{\mathit{\text{lnGFCF}}}_{\mathit{\text{it}}}+{\mathrm{ }\beta }_7{\mathit{\text{lnPop}}}_{\mathit{\text{it}}}+{\beta }_8{\mathit{\text{lnGDP}}}_{\mathit{\text{it}}}+{\beta }_9{\mathit{\text{lnPolity}}}_{\mathit{\text{it}}}+{\beta }_{10}\mathit{\text{Dy}}82+{\beta }_{11}\mathit{\text{Dy}}91+{\beta }_{12}\mathit{\text{Dy}}07+{\beta }_{13}\mathit{\text{Dy}}08+{\beta }_{14}\mathit{\text{Dy}}09+{\epsilon }_{\mathit{\text{it}}}$$(5) 's error structure would follow the characteristics of most panel data models. The model would be impacted by statistical issues such as heteroscedasticity, autocorrelation, and cross-sectional dependency (Ramoutar, 2017). Numerous dynamic panel data estimators, namely the Arellano and Bond (1991) GMM-DIF estimator, the Blundell & Bond (1998) GMM-SYS estimator, and the Bruno (2005) bias-corrected least squares dummy variable (LSDVC) estimator, can be considered in this case. However, panel data with a small number of cross-sectional units can cause the GMM estimators to be significantly skewed (Bruno, 2005). Despite the fact that employing small samples in macro-panels would help to reduce heterogeneity, employing GMM in this situation would result in inconsistent estimates. The LSDVC estimator often outperforms the GMM estimators in terms of bias correction (Bruno, 2005). For finite N and T, the LSDV would not, however, provide consistent estimates.

By including terms of at most order N^-1 T^-1, Kiviet (1995) uses higher-order asymptotic expansion techniques to approximate the small sample bias of the LSDV estimator. Later, Bun & Kiviet (2003) proved that bias approximations in the first-order approximation term of the LSDVC are capable of accounting for more than 90% of the real bias by simplifying the estimates in Kiviet (1995). However, unbalanced panels cannot be used with the procedures in Bun & Kiviet (2003). This is extended by Bruno (2005) so that the method takes into account panel data that is not balanced. He evaluated the performance of the LSDVC model using estimators from Anderson & Hsiao (1982) (AH), Arellano and Bond (1991) (AB), and Blundell & Bond (1998) (BB) for unbalanced panels with small N. The AH procedure transforms the model through first differences to eliminate unobserved individual heterogeneity (Anderson & Hsiao, 1982). AB uses a GMM estimator for the first-differenced model, which is more efficient than AH as it relies on a greater number of internal instruments (Anderson & Hsiao, 1982). The first differenced IV or GMM estimators may experience a strong small-sample bias due to poor instruments with extremely persistent data. To control this, BB uses a system GMM estimator with instruments in levels for the first-differenced equation and first-differenced instruments for the equation in levels (Blundell & Bond, 1998). The three versions of the LSDVC outperform in terms of bias and RMSE, and the model emerges as the preferred estimator for dynamic panel data models with a small N (Bruno, 2005). As such, given the small sample size of the data in this study, we opt to rely on Bruno (2005)’s bias-corrected LSDV model. The model is robust against heteroscedasticity, autocorrelation, and cross-sectional dependence.

3.4. Descriptive statistics

Graphically, the trends of the change in AVA are demonstrated in Figure 1. The figure shows that over the study period, AVA increased rapidly across most of the countries. In Burundi and Rwanda, slowly rising trends of AVA were observed. Smoothly rising trends were witnessed in Uganda and Tanzania. Ethiopia and Sudan had registered rapidly rising trends during the two and a half decades of the studied period. Similarly, a rising trend of agricultural value addition was exhibited in Kenya, with modest fluctuations.

Figure 1. Trends of agricultural value added.

The trends of the NACs are presented in Figure 2. For all of the countries, quite similar trends in NAC in both aggregate and exportable agricultural products were demonstrated. Initially, little deviation was observed between exportable NAC and aggregate NAC, particularly at the beginning of the 1980s in Ethiopia, Tanzania, and Uganda, which persisted until the beginning of the 1990s. By then, these deviations generally appeared to be converging.

Figure 2. Trends of NACs.

Similarly, the trends of the NRP show moderate fluctuations across the sampled countries. Rwanda had the most protected agriculture prior to 2010, particularly between 2008 and 2010. During the same period, Burundi had the least protected agriculture. A sharply rising trend of agricultural protection was then demonstrated in Burundi between 2007 and 2016. Except for Ethiopia, which has all negative NRP, no other country has either all positive or all negative NRP over the studied period. The results are presented in Figure 3.

Figure 3. NRP for agriculture.

Furthermore, the descriptive statistics results (Appendix A) show that the overall average arable land was about 10 million hectares. Proportionally, the area of arable land was about 26% in the region. The average number of people practicing farming in this region was about 26 million, and the regional average proportion of employment in agriculture was about 72%. Regionally, the average budget allocated for education was about 3% of gross national savings, while the average total population was about 37 million. The regional average real per capita GDP was about $28 billion, while the regional average polity index was about 0.4.

5. Results of the econometric analysis

5.1. Unit root analysis

We assessed the time series properties of the variables used in model 1 and model 2. For the variables used in model 3, the average length of the series is about 13 years, and unit root tests would be less reliable. The unit root test has low power for a small sample (Hendricks et al., 2023). The average length of time involved in model 1 and model 2 is 29 years, which is worth analyzing its time-series properties. Otherwise, the presence of unit roots would indeed generate inference problems for the LSDVC estimates. For instance, during the initialization of the LSDVC model with the AH procedure, all the variables are transformed through first differences to get rid of the unobserved individual heterogeneity that would be influenced by panel unit roots, which is worth investigating. In this study, the Fisher-type unit root test presented in Maddala & Wu (1999) is used. The method is preferred as it handles both balanced and unbalanced panel data. The results show that the variables in the models are a mix of stationary both at level (I(0)) and at order one (I(1)). The results thus imply that the employed LSDVC specification, initialized by the GMM estimator, would not provide spurious estimates as the panel unit root tests all reject the null hypothesis that all panels contain unit roots, at most at first differences. The results are presented in Table 2.

Table 2. Unit root test results.

	At level				At 1^st difference
	Const		Const. & trend		Const		Const. & trend
Variables	Inverse χ2	p > χ2	Inverse χ2	p > χ2	Inverse χ2	p > χ2	Inverse χ2	p > χ2
NACTotal	36.56***	0.000	47.05***	0.000	–	–	–	–
NACExportable	9.70	0.467	27.29***	0.002	166.68***	0.000	–	–
Ln(AVA)	0.79	0.999	13.14	0.216	120.82**	0.000	98.13***	0.000
Ln(Land)	3.60	0.964	3.74	0.958	93.35***	0.000	79.65***	0.000
Ln(AgEmp)	29.12***	0.001	5.44	0.860	–	–	71.77***	0.000
Ln(Educ)	10.65	0.386	18.47**	0.048	120.24***	0.000	100.25***	0.000
Ln(GFCF)	14.30	0.160	24.95***	0.005	133.58***	0.000	–	–
Ln(PopTot)	109.16***	0.000	24.73***	0.006	–	–	–	–
Ln(GDP)	0.91	0.999	7.90	0.639	75.10***	0.000	69.32***	0.000
Ln(Polity)	7.03	0.723	5.96	0.819	110.53***	0.000	86.83***	0.000

Note. ***, **, and * show significance levels at 1%, 5%, and 10%, respectively.

5.2. Effects of distorted agricultural price incentives on AVA

The econometric results of the effects of agricultural price incentives on AVA are presented in Table 3. The results indicate that agricultural price incentives have positive and significant effects on the AVA, regardless of the dimensions of the incentives considered. The results show that for every one unit increase in the aggregate NAC, AVA increases by about 11%, ceteris paribus. The result is statistically significant at 1% significance level. Similarly, if all country outputs were exportable, every one unit increase in the NAC would have an incremental effect on AVA of about 9%, ceteris paribus. The result is statistically significant at 5% significance level. In other words, suppose that the initial NAC is undistorted, these results indicate that every 100% increase in each of the current aggregate and exportable product domestic prices that producers effectively receive is associated with an increase in AVA of about 11% and 9%, respectively, if all other prices remain fixed. Similarly, a 1% increase in the NRP increased AVA by about 3%, ceteris paribus. The result is statistically significant at 10% significance level. The larger effect of the price incentives on AVA was reported for the NRP compared to that of the NAC.

Table 3. The effects of distortion of agricultural price incentives on AVA.

	Model 1		Model 2		Model 3
Ln(AVA)	Coef	Std. Err	Coef	Std. Err	Coef	Std. Err
NACTotal^3	0.102***	0.037	–	–	–	–
NACExportable^4	–	–	0.089**	0.038	–	–
NRP^5	–	–	–	–	0.026*	0.015
Ln(AVA)_1	0.793***	0.048	0.796***	0.050	0.664***	0.082
Ln(Land)	0.153*	0.082	0.168**	0.084	−0.009	0.092
Ln(PopAgric)	−0.193	0.181	−0.289	0.184	–	–
Ln(AgEmp)	–	–	–	–	−0.003**	0.001
Ln(Educ)	−0.056**	0.022	−0.047**	0.023	−0.0001	0.008
Ln(GFCF)	−0.009	0.021	−0.022	0.023	−0.001	0.001
Ln(PopTotal)	0.293*	0.173	0.329*	0.182	−0.106	0.118
Ln(GDP)	0.049***	0.017	0.049***	0.016	0.219**	0.091
Ln(Polity)	−0.012	0.013	−0.007	0.012	0.054**	0.025
Dy82	0.024	0.030	0.009	0.029	–	–
Dy91	0.034	0.026	0.032	0.025	–	–
Dy07	0.001	0.029	0.018	0.030	−0.013	0.012
Dy08	−0.012	0.031	0.006	0.031	−0.010	0.012
Dy09	−0.027	0.028	−0.026	0.031	−0.004	0.011
Numb. Obs	145		145		78
Av. Period	29		28.8		13

Note. Bias correction is initialized by AB, with bias correction up to order O(1/NT). The standard errors are bootstrapped using 200 iterations. ***, **, and * show significance levels at 1%, 5%, and 10%, respectively.

Furthermore, among the control variables, arable land influenced AVA positively and significantly at 10% and 5% probability levels in Model 1 and Model 2, respectively. Agricultural employment had a negative and significant effect on AVA at 5% probability level in model 3. Education expenditure influenced AVA negatively and significantly at 5% probability level, both in Model 1 and Model 2. Similarly, population size influenced AVA positively and significantly at 10% probability level, both in Model 1 and Model 2. Per capita real GDP influenced AVA positively and significantly at 1% probability level in Model 1 and Model 2 and at 5% probability level in Model 3. The polity index influenced AVA positively and significantly at 5% probability level in Model 3. Nevertheless, we do not put much emphasis on the coefficients of control variables, as they are included in the models to capture the effect of factors other than agricultural price incentives that may have correlations with AVA.

5.3. The error structure diagnosis

To check the reliability of the findings, we examine the structure of the error terms. The presence of FE and RE was investigated first (Appendix B). The F-test results confirm the presence of FE in all the models. Of course, when the selection of individuals in the panel is not random, the inclusion of FE is appropriate (Trabelsi, 2016). The LSDVC model is able to account for the presence of FE in the estimates. The Breuch and Pagan Lagrangian multiplier tests show no evidence of RE in the models. The presence of heteroskedasticity, cross-sectional dependence, and auto-correlation is then examined. Panel data structures frequently deviate from the standard assumptions related to these issues (Podestà, 2002). The likelihood ratio test results show that heteroskedasticity exists in each model. Similarly, Pesaran’s tests of cross-sectional independence show that no model exhibits cross-sectional dependence. The presence of first-order autocorrelation is confirmed in all the models. Based on the findings, the error structure is typically classified as panel heteroskedastic and auto-correlated, implying that the use of the LSDVC model is a valid estimation strategy. Additionally, model performances were evaluated using different bias correction terms along with a range of iterations. The models initialized with AB, bias 2, and iteration 200 were then selected to estimate the models (for details of the procedure, see Bruno, 2005). The results of the error structure analysis are presented in Table 4.

Table 4. Error structure.

	Group-wise hetero		Cross-sectional depend		Auto-corr
Models	χ2 – value	p > χ2	Test stat	p > t	F-value	p > F
Model 1	380.29***	0.000	0.388	0.6980	17.200**	0.0143
Model 2	37.56***	0.000	−0.684	1.506	16.561**	0.0152
Model 3	36.81***	0.000	−0.903	1.6334	33.694***	0.0021

Note. *** & ** show significance levels at 1% & 5%, respectively.

6. Discussion

Our results indicate that agricultural price incentives in favor of agriculture have positive effects on AVA, regardless of their type, whether they are aggregated or only for exportable products. These results justify the importance of agricultural incentives in the growth of regional AVA, though with varying degrees of significance and magnitude. The positive association could be related to the fact that agricultural price incentives increase farmers’ production capacity. It increases the use of farm inputs, which in turn enhances agricultural yield and encourages value addition. Favorable agricultural product prices promote local production by improving farmers’ income and their overall purchasing power of production technologies. Increased farmer’s incomes in this case would encourage innovation in agricultural production, providing fertile grounds for enhanced local agricultural value addition and also overall sectoral value addition in an economy. An exclusive positive effect of exportable products’ NAC supports the findings of Hendricks et al. (2023), who note that countries in SSA that reduced anti-agricultural policies on exportables had increased agricultural production. It implies that in east Africa, a reduced anti-agricultural price incentive for exportable agricultural products had significant incremental effects on the growth of AVA. According to Anderson et al. (2013), the reduction in anti-agricultural policies for exportable agricultural products in SSA was primarily due to reduced export taxes. In support of this, several other studies present the positive role of agricultural incentives in the agricultural sector (Bollman & Ferguson, 2018; Hemming et al., 2018; Ratii, 2016; Seitov, 2023; Ye et al., 2023). In contrast, Allcott et al. (2006) show that some agricultural incentive schemes, such as non-social subsidies, reduce the growth of agricultural GDP. Closely similar results were presented by Omeje et al. (2019), who note that agricultural protection has a negative effect on the growth of agricultural outputs. The negative effect emerges when agriculture does not make up a significant share of GDP.

Furthermore, arable land encourages AVA, while education expenditure and agricultural employment discourage it. Population size, per capita real GDP, and polity contribute to a larger AVA, though the effects vary across the models. The interpretation of these factors is not the focus of the current study. However, to highlight this, higher government expenditure on education has a positive implication for the growth of AVA (Badri et al., 2017). Educating humans is vital to flourishing the hidden talents that support value addition in agriculture (Badri et al., 2017). Similarly, countries with higher levels of economic development tend to participate more in value addition (Barbero & Rodríguez-Crespo, 2020). This might be the reason why improvement in the level of GDP is supportive of agricultural value addition in this study. A larger number of rural populations encourage AVA. This is in line with the population-led output expansion hypothesis of Furuoka (2009). A larger population creates better market opportunities, positively influencing value addition. This, however, is against the arguments of Salehi-Isfahani (1993), who note that population growth in rural areas is associated with low productivity in Africa due to the continent’s lower labor absorption capacity.

The results generally imply that the extent of the effect of agricultural price incentives on AVA depends on the dimensions of the incentives considered, although they all contribute positively. The AVA responds largely to the change in NRP, followed by the aggregate NAC. A higher AVA is associated with incentives targeting different stages of value addition than those targeting agricultural outputs. Understanding the different dimensions of incentives is thus vital to supporting value addition in the agricultural sector. The results are suggestive of the causal effects of agricultural price distortions on AVA. However, we cannot necessarily claim this relationship given the theoretical understanding that intervention through distortive measures reduces production efficiency. The study’s contributions in this regard, however, are not without drawbacks, as, for instance, it entirely depends on secondary data, sharing the limitations related to the use of such data. Another potential limitation is that disentangling measures of agricultural incentive at different levels of AVA is challenging, which can be considered in future related studies. Besides, the current study considered a shorter time period, limited by a lack of sufficiently long time series data to undertake commendable country-specific time series analysis. Although the biases that may arise in this regard were attempted through model selection, future studies would still make use of more advanced panel data models by employing data over a relatively extended period of time. Furthermore, to provide more holistic policy options in this regard, an investigation into the extent to which distorted agricultural incentives could be translated into household-level agricultural value addition at the micro-level is vital in future related studies. Finally, our results do not imply that price distortions are the only policies that matter in AVA. We, however, do not account for the potential interactions of a wide array of agricultural policies that matter in agricultural value addition, and this could also be considered in future related studies.

7. Conclusion and policy implications

In East Africa, agricultural incentives have been biased against the agricultural sector. Price incentives rendered to agricultural commodities, including those that were exclusively devoted to exportable agricultural products, appeared to be vital in enhancing the regional growth of AVA. An incentive targeting agricultural products at different levels of value addition has a larger effect than one targeting aggregate agricultural outputs and exportable agricultural outputs. This implies that agricultural incentive policies and market conditions that support local producers are vital to enhancing AVA in East Africa. Price incentives targeting exportable agricultural products increase farmers’ incomes for those engaged in the production of exportable products, which enhances the use of farm resources and, in turn, facilitates the process of agricultural value addition locally. Besides this, a larger area of arable land, a lower share of agricultural employment, and a larger GDP index significantly contributed to an increase in regional AVA. Given the importance of the agricultural sector for the livelihoods of the majority of society in this region, the findings have vital implications for persistent regional anti-agricultural biased policies. To further increase agricultural value addition in this region, governments should generally consider revising agricultural policies in a pro-agricultural way. Enhancing agricultural price support needs to be a crucial element of policy revisions in the region.

Authors’ contributions

Conceptualization, data collection, formal analysis, and writing the original draft were all undertaken by the first author. The second and third authors supervised this work, revised and engaged in providing intellectutal support. All authors have read and agreed to the submitted version of the manuscript for publication. All authors agree to be accountable for all aspects of this work.

Disclosure statement

The authors report there are no competing interests to declare.

Data availability statement

The employed dataset is publicly available. The NRA and Pop_Agric, are retrieved from the World Bank datasheet at: www.worldbank.org/agdistortions. The NRP is retrieved from the IFPRI datasheet at: www.agincentives.org/nominal-rate-of-protection. The AVA, Educ, GFCF, and Pop_Total, are retrived from the World Bank datasheet at: https://databank.worldbank.org/. The Land, AgEmp, and GDP, are obtained from the FAO datasheet at: www.fao.org/faostat/en/. Polity data is retrieved from the CSP datasheet at: www.systemicpeace.org/inscrdata.html.

Additional information

Funding

This study was financed by Jimma University.

Notes on contributors

Biru Gelgo Dube

Biru Gelgo Dube is a PhD candidate at Jimma University. He is interested in researches addressing political economy, agricultural policies, institutions, and agricultural technology adoption.

Adeba Gemechu Gobena

Adeba Gemechu Gobena is an associate professor of Agricultural Economics in the Department of Agricultural Economics and Agribusiness Management at Jimma University, Ethiopia. He has ample experiences on micro-economic and macro-economic researches. Specifically, he is more interested in researches addressing natural resource economics and development economics including agricultural policies, institutional economics, agricultural technology adoption, agricultural marketing, climate smart agriculture, and agricultural finances among money others.

Amsalu Bedemo Beyene

Amsalu Bedemo Beyene is an associate professor of economics at the Department of Policy Studies, Ethiopian Civil Service University (ECSU). The areas of his research interest include macroeconomic analysis, governance, institutions, and economic growth/development, political economy of financial development, agricultural development and structural transformation in Ethiopia, poverty, income inequality, and growth.

Notes

1 We compute exportable agricultural products NAC following similar procedures for agreggating NAC. That is: NAC_Exportable = NRA_Exportable +1.

2 A normalized polity data is used for the employed models to cease. The transformed series falls between two extreme values of 0 and 1. Both minimum (min) and maximum (max) values were theoretical and computed as follows:$\mathit{\text{Polity}}=({\mathit{\text{Actual Polity}}}_{\mathit{\text{it}}}-{\mathrm{\text{min}} \mathit{\text{Polity}}}_{\mathit{\text{it}}})/({\mathrm{\text{max}} \mathit{\text{Polity}}}_{\mathit{\text{it}}}-{\mathrm{\text{min}} \mathit{\text{Polity}}}_{\mathit{\text{it}}}).$

3 The NAC_Total coefficient is recalculated accounting for log-transformed AVA. We exponentiate the coefficient, subtract 1 and multiply by 100 to find % change. That is: exp ((0.102)-1)*100 % = 10.74%.

4 Similarly, we interpret the coefficient of NAC_Exportable as: exp ((0.089)-1)*100% = 9.31%.

5 The coefficient of NRP is: exp ((0.026)-1)*100% = 2.63%.

Appendix A.

Descriptive statistics (mean)

Table

Variable	Burundi	Ethiopia	Kenya	Sudan	Rwanda	Tanzania	Uganda	Overall
NACTotal	–	0.84	1.00	0.78	–	0.74	0.96	0.87
	–	[0.16]	[0.12]	[0.26]	–	[0.21]	[0.07]	[0.20]
NACExportable	–	0.66	0.90	0.63	–	0.49	0.75	0.69
	–	[0.14]	[0.13]	[0.27]	–	[0.22]	[0.28]	[0.25]
NRP (%)	−23.63	−32.11	5.90	–	−0.09	−3.60	−3.48	−9.50
	[40.78]	[14.77]	[20.55]	–	[26.87]	[28.74]	[26.64]	[30.02]
AVA (mil.)	995.98	11400.50	12806.09	13973.65	1736.87	7473.86	3949.14	9920.65
	[47.18]	[6259.05]	[3902.51]	[8100.45]	[365.58]	[3477.15]	[1409.82]	[6327.94]
Land (1000hec)	–	12342.31	5102.05	16408.97	–	10008.73	5643.95	9901.20
	–	[2183.05]	[595.36]	[2966.63]	–	[1948.87]	[873.97]	[4648.50]
Land (%)	41.67	13.15	9.87	–	46.01	13.32	33.26	26.21
	[4.81]	[0.97]	[0.43]	–	[1.05]	[1.17]	[1.64]	[14.87]
AgEmp (%)	87.77	72.62	58.98	–	75.09	69.51	69.41	72.23
	[1.51]	[3.65]	[2.13]	–	[7.86]	[2.53]	[2.70]	[9.45]
PopAgric	–	49359467	21158100	18478767	–	24558033	17049067	26140000
	–	[9135483]	[4277590.20]	[2311021.10]	–	[4938955.10]	[4060319.50]	[13100000]
Educ (%)	5.06	2.68	5.63	2.05	3.70	2.98	2.76	3.22
	[0.73]	[0.68]	[0.58]	[0.73]	[0.43]	[0.61]	[0.86]	[1.43]
PopTotal	4784413.40	67552574	32526313	27327661	5880865.40	34671691	24777938	37400000
	[665751.63]	[22069213]	[10257135]	[7981421.10]	[697885.61]	[10910341]	[8701477]	[20200000]
GDP (mil.)	2519.94	23672.54	37074.08	44811.29	6852.37	24132.98	11811.68	28300.51
	[399.08]	[22150.60]	[15802.33]	[24353.42]	[1958.93]	[14525.11]	[8012.99]	[21158.15]
GFCF (%)	12.40	21.75	17.56	14.425	19.80	29.58	20.83	20.83
	[1.45]	[9.34]	[2.17]	[4.437]	[3.50]	[8.22]	[6.27]	[8.29]
Polity index	0.70	0.37	0.48	0.29	0.34	0.34	0.35	0.36
	[0.16]	[0.19]	[0.33]	[0.21]	[0.02]	[0.12]	[0.16]	[0.22]

Note. [ ] standard deviations.

Appendix B.

Test results for the presence of FE and RE in the models

Table

	F – test		B – P LM test
Models	F – value	p > F	χ2 – value	p > χ2
Model 1	113.00***	0.000	0.00	1.000
Model 2	106.11***	0.000	0.00	1.000
Model 3	23.96***	0.000	0.00	1.000

Note. *** shows significance level at 1%.