Analysing the Impact of
Agro-Industry Development:
A Structural Change Perspective
Master Thesis: International Economics and Business (MSc.)
University of Groningen
Faculty of Economics and Business
Supervisor: Bart van Ark
Co-assessor: Tarek Harchaoui
ii Abstract
This master thesis aims to make a start in closing the existing gap between theory, research and practice regarding agro-industry development. Therefore, the impact of agro-industry development on agricultural labour productivity as well as agricultural employment and agro-industrial employment is investigated. These three research questions are derived from explaining the agro-industry development framework from a structural change perspective. The impact of agro-industry development on all three components is analysed by linear fixed effects regressions. Publicly available data from the FAO, UNIDO, World Bank and ILO have been combined in a newly constructed panel data set, which includes data on 15 countries from Africa, America, and Asia between 1990 and 2013. Food and beverages sector intensity operationalises agro-industry development being the central explanatory variable throughout this thesis.
All three hypotheses can be cautiously confirmed as food and beverages intensity appears to increase agricultural labour productivity and employment in the food sector, while agricultural employment is decreased. However, especially the first regression shows low robustness and only weak levels of significance. Overall, it is concluded that more research is needed to adequately justify agro-industry development programmes.
iii Table of Contents
1 Introduction ... 1
2 Agro-Industry Development ... 2
2.1 Structural Change ... 2
2.2 Smallholder Farmers and Commercialisation ... 3
2.3 Agro-Industry Development ... 5
2.4 Implication for this research ... 9
3 Methodology and Data ... 10
3.1 Methodology ... 11
3.2 Data Reliability ... 13
3.3 Appropriateness of the specification ... 18
4 Analysis ... 19
4.1 Agro-Industry Development, Productivity Gap, FDI and Employment Trends ... 19
4.2 The Impact of Agro-Industry Development ... 22
4.3 Robustness Tests ... 24
5 Conclusion ... 31
6 Publication bibliography ... 33
1 1 Introduction
Agricultural development is seen as a key element to poverty alleviation, food security and employment generation in developing countries (The World Bank 2007). Subsistence farmers are the largest group among the extreme poor worldwide having made them a prime target group for development programmes. However, low success rates have increasingly spurred doubts that smallholder commercialisation is the best way to facilitate agricultural development and poverty reduction (Collier and Dercon 2014). Additionally, awareness that subsistence farming should not be preserved over the course of development has risen (da Silva et al. 2009; FAO 2014). Instead, agro-industry development is increasingly mentioned as an alternative way to facilitate agricultural development (Barrett 2008).
The idea of agro-industry development is closely linked to the relatively recent approach of thinking in terms of activities as parts of value chains rather than merely thinking of agriculture products. Indeed, the agro-industry sector comprises all actors engaged in pre- or post-production activities of agricultural products. Especially handling and processing of agricultural raw materials are believed to hold great income generating potential in an agricultural development context. However, there appears to be a considerable gap between academic research and development practice. Aiming to reduce that gap this thesis attempts to answer the following three research questions.
The first research question investigates if agro-industry development has a positive impact on agricultural labour productivity. Furthermore, recent research on structural change found that enough manufacturing jobs need to be provided to realise the potential of agricultural productivity growth (Diao et al. 2017). Consequently, the second research question asks if agro-industry development has a positive impact on agro-industrial employment. Additionally, structural change theory assumes changes in the relative employment shares between both sectors. The third research question investigates therefore if there is any significant impact of agro-industry development on agricultural employment.
Answering these questions required the construction of a new panel data set, which includes 15 developing and emerging countries from Africa, America, and Asia, covering a time period of 1990-2013. The analysis utilises fixed effects linear multivariate regressions and confirms all hypotheses even though they partially show very low levels of robustness.
2 2 Agro-Industry Development
The recent paradigm change in economics towards thinking in global value chains and in activities rather than product types offers a new perspective on modernised agricultural value chains and their potential role in development processes. Traditional approaches to agricultural development have shown low rates of success, which inspires the international development community to look for additional approaches. Additionally, smallholder farming increasingly loses its positive image of environmentally sustainable, local production (Goswami et al. 2017). Instead, it is more and more recognised that smallholder farming limits specialisation and therefore poses an allocative inefficiency of resources, which can be reduced by agro-industry development and thereby facilitated structural change (Babu and Shishodia 2018; FAO 2014; Gollin et al. 2014; The World Bank 2007, 2008).
Therefore, the following chapter provides a structural change perspective of the idea to utilise agro-industry development for income generation, economic growth and development in general.
2.1 Structural Change
The notion of structural change as a driver of economic growth is based on the Lewis model and the observation that modern manufacturing sector is more productive than the traditional agricultural sector (Lewis 1954; Ranis and Fei 1961). Thus, the same amount of output can be produced by fewer workers and employment is “freed-up” if agricultural labour productivity increases. If this employment reallocates to the modern sector, its workforce is used more efficiently, aggregate labour productivity increases and the economy expands. It was found that reducing the gap in industrial labour productivity explains about 50% of aggregate labour productivity increases across countries (Margarida Duarte et al. 2010). Likewise, can all productivity stagnations be explained by the absents of industrial productivity increases as well as low services productivity (Margarida Duarte et al. 2010). Furthermore, the importance of structural transformation for rapidly growing economies such as China has been established by other studies (Cao and Birchenall 2013).
3
perspective on economic growth, which emphasises disembodied innovation as a growth engine, as proposed in the Solow model (Solow 1956). The latter is referred to as “structural change” term, which is an additional driver of economic growth (McMillan and Harttgen 2014; McMillan et al. 2017). Developed countries have followed an inverted U-shape pattern in their development of manufacturing shares in the economy over time (Margarida Duarte et al. 2010; McMillan and Harttgen 2014). In relatively low-income situations income increases lead to a higher demand for manufacturing goods and, consequently, increased manufacturing activity. If income levels augment further the demand for services becomes increasingly important causing a relative de-industrialisation as the share of manufacturing in the economy starts to decline. However, a recent analysis of developing countries has shown that positive agricultural productivity shocks, which traditionally freed up labour for industrial development can have negative impacts on aggregate labour productivity if workers move into less productive occupations for example in the (informal) service sector, which is sometimes referred to as premature de-industrialisation (Diao et al. 2017). Although agricultural development itself is promoted as another potential engine of growth, detailed policy recommendations are rarely discussed in the structural change literature. Instead, there is a strong focus on infrastructure improvements (Diao et al. 2003) and industrialisation in general.
2.2 Smallholder Farmers and Commercialisation
4
factor even though other definitions, such as access to input markets,would be more desirable and accurate to compare farms in a more differentiated analysis (Braun 2005).
Technological change, which raises total factor productivity, and commercialisation are key drivers of agricultural productivity (Braun 1995). Commercialisation refers to increased integration in nonfarm labour markets and market transactions on input and output markets in general. Increased commercialisation is believed to facilitate farmers and farm communities to engage in specialisation and to realise the benefits stemming from specialisation such as improved allocative efficiency and improved competitive advantages (Braun 1995; Barrett 2008; Carletto et al. 2017; Reardon et al. 2012; Strasberg et al. 1999; Tickner 1995). Traditionally, commercialisation has been associated with a transformation from food and staple crop sales towards cash crops exposing farmers increasingly to risks of food insecurity (Szirmai 2005). However, cash crop commercialisation has large positive spill-overs on food production as, for example, cash crop production may improve access to trade and delivery channels, which can be utilised for food production as well (Strasberg et al. 1999). Following the same idea, some scholars emphasise that commercialisation can happen in many different ways and there is, for instance, a distinction to be made between commercialisation on the input side and commercialisation on the output side (Tickner 1995). Consequently, commercialisation must be understood as more than transforming cropping patterns to cash crops. This aspect is further supported by the increased relevance of non-traditional agricultural products such as horticulture products or by fair trade or organic farming differentiated export crops (da Silva et al. 2009; Diao et al. 2003; Humphrey and Memedovic 2006; ILO 2007; Pingali and Rosegrant 1995).
5
labour can be quite high if markets are not functioning well. Such market failures cause artificial diseconomies outweighing the inherent positive economies of scale that could be realised with specialisation and commercialisation (Collier and Dercon 2014; Szirmai 2005; Braun 2005; Barrett 2008). Thus, it is increasingly postulated that development strategies should look beyond smallholder development (Dercon and Gollin 2014; Humphrey and Memedovic 2006; Collier and Dercon 2014). Nevertheless, smallholder commercialisation is still believed to be a crucial component of rural poverty reduction especially if productivity gains can be realised requiring lower transaction and transport costs in the rural markets (Braun 2005; Christiaensen et al. 2011; Diao et al. 2003; The World Bank 2007, 2008).
Linking smallholders to regional or global markets has proven to be difficult as these markets are increasingly associated with higher requirements in terms of scale, reliability and constant quality, which smallholder cannot easily provide on their own (Humphrey and Memedovic 2006; Rehber 2007; Swinnen 2005; The World Bank 2007, 2008; UNIDO 2012; Weatherspoon and Reardon 2003; Wiggins et al. 2011; Wiggins 2014). As a result, development agencies are increasingly searching for supplementary or alternative development strategies of which agro-industry development is one, which will be discussed in the next section.
However, the literature’s focus on micro-level surveys researching commercialisation in the context of nutrition and food security or crop-specific aspects (Carletto et al. 2017; Strasberg et al. 1999; Tickner 1995; Govereh et al. 1999; Braun 1995, 2005) implies a significant gap within the academic and practical literature. It appears that no studies are comparing the impact of commercialisation on labour productivity at national levels over time or applying structural change perspectives in a more general way.
2.3 Agro-Industry Development
6
of scale. These required services are the same services, which other (manufactured) products require. Therefore, modernised agricultural value chains can be assumed to share, as “industries without smokestacks”, some characteristics with traditional industries, such as value added distribution along the value chain or forward and backward linkages (Page 2011; UNIDO 2012). Especially non-traditional export products like horticulture and flowers are often discussed within the literature as examples for value chains with high value added through a continuous cool chain as well as adequate packaging and delivery for western markets (FAO 2014; Humphrey and Memedovic 2006; Page 2011; The World Bank 2007, 2017; UNIDO 2012).
7
Reardon 2003; Wiggins et al. 2011; Swinnen 2005; The World Bank 2008). Increased urbanisation and rising income increasingly leads to a “westernisation” of diets and consumption habits. Thus, national markets for food and beverages are growing rapidly in numerous developing countries, which is also reflected by the increased emergence of supermarkets (Reardon and Timmer 2012). At the same time, however, increasingly high standards on international food markets and delivery requirements make an integration of smallholders difficult (Collier und Dercon 2014). Consequently, it is argued that processing agro-enterprises can act as an intermediary. Nevertheless, such advanced requirements pose explicit market entry barriers to smallholder farmers and smaller agro-enterprises. Fulfilling rising standards requires learning and innovation processes, which in turn require time and access to information giving larger firms and enterprises with existing experience a competitive advantage. Therefore, foreign direct investment (FDI) is seen as a pivotal to facilitating knowledge transfer and innovation processes (Humphrey and Memedovic 2006; Swinnen 2005). FDI is believed to hold great potential for integrating smallholder farmers in markets as their technological knowledge can be transferred over time and access to input and output markets can be provided. Existing research suggests that the required learning processes do not happen without prominent private sector participation and are often transmitted through contract farming (The World Bank 2007). However, sector-level FDI data are scarce hampering an econometric analysis of the importance of FDI.
The international paradigm shift towards agro-industry development is also rooted in a recognition of subsistence farming’s macro-economic inefficiency and that it can only be part of food security solutions in early development stages (FAO 2014). Although being apparent in publications of other institutions as well, the FAO framework for sustainable food value chain development (SFVCD) perhaps formulates this idea the clearest:
“There is a fine line between helping smallholder farmers to survive in the short term and prolonging their misery in the long term. The objective of SFVCD is not to preserve smallholder farming, it is about broad-based job creation, income growth, and wealth accrual.” (FAO 2014, p. 32).
8
produce more in the same amount of time or to be able to spend additional time in non-farm activities while producing the same output. Either way, additional income is generated, which can be spend in multiple ways. At lower income levels this additional income is principally used to improve and diversify nutrition (Carletto et al. 2017; Jaffee et al. 2003; Braun 1995). Therefore, income increases in deprived areas are likely to raise demand for food and beverages as well as some other agro-industrial consumption goods. This in turn, ties right into the already changing food demand structures in developing countries (Reardon and Timmer 2012). Moreover, additional income from raised agricultural labour productivity can be reinvested into inputs or tools, which further facilitates productivity growth and the transition from subsistence to specialised agriculture. This process is referred to as an “investment loop” in the FAO framework on sustainable food value chains, which appears to correspond to what commercialisation of agriculture is about in general. However, it is recognised that only about 10 to 30% of smallholder farmers can be expected to become agricultural entrepreneurs. The other 70 to 90% of farmers are expected to search for alternative sources of income, for example, in newly created agro-industry jobs (FAO 2014). However, not only forward linkages exist but demand for intermediate inputs and capital goods such as machinery is generated through backward linkages as well (UNIDO 2012).
9
food and beverages is by far the most important subsector of agro-industrial manufacturing (UNIDO 2012). Due to the limited scope of this work, the analysis will, therefore, be restricted to the food and beverages sector.
Despite having well-designed frameworks for agro-industry development programmes, the FAO admits that there is a lack of scientific research and foundation for such programmes (FAO 2014, p. 15). The aim of this master thesis is to start filling this gap between economic theory, economic research, development theory and development practice.
2.4 Implication for this research
This thesis investigated whether there are any indications that the assumed links between agro-industry development, agricultural labour productivity and structural change exist. Validating those correlation in a cross-country context can be interpreted as a necessary criterion to justify, from a structural change perspective, development programmes targeting the agro-industrial sector. This does not implicate that research on socio-economic effects on welfare such as income, nutrition and food security changes are unimportant. The focus on macro-economic impacts and possible structural change effects is chosen because it is rarely included in existing research but, nevertheless, might be pivotal in our current understanding of long-term economic development.
The literature review clearly indicates that development programmes in the agricultural sector are only helpful from a structural change perspective if agricultural productivity is increased as a result. For this reason, the first research question is investigating if agro-industry development has a positive impact on agricultural labour productivity.
Furthermore, current research indicates that positive agricultural labour productivity shocks only translate into increased aggregate productivity if enough high productive employment is generated (Diao et al. 2017). Likewise, income growth through agro-industrial employment creation plays a pivotal part within the FAO’s framework on sustainable food value chains (FAO 2014). Subsequently, a second research question investigates if agro-industry development creates additional employment opportunities in the agro-industry sector.
10
engage in off-farm activities. The third research question therefore asks if agro-industry development has a significant impact on agricultural employment.
Overall, it needs to be mentioned that even finding the expected relationship is not a sufficient reason for implementing agro-industry development programmes as, for example, other externalities and efficiencies need to be considered as well. However, not validating these expectations would cast doubt on such programmes adequateness from an economic perspective.
3 Methodology and Data
Analysing the impact of agro-industry development requires a precise definition and measurements. Unfortunately, measuring agribusiness is difficult and the international development community has still not reached a consensus on how to measure agribusiness itself adequately (FAO and UNIDO 2015). However, throughout this work agro-industries are limited to the food and beverages sector. Additionally, the focus is on agro-industry development in the sense that the food and beverages sector (FB) becomes, relative to the agriculture, forestry and fishing sector (AFF), more important over time. As a result, agro-industry development is operationalised through FB intensity of agricultural value chains, which becomes the central explanatory variable throughout the analysis. This measure is based on a working paper by the World Bank and has also been used by the FAO. It divides the share of food and beverages in GDP over the share of agriculture in GDP (Jaffee et al. 2003; da Silva et al. 2009).
FB Intensity= value addedvalue addedFB share in GDP
AFF share in GDP *100
=value addedGDP FB value addedGDP AFF*100 = value addedvalue addedFB
AFF*100
11
well. Consequently, the intensity share is believed to represent the level of commercialisation of food value chains, the general professionalisation and specialisation within value chains, the importance of distribution activities as well as the sectors potential for value adding (da Silva et al. 2009, p. 49).
The next chapter will introduce the methodology and specifications of the research questions. Afterwards the constructed data set is introduced. On this basis the appropriateness of the specifications is discussed.
3.1 Methodology
Chapter 2.4 already introduced the three research questions and the reasoning behind them. This section will present the specifications, expected signs of correlations according to the hypotheses and the economic models. All three questions will be analysed by utilising fixed effects linear multivariate regressions. Therefore, all economic models will have the same appearance:
= + ∗ + !
Where Y represents the dependent variable of interest, the constant, X a matrix of independent variables and a vector of corresponding coefficients as well as an error term !, which comprises the difference between estimated and observed value for each point of estimation.
In order to investigate the first research question, whether agro-industry development (FB Intensity) has a positive impact on agricultural labour productivity (value added per worker), the following specifications are made:
ln( labour productivity_AFF)y+8 =
β0 + β1* FB Intensity + β2* ,exportsoutputAFF
AFF * 100 . + β3 *(
gross fixed capital formationAFF
outputAFF *100)
+ β4* fertiliser consumption output
AFF + β5*
GDP
population + ϵ
12
years until effects become apparent in the data. The logarithm is used to flatten the distribution of variables to allow for a linear regression and improve overall comparability. For example, Chinese employment figures and Ethiopian employment figures will always be on very different levels. The logarithm allows to interpret coefficients in percentage changes instead of absolute scale units like 1000 workers. Following the argumentation in chapter 2.4, FB intensity is expected to show a positive coefficient. Although FB intensity is the key variable of interest agricultural labour productivity is likely to be influenced by other factors as well. One other potential influence are export shares of agricultural products, which represent a certain degree of international connectivity and serve as an indicator for organisational capabilities. As discussed previously, exporting includes special challenges for developing countries, which smallholder farmers cannot meet on their own. Thus, high export shares may indicate higher levels of cooperation and organisational efficiency among exporting farmers. Besides, gross fixed capital formation in the agricultural sector is included to control for labour productivity increases due to mechanisation and other forms of capital accumulation. Likewise, increased fertiliser usage is expected to enhance agricultural labour productivity. As discussed previously, increased use of inputs such as fertilisers are used as indicators for commercialisation. Hence, fertiliser consumption per million of output, which is constructed as a two-year average, and capital formation are controlling for the effects of commercialisation. All time-invariant systematic effects are captured by the constant (β0). Moreover, GDP per capita is included to control for non-constant country effects.
The second question referring to employment effects of agro-industry development is investigated by the following model:
ln(FB employment)y+1 = β0 + β1*FB intensity + β2*( exportsoutputFB
FB *100)
+β3*(gross fixed capital formationoutput FB
FB *100) + β4* ln(abour productivity_AFF) + β5*
GDP
population + ϵ
13
differences. Agricultural labour productivity is included as another variable of interest. If the first regression confirms the hypothesis that FB sector development increases agricultural labour productivity and the productivity coefficient in the second is positive as well, it is a first indication that reinforcement mechanisms, which are assumed by the FAO and have been discussed in chapter 2.3, may exist.
In order to test if agro-industry development spurs employment migration out of agricultural qualifying as the third research question the following specifications are made:
ln( employment_AFF)y+1= β0 + β1* FB intensity + β2 *(exportsoutputAFF
AFF *100)
+ β3*(gross fixed capital formationoutput AFF
AFF *100)+ β4* ln(labour productivity_AFF) + β5*
GDP
population + ϵ
This specification assumes again it takes some time until effects of increased FB intensity become apparent in absolute employment movements. Therefore, again a one-year lag is applied. Like in the previous specification agricultural labour productivity is of additional interest to identify the possibility of reinforcing or adverse effects of both. FB intensity is expected to show a negative coefficient. Agricultural export shares, fixed capital formation and GDP per capita are included as control variables like in specification 1.
3.2 Data Reliability
14
dataset: Bolivia, China, Costa Rica, Ecuador, Ethiopia, India, Malaysia, Malawi, Mexico, Morocco, Philippines, Tanzania, Tunisia, Uruguay, and Vietnam.
A complete table listing the source of all variables as well as their measurement units and definitions can be found below (table 1). Data was mainly retrieved from the FAO database and the UNIDO, which provided requested parts of their INDSTAT 2 dataset. The INDSTAT 2 database uses less specific subsectors (2-digit) but has more data available for the chosen time period in comparison to other databases e.g. INDSTAT 4. Other data were obtained from the “World Development Indicators” Database of the World Bank and agricultural employment data were retrieved from the ILO. The retrieved data were comprised into one data set and organised into a panel structure. All monetary values have been converted from current USD into constant 2010 USD using adequate deflators.
TABLE 1: Variables, Data Sources, Definitions and Missing Values in the constructed Dataset
Variable Source Units Definition Comments
Food and Beverages Industry Intensity
Own calculations In percent
FB_intensity = 6789: 7;;:;<=/?@A 6789: 7;;:;B<</?@A *100 = 6789: 7;;:;<= 6789: 7;;:;B<< *100
Proxy for “Agro-industry Development” Agricultural Labour Productivity World Development Indicators Database In USD (constant 2010; AFF deflator) Labour Productivity_AFF= CDEFG DHHGH_IJJ KLMENOLGPQ_IJJ
Derived from WB national accounts files, FAO, Production Yearbook and data files.
Value added and employment were only available from different sources. Division of value added by employment gives slightly deviating results
Agricultural
Employment ILO
In thousand
workers Employment_AFF
15 Employment in the FB industry UNIDO (“INDSTAT2”) In thousand workers Employment_FB
Missing values for Costa Rica, Mexico, Tanzania, Tunisia, Vietnam Gross Domestic Product FAOstat (sourced from UNSD) In million USD (constant 2010, GDP deflator) GDP GDP per capita World Development Indicators Database In USD (constant 2010) ARS987TURV?@A
sourced from national accounts of WB and OECD Total Agricultural Output FAOstat (“Production”) In million USD (constant 2010; AFF deflator) Output_AFF
“Aggregate, may include official, semi-official, estimated or calculated data” Total Value Added in Agriculture, Forestry and Fishing (AFF) FAOstat (“Macro-Indicators”) In million USD (constant 2010, AFF deflator) Value added_AFF = Output_AFF - interm. inputs_AFF
based on national accounts data of UNSD Agricultural Export Value FAOstat (“Trade-Indices”) In thousand USD (constant 2010; AFF deflator) Export_AFF
“Aggregate, may include official, semi-official, estimated or calculated data” Export share of agricultural products
Own calculation In percent
Sh_Exports_AFF = KWMNXQY_IJJ ZNQDE [FQMFQ_IJJ
* 100
Should not be possible to become larger than 100%. Some values (Malaysia, Uruguay) above probably due to estimation errors
Agricultural Gross Fixed Capital Formation FAOstat (“Investment”) In million USD (constant 2010; GFCF deflator)
GFCF_AFF FAO data imputation methodology Agricultural Gross Fixed Capital Formation Ratio
Own calculations In percent
Investment ratio_AFF = \J]J_IJJ [FQMFQ_IJJ ∗ 100 Agricultural Foreign Direct Investment FAOstat (“Investment”; based on UNCTAD and INTRACEN) In million USD (constant 2010; CPI deflator) FDI_AFF
Available for less than 50% of years in: Bolivia, Ethiopia, India, Malawi, Malaysia, Tanzania, Uruguay, Vietnam Fertiliser Consumption per Output FAOstat (“Inputs”) In tonnes per million of Output Fertiliser = 0.5 ∗
16
(d9TS9T e___:`TU8Ua:`bc
bf
+ d9TS9T e___:`TU8Ua:`bgh
bgh)
Due to correlation with the GFCF_AFF share later only used as total values and not as share.
Total Output in the Food and Beverages (FB) Industry Sector UNIDO (“INDSTAT2”) In million USD (constant 2010, manufacturin g deflator) Output_FB Due to manufacturing deflator gaps: missing values for China prior to 2005; some missing values for Tunisia and Vietnam in the data Total Value Added in Food and Beverages UNIDO (“INDSTAT2”) In million USD (constant 2010, manufacturin g deflator) Value added_FB = Output_FB – interm. inputs
missing values for China prior to 2005 (deflator gap); Export Value of Beverages FAOstat (“Trade-Indices”) In thousand USD (constant 2010, manufacturin g deflator)
Exports_FB Some missing values for Tanzania
Food and Beverages Export share
Own calculations In percent
Sh_Exports_FB = KWMNXQY_Ji
[FQMFQ_Ji ∗ 100
Missing values due to gaps in both components (China, Tanzania, Tunisia, Vietnam) Gross Fixed Capital Formation in the FB industry UNIDO (“INDSTAT2”) In million USD (constant 2010; GFCF deflator) GFCF_FB
Many missing values especially in Bolivia, China, Costa Rica (no data), Vietnam FB industry Gross Fixed Capital Formation Ratio
Own calculations In percent
Investment ratio_FB =
\J]J_Ji [FQMFQ_Ji ∗ 100
Combined missing values of both components; Foreign Direct Investment in the FB industry FAOstat (“Investment”) In million USD (constant 2010, CPI deflator) FDI_FB
based on UNCTAD and INTRACEN; Over 10 values only available for: Mexico and Uruguay
17
respectively. The loss in observations causes significance tests to become less reliable. Moreover, the deletion of observations due to missing values can lead to distortions in the country representation. Most missing variables are spread across the data set. However, gross fixed capital formation is not available for Costa Rica and only partially available for some other countries. Therefore, an alternative estimation of the second specification will be discussed in chapter 4.3. Other clusters of missing values are mentioned in the table above. Furthermore, the data itself may not always be reliable. Most included variables are sourced from national accounts data. However, the reliability and quality of statistic agencies profoundly differ among developing countries and even can be very volatile on subnational levels. Furthermore, many included variables are estimates. Total agricultural output, for example, is estimated by the FAO based on fertiliser and seed usage. Throughout this work labour productivity will be defined as each sector’s value added divided by employment. Hours worked would be preferred but it appears that employment data are already contested. Especially agricultural employment data appear to be unreliable as estimates from the ILO and FAO partially show significant differences. Considering the context of high rates of informality in developing countries and family labour characteristic of smallholder farmers it is not surprising that estimation methods produce different results. Higher continuity has led to use the ILO data on agricultural employment.
Furthermore, the nature of this dataset has led to the decision to not using any statistical data imputation methods. While adding additional problems by itself, any data imputation based on estimates and partially unreliable data is not believed to add to the robustness of the dataset. As discussed in section 2.3, FDI is of high importance for the internationalization of agro-industry development. Unfortunately, sector-specific FDI data are scarce for developing and emerging countries. While there will be a short descriptive analysis of the available FDI data (chapter 4.1), it had to be dropped from the regressions. Instead, FB export shares act as a proxy for international relationships as discussed in the previous section.
18 3.3 Appropriateness of the specification
Planning to work with multivariate linear regressions involves the fundamental question whether fixed or random effects estimation should be used. Implementing each estimation as a fixed regression allows to control for all time-invariant systematic differences between countries. Intuitively, there is no reason to assume that countries randomly differ without any systematic effects. Statistically, Hausman tests, which compare both variants, reject a random effects estimation for the first and third regression (Appendix I). Therefore, the use of fixed effects estimation appears to be justified. It appears unreasonable to use different methods within one data set and change fundamental assumption about country-specific influence, which led to use the fixed effects estimation for the second regression as well.
Another problem, which is typically encountered with multivariate regressions, is heteroscedasticity. The standard regression assumes homoscedasticity. Not controlling for this leads to an overestimation of standard errors and, consequently, confidence intervals and significance levels. Indeed, graphical analysis plotting residuals against the regression lines clearly indicate that residuals are not randomly and evenly distributed around the estimated values. Instead it appears that there is some systematic distortion in all three regressions (Appendix II). Hence, robust standard errors are used to control for this effect and creating more reliable significance levels.
Furthermore, multivariate regression assumes that no independent variables are strongly correlated with each other. However, listing all correlations of the independent variables reveals that fertiliser consumption per output and agricultural gross fixed capital formation shares become highly correlated due to their shared denominator. Therefore, it was decided to use just a two-year average of fertiliser consumption instead of the ratio. The correlation table of the final data set can be found in Appendix III.
19 4 Analysis
Aiming to better connecting economic research with theory and development practice this study examines the proposed correlations between agro-industry development and agricultural labour productivity, industry employment and labour movements between agriculture and agro-industry.
This chapter consists of three parts. The first section provides a descriptive analysis of some crucial features. The regression results will be analysed and interpreted in the second section. Finally, those results will be discussed.
4.1 Agro-Industry Development, Productivity Gap, FDI and Employment Trends
Table 2 provides some summary statistics, which again reveals amongst other things the unreliable nature of the data.
For instance, Uruguay and Malaysia show agricultural export rates above 100%, which would mean that exports surpass production in the country and could happen due to two different reasons: underestimation of agricultural production values or trading of products without local production activities. To avoid massive distortions without excluding another entire country from the analysis all values higher than 140% have been manually excluded from the regressions. Chapter 4.3 provides a comparison of results with agricultural export shares in-
TABLE 2: Summary Statistics
Observations Mean Median Standard
Deviation Maximal Value Minimal Value GDP 360 388096.00 45258.76 989667.85 7726483.65 3350.32 GDP per capita 360 3336.76 1977.772 3125.90 13467.46 146.7 FB Intensity 326 27.14 21.36451 22.44 102.35 1.28 Labour Productivity AFF 352 4176.44 1159.977 11912.65 190516.03 3.98 Labour Productivity FB 301 23072.12 19638.51 26536.99 309394.16 1578.36 Employment AFF 345 40189.04 4250.968 86389.25 352323 51 Employment FB 316 552022.00 72.746 1281.74 7769 7.7
Export Shares AFF 322 42.61 28.83181 35.01 139.29 3.39
Export Shares FB 313 0.55 0.203103 0.96 4.97 0
Gross Fixed Capital
Formation AFF 360 9773.72 654.1715 74393.80 1379353.43 3.58
Gross Fixed Capital
Formation FB 253 3793.71 333.7863 20971.94 176766.64 6.71
20
and excluded. However, the substantial difference of 14 percentage points between mean and median reveals that those large numbers are indeed outliers. In contrast, food and beverages export shares show a different pattern as they are generally very low. Hence, interpreting regression coefficients of export shares needs to be done cautiously for both sectors.
Food and beverages intensity as well as fertiliser usage per generated million worth of output reveal a high heterogeneity among the countries. Commercialisation and agro-industry development levels, as assumed, appear to differ in the selected sample of countries, which is also confirmed by a graphical overview of FB intensity across countries over time (Figure 1). The strongest agro-industry developments can be found in Latin America with Mexico, Costa Rica and Ecuador (prior to 2010) having high intensification rates. Vietnam, China and Bolivia also demonstrate strong positive changes. In contrast, Uruguay shows a strong declining pattern, which is also apparent in the Philippines albeit not as stark. However, Tanzania is often discussed in the context of agro-industry development studies but shows only very modest improvements.
Figure 1: FB Intensity Development by Country over Time
21
(Figure 2). Despite being higher than agricultural labour productivity, productivity in the food and beverages sector does not show a clear overarching trend. However, the rapid increase in Tanzania might give an indication why agro-industry development is often discussed in the context of Tanzania in Africa.
Figure 2: (logarithmic) Labour Productivity Gap
Creating the same figure for FDI in the food and beverages sector is less informative. Nevertheless, it demonstrates how little data is available (Appendix IV). Mexico is the only country in the sample having continuous data. However, it may not be a pure coincidence that there is continuous data available for Mexico, which shows the most impressive food and beverages intensity development in this sample. Mexico’s FDI pattern resembles the ups and downs of its labour productivity development. Therefore, this simple graphic test appears to confirm the assumed relevance of FDI. Unfortunately, that is the only message to be obtained from this data as there are too many gaps to include the measure in the analysis.
22 4.2 The Impact of Agro-Industry Development
As stated previously, this thesis follows three research questions and hypotheses. Firstly, a positive correlation between food and beverages intensity and agricultural labour productivity is expected. Secondly, a positive correlation is expected between food and beverages intensity and employment in the same sector. Additionally, a positive coefficient of agricultural labour productivity is expected, which may be a first indicator for an existing reinforcement mechanism between the two. Thirdly, it is expected that increased FB intensity has negative impact on agricultural employment. Table 3 presents the results for all three regressions. Overall, the regression coefficients show the expected signs and are at least significant on a 10% level.
More specifically, the first model’s estimations are presented in column 1. As stated above, the food and beverages sector intensity variable has a slightly significant, positive correlation coefficient. The coefficient indicates that a 1 percentage point increase of FB intensity is associated with a 1.9% increase of agricultural labour productivity 8 years later, which is statistically significant only at a 10% level (p-value of 0.077). Unfortunately, the only other significant variable is GDP per capita, where an increase of 100 USD per capita can be associated with an 5.6% increase in agricultural labour productivity. Country-specific fixed effects appear to strongly influence this relationship. The within R2 indicates that 35.4% of the
variation within the panel data can be explained, which is relatively low. Nevertheless, the first hypothesis can be confirmed. The data show that agro-industry development, or at least a higher food and beverages intensity, has a positive impact on agricultural labour productivity.
23
sector. In contrast, an increase in fixed capital formation shares appears to have a negative impact on employment.
Additionally, agricultural labour productivity shows a significant positive coefficient. In combination with the results of the first regression this indicates that reinforcement
TABLE 3: Regression Results of the Analysis
(1) (2) (3)
Variables Ln(Lab. Productivity AFF)y+8 Ln(Employment FB) y+1 Ln(Employment AFF) y+1
FB Intensity 0.0194* 0.0102*** -0.00345*
(0.0102) (0.00276) (0.00170)
Export Shares AFF 0.0146 -0.00230
(0.0209) (0.00187)
GFCF Shares AFF 0.000589 0.0137**
(0.0187) (0.00587)
Fertiliser Consumption 1.08e-08
(2.50e-08)
GDP per capita 0.000561** 9.54e-05** 1.88e-05
(0.000230) (3.72e-05) (4.52e-05)
Export Shares FB 0.119***
(0.0251)
GFCF Shares FB -0.00152***
(0.000390)
Ln(Lab. Productivity AFF) 0.0804** 0.0936**
(0.0361) (0.0426) Constant 3.531*** 10.29*** 14.68*** (0.812) (0.222) (0.244) Observations 201 209 279 Within R-squared 0.354 0.448 0.269 Number of Countries 15 14 15
24
mechanisms, as assumed by the FAO, may exist. The within R2 is relatively high and indicates
that 44.8% of variation within the data can be explained. Overall, the second hypothesis is confirmed as well. Food and beverages sector development has a positive impact on the sector’s employment. The magnitude of the impact may confirm the FAO’s intuition that the manufacturing employment generation is probably the most important feature of agro-industry development as discussed in chapter 2.3.
Finally, the third column confirms the expectations as well. A 1 percentage point increase of food and beverage intensity is associated with a 0.35% decrease of agricultural employment. However, this relationship is again only significant at a 10% level. In contrast, increased agricultural exports and labour productivity have positive impacts on agricultural employment. Each 1 percentage point increase of agricultural exports can be associated with an 1.37% increase of agricultural employment, which is significant at a 5% level. It is interesting that the impact of agricultural labour productivity increases roughly appear to be of the same magnitude for both sectors. The within R2 indicates that 26.9% of the variance within the data can be explained, which is relatively low. Nevertheless, the third hypothesis can be confirmed as well.
4.3 Robustness Tests
The previous analysis confirms the assumptions, which have been made in the context of agro-industry development programmes. In this chapter a few aspects of the previous analysis will be discussed especially regarding the robustness of results and associated caveats.
25
interpret any results cautiously and to scrutinise this relationship, which is fundamental for any development programmes including agro-industry facilitation.
TABLE 4: Robustness Test of Regression 1 by Time-Lag Comparison
(1) (2) (3) (4) (5) (6) (7)
Ln (Lab. Productivity AFF)
Y Y+1 Y+5 Y+7 Y+8 Y+9 Y+10
FB Intensity 0.00102 0.00420 0.0109 0.0196 0.0194* 0.0171* 0.0149 (0.00641) (0.00843) (0.0123) (0.0116) (0.0102) (0.00893) (0.00963) Export Shares AFF 0.0132 0.0143 0.00469 0.00747 0.0146 0.0197 0.0207 (0.00913) (0.0103) (0.0260) (0.0271) (0.0209) (0.0181) (0.0173) GFCF Shares AFF -0.0287** -0.000695 0.00295 0.00435 0.000589 -0.0101 -0.0188 (0.0127) (0.0124) (0.0173) (0.0162) (0.0187) (0.0252) (0.0296) Fertiliser
Consumption 3.55e-09 6.47e-11 -5.85e-09 2.64e-09 1.08e-08 1.73e-08 2.33e-08
(1.78e-08) (1.77e-08) (2.76e-08) (2.67e-08) (2.50e-08) (2.43e-08) (2.49e-08)
GDP per capita 0.000397** 0.000389 0.000609** 0.000581** 0.000561** 0.000594** 0.000631** (0.000152) (0.000197) (0.000237) (0.000242) (0.000230) (0.000236) (0.000236) Constant 5.752*** 5.230*** 4.394*** 3.875*** 3.531*** 3.243*** 3.098** (0.406) (0.378) (0.711) (0.877) (0.812) (0.783) (0.830) Observations 288 287 240 214 201 188 175 Within R-squared 0.503 0.473 0.366 0.337 0.354 0.373 0.379
Robust standard errors in parentheses *** p < 0.01, ** p < 0.05, * p < 0.1
26
the validity of the second regression and the confirmation of the second hypothesis as well as an indication that agricultural labour productivity may reinforce the impact of FB intensity.
TABLE 5: Robustness Test of Regression 2 by Time-Lag Comparison
(1) (2) (3) (4) (5) (6) (7)
Ln(Employ-ment FB) Y Y+1 Y+3 Y+5 Y+7 Y+9 Y+10
FB Intensity 0.0136*** 0.0102** 0.0122** 0.0115** 0.00753 0.00427 0.00577 (0.00266) (0.00276) (0.00315) (0.00385) (0.00553) (0.00509) (0.00550) Export Shares FB 0.00889 0.119*** 0.0571* 0.119* 0.165 0.445 0.493** (0.0504) (0.0251) (0.0826) (0.0664) (0.109) (0.250) (0.175) GFCF Shares FB 0.00201* -0.00152** -0.00534*** -0.00935*** -0.00832*** -0.00703 -0.0113 (0.000773) (0.000390) (0.00109) (0.00121) (0.00157) (0.00882) (0.0119) Ln(Lab. Productivity AFF) 0.0694** 0.0804** 0.0634 0.0662 0.0994* 0.179** 0.210** (0.0319) (0.0361) (0.0456) (0.0543) (0.0519) (0.0611) (0.0924) GDP per capita 0.000137** 0.0000954* 0.000110** 0.0000692 0.0000375 -0.00000208 -0.00000442 (0.0000414) (0.0000372) (0.0000390) (0.0000494) (0.0000707) (0.0000802) (0.0000902) Constant 10.20*** 10.29*** 10.36*** 10.47*** 10.39*** 9.798*** 9.472*** (0.227) (0.222) (0.329) (0.390) (0.431) (0.523) (0.761) Observations 221 209 190 169 148 128 119 Within R-squared 501 0.448 0.412 0.394 0.268 0.164 0.151
Robust standard errors in parentheses *** p < 0.01, ** p < 0.05, * p < 0.1
27
with some labour migration out of farming while agricultural labour productivity leads to more absolute employment in the agricultural sector over time.
TABLE 6: Robustness Test of Regression 3 by Time-Lag Comparison
(1) (2) (3) (4) (5) (6) (7)
Ln(Employment
AFF) Y Y+1 Y+5 Y+7 Y+8 Y+9 Y+10
FB Intensity -0.00418** -0.00345* -0.00155 -0.000704 -0.000283 -0.000176 0.00111
(0.00166) (0.00170) (0.00193) (0.00185) (0.00197) (0.00204) (0.00175)
Export Shares AFF -0.00134 -0.00230 -0.00362** -0.00266 -0.00212 -0.00181 -0.00187
(0.00193) (0.00187) (0.00163) (0.00152) (0.00174) (0.00193) (0.00167) GFCF Shares AFF 0.0121* 0.0137** 0.0145*** 0.0154*** 0.0143*** 0.0143*** 0.0151*** (0.00616) (0.00587) (0.00295) (0.00281) (0.00334) (0.00300) (0.00226) Ln(Lab. Productivity AFF) 0.0811 0.0936** 0.125*** 0.146*** 0.151*** 0.135*** 0.116*** (0.0464) (0.0426) (0.0230) (0.0214) (0.0295) (0.0278) (0.0237) GDP per capita 0.00000665 0.0000188 0.00000691 -0.0000153 -0.0000282 -0.0000261 -0.0000273 (0.0000442) (0.0000452) (0.0000239) (0.0000173) (0.0000186) (0.0000177) (0.0000177) Constant 14.81*** 14.68*** 14.47*** 14.31*** 14.28*** 14.37*** 14.47*** (0.248) (0.244) (0.203) (0.172) (0.211) (0.205) (0.167) Observations 290 279 234 208 194 181 167 Whithin R-squared 0.251 0.269 0.341 0.344 0.294 0.286 0.287
Robust standard errors in parentheses *** p < 0.01, ** p < 0.05, * p < 0.1
The three original estimations vary in their lag-sensitivity. However, the comparisons above suggest that the adaptation processes may happen in different time frames. Labour market reactions to agro-industry development appear to happen relatively fast. In contrast, the adaption of agricultural labour productivity to FB intensity changes and reactions of labour markets to productivity changes appear to require at least seven years to become apparent in the data set.
28
is only included in the second regression, may have distorting effects. Table 7 provides a direct comparison of the second regression with and without this variable.
TABLE 7: Comparison of Regression 2 with and without Gross Fixed Capital Formation in the food and beverages sector
(GFCF FB)
(1) (2)
Variables Ln(Employment FB) y+1 Ln(Employment FB) y+1
FB Intensity 0.0102*** 0.0119** (0.00276) (0.00292) Export Shares FB 0.119*** -0.000625 (0.0251) (0.0416) GFCF Shares FB -0.00152*** (0.000390)
Ln(Lab. Productivity AFF) 0.0804** 0.104*
(0.0361) (0.0495) GDP per capita 0.0000954** 0.000111** (0.0000372) (0.0000403) Constant 10.29*** 9.921*** (0.222) (0.295) Observations 209 262 Within R-squared 0.448 0.393 Number of Countries 14 15
Robust standard errors in parentheses *** p < 0.01, ** p < 0.05, * p < 0.1
Excluding the capital formation shares from the regression induces a slight magnitude increase in the intensity and labour productivity coefficients and a slight increase in their p-values. Simultaneously, the extremely significant and positive coefficient of export shares becomes totally insignificant and negative. Within R2, which is not adjusted for the number of variables, decreases by 5.5 percentage points. Overall, the original regression including capital formation shares appears to better explain impacts on FB employment while having lower values for our main variables of interest. Therefore, the original regression appears to be preferable to its counterpart.
29
regressions excluding the export shares of agricultural products for key regressions of interest. The original regression can be found in column 1 (regression 1) and column 5 (regression 3). Dropping export shares from the first regression causes p-values to go slightly beyond the 10% significance level (0.122, 0.115, 0.115) but the coefficients do not change their magnitude. Additionally, the impact of the agricultural gross fixed capital formation share becomes extremely significant and negative for the eight-year lag regression but become insignificant afterwards. Within R2 increases in direct comparison to the original regression, which is unusual when dropping a variable.
TABLE 8: Comparing the impact of excluding agricultural export shares from the first and third regression
Ln(Lab. Productivity AFF) Ln(Employment AFF)
(1) (2) (3) (4) (5) (6) (7) (8)
Variables Y+8 Y+8 Y+9 Y+10 Y+1 Y+1 Y+8 Y+8
FB Intensity 0.0194* 0.0194 0.0183 0.0173 -0.00345 -0.00537** -0.000283 -0.00110 (0.0102) (0.0118) (0.0109) (0.0103) (0.00170) (0.00184) (0.00197) (0.00226) Export Shares 0.0146 -0.00230 -0.00212 AFF (0.0209) (0.00187) (0.00174) GFCF Shares 0.000589 -0.0204*** -0.0243 -0.0166 0.0137** -0.000245 0.0143*** 0.00550 AFF (0.0187) (0.00403) (0.0155) (0.0286) (0.00587) (0.000338) (0.00334) (0.00382)
Fertiliser 1.08e-08 1.54e-08 1.98e-08 2.39e-08 Consumption (2.50e-08) (2.71e-08) (2.75e-08) (2.78e-08)
GDP per capita .000561* .000616** .000660** .000657* .0000188 .0000751 -.0000282 -.00000625 (0.000230) (0.000168) (0.000196) (0.000232) (0.0000452) (.0000725) (.0000186) (.0000208) Ln (Labour Productivity 0.0936* 0.0317 0.151*** 0.118** AFF) (0.0426) (0.0655) (0.0295) (0.0384) Constant 3.531*** 4.211*** 4.009*** 3.838*** 14.68*** 14.96*** 14.28*** 14.35*** (0.812) (0.627) (0.581) (0.572) (0.244) (0.311) (0.211) (0.310) Observations 201 209 195 181 279 293 194 201 Within R-squared 0.354 0.394 0.376 0.356 0.269 0.235 0.294 0.187
30
In the third regression excluding export shares causes FB intensity to slightly increase in magnitude and to become significant at the 5% level. In contrast, the eight-years impact of agricultural labour productivity decreases and becomes slightly less significant. In the original estimation comparison R2 does not change too much. However, for the long-term comparison R2 drops by over 10 percentage points and capital formation shares loses its significance.
Overall, these regressions might have been a viable alternative. Although FB intensity loses its already weak significance level for the first regression the p-values remain within a reasonable range and the coefficients remain of the same magnitude. Therefore, the alternative regression series slightly adds to the confidence in an existing correlation, which has a relative high error probability.
Another issue, which needs to be discussed, is causality. Correlations do not implicate any form of direction. Therefore, finding a significant positive correlation between, for instance, intensity and agricultural labour productivity does not implicate that intensity causes a productivity increase. Instead it only measures that if one increases the other one increases as well by a certain factor. A one-year lag is certainly not enough to exclude effects of reversed causality. However, the results of the longer lags appear to be sufficient to at least assume that the coefficients are not exclusively due to reversed causality, which describes a situation where the coefficient is explained by the effect the assumed dependent variable (Y) has on the assumed independent variable (X) instead of the assumed impact of X and Y. Nevertheless, it is likely that there is causality in both directions even if all relationships would have assumed one-sided causality. If there is a directed impact from FB intensity on agricultural labour productivity and productivity in turn is found to have a positive impact on FB employment, it is likely that the sector increases its output and, ceteris paribus, FB intensity is increased. Indeed, assuming the existence of multiplier effects, as the FAO does, requires causality in both directions.
31 5 Conclusion
Throughout this thesis the macro-economic impact of agro-industry development has been analysed. Agro-industry is a relatively new focus area of development strategies, which is discussed in many different dimensions. While it is rarely discussed in terms of macro-economic indicators such as aggregate labour productivity, it has become apparent that agro-industry development holds great growth potential from a structural change perspective. Existing literature mostly focus on slightly different topics such as impacts on nutritional and food security, which appears to be very reasonable in the context of the MDGs. However, only very few papers look at aggregate data and mostly compare the results across countries in a cross-section or one country over time, but none appears to combine both.
In this thesis three research questions were posed aiming at closing that gap between economic theory, research and development practice. Therefore, the impact of agro-industry development on agricultural labour productivity, agro-industrial employment and agricultural employment was analysed through constructing a new data set and by applying linear multivariate fixed effects estimations. Agro-industry development is measured by the value added share in GDP of the food and beverages sector over the same share in agriculture.
32
Nevertheless, it appears to me that agro-industry development bears excellent growth potential for developing countries having difficulties to industrialise. Additional spill-over effects, which are based in improvements alongside the agro-industry value chains such as establishing financial services, can be expected in other industries and therefore should be investigated as well. Furthermore, the low robustness of the first relationship should be further investigated by trying to identify facilitating as well as hampering factors. Further researching the impact of agro-industry development by using structured equations appears to be promising as well. To conclude, it seems that a paradigm change in the development regime is beginning. However, more emphasis should be put on testing the underlying models, especially in the macroeconomic context of developing countries. Perhaps this thesis can provide a small contribution to this or may even be an inspiration for future research.
33 6 Publication bibliography
Babu, Suresh Chandra; Shishodia, Mahika (2018): Measuring Agribusiness Competitiveness: An Application to African Countries. In Ifedapo Adeleye, Mark Esposito (Eds.): Africa’s Competitiveness in the Global Economy. Springer International Publishing, pp. 169–193. Baldwin, R. (2016): The Great Convergence. Information Technology and the New Globalization: Belknap Press of Harvard University Press.
Barrett, Christopher B. (2008): Smallholder market participation. Concepts and evidence from eastern and southern Africa. In Food Policy 33 (4), pp. 299–317.
Binswanger, Hans P.; Rosenzweig, Mark R. (1986): Behavioural and material determinants of production relations in agriculture. In The Journal of Development Studies 22 (3), pp. 503– 539.
Boserup, E. (1965): The Conditions of Agricultural Growth. The Economics of Agrarian Change under Population Pressure. London: George Allen & Unwin LTD.
Braun, Joachim von (1995): Agricultural commercialization: impacts on income and nutrition and implications for policy. In Agricultural Commercialization and Diversification 20 (3), pp. 187–202.
Braun, Joachim von (2005): Small-Scale Farmers in Liberalised Trade Environment. In Agricultural Policy Publications 38, pp. 21–52.
Cao, Kang Hua; Birchenall, Javier A. (2013): Agricultural productivity, structural change, and economic growth in post-reform China. In Journal of Development Economics 104, pp. 165– 180.
Carletto, Calogero; Corral, Paul; Guelfi, Anita (2017): Agricultural commercialization and nutrition revisited: Empirical evidence from three African countries. In Food Policy 67, pp. 106–118.
Chayanov, A. (1966): The Theory of Peasant Economy. With assistance of D. Thorner, B. Kerblay, R. Smith. Homewood, Illinois: Richard D. Irwin, Inc.
Christiaensen, Luc; Demery, Lionel; Kuhl, Jesper (2011): The (evolving) role of agriculture in poverty reduction—An empirical perspective. In Journal of Development Economics 96 (2), pp. 239–254.
Collier, Paul; Dercon, Stefan (2014): African Agriculture in 50 Years. Smallholders in a Rapidly Changing World? In World Development 63, pp. 92–101.
da Silva, C. A.; Baker, Doyle; Shephard, A.; Jenane, C.; Da Cruz, S. M. (2009): Agro-Industries for Development. Rome: FAO.
Dercon, Stefan; Gollin, Douglas (2014): Agriculture in African Development. Theories and Strategies. In Annu. Rev. Resour. Econ. 6 (1), pp. 471–492.
Diao, Xinshen; Dorosh, Paul; Shaikh Mahfuzur Rahman (2003): Market Opportunities for African Agriculture: An Examination of Demand-side Constraints on Agricultural Growth. In Development Strategy and Governance Division - Discussion Paper (1).
Diao, Xinshen; McMillan, Margaret; Rodrik, Dani (2017): The Recent Growth Boom in Developing Economies: A Structural Change Perspective. Cambridge, MA: National Bureau of Economic Research.
34
FAO; UNIDO (2015): Working group on Agro-Industry Measurement. Available online at http://www.fao.org/economic/ess/ess-events/fao-unido2015/en/, checked on 1th May 2018. Gollin, Douglas; Lagakos, David; Waugh, Michael E. (2014): The Agricultural Productivity Gap. In The Quarterly Journal of Economics 129 (2), pp. 939–993.
Goswami, Rupak; Saha, Subhrajit; Dasgupta, Purnabha (2017): Sustainability assessment of smallholder farms in developing countries. In Agroecology and Sustainable Food Systems 41 (5), pp. 546–569.
Govereh, J.; Jayne, T. S.; Nyoro, J. (1999): Smallholder Commercialization, Interlinked Markets and Food Crop Productivity. Cross-Country Evidence in Eastern and Southern Africa. In MSU International Development Working Papers.
Holt, T.; Pryor, S. (1999): Agribusiness as an Engine of Growth in Developing Countries. Humphrey, J.; Memedovic, O. (2006): Global Value Chain in the Agrifood Sector. In UNIDO Working Papers.
ILO (2007): The impact of global food chains on employment in the food and drink sector. Issues paper for discussion at the Tripartite Meeting to Examine the Impact of Global Food Chains on Employment. In Issues paper for discussion at the Tripartite Meeting to Examine the Impact of Global Food Chains on Employment.
Jaffee, S.; Kopicki, R.; Labaste, P.; Christie, I. (2003): Modernizing Africa’s Agro-Food Systems: Analytical Framework and Implications for Operations. In Africa Region Working Papers Series 44.
Lewis, W. A. (1954): Economic Development with Unlimited Supplies of Labor. In Manchester School of Economic and Social Studies 22 (2), pp. 139–191.
Margarida Duarte; Diego Restuccia; University of Toronto (2010): The Role of the Structural Transformation in Aggregate Productivity. In The Quarterly Journal of Economics 125 (1). McMillan, Margaret; Harttgen, Kenneth (2014): What is driving the 'African Growth Miracle'? In NBER Working Papers 20077.
McMillan, Margaret; Rodrik, Dani (2011): Globalization, Structural Change and Productivity Growth. In NBER Working Papers 17143.
McMillan, Margaret; Rodrik, Dani; Sepúlveda, C. (2017): Structural change, fundamentals, and growth: A framework and case studies. Washington, DC: International Food Policy Research Institute.
Page, J. (2011): Can Africa Industrialise? In Journal of African Economies 21 (suppl 2), pp. 86-124.
Pingali, Prabhu L.; Rosegrant, Mark W. (1995): Agricultural commercialization and
diversification: processes and policies. In Agricultural Commercialization and Diversification 20 (3), pp. 171–185.
Ranis, Gustav; Fei, John C. H. (1961): A Theory of Economic Development. In The American Economic Review 51 (4), pp. 533–565.
35
Reardon, Thomas; Timmer, C. Peter (2012): The Economics of the Food System Revolution. In Annu. Rev. Resour. Econ. 4 (1), pp. 225–264.
Rehber, E. (2007): Contract Farming. Theory and Practice. Punjagutta, India: The Icfai University Press.
Schultz, Theodore William (1964): Transforming traditional agriculture. In Transforming traditional agriculture.
Solow, R. M. (1956): A Contribution to the Theory of Economic Growth. In The Quarterly Journal of Economics 70 (1), pp. 65–94.
Strasberg, P. J.; Jayne, T. S.; Yanoma, T.; Nyoro, J.; Karanja, D.; Strauss, J. (1999): Effects of Agricultural Commercialization on Food Crop Input Use and Productivity in Kenya. In MSU International Development Working Papers (71).
Swinnen, J. F. M. (2005): When the market comes to you-or not: The dynamics ofvertical coordination in agri-food chains in transition. Final report of the World Bank (ECSSD) ESW on 'Dynamics of Vertical Coordination in transitionfor Policy and Bank Operations
ECA.Agrifood Chains: Implications for Policy and Bank Operations. In The World Bank (ECSSD).
Szirmai, Adam (Ed.) (2005): The Dynamics of Socio-Economic Development. Cambridge: Cambridge University Press.
The World Bank (2007): World Development Report 2008. Agriculture for development: The World Bank.
The World Bank (2008): Awakening Africa’s Sleeping Giant.
The World Bank (2017): Enabling the Business of Agriculture 2017. Washington, DC: The World Bank.
Tickner, Vincent (1995): Agricultural commercialization, economic development, and nutrition. In IFPRI publications 20 (2).
Timmer, M. P.; Los, B.; Stehrer, R.; Vries, G. de (2013): Fragmentation, Incomes and Jobs. An Analysis of European Competitiveness. In Economic Policy 28 (631-661).
UNIDO (2012): The structure and growth pattern of agro-industry of African countries. In UNIDO Working Papers (09).
Weatherspoon, D.; Reardon, T. (2003): The Rise of Supermarkets in Africa: Implications for Agrifood Systems and the Rural Poor. In Development Policy Review 21 (3).
Wiggins, Steve (2014): African Agricultural Development. Lessons and Challenges. In J Agric Econ 65 (3), pp. 529–556.
Wiggins, Steve; Argwings-Kodhek, G.; Leavy, J.; Poulton, C. (2011): Small farm
36 Appendices
I. Hausman Test Results
Appendix Figure 1: Hausman test for regression 1
Appendix Figure 2: Hausman test for regression 2
Prob>chi2 = 0.0000 = 28.35
chi2(4) = (b-B)'[(V_b-V_B)^(-1)](b-B) Test: Ho: difference in coefficients not systematic
B = inconsistent under Ha, efficient under Ho; obtained from xtreg b = consistent under Ho and Ha; obtained from xtreg GDPcapita .0005614 .0003059 .0002556 .000066 f_2 1.08e-08 3.14e-08 -2.06e-08 2.19e-08 sh_GFCF_AFF .0005889 .028684 -.028095 .0085051 sh_exp_AFF .0146489 .0070532 .0075957 .0047095 intensity_FB .0194139 .0152427 .0041712 .0035193 labprod_fe labprod_re Difference S.E.
(b) (B) (b-B) sqrt(diag(V_b-V_B)) Coefficients
Prob>chi2 = 0.9444 = 1.20
chi2(5) = (b-B)'[(V_b-V_B)^(-1)](b-B) Test: Ho: difference in coefficients not systematic
B = inconsistent under Ha, efficient under Ho; obtained from xtreg b = consistent under Ho and Ha; obtained from xtreg GDPcapita .0000954 .0000935 1.86e-06 2.47e-06 lnlabprod_~F .0803628 .0810128 -.00065 .0007258 sh_GFCF_FB -.001521 -.0015182 -2.84e-06 .0000335 sh_exp_FB .118934 .1191938 -.0002598 .002331 intensity_FB .0101969 .0101199 .000077 .0000962 L_FB_fe L_FB_re Difference S.E.
37 Appendix Figure 3: Hausman test for regression 3
II. Residual Plots for Graphical Detection of Heteroscedasticity
Appendix Figure 4: Residual Plot Regression 1
Prob>chi2 = 0.0000 = 49.01
chi2(5) = (b-B)'[(V_b-V_B)^(-1)](b-B) Test: Ho: difference in coefficients not systematic
B = inconsistent under Ha, efficient under Ho; obtained from xtreg b = consistent under Ho and Ha; obtained from xtreg GDPcapita .0000188 .0000144 4.35e-06 1.51e-06
lnlabprod_~F .09361 .0967558 -.0031458 .0013292 sh_GFCF_AFF .0136662 .013518 .0001482 .0001385 sh_exp_AFF -.0023006 -.0025555 .0002549 .0000633 intensity_FB -.0034464 -.0036551 .0002086 .0000404
L_AFF_fe L_AFF_re Difference S.E.
38 Appendix Figure 5: Residual Plot Regression 2
