You are on page 1of 11

SPC: 268 Model 5G pp. 1–11 (col.

fig: NIL)

Sustainable Production and Consumption xxx (xxxx) xxx

Contents lists available at ScienceDirect

Sustainable Production and Consumption


journal homepage: www.elsevier.com/locate/spc

What are the main factors that determine post-harvest losses of


grains?

William Eduardo Bendinelli, Connie Tenin Su , Thiago Guilherme Péra,
José Vicente Caixeta Filho
Group of Research and Extension in Agroindustrial Logistics (ESALQ-LOG), Department of Economics, Administration and Sociology, College of
Agriculture Luiz de Queiroz (ESALQ), University of Sao Paulo (USP), Av. Padua Dias, 11 - CEP 13418-900, Piracicaba, SP, Brazil

article info a b s t r a c t

Article history: Reducing post-harvest losses (PHL) permits the improvement of food security and food safety,
Received 12 April 2019 reduction of unnecessary resource use and increase of food supply chain actors’ profits. Most published
Received in revised form 17 September 2019 studies have addressed the problem qualitatively, mainly due to difficulty obtaining necessary data.
Accepted 30 September 2019
This paper seeks to understand how macroeconomic conditions influence PHL of grains (rice, maize,
Available online xxxx
soybeans and wheat), which are the main source of food for humans and animals, through the
JEL classification: construction of an econometric model using global level panel data from publicly available databases.
O13 Results suggest that increasing production to feed the increasing population often involves a difficult
Q18 trade-off. Some countries seeking on-farm production gains lack post-harvest infrastructure, especially
C5 in food storage and food marketing, which contribute to a sharp increase in the PHL level. There is
Keywords: also evidence that economic development non-linearly reduces grains’ PHL in all dimensions.
Post-harvest loss © 2019 Published by Elsevier B.V. on behalf of Institution of Chemical Engineers.
Economic development
Econometric model
Agribusiness

1 1. Introduction Parfitt et al., 2010; Shafiee-Jood and Cai, 2016). In developing and 19

transitional countries, early stage FSC loss is generally greater 20

2 There are several definitions of post-harvest losses (PHL), a than food waste, while food waste is greater than food loss in 21

3 term that can be used interchangeably with food loss and waste developed and industrialized countries (Cardoen et al., 2015a; 22

4 (FLW) (Chaboud and Daviron, 2017; Sheahan and Barrett, 2017). Gustavsson et al., 2011). Fig. 1 shows the difference in food loss 23

5 In this work, PHL are defined as the unintentional decrease in levels between developing countries (lower income levels) and 24

6 the quantity of food produced for human consumption at all developed countries (higher income levels) for grains (rice, maize, 25

7 stages of the food supply chain (FSC) regardless of the cause or soybeans and wheat). 26

8 destination,1 not including retail and final consumption stages. The reduction of PHL levels can be motivated by four main 27

9 FSC consists of a series of activities needed to deliver food through objectives. Firstly, a decrease in PHL could improve all four pillars 28

10 multiple stages (production, handling and storage, processing, of food security, through increase in food availability (quantity), 29

11 distribution, retailing) from farms to the final consumer (HLPE, improvement of food access due to reduction in prices, increase 30

12 2014; Shafiee-Jood and Cai, 2016). Gustavsson et al. (2011) esti- of food utilization owing to the decrease of qualitative PHL, and 31

13 mated that almost one-third of global food production in terms higher stability obtained by an increase in stored food (Affognon 32

14 of weight is lost. et al., 2015; Sheahan and Barrett, 2017). 33

15 The literature usually distinguishes food loss, that occurs in the Another objective is the improvement of food safety. Some 34

16 early stages of the FSC, from food waste, that happens in retail PHL occur because of fungal or pest infestations, which can cause 35

17 markets or after it reaches consumer and is generally related to severe human health implications if spoiled or contaminated food 36

18 behavior, i.e., the intentional act of a person (Flanagan et al., 2019; is consumed (Sheahan and Barrett, 2017). A third objective is to 37

increase environmental sustainability, through the reduction of 38

∗ Corresponding author. agricultural residues’ waste stream and greenhouse gas (GHG) 39

emissions (Cardoen et al., 2015b,a; Flanagan et al., 2019; Porter 40


E-mail addresses: wbendinelli@gmail.com (W.E. Bendinelli),
connietsu@esalqlog.esalq.usp.br (C.T. Su), thiago.pera@usp.br (T.G. Péra), et al., 2016) and the reduction of unnecessary resource use. It 41

jose.caixeta@usp.br (J.V. Caixeta Filho). includes agricultural input resources (water, chemical fertilizer, 42
1 There are relevant works about edible and non-edible residues’ generation agrochemicals, labor, and land) and FSC resources (fuel and en- 43

and their potential uses and waste stream (Cardoen et al., 2015b,a). ergy) (Affognon et al., 2015; Shafiee-Jood and Cai, 2016; Sheahan 44

https://doi.org/10.1016/j.spc.2019.09.002
2352-5509/© 2019 Published by Elsevier B.V. on behalf of Institution of Chemical Engineers.

Please cite this article as: W.E. Bendinelli, C.T. Su, T.G. Péra et al., What are the main factors that determine post-harvest losses of grains?. Sustainable Production and
Consumption (2019), https://doi.org/10.1016/j.spc.2019.09.002.
SPC: 268

2 W.E. Bendinelli, C.T. Su, T.G. Péra et al. / Sustainable Production and Consumption xxx (xxxx) xxx

Fig. 1. Average food loss of grains (2000–2011) versus gross domestic product per capita (2011).
Source: Data from FAOSTAT and World Bank.

1 and Barrett, 2017). Kummu et al. (2012) found that around one- Nations Educational, Scientific and Cultural Organization (UN- 34

2 quarter of the consumed water, cropland, and fertilizer used ESCO) also provide access to indicators related to PHL. Data from 35
3 to produce food have been wasted. Lastly, PHL reduction can these sources can be used to estimate an econometric model to 36
4 lead to economic benefits and well-being for FSC actors, through
test hypotheses regarding PHL and the effects of its determinants. 37
5 increased profits or lower costs (Sheahan and Barrett, 2017).
Non-perishable food crops, such as cereal grains, oilseeds and 38
6 PHL can be attributed to different FSC related processes and
7 operations (e.g., harvesting, drying, storage, transportation) or legumes (hereinafter referred to as grains), constitute the ba- 39

8 to external forces that induce PHL (e.g., market economy, re- sic elements for human diets in most developing and transi- 40

9 source limitations, climate, legislation and cultural differences) tional countries and are important products for smallholder farms 41

10 (Parfitt et al., 2010; Shafiee-Jood and Cai, 2016). Thus, PHL can be (World Bank, 2011). Grains are also the main component in most 42
11 analyzed through economic, cultural, social, and environmental animal feed diets, thus having the highest crop production among 43
12 perspectives. Economic elements are divided into micro, which
food commodity groups (FAO, 2018a; Neethirajan et al., 2007). 44
13 reflects the behavior of individual actors in the FSC, and macro,
Maize, rice, wheat, and soybeans have the highest production of 45
14 where the aggregated variables are studied instead of the single
15 operator’s. Micro and macroeconomic dimensions are interre- all the grains, respectively, and are important in international 46

16 lated, as the aggregated variables can be described as the sum food markets (AMIS, 2018; FAO, 2018b). It is estimated that 47

17 of the individual economic actors’ variables (Segrè et al., 2014). grains’ PHL vary from 20% to 35% across the world’s different 48

18 Several qualitative published works focused on the global geographical regions (Gustavsson et al., 2011). 49
19 determinants of PHL (Chaboud and Daviron, 2017; Godfray et al., The goal of this paper is to understand how macroeconomic 50
20 2010; Hodges et al., 2011; Parfitt et al., 2010; Rutten, 2013;
conditions influence grains’ (rice, maize, soybeans and wheat) 51
21 Segrè et al., 2014; Shafiee-Jood and Cai, 2016; Sheahan and Bar-
PHL, through the construction of an econometric model using 52
22 rett, 2017; World Bank, 2011), but only a handful of works
23 present international quantitative estimates of PHL determinants’ global level panel data with the necessary corrections. The re- 53

24 coefficients (KC et al., 2016; Rosegrant et al., 2015). Data on sults should indicate the main macroeconomic determinants of 54

25 PHL are scarce, old, mostly unreliable, and often not comparable PHL, contributing to a better understanding of this subject, and 55

26 (Affognon et al., 2015; Chaboud and Daviron, 2017; Gustavsson stimulating new studies and policies. 56
27 et al., 2011). Methods based on direct measurement are accurate This paper is structured in the following order. Section 2 57
28 but demand high levels of effort and cost, making it difficult to
presents the macro-economic conditions that impact PHL. Sec- 58
29 measure the PHL of several food crops in numerous countries
tion 3 shows the research design, detailing the selected data, the 59
30 (World Resources Institute, 2016).
31 On the other hand, Food and Agriculture Organization of the econometric model and the estimation strategy. Section 4 con- 60

32 United Nations Statistics (FAOSTAT) provides access to a large tains results, discussions, and the robustness checks performed. 61

33 food and agriculture database. The World Bank and the United Lastly, Section 5 presents the concluding remarks. 62

Please cite this article as: W.E. Bendinelli, C.T. Su, T.G. Péra et al., What are the main factors that determine post-harvest losses of grains?. Sustainable Production and
Consumption (2019), https://doi.org/10.1016/j.spc.2019.09.002.
SPC: 268

W.E. Bendinelli, C.T. Su, T.G. Péra et al. / Sustainable Production and Consumption xxx (xxxx) xxx 3

1 2. Macro-economic conditions Bennett’s Law (Bennett, 1941), in combination with deficiency of 63

infrastructure, tends to increase food loss and waste, especially 64

2 2.1. Infrastructure of shorter shelf-life items (Cardoen et al., 2015a; Godfray et al., 65

2010; Parfitt et al., 2010; Rae, 1997; Segrè et al., 2014; Sheahan 66

3 In developing and transitional countries, the main causes of and Barrett, 2017), and decrease starchy staple food loss. The Ben- 67

4 food loss are related to outdated harvest techniques and lim- nett’s Law states that the food share of starchy staples decreases 68

5 ited post-harvest handling and infrastructure. The improvement with an increase in income (Bennett, 1941). 69
6 of harvest techniques without the proper infrastructural invest- The Malthusian theory addresses the dynamics between pop- 70
7 ments in food storage and transportation can lead to the creation ulation and economic growth. Its central hypothesis states that 71
8 of food surplus, which can culminate in lower food prices and in technical improvements and resource expansion in the preindus- 72
9 higher food losses due to difficulties in the conservation of the trial era had generated temporary increases in income per capita, 73
10 oversupply. In addition, there will be higher food losses where while leading to a larger but not richer population. Recently, 74
11 the links between producers and consumers are frail (Cardoen the Malthusian urban dynamics describes the urban growth in 75
12 et al., 2015a; Hiç et al., 2016; Hodges et al., 2011; Kaminski and developing countries mainly due to migration of population ‘‘ex- 76
13 Christiaensen, 2014; Liu et al., 2013; Parfitt et al., 2010; Segrè pulsed’’ from rural areas. When the population growth is not 77
14 et al., 2014; Shashi et al., 2017). appropriately accompanied by investments in urban infrastruc- 78
15 On the other hand, developed and industrialized countries ture, it leads to congestion costs that surpasses the benefits from 79
16 have an advanced post-harvest infrastructure, thus their food agglomeration (Ashraf and Galor, 2011; Castells-Quintana, 2017; 80
17 loss is extremely low; although, the food surplus and relatively Weil and Wilde, 2009). 81
18 low cost of food in these countries do foster food waste. This
19 more efficient infrastructure system led to a substantial growth 2.3. Trade and globalization 82
20 of the food-processing sector, allowing producers to diversify
21 their incomes, while increasing the distance between economic
Increased global agricultural trade or increased receipt of food 83
22 agents (Hodges et al., 2011; Parfitt et al., 2010; Segrè et al., 2014;
aid can either reduce or increase PHL in developing and tran- 84
23 Thyberg and Tonjes, 2016). Table 1 provides an overview of the
sitional countries. New opportunities for international agricul- 85
24 characteristics of PHL according to a country’s stage of economic
tural product trade can lessen PHL (Godfray et al., 2010; Parfitt 86
25 development.
et al., 2010) as new clients bring increasing sales volume and 87
26 Food marketing infrastructure plays an important role in PHL
the demands of international trade require food and packaging 88
27 (Hodges et al., 2011; Sheahan and Barrett, 2017). Better roads
standardization, especially for the use of containerized shipments, 89
28 generate lower transport costs and improve access to markets,
which reduces product degradation and lowers PHL. Interestingly, 90
29 which could accelerate trade and reduce the need for on-farm
globalization could lower PHL due to the increase in efficiency of 91
30 storage. Better access to electricity could improve local drying
the whole FSC, even considering the increase in the number of lo- 92
31 and processing. Better access to local markets, combined with
gistic operations and longer travel times. Global food production 93
32 policies focused on increasing consumer purchasing power, could
efficiency could also increase because of regional specialization in 94
33 increase the domestic demand, whilst the creation of adequate
the production of the locally most suitable crops (Godfray et al., 95
34 relations to the international markets could help absorb the entire
2010). 96
35 offer (Godfray et al., 2010; KC et al., 2016; Segrè et al., 2014;
Alternatively, trade globalization and food aid can raise PHL 97
36 Sheahan and Barrett, 2017; World Bank, 2011).
due to food price reduction caused by the importation of in- 98
37 Not only can poor infrastructure cause losses through prod-
expensive, high quality imports that can put producers in the 99
38 uct deterioration and spillage, it can also slow the adoption or
destination markets at a disadvantage and reduce their incentive 100
39 scaling-up of technologies that potentially reduce PHL (Affognon
to manage PHL, even in poorer countries. Globalization also puts 101
40 et al., 2015; Rosegrant et al., 2015). For instance, in Bihar, India,
local distributors at a disadvantage as multi-national supermarket 102
41 the adoption of cold storage practices significantly increased after
42 the local government improved physical and social infrastructure. chains expand into developing and transitional countries, which 103

43 The government there reformed policy, increased enforcement of impacts FSC dynamics and, consequently, PHL (Parfitt et al., 2010; 104

44 the rule of law, strengthened governance, and invested in infras- Shafiee-Jood and Cai, 2016). 105

45 tructure, particularly roads, thereby enhancing market access for


46 farmers in remote areas (Minten et al., 2014; Rosegrant et al., 2.4. Food price inflation and unemployment 106

47 2015; Shashi et al., 2017; Sheahan and Barrett, 2017).


Food price inflation, as well as international food price peaks 107

48 2.2. Urbanization and socio-economic growth (for instance, during the 2007/2008 and the 2011/2012 food 108

crisis), can cause a decrease or a change in the food consumption, 109

49 Economic development normally results in rapid urbanization, especially for groups with limited purchasing power (Chaboud 110

50 agricultural sector contraction and changes in the employment and Daviron, 2017; Rosegrant et al., 2015; Rutten, 2013; Segrè 111

51 structure. Urbanization demands that the FSC be extended to et al., 2014). In the short term, a reduction in the food demand can 112

52 provide sufficient affordable food to meet the needs of a growing cause an imbalance with the food supply, leading to higher food 113

53 urban population. However, FSC efficiency and PHL levels rely losses. A re-orientation of consumption leads to a higher demand 114

54 heavily on road, rail, and market infrastructure improvements for staple foods and inexpensive food products that are usually 115

55 (Aldred, 2012; Parfitt et al., 2010; Segrè et al., 2014; Shashi et al., high in calories and fat, with a possible progressive reduction in 116

56 2017). the demand for perishable and other more expensive food items 117

57 Furthermore, these modifications in employment structure (Segrè et al., 2014). 118

58 frequently lead to changes in the social organization, in consump- Even though global gross food prices have fallen over the past 119

59 tion habits and in the relationship with food. As urbanization century, the food price peaks are an indication of a forthcoming 120

60 and household income grow simultaneously, the consumption of period of rising and more volatile food prices, driven primar- 121

61 non-perishable food declines while the consumption of perish- ily by increased demand from rapidly developing countries and 122

62 able food increases. This dietary transition, which conforms to first-generation biofuels production (Godfray et al., 2010). 123

Please cite this article as: W.E. Bendinelli, C.T. Su, T.G. Péra et al., What are the main factors that determine post-harvest losses of grains?. Sustainable Production and
Consumption (2019), https://doi.org/10.1016/j.spc.2019.09.002.
SPC: 268

4 W.E. Bendinelli, C.T. Su, T.G. Péra et al. / Sustainable Production and Consumption xxx (xxxx) xxx

Table 1
Characteristics of PHL according to a country’s stage of technological/economic development.
Source: Adapted from Parfitt et al. (2010) and Hodges et al. (2011).
Analysis Developing countriesa Transitional countries Developed industrialized countries
dimension
Bottleneck in the On-farm (mainly storage). From farm to the wholesale stage Consumer.
supply chain (link) (mainly transportation).
Supply chain Poor integration with local markets, many Tighter integration of growers, Use of highly integrated systems between
characteristics intermediaries suppliers, processors and distribution growers and supply chain; more seasonal
supplying urban markets. systems is required. produce imported; more secondary
processing of food.
Type of growers Smallholders, including subsistence farmers. Small-scale farmers who often have Medium- and large-scale farmers.
access to
limited post-harvest specific
infrastructure.
Markets and Local markets: mostly meeting Produce of variable quality, target Meet quality, safety, volume and timeliness
quality household/village food requirements; both local demands of local retailers (particularly
limited access to international markets. (including supermarkets) and, supermarkets/convenience store chains) and
increasingly, export markets in a export markets.
number of countries.
Technological Simple technologies, labor-intensive, Packing houses, refrigeration and Access to relatively sophisticated
development traditional storage systems and harvesting storage facilities systems side-by-side technologies, e.g., packing-house equipment
techniques. with elements of traditional systems. and cold chains; losses still occur;
harvesting highly mechanized.
Incentives Loss reductions have directly improved Combination of developing and Need for financial incentives to reduce
livelihoods, food security, food quality and developed country initiatives, losses and waste since the low-margin
health benefits. Improvements in depending on the country’s industries and low proportion of spending
mechanization free up time to spend on technological level and the market. by families on food do not encourage
more profitable off-farm activities. agents to cut losses or penalize for
overproduction and waste.
Strategies Wide education of farmers on the main Consumer education campaigns, waste
causes of losses, infrastructure tariffs and the development of
improvements to connect small farmers to public–private partnerships (PPPs) to
markets, more effective supply chain with reduce and divide responsibility for losses
access to microcredit to improve the and waste.
technological level and public–private
partnerships (PPPs) sharing investment
costs and risks in the market targeted
interventions.
a
Hodges et al. (2011) use different nomenclature for countries: less developed countries, intermediate developing countries and developed countries, respectively.
These three categories are also equivalent to the classification according to the country’s Gross Domestic Product (GDP) per capita, that is, in countries of low, middle
and high income per capita.

1 In addition, a predictable price and policy environment could Despite all the problems, the FBS do provide an approximate 26

2 support investments in off-farm storages, providing affordable picture of the overall food situation in a country and can be of 27

3 drying and storage services to smallholders (World Bank, 2011). use in economic and nutritional studies, in development plans, 28

4 Unemployment causes a decrease in the consumer purchas- and in the formulation of related projects (FAO, 2001). 29

5 ing power, leading to a reduction and a re-orientation in food In this current study, the dependent variable is the PHL per- 30

6 consumption, with an effect similar to food price inflation. How- centage, calculated according to the formula utilized in Gustavs- 31

7 ever, its impacts in the formation of PHL might have different son et al. (2013): 32

8 consequences, with higher food losses due to a less efficient FSC Losses Losses
9 and lower consumer food waste owing to the need of better food %PHL = = (1) 33

10 management at the household level (Segrè et al., 2014). Supply Production + Import + Stock Variation
All data used to calculate the dependent variable were from 34

11 3. Research design the FBS. The loss data available in the FBS comprise the food loss 35

at all stages after harvest and prior to consumption, i.e., losses 36

12 3.1. Data during storage and transportation. The food supply is equal to 37

the sum of production (primary crops), import quantity and stock 38

13 The dataset consists of panel data from 2000 through 2011 for variation, i.e., changes between the opening and closing stocks at 39

14 82 countries with at least one percent of their geographical re- all levels, but mostly of stocks held by governments (FAO, 2001; 40

15 gion’s domestic supply of grains. The domestic supply data were Gustavsson et al., 2013). 41

16 obtained from the FAO Food Balance Sheets (FBS at FAOSTAT, The independent variables were selected based on the macroe- 42

17 2017). conomic conditions explored in Section 2 and will be presented 43

18 The FBS present a comprehensive picture of a country’s food in Section 3.2. All data utilized in this study are publicly available 44

19 supply and its utilization for each food item during a specified from FAOSTAT, the World Bank, and UNESCO. 45

20 reference period. They have been published since 1945 and are Observations with at least one missing data and/or outliers 46

21 continually improved. The FBS data are assembled from a variety were excluded, because missing data would cause the matrix to 47

22 of sources, so data quality and data coverage varies considerably be invertible and then the parameters could not be calculated. 48

23 among countries and commodities (FAO, 2001). Also, adjustments These exclusions resulted in an unbalanced dataset as countries 49

24 to the FBS basic data and missing data filling are needed to main- may have a different number of observations. In order to min- 50

25 tain a certain degree of consistency, completeness, and reliability. imize the number of excluded observations, missing data was 51

Please cite this article as: W.E. Bendinelli, C.T. Su, T.G. Péra et al., What are the main factors that determine post-harvest losses of grains?. Sustainable Production and
Consumption (2019), https://doi.org/10.1016/j.spc.2019.09.002.
SPC: 268

W.E. Bendinelli, C.T. Su, T.G. Péra et al. / Sustainable Production and Consumption xxx (xxxx) xxx 5

1 filled with data from the same country but from a different pe- et al., 2016), e.g., rural area with better access to markets 59

2 riod; or when there was no individual country data, with adapted and higher percentage of paved roads. Source: Suite of Food 60

3 data from figures coming out of the country’s geographical region. Security Indicators (FAOSTAT, 2017). 61

4 For some independent variables, both country and regional data • ln railit : it is the natural logarithm of the rail lines density 62

5 were unavailable, which made that specific dataset inestimable of country i in year t (unit of measure: km of route per 63

6 and its input impossible. All data over three standard deviations 100 square km of land area). This variable measures the 64

7 were considered as outliers. possibility of physical access to markets (FAOSTAT, 2017). 65

It also expresses part of a country’s logistics infrastructure 66

8 3.2. Econometric model (Liu et al., 2013). Source: Suite of Food Security Indicators 67

(FAOSTAT, 2017). 68

9 Eq. (2) presents the empirical PHL model: • ln urbanit : it is the natural logarithm of the urban population 69

of country i in year t (unit of measure: % of total population).


ln yit = β0 + ln Xit β + dummiesit β + µg + λt + νit
70
10
An increase in the rate of urbanization demands extended 71
11 i = 1, . . . , N ; t = 1, . . . , T (2) food supply chains, which has implications on PHL (Parfitt 72

et al., 2010). Source: UNESCO (2017). 73


12 where ln yit is the natural logarithm percent of grains postharvest
• ln incomeit : it is the natural logarithm of the gross domestic
loss in the country i in year t, β0 is the constant term, ln Xit is the
74
13
product (GDP) per capita of country i in year t (unit of 75
14 set of natural logarithm explanatory variables in the country i in
measure: constant 2011 international $). This variable mea- 76
15 year t, dummiesit is the set of dummy explanatory variables in the
sures the economic possibility of market access (FAOSTAT,
country i in year t, β is the set of regression coefficients of the
77
16
2017) and expresses the economic development level of a
explanatory variables with respect to the dependent variable, µg
78
17
country. Also, growth in household income, which is highly
is the unobservable group-specific effect, λt is the unobservable
79
18
related to growth of GDP per capita, is associated with a
time effect and νit is the remainder stochastic disturbance term
80
19
dietary transition from starchy food staples to perishable
with νit ∼ IID(0, σv2 ).
81
20
food (Parfitt et al., 2010). Source: Suite of Food Security 82
21 The natural logarithm of the dependent and the explanatory
Indicators (FAOSTAT, 2017). 83
22 variables was used in order to capture the nonlinearities of the
• ln tradeit : it is the natural logarithm of the sum of the base 84
23 problem and to directly calculate the coefficients’ elasticities,
price value of agricultural and food products imports and 85
24 since the estimated parameters would already represent them.
exports divided by the agriculture GDP of country i in year t 86
25 In this way, one more estimation step was avoided.
(unit of measure: %). This variable measures the country’s 87
26 The variables that best represent the macroeconomic con-
openness to international trade of agricultural and food 88
27 ditions that are the basis for the PHL determinants’ analysis
products, which influences PHL (Parfitt et al., 2010). Sources: 89
28 were selected as the explanatory variables (Chaboud and Davi-
Trade Indices (FAOSTAT, 2017) and World Bank (2017) . 90
29 ron, 2017; Godfray et al., 2010; Hiç et al., 2016; Kaminski and
• ln priceit : it is the natural logarithm of the domestic food 91
30 Christiaensen, 2014; KC et al., 2016; Liu et al., 2013; Parfitt et al.,
price volatility of country i in year t (unit of measure: index). 92
31 2010; Rosegrant et al., 2015; Rutten, 2013; Segrè et al., 2014;
This variable expresses the variability in the relative price of 93
32 Shafiee-Jood and Cai, 2016; Sheahan and Barrett, 2017; Thyberg
food in a country and reflects food price inflation (FAOSTAT, 94
33 and Tonjes, 2016; World Bank, 2011).
2017). Source: Suite of Food Security Indicators (FAOSTAT, 95

34 • ln %PHLit : it is the natural logarithm of the losses quantity 2017). 96

35 divided by the sum of production, imports and stock vari- • crisis : equal to 1 if year t ≥ 2008 and t ≤ 2011, 0 otherwise. 97

36 ation of grains of country i in year t (unit of measure: %). This is a dummy variable that represents the 2008–2011 98

37 Source: Food Balance Sheets (FAOSTAT, 2017). economic crisis and food price spikes. 99

38 • ln supplyit : it is the natural logarithm of the dietary energy • low _incomeit , low _middle_incomeit , upper_middle_incomeit 100

39 supply per capita divided by the average dietary energy re- and high_incomeit : dummy variables that represent four 101

40 quirement per capita of country i in year t (unit of measure: clusters of incomeit , generated by the software Stata with 102

41 %). It expresses average dietary energy supply adequacy and medians as the basis. 103

42 measures a country’s food surplus or deficit, which reflects • low _tradeit and high_tradeit : dummy variables that repre- 104

43 its infrastructure (FAOSTAT, 2017). Source: Suite of Food sent two clusters of tradeit , generated by the software Stata 105

44 Security Indicators (FAOSTAT, 2017). with medians as the basis. 106

45 • ln food_sectorit : it is the natural logarithm of the food net • low _urbanit , middle_urbanit and high_urbanit : dummy vari- 107

46 production value per capita of country i in year t (unit ables that represent three clusters of urbanit , generated by 108

47 of measure: international $2 ). This variable expresses the the software Stata with medians as the basis. 109

48 relative economic size of the food production sector in the


In the robustness check of results (Section 4.3), the follow- 110
49 country and it is an indication of infrastructure and de-
ings dependent variables were also considered as alternative 111
50 velopment (FAOSTAT, 2017). Source: Suite of Food Security
specification of ln %PHLit : 112
51 Indicators (FAOSTAT, 2017).
52 • ln roadit : it is the natural logarithm of the road density of • ln PHLit : it is the natural logarithm of the losses quantity of 113
53 country i in year t (unit of measure: km of road per 100 grains of country i in year t (unit of measure: tons). Source: 114
54 square km of land area). This variable measures the possi- Food Balance Sheets (FAOSTAT, 2017). 115
55 bility of physical access to markets (FAOSTAT, 2017). It is • ln PHL_capitait : it is the natural logarithm of the losses quan- 116
56 considered as a proxy of a country’s logistics infrastructure tity of grains divided by the population of country i in year t 117
57 and it is assumed that a country with a higher road density (unit of measure: kg per capita). Source: Food Balance Sheets 118
58 has a higher probability of an efficient transportation (KC and Suite of Food Security Indicators (FAOSTAT, 2017). 119

2 ‘‘An international dollar would buy in the cited country a comparable Table 2 presents descriptive statistics for the dataset with- 120

amount of goods and services a U.S. dollar would buy in the United States’’ out outliers (556 observations). Observations with at least one 121

(World Bank, 2017). missing data were not excluded in the descriptive statistics. 122

Please cite this article as: W.E. Bendinelli, C.T. Su, T.G. Péra et al., What are the main factors that determine post-harvest losses of grains?. Sustainable Production and
Consumption (2019), https://doi.org/10.1016/j.spc.2019.09.002.
SPC: 268

6 W.E. Bendinelli, C.T. Su, T.G. Péra et al. / Sustainable Production and Consumption xxx (xxxx) xxx

1 3.3. Estimation strategy shown in the robustness checks (Section 4.3), the dependent vari- 61

able ln PHL_capitait was used, maintaining the independent vari- 62

2 First, panel unit roots tests were performed to avoid spuri- able ln supplyit in the specifications (Table 5). Although the cor- 63

3 ous regression. In the context of panel data unit-root tests, a relation between the main dependent variable ln %PHLit and the 64

4 Fisher-type test performs a unit-root test on each panel’s series independent variable ln supplyit is low (−0.20), these approaches 65

5 separately, and then combines the p-values to obtain an overall allow to minimize this potential endogeneity problem. 66

6 test of whether the panel series contains a unit root (Choi, 2001). In addition, when the independent variable ln incomeit is used, 67

7 All the tests subtract the cross-sectional averages from the series the related independent variables ln priceit and ln tradeit were 68

8 to mitigate the impact of cross-sectional dependence. All series excluded (specification (1) of Table 3). However, the GDP per 69

9 rejected the null hypothesis that all panels contain a unit root of capita information was transformed in four dummy variables 70

10 at least 10 percent. As the panels are trend or drift stationary, it indicating a country’s income level, and when using these dummy 71

variables instead of the original variable, it is possible to tackle 72


11 is not necessary to employ panel cointegration tests.
the endogeneity problem and consider the independent variables 73
12 In the residuals, tests of panel heteroskedasticity and au-
ln priceit and ln tradeit in the model (specifications (3) to (6) of 74
13 tocorrelation were performed. It is important to differentiate
Table 3). 75
14 heteroskedasticity across panels and autocorrelation within pan-
15 els. It is common to have different variance for each panel when
4.2. Discussions 76
16 working with data from many countries; however, autocorre-
17 lation is something to be wary of. It is reasonable to assume
Table 3 contain some important findings. The main one is that 77
18 that laws and policies affect people’s behavior, both in the short
there are evidences to find quantitatively the existence of the 78
19 and long terms; and although these laws and policies may have relationships between PHL and the aforementioned interrelated 79
20 positive externalities derived from other countries, their effect macroeconomic conditions. The obtained results are consistent 80
21 is stronger within the country of their origin. Because of this, a with the literature on PHL. The estimation strategy aimed at 81
22 unique correlation parameter for each panel was structured. reducing modeling biases and generating more calibrated pa- 82
23 Following Green (2000), a modified Wald statistic to test rameters, which provides a better indication of intensity, both 83
24 for group-wise heteroskedasticity in the fixed effect regression relatively and absolute, and sign of the PHL determinants’ coeffi- 84
25 model’s residuals was employed. The null hypothesis of ho- cients. 85
26 moscedastic disturbances was strongly rejected. A Wooldridge Considering the baseline models (columns (4) and (6) of Ta- 86
27 test for autocorrelation in panel data was also implemented. This ble 3), the highest impact on PHL are due to variation on GPD 87
28 test indicated the presence of first-order autocorrelation. The per capita. An increase on a country’s GPD per capita leads to 88
29 procedure designed by Newey–West was employed to adjust the an increase in economic access to food for the poorest, a dietary 89
30 panels’ standard error estimates. transition from starchy food staples to perishable food, economic 90
31 It is important to emphasize that, to the best of found knowl- development, and investments on logistics infrastructure (Liu 91

32 edge, this study is a pioneer in that it takes into account a world et al., 2013), intensely reducing grains PHL. This impact is non- 92

33 data panel and corrections to reduce parameter bias. There were linear and varies according to economic development cluster and 93

34 only two other works that attempted to quantify the determi- culture, as it can be seen in Table 3 (dummy variables low _income, 94

35 nants of PHL worldwide (KC et al., 2016; Rosegrant et al., 2015). low _middle_income and upper_middle_income). In order to test 95

36 Both works present parameters with higher bias and a higher the non-linear relationship between economic development and 96

37 probability of type-II error due to the lack of adequate correction PHL reduction, the squared value of the independent variable 97

38 and the omission of relevant variables. ln incomeit was added (variable ln_square_income). Thus it was 98

found the relationship is indeed non-linear and it is compliant 99

with the Kuznets’ curve, in which the environmental degradation 100


39 4. Results and discussions
increases with the income per capita, until the inflexion point, 101

and then gradually decreases (Madden et al., 2019), as it can be 102


40 4.1. Results
seen in specifications (1) and (2) of Table 3. In addition, the non- 103

linearity was shown with the use of logarithm on both dependent 104
41 Table 3 presents estimation results from the empirical PHL and independent variables. 105
42 determinants model. Six different specifications are considered. Urbanization (variable ln_urban) has a moderate effect3 on 106
43 In the first, there are no corrections so that the results can be reducing PHL. Its several dimensions, such as agricultural sector 107
44 compared with those from the studies noted in Section 3.3. In contraction, pressure on infrastructure, and more efficient supply 108
45 specification (2), time fixed effects correction is added. In specifi- chains, contribute to reduce PHL. When comparing countries 109
46 cations (3) and (4), one-way and two-way fixed effect correction that have high levels of urbanization with countries that have 110
47 are added, respectively, and the explanatory variable ln incomeit low (dummy variable low _urban) or medium (dummy variable 111
48 is replaced with its four dummy variables, using high_incomeit middle_urban) levels of urbanization, one finds that the less ur- 112
49 as the base group. Lastly, the two-way fixed effect correction is banized countries show a 29.9 percent and 23.6 percent higher 113

50 kept in specifications (5) and (6) while adding the trade openness level of PHL, respectively. 114

51 dummies to specification (5) and the urbanization rate dummies The 2008–2011 crisis and the country’s openness to inter- 115

52 to specification (6) to estimate how these economic develop- national trade of food have a moderate impact on PHL. Lack 116

53 ment indicators affect PHL percentages. Clusters high_tradeit and of stability due to the 2008–2011 crisis (dummy variable crisis) 117

54 high_urbanit are the base groups in (5) and (6). Columns (4) and increased PHL by affecting consumer purchasing power and by 118

55 (6) contain the baseline models. putting investments on hold, e.g. in off-farm storages. 119

56 As both the main dependent variable ln %PHLit and the inde- There is some evidence that countries that are more integrated 120

57 pendent variable ln supplyit contains food production information, into global markets suffer fewer PHL (variable ln_trade). As it is 121

58 two approaches were used: first, the dependent variable ln %PHLit


59 was selected, but the independent variable ln supplyit was ex- 3 In this work, moderate effect is considered when the estimated coefficient
60 cluded from the specifications, as shown in Table 3. Secondly, as is between 0.15 and 0.40, while low effect occurs when it is lower than 0.15.

Please cite this article as: W.E. Bendinelli, C.T. Su, T.G. Péra et al., What are the main factors that determine post-harvest losses of grains?. Sustainable Production and
Consumption (2019), https://doi.org/10.1016/j.spc.2019.09.002.
SPC: 268

W.E. Bendinelli, C.T. Su, T.G. Péra et al. / Sustainable Production and Consumption xxx (xxxx) xxx 7

Table 2
Descriptive statistics for the dependent and explanatory variables (without outliers).
Variables (1) (2) (3) (4) (5) (7) (8) (9) (10) (11) (12)
Pearson correlation
%PHL (1) 1.00
PHL (2) 0.14 1.00
PHL_capita (3) 0.52 0.08 1.00
supply (4) −0.20 −0.07 −0.01 1.00
food_sector (5) −0.33 −0.05 0.19 0.53 1.00
road (7) −0.42 −0.01 −0.20 0.28 0.35 1.00
rail (8) −0.42 −0.14 −0.07 0.31 0.31 0.80 1.00
income (9) −0.58 −0.17 −0.18 0.63 0.60 0.56 0.57 1.00
price (10) 0.29 −0.01 0.19 −0.14 −0.24 −0.22 −0.20 −0.32 1.00
trade (11) −0.39 −0.22 −0.12 0.37 0.38 0.54 0.64 0.60 −0.25 1.00
urban (12) −0.41 −0.17 −0.02 0.57 0.53 0.23 0.34 0.72 −0.23 0.44 1.00

Univariate statistics
Mean 0.05 1,258,122 16.06 123.85 324.31 49.77 2.09 15,359 9.20 130.18 59.45
Standard deviation 0.03 3,046,840 18.34 15.31 168.71 55.89 2.35 12,373 5.50 121.01 20.49
Minimum 0.00 17,467 1.03 89.00 63.00 1.20 0.00 808 0.00 8.08 12.25
Maximum 0.15 23,500,000 234.35 158.00 921.00 216.00 10.30 51,011 43.80 678.18 91.89

Table 3
Estimation results for the empirical model of PHL determinants.
Variables (1) (2) (3) (4) (5) (6)
% PHL % PHL % PHL % PHL % PHL % PHL
ln_food_sector 0.152*** 0.0364 0.261*** 0.205*** 0.0263 −0.0544*
(0.0394) (0.0636) (0.0426) (0.0346) (0.0364) (0.0309)
ln_road −0.0860*** −0.0139 0.00708 0.00803 −0.00604 −0.00423
(0.0217) (0.0200) (0.0182) (0.0152) (0.0166) (0.0172)
ln_rail −0.0174 −0.0538** −0.0989*** −0.0865*** −0.0589*** −0.0740***
(0.0192) (0.0259) (0.0179) (0.0142) (0.0174) (0.0200)
ln_income 2.017*** 2.899***
(0.330) (0.344)
ln_square_income −0.135*** −0.184***
(0.0184) (0.0199)
ln_price 0.0595*** 0.0715*** 0.0873*** 0.0814***
(0.0144) (0.0152) (0.0167) (0.0171)
ln_trade −0.216*** −0.204***
(0.0184) (0.0184)
ln_urban −0.312*** −0.113 −0.337*** −0.338*** −0.459***
(0.0563) (0.0828) (0.0478) (0.0365) (0.0481)
crisis 0.0670*** 0.233*** 0.00552 0.138*** 0.332*** 0.292***
(0.0108) (0.0390) (0.0164) (0.0417) (0.0407) (0.0387)
low_income 0.430*** 0.452*** 0.673*** 0.731***
(0.0713) (0.0672) (0.0633) (0.0769)
low_middle_income 0.481*** 0.501*** 0.752*** 0.706***
(0.0627) (0.0573) (0.0496) (0.0701)
upper_middle_income 0.337*** 0.319*** 0.592*** 0.534***
(0.0612) (0.0575) (0.0576) (0.0684)
low_trade 0.249*** 0.239***
(0.0302) (0.0308)
low_urban 0.299***
(0.0621)
middle_urban 0.236***
(0.0522)
constant −9.677*** −14.03*** −2.893*** −2.724*** −2.659*** −4.192***
(1.392) (1.418) (0.248) (0.227) (0.302) (0.207)
Group fixed effect No No Yes Yes Yes Yes
Time fixed effects No Yes No Yes Yes Yes
Wald χ 2 (p-value) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Nr observations 546 546 534 534 534 534
Nr countries 69 69 68 68 68 68

Notes:
1. Results produced by feasible generalized least squares (FGLS); statistics robust to heteroskedasticity and serial correlation
where the correlation parameter is unique for each panel.
2. Standard errors of the estimated coefficients (in brackets).
3. P-value representations: ***p < 0.01, **p < 0.05, *p < 0.10.

1 known, trade on a global scale opens opportunities for the expor- with a low level of global trade (dummy variable low _trade) have 8

2 tation of agricultural products, and investment on a global scale 23.9 percent more PHL when compared to countries with high 9

3 has led to the establishment of large, internationally connected levels of global trade. This is most likely related to a greater use 10

4 supermarkets in several countries. Although economic openness of standardized operations and packaging procedures to meet 11

5 may threaten the development of domestic markets, the model global trading requirements in countries more interconnected 12

6 suggests that there is a positive net benefit relative to PHL that with global markets, and these procedures lead to reduced PHL 13

7 is related to greater economic openness. For example, countries (Jedermann et al., 2014). 14

Please cite this article as: W.E. Bendinelli, C.T. Su, T.G. Péra et al., What are the main factors that determine post-harvest losses of grains?. Sustainable Production and
Consumption (2019), https://doi.org/10.1016/j.spc.2019.09.002.
SPC: 268

8 W.E. Bendinelli, C.T. Su, T.G. Péra et al. / Sustainable Production and Consumption xxx (xxxx) xxx

Table 4
Robustness checks: alternative concepts of dependent variable (ln PHLit ) and alternative selection of independent variables.
Variables (1) (2) (3) (4) (5) (6) (7)
% PHL PHL PHL PHL PHL PHL PHL
ln_supply 1.422*** 1.305*** 1.563***
(0.342) (0.393) (0.378)
ln_food_sector −0.0544* −0.00255 0.464*** 0.188** 0.0639 0.345*** 0.163**
(0.0309) (0.0760) (0.0867) (0.0863) (0.0738) (0.0645) (0.0785)
ln_road −0.00423 0.0390 0.0570 0.0713 0.114*** 0.0971** 0.0551
(0.0172) (0.0411) (0.0446) (0.0443) (0.0312) (0.0403) (0.0438)
ln_rail −0.0740*** −0.0384 0.0529 0.151*** −0.0284 0.131*** 0.185***
(0.0200) (0.0371) (0.0479) (0.0429) (0.0351) (0.0381) (0.0414)
ln_income 7.251*** 8.122***
(0.512) (0.358)
ln_square_income −0.379*** −0.416***
(0.0280) (0.0199)
ln_price 0.0814*** −0.0187 −0.00520
(0.0171) (0.0211) (0.0182)
ln_trade −0.432*** −0.488***
(0.0342) (0.0357)
ln_urban −1.526*** 0.275*** −1.870*** 0.344***
(0.140) (0.0855) (0.130) (0.113)
crisis 0.292*** 0.398*** 0.0593 0.365*** 0.424*** 0.0770 0.435***
(0.0387) (0.0483) (0.0608) (0.0567) (0.0463) (0.0596) (0.0617)
low_income 0.731*** 0.339*** −0.149 0.397*** −0.154
(0.0769) (0.116) (0.144) (0.116) (0.140)
low_middle_income 0.706*** 0.191** 0.182* 0.306*** 0.168*
(0.0701) (0.0804) (0.107) (0.0739) (0.102)
upper_middle_income 0.534*** 0.143*** 0.157* 0.181*** 0.177**
(0.0684) (0.0521) (0.0844) (0.0506) (0.0844)
low_trade 0.239*** 0.392*** 0.334***
(0.0308) (0.0622) (0.0600)
low_urban 0.299*** 0.691*** 0.635***
(0.0621) (0.129) (0.127)
middle_urban 0.236*** 0.139 0.166*
(0.0522) (0.0898) (0.0864)
constant −4.192*** −22.21*** 4.648*** 3.395** −19.48*** 11.33*** 11.09***
(0.207) (2.669) (1.594) (1.696) (1.524) (0.565) (0.481)
Group fixed effect Yes No Yes Yes No Yes Yes
Time fixed effects Yes Yes Yes Yes Yes Yes Yes
Wald χ 2 (p-value) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Nr observations 534 546 546 546 546 534 534
Nr countries 68 69 69 69 69 68 68

Notes:
1. Results produced by feasible generalized least squares (FGLS); statistics robust to heteroskedasticity and serial correlation
where the correlation parameter is unique for each panel.
2. Standard errors of the estimated coefficients (in brackets).
3. P-value representations: ***p < 0.01, **p < 0.05, *p < 0.10.

1 The three remaining determinants [size of the food produc- the model [specifications (3) and (6)]. Specifications (4) and (6) 23

2 tion sector (variable ln food_sector), food price volatility (variable contains the dummy variables that represent ln incomeit , ln tradeit 24

3 ln price), and rail (variable ln rail)] influence PHL in a low level and ln urbanit . 25

4 with its signs consistent with the literature, although the sign Lastly, the causally related independent variables ln supplyit 26

5 of the ln food_sector coefficient changed in specification (6). The and ln priceit are not considered together in the same specifica- 27

6 road density (variable ln road) was not significant in the baseline tion. Thus, specifications (2), (3) and (4) considers ln supplyit and 28

7 models. Still, the empirical model is robust, as will be seen in the specifications (5), (6) and (7), uses ln priceit . 29

8 next section. These approaches led to 12 specifications, which generated 30

robust results indicating that our empirical model is robust. 31

In specifications (2) to (7) of Table 4, the PHL dependent 32


9 4.3. Robustness checks
variable was changed from percent of production loss to tons of 33

production loss. The alteration led the road, rail and urbanization 34
10 Two sets of robustness checks were carried out to assess variables’ coefficients to change sign. This behavior was expected 35
11 the validity and sensitivity of the results. First, alternative PHL since the dependent variable is not weighted; in other words, 36
12 concepts were employed as dependent variables (Tables 4 and it should be expected that countries where there are a greater 37
13 5). Secondly, some variables were dropped from the empirical number of logistical operations or that move a greater amount of 38
14 model (Table 6). The resulting changes to the variables’ estimated agricultural product present a greater amount of loss in absolute 39

15 coefficients were analyzed. terms. Because of this, another weight (country population) was 40

16 The main results are not changed when the PHL concept used to compare specifications (2) to (7) of Table 5 with the 41

17 was changed. In both Tables 4 and 5, when the independent baseline specifications shown in Table 3.4 As one can see, only the 42

18 variable ln incomeit is used, the related independent variables coefficient of the variables size of the food production sector and 43

19 ln priceit and ln tradeit were excluded [specifications (2) and (5)]. rail density changed sign, which shows that the baseline model 44

20 When the GDP per capita information was transformed in four is robust to most specification changes. 45

21 dummy variables, the endogeneity problem is dealt with and the


22 independent variables ln priceit and ln tradeit are considered in 4 Specification (1) in Tables 4–6 is the same as specification (6) in Table 3.

Please cite this article as: W.E. Bendinelli, C.T. Su, T.G. Péra et al., What are the main factors that determine post-harvest losses of grains?. Sustainable Production and
Consumption (2019), https://doi.org/10.1016/j.spc.2019.09.002.
SPC: 268

W.E. Bendinelli, C.T. Su, T.G. Péra et al. / Sustainable Production and Consumption xxx (xxxx) xxx 9

Table 5
Robustness checks: alternative concepts of dependent variable (ln PHL_capitait ) and alternative selection of independent variables.
Variables (1) (2) (3) (4) (5) (6) (7)
% PHL PHL_capita PHL_capita PHL_capita PHL_capita PHL_capita PHL_capita
ln_supply 1.343*** 1.382*** 0.954***
(0.215) (0.222) (0.272)
ln_food_sector −0.0544* 0.560*** 0.621*** 0.621*** 0.691*** 0.771*** 0.837***
(0.0309) (0.0470) (0.0465) (0.0630) (0.0526) (0.0409) (0.0318)
ln_road −0.00423 −0.158*** −0.211*** −0.155*** −0.158*** −0.193*** −0.106***
(0.0172) (0.0249) (0.0284) (0.0252) (0.0175) (0.0286) (0.0226)
ln_rail −0.0740*** 0.199*** 0.158*** 0.154*** 0.164*** 0.136*** 0.115***
(0.0200) (0.0200) (0.0262) (0.0194) (0.0182) (0.0221) (0.0226)
ln_income 4.800*** 4.747***
(0.258) (0.280)
ln_square_income −0.287*** −0.279***
(0.0143) (0.0158)
ln_price 0.0814*** 0.0999*** 0.0951***
(0.0171) (0.0218) (0.0188)
ln_trade −0.125*** −0.125***
(0.0258) (0.0228)
ln_urban −0.381*** −0.451*** −0.374*** −0.239***
(0.0869) (0.0831) (0.0874) (0.0884)
crisis 0.292*** 0.302*** 0.168*** 0.241*** 0.319*** 0.219*** 0.279***
(0.0387) (0.0375) (0.0451) (0.0459) (0.0362) (0.0521) (0.0468)
low_income 0.731*** 0.467*** 0.479*** 0.438*** 0.349***
(0.0769) (0.0908) (0.0926) (0.0728) (0.0706)
low_middle_income 0.706*** 0.587*** 0.560*** 0.692*** 0.575***
(0.0701) (0.0763) (0.0816) (0.0652) (0.0707)
upper_middle_income 0.534*** 0.366*** 0.283*** 0.395*** 0.306***
(0.0684) (0.0573) (0.0621) (0.0525) (0.0573)
low_trade 0.239*** 0.185*** 0.296***
(0.0308) (0.0359) (0.0357)
low_urban 0.299*** 0.351*** 0.343***
(0.0621) (0.0805) (0.0706)
middle_urban 0.236*** 0.265*** 0.245***
(0.0522) (0.0640) (0.0630)
constant −4.192*** −24.95*** −5.184*** −5.987*** −19.40*** −0.492 −2.963***
(0.207) (1.383) (1.009) (1.227) (1.083) (0.360) (0.189)
Group fixed effect Yes No Yes Yes No Yes Yes
Time fixed effects Yes Yes Yes Yes Yes Yes Yes
Wald χ 2 (p-value) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Nr observations 534 546 546 546 546 534 534
Nr countries 68 69 69 69 69 68 68

Notes:
1. Results produced by feasible generalized least squares (FGLS); statistics robust to heteroskedasticity and serial correlation where
the correlation parameter is unique for each panel.
2. Standard errors of the estimated coefficients (in brackets).
3. P-value representations: ***p < 0.01, **p < 0.05, *p < 0.10.

1 The independent variable food surplus (variable ln supply), and globalization), respectively, were removed. When comparing 26

2 added in (2), (3) and (4) of Tables 4 and 5, strongly generates the baseline PHL model to these specifications, the results show 27

3 higher PHL, as expected. Communities facing food deficit may that it is robust to most changes in specification. The variable that 28

4 lack infrastructure to deal with the surplus, while places where expresses the road density is the only variable that presents high 29

5 food available is higher than food demand have low economic sensitivity to the sub-specification, although it was not significant 30

6 incentives to prevent PHL (Hiç et al., 2016; Liu et al., 2013; Segrè in the baseline model. As other studies have not completely spec- 31

7 et al., 2014). ified these macroeconomic conditions in their empirical models 32

8 It is expected that the world population reach 9 billion people (KC et al., 2016; Rosegrant et al., 2015), it is concluded that any 33

9 by 2050, which will require 70% more food (Hodges et al., 2011). sub-specification related to variables may cause relevant prob- 34
10 The obtained results show that efforts focused on increasing food lems of inconsistent estimation and inference that can lead to 35
11 production cannot be restricted to on-farm production, as it will unpleasant false negatives. 36
12 lead to a food surplus and incur in much higher levels of PHL,
13 shown by the high impact of the variable ln supply. Increase in 5. Concluding remarks 37

14 food production should be complemented by investments on


15 post-harvest infrastructure, especially in food storage and food The results suggest that there is a difficult trade-off involved 38

16 marketing, in order to achieve an advanced stage just as in when countries seek to increase the availability of basic foods 39

17 high-income countries and counterbalancing this PHL increase to feed the increasing population. Without proper post-harvest 40

18 (Flanagan et al., 2019; Segrè et al., 2014). The rapid rise of modern infrastructure, especially in food storage and food marketing, a 41

19 FSC, with vertical integration and computer-based modeling and country’s efforts to increase food supply can lead to an increase 42

20 monitoring systems, is associated with economic development in the level of PHL. There was also evidence that as a country 43

21 and urbanization and can also reduce PHL rates (Flanagan et al., develops economically there is concurrent, non-linear PHL reduc- 44

22 2019; Sheahan and Barrett, 2017). tion in all analyzed dimensions. The construction of infrastructure 45

23 In specifications (2), (3) and (4) in Table 6, the set of variables such as storage facilities in developing countries, along with 46

24 representing food sector infrastructure and food price variation, the transfer of knowledge and technologies from industrialized 47

25 logistics infrastructure, and economic development (urbanization countries, has led to the reduction in PHL. 48

Please cite this article as: W.E. Bendinelli, C.T. Su, T.G. Péra et al., What are the main factors that determine post-harvest losses of grains?. Sustainable Production and
Consumption (2019), https://doi.org/10.1016/j.spc.2019.09.002.
SPC: 268

10 W.E. Bendinelli, C.T. Su, T.G. Péra et al. / Sustainable Production and Consumption xxx (xxxx) xxx

Table 6 assumptions and that their PHL determinants are different. Al- 32
Robustness checks: variables dropped off the empirical model. though the conclusions remarked in this work cannot be directly 33
Variables (1) (2) (3) (4) extrapolated to fruits and vegetables, it could lead to a future 34
% PHL % PHL % PHL % PHL
work focusing in this food group. 35
ln_food_sector −0.0544* −0.0961*** −0.0838** To the best of our knowledge, there is not a publicly avail- 36
(0.0309) (0.0327) (0.0351)
ln_road −0.00423 −0.0291 0.0295*
able food storage variable with global level panel data, thus we 37

(0.0172) (0.0271) (0.0151) used the road and rail lines density variables as a proxy for a 38

ln_rail −0.0740*** −0.0992*** −0.123*** country’s infrastructure. Although obtained model using these 39
(0.0200) (0.0278) (0.0191) proxy variables was robust, it would be important to pursue the 40
ln_price 0.0814*** 0.0895*** 0.0820***
measurement of a more accurate food storage variable, which 41
(0.0171) (0.0157) (0.0174)
crisis 0.292*** 0.180*** 0.212*** would benefit future works. 42

(0.0387) (0.0397) (0.0407) In this work, PHL does not consider food waste, which happens 43
low_income 0.731*** 0.549*** 0.649*** 0.834*** in retail markets or after it reaches consumer and is generally 44
(0.0769) (0.105) (0.0859) (0.0773) related to retailers and consumers’ behavior. Due to this defini- 45
low_middle_income 0.706*** 0.562*** 0.531*** 0.814***
tion of PHL, consumer behavior explanatory variables were not 46
(0.0701) (0.0956) (0.0808) (0.0727)
upper_middle_income 0.534*** 0.370*** 0.380*** 0.564*** included in this model, although they should be considered in an 47

(0.0684) (0.0880) (0.0755) (0.0739) empirical model studying food waste. 48


low_trade 0.239*** 0.182*** 0.180***
(0.0308) (0.0329) (0.0331) Declaration of competing interest 49
low_urban 0.299*** 0.335*** 0.234***
(0.0621) (0.0689) (0.0605)
middle_urban 0.236*** 0.266*** 0.197*** The authors declare that they have no known competing finan- 50

(0.0522) (0.0560) (0.0519) cial interests or personal relationships that could have appeared 51
constant −4.192*** −3.978*** −3.640*** −3.725*** to influence the work reported in this paper. 52
(0.207) (0.124) (0.213) (0.216)
Group fixed effect Yes Yes Yes Yes
Time fixed effects Yes Yes Yes Yes
Acknowledgment 53

Wald χ 2 (p-value) 0.0000 0.0000 0.0000 0.0000


Nr observations 534 546 544 534 Funding 54
Nr countries 68 69 69 68

Notes: This work was supported by the São Paulo Research Founda- 55

1. Results produced by feasible generalized least squares (FGLS); statistics robust tion (FAPESP) [grant number 2015/22097-1]. 56
to heteroskedasticity and serial correlation where the correlation parameter is
unique for each panel.
References 57
2. Standard errors of the estimated coefficients (in brackets).
3. P-value representations: ***p < 0.01, **p < 0.05, *p < 0.10.
Affognon, H., Mutungi, C., Sanginga, P., Borgemeister, C., 2015. Unpacking 58
postharvest losses in sub-Saharan Africa: A Meta-Analysis. World Dev. 66, 59
49–68. http://dx.doi.org/10.1016/j.worlddev.2014.08.002. 60

1 Economic growth leads to increased urbanization and food Aldred, D., 2012. Urbanization: a major driver of infrastructure spending. Citi 61
Perspect. 3, 6–39. 62
2 diversification. Although urbanization may cause the contraction AMIS, 2018. AMIS - Agricultural Market Information System [WWW Document]. 63
3 of a country’s agricultural sector, the benefits derived from an URL http://www.amis-outlook.org/. 64
4 improved infrastructure and a more efficiently operated supply Ashraf, Q., Galor, O., 2011. Dynamics and stagnation in the malthusian epoch. 65

5 chain that result from urbanization bring about a reduction of Amer. Econ. Rev. 101, 2003–2041. 66
Bennett, M.K., 1941. Wheat in national diets. Wheat Stud. Food Res. Inst. 18, 67
6 PHL that more than counterbalances the migration of agricultural
37–76. 68
7 workers to growing urban centers. In the same vein, economic Cardoen, D., Joshi, P., Diels, L., Sarma, P.M., Pant, D., 2015a. Agriculture biomass 69
8 development normally leads to an increase in foreign agricultural in India: Part 2. Post-harvest losses, cost and environmental impacts. Resour. 70
9 commodity trade. In order to compete globally in the commodi- Conserv. Recycl. 101, 143–153. http://dx.doi.org/10.1016/j.resconrec.2015.06. 71

10 ties market, packaging and operational procedures must meet 002. 72


Cardoen, D., Joshi, P., Diels, L., Sarma, P.M., Pant, D., 2015b. Agriculture biomass 73
11 international norms. The standards demanded when trading in in India: Part 1. Estimation and characterization. Resour. Conserv. Recycl. 74
12 the international market are often an improvement over local 102, 39–48. http://dx.doi.org/10.1016/j.resconrec.2015.06.003. 75
13 practices in developing countries and more appropriate to the Castells-Quintana, D., 2017. Malthus living in a slum: Urban concentration, 76

14 maintenance of product attributes, both nutritionally and in ap- infrastructure and economic growth. J. Urban Econ. 98, 158–173. http://dx. 77
doi.org/10.1016/j.jue.2016.02.003. 78
15 pearance, thereby lowering PHL, advancing food security, and
Chaboud, G., Daviron, B., 2017. Food losses and waste: Navigating the incon- 79
16 securing food safety. sistencies. Glob. Food Sec. 12, 1–7. http://dx.doi.org/10.1016/j.gfs.2016.11. 80
17 All used data come from reliable, internationally recognized 004. 81
18 sources, being the results robust to shocks; however, the impor- Choi, I., 2001. Unit root tests for panel data. J. Int. Money Financ. 20, 249–272. 82

19 tance of further discussion regarding data standardization and http://dx.doi.org/10.1016/S0261-5606(00)00048-6. 83


FAO, 2001. Food balance sheets - A handbook, Rome, Food and Agriculture 84
20 improvements in the quality of PHL data should not be dimin- Organization of the United Nations. 85
21 ished or overlooked. Quite the contrary, better quality data will FAO, 2018a. World Food and Agriculture - Statistical Pocketbook 2018. Food and 86
22 reduce estimation bias and allow agents and governments to Agriculture Organization of the United Nations, Rome. 87

23 make more informed decisions and researchers to make more FAO, 2018b. INPhO - Information on post-harvest operations [WWW Document]. 88
Food Agric. Organ. United Nations. URL http://www.fao.org/in-action/inpho/ 89
24 accurate estimates.
home/en/. 90
25 Unilateral analyzes should be avoided due to the problem of FAOSTAT, 2017. FAO database for food and agriculture [WWW Document]. Food 91
26 model sub-specification. In the same sense, cross-section data Agric. Organ. United Nations. URL http://www.fao.org/faostat/en/#home. 92
27 that lack adequate corrections and have a functional form without Flanagan, K., Robertson, K., Hanson, C., 2019. Reducing food loss and waste. 93

28 error term raise the problem of inferring false negatives. Finally, Godfray, H.C.J., Beddington, J.R., Crute, I.R., Haddad, L., Lawrence, D., Muir, J.F., 94
Pretty, J., Robinson, S., Thomas, S.M., Toulmin, C., 2010. Food security: The 95
29 as suggestions for future works, perishable food crops, such as challenge of feeding 9 billion people. Science (80-. ) 327, 812–818. http: 96
30 fruits and vegetables, have different supply chain requirements //dx.doi.org/10.4337/9780857939388. 97
31 than grains; hence, it is likely that they require a different set of Green, W., 2000. Econometric Analysis. Prentice-Hall, New York. 98

Please cite this article as: W.E. Bendinelli, C.T. Su, T.G. Péra et al., What are the main factors that determine post-harvest losses of grains?. Sustainable Production and
Consumption (2019), https://doi.org/10.1016/j.spc.2019.09.002.
SPC: 268

W.E. Bendinelli, C.T. Su, T.G. Péra et al. / Sustainable Production and Consumption xxx (xxxx) xxx 11

1 Gustavsson, J., Cederberg, C., Sonesson, U., Emanuelsson, A., 2013. The Methodol- Parfitt, J., Barthel, M., Macnaughton, S., 2010. Food waste within food supply 42
2 ogy of the FAO Study : Global Food Losses and Food Waste - Extent, Causes chains: quantification and potential for change to 2050. Phil. Trans. R. Soc. 43
3 and Prevention. FAO, 2011. B 365, 3065–3081. http://dx.doi.org/10.1098/rstb.2010.0126. 44
4 Gustavsson, J., Cederberg, C., Sonesson, U., Van Otterdijk, R., Meybeck, A., 2011. Porter, S.D., Reay, D.S., Higgins, P., Bomberg, E., 2016. A half-century of 45
5 Global Food Losses and Food Waste – Extent, Causes and Prevention. FAO, production-phase greenhouse gas emissions from food loss & waste in the 46
6 Rome. global food supply chain. Sci. Total Environ. 571, 721–729. http://dx.doi.org/ 47
7 Hiç, C., Pradhan, P., Rybski, D., Kropp, J.P., 2016. Food surplus and its climate 10.1016/j.scitotenv.2016.07.041. 48
8 burdens. Environ. Sci. Technol. 50, 4269–4277. http://dx.doi.org/10.1021/acs. Rae, A.N., 1997. Changing food consumption patterns in east asia: Implications 49
9 est.5b05088. of the trend towards livestock products. Agribus. Int. J. 13, 33–44. 50
10 HLPE, 2014. Food losses and waste in the context of sustainable food systems, Rosegrant, M.W., Magalhaes, E., Valmonte-Santos, R.A., Mason-D’Croz, D., 2015. 51
11 A report by the High Level Panel of Experts on Food Security and Nutrition Returns to Investment in Reducing Postharvest Food Losses and Increasing 52
12 of the Committee on World Food Security, Rome. Agricultural Productivity Growth, Food Security and Nutrition Assessment 53
13 Hodges, R.J., Buzby, J.C., Bennett, B., 2011. Postharvest losses and waste in Paper. https://doi.org/http://purl.umn.edu/235879. 54
14 developed and less developed countries: Opportunities to improve resource Rutten, M.M., 2013. What economic theory tells us about the impacts of reducing 55
15 use. J. Agric. Sci. 149, 37–45. http://dx.doi.org/10.1017/S0021859610000936. food losses and/or waste: implications for research, policy and practice. 56
16 Jedermann, R., Nicometo, M., Uysal, I., Lang, W., 2014. Reducing food losses Agric. Food Secur. 2, http://dx.doi.org/10.1186/2048-7010-2-13. 57
17 by intelligent food logistics. Phil. Trans. R. Soc. A 372, 20130302. http: Segrè, A., Falasconi, L., Politano, A., Vittuari, M., 2014. Background paper on the 58
18 //dx.doi.org/10.1098/rsta.2013.0302. economics of food loss and waste (unedited working paper). FAO. 59
19 Kaminski, J., Christiaensen, L., 2014. Post-harvest loss in sub-Saharan Africa-what Shafiee-Jood, M., Cai, X., 2016. Reducing food loss and waste to enhance 60
20 do farmers say?. Glob. Food Sec. 3, 149–158. http://dx.doi.org/10.1016/j.gfs. food security and environmental sustainability. Environ. Sci. Technol. 50, 61
21 2014.10.002. 8432–8443. http://dx.doi.org/10.1021/acs.est.6b01993. 62
22 KC, K.B., Haque, I., Legwegoh, A.F., Fraser, E.D.G., 2016. Strategies to reduce Shashi, Singh, R., Shabani, A., 2017. Value-adding practices in food supply chain: 63
23 food loss in the global south. Sustain. 8, 1–13. http://dx.doi.org/10.3390/ Evidence from indian food industry. Agribus. Int. J. 33, 116–130. 64
24 su8070595. Sheahan, M., Barrett, C.B., 2017. Review: Food loss and waste in Sub-Saharan 65
25 Kummu, M., de Moel, H., Porkka, M., Siebert, S., Varis, O., Ward, P.J., 2012. Lost Africa. Food Policy 70, 1–12. http://dx.doi.org/10.1016/J.FOODPOL.2017.03. 66
26 food, wasted resources: Global food supply chain losses and their impacts 012. 67
27 on freshwater, cropland, and fertiliser use. Sci. Total Environ. 438, 477–489. Thyberg, K.L., Tonjes, D.J., 2016. Drivers of food waste and their implications 68
28 http://dx.doi.org/10.1016/j.scitotenv.2012.08.092. for sustainable policy development. Resour. Conserv. Recycl. 106, 110–123. 69
29 Liu, J., Lundqvist, J., Weinberg, J., Gustafsson, J., 2013. Food losses and waste in http://dx.doi.org/10.1016/j.resconrec.2015.11.016. 70
30 china and their implication for water and land. Environ. Sci. Technol. 47, UNESCO, 2017. UNESCO Institute for Statistics data base. 71
31 10137–10144. http://dx.doi.org/10.1021/es401426b. Weil, D.N., Wilde, J., 2009. How relevant is malthus for economic development 72
32 Madden, B., Florin, N., Mohr, S., Giurco, D., 2019. Using the waste Kuznet’s today?. Amer. Econ. Rev. 99, 255–260. 73
33 curve to explore regional variation in the decoupling of waste generation World Bank, 2011. Missing food: The case of postharvest grain losses in 74
34 and socioeconomic indicators. Resour. Conserv. Recycl. 149, 674–686. http: Sub-Saharan Africa. World Bank. https://doi.org/ReportNo.60371-AFR. 75
35 //dx.doi.org/10.1016/j.resconrec.2019.06.025. World Bank, 2017. World Bank Open Data. 76
36 Minten, B., Reardon, T., Singh, K.M.M., Sutradhar, R., 2014. The new and changing World Resources Institute, 2016. Food Loss and Waste Accounting and Reporting 77
37 roles of cold storages in the potato supply chain in Bihar. Econ. Polit. Wkly. Standard 160. 78
38 49, 98–108. http://dx.doi.org/10.1227/01.NEU.0000349921.14519.2A.
39 Neethirajan, S., Karunakaran, C., Jayas, D.S., White, N.D.G., 2007. Detection
40 techniques for stored-product insects in grain. Food Control 18, 157–162.
41 http://dx.doi.org/10.1016/j.foodcont.2005.09.008.

Please cite this article as: W.E. Bendinelli, C.T. Su, T.G. Péra et al., What are the main factors that determine post-harvest losses of grains?. Sustainable Production and
Consumption (2019), https://doi.org/10.1016/j.spc.2019.09.002.

You might also like