Exploring the Dynamic Effects of Agricultural Subsidies on Food Loss: Implications for Sustainable Food Security

Abstract

This paper analyzes the dynamic effects of agricultural subsidies on food loss using the two-stage dynamic panel model. The results reveal that dynamic adjustments exist in agricultural productivity (0.56) and food loss (0.58), with a U-shaped curve between them. That is, food loss declines as productivity grows, but it rises after reaching a certain productivity level. In addition, the results show that agricultural subsidies induce an increase in food loss in the short and long terms. Particularly in the short term, agricultural subsidies increase food loss directly by about 0.09%, and increase it indirectly by about 0.33% through changes in agricultural productivity. The long-term direct and indirect effects are estimated to be about 0.21% and 4.06%, respectively. While the indirect effects are greater than the direct effects, it is found that food loss responds more sensitively to agricultural subsidies in the long term.

1. Introduction

According to the Food and Agriculture Organization (FAO), about 193 million people have experienced food crises worldwide, and acute food insecurity is expected to worsen in the future [1]. While there are lots of factors affecting food insecurity, food loss and waste are considered to be detrimental because about one third of the food produced in the world is lost or wasted [2,3,4]. The costs related to food loss and waste are also estimated to reach about USD 2.6 trillion per year, including environmental (about USD 700 billion) and social (about USD 900 billion) costs [5]. As food loss and waste cause serious food insecurity and environmental problems [6,7,8], the Sustainable Development Goals (SDGs) of the United Nations aim to reduce food waste at the distribution and consumption levels and reduce food loss in the production levels by 2030 [9]. In particular, food loss reduction in the supply chain has great potential to contribute to the transition of agriculture toward a sustainable and circular sector.

Meanwhile, many countries have implemented various types of agricultural support policies to help producers’ long-term decision making for cultivation and harvesting. Since agricultural producers play an important role in supporting the economy and contributing to food security, many policy makers have implemented policies mainly to protect their own agriculture, support farmers’ income, secure a stable food supply, prevent poverty, and achieve the environmental sustainability of their agricultural industry [10]. Due to the importance of agriculture, governments around the world have provided producers with agricultural subsidies. On average, the annual net subsidies paid to agricultural producers currently amounts to about USD 540 billion per year, accounting for 15% of total agricultural production [11]. Admittedly, the World Trade Organization (WTO) clearly prohibits direct agricultural subsidies, but agricultural policies in most countries continue to support producers with payments decoupled from production decisions [12].

Most agricultural subsidies were considered necessary and positive for agriculture and farmers’ incomes. However, some studies found that agricultural subsidies would negatively affect producers’ behavior and agricultural productivity. As agricultural subsidies were intended to support producers’ income, producers had no incentive to produce farm crops optimally, due to their expectation of free financial support in the future [13]. That is, producers tended to just produce subsidized farm crops because the repeated payments from the government guaranteed their substantial profits [12,14,15,16,17]. Moreover, producers’ inefficient production decisions resulted in low productivity due to the abuse of resources and misguided distribution [18,19,20]. Or there were poor-quality outputs that were wasted, which also caused low productivity. In this regard, many studies supported the negative effects of agricultural subsidies on agricultural productivity [21,22,23,24,25,26,27,28,29].

Since agricultural subsidies are negatively associated with producers’ behavior and agricultural productivity, it is expected that agricultural production distorted by agricultural subsidies can result in substantial food loss. Food loss can be determined by producers’ careless behavior in discarding crops or low efficiency in production. Considering that the loss function consists of agricultural outputs, food loss can be linearly or non-linearly proportional to output levels affected by agricultural subsidies. However, to the best of our knowledge, the literature has paid little attention to the relationships of food loss with both agricultural productivity and subsidies. Most studies have focused only on the current status and problems of food loss in the production, distribution, and storing processes, and they discussed the challenges arising from food loss, providing potential solutions to those problems [1,2,3,4,5,6,17]. Considering the linkages of food loss with both agricultural productivity and subsidies, it is necessary to understand the mechanism that generates food loss under agricultural support policies.

Therefore, this paper aims to examine the dynamic effects of agricultural subsidies on food loss. Specifically, this paper explores the dynamics of agricultural productivity and food loss using the two-stage dynamic panel approach. Using this approach, this paper examines how agricultural productivity is associated with food loss. We determine whether there is a U-shaped curve between food loss and agricultural productivity. In addition, we determine whether agricultural subsidies increase food loss in the short and long terms. In particular, this paper decomposes the total effects into direct and indirect effects in order to understand how agricultural subsidies increase food loss directly and indirectly through changes in agricultural productivity. The decomposition results can contribute to identifying the mechanism causing food loss to be associated with both productivity and subsidies.

The structure of this paper is as follows: Chapter 2 introduces the two-stage dynamic panel model and the data used for empirical analyses. Chapter 3 presents the estimation and decomposition results in terms of elasticities. Finally, Chapter 4 discusses the findings of this paper and concludes with policy implications for sustainable food security.

2. Materials and Methods

2.1. Two-Stage Dynamic Panel Model

This paper uses the two-stage dynamic panel approach to examine the direct and indirect effects of agricultural subsidies on food loss [30,31]. For the dynamic panel model, we add the lagged dependent variable to the conventional static panel model. While the coefficient of the dynamic term represents the adjustment in the dependent variable, we construct the dynamic panel models in two stages: the dynamic specifications for agricultural yield and food loss functions. The first-stage model specifies the function for agricultural production per area, and the second-stage model specifies the function for food loss per area. Since food loss is non-linearly proportional to agricultural yield, the second-stage model includes the variables of agricultural yield and its squared term. In this stage, as agricultural yield is determined endogenously, we used the predicted value of agricultural yield obtained in the first stage.

Specifically, in the first stage, we construct the dynamic model with a focus on the effects of agricultural subsidies on agricultural productivity; agricultural yield (outputs per area) is used to represent agricultural productivity. The first-stage dynamic panel model is written as

$$\ln Q_{it}={\alpha }_0+\theta \ln Q_{it-1}+{\alpha }_1\ln K_{it}+{\alpha }_2\ln L_{it}+{\alpha }_3\ln SU{B}_{it}+{\alpha }_4\ln RN{D}_{it}+{\gamma }_i+{\delta }_t+u_{it}$$

(1)

for country $i$ at time $t$. In Equation (1), $Q$ is agricultural outputs per area, $K$ is capital per area, $L$ is labor per area, $SUB$ is agricultural subsidies per area, and $RND$ is the investment in research and development (R&D) per area. Parameters $\theta$ and $\alpha$s are to be estimated. Equation (1) is constructed using the conventional econometric form of production function with a dynamic adjustment in the outputs [32]. Due to the nature of panel analyses, $u$ denotes the error term, $\gamma$ denotes the unobserved country-specific term, and $\delta$ denotes the unobserved time-specific term.

After obtaining the predicted values of agricultural productivity in the first stage, the predicted values are used as explanatory variables in the second stage. In this stage, we focus on the determinants of food loss. The second-stage dynamic panel model is specified as

$$\ln LOS{S}_{it}={\beta }_0+\rho \ln LOS{S}_{it-1}+{\beta }_1\ln {\widehat{Q}}_{it}+{\beta }_2{\left(\ln {\widehat{Q}}_{it}\right)}^2+{\beta }_3\ln SU{B}_{it}+{\beta }_4\ln RN{D}_{it}+{\beta }_{5}D{I}_{it}+{\mu }_i+{\xi }_t+{ϵ}_{it}$$

(2)

for country $i$ at time $t$. In Equation (2), $LOSS$ denotes the volume of food loss per area, $\widehat{Q}$ denotes the predicted values of agricultural productivity, and $DI$ denotes the dummy variable for high-income countries. Parameters $\rho$ and $\beta$s are to be estimated. Equation (2) represents the functional form of food loss that is non-linearly proportional to the agricultural outputs. As in Equation (1), $ϵ$ is the error term, $\mu$ is the unobserved country-specific characteristic, and $\xi$ is the unobserved time characteristic.

Through Equation (2), we test for the non-linear relationships between food loss and agricultural productivity and examine how agricultural subsidies affect food loss. We determine whether there is a U-shaped curve between food loss and agricultural productivity. Moreover, as agricultural subsidies affect food loss directly through producers’ behavior and indirectly through changes in agricultural productivity, we decompose the total effect into direct and indirect effects. While the estimate of agricultural subsidies obtained in Equation (2) represents the direct effect, the indirect effect is obtained by being combined with the estimates in Equation (1). In other words, the total effect of agricultural subsidies on food loss can be calculated by

$$\frac{d\ln LOSS}{d\ln SUB}=\frac{\partial \ln LOSS}{\partial \ln SUB}+\frac{\partial \ln LOSS}{\partial \ln \widehat{Q}}\frac{\partial \ln Q}{\partial \ln SUB}$$

(3)

where the first and second terms indicate the direct and indirect effects, respectively. While the direct effect is constant regardless of agricultural productivity, the indirect effect varies with agricultural productivity due to the nature of the quadratic form. In this paper, the indirect effect is evaluated at the sample mean of agricultural productivity; the interpretation of the total effect is based on the average agricultural productivity. In addition, the direct and indirect effects are calculated for the short and long terms. In the dynamic panel, the short-term effect is directly obtained in the form of elasticity $\left(\beta \right)$ from Equation (2), and the long-term effect is obtained by combining the estimates with the dynamic adjustment estimate ($\alpha /\left(1-\theta \right)$ and $\beta /\left(1-\rho \right)$) [33]. The long-term effects are generally greater than the short-term effects.

The empirical estimation is conducted using the system-generalized method of moments (system-GMM) in each stage. While the fixed- and random-effect models are used for the static panel analysis, they are not appropriate for the dynamic panel analysis. Since the dynamic panel model includes the lagged dependent variable in the specification, the pooled ordinary least squares and fixed effect estimators can yield inconsistent estimates due to the endogeneity problem. This problem can be solved using the system-GMM [34,35]. The system-GMM combines the level equations with the first difference equations to use them as instrumental variables, which addresses the endogeneity problem and produces consistent estimates. Using the system-GMM, we estimate the yield function with a focus on the effect of agricultural subsidies in the first stage. Using the predicted value of agricultural productivity, we estimate the loss function that includes the predicted productivity and its squared term, agricultural subsidies, and other explanatory variables.

2.2. Data

The empirical analysis is conducted for 20 countries covering the period from 2010 to 2019. Due to the limited availability of data, our analysis focuses on the selected countries and years: the countries include Australia, Brazil, Canada, Chile, China, Colombia, Costa Rica, Indonesia, Israel, Japan, Korea, Mexico, Netherlands, Norway, Philippines, Russia, South Africa, Switzerland, Turkey, and the United States. We collect the data for food loss, agricultural outputs, total capital formation, number of workers, cultivated area, and R&D investment in the supply side from the FAO. In addition, we obtain and use the data for agricultural producer support from the OECD. We use the producer support estimate available for only a limited number countries, which represents the monetary transfers from consumers to producers in the agricultural sector [36].

Regarding the income dummy variable, we classify countries into high-income and non-high-income countries according to the standards of the World Bank. According to the standards of the World Bank, high-income countries are those with a GNI per capita more than USD 13,205, and in this sample, high-income countries include Australia, Canada, Chile, Israel, Japan, Korea, Netherlands, Norway, Russia, Switzerland, Turkey, and the United States, while non-high-income countries include Brazil, China, Colombia, Costa Rica, Indonesia, Mexico, Philippines, and South Africa. A detailed description of the data is provided in

Table 1. Data description.

Variable	Unit	Mean	Std. Dev.	Min	Max
Food loss	1000 tons	31,278	62,562	262	237,802
Agricultural yield	1000 tons	578,210	996,585	12,967	4,269,238
Agricultural subsidy	Million dollars	18,787	40,611	92	221,506
Gross capital formation	Million dollars	14,286	29,656	259	150,976
Number of workers	100 people	16,058	50,423	34	278,066
Cultivated area	1000 km2	1083	1547	0.010	5289
Agricultural R&D	Million dollars	1.771	4.140	0.018	19.941
Income dummy	High-income country = 1	0.500	0.501	0.000	1.000

Sources: FAO, OECD, and World Bank.

. The average food loss amounts to about 31 million tons, while the average agricultural yield is approximately 578 million tons. The average agricultural subsidy is about 19 billion dollars.

Figure 1 depicts agricultural yield, food loss, and agricultural subsidies per area over the period between 2010 and 2019. In this figure, agricultural outputs increased by about 16.7%, from about 10 billion tons in 2010, to about 12 billion tons in 2019. Meanwhile, food loss increased by about 22.6% over the same period, faster than the growth rate of agricultural productivity, from about 550 million tons in 2010, to about 674 million tons in 2019. It seemed that food lost in the supply chain became greater than outputs produced. Regarding the per-area values, agricultural yield and food loss per km² increased by only about 8.0% and 5.6%, respectively, which was attributed to the reduction in cultivated areas of about 0.4% over the time period measured.

Meanwhile, the total producer subsidies in the agricultural sector increased by 22.0%—from USD 316 billion in 2010, to USD 386 billion in 2019—but the per-area subsidies slightly decreased by 4.9% over the same period. In addition, the R&D investment in agriculture increased by 37.6%, from USD 30 million (USD 100 per km²), to USD 41 million (USD 130 per km²). The capital formulation increased by 33.8%, whereas labor decreased by 25.4% due to the recent tendency of replacing labor with machinery. The per-area capital also increased by 7.4%, but the per-area labor declined by 14.4%

3. Results

Table 2. First-stage estimation results.

	Pooled OLS	Fixed Effect	System-GMM
$\ln Q_{t-1}$			0.559 ***
			(0.111)
$\ln K_t$	0.320 ***	0.081 **	0.148 **
	(0.053)	(0.037)	(0.053)
$\ln L_{it}$	0.462 ***	0.168 ***	0.194 ***
	(0.025)	(0.064)	(0.048)
$\ln SU{B}_{it}$	−0.167 ***	−0.025	−0.061 **
	(0.032)	(0.019)	(0.022)
$\ln RN{D}_{it}$	−0.061 **	0.060 **	0.103 **
	(0.022)	(0.029)	(0.045)
$Constant$	7.248 ***	6.796 ***	2.978 ***
	(0.390)	(0.248)	(0.856)
AR(1)/AR(2)			0.002/0.911
Hansen test			0.991

Notes: Robust standard errors are in parenthesis; ** and *** denote significance at the 5% and 1% levels, respectively.

reports the estimation results of the first-stage dynamic panel model in terms of elasticities. The estimation of the dynamic panel model requires the use of the Arellano–Bond and Hansen J tests to confirm that there are no serial correlation and over-identification problems. As in Table 2, the test results confirm that there is no second-order autocorrelation and the instrumental variables are used appropriately. In the results of system-GMM, the estimated lagged dependent variable is about 0.56, indicating the suitability of the dynamic panel model for the data, which reflects the existence of dynamic adjustment in agricultural productivity. Regarding the production factors, the results show that capital (0.15) and labor (0.19) contribute to agricultural production, which is consistent with the theory that the marginal products of capital and labor inputs are positive.

On the other hand, agricultural subsidies are found to affect productivity negatively. This means that if producers’ investment decisions are dependent on subsidies, they may conduct inefficient agricultural activities with the abuse of resources or inappropriate distribution activities. Low productivity can occur because producers can just increase production factors or cultivated areas inappropriately without considering productivity once they receive subsidies. Moreover, it is possible to yield outputs that can be discarded after harvesting, that is, agricultural producers may generate worse quality crops or lose outputs due to inefficient distribution channels. The results imply that optimal decision making through cost minimization may not be made due to agricultural subsidies, which is consistent with previous studies [21,22,23,24,25,26,27,28,29].

Meanwhile, agriculture around the world is facing a labor shortage problem, and in order to overcome this limitation, the need for a shift to a technology-intensive production structure has been raised. As R&D investment acts as an important factor inducing technological innovation, it is found that R&D investment in the agricultural sector can increase agricultural productivity, thereby innovation in machinery, fertilizer and irrigation facilities, and genetic development such as breeding technologies [37]. It is also possible that R&D investment in artificial intelligence, machine learning, and Internet of Things (IoTs) sensors can optimize the efficiency of irrigation and fertilizer use and reduce pests and diseases [38,39,40,41]. These technologies can help farmers and food processors enhance productivity and safety, maintaining economic gains in supply chains [41,42].

Table 3. Second-stage estimation results.

	Pooled OLS	Fixed Effect	System-GMM
$\ln LOS{S}_{t-1}$			0.577 ***
			(0.158)
$\ln {\widehat{Q}}_{it}$	0.683 ***	4.146	−0.358 **
	(0.125)	(2.604)	(0.155)
${\left(\ln {\widehat{Q}}_{it}\right)}^2$	0.015 ***	−0.142	0.028 *
	(0.005)	(0.180)	(0.015)
$\ln SU{B}_{it}$	0.053	0.194 ***	0.090 **
	(0.038)	(0.059)	(0.038)
$\ln RN{D}_{it}$	0.016	−0.175 **	0.099
	(0.057)	(0.088)	(0.102)
$D{I}_{it}$	−0.746 ***		−0.376 *
	(0.131)		(0.198)
$Constant$	−1.597 *	−20.031 *	3.102 **
	(0.883)	(10.168)	(1.119)
AR(1)/AR(2)			0.027/0.205
Hansen test			1.000

Notes: Robust standard errors are in parenthesis; *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

shows the results of the second-stage dynamic panel model in terms of elasticities. The estimated lagged dependent variable shows that dynamic adjustment in food loss exists (0.58). In addition, due to the squared term of the predicted productivity, the results reveal that food loss has a non-linear relationship with agricultural productivity. As productivity grows, food loss tends to decline initially, but then rise after achieving a certain level of productivity. This means that, as productivity rises above a certain level, the disposal of outputs during production, transportation, and storage can increase. The results imply that active efforts towards food loss reduction are required throughout the entire food value chain, from food production to transportation and storage; logistics may play a fundamental role in reducing food loss [43].

Meanwhile, the effect of agricultural subsidies on food loss is estimated to be about 0.09, indicating that agricultural subsidies increase it directly. This means that agricultural support can induce a lack of care about food lost on farms and food loss during distribution and storage. If producers receive agricultural subsidies, they may then have no incentives to manage food loss. The results reflect producers’ careless behavior in wasting excessive surplus of food crops when they are financially supported. On the other hand, R&D investment in the agricultural sector shows an insignificant effect on food loss. Since R&D investment is mainly focused on increasing productivity, it does not seem to have a significant effect on activities to reduce food loss. Regarding the effects by income, the results reveal that high-income countries tend to generate less food loss than non-high-income countries.

From the first- and second-stage results, we estimate the short- and long-term effects of agricultural subsidies on food loss. As the total effects are evaluated using the average value of agricultural productivity, the results are interpreted as the average total effects. In

Table 4. The effects of agricultural subsidies on food loss.

	Direct Effect	Indirect Effect	Total Effect
Short-run effect	0.090 **	0.333 ***	0.423 ***
	(0.038)	(0.086)	(0.094)
Long-run effect	0.213 ***	4.061 ***	4.274 ***
	(0.033)	(0.052)	(0.053)

Notes: Robust standard errors are in parenthesis; ** and *** denote significance at the 5% and 1% levels, respectively.

, the results show that a 1% increase in agricultural subsidies increases, on average, food loss by about 0.42% in the short term and by about 4.27% in the long term. In particular, in the short term, while the direct effect of agricultural subsidies is positive (0.09), agricultural subsidies also increase food loss indirectly (0.33).

Even in the long term, both direct (0.21) and indirect (4.06) effects are found to be statistically significant and positive. As the indirect effects are much higher than the direct effects, the total effects represent that food loss can, on average, increase due to agricultural subsidies in the short and long terms. In the literature, it appears that agricultural subsidies can affect productivity negatively [21,22,23,24,25,26,27,28,29]. Accordingly, the results of the current work show that agricultural subsidies can result in substantial food loss if they reduce agricultural productivity, because low productivity is associated with poor-quality outputs, inefficient production processes, and inappropriate transportation and storage in the food supply chain.

4. Discussion

Food loss is of importance because its eradication is critical in order to achieve sustainable food security and a circular economy [4]. According to the FAO, preventing food loss and waste has the potential to feed about 1.26 billion people every year, while contributing to the addressing of negative externalities in the food supply chain [2,3,4]. It seems to be very urgent to come up with a plan to solve the problems related to food loss and waste. As shown in Figure 2, the prevalence of severe food insecurity in the population rose from 8.2 in 2015 to 10.5 in 2019, while the total volume of food loss increased from about 5.5 billion tons to 5.8 billion tons over this same period. Considering the potential contribution to food insecurity, it is necessary to reduce food loss under the continuous agricultural subsidy policies. However, as policies are generally focused on reducing the food waste of consumers and households [44], the policy and academic interests in reducing food loss have been somewhat insufficient.

The findings in this paper are meaningful in that they enable us to understand how agricultural subsidies actually cause food loss directly and indirectly via changes in agricultural productivity. Agricultural subsidies increase food loss directly by stimulating producers’ careless behavior, and they also increase it indirectly by a reduction in agricultural productivity. Agricultural subsidies may hinder producers in making optimal decisions because producers who receive subsidies tend not to care about losing food crops and not to use the optimal resources for production. The findings suggest that policy makers take account into the policy effects on food loss, rather than provide farmers with unconditional agricultural subsidies. It is also necessary to build well-managed food supply chains and encourage loss-reducing actions when agricultural subsidy policies are implemented [45].

Moreover, food loss can occur when it is not possible to access appropriate harvesting machines, farmer training courses, extended services, or research institutions [46,47,48]. Considering the interactions between universities, industries, and the government in promoting innovation in a knowledge-based society, decision makers should consider cooperation between these stakeholders to produce innovative strategies that could contribute to reducing food loss [49]. Universities can share professional research and technology with farmers, while the public sector can help them finance agricultural extension services [50]. Moreover, technologies such as aerial imaging technology that can detect crop-damaging weather, blockchain and wireless sensors that can monitor pest infections, and radio-frequency identification tags that can identify potential losses in the food supply chain can be developed and shared to reduce food loss [51,52,53]. In particular, due to the perishability of food, new technologies for well-established transportation and storage processes are also needed to reduce food loss [54]. If we apply these technologies in order to reduce food loss throughout the food supply system, we can achieve sustainable development and a circular economy [45,55,56,57,58,59].

5. Conclusions

This paper establishes a balanced panel dataset for 20 countries from 2010 to 2019 and empirically analyzes agricultural yield and food loss functions using the two-stage estimation method. The main findings show that, as productivity increases, food loss initially decreases, but rises after achieving a certain productivity. Moreover, in the short and long terms, agricultural subsidies increase food loss directly and indirectly through reduced productivity. While the indirect effects are stronger than the direct effects, it is found that food loss responds more sensitively to agricultural subsidies in the long term. The world is experiencing food shortage and food loss at the same time, so each country needs to make efforts to improve awareness regarding food loss [60]. As the SDGs emphasize the importance of reducing food loss, we need to improve environmental perceptions and develop comprehensive plans for food loss management. Moreover, if food security and environmental factors are taken into consideration from a long-term perspective, the findings suggest that subsidy policies reflect behavioral changes of producers, and economic incentives for improving productivity with less food loss must be considered to achieve sustainable food security.