2. Farm input subsidy in Malawi

Malawi is a landlocked southeastern African country with a population of over 19 million people (World Population Review, 2018). It has two main seasons: the cold-dry and hot-wet seasons, and temperatures ranging from 14–32°C (United Nations, 2014). The country is ranked as one of the world's poorest, 170^th out of 187, it suffers from low levels of nutrition, and it is vulnerable to weather shocks (United Nations, 2014; World Bank, 2018).

From the time of Malawi's independence in 1964 up to the mid-1980s when the structural adjustment programme was introduced, the food security policy was the main guide for agricultural plans and strategies in the country. Following this, a number of policies have evolved over the years, such as the Agriculture and Livestock Development Strategy and Action Plan established in 1995; the Malawi Agricultural Sector Investment Programme of 1999; the Agricultural Development Programme of 2006; and the Agricultural Sector Wide Approach of 2007–2009 and 2010–2015 (Ministry of Agriculture and Food Security, 2010; FAO, 2014).

In 2010, the government decided to establish a policy with the responsibility for harmonising the various agricultural development strategies. This was named the National Agricultural Policy Framework, and it was tasked with the responsibility for promoting agricultural productivity and realising national food security, amongst others (FAO, 2014).

Malawi's FISP is an offshoot of the Agricultural Input Subsidy Programme, a small-scale targeted input subsidy programme popularly referred to as the Starter Pack Scheme which was initiated in 1998. The FISP became popular in 2005 after a severe drought in the country that led to the programme being augmented from only a few farmers to about 50 per cent of the country's farmers, and to over 70 per cent of farmers in recent years (Harou, Reference Harou2018).

The objective of the programme is to give farmers access to improved agricultural inputs which can bring about food self-sufficiency and enhance rural income via higher levels of food and cash crop production (Dorward and Chirwa, Reference Dorward and Chirwa2011; Lunduka et al., Reference Lunduka, Ricker-Gilbert and Fisher2013), by handing out vouchers and coupons to smallholder farmers who own their farmlands and reside legitimately in their own villages, for the purchase of farm inputsFootnote ⁴ at subsidised rates (Chibwana and Fisher, Reference Chibwana and Fisher2011; Dorward and Chirwa, Reference Dorward and Chirwa2011; Dorward et al., Reference Dorward, Chirwa, Jayne, Chunan-Pole and Angwafo2011; Harou, Reference Harou2018).

The distributions of the vouchers are carried out at two levels (Ricker-Gilbert and Jayne, Reference Ricker-Gilbert and Jayne2017). First, the fertiliser and seeds are officially allocated to regions and districts based on agricultural cultivation area and the number of smallholders in such locations. Second, at the community-level, the community and the village heads are then involved in determining the eligible smallholders. The original allocation strategy for the vouchers targeted smallholders who are full-time farmers, and who are unable to purchase at most two bags of fertilisers at the prevailing commercial price in the community as determined by local leaders (Dorward et al., Reference Dorward, Chirwa, Matita, Mhango, Mvula, Taylor and Thome2013). From 2008 onward, the target group was defined as the ‘vulnerable’ group, including resource-poor households, disabled, elderly, female and child-headed households (see Ricker-Gilbert and Jayne, Reference Ricker-Gilbert and Jayne2017).

The voucher can be used to purchase agricultural inputs at a subsidised price from participating private retail stores, and such retailers then submit the voucher and receipt to the government for payment. Each smallholder who participates in the FISP is eligible to receive two vouchers, used for one 50-kg bag of fertiliser at a discounted price and for between 5 and 10 kg of improved maize seed (see Ricker-Gilbert and Jayne, Reference Ricker-Gilbert and Jayne2017).

Existing studies show that since the implementation of the FISP, there has been improvement in agricultural output among smallholder farmers, such that maize increased from a 43 per cent deficit in 2005 to a 53 per cent surplus by 2007 (Chibwana and Fisher, Reference Chibwana and Fisher2011; Harou, Reference Harou2018). However, due to the declining food insecurity levels in Malawi, from about 57 per cent in 2004/05 to 42 per cent in 2010/11 (Government of Malawi, 2005, 2012; Sibande et al., Reference Sibande, Bailey and Davidova2015), and the enormity of the FISP,Footnote ⁵ there has been criticism with regard to beneficiaries of the FISP adopting poor climate-resilient farming systems such as unsustainable land management practices and poor crop diversification strategies (see Zulu, Reference Zulu2017).

Another criticism of the FISP is the cultural sentiment associated with the distribution of the agricultural input, such that non-beneficiary neighbours could still benefit from the programme (Holden and Lunduka, Reference Holden and Lunduka2013).

There are some issues about the effectiveness of the targeting system of the FISP, including the frequent exclusion of poor households (Holden and Lunduka, Reference Holden and Lunduka2013), high administration costs (Lunduka et al., Reference Lunduka, Ricker-Gilbert and Fisher2013), tacit exclusion of female farmers from the programme (Ricker-Gilbert et al., Reference Ricker-Gilbert, Jayne and Chirwa2011; Chibwana et al., Reference Chibwana, Fisher and Shively2012), and consideration of households' wealth and agricultural land holding (Kilic et al., Reference Kilic, Whitney and Winters2013; Fisher and Kandiwa, Reference Fisher and Kandiwa2014). In some other cases, there are instances of elite capture, village leaders reducing the number of coupons per beneficiary household and some villages that are excluded based on egalitarian bias (Holden and Lunduka, Reference Holden and Lunduka2013). In light of these issues, there is a need for better targeting of the FISP based on a random and universal framework for the distribution of coupons. Such efforts will be necessary to ensure that the programme is actually targeting the intended group, which are the rural poor.

3. Empirical literature

This study is related to two strands of literature which are discussed in this section. First, we consider studies that focus on agricultural policies and food security, paying attention to smallholders in Africa. Second, we focus on weather shocks and their impact on smallholders' outcomes.

Considering the first strand, some studies such as Daidone et al. (Reference Daidone, Davis, Knowles, Pickmans, Pace and Handa2017) highlight the need for agricultural programmes to consider both cash transfers and input programmes for efficient improvement of the outcome of beneficiary farmers. Although the authors focus on welfare outcomes like poverty and hunger, they conclude that the efficiency of such programmes will largely depend on other factors, including the prevailing climatic conditions.

Further, Sibande et al. (Reference Sibande, Bailey and Davidova2015) also consider a similar agricultural policy in Malawi on both the broad measure of food security, and per capita consumption of smallholder farmers. The authors find a significant improvement in these two outcome variables for smallholders who are beneficiaries of the agricultural input programme. Similar findings are echoed in Chirwa and Dorward (Reference Chirwa and Dorward2013), Dorward et al. (Reference Dorward, Chirwa, Matita, Mhango, Mvula, Taylor and Thome2013) Asfaw et al. (Reference Asfaw, Cattaneo and Pallante2016) and Ricker-Gilbert and Jayne (Reference Ricker-Gilbert and Jayne2017). Malhotra (Reference Malhotra2015) accounts for the impact of a different programme – the National Agriculture Input Voucher Scheme – on smallholders' food security in Tanzania. She notes that this programme increases farmers' level of food security. However, this impact was noted to be through some important individual characteristics of the smallholders such as educational attainment, among others. For Nigeria, Ayoade et al. (Reference Ayoade, Ogunwale and Adewale2011) find the National Special Programme for Food Security to be highly effective for poverty reduction, while taking into account the gender dimension of their study.

Finding a positive and significant impact of agricultural-related programmes that target farm production efficiency is only logical considering that such programmes facilitate farm productivity by improving planting, reducing farm-related cost in farm input acquisition, or even by improving farm processes. However, this is not always the case, as studies which are beginning to advocate for the consideration of other non-controllable factors that affect the efficiency of agricultural programmes (see Giller et al., Reference Giller, Tittonell, Rufino, van-Wijk, Zingore, Mapfumo, Adjei-Nsiahe, Herrero, Chikowo, Corbeels, Rowe, Baijukya, Mwijage, Smith, Yeboah, van der Burg, Sanogo, Misiko and Vanlauwe2011; Daidone et al., Reference Daidone, Davis, Knowles, Pickmans, Pace and Handa2017) note that agricultural programmes in Africa have had mixed results due to ecological variability within farming systems in this region. Hence, the need to consider the impact of weather shocks on smallholders' agricultural outcomes.

Weather shocks include those unpredicted natural and environmental occurrences that directly affect farm yield, such as floods, droughts, frost and hailstorms, and can also have significant and severe consequences on agricultural productivity and general household welfare. The literature on this linkage abounds with important evidence from developing countries (Jayachandran, Reference Jayachandran2006; Yang and Choi, Reference Yang and Choi2007; Dell et al., Reference Dell, Jones and Olken2009; Schlenker and Lobell, Reference Schlenker and Lobell2010; Björkman-Nyqvist, Reference Björkman-Nyqvist2013). For instance, Badolo and Kinda (Reference Badolo and Kinda2012) and Benton and Bailey (Reference Benton and Bailey2015) note that weather variability increases the frequency and severity of devastating impacts on the sufficiency of food production in developing countries.

However, when smallholder farmers become vulnerable to certain weather shocks, studies have noted that having access to some agricultural inputs that can help reduce the impact of such shocks will have beneficial effects on farm performance. With the perceived adverse effect from weather shocks, and the need for farmers' access to some agricultural input, this study is set up to investigate the shock-cushioning capacity of specific agricultural policy in Malawi – the Farm Input Subsidy Programme – on food security outcomes of smallholder farming households.

4. Data sources and empirical methodology

4.2 Other sources of information: rainfall data

For the rainfall data used in our analysis, we match the household survey data with rainfall data obtained from the National Oceanic and Atmospheric Administration Climate Prediction Centre's African Rainfall Estimation Algorithm version 2.0. Seasonal precipitation data gathered from the Malawian meteorological weather stations are used in the interpolation of the global positioning system of the surveyed households. These data include annual and wet season precipitation measures respectively, and spatial distribution of households included in the LSMS-ISA survey for Malawi enhances the credibility of the rainfall variation at the enumeration area (EA) level.

The measure of rainfall shocks we used for precipitation data was provided by the World Bank (along with the LSMS data). We follow Maccini and Yang (Reference Maccini and Yang2009), Björkman-Nyqvist (Reference Björkman-Nyqvist2013) and Rocha and Soares (Reference Rocha and Soares2015) in constructing rainfall shocks and creating measures of deviations in rainfall from the long-run mean rainfall for an area by constructing rainfall shock in the following way:

(1)\begin{equation}{\rm Rainshoc}{{\rm k}_{ht - 1}} = ln{R_{ht - 1}} - \overline {ln{R_h}}, \end{equation}

where $\; ln{R_{ht - 1}}$ indicates the yearly rainfall in household h for the preceding year's planting season and is the average historical yearly rainfall in household h. The average historical yearly rainfall was calculated from 2001 to 2013. Thus, ${\rm Rainshoc}{{\rm k}_{ht - 1}}$ is the shock measure used for deviation of the natural logarithm of the total rainfall in the 12 months prior to the 2009/2010 and 2012/2013 periods and the natural logarithm of the average yearly historical rainfall in household h prior to the corresponding years.

The rainfall deviation basically implies a percentage deviation from mean rainfall. However, we use negative shock in the regressions, which is measured as the absolute value of the deviation if a negative deviation exists between rainfall deviation from the historical norm and 0 if otherwise. We use this rainfall shock definition for the following reasons: (1) to take care of the outlier issue, and (2) to easily interpret the result as a percentage deviation with reference to the historical rainfall information.

4.3 Summary statistics

For the household welfare measure, we use the following outcomes in our analysis: (i) log of per capita food expenditure, (ii) log of non-food consumption per capita, (iii) log of total consumption per capita, and (iv) Food Consumption Score (FCS). We compute an FCS following the World Food Programme guidelines that captures both dietary diversity and food frequency. It is the weighted sum of the number of days the household consumed foods from eight food categories in the last week. The score is calculated based on the sum of weighted number of days in the last week the household ate food from eight food groups: (2 × number of days of cereals, grains, maize grain/flour, millet, sorghum, flour, bread and pasta, roots, tubers and plantains) + (3 × number of days of nuts and pulses) + (number of days of vegetables) + (4 × number of days of meat, fish, other meat and eggs) + (number of days of fruits) + (4 × number of days of milk products) + (0.5 × number of days of fats and oils) + (0.5 × number of days of sugar, sugar products and honey). Spices and condiments are excluded. The maximum value of the FCS is 126.

Table 1 presents the summary statistics for the dependent variables used in our analysis. For the two periods used (2010–2013) in the analysis, the average total expenditure, food expenditure and non-food are approximately 4,453, 3,496 and 485 Malawian Kwacha respectively.

Table 1.

Summary statistics

Notes: Authors' computation from the Malawi Integrated Household Panel Survey 2010–2013.

Also, the average FCS reported by the households is 51 out of a maximum score of 126. For the rainfall variable, the 3-year average rainfall across rural households for the periods of our data (2009/10 and 2012/2013) is 822 mm and the average historical rainfall (2001–2012) is 8,649 mm. The negative rainfall shock is 0.078 (in absolute deviation). Household size is 5 with an average household head age of 44 years.

Table 2 shows mean characteristics by household's receipt of Farm Input Subsidy. The number of households that planted crops in the previous season and received FISP is larger than the number of households that did not receive FISP. In terms of the average age of household head, the mean difference shows no statistically significant difference between the recipients of FISP and the non-recipients of FISP. Moreover, households that received FISP reported larger farm size of 2.32 acres compared to households without FISP that reported 1.892 acres. Also, the mean difference shows that 39 per cent of households in communities with a public works programme such as the Malawi Social Action Fund (MASAF) received FISP compared to 46 per cent of households in communities without a public works programme.

Table 2.

Test of mean difference of household characteristics and receipt of FISP

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

4.4 Empirical methodology

We adopt an identification strategy similar to that of Björkman-Nyqvist (Reference Björkman-Nyqvist2013) and Yang and Choi (Reference Yang and Choi2007). We exploit the exogenous variations in seasonal precipitation patterns to investigate the mitigating role of the FISP due to negative rainfall shocks on rural households' welfare. We use data on the FISP provided by Malawi's Integrated Panel Survey to investigate whether Malawi's FISP mitigates the impact of negative rainfall shocks on household welfare in rural Malawi. The primary focus of the estimation is to model the interaction of the FISP on the relationship between negative rainfall shocks and households' welfare.

The paper analyses negative rainfall shocks, FISP and household welfare nexus by estimating the following equation:

(2)\begin{equation}{Y_{ht}} = {\delta _t} + {\emptyset _f}N{S_{ht - 1}} + {\varphi _f}({{\rm N}{{\rm S}_{ht - 1}} \times {\rm FIS}{{\rm P}_{ht}}} )+ {\rm FIS}{{\rm P}_{ht}} + X_{ht}^{\rm \prime } + {\varepsilon _{ht}}, \end{equation}

where ${Y_{ht}}$ denotes household welfare measures for household h at time t, and ${\delta _t}$ represents year fixed effects. Also, parameter $\emptyset$ denotes the direct effect of negative rainfall shocks on household welfare, $\varphi$ captures our parameter of interest – the interaction of rainfall shocks (NS) and dummy for receipt of farm input subsidy by the household (FISP) – a dummy variable equals one if a household redeemed at least one input subsidy in the previous planting season and zero if otherwise. Evaluating the relationship between the interactions of coefficient estimates from $\emptyset$ and $\varphi \;$is the underlying basis for equation (2). Lastly, $X_{ht}^{\rm \prime }$ denotes household and community covariates used in the estimation and ${\varepsilon_{ht}}$ is the error term which is assumed to be normally distributed. The error term is assumed to be independent and identically distributed (iid) between villages, but correlated within EAs; hence, we clustered the standard errors at the EAs for all estimations.

4.4.1 Endogeneity of households' receipt of farm input subsidy programme voucher

The potential threat to the identification of the mitigating role of the FISP is non-random assignment of FISP recipients across the rural households. Hence, households that received the FISP voucher are likely to systematically differ from non-recipient households in some other ways. To allay this concern, we use an interaction instrumental variable by interacting the share of FISP vouchers distributed in a district with negative rainfall shock as an instrumental variable for the interaction of a dummy variable of receipt of FISP vouchers by each household with negative rainfall shock. The use of district shares of FISP vouchers received as an instrument for FISP vouchers received by a household is closely related to that which is used by Mason and Ricker-Gilbert (Reference Mason and Ricker-Gilbert2013), and Harou (Reference Harou2018).Footnote ⁷

The argument for the use of district shares of FISP vouchers receipt is that household receipt of FISP vouchers is likely to be positively correlated with the share of vouchers distributed to a district. But the district share of FISP vouchers is unlikely to affect households' welfare directly, except through households' receipt of FISP vouchers. Although we cannot empirically test whether there is correlation between the shares of FISP vouchers allocated to a district and unobserved factors that could potentially affect households' welfare, we are making the argument that any uncontrolled factors in our regressions that could affect the instrument are also likely to affect households' receipt of FISP vouchers.

Furthermore, in an instrumental variable analysis, two conditions should be fulfilled in order for the instrument to be relevant and valid. These conditions include the following: (i) the instrument must be relevant, i.e., ${\rm cov}({Z_i},\,{X_i}) \ne 0$; (ii) the instrument must fulfil the exclusion restriction condition such that ${\rm cov}({{Z_i},{\varepsilon_i}} )= 0$, that is, the instrument does not directly affect household welfare. The first-stage equations using district share of FISP received, and district share of FISP interacted with negative rainfall shock are shown below:

(3)\begin{align}{\rm Dummy\; of\; FIS}{{\rm P}_{ht}} \times {\rm Shoc}{{\rm k}_{ht}} &= {\alpha _0} + {\alpha _1}Z \times {\rm Shoc}{{\rm k}_{ct}} + {\alpha _2}X_{ht}^{\prime } + {\varepsilon _{ht}}, \end{align}

(4)\begin{align}{\rm Dummy\; of\; FIS}{{\rm P}_{ht}} &= {\alpha _0} + {\alpha _1}{Z_{ct}} + {\alpha _2}X_{ht}^{\prime } + {\varepsilon _{ht}}. \end{align}

Table A1 in the online appendix shows the results of the first-stage regressions for the relationship between receipt of FISP vouchers redeemed by households and the district shares of FISP received. The first-stage results show a positive and strong correlation between the dummy variable for FISP voucher redeemed by households and the share of FISP received in a district.

The first condition for the relevance of the instrument is established in the first-stage estimations. The second requirement for an instrumental variable is the validity or exclusion restriction, which is that the instrument should have no effect on household welfare other than through the first-stage channel. Moreover, the F-statistics of the excluded instrument suggest that district shares of FISP received is not a weak instrument as the value ranges between 114.07 and 85.57 for the Cragg-Donald Wald F statistic and Kleibergen-Paap rk Wald F statistic. See table A1 in the online appendix for details. In addition, we present the results of the reduced form regression in table A2 in the online appendix. The results show a significant relationship between district level receipt of vouchers and household welfare.