Are smallholders disadvantaged by ‘double sell low, buy high’ dynamics on rural markets in Madagascar?

ABSTRACT

Local markets in rural areas constitute the main means of market integration for smallholders in developing countries. They are used for selling and buying agricultural products and basic necessities. Frequently infrastructure is poor and transport costs high so that farmers’ access is restricted to few local markets. To understand local market dynamics we investigated the Mahafaly Plateau region in Madagascar as an example of a rural region in a developing country, where farmers depend on small local markets. We collected data on usage of markets and monitored prices for crops and livestock for two years on five markets. We find an extreme seasonality of market prices for crops and livestock. Given prevailing marketing strategies and price dynamics, farmers face a ‘double sell low, buy high’ challenge for crops and livestock leading to welfare losses, increased food insecurity and seasonal hunger.

KEYWORDS:

• Developing countries
• seasonality
• food security
• econometric analysis
• network analysis
• market monitoring

1. Introduction

Smallholders represent 50% of the world's starving people, especially in Asia and Sub-Saharan Africa (Sanchez & Swaminathan, 2005). For them, local markets in rural areas are the main mechanism of market integration. Smallholders mainly live from small-scale agriculture and livestock breeding, selling their surplus produce and purchasing foodstuffs as seasonally required (Barrett, 2008; Jayne et al. 2010). Thus, price developments on rural markets for local agricultural products impact the food security of smallholders (Akter & Basher, 2014). This holds particularly true for poor households who spend a major share of their income on food (Matz et al., 2015; Svanidze et al., 2019).

Especially on closed remote rural markets in developing countries, prices for staple food crops are subject to substantive seasonal variations (Sahn, 1989; Minten et al., 1997; Van Campenhout et al., 2015; Burke et al., 2019). Due to shifting supply and demand patterns in the agricultural crop cycle prices are generally low in the harvest season and high in the lean season (Minten et al., 1997; Van Campenhout et al., 2015). Thus, smallholders are forced to sell their harvest at low prices after the harvest, but in the lean season when they need to buy food on the market, they pay much higher prices. This ‘sell low, buy high’ behaviour of smallholders (Stephens & Barrett, 2011; Aker, 2012; Burke et al., 2019) is strongly disavantageous for them since it exacerbates food insecurity and malnutrition during lean season times (Stephens & Barrett, 2011; Bacon et al., 2014; Burke et al., 2019).

Livestock markets similarly exhibit seasonal price variations triggering also a ‘sell low, buy high’ behaviour of smallholders. However, demand and supply structures indicate that seasonal variations in livestock markets show an inverse pattern compared to that of crops (Hesse, 1987; Turner & Williams, 2002), meaning that livestock prices are highest in the harvest season and thus coincide with low crop prices, while in the lean season low livestock prices coincide with high crop prices. Frequently, livestock is the only value storing asset available for consumption smoothing and wealth accumulation in remote rural areas (Davis et al., 2010; Kocho et al., 2011; Herrero et al., 2013; Wong et al., 2017), meaning that in the harvest season stakeholders mainly buy livestock to invest cash earned by crop sales and in the lean season sell livestock to buy staple food. Again, this behaviour in combination with the market price dynamics impact smallholders’ food security negatively as they buy livestock at high prices and are forced to sell it for much lower prices.

Another factor that potentially impacts price dynamics is the remoteness of rural markets. Frequently, infrastructure is underdeveloped in remote regions in developing countries. It becomes more burdensome to transport products to markets as this can only be done by foot or oxcart (Barrett, 2008; Zanello et al., 2014). Transportation challenges may substantially increase marketing costs and thereby significantly reduce the profits of smallholder sellers and increase food prices in remote rural markets. These market dynamics are known in the literature (Hesse, 1987; Sahn, 1989; Turner & Williams, 2002; Barrett, 2008; Stephens & Barrett, 2011; Aker, 2012; Zanello et al., 2014; Burke et al., 2019).

What is to our knowledge lacking is an analysis that combines the effects of both, crop and livestock price dynamics on remote markets from the perspective of smallholders, which is what we do in the following: We propose that price dynamics on rural crop and livestock markets lead to a ‘double sell low, buy high’ behaviour of smallholders as the seasonal pattern of demand and supply in rural markets force them to exchange crops and livestock twice at unfavourable relative prices. First, smallholders sell their crops produce surplus at low prices to invest in livestock whose prices are high in the harvest season. Secondly, they may sell livestock for staple food in the lean season when prices of livestock are low and crops prices high. Thus, we suspect that the combined effects of seasonal price fluctuations of crops and livestock have an even more severe effect on smallholders’ food security than the price dynamics in crops and livestock markets alone. In addition, higher price levels on remote markets may aggravate the negative impacts on food security.

Understanding the interconnectedness of these dynamics is a central precondition to design measures to increase food security (Barrett, 2008; Jayne et al., 2010). Measures that only tackle one of the described mechanisms may not effectively increase food security. E.g. a measure that one-sidedly increases crops prices during harvest season may be able to increase revenues of smallholders in the harvest season. However, if this measure is not complemented by measures that increase livestock supply during harvest season higher crops prices may simply increase livestock prices instead of food security.

To provide empirical evidence for the ‘double sell low, buy high’ behaviour of smallholders we investigate seasonal price developments for agricultural produce and livestock in rural markets in the Mahafaly Plateau region in south-western Madagascar. Devereux (2010) and Minten et al. (1997) postulated opposite fluctuations of crop and livestock prices, however, without a detailed statistical analysis and without relating it to the consequences for smallholder market integration. Thus, to our knowledge, our study is the first combined analysis of price fluctuations for crops and livestock. In line with these objectives, we formulate two hypotheses:

H1: The combined impact of price fluctuations for staple crops and livestock products leads to a ‘double sell low, buy high’ challenge for smallholders, having a double negative impact on food security.

H2: Remote markets show higher price levels than central markets in the Mahafaly Plateau region.

We test our hypotheses with data from the Mahafaly Plateau region, which serves as a prime study area as it is a particular poor and remote region in one of the poorest developing countries of the world, where 62% of the rural population suffer from extreme poverty, i.e. are unable to obtain sufficient basic foodstuffs throughout the year (Dostie et al., 2002; INSTAT, 2011) and are particularly vulnerable to food insecurity (Krishnamurthy et al., 2014).

We used a market network analysis to identify central and remote rural markets used by locals in the Mahafaly Plateau region. On five representative markets we monitored prices for crops and livestock for two years. To scrutinise the existence of ‘double sell low, buy high’ behaviour of smallholders and the impact of market remoteness on price formation we applied panel data regression.

Methodologically we go beyond existing studies by combining information on the usage of markets by locals with monitoring of price data over a period of two years. In developing countries and particularly for rural markets, there is often a lack of information about long-term price developments, although this is an important indicator for market-related food insecurity (Barrett et al., 2009). Development projects and scientific studies thus often rely on statistical data, which is available only for central markets or large administrative units, e.g. communes (Minten & Randrianarison, 2003), which may, however, underestimate the severity of seasonal price changes on remote markets (Brown et al., 2012). When primary data is collected, recall interviews are frequently chosen, although these are often subject to large bias (Minten et al., 1997). Alternatively, price monitoring offers better options for obtaining spatially and temporally detailed data on prices for a thorough analysis, but is seldom done due to high costs of data collection. Thus, the price monitoring data over 2 years presented here represents a unique panel data set on market prices from one of the poorest rural regions in Madagascar and the world.

Moreover, we combine price monitoring data with data on the usage of markets and qualitative follow up interviews with market monitors. Information on the usage of markets was collected with a market network approach, which provides important spatial information on market use but is seldom applied in the literature on African markets and trade (Walther, 2014). For our study, it provides important preliminary information on importance and remoteness of markets.

We also add information from qualitative interviews with market monitors. The qualitative approach can provide preliminary explanations for causal mechanisms on market price developments and empirical evidence for the ‘double sell low, buy high’ behaviour of smallholders, which would remain unclear with the statistical analysis of price data only (Ngulube & Ngulube, 2015).

2. Materials and methods

2.1. Study region

The studied markets are located in the Mahafaly Plateau region in south-western Madagascar, which is part of the Atsimo Andrefana economic region. This region is exceptional within Madagascar with regard to poverty, prevailing agricultural systems and staple crop consumption patterns. Atsimo Andrefana (66,236 km² and 1.35 million inhabitants) is among the poorest regions in Madagascar with approximately 88% of rural households having an annual per capita income of below 468 800 Ariary (approx. 200 USD) (INSTAT, 2011). 89% of households cultivate cassava while rice cultivation accounts for only 19% (INSTAT, 2011; Rahaingo Vololona et al., 2013). In slash-and-burn agricultural systems, e.g. in the south-west, the main crop is maize (Casse et al., 2004). In Madagascar as a whole, rice dominates with 40% of total food consumption. Tubers (among them cassava) make up 25%: Other cereals (among them maize) 4% (INSTAT, 2011). In rural areas, tuber consumption is 20% higher than in urban areas amounting to 67.1 kg per individual, while rice consumption is slightly lower amounting to 96.6 kg per individual (INSTAT, 2011).

The exceptional poverty of the rural population coincides with high biodiversity values in the region. The ecoregion is characterised by a semi-arid climate with an annual average rainfall between 300 and 600 mm, which falls mostly in a short rainy season (December-February). Precipitation fluctuates strongly inter-annually. The region is divided into a coastal region with sandy soils and lower precipitation and an elevated plateau region with loamy soils receiving higher precipitation (Hanisch et al., 2015). This climate contributes to the richness of local biodiversity, a dry and spiny forest (Ratovonamana et al., 2011). However, the biodiversity faces increasing anthropic pressures, mainly caused by slash-and-burn agriculture (teteke), coupled with population growth and regional poverty (Ratovonamana et al., 2013; Waeber et al., 2015).

Households in the Mahafaly Plateau region are mostly agro-pastoralists and engage in semi-subsistence farming, with livestock and especially zebu playing a crucial role as insurance against droughts (Wüstefeld, 2004; FAO, 2009; Rich et al., 2011; Goetter & Neudert, 2016). However, 39% of households do not keep livestock being therefore especially poor and vulnerable (Neudert et al., 2015). Most households earn a high proportion of their income from cultivating cassava, maize, beans, and other vegetables (Neudert et al., 2015). A strong seasonality of market purchases of basic foodstuffs is observed: While only 10–30% of households buy basic foodstuffs during harvest seasons, more than 60–80% buy cassava or rice during lean seasons (December-January) (Neudert et al., 2015; Noromiarilanto et al., 2016). Poor infrastructure characterises the region, e.g. lacking roads and clean water supply, and underdeveloped educational and medical care facilities (ILO program, 2001).

2.2. Theoretical framework

To derive our hypothesis of a ‘double sell low, buy high’ behaviour we model rural markets in Sub-Saharan-Africa as small and closed markets. In this market type, equilibrium prices are sensitive to small variations in supply and demand as already small changes significantly alter total supply and demand (Mankiw & Taylor, 2014). In addition, supply and demand are rather price inelastic as market closedness hinders that intermarket trade levels out price fluctuations caused by variations in local supply and demand (Svanidze et al., 2019). We assume that only two types of goods are traded: crops and livestock.

Figures 1(a) and (b) show the market dynamics for crops and livestock in the harvest season in a small and closed rural market compared to a reference market which does not experience seasonal shifts in demand and supply. In Figure 1(a), demand for crops shifts from $D_{rm}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}$ to $D_{hs}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}$ and supply from $S_{rm}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}$ to $S_{hs}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}$ as self-sufficient smallholders reduce their demand and market their harvest surplus due to lacking storage facilities (Barrett & Dorosh, 1996). As a consequence, the market equilibrium settles in A where smallholders are forced to sell crops at a lower price ($p_{rm}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}>p_{hs}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}$). In the livestock market in Figure 1(b) demand increases from $D_{rm}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}$ to $D_{hs}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}$ as smallholders invest income generated from crop surplus selling in livestock. Due to lacking alternative investment and saving opportunities livestock is frequently considered the only value storing asset available for consumption smoothing and wealth accumulation in remote areas (Ndoro & Hitayezu, 2014). In the new market equilibrium B, livestock price increased from $p_{rm}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}$ to $p_{hs}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}$. Thus, market dynamics in small and closed rural markets force smallholders to exchange crops and livestock at unfavourable relative prices in the harvest season: $\frac{p_{hs}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}}{p_{hs}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}}>\frac{p_{rm}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}}{p_{rm}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}}.$

Figure 1. Crop and livestock markets in the harvest season.

In the lean season, smallholders reverse their selling and buying behaviour. In Figure 2(a), crop supply decreases from $S_{rm}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}$ to $S_{ls}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}$ as little crops are produced and marketed in the lean season. Demand increases from $D_{rm}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}$ to $D_{ls}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}$ as smallholders are short of crops and need to buy staple food. This shifts crop market equilibrium to C where smallholders pay higher prices ($p_{ls}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}>p_{rm}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}$). Simultaneously, the livestock price settles in the market equilibrium D in the lean season at a lower level ($p_{ls}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}<p_{rm}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}$) as smallholders sell livestock to buy crops (Figure 2(b)). Like in the harvest season, smallholders exchange crops and livestock at unfavourable relative prices: $\frac{p_{ls}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}}{p_{ls}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}}<\frac{p_{rm}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}}{p_{rm}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}}$This leads to the ‘double sell low, buy high’ behaviour auf smallholders as smallholders face twice unfavourable relative prices: $\frac{p_{ls}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}}{p_{ls}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}}<\frac{p_{rm}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}}{p_{rm}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}}<\frac{p_{hs}^{\mathrm{l}\mathrm{i}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{o}\mathrm{c}\mathrm{k}}}{p_{hs}^{\mathrm{c}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{s}}}.$

Figure 2. Crop and livestock markets in the lean season.

In bigger and more open markets, seasonal changes in relative prices may be less pronounced as seasonal variation in supply and demand alters total supply and demand less significantly. In addition, demand and supply may be more price elastic as big markets are better integrated with the rest of the economy and income sources and demand patterns of market participants are more diversified. Ceteris paribus, this may stabilise crop prices as total demand is decoupled from seasonal agricultural patterns (Svanidze et al., 2019). Smallholders which are better integrated in the rest of the economy may also have the choice to invest in a variety of saving vehicles other than livestock. In this case, seasonal supply and demand for livestock may become more price elastic as smallholders decide to invest in assets whose prices are less sensitive to the seasonal agricultural demand and supply cycle. As a consequence, variations in income of smallholders during the harvest and lean season impact livestock prices less.

2.3. Market network analysis

To assess market existence and use, data was gathered in focused group discussions in 2011 (Brinkmann et al., 2014). Among other information, the questionnaire included markets used by local population ordered by frequency of visits in four categories. 79 settlements in an area of approx. 4000 km² in the communes Beheloka, Beantake, Masiaboay and the southern parts of Betioky Sud and Soalara were covered. Interviews were conducted in groups with 15–20 inhabitants, among them the local representative of the community (president of fokontany). Respondents were selected as a convenience sample due to time constraints and lacking preliminary information (Krueger & Casey, 1994). Data on visit frequencies was the input for our market network analysis. We used the software Ucinet (Borgatti et al., 2002) to identify market clusters and central markets with visitors from different regions and compiled a matrix containing settlements and markets. If a settlement has a market, it occurs twice in the dataset, as settlement and market. We applied Ucinet Netdraw to create a visual depiction of the market network and used Girvan-Newman clustering to identify market clusters within the network (Borgatti et al., 2002).

2.4. Market monitoring of crops and livestock products

We monitored prices over 107 weeks from mid-January 2013 to mid-January 2015 on five markets in the Mahafaly Plateau region. The selection of markets was based on the market network analysis (see results in section 3.1, Figure 4) with the selection criteria: remoteness/centrality and coverage of different regions. Efoetse and Marofijery were selected as remote markets in the littoral region, Andremba and Itomboina as remote markets in the plateau region. Ambatry qualified as market of regional importance.

The currency of recordings was Madagascar Ariary (MGA; 1 US$ = 2,474 MGA: 4 September 2014, www.oanda.com). Local assistants monitored markets weekly or biweekly. Data collection included prices of available agricultural commodities such as crops, livestock as well as local alimentary plants and tubers. The first price given by the seller was taken. This price may deviate from the actual selling price since bargaining is common. Recording of prices started usually around 8.00 am. The full data set was organised in weekly intervals. In our analysis, we cover the most frequently available products (see Table 1).

Table 1. Crop availability in % of moving 4-week intervals on 5 different markets in Mahafaly Plateau region and mean prices 2013–2014.

Crops			Availability in % of moving 4-week intervals						Mean price
Crop type	Crop (Malagasy name)	Latin name	Ambatry	Andremba	Efoetse	Itomboina	Marofijery/Ankilibory	Unit	Mean	(Stddev)
Beans	Cowpea (lojy)	Vigna unguiculata	94.1	100.0	88.2	100.0	92.2	MGA/kp^a	324	(103)
	Hyacinth bean (antake)	Lablab purpureus	51.0	51.0	23.5	58.8	35.3	MGA/kp	260	(48)
	Mung bean (antsamby)	Vigna radiata	92.2	66.7	56.9	84.3	64.7	MGA/kp	465	(190)
	Rice bean (antsabim-bazaha)	Vigna umbellata	60.8	11.8	25.5	37.3	7.8	MGA/kp	408	(130)
	Bean (tsaramaso)	Phaseolus vulgaris	31.4	0.0	9.8	0.0	13.7	MGA/kp	711	(93)
Tubers	Cassava, dried (balahazo)	Manihot esculenta	88.2	76.5	76.5	78.4	78.4	MGA/kg	713	(302)
	Sweet potatoes (bele)	Ipomoea batatas	54.9	58.8	21.6	62.7	43.1	MGA/kg	1003	(353)
Grains	Maize (tsako)	Zea mays	92.2	100.0	92.2	100.0	94.1	MGA/kp	252	(69)
	Millet (bajiry)	Pennisetum glaucum	0.0	0.0	19.6	3.9	43.1	MGA/kp	310	(25)
	Sorgho (ampemba)	Sorghum bicolor	7.8	0.0	5.9	0.0	7.8	MGA/kp	363	(143)
Melons and pumkins	Calabash (mody)	Lagenaria siceraria	60.8	7.8	5.9	0.0	15.7	MGA/piece	166	(89)
	Pumpkin (taboara)	Cucurbita maxima var.	82.4	39.2	35.3	49.0	39.2	MGA/piece	628	(270)
	Sugar melon (voantango)	Cucumis melo var.	25.5	21.6	25.5	33.3	19.6	MGA/piece	472	(209)
	Sugar melon (kiseny)	Cucumis melo var.	43.1	27.5	7.8	52.9	19.6	MGA/kg	192	(26)
	Water melon (vamanga, voazavo)	Citrullus lanatus var	35.3	19.6	29.4	33.3	23.5	MGA/piece	406	(151)
Others	Bambara groundnut (voanjobory)	Vigna subterranea	90.2	17.6	29.4	11.8	3.9	MGA/kp	364	(78)
	Peanut, dried (kapiky)	Arachis hypogaea	31.4	68.6	49.0	88.2	66.7	MGA/kp	171	(45)
	Tomatoes (voatabia)	Solanum lycopersicum var.	80.4	33.3	33.3	43.1	37.3	MGA/kg	255	(84)
Mean availability of crop products			56.8	38.9	35.3	46.5	39.2

^a kp = kapoaka, a tin cup with a volume of 0.39 l.

We defined standard units for different crops and converted prices recorded for other units to standard units based on conversion factors derived from weightings. For some products, e.g. melons and pumpkins, a price range was recorded due to the different sizes of the fruits. Here, we used mean prices. For testing seasonality in crop price data, the harvest season and the lean season were predefined according to local information and dataset limitations. Figure 3 depicts the definition of harvest and lean seasons for different crops (SuLaMa, 2011). Since many crops were sold only seasonally, a statistical analysis was only possible for most frequently traded crops. For the analysis of availability, prices were arranged in four-week intervals moving in steps of 2 weeks, i.e. the interval from week 3–6 is followed by the interval from week 5–8.

Figure 3. Harvest and lean season periods for different crops.

For livestock, collected data included prices for living animals (sheep, goats and zebu) and products (meat and milk). Live animals were recorded in 72 locally used categories for zebu and 44 categories for sheep and goats. Categories are mainly based on age, body condition and, for females, reproductive performance. For zebu, also castration and use as draught animal define categories. For descriptive statistics, categories were summarised as illustrated in Table 2. To analyse price variability, we calculated price spreads as follows: (1) $\text{S}=\frac{p_{\mathrm{m}\mathrm{a}\mathrm{x}}-p_{\mathrm{m}\mathrm{i}\mathrm{n}}}{p_{\mathrm{m}\mathrm{a}\mathrm{x}}}\times 100$(1) where S denotes the price spread and p_max and p_min the maximum and minimum prices for products. Price spreads are calculated from averages over all markets.

Table 2. Livestock availability in % of moving 4-week intervals on 5 different markets in Mahafaly Plateau region and mean prices 2013–2014.

Livestock			Availability in % of moving 4-week intervals					Mean price
Category	Type	Age class	Ambatry	Andremba	Efoetse	Itomboina	Marofijery/Ankilibory	Unit	Mean	Stddev
Goats	Castrated	up to 1 year	16.0	8.7	14.6	24.9	19.6	MGA/head	21,128	(4,245)
		older than 1 year	64.7	11.5	34.6	24.8	51.0	MGA/head	65,981	(16,865)
	Female	up to 1 year	26.9	28.9	31.7	49.3	59.1	MGA/head	17,568	(4,579)
		older than 1 year	51.6	22.5	16.7	36.9	61.9	MGA/head	34,338	(6,714)
	Male	up to 1 year	39.2	36.4	27.2	54.9	67.8	MGA/head	20,285	(8,547)
		older than 1 year	52.0	6.1	7.8	21.1	46.1	MGA/head	50,891	(13,626)
Sheep	Castrated	up to 1 year	14.8	0.0	13.4	13.2	19.3	MGA/head	24,113	(6,318)
		older than 1 year	61.3	2.9	25.0	22.3	44.9	MGA/head	72,378	(34,194)
	Female	up to 1 year	42.9	15.4	32.8	46.8	57.1	MGA/head	21,137	(7,240)
		older than 1 year	48.4	5.7	17.5	27.8	63.4	MGA/head	40,897	(11,709)
	Male	up to 1 year	35.9	17.6	31.7	45.9	63.9	MGA/head	21,579	(6,328)
		older than 1 year	46.3	2.0	13.5	15.7	47.8	MGA/head	49,541	(12,499)
Zebu	Castrated	up to 2 years	31.1	0.5	7.6	15.0	20.1	MGA/head	164,386	(53,882)
		older than 2 years	54.5	0.0	8.7	14.8	4.5	MGA/head	632,231	(202,905)
	Female	up to 3 years, no parturitions	35.7	1.3	23.3	38.1	28.1	MGA/head	182,235	(49,600)
		females with parturitions	40.8	0.0	18.8	13.3	24.3	MGA/head	276,696	(71,994)
	Konda	up to 2 years	30.8	0.0	12.6	35.9	10.9	MGA/head	152,395	(30,756)
		older than 2 years	45.8	0.4	11.2	20.7	8.7	MGA/head	647,763	(154,780)
	Male	up to 2 years	42.6	0.0	29.7	42.0	40.3	MGA/head	146,653	(40,148)
		older than 2 years	43.5	0.0	13.0	8.4	9.1	MGA/head	501,897	(86,259)
Mean availability of live animals			41.2	8.0	19.6	28.6	37.4
Meat	Goat		86.3	37.3	80.4	84.3	92.2	MGA/kg	4,411^a	(464)
	Sheep		86.3	3.9	31.4	11.8	21.6	MGA/kg	4,173^a	(470)
	Zebu		86.3	7.8	72.5	76.5	37.3	MGA/kg	3,775^a	(450)
Milk	Goat		3.9	45.1	92.2	58.8	86.3	MGA/kp	256	(52)
	Zebu		58.8	43.1	35.3	41.2	15.7	MGA/kp	209	(28)

^a excluding Andremba market due to inconvertible unit (piece).

2.5. Follow-up interviews with local monitors

Market monitors were asked in single-person interviews for feedback on diagrams showing crop and livestock data recorded in 2013. Using semi-structured interview guidelines we posed questions on phenomena visible in the diagrams (e.g. periods of especially high prices, comparisons between different market places). This feedback allowed us to cross-check findings from the data and collect additional information on local perceptions of possible reasons for price differences and fluctuations. Monitors understood the logic of the diagrams as most of them had basic knowledge on the market logic of supply and demand.

2.6. Econometric analysis

To scrutinise empirically the factors suspected to have an influence on crop and livestock price formation, we applied the linear panel model with panel-corrected standard errors (PCSE) developed by Beck & Katz (1995). We used pooled OLS regression as the Breusch–Pagan LM test favours this type of panel model over random and fixed effects models. Out of the set of pooled OLS models we selected the PCSE model to account for diagnosed cross-sectional dependence, temporal autocorrelation and heteroscedasticity in the underlying data (Hoechle, 2007). Commonly, contemporaneous correlation of the residuals of a panel model is an issue if the panel – as in our case – includes a long time series and relatively few cross-sectional observations (Baltagi, 2013). Since the used panel is unbalanced, we applied the pairwise option to ensure that all available observations are considered in the model.

Mathematically, the PCSE model may be written as (Bailey & Katz, 2011): (2) $y_{i,t}=x_{i,t}\text{β}+{ϵ}_{i,t}\text{with}i=1,\dots ,M\text{and}t=1,\dots ,T$(2) where $y_{i,t}$ is the dependent variable, $x_{i,t}$ a vector of independent variables and ${ϵ}_{i,t}$ an error term which is potentially auto-correlated across $t$ and contemporaneous correlated across $i$. $M$ represents the number of units and $T$ the number of periods.

To explain price developments for crops, we ran the PCSE model for the four most important crops, namely maize, cassava, cowpea and mung beans. To capture the impact of intra-annual fluctuations in demand and supply on prices and to test the ‘sell low, buy high’ hypothesis we incorporated dummy variables for the lean and harvest season as explanatory variables (see Figure 3). To account for market centrality we controlled for whether crops were sold in Ambatry. Based on our previous findings, Ambatry is the most central market (see section 3.1). In addition, we included weekly rainfall as explanatory variable to control for the effect of seasonal variation in precipitation on price formation. Since sowing of crops frequently takes place after rain events, smallholders need to obtain inputs for cropping during this period (Sulama, 2011). We therefore hypothesise that rainfall events increase the demand and, as a result, the prices for crops which can also be used as seeds. We recorded rainfall in 30 min intervals using a Hobo Weather Station (Onset Company, Bourne, MA, USA) in Efoetse, Andremba and Miarintsoa (located 2 km from Itomboina). Data was transformed into weekly intervals to match the time unit of the dependent variable. Finally, to control for the impact of monthly inflation on price trends in rural markets we included monthly data on the Madagascar Consumer Price Index (MCPI) for 2013 and 2014 into the model (Trading Economics, 2017).

To explain price developments for livestock, we ran the PCSE model for goats, sheep and zebu, each for the local age category with most available data. We included MCPI and a dummy variable for Ambatry to account for inflation and market centrality. To account for the impact of supply and demand fluctuations across seasons we included dummy variables for lean and harvest seasons of cassava, which constitutes the main staple crop in the region (Sulama, 2011). Since the sex of an animal may impact weight and meat quality and, as a result, prices of sold animals we control for male livestock.

3. Results

3.1. Market network and relative importance

Markets in the region take place regularly, i.e. once every week or every 10 days. The network analysis of visited markets (Figure 4) depicts markets visited by local villagers and their relative importance. Although the geographic location of villages was no model input the network model reproduces a regional map, indicating that locals tend to use markets in their vicinity. The cluster analysis clearly distinguishes markets in the coastal region and the plateau region due to access limitations. Two markets (Itomboina and Beroy) have visitors from the coastal and plateau region. The network model also identifies two markets of high importance (Betioky and Ambatry).

Figure 4. Market relations of 79 villages in the Mahafaly Plateau region.

Notes: Dots show villages, squares markets. Lines indicate which market is visited by the inhabitants of the village. Size of squares depicts market importance.

The network cannot depict trade relations (i.e. transport and resale of goods) between local markets comprehensively. Market monitors stated that traders from Toliara and Itampolo mainly visit Efoetse and Marofijery/Ankilibory, while traders and farmers from Betioky and the Onilahy region mainly frequent plateau markets. Trade is particularly strong between the coastal zone and Itomboina with decreasing intensity though during lean seasons due to security problems with cattle raiders (malaso).

3.2. Traded crops and livestock products

We recorded market data for 19 locally grown crops: tubers (cassava and sweet potatoes), grains (maize, millet and sorghum), different kinds of beans, melons and pumpkins, and other crops (Bambara groundnuts, peanuts and tomatoes). Maize, cowpea and cassava are most frequently offered in markets (Table 1). Other products, e.g. pumpkins and melons, are only seasonally or occasionally available. Availability of sorghum, millet and calabash is lowest. The central Ambatry market provides the most constant range of products.

Table 2 depicts availability and mean prices of livestock and livestock products on the five monitored markets. Living animals are most constantly available in the moving 4-week intervals in Ambatry followed by Marofijery. Andremba has drastically low availability for live zebu cattle and sheep, indicating its close-to-dysfunctionality. Standard deviations for mean prices are comparably high on all markets, indicating that livestock characteristics not captured by age categories influence prices. Monitors confirmed this by stating that e.g. body conditions strongly influence prices.

Sheep milk is taboo (fady) among Mahafaly and Tanalana ethnicities and therefore not sold. Most frequently butchers sell goat meat in markets. Only in Ambatry sheep meat is sold as often as goat meat while it is only rarely available at Itomboina and Andremba.

Local age categories for living animals have a strong influence on prices. Supplementary material S-1 and S-2 depict category prices for female goats and male zebu cattle. While prices for females only increase until they reach reproductive age and stagnate afterwards, prices for males increase also in higher age classes. Monitors confirmed this relationship indirectly by specifying body condition (fatness) as main price determinant, which increases with age of males. Zebu prices are furthermore positively influenced by colour (particularly red) and horn size. Particularly the Mahafaly ethnicity around Ambatry seems to value big horns.

3.3. Factors influencing crop prices

Table 3 contains the results of the PCSE panel model. We scrutinised empirically the impact of weekly rainfall, seasonality, MCPI and market centrality on the prices of four crops. Models are statistically significant as the probability of obtaining the reported chi-square values is very low for all crops (Prob > chi2 in Table 3). The overall model fit is reasonable for cowpea, mung bean and maize as the variation in independent variables explains about half of price variations. It is less good for dried cassava where it explains one third of price variations (R² = 32.5%).

Table 3. Parameter estimates of the linear panel model with panel-corrected standard errors (PCSE) on the prices of 4 crops in the Mahafaly Plateau region in 2013 and 2014.

	Price of cowpea (Ar/kp)	Price of mung bean (Ar/kp)	Price of maize (Ar/kp)	Price of dried cassava (Ar/kg)
Rainfall (weekly in mm)	.70*** (0.001)	−.01 (0.970)	.20 (0.130)	−1.87** (0.023)
Consumer price index (MCPI)	2.11** (0.013)	3.98** (0.013)	1.58*** (0.006)	−9.78** (0.018)
Dummy for lean season	86.46*** (0.000)	159.98*** (0.000)	33.34*** (0.010)	105.12 (0.220)
Dummy for harvest season	.27 (0.989)	−52.74 (0.117)	−11.17 (0.480)	−165.88* (0.056)
Dummy for Ambatry market	30.31 (0.330)	−19.93 (0.658)	81.55*** (0.000)	173.47 (0.151)
Constant	−420.71 (0.137)	−908.77* (0.087)	−300.12 (0.116)	3978.72*** (0.004)
R2	0.5347	0.5071	0.5303	0.3251
No. Obs.	192	116	192	158
Prob > chi2	p < 0.001	p < 0.001	p < 0.001	0.0018

p-values in brackets; *** significant at 1% p-level, ** significant at 5% p-level, * significant at 10% p-level.

Seasonality is confirmed having a significant impact. Prices of cowpea, mung bean and maize are significantly higher during lean seasons (p < 0.01). Harvest seasons impact dried cassava prices significantly negatively (p = 0.056). There is no empirical evidence that prices of cowpea, mung bean and maize differ significantly between harvest and normal seasons though. Also dried cassava prices do not alter significantly during lean seasons. Models’ outcomes confirm cowpeas price patterns, as illustrated in Figure 5. Prices rise strongly during lean seasons across all markets and remain rather constant in other periods. Mean price spreads range between 206% (2013) and 167% (2014) for cowpea, 123% (2013) and 120% (2014) for maize and 186% (2013) and 257% (2014) for cassava.

Figure 5. Price development for cowpea in 2013 and 2014 on five markets in the Mahafaly Plateau region, data organised in 4-week moving intervals.

Follow-up interviews with monitors confirmed seasonal price variations, especially price increases during lean seasons when households run out of stocks and buy staple food. They also traced fine-scale price variations between markets to differences in climatic conditions (especially rainfall occurrence). Respondents associated localised high prices to single events, e.g. external traders’ visits.

Estimates of rainfall effects on prices varied substantially among investigated crops. A positive effect on cowpea prices (p = 0.01) and a negative effect on dried cassava prices (p = 0.023) was confirmed. At the same time, rainfall does not seem to influence mung bean and maize prices significantly. Possibly, rainfall impacts are already included in lean season effects as rainy and lean seasons tend to coincide for these two crops. This interpretation may be supported by the fact that the rainfall parameter estimate is significant at the 10% level if the price of maize is only regressed on rainfall. For mung bean prices, the parameter estimate of rainfall remains insignificant though if rainfall is the only included independent variable.

In the follow-up interviews, monitors confirmed the occurrence of cowpea price shocks after rainfall events. Same patterns were also reported for maize, which however could not be confirmed statistically. According to market monitors, price shocks are caused by the demand for seeds people need to buy for sowing after rainfall events as they have consumed all of the preceding harvest. Thus, the effect occurs with crops where the harvested product can be re-used as seeds, such as maize and beans, while the effect does not occur with cassava, which is re-planted from stem cuttings.

As expected, models estimated a significant positive relationship between MCPI and the price of cowpea, mung bean and maize. One expects inflation to trigger an increase in crop prices to adjust for increased general price levels. In contrast, the relation between dried cassava and MCPI was estimated to be negative (p = 0.018). As this is contrary to common expectations, a different factor is likely to override the MCPI effect.

With the exception of maize, the model does not find a significant relationship between crop prices and market centrality. Only maize seems to be sold at significantly higher prices on the most central market Ambatry. The presence of national and international retailers might influence prices there.

3.4. Factors influencing livestock prices

Table 4 depicts the estimation results of the PCSE panel models for the impacts of seasonality, MCPI, sex of the animal and market centrality on the prices of three livestock species in the Mahafaly Plateau region. Models are statistically significant as the probability of obtaining the reported chi-square values is very low for all livestock types (Prob > chi2 in Table 4). In addition, the overall fit of the model is reasonable for sheep and goats as the variation in independent variables explains more than half of price variations. For zebu, the regression model explains more than a third of the price fluctuations (R² = 38%).

Table 4. Parameter estimates of the linear panel model with panel-corrected standard errors (PCSE) on the prices of 3 livestock animals in 4 markets in the Mahafaly Plateau region in 2013 and 2014.

	Price of sheep (vibine)	Price of goats (vibine)	Price of Zebu (sakany)
Dummy for male sex	−297.72 (0.719)	174.95 (0.807)	8347.32 (0.360)
Consumer price index (MCPI)	−162.63*** (0.002)	−60.03 (0.141)	−1942.42*** (0.000)
Dummy for lean season	−3314.36*** (0.004)	−2845.81*** (0.003)	−21849.98* (0.075)
Dummy for harvest season	1347.44 (0.235)	2232.28** (0.021)	−2264.79 (0.853)
Dummy for Ambatry market	−3308.85** (0.020)	−5436.38*** (0.000)	−15603.38 (0.117)
Constant	77596.56*** (0.000)	41418.7*** (0.002)	832397.4*** (0.000)
R2	0.5825	0.5566	0.3752
No. Obs.	163	194	117
Prob > chi2	0.0001	p < 0.0001	0.0003

p-values in brackets; *** significant at 1% p-level, ** significant at 5% p-level, * significant at 10% p-level.

Like for crops, seasonality plays a significant role in the formation of livestock prices. Patterns are reversed though as livestock prices are significantly lower during lean seasons. Negative parameter estimates of lean season dummy variables are significant at the 1% level for sheep and goat prices and significant at the 10% level for zebu prices (see Table 4). In addition, goat prices are significantly higher during harvest seasons (p = 0.0215). To some extent, model outcomes confirm price developments for female goats depicted in Figure 6. While prices are elevated during harvest seasons, prices fall significantly during lean seasons compared to the rest of the year. Price spreads range between 81% (2013) and 106% (2014) for male sheep and 121% (2013) and 211% (2014) for male zebu.

Figure 6. Price development for female vibine goats (6 months to 1 year and 2 months old) in 2013 and 2014 on five markets in the Mahafaly Plateau region, data organised in 4-week moving intervals.

In the follow-up interviews market monitors confirmed high goat and sheep prices in harvest seasons as people sell their harvest and invest cash in small stock. At that time animals are also slaughtered at social events. Harvest failure (kere) corresponds with very low prices for animals. For the respondents, it was understandable to see fluctuations to be less strong for zebu cattle since most people do not earn enough cash with their harvest to invest in zebu cattle. They explained that for local households acquiring zebu cattle for later consumption at funerals or other social events requires more time. Seasonally, prices vary therefore less.

In contrast, the sex of an animal does not seem to have a significant influence on prices. Also puzzling are the significant negative MCPI parameter estimates for vibine sheep and sakany zebu (p < 0.01). As this runs contrary to common expectations, we suspect that other factors override MCPI effects. As the inter-annual decrease in price levels is systematic across all livestock types and markets (data not shown), the factor must have at least regional scope, which points to a climatic effect or political factor.

Finally, market centrality significantly impacts sheep and goats prices negatively. This indicates that sheep and goat livestock supply might be more readily available on the central market of Ambatry.

In the follow-up interviews, monitors saw differences between markets particularly regarding the availability of animals. Itomboina is well known in the study region as livestock market attracting buyers from the entire Onilahy region.

4. Discussion and conclusion

To understand local market dynamics and their impact on smallholder market participation in south-western Madagascar we monitored and analysed prices for crops and livestock on five markets in the Mahafaly Plateau region over 107 weeks between January 2013 and January 2015. The monitoring of prices reveals insights into factors affecting prices of rural markets in the Mahafaly region which are in line with the existing literature (e.g. Dostie et al., 1999, 2002) and provide support for the ‘double sell low, buy high’ behaviour of smallholders which exacerbates food insecurity and poverty in the region.

We find empirical evidence for strong seasonality of crop prices. Price spreads for crops ranged between 106 and 257%. Compared to price spreads for agricultural products reported in the literature (Sahn & Delgado, 1989; Minten et al., 1997), spreads of more than 250% are extreme. Possible reasons are losses due to infestation, especially for cassava, high interest rates and low availability of capital in the study region (Manon, 2015; Randrianarison et al., 2017). Our results suggest that rainfall events in the lean season lead to price increases for crops which can also be used as seeds, indicating a strong demand for seeds during these periods. In addition, prices might vary according to unobserved characteristics, e.g. crop quality. For the local households, the marked seasonality leads to strong ‘sell low, buy high’ dynamics (Ellsworth & Shapiro, 1989; Dostie et al., 2002).

The ‘sell low, buy high’ phenomenon is also present in local livestock markets, especially for goats and sheep, and to a lesser extent for zebu cattle. We find seasonal price spreads of 81–211%. Seasonal livestock price variations are reversed compared to observed crop price patterns: Prices are lowest during lean seasons and highest during harvest seasons. The price spreads found in our data are much higher than the approx. 50% spreads reported for zebu cattle (Minten et al., 1997). This might be also related to the fact that price seasonality and thus nutritional stress are reported to have increased in last decades in Madagascar (Khan et al., 1993; Dostie et al., 2002).

Thus, households in the Mahafaly Plateau region even face a ‘double sell low, buy high’ challenge, confirming H1 (Figure 7). Price variations in crop and livestock markets are even stronger than previously reported in the literature. Likely causes of the extreme price variations are the extreme rurality of monitored markets, poverty among local population, and strong crops seasonality. Especially very poor households are likely to be negatively affected by strong price fluctuations while more wealthy households may be able to profit by delaying sales and purchases (Van Campenhout et al., 2015).

Figure 7. The double ‘sell low, buy high’ challenge: Farmers behaviour and market prices result in cyclical income losses for farmers.

The severity of the ‘double sell low, buy high’ challenge for smallholders can be illustrated by a small example: During the harvest season in May/June 2014, a smallholder in our study region needed to sell approx. 31 kg of cowpeas or 32 kg of maize to buy one young goat. In the lean season 2015, when he wanted to sell the same goat to buy food he got for the goat only 8 kg of cowpeas or 10 kg of maize. This illustrates that the exchange value of staple food in relation to livestock more than triples in the lean season compared to the harvest season, which in the worst case reduces the availability of food in the smallholder household by the factor 3. Thus, the ‘double sell low, buy high’ challenge exacerbates the extreme poverty and seasonal hunger observed in the region in general.

In contrast to seasonality, market centrality as a measure for market remoteness has a less clear effect on prices. While livestock price levels tend to be significantly lower in the region's most central market Ambatry, maize was sold at a significantly higher price level. Thus, observed livestock prices provide support for H2 that marketing costs and, as a consequence, price levels are lower in more central markets, while observed maize prices provide support for rejecting H2 for crops. One explanation might be the presence of external traders buying maize for national markets and export in Ambatry (Stifel & Randrianarisoa, 2006).

Overall, our results underline the extreme rurality of markets in the Mahafaly Plateau region. Considering the strong price volatility we find in our study, smallholders are most likely caught in a semi-subsistence poverty trap, where the exclusive access to dysfunctional markets discourage productivity increases and economic development (Barrett, 2008; Jayne et al., 2010). Insufficient infrastructural access, high transport costs and little trans-regional trade opportunities leave subsistence farmers as the main suppliers and buyers in those markets. Supposedly, this creates the seasonal supply and demand and price patterns which levy the ‘double sell low, buy high’ burden on smallholders.

The following policy recommendations follow from our analysis:

1. Measures for increasing market participation of smallholders alone are likely to be counterproductive (Chilowa, 1998), and should be accompanied by measures that improve market institutions and transport infrastructure to integrate markets trans-regionally (Poulton et al., 2006; Barrett, 2008; Jayne et al., 2010). A continued monitoring of prices should be applied to check if markets become more functional and the strong price seasonality decreases.

2. In-kind food aid by international donors and payments for ecosystem services should be targeted precisely to periods with food shortages (Barrett et al., 2009; Randrianarison et al., 2017). While international donors can buy food on world markets with less seasonal price spreads, the local benefits achieved among the receiving population are highest.

3. For crops, improved storage facilities (Kotu et al., 2019) and micro-credit systems (Barrett, 2008) can help to mitigate the strong seasonality of sales and purchases. The establishment of micro-banking may allow smallholders to invest crops surpluses into assets whose prices are decoupled from unfavourable price fluctuations in rural markets. This may shield them from the ‘double sell low, buy high’ challenge they face in rural livestock markets.

Acknowledgements

The study was carried out within the SuLaMa project under the ‘Accord de Collaboration’ between the Universities of Antananarivo, Madagascar National Parks and the German partner universities and under the cooperation agreement between the German Universities of Hamburg and Cottbus-Senftenberg. The authors are grateful to the World Wide Fund for Nature (WWF) and the SuLaMa project team for supporting their work. Market monitors and field assistant made data collection feasible in 5 villages, particularly Leopold Andrianjohary. We thank Frank Wätzold and one anonymous reviewer for constructive comments that helped to further improve the manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The study was funded by the German Federal Ministry of Education and Research (BMBF) as part of a project on sustainable land management of the Mahafaly Plateau (SuLaMa; grant numbers 01LL0914G, 01LL0914A).